CONCEPTUAL DATA

MODELING AND
DATABASE DESIGN:
A FULLY ALGORITHMIC
APPROACH
Volume 1
The Shortest Advisable Path
This page intentionally left blank
 CONCEPTUAL DATA
MODELING AND
DATABASE DESIGN:
A FULLY ALGORITHMIC
APPROACH
Volume 1
The Shortest Advisable Path

Christian Mancas
Mathematics and Computer Science Department,
Ovidius State University, Constanta, Romania
Computer Science Taught in English Department,
Politehnica University, Bucharest, Romania
Software R&D Department, Asentinel International and
DATASIS ProSoft, Bucharest, Romania
CRC Press Apple Academic Press, Inc
Taylor & Francis Group 3333 Mistwell Crescent
6000 Broken Sound Parkway NW, Suite 300 Oakville, ON L6L 0A2
Boca Raton, FL 33487-2742 Canada
© 2016 by Apple Academic Press, Inc.
Exclusive worldwide distribution by CRC Press an imprint of Taylor & Francis Group, an Informa
business

No claim to original U.S. Government works

Version Date: 20150914

International Standard Book Number-13: 978-1-4987-2844-7 (eBook - PDF)

This book contains information obtained from authentic and highly regarded sources. Reason-
able efforts have been made to publish reliable data and information, but the author and publisher
cannot assume responsibility for the validity of all materials or the consequences of their use. The
authors and publishers have attempted to trace the copyright holders of all material reproduced in
this publication and apologize to copyright holders if permission to publish in this form has not
been obtained. If any copyright material has not been acknowledged please write and let us know so
we may rectify in any future reprint.

Except as permitted under U.S. Copyright Law, no part of this book may be reprinted, reproduced,
transmitted, or utilized in any form by any electronic, mechanical, or other means, now known or
hereafter invented, including photocopying, microfilming, and recording, or in any information
storage or retrieval system, without written permission from the publishers.

For permission to photocopy or use material electronically from this work, please access www.
copyright.com (http://www.copyright.com/) or contact the Copyright Clearance Center, Inc.
(CCC), 222 Rosewood Drive, Danvers, MA 01923, 978-750-8400. CCC is a not-for-profit organiza-
tion that provides licenses and registration for a variety of users. For organizations that have been
granted a photocopy license by the CCC, a separate system of payment has been arranged.

Trademark Notice: Product or corporate names may be trademarks or registered trademarks, and
are used only for identification and explanation without intent to infringe.
Visit the Taylor & Francis Web site at
http://www.taylorandfrancis.com
and the CRC Press Web site at
http://www.crcpress.com
For information about Apple Academic Press product
http://www.appleacademicpress.com
 To the loving memory of my parents,
to my revered Professors,
to my students, colleagues, love, children, and friends.

Our standards in databases (and not only) should always be at least equal to
the maximum of the corresponding highest academic and customer ones.

Laudamus veteres, sed nostris ultimur annis:

mos tamen est aeque dignus uterque coli.
—Publius Ovidius Naso
We cannot have a future without respecting our past.
—King Michael I of Romania
What we think, we become.
All that we are arises with our thoughts.
With our thoughts, we make the world.
—The Buddha
My prayer:
Lord, do not ever let me be satisfied with myself.
—Lucian Blaga
THINK.
—Thomas J. Watson
As we look ahead into the next century,
leaders will be those who empower others.
—Bill Gates
I believe people have to follow their dreams—I did.
—Larry Ellison
The cave you fear to enter holds the treasure you seek.
—Joseph Campbell
The Key
The whole Key
and nothing but the Key,
so help me Codd.
—David J. Emmick
This page intentionally left blank
CONTENTS

Dedication..................................................................................................... v
About the Author.......................................................................................... ix
Foreword by Professor Bernhard Thalheim................................................. xi
Foreword by Professor Dan Suciu...............................................................xv
Preface...................................................................................................... xvii
Acknowledgments..................................................................................... xxiii
List of Symbols...........................................................................................xxv
1. Data, Information and Knowledge in the Computer Era....................... 1
2. The Quest for Data Adequacy and Simplicity:
The Entity-Relationship Data Model (E-RDM)..................................... 35
3. The Quest for Data Independence, Minimal Plausibility,
and Formalization: The Relational Data Model (RDM)..................... 115
4. Relational Schemas Implementation and Reverse Engineering......... 341
5. Conclusion................................................................................................ 595
Appendix: Mathematic Prerequisites for the Math Behind....................... 615
Index.......................................................................................................... 635
This page intentionally left blank
ABOUT THE AUTHOR

Christian Mancas, PhD, is currently an Associate Professor with both the

Mathematics and Computer Science Department of Ovidius University,
Constanta, Romania, and the Engineering Taught in Foreign Languages
Department (Computer Science and Telecommunications in English
stream) of Politehnica University, Bucharest, Romania (as an invited Pro-
fessor). Since 2012, he is also a database architect with Asentinel Interna-
tional srl, Bucharest, a subsidiary of Asentinel LLC, Memphis, Tennessee.
His specialties include university teaching, R&D, business analysis,
conceptual data and knowledge modeling and querying, client-server
and hierarchical software architecture, object-oriented and event-driven
design, structured development, complex projects and small IT company
management, Datalog, SQL, C#, XML programming, etc.
Professor Christian Mancas has published dozens of scientific papers
(in Romania, USA, Austria, and Greece), which have been indexed by
ACM Digital Library, Zentralblatt MATH, Scopus, DBLP, Arnetminer,
Researchr, TDGS, SCEAS, etc. He has also published three books in
Romanian and dozens of reviews (mostly in USA, including ACM
Reviews). He was a program committee member and session chairman
for several software conferences in the USA, Austria, and Romania, and
he is a member of several associations (including ACM, the Romanian
Mathematics Sciences Society, and the International Who’s Who of Pro-
fessionals). Since 2006, his biography is included in Marquis Who’s Who
in the World and Who’s Who in Science and Technology and Hubners’
Who’s Who in Romania.
Since 1990, he also worked for several IT startups, including his own
DATASIS Consult srl (co-owned with his good friend and faculty col-
league Ion Draghicescu) and DATASIS ProSoft srl (which had 25 pro-
grammers working for the design and development of several ERP-type
database applications for customers from France, UK, Switzerland, USA,
Israel, Greece, and Romania).
x About the Author

His main research areas are the conceptual data and knowledge mod-
eling and querying, database design, implementation, and optimization,
as well as the architecture, design, development, fine-tuning, and mainte-
nance of data and knowledge base management systems.
Dr. Mancas graduated in 1977 from the Computers Department at
Politehnica University of Bucharest, Romania, with a thesis on Generating
Parsers for LR(k) Grammars, under the supervision of Professor Dan Luca
Serbanati. Up until the fall of communism in 1990, he worked as a soft-
ware engineer and, since 1980, had been R&D manager of a state-owned
computer center in Bucharest (contributing to the design, development,
and maintenance of a dedicated ERP), and conducted (from time to time)
computer programming labs at Politehnica University of Bucharest, but,
for political reasons, he was not accepted for PhD studies. He started this
program under the supervision of Professor Cristian Giumale in 1992 and
obtained his PhD degree in 1997 from the same as above department, with
a thesis on Conceptual Data and Knowledge Modeling.
FOREWORD BY
PROFESSOR BERNHARD THALHEIM

What Christian Mancas wanted to do is to write the best possible book

on real, pragmatic database design available, bar none. He succeeded.
Christian had most of these tricks and approaches in his head when he
planned this book more than a dozen years ago. When you work on a
new approach, you tend to pick up things that have been developed in the
past by researchers. Mimicking accepted approaches without reconsidera-
tion, reorganization and renovation is a common unculture in computer
science. The area of database and information system design is not an
exception. There are at least three-score books at present that follow the
main agenda and repeat what has been done in the past with some minor
changes in scope and order. However, if you are smart and clever, you col-
lect main results and ideas, work on them, and figure out whether the clas-
sical approach is really the way how topics must be given in a course book.
But what if the classical approach is not the most appropriate one? What if
the classical approach leaves pragmatism completely out of scope? What
if the classical approach is mainly based on syntax? It is very hard for a
person new to a language to walk into a discipline knowing only syntax
and some few integrity constraints and design a complex database schema.
Ever try to get along in a foreign country with only a small dictionary that
gives word-to-word transformations and additionally a small rulebook?
Okay, now we have the purpose of this book. In order to develop data-
base applications one must have a methodological background, a number
of well-developed techniques, and the best possible theory in the back-
ground. The next step is how to write so that everybody can use the book
as a main source for database design and development. It seems that in the
age of the Internet someone has only to look into the ocean of sources in the
Internet and will succeed in this task. The agility of criticism on the Internet
seems to provide the path for improvement of any book. Anybody can pro-
pose new topics that must be covered, can add what has been left out, can
xii Foreword by Professor Bernhard Thalheim

correct typos and real errors, and propose some kind of insight how things
should be developed for real applications.
So, we have now a glimpse into the real need of practical database
development. One must start with a good understanding of the topic and the
application, must use the best language, must be supported by most appro-
priate tools for development, and must be completely sure that the approach
is well founded and will not add errors due to the partial background of
the language. The classic approach uses some extension of the entity-rela-
tionship modeling language for introduction to conceptual database design,
allows to transfer schemata developed in such language to another lan-
guage—say some abstraction of relational database management system
language, and finally derives the description within a system language. This
book follows such order as well. Whether the given extension is a most
appropriate is a matter of culture, education and background. Whether the
transfer or mapping works well is a matter of insight. This book is, how-
ever, based on methods—called algorithms—for database design and on
stewardship by good guidelines—called good practices. I know only four
books in our area that follow such approach. Practical people wrote all four
books during the last almost three decades. And that is it. All four books
provide an insight what should be done for database development. They do,
however, not provide an insight how database design should be backed by
an appropriate language and theoretical underpinning.
Since I have a reputation for expertise in object-relational database
design and development, in database theory, in database programming
and in methodology development for this area, I was one of the people
Christian Mancas asked for a review of the manuscript. And I was sur-
prised and satisfied while discovering stuff and tricks I have not seen
before. While improving the co-design methodology to information sys-
tems development with the goal to reach maturity at SPICE level 3, we
had to realize that database design is an engineering activity far beyond
an art and must be based on a well-founded technology of design. Many
areas of knowledge begin as an art and only later become scientific. The
maturity of such an area can be measured by the existence of well-defined
methods, by a chance for being repeatable, by becoming manageable,
and by logistics on the basis of methods for continuous optimization.
Currently, database design is still an art. An artisan or design mechanic
Foreword by Professor Bernhard Thalheim xiii

is supported by constructs mainly at the syntactic level. So, a database

developer knows templates for structure description and templates for
coherence specification. The second kind of templates is called integrity
constraints and considered to provide semantics. It is, however, only the
starting point for semantics description. Artisans apply myriad rules of
thumb, each rule with only a narrow scope. So, we require maturation of
the area. We need a small number of small powerful methods that can be
applied systematically. Systematic application must be guided.
This book will find its way into the literature on database design and
development. It has a good number of ideas that must be considered in any
design task. It uses a sample-based approach and is thus easy to under-
stand. It supports digestion due to nice exercises. And, finally it discusses
in details also the result of a design in different DBMS languages. So, a
reader can be sure that the book guides to the right track.
Enjoy it! Use it!

Bernhard Thalheim
Professor with the Department of Computer Science,
Christian-Albrechts-University Kiel, Germany
E-mail: [email protected]
October 2014
This page intentionally left blank
FOREWORD BY
PROFESSOR DAN SUCIU

For a computer professional today, data management skills are no lon-

ger just nice to have, but a necessity. Companies collect ever-increasing
amounts of data, and make critical business decisions based on sophisti-
cated analysis of this data. They need experts that understand how to model
and store the data, how to query and analyze it, and how to be advanced
users of modern database management systems. Professionals with strong
data management skills are in high demand today, including not only the
traditional database administrator (DBA), but also newer skills like the
data analyst, the data scientist, the data steward, or the data enthusiast.
The first volume of this book, by Prof. Christian Mancas, covers the
classical data management topics that any computer professional should
master. Based on courses taught by Prof. Mancas over several years at the
University Ovidius in Constanta and the Polytechnic University in Bucha-
rest, Romania, this volume is a gentle yet rigorous and extensive introduc-
tion to the main topics in data management, with concrete examples on
several popular database systems. There are lots of detailed examples, and
each concept is covered in detail, and from several perspectives, using
alternative definitions or notations where needed. The book ensures that
no reader is left behind, and all potential questions are answered. Basic
yet fundamental data management tasks like designing an E/R diagram,
converting it into relations, checking for normal forms, are described in
full detail and with lots of examples. The book covers all major normal
forms, 1st, 2nd, 3rd, BCNF, 4th, and 5th, which one rarely finds described
in an introductory textbook to databases. Special treatment is given to the
relational algebra, which the book defines rigorously, using mathematical
concepts, then describes with the same rigor the connection between the
relational algebra and SQL. The SQL chapter has case studies using sev-
eral major commercial and open source database management systems.
xvi Foreword by Professor Dan Suciu

The first volume of this book is best suited for the practitioner who
wants to achieve a thorough understanding of the fundamental concepts
in data management. The increasingly central role that data, sometimes
called big data, plays in our society today makes it imperative for every
computer professional to be skilled in the topics covered by the book. One
should think of this first volume as an important first step in understanding
the complexities of data today.
Dan Suciu
Professor
University of Washington, Seattle, Washington, USA
September 2014
PREFACE

You can have the vastest horizons from the mountains

that obliterated them until lately.
—Lucian Blaga

There are those who believe that databases (dbs) do not need architec-
ture and design. In the best such case, they consider that it is enough to
build and design your software application, then develop it accordingly
by using a CASE1 ORM2 development environment, like, for example,
Java’s Hibernate, which automatically generates the needed underlying db
scheme too.
If the application’s architecture is correct, then architecture of the cor-
responding db is also correct; however, generally, even the best software
architects who are not also db ones are extremely tempted, for example in
the case of commercial applications, to treat cities, states, and sometimes
even countries as character strings, instead of objects, and/or to semanti-
cally overload classes, which will result in db instance anomalies and/or
semantically overloaded corresponding db tables, both of them prone to
implausible data storing.
Moreover, even if the db architecture is perfect, it is not enough: in
order to guarantee data plausibility, its design should also take into con-
sideration all existing business rules (constraints) in the given universe of
discourse.
Consequently, even when perfectly using such CASE ORM tools, the
generated db scheme should then be refined.
In the worst such case, they simply launch a database management
system (DBMS), like SQLite, MySQL, MS Access, or even IBM DB2,
Oracle Database (Oracle, for short), Sybase Advantage Database Server
(former SQL Server), or MS SQL Server, etc. and start creating tables and
filling them with data.
1
The acronym for Computer Aided Software Engineering.
2
The acronym for Object-Relational Mapping.
xviii Preface

Generally, such persons believe that dbs are only some technological
artifacts and almost completely ignore the old software adage “garbage
in, garbage out”, which is equivalent to, for example, building not only
houses, but even skyscrapers, without architecture and design.
Unfortunately, in dbs there is no law prohibiting such “constructions”,
and the day when mankind will realize that data is, after health, education,
and love, perhaps our most valuable asset (and so it deserves legal regula-
tions too) is not yet foreseeable.
Then, there are those who consider db architecture and design as being
rather an art, or even a “mystery”, but definitely not a science.
This book advocates a dual approach: conceptual data modeling,
strengthened with CASE algorithms, is db architecting, while db design
and querying is applied mathematics, namely the (semi-naïve) algebraic
theory of sets, relations, and functions, plus the first-order predicate logic.
The main four chapters of this book are devoted to four data models
that I’m considering as being cornerstones in this field: the relational, the
entity-relationship (E-R), the logical, and a(n elementary) mathematical
ones.
Two chapters are devoted to implementing optimized relational dbs
(rdbs), by using five of the currently most widely used relational DBMSs
(RDBMSs): IBM DB2, Oracle Database and MySQL, Microsoft SQL
Server and Access.
Essentially, this book emphasizes that using RDBMSs, so, generally,
technology mastering is a must, but much more important are the previous
two steps of db architecture and design: it almost does not matter what pen
(and RDBMSs too are mainly pens…) you write with, it is what you write
that matters the most.
The proposed approach is fully algorithmic: Entity-Relationship (E-R)
diagrams and business rule (restriction) sets are obtained in the concep-
tual data modeling architectural phase by using an assistance algorithm;
they are next algorithmically translated into mathematical schemes, which
are then refined by applying other five algorithms; next, mathematical
schemes are algorithmically translated into relational ones and associated
sets of nonrelational constraints, which, in their turn, are then algorithmi-
cally translated into actual RDBMSs rdbs. Finally, rdbs are optimized with
a couple of other algorithms.
Preface xix

Moreover, this book also includes “shortcut” algorithms, like, for

example, the one for directly translating E-R diagrams and business rule
sets into relational schemas (or even dbs on a particular RDBMS version)
and associated sets of nonrelational constraints, as well as reverse engi-
neering (RE) algorithms, like, for example, the duals of the above men-
tioned forward (FE) ones.
Except for those in the case studies from the Appendices of the sec-
ond volume, all algorithms are written in an easy to understand and com-
mented pseudocode.
Almost no R, S, a, b, X, Y, etc. type (i.e., purely syntactic) examples
are included: except for some exercises from the third chapter of this first
volume that are borrowed from other books, all other exercises, exam-
ples, and case studies provided in this book are everyday life ones, full of
semantics.
Informal and formal aspects are always clearly isolated: informal ones
always precede corresponding formal ones, as (only relevant) math, if any,
is presented in dedicated sections placed at the end of the chapters; needed
math prerequisites are grouped in a dedicated Appendix of the first volume.
No other theoretical or technological prerequisites are needed; how-
ever, operating systems basic notions and (especially object-oriented and
event-driven) computer programming experience, as well as math prereq-
uisite knowledge are very useful for a faster and deeper understanding of
this book.
For founding solid architecture, design, implementation, manipulation,
and optimization, as well as criteria for objective comparisons between
possible alternative solutions, a set of db axioms are provided too.
Exemplified best practice rules are also provided for all levels: those
that are only searching for such practical receipts might skip everything
else from this book, perhaps except for some examples and case studies.
Equivalences between informal, E-R, relational, math, and RDBMS
jargons are also provided. Bibliographic notes are adding optional con-
cepts and results, as well as retracing domain mainstream history. Refer-
ences include both up-to-date and seminal relevant papers and books.
On one hand, given the immense quantity of published work in this
field, it is impossible to exhaustively cite them; on the other, online search-
able bibliographies cover essentially all recent ones. Thus, in this book we
xx Preface

rather mention only the papers of historical importance, major secondary

sources, and useful surveys.
There are some well-known proposed exercises, especially in the third
chapter, but most of them are original. Those marked with a star are more
difficult; those that have two stars are very difficult. For all chapters, at the
beginning of the exercises section, problems of all types are solved.
The subject part of the index is, generally, built dually: for example,
both “function, domain” and “domain, function” are included; moreover,
the index also includes abbreviations and relational and math symbols
(thus serving as a legend too), as well as paper, book, and motto authors.
Mottos without authors come from previous books by this book’s author.
Essentially, I am using the first volume (together with my book on SQL
ANSI-92 Programming) for my undergraduate database lectures, and the
second one (together with my books on Conceptual Data and Knowledge
Modeling and Querying and on Database Management Systems) for the
postgraduate Conceptual Data and Knowledge Modeling and Querying and
Architecture, Design and Development of Database Applications lectures,
at both Ovidius State University Math and Computer Science Dept., Con-
stanta (Romanian and English) and Bucharest Politehnica University Com-
puter Science taught in Foreign Languages Department, English stream.
The first volume describes the shortest advisable path for conceptual
data modeling, db design and implementation, which completely skips
both the intermediate mathematical level (except for algorithm A7/8–3, an
adaptation of A3—see Figs. 1.1 and 1.2 in Section 1.7) and the additional
logic one.
The second volume of this book, scheduled to appear next year, con-
tains advanced topics (on mathematic schemas refinement, recursive SQL,
knowledge bases, nonrelational constraint enforcement and db optimiza-
tions) and is also made up of five chapters:
1. The quest for knowledge: Recursive SQL, Datalog, and knowledge
bases (KBs, also including Datalog and deductive dbs).
2. The quest for rigor: The (Elementary) Mathematical Data Model
(E)MDM—including math db schemas and instances, algorithms
A1 to A7, REA1 and REA2 –see Figs. 1.1 and 1.2 in Section 1.7, (E)
MDM models of the E-RDM, RDM, and itself, E-R and relational
data models of the (E)MDM, the equivalence between E-R and
Preface xxi

functional data modeling, as well as a presentation of the corre-

sponding prototype KDBMS MatBase.
3. The quest for ultimate data plausibility: Non-relational constraints
enforcement (which includes all (E)MDM constraints related to
sets, mappings, homogeneous binary relations, and objects, as well
as the family of algorithms AF9/9′—see Figs. 1.1 and 1.2 in Sec-
tion 1.7).
4. The quest for speed: Optimizing schemas and data manipulation
(including the families of algorithms AF10 and AF11—see Figs.
1.1 and 1.2 in Section 1.7).
5. Conclusion.
Naturally, second volume chapters also include corresponding best
practice rules, math behind, conclusion, past, present, references, and
solved, as well as proposed exercises sections.
Moreover, besides its index, the second volume also includes two
appendices, each one containing a full case study: one for a reigns and the
other one for a geographical db software applications.
There are, however, sections that are only optional supplementary
reading for my students: from the first volume, 2.14, 3.10, 4.4, and 4.12;
from the second one, all ones on the past, present and references.
Overall, this book embeds more than 40 years of db industry and
scientific research, as well as 20 years of university teaching experi-
ence. Besides the author’s original results it contains (e.g., most of the
25 proposed (family of) algorithms, formalizations of the Key Propa-
gation Principle and referential integrity constraints (typed inclusion
dependencies/foreign keys), E-RDM, and RDM, the (E)MDM, their
metamodels (i.e., E-R, relational, and mathematical data models of the
E-RDM, RDM, and (E)MDM), the equivalence between the E-R and
the functional data modeling (which proves that the only fundamental
mathematical relations for data modeling are the functions), relation-
ship hierarchies, restriction sets, and descriptions associated to E-R dia-
grams, the relational Domain-Key-Referential-Integrity Normal Form
(DKRINF), the (E)MDM Semantic NF, the MatBase KDBMS proto-
types, the proof that any table having n columns may have at most C(n,
[n/2]) keys, the orthogonality of keys and foreign keys, the formalization
of the SQL GROUP BY clause with quotient sets computed by the kernel
xxii Preface

of mapping products equivalence relations, case studies, examples, and

exercises, many database axioms and best practice rules, etc.), the main
difference between it and the most of the other similar books is the com-
pletely algorithmic approach it advocates.
The intended audience is diverse: computer science students and pro-
fessors; db architects, designers, developers and, partially, administrators;
and, as it also contains (mainly in the second volume) open problems and
new points of view in this field, db scientific researchers as well.
Obviously, some might wish to study and use only the 25 algorithms
and the almost 100 best practice rules (many of them having several sub-
rules), others only the examples, or only the math behind, or only the
plethora of relevant technological details on the latest versions of the 5
flagship RDBMSs considered, etc.
Although I proofread this first volume several times, I fear that it still
may contain both typo errors and inaccurate details. Thanks in advance
to all those who will contribute to make subsequent editions better and
better!
ACKNOWLEDGMENTS

The wise man does not lay up his own treasures.

The more he gives to others, the more he has for his own.
—Lao Tzu

Acknowledgements are firstly due to those who seeded in me the love

for mathematics and computer science—my wonderful parents, Elena
and Ioan, and professors, chronologically, who were the most influential:
Lucia Ţene, Octavian Stănăşilă, Gheorghe Şabac, Alexandru Dincă, Ana
Burghelea, Luca Dan Şerbănaţi, Pierre (Petre) Dimo, Irina Athanasiu, and
Christian Giumale.
Among the colleagues with whom I had the privilege to learn from and
work with, in the same order, I’m especially indebted to Professors Val
Tannen (Valeriu Breazu), Victor Markowitz (Mănescu), and Dan Suciu;
to my teaching assistants Lecturer Dr. Elena Băutu and PhD student Alina
Iuliana Dicu; as well as to Lavinia Crasovschi, Edward Sava-Segal, Adrian
Lungu, Vladimir Ivanov, Vadim Leandru, and Adrian Raicu.
I will always recall with great pleasure the effervescent atmosphere
and fruitful discussions that I had in the 80s with the outstanding members
of the Miller seminar of the mathematics and computer science faculty of
the Jassy (Iaşi) “A. I. Cuza” University (Professors Sorin Istrail, Gabriel
Ciobanu, Christian Masalagiu, Victor Felea, etc.); moreover, they offered
me too access to the seminar’s extraordinary library, an incredible one in
those dark communist years, when we were almost completely isolated
from the rest of the world.
Thanks to my students from the mathematics and computer science
department of the Ovidius State University at Constanta and from the com-
puter-science-taught-in-foreign-languages department, English stream, at
the Bucharest Politehnica University who inspired my teaching: for con-
vincing them of the beauty of conceptual data modeling and querying, I
always had to find better and better algorithms, examples, and exercises.
xxiv Acknowledgments

Thanks to God for the hundreds of customers’ projects on which I worked

and which provided me with a wealth of such interesting examples.
Thanks for their patience and kindness to Sandra Sickels and Ashish
Kumar at Apple Academic Press.
Last only chronologically, but first in my mind and heart, special thanks
and all my gratitude to Professors Dan Suciu and Bernhard Thalheim, who
had the kindness and patience to be the first readers of this first volume and
to write the forewords that honor it, as well as to all of its reviewers, and
especially to Professors Cristian Calude and Andra Hugo.

Bucharest, Romania, October 1, 2014

LIST OF SYMBOLS

 universal quantifier (for any)

 existential quantifier (there exists)
 (element) belongs to (set)
 (element) does not belong to (set)
 sum operator symbol
 function product operator
! there exists a unique (element of a set)
& string concatenation operator
[x] integer part (of x)
\ difference set operator
 difference (set, numbers) operator
|P| arity (of P)
π projection (RA) operator
ρ renaming (RA) operator
σ selection (RA) operator
 (inner)join (RA) operator
L
 left (outer) join (RA) operator
 right (outer) join (RA) operator
L
 R
full (outer) join (RA) operator
|> left anti-semijoin (RA) operator
<| right anti-semijoin (RA) operator
|>< left semi-join (RA) operator
><| right semi-join (RA) operator
÷ division (RA) operator
, SQL set Cartesian product operator
 (math) set Cartesian product operator
ε, λ empty string
 logic implication (between sets of formulas)
° function composition operator
• function (Cartesian) product operator
xxvi List of Symbols

 direct sum set operator

Ø empty set
 intersection set operator
 union set operator
 (set) inclusion relation
 (set) strict inclusion relation
 (set) not inclusion relation
 (logical) not operator
 (logical) and operator
 (logical) or operator
 (logical) implication operator
 not equal
<> not equal
x! factorial of x
ASCII(n) the set of all strings of ASCII characters of length at most n
C(n, j) n choose j (combinations of n taken j)
card the (math) cardinal function
COUNT the (SQL aggregation) cardinal function
f –1 (mapping) f’s inverse
iff if and only if
Im function image
INT(n) the set of all integers having at most n digits
ker (mapping) kernel operator
S/~ the quotient set of S with respect to the equivalence relation ~
max the maximum function
NAT(n) the set of all naturals having at most n digits
RAT(n,m) the set of all rationals having at most n digits before the
decimal dot (comma, in Europe) and m after it
UNICODE(n) the set of all strings of UNICODE characters of length at
most n
CHAPTER 1

DATA, INFORMATION AND

KNOWLEDGE IN THE COMPUTER ERA

Data is not information, Information is not knowledge,

Knowledge is not understanding, Understanding is not wisdom.
—Clifford Stoll

CONTENTS

1.1  Data, Information and Knowledge.................................................... 2

1.2  Data Analysis and Conceptual Modeling.......................................... 6
1.3  Data and Knowledge Bases.............................................................. 9
1.4  Constraints (Business Rules)...........................................................11
1.5 Data and Knowledge Base Management Systems
(DBMS, KBMS)............................................................................. 15
1.6  Static and Dynamic Aspects of Databases...................................... 17
1.7  Conclusion and Book Overview..................................................... 18
1.8  Past and Present.............................................................................. 25
Keywords................................................................................................. 29
References................................................................................................ 31

After studying this chapter you should be able to:

• Concisely define informally the following key terms: data (analy-
sis, model, aggregation, generalization, factorization, plausibility,
positive, negative, incomplete, semistructured, big, homogeneous,
heterogeneous), metadata, information, knowledge, assumption
2 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

(closed world, closed domain, unique objects), object set (prop-

erty), database (fundamental, derived, scheme, instance—both
consistent and inconsistent), inference rule, knowledge base, busi-
ness rule (constraint) (fundamental, implied, trivial, redundant, sat-
isfaction, violation, plausible, implausible), constraint set (closure,
equivalence, minimal, redundant, coherent, incoherent), implica-
tion problem, conceptual data modeling, data models (Relational
(RDM), Entity-Relationship (E-R) (E-RDM), elementary mathe-
matical ((E)MDM), logical (LDM)), database management system
(DBMS), knowledge-based management system (KBMS), Struc-
tured Query Language (SQL) (stored methods, Data Definition
Language (DDL), Data Manipulation Language (DML)), Datalog,
NoSQL systems.
• Give examples of both commercial and open source DBMSs and
KBMSs of all existing types.
• Outline the steps involved in data analysis, conceptual data mod-
eling, database design, implementation, and optimization, as well
as the proposed corresponding forward and reverse engineering
(families of) algorithms.

1.1  DATA, INFORMATION AND KNOWLEDGE

The goal is to turn data into information, and information into insight.
—Carly Fiorina

Data comes from the Latin datum, which means both a fact (known from
direct observation) and a premise (from which conclusions are drawn).
Information also comes from Latin: informatio (meaning formation, con-
ception, education) is the participle stem of informare (from where “to
inform” descends).
Generally, in Information Technology (IT), we mean by data facts’
(objects, concepts, etc.) property descriptions (through some values, gen-
erally of type character strings, numbers, calendar dates, etc.) which are
worth being noticed and (electronically) stored for later retrieving and
processing, while by information—increments of knowledge that can be
derived from data.
Data,
 Information and Knowledge in the Computer Era 3

For example, all cells in Table 1.1 contain data: those in the first row
(of type character string) store the names of the corresponding columns
(the so-called table header or scheme), which in fact, together with the
table name, is metadata—that is, data on data—, while all of the others
(the table instance or content) contain data on some Carpathians peaks.
Obviously, for example, line two of this table embeds the informa-
tion that in the (Carpathians) “Făgăraşi” mountains there is a peak called
“Moldoveanu” whose altitude (in meters) is 2544. Generally, each (data,
not metadata) line of a table embeds the information that there is (in the
corresponding set of objects) an element (object, item, etc.) having cor-
responding values for the corresponding properties.
But more information, of higher degrees, may also be derived from this
table, for example:
1. There are 3 mountains and 5 peaks for which data is stored, out of
which 2 from the “Făgăraşi” and “Retezat”, and 1 from the “Piatra
Craiului” mountains.
2. “La Om (Piscul Baciului)” is less high than “Retezat”, which is
less high than “Peleaga”, which is less high than “Negoiu”, which
is less high than “Moldoveanu”; “Moldoveanu” is the highest,
while “La Om (Piscul Baciului)” is the lowest.
3. Only 3 peaks are at least 2500 m, with two from “Făgăraşi” and
one from “Retezat”.
4. Overall average altitude is 2461.6 m; for “Făgăraşi” it is 2539.5 m,
for “Retezat” it is 2495.5 m, and for “Piatra Craiului” it is 2238 m.
5. There is only one peak (“Retezat”) that gives its name to its mountain.
Traditionally, data storage uses some communication means (e.g.,
images, languages, sculptures, shrines, computers, etc.) and media
(lime stone, granite, papyrus, paper, HDD/CD/DVD/SSD, etc.). Getting
TABLE 1.0  Carpathians Peaks
4 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

information generally needs retrieving and processing data (e.g., arith-

metically, statistically, logically, etc.).
Note that such stored data is always positive (i.e., we always store the
fact that somebody has a child, and never that somebody does not have
children) and that the following three implicit assumptions are governing
both storage and processing:
a. closed world: facts that are not known to be true (i.e., for which
there is no stored data) are implicitly considered as false;
b. closed domains: there are no other objects than those for which
there is stored data;
c. unique objects: objects are uniquely identifiable by their corre-
sponding stored data.
Moreover, conventional IT information is never obtained through
recursion.
Our minds, our books, our libraries have finite capabilities of both
storing data and deriving information. Our computers, even if some of
them have today more storage capacity and faster processing speeds than
humans, are and will always be in the foreseeable future finite too.
Most of the object sets we are currently dealing with (e.g., people,
organizations, services, products, etc.) are, fortunately, finite; even when
they might be infinite (as, for example, the number of universe’s particles)
most of them are countable and we know that we can only store and pro-
cess finite subsets of them.
We know how to define and characterize infinite not countable sets too
(like, for example the rational, real, and complex numbers), but when it
comes to manipulating we have to digitize them. Probably for all practical
purposes, from measuring it to real-time management systems (be them
nuclear power plants or missiles batteries), even Time is digitized.
Most of the real and complex mappings (as, for example, sine, cosine,
tangent, arc cotangent, exponential, logarithm, fractals, etc.) are, first of
all, only ideally encountered (e.g., when no friction, no gravity, no obsta-
cles, etc. exist) and, secondly, are, practically, only approximately comput-
able, with, at most, a maximum number of digits both before and after the
digital dot, using discrete computations.
Fortunately, just like for sets, the vast majority of currently encoun-
tered mappings are finite too. Moreover, most of the fundamental ones
Data,
 Information and Knowledge in the Computer Era 5

(e.g., names, birth dates, e-mail addresses, population, surface, latitude

and longitude, altitude, minimum and maximum temperatures, etc.) do not
have any other computation “law” than their graphs (i.e., the correspond-
ing set of pairs <x, f(x)>).
Consequently, we may use any finite data structure for storing data: (inde-
pendent) variables, records, lists (vectors), stacks, matrices, trees, acyclic or
cyclic graphs, etc. Not only for their extremely powerful intuitive appeal or
algebraic naturalness, but especially as manipulating data (both for adding,
modifying, removing, and querying it for deriving information) is, by far, as
we will see, the simplest possible for tables (which are matrices extended
with headers for storing metadata too), data is since decades and will con-
tinue to be in the foreseeable future stored in and processed from tables, even
if, for example, graphs have lately reemerged as a very powerful alternative.
Knowledge (coming from old English) generally means facts, informa-
tion and skills acquired through experience and/or education, as well as
theoretical and/or practical understanding of a subject. In IT, knowledge
includes positive data and information, but the latter is also obtainable from
data through recursion, by applying inference rules (e.g., anybody who is
a child of one of somebody’s descendants is that somebody’s descendant
too, for any somebody); moreover, data may be also stored as first-order
logic predicates, which allow for negative (e.g., “Iosif Visarionovici loved
nobody”) and incomplete data too.
For example, incomplete data is used for asserting that:
• something has a given property even if no values are available for
it (e.g., “John Adam has a child”);
• all elements of a set, which have not to be enumerated, have a cer-
tain property (e.g., “all children have parents”);
• something has either one or another value for a property, without
having to be sure which one (e.g., “Mary Eve is 29 or 30”);
• two distinct expressions (e.g., “Mary Adam” and “Adam’s wife”)
are referring or not to a same object.
Generally, positive data is homogeneous (i.e., for all objects of inter-
est, same types of data are desired, like, for example, in Table 1.1), while
negative and incomplete one is not. Both theory and experience proved
that, for obtaining best possible querying speed and storage size, positive,
homogeneous data should be stored in tables, leaving to logic predicates
6 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

only negative and incomplete data, as well as inference rules. When not
otherwise stated, this book deals with homogeneous data.

1.2  DATA ANALYSIS AND CONCEPTUAL MODELING

Given a large mass of data, we can by judicious selection

construct perfectly plausible unassailable theories—
all of which, some of which, or none of which may be right.
—Paul Arnold Srere

As usual, we are not happy only seeing trees (data): we want to see espe-
cially forests (information). As such, some data interpretation is always
needed that should be enough abstract and powerful in order to understand
not only data values, but especially data interconnections.
Generally, models are such intellectual tools, intensively used in sci-
ence for abstracting details and outlining structure; from mathematical and
physical ones (e.g., Big Bang, Big Crunch, relativity, quantum, strings,
unification theories, etc.), economics (e.g., Nash equilibrium), geography
and meteorology (e.g., maps), and up to everyday life (e.g., phone books,
yellow pages, time schedules and tables, etc.) models allow us to at least
partially derive and understand information.
Data models existed long before computers: from maps to phone
books and all color pages, from Alumni to cook books, and from ships/
trains/planes/buses time tables to accounting worksheets mankind stored
data according to several data models since at least the ancient Egyptians,
Babylonians, Mayans, and Chinese. We will, however, focus here only on
computer manageable ones.
Such models are structuring elementary data items on which a pre-
defined set of operations is allowed. Elementary data items are of type
<object set, object unique id, object properties, properties’ values, time-
stamp>, as objects descriptions are finite and characterized by sets of
aggregated properties’ values that may vary in time. For example, MOUN-
TAIN_PEAKS objects from Table 1.1 are aggregated from the Mountain,
Peak, and Altitude properties. PEOPLE might be aggregated from SSN,
FirstName, LastName, Sex, BirthDate, BirthPlace, and e-mail.
Data,
 Information and Knowledge in the Computer Era 7

As in most of the cases we are only interested in current values or, at

most, in the history of few some of them (e.g., hire/marriage/divorce/start/
end dates), generally timestamps are dropped or replaced by calendar date/
time values.
The sets of allowed operations generally include create, alter, and
drop—for object set collections, objects, users, and user group sets, etc.,
select, insert, update, and delete—for elementary data items aggregated in
instances of above collections and sets, grant and revoke rights on object
collections and sets to/from users and groups, etc. 12

Besides properties aggregation,3 data structuring is obtained by hierar-

chies of categorizations/classifications/generalizations4 and factorization
(of common properties, on appropriate levels). For example, we might
characterize persons as the aggregation of SSN5, first and last name, birth
date, birth place, sex, and e-mail address; some catalog items may be clas-
sified as water, wine, beer, cheese, vegetables, women cosmetics, men
cosmetics, etc.; then, the first two above may be categorized as beverages,
the following two as food, and the last two as cosmetics; beverages and
food may then be generalized as aliments, aliments and cosmetics may
be generalized as products, etc.; corresponding classes, categories, types,
families, etc. names and properties should be factorized on their appropri-
ate levels of this hierarchy.
This is why it is no wonder that the first data model was the hierarchi-
cal one, especially as—from states, armies, and organizations to biology,
botany, and zoology—we are living in a world of hierarchies.
Then, hierarchies (trees) were extended to the more general graphs in a
network data model, where nodes are elementary data items and edges are
connections (links, relationships, etc.) between them.
The first mathematical data model was the relational (RDM) one, pre-
sented in the third chapter of this first volume: initially, although several
equivalent interpretations are possible (and we will provide one more in
Section 3.12), it was introduced as sets of representatives of mathemati-
cal relations (defined over data domains/types) equivalence classes (where
31
Mathematically, aggregations are function (Cartesian) products.
42
Mathematically, such hierarchies
aggregations are are function
function compositions;
(Cartesian) in Artificial Intelligence (AI), categori-
products.
3zations/classifications/generalizations are IS-As.
Mathematically, aggregations are function (Cartesian) products.
5
The acronym for Social Security Number.
8 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

each such class contains all possible permutations of underlying sets of a

relation), interconnected by common data domains. Ultimately, although,
again, several equivalent representations are possible, the one consisting
of sets of tables (each made out of a collection of columns) interconnected
by foreign keys prevailed, as it is the most natural and simple one for rep-
resenting finite data.
Essentially, RDM is purely syntactic: for example, theoretically, no
table should correspond to any actual set of objects. This is why designing
RDM schemas proved almost intractable: which column should belong to
which table and why? Which tables should we add and why? This is why,
hundreds of semantic data models aroused in these last four decades and
this will continue to happen.
One of the most still used is graphical, so very powerful if correctly
used: the entity-relationship data model (E-RDM), presented in the next
chapter, uses rectangles for atomic object sets, diamonds for mappings
and relations between them, and ellipses for their base properties; due to
their simplicity, its E-R diagrams (E-RDs) can be best understood by our
customers too.
However, E-RDM is fine only for depicting the structure of data
(which object sets are of interest, what are their properties and intercon-
nections between them), but not also for embedding all of the business
rules that are governing any given subuniverse of discourse. This is why
hundreds of higher level (semantic) more sophisticated data models were
proposed too.
For example, this book proposes the (elementary) mathematical one
((E)MDM), presented in the second chapter of its second volume, for
refining the conceptual data modeling such as to guarantee not only the
best suited structure, but also the highest possible level of data plausibility.
In (E)MDM, for example, one can also express constraints like “any
capital of a country must be a city of that country” and “any hotel room
may be reserved for any day by only one person”, which are not express-
ible either in E-RDM or in RDM.
Before using any data model, the first and one of the most important
steps is data analysis: understand exactly what data is needed in order to
correctly model a given subuniverse of discourse (e.g., an e-commerce, a
bank, a supermarket, etc.) and what are the business rules that govern it.
Data,
 Information and Knowledge in the Computer Era 9

The next chapter is devoted to both data analysis and its deliverables: an
exhaustive, concise and clear informal description of the needed data and
existing business rules, as well as a corresponding set of E-RDs.
The shortest advisable path for db design and implementation needs
then an algorithm to translate E-RDs and associated business rules into rdb
schemas and (possibly empty) sets of associated nonrelational constraints
(i.e., business rules that are not expressible in the RDM framework and,
consequently, need to be enforced by software applications built on top the
corresponding database).
However, the best approach to conceptual data analysis and modeling,
as well as database design and implementation is to first translate E-RDs
and associated business rules into a higher level semantic data model
scheme, refine it as much as possible, and only afterwards translating this
refined scheme into RDM schemas and (possibly empty) sets of associated
nonrelational constraints.
Finally, RDM schemas can be implemented as databases (through
corresponding translation algorithms presented in the fourth chapter), by
using any desired (or imposed) relational database management system
(RDBMS) version.
Knowledge is stored both using RDM and logic (for negative and
incomplete data) and inferred by logical data models (presented in the
first chapter of the second volume), based on first order logic.

1.3  DATA AND KNOWLEDGE BASES

It is a capital mistake to theorize before one has data.

Insensibly one begins to twist facts to suit theories,
instead of theories to suit facts.
—Sir Arthur Conan Doyle

Any data collection structured according to a data model is called a data-

base (db). This is why, on one hand, there are hierarchical, network, rela-
tional (rdb), object-oriented (O-O), etc. dbs and, on the other, not any data
collection (e.g., not even a “relational database” storing aberrant data) is
10 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

a db. As such, dbs are models representing subuniverses of interest to at

least some of us.
For example, a geographic rdb could be made of the interrelated
tables CONTINENTS, ARCHIPELS, ISLANDS, MOUNTAIN_RANGES,
MOUNTAINS, PEAKS, RIVERS, LAKES, SEAS, OCEANS, COUNTRIES,
STATES, CITIES, etc.
Databases are three-dimensional objects: their schemes store the struc-
ture and the associated constraints (i.e., their metadata), their instances
store actual data, whereas their SQL stored programs (views, triggers,
functions, procedures, etc.) are modifying and/or querying instances.
For example, such a stored view could compute, in alphabetic order,
the set of triples <country name, state name, city name>; a stored trigger
could automatically add to STATES an “UnknownState” whenever a new
country is added to COUNTRIES; a stored procedure might compute the
number of states, the corresponding number of cities, as well as their total
number for any country given to it as a parameter.
Beware of notational abuses, as, for example, instances are often
referred to also as “databases”.
Modern knowledge bases (kb) extend databases with inference rules
(IR), adding to them the power of recursiveness and negation.
For example, a genealogic rdb, containing a PEOPLE table also includ-
ing columns Mother and Father, could be extended with Datalog6 pro-
grams for computing the set of all genealogic ancestor and/or descendent
relationships7, or all (male and/or female) ancestors and/or descendants of
a given person (see Exercise II.1.1 in the second volume).
Note that conventional dbs do not embed inference rules, so can only
infer information, not knowledge.
Also note that there are fundamental dbs (which store data which
cannot be retrieved from other dbs) and derived dbs (like, for example,
the Online Analytical Processing (OLAP) and/or Business Intelligence
(BI) ones, which store only data that can be obtained from other dbs).
Except when otherwise explicitly stated, this book deals only with fun-
damental ones.
6
Datalog (see the first chapter of the second volume) is a happy marriage between rdbs, the program-
ming language based on first order logic Prolog, and a much more efficient computational semantics
than Prolog’s one.
7
The so-called transitive closure of Mother and/or Father.
Data,
 Information and Knowledge in the Computer Era 11

1.4  CONSTRAINTS (BUSINESS RULES)

Mathematics may be compared to a mill of exquisite workmanship,

which grinds you stuff of any degree of fineness; but, nevertheless, what
you get out depends upon what you put in; and as the grandest mill in the
world will not extract wheat-flour from peasecods, so pages of formulae
will not get a definite result out of loose data.
—Thomas Henry Huxley

Besides interrelated tables simplicity and naturalness, besides the

extremely powerful associated querying formalism also introduced, RDM
had the major contribution of introducing and formalizing constraints
(business rules) too: logic formulas that are added to dbs schemas and
enforced by RDBMSs in order to reject attempts to store undesired data in
their instances.
For example, if we leave Altitude from Table 1.1 above to get inte-
ger values (as most computer programmers, but also too many database
designers are still doing), end-users might enter peaks having either nega-
tive, less than 1000, or greater than 8850 ones too.
Some are still speaking about data correctness in this respect, although
not only me is inclined to not ever use this syntagm: even today, for exam-
ple, there is still debate even on mountain peaks’ altitudes, as it depends
on measurement instruments, meteorological conditions, Earth tectonic
plates’ movements, etc. Moreover, even if we, humans, were in the posses-
sion of only relatively correct altitude values for all peaks, how could an
automaton (e.g., a computer program) reject human data entering mistakes
(e.g., typo ones or swapping between altitudes of two peaks)?
But even if we do not know (or agree on the fact) whether, for example,
peak Moldoveanu has 2543 m or 2545 m (as some newer measurements
reportedly suggest), so we cannot instruct our software to accept or reject any
such value, we can at least instruct it that only values between 1000 (by geo-
graphical convention, the minimum altitude in meters for a peak in order to
consider it a mountain one) and 8850 (Everest’s altitude) are acceptable, as,
otherwise, they are not plausible. Of course that, for enhancing plausibility,
we could even store maximum elevations per mountains and instruct soft-
ware to dynamically compute maximum thresholds mountain per mountain.
12 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

In order to always guarantee data plausibility, data model schemas

should also include all needed constraint types. For example, the above
“peaks should have altitudes between 1000 and 8850” is formalized in
RDM as the “domain” constraint “Altitude between [1000, 8850]”.
When enforced, constraints are always satisfied by (or hold in) any
corresponding db instance, as any attempt to violate them is rejected. For
example, above constraint is satisfied by Table 1.1’s instance. When not
enforced, some instances may violate (or do not satisfy) them. For exam-
ple, if we replace in Table 1.1 any altitude with 900 or 9000 (or any other
natural less than 1000 or greater than 8850), then above constraint would
be violated.
An instance is said to be consistent (valid) if it satisfies all (table/db)
constraints; otherwise, it is called inconsistent (invalid).
Note that constraints exist independently of whether we know and/
or ignore them and that only adding all existing ones in our data models
may guarantee data plausibility: “garbage in, garbage out” applies in this
context too if and only if there is at least one missing constraint from our
corresponding model.

1.4.1  IMPLIED CONSTRAINTS AND CONSTRAINT SET

CLOSURES

You can have data without information,

but you cannot have information without data.
—Daniel Keys Moran

As constraints are first order logic formulas,8 one of them or a set of con-
straints may imply other constraints as well: a set of constraints C implies
a constraint c if c holds in all instances in which C holds (dually, C does
not imply a constraint c if there is an instance for which c does not hold,
but C holds).
The standard logical notation for formulas implication is C |— c; c
is called an implied constraint. For example, in Table 1.1, constraint set
8
In particular, closed ones, that is, formulas whose variable occurrences are bound to at least one logic
quantifier (be it “for any” or “there is”). Dually, open formulas have at least one occurrence of a vari-
able free (i.e., not bound to any logic quantifier), which, in dbs, are formalizing queries.
Data,
 Information and Knowledge in the Computer Era 13

{Altitude > 1000, Altitude < 2550} implies constraint Altitude [ (1000,
2550); trivially, the vice-versa is also true. Constraints that are not implied
are called fundamental.
We should never enforce implied constraints in db schemas: it would
only be superfluously time consuming.
Constraint implication is not important only for minimizing constraint
sets, but also for defining constraint sets closures that are needed for defin-
ing many important db theoretical notions, including the highest RDM
normal form: given any constraint class CC, its subset containing all con-
straints implied by any of its subsets C is called the closure of C (with
respect to that class) and is denoted by (C)CC+ (or, if CC is understood from
the context, simply C+). For example, if C = {S # T, T # U}, its transitive
closure is C+ = {S # T, T # U, S # U}.
Closures of a set with respect to different classes are generally different
(e.g., the reflexive closure of the above C is C+ = {S # T, T # U, S # S, T
# T, U # U}).
Note that any set is included in its closure and that the closure of a
closure is itself (i.e., for any C and CC, (C+)+ = C+). Moreover, the closure
of a subset is a subset of the closure of the corresponding superset (i.e., for
any C, D, and CC, C # D  C+ # D+), whereas the union of two closures
is distinct of the closure of the corresponding union (i.e., for any C, D, and
CC, C+ ø D+ ≠ (C ø D)+).
A set which is equal to its closure is called closed with respect to impli-
cation (e.g., C’ = {S # T, T # U, S # U} is transitively closed).
Two sets having same closures are called equivalent and are said to cover
each other (e.g., C and C’ above are equivalent or covering each other).

1.4.2  REDUNDANT CONSTRAINTS AND MINIMAL

CONSTRAINT SETS

There is an enormous redundancy in every well-written book.

With a well-written book I only read the right-hand page
and allow my mind to work on the left-hand page.
With a poorly written book I read every word.
—Marshall McLuhan
14 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

A constraint is said to be redundant in a set when the set’s closure remains

the same after removing it (e.g., S # U is redundant in C’ above).
Any cover not containing redundant constraints is said to be minimal
(e.g., C above is minimal under both transitivity and reflexivity).

1.4.3  THE IMPLICATION PROBLEM

To think out in every implication the ethic of love for all creation…
this is the difficult task which confronts our age.
—Albert Schweitzer

Testing for constraint sets equivalence and/or constraints redundancy

can be reduced to the well-known implication problem: given any con-
straint set C and constraint c, does C imply c? (e.g., does C above imply
S # U? does it imply T # U too?)
Note that, depending on the constraint classes, this problem may be
solved very quickly (i.e., linearly), very slowly (i.e., exponentially), or be
impossible to solve (even undecidable9).

1.4.4  TRIVIAL CONSTRAINTS

Science is an organized pursuit of triviality.

Art is a casual pursuit of significance.
Let’s keep it in perspective.
—Vera Nazarian

A very particular case of implied constraints are obtained when C is empty:

a constraint implied by the empty set is said to be trivial, as it always holds
in any db instance.
For example, (A = B ~ A ≠ B) is trivial for any sets A and B.

9
A problem is said to be undecidable if it is neither provable, nor refutable or for which it is impos-
sible to design an algorithm that always (i.e., in any context) correctly answers to its questions (see
Appendix).
Data,
 Information and Knowledge in the Computer Era 15

1.4.5  PLAUSIBLE AND IMPLAUSIBLE CONSTRAINTS:

COHERENT AND INCOHERENT CONSTRAINT SETS

Civilization is a progress from an indefinite, incoherent homogeneity

toward a definite, coherent heterogeneity.
—Herbert Spencer

Given any db and its associated set of constraints C, C is coherent (with

respect to the db scheme) if for any table of the db there is at least a non-
void instance that satisfies C and incoherent otherwise.
For example, if a db scheme contains the above MOUNTAIN_PEAKS
table and constraints C1: Altitude ≥ 1000 and C2: Altitude < 1000 (or,
equivalently, constraint C: Altitude ≥ 1000 ` Altitude < 1000), then the
only possible instance for MOUNTAIN_PEAKS is obviously the empty
set, which means that the corresponding set of constraints C is incoherent.
Trivially, from any incoherent set a coherent one may be obtained
by removing any pair of contradicting (sub)constraints (C2 in the above
example).
Note that, depending on the context, constraints too may be plausible
(i.e., corresponding to an existing business rule governing that context) or
implausible (dictatorial, aberrant: that is not corresponding to an existing
business rule governing that context): for example, C1 above is plausible
in any mountains context, while C2 is not. Obviously, we must always
include in our data models only plausible constraints.

1.5  DATA AND KNOWLEDGE BASE MANAGEMENT SYSTEMS

(DBMS, KBMS)

The only source of knowledge is experience.

—Albert Einstein

Any software tool providing db schemas definition facilities (i.e., a data

definition language (DDL), including create, add, alter, drop, grant, and
revoke statements), querying and processing instances facilities (i.e., a
data manipulation language (DML), including select, insert, update, and
delete statements), as well as other needed facilities (e.g., at least, con-
16 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

straint enforcement (CE), data access control (DAC), concurrency control

(CC), and data and programs backup and restore (DPBR)) is said to be a
database management system (DBMS).
For example, IBM DB2, Oracle Database (Oracle, for short) and
MySQL, Sybase Advantage Database Server, Microsoft SQL Server and
Access, Tandem NonStop SQL Server, Apache Hadoop10 (actually, its
DBMSs components Cassandra, HBase, and Hive), etc. are all DBMSes.
All of them provide a variant of the ANSI11 standard relational language
SQL12, which include both a DDL and a DML. The relational DBMS
(RDBMS) cores (i.e., SQL parsers, compilers/interpreters, optimizers,
evaluators, etc.) are called SQL engines.
Note that binary, assembly, and high level programming languages
(PLs) are manipulating data as bits, bytes, words, records, record sets,
and/or files (stored and managed by operating systems (OS) file systems);
hierarchical, network, and relational DBMSs are storing and manipu-
lating data values; O-ODBMSs are storing and manipulating objects,
while knowledge base management systems (KBMS, for example: LDL,
Coral++, etc.) are storing and manipulating knowledge. There are also
hybrid products, like, for example, MatBase, based on both Datalog, (E)
MDM, E-RDM and RDM that is storing and manipulating both objects
and knowledge (which is why they are called knowledge and DBMS,
KDBMS for short).
Please note that, just like for OSs, PLs, or any other system software,
(K)DBMSs are just tools, let’s say pens, and that it is important to know
how to use them the best possible, but do not ever forget that it does not
matter with what pen you write: what matters is what you write.
Moreover, for all tools, including the above mentioned ones, keep in
mind that, during your professional lifetime, they are continuously evolv-
ing, sometimes dying, others appearing and that, generally, you will have
to switch several times to other positions, asking from you to work with
other tools than those you were already acquainted with: consequently, the
10
Open source framework developed in Java that stores, manages and processes gigantic amounts of
data in a highly parallel manner, on clusters of commodity machines, and supporting large scale data
analysis by allowing question decomposition into discrete chunks that can be executed independently
very close to slices of the data in question and, ultimately, reassembled into an answer to the posed
question; used by Facebook too.
11
The acronym for American National Standards Institute.
12
The acronym for Structured Query Language.
Data,
 Information and Knowledge in the Computer Era 17

most important things are to master conceptual data analysis, db design,

implementation and optimization, as well as querying and manipulations
principles, algorithms and best practice rules, and mastering technologies
comes only last (even if this is a must too).
This is why, although necessarily having to exemplify with some of
the most currently used RDBMS versions, this book is not about IBM
DB2 or Oracle or MySQL or MS SQL Server or Access or whatever other
such tool.
Also note that using a DBMS does not guarantee at all that your out-
come is a database, just like using wood and appropriate carpenter tools
does not guarantee at all that your outcome is, at least, furniture. Without
proper data analysis, conceptual and physical db design, implementation
optimization, one cannot equal in this field a wood artist (be it sculptor
or cabinetmaker) and not even a good furniture manufacturer, but simply
remains a woodcutter.

1.6  STATIC AND DYNAMIC ASPECTS OF DATABASES

Hiding within those mounds of data is knowledge

that could change the life of a patient, or change the world.
—Atul Butte

Let us consider stores also selling CDs/DVDs/Blu-rays and potential buy-

ers trying to get answers (either directly, from a db application provided
on public available computers, or indirectly, through store employees) to
their following possible questions:
• “What recordings of Dinu Lipatti do you have?”
• “Who else is also costarring in these recordings?”
• “How much do they cost?”
• “What are the opuses contained by each of them?”
• “What are their corresponding recording houses?”
• “What are their recording quality ratings?”
This typical example involves two subtopics:
• a static one: what data should be stored and how, such that
answers to these queries (and any other similar ones) may be
provided correctly and promptly (keeping in mind too that stocks
18 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

are varying in real time)? (i.e., what should be the scheme of the
needed db and how should it be implemented on a DBMS? What
columns and constraints in what tables?) For example, Table 1.1
should be split into two tables, one for mountains and the other
for peaks, and all existing constraints should be added to both of
them.
• a dynamic one: how should the corresponding db be queried for
getting correct and prompt answers? (i.e., what should be the archi-
tecture, design, and development of the needed application built on
top of the above db?)
Generally, dbs have four dimensions: design, implementation, optimi-
zation, and usage (i.e., programming for querying and modifying data).
The design, implementation and optimization are static, while usage is
dynamic.
Note that, generally, it is true that daily db tasks involve in average
some 90% usage, 3% design and implementation, and 7% optimization;
consequently, at a first glance, it might seem that mastering only SQL’s
DML is almost the only key to success in the db field.
However, both design, implementation and optimization are crucial
and can make the huge difference between a very poor and cumbersome
set of interrelated tables and a true database: who would not love to deco-
rate and live in a beautifully and intelligently architectured, designed, and
built spacious home full of light and splendid views around, but prefer
instead a dark underground labyrinth shared with some minotaur?

1.7  CONCLUSION AND BOOK OVERVIEW

There are two modes of acquiring knowledge,

namely, by reasoning and experience.
Reasoning draws a conclusion and makes us grant the conclusion,
but does not make the conclusion certain, nor does it remove doubt
so that the mind may rest on the intuition of truth
unless the mind discovers it by the path of experience.
—Roger Bacon
Data,
 Information and Knowledge in the Computer Era 19

Data was stored since the dawn of humanity on various media in order
to infer from it information and knowledge. Most of the data is discrete,
which makes tables and graphs ideal candidates to store it, with tables
having great advantages on both storing and querying, which are still the
simplest and fastest.
Especially in the computer era, where the amount of stored data is huge
(think, for example, of the astronomical one) and there may be several
thousands of people simultaneously updating it, but especially as crucial
global and local decisions, be them in politics, intelligence, military, sci-
ence, health, commerce, etc., are more and more heavily based on data,
guaranteeing at least its plausibility is of paramount importance.
Consequently, on one hand, more and more sophisticated data models
were introduced, even if, commercially, from the implementation point of
view, the vast majority of the existing database management systems are
relational. This means that only part of the existing business rules may be
enforced through them and for the rest we have to develop software appli-
cations built on top of the corresponding databases.
On the other hand, just as starting to directly write Java (or C# or
whatever other PL) code, immediately after getting some desired soft-
ware application requirements, without proper problem analysis, software
architecture and design is a huge mistake, starting to directly create tables
in a database on a RDBMS is an even greater one: databases are the foun-
dations of db software applications.
Just like no building resists earthquakes, tsunamis or even significant
winds or rains if its foundations are not solid enough, no db software
application may be easily designed, developed, maintained, and used if its
underlying db was not properly designed and implemented.
As such, whenever a new database is needed or an existing one has to
be expanded, first of all, besides establishing exactly what are the corre-
sponding data needs, data analysis has to also discover all business rules
that govern the corresponding subuniverse.
Then, these findings should be structured according to a data model
that, even if it is not as powerful as needed in order to guarantee data plau-
sibility, is comprehensible by the customers too. We will see in the next
chapter that E-R data models (i.e., E-RDs accompanied by restriction sets
20 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

and informal descriptions of the corresponding subuniverse) are an excel-

lent candidate for this first step.
Chapter 3 of this first volume provides enough evidence that RDM is
indeed excellently suited for elegantly and efficiently data both storing and
manipulating.
Finally, Chapter 4 hopefully proves not only that for any RDBMS ver-
sion there is an algorithm that translates any relational scheme into a cor-
responding db, but also that it is worthwhile first thoroughly analyzing the
corresponding subuniverse, obtaining its E-RDs and associated restriction
set, translating it into a relational scheme, and only then translating this
scheme into a db.
The second volume first tackles recursive queries, a logic data model
for also getting knowledge from data, not only information, and the high-
est relational normal forms that we are proposing; then, it introduces an
elementary mathematical data model, for refining as much as possible
E-RDs and restriction sets before translating them into relational schemas.
Its final two chapters deal with enforcing nonrelational constraints and
optimizations of dbs, respectively.
Obviously, forward engineering algorithms should be applied when-
ever new dbs need to be designed, implemented, optimized, and extended.
However, much more frequently than in almost any other industry, we are
dually confronted with the need to exploit, maintain, and extend legacy
dbs, which, generally, lack either completely or at least partially corre-
sponding design, implementation, and optimization adequate documenta-
tion. This is why corresponding dual reverse engineering algorithms are
very useful (and sometimes even crucial) too.
Figure 1.0 presents the 25 algorithms proposed in this book. Figure 1.1
shows an overview of the proposed fully algorithmic data analysis, con-
ceptual data modeling, database design, and optimization forward engi-
neering process, as well as its dual reverse engineering one.
A7/8–3 needs to be applied at the RDM level only if neither A3, nor A3′
were previously applied (i.e., when the shortcut A1–7 was taken instead
of the full normal path A1, A2, A3, A4, A5, A6, A7); similarly, it needs to
be applied at the RDBMS level only if neither A3 nor A3′ were previously
applied and if it was not applied at the RDM level either (i.e., when the
shortcut AF1–8 was taken instead of the full normal path A1, A2, A3, A4,
A5, A6, A7, A8, AF8′).
Data,
 Information and Knowledge in the Computer Era 21

A0—The algorithm for assisting data analysis and initial conceptual

data modeling (in order to obtain E-R data models)
A1—The algorithm for translating E-R data models into mathemati-
cal schemas
A2—The algorithm for assisting validation of the initial E-R data
model
A3—The algorithm for assisting keys discovery (mathematical level)
A3′—The improved algorithm for assisting keys discovery (math-
ematical level)
A4—The algorithm for assisting analysis of E-RD cycles
A5—The algorithm for guaranteeing the coherence of constraint sets
A6—The algorithm for minimization of constraint sets
A7—The algorithm for translating mathematical schemas into rela-
tional ones and associated nonrelational constraint sets
A7/8–3—The algorithm for assisting keys discovery (relational/
RDBMS levels)
A8—The algorithm for translating RDM schemas into SQL ANSI
DDL scripts
AF8′—The family of algorithms for translating ANSI DDL scripts
into rdbs
AF9—The family of algorithms for assisting enforcement of nonre-
lational constraints through SQL triggers
AF9′—The family of algorithms for assisting enforcement of non-
relational constraints through high level PL embedding SQL
trigger-type methods
AF10—The family of algorithms for assisting initial optimization
of rdbs
AF11—The family of algorithms for assisting regular optimizations
of rdbs
A1–7—The algorithm for translating E-R data models into relational
schemas and associated nonrelational constraint sets*
AF1–8—The family of algorithms for translating E-R data models
into rdbs and associated nonrelational constraint sets*
REAF0′—The family of algorithms for translating rdb schemas into
SQL ANSI DDL scripts
22 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

REA0—The algorithm for translating SQL ANSI DDL scripts into

relational schemas
REA1—The algorithm for translating relational schemas into math-
ematical ones
REA2—The algorithm for translating mathematic schemas into E-R
data models
REAF0–1—The family of algorithms for translating rdb schemas
into mathematical ones
REAF0–2—The family of algorithms for translating rdb schemas
into E-R data models
REA1–2—The algorithm for translating relational schemas into E-R
data models
*: without the refinements brought by the algorithms A2 to A6.

FIGURE 1.0  Proposed algorithms for data analysis, db design, implementation, and
optimization, as well as their dual reverse engineering ones.

In fact, out of these 25 algorithms 9 (AF8’, AF9, AF9’, AF10, AF11,

AF1–8, as well as REAF0’, REAF0–1, and REAF0–2) are family of algo-
rithms (with the ‘F’ in their name from “Family”), each of them including
one corresponding algorithm per RDBMS version. For example, all of
them include the subfamilies DB2, Oracle, SQL Server, Access, etc.; in
their turn, for example, all Oracle subfamilies include algorithms for its
versions 12c, 11g, 10g, 9i, etc. For each such family, this book actually
presents, as an example, only a couple of algorithms and invites its readers
to similarly design their own ones for the rest of the RDBMS versions.
18 algorithms are of forward and 7 of reverse engineering types. Reverse
engineering algorithms are represented with upwards dashed arrows and
their names are prefixed with “REA”; forward ones have downwards not
dashed arrows and their names are prefixed with “A”.
19 algorithms are fundamental and 6 are derived ones: these latter
“shortcut” ones are composed: A1–7 is the composition of A1 followed
by A7; any algorithm from AF1–8 is the composition of A1–7 followed by
A8, and one from AF8′; REA1–2 is the composition of REA1 followed by
REA2; any algorithm from REAF0–1 is the composition of one algorithm
Data,
 Information and Knowledge in the Computer Era 23

FIGURE 1.1  Overview of proposed algorithms for data analysis, db design,

implementation, and optimization, as well as their dual reverse engineering ones.

from REAF0′ followed by REA0 and REA1; finally, any algorithm from
REAF0–2 is the composition of one algorithm from REAF0–1 followed
by REA2.
16 algorithms are of translation and 9 of refinement types: A2, A3, A3’,
A4, A5, A6, A7/8–3, AF10, and AF11 are refining-type algorithms; all of the
others are translation-type (from one formalism to another) ones. 15 algo-
rithms need not human intervention, while 10 (the assisting ones) do need
it: A0, A2, A3, A3’, A4, A7/8–3, AF9, AF9’, AF10, and AF11 are assisting
algorithms, as only humans may take corresponding needed decisions.
All of these 25 algorithms may and should be embedded in any power-
ful DBMS. Unfortunately, today’s RDBMSs are only providing A8, AF8′
and, partially, REAF0–2 and/or AF10/AF11; only a couple of them (e.g.,
DB2 and Oracle) are also offering AF1–8 (but without the refinements
brought by the algorithms A2 to A6 or, at least, by A7/8–3).
24 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

Theoretically, E-RDM might be dropped (in which case an A0–1 algo-

rithm should replace A0 and A1), but, in my opinion, it would not only be a
pity, but neither users and nor even business analysts and db designers would
be prepared or at least ready to get prepared for it in the foreseeable future.
(E)MDM might be replaced with any other higher-level conceptual data
model that provides at least equivalent to A1, A2, A3/A3’, A4, A5, A6, A7,
AF9, and AF9′ algorithms. Dropping this level would generally result in
semantic overloading of object sets and/or much, much poorer data plau-
sibility, as well as possible constraint sets incoherence and/or redundancy.
RDM might be replaced as well (e.g., with a modern graph-based data
model, an O-ODM, a semistructured, or a heterogeneous13 one), provided
that A7, A7/8–3, if needed, and AF8′ are replaced with corresponding
equivalent ones. Dropping this level would compromise the portability of
the conceptual data modeling solutions on any number of existing and to
come RDBMS versions. Moreover, learning RDM is for RDBMSs similar,
for example, to learning formal languages for PLs: you may then extremely
easily navigate between any versions of any manufacturer, only having to
pay attention to their minor syntactical and technological peculiarities.
RDBMS versions might be replaced, for example, with O-ODBMS
or NoSQL14 ones, provided that AF8’, AF9, AF10, and AF11 are replaced
with corresponding equivalent ones.
This first volume introduces only the E-RDM, RDM, RDBMS dbs, and
(the family of) algorithms A0, A1–7, A7/8–3, A8, AF8’, AF1–8, REAF0’,
REA0, REA1–2, and REAF0–2; the (E)MDM and the rest of the algorithms
are presented in the second volume.
In my opinion, the shortest forward advisable path is A0, AF1–8,
A7/8–3, AF10, while the professional one is A0, A1, A2, A3/A3’, A4, A5,
A6, A7, A8, AF8’, AF10, AF11.
Unfortunately, the most frequent path taken today (if any!) is AF1–8,
with some pros also considering some partial implicit variants of A0, AF10
and AF11.

13
Data is heterogeneous when has several format types (e.g., a mixture of at least two of the follow-
ing: binary, contiguous, graph, image, map, audio, video, ranking, URL, etc.). Heterogeneous DBMS
(HDBMS) are systems able to integrate heterogeneous data managed by other DBMSs.
14
NoSQL DBMS (NoSQL = Not Only SQL) are systems that replaced RDM with some other types of
less restrictive data models (e.g., based on pairs of type <key, value>, or graphs, or documents, etc.),
but still use SQL-like languages (for both declaring and processing data).
Data,
 Information and Knowledge in the Computer Era 25

1.8  PAST AND PRESENT

I like the dreams of the future better than the history of the past.
—Thomas Jefferson

Real generosity toward the future lies in giving all to the present.
—Albert Camus

In my opinion, modern data modeling era opened with Mealy (1967) and
Quillian (1968).
Codd (1970) introduced the RDM.
Chen (1976) introduced the E-RDM.
Smith and Smith (1977) was the first to explore using generalization
and aggregation as data modeling tools.
Gallaire and Minker (1978) first married logic and databases: knowl-
edge bases (in particular, deductive databases) sprung from it.
Hammer and McLeod (1978) proposed a first semantic data model.
Mancas (1985) introduced the (E)MDM, and Mancas (1990, 1997,
2002) extended it.
The two main prototype implementations that immensely contributed
to the technology of the RDBMS field were the System R project at IBM
Almaden Research Center (Astrahan, 1976) and the INGRES project at
Berkeley University (Stonebraker et al., 1976). Both of them were crucial
for establishing this type of DBMS as the dominant db technology of today,
including the relational languages SQL (Chamberlin and Boyce, 1974), the
de facto standard one today, and its rival Quel (Stonebraker et al., 1976).
IBM DB2 (http://www-01.ibm.com/software/data/db2/), System R’s
direct descendant, celebrated a couple of years ago its 30 years of success
and superior innovation (Campbell et al., 2012).
DB2′s main competitor is another System R’s descendant, the Oracle
Database (http://www.oracle.com/us/products/database/overview/index.
html), or, simply, Oracle, which managed to reach the market some two
years before DB2 and which continued ever since to challenge it in a beau-
tiful race of innovation and excellence (Ashdown and Kyte, 2014).
INGRES too spawned several very successful commercial RDBMSs: the
current open source commercial Ingres supervised by Actian Corporation
26 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

(http://www.actian.com/products/operational-databases/), Sybase (now a SAP

company) SQL Server (currently known as the Advantage Database Server,
http://www.sybase.com/products/databasemanagement/advantagedatabas-
eserver, after Microsoft bought its license in 1992), Tandem NonStop SQL
(http://h17007.www1.hp.com/ro/en/enterprise/servers/integrity/nonstop/
nonstop-sql.aspx#.U8BUe7EhHCI), now owned by Hewlett-Packard, the
post-Ingres family of products Postgres, which evolved into the current open
source PostgreSQL (http://www.postgresql.org/), and Informix Database,
now IBM owned (http://www-01.ibm.com/software/data/informix/) are the
best known ones.
Besides the SQL Server (http://www.microsoft.com/en-us/server-
cloud/products/sql-server/#fbid=o6jljjvm1J0, http://www.sqlservercen-
tral.com/), now the biggest competitor for DB2 and Oracle (all three of
them currently having cloud-empowered solutions available too), Micro-
soft also still offers its genuine smaller scale RDBMS Access (http://office.
microsoft.com/en-us/access/).
MySQL (http://www.mysql.com/), one of the most popular open source
small scale RDBMS, is now under Oracle Corp.’s umbrella too.
Although they use their powerful in-house developed DBMS BigTable
(https://developers.google.com/appengine/articles/storage_breakdown),
not distributed outside the company, but to which users of the Google App
Engine cloud platform as a service (https://developers.google.com/appen-
gine/) may have access however, Google is offering their MySQL’s Cloud
SQL version too (https://developers.google.com/cloud-sql/).
Nowadays, on one hand, even smartphones are equipped with
RDBMSs (e.g., Android is using SQLite, http://www.sqlite.org/, an open
source small RDBMS engine sponsored mainly by Oracle, Bentley,
Mozilla, Nokia, Adobe, and Bloomberg, which, probably, is the current
most widely deployed mobile db engine worldwide).
On the other, the so-called big data15 companies are using, just like
Google, so-called NoSQL DBMSs, like Apache Cassandra (http://cas-
sandra.apache.org/, developed by Facebook when MySQL proved to not
be satisfactory anymore and currently used by more than other 1,500
15
Big data is denoting data sets generally unstructured and having sizes of petabytes (PB, 1015 bytes =
103 terabytes), which are consequently beyond the ability of commonly used software tools to manage
and process within user tolerable periods of time.
Data,
 Information and Knowledge in the Computer Era 27

companies too, among which, for example, are CERN, the Swiss-based
European Council for Nuclear Research, which built and uses the Large
Hadron Collider, eBay, and The Weather Channel), HBase (http://www.
webuzo.com/sysapps/databases/HBase), the main Hadoop db engine
(http://hadoop.apache.org/, whose underlying technology was invented
by Google and then developed by Yahoo), and Hive (http://hive.apache.
org/, of SQL-type, born within the Hadoop project too, but now being a
project of its own), all of them open source and best suited for mission-
critical data that is replicated in hundred worldwide scattered data centers.
For example, (Redmond and Wilson, 2012) presents 7 NoSQL open
source solutions (including HBase and Postgres). Moreover, there are also
bridges built between the complementary SQL and NoSQL solutions, for
bringing speed and BI capabilities to querying heterogeneous big data,
like, for example, the one between the MS SQL Server 2012 and Hadoop
(Sarkar, 2013).
O-ODBMS were introduced starting with the 1982 Gemstone proto-
type (see http://gemtalksystems.com/index.php/products/gemstones/).
Even if they only managed to conquer niche areas (CAD, multimedia pre-
sentations, telecom, high energy physics, molecular biology, etc.), their
influence on RDM and RDBMSs, at least for the hybrid object-relational
(O-R) systems was significant. Out of them, probably the most famous,
especially after IBM bought it (from Illustra), is Informix (http://www-01.
ibm.com/software/data/informix/).
Only some 10 years after the first commercial RDBMSs were launched,
Chang (1981) described a first system, which is marrying rdbs and infer-
ence rules. Most probably, the name Datalog was first coined to the cor-
responding language in Maier (1983).
It then took almost another decade until Naqvi and Tsur (1989), after
a 5 years efforts of MCC, introduced LDL, a language extending Datalog
with sets, negation, and data updating capabilities, which was the founda-
tion of the first widely known homonym KDBMS (Chimenti et al., 1990).
The University of Wisconsin at Madison started in 1992 the CORAL
project (Ramakrishnan et al., 1992), experimenting a marriage between
SQL and Prolog, which was implemented in a corresponding homonym
KDBMS one year later (Ramakrishnan et al., 1992). CORAL++ (Srivas-
tava et al., 1993) is its object-oriented extension.
28 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

Today, there are lot of prototype KDBMS and very few commercial
ones. Among the former, for example, there are the notable open source
semantic web framework Jena (http://jena.apache.org//), developed
(first by HP, but then by Apache) in Java and also embedding Datalog
and OWL, the Datalog educational system DES (http://www.fdi.ucm.es/
profesor/fernan/des/), licensed by the Free Software Foundation, and the
Apache Clojure and Java library for querying data stored on Hadoop clus-
ters Cascalog (http://cascalog.org/).
Among the latter, for example, there are QL (de Moor et al., 2008), a
commercial object-oriented variant of Datalog created by Semmle Ltd.
(http://semmle.com/), a free-of-charge database binding for pyDatalog (a
logic extension for Python, see https://sites.google.com/site/pydatalog/)
from FoundationDB (https://foundationdb.com/key-value-store/docu-
mentation/datalog.html), and a distributed db system for scalable, flexible
and intelligent applications running on cloud architectures, which uses
Datalog as the query language, called Datomic (http://www.datomic.com/
rationale.html), from Cognitect (http://cognitect.com/).
Even Microsoft Research developed and provides a security policy
language based on Datalog called SecPal (http://secpal.codeplex.com/).
MatBase has several prototype versions (MS C and Borland Paradox,
MS Access, C# and SQL Server) and was presented in a series of papers:
Mancas et al. (2003), Mancas and Dragomir (2004), and Mancas and Man-
cas (2005, 2006). Its users may either work in (E)MDM (and the system
automatically translates everything into both E-RDM and RDM), which
also includes a Datalog engine, in E-RDM (and the system automatically
translates everything into both (E)MDM and RDM), or in RDM (and
the system automatically translates everything into both E-RDM and (E)
MDM).
The excellent (Tsichritzis and Lochovsky, 1982) and the more recent
(Hoberman, 2009; Hoberman et al., 2009; Simsion and Witt, 2004; Sim-
sion, 2007), etc. are devoted to data modeling.
The recently revived graph-based database models are discussed in
depth, for example, in Robinson et al. (2013).
For data model examples (even if not all of them correct and/or opti-
mal), see, for example, http://www.databaseanswers.org/data_models/
index.htm.
Data,
 Information and Knowledge in the Computer Era 29

Not only in my opinion, the current “bible” version for database the-
ory still remains (Abiteboul et al., 1995), while (Garcia-Molina et al.,
2014) is the one for DBMSs.
Other remarkable books in this field are, for example, Date
(2003, 2011, 2012, 2013), Churcher (2012), Hernandez (2013), Light-
stone et al. (2011), Ullman, (1988, 1989), and Ullman and Widom
(2007).
Especially in O-ODBMSs, but not only, data is frequently
stored in XML; as such data is somewhere at the half-road between
dbs and plain text documents it is called semistructured; again not
only in my opinion, the current “bible” in this field remains (Abite-
boul et al., 2000).
DBLP (http://www.sigmod.org/dblp/db/index.html), started and
still managed by Michael Ley, lists now more than 1,200,000 pub-
lications in the db field. A smaller searchable index of db research
papers (only some 40,000 entries, but out of a total of more than
3,000,000 ones in computer science, with more than 600,000 also
having links to the full papers), http://liinwww.ira.uka.de/bibliog-
raphy/Database/, is maintained by Alf-Christian Achilles.
All URLs mentioned in this section were last accessed on July
11th, 2014.

KEYWORDS

•• big data
•• business rule (BR)
•• closed domain assumption
•• closed world assumption
•• coherent constraint set
•• conceptual data modeling
•• consistent database instance
•• constraint
•• constraint satisfaction
•• constraint set
•• constraint set closure
30 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

•• constraint sets equivalence

•• constraint violation
•• data
•• data aggregation
•• data analysis
•• Data Definition Language (DDL)
•• data factorization
•• data generalization
•• Data Manipulation Language (DML)
•• data model
•• data plausibility
•• database (db)
•• database instance
•• database management system (DBMS)
•• database scheme
•• database stored methods
•• Datalog
•• derived database
•• E-R data model
•• (Elementary) Mathematical Data Model ((E)MDM)
•• Entity-Relationship (E-R) Data Model (E-RDM)
•• fundamental constraint
•• fundamental database
•• heterogeneous data
•• homogeneous data
•• implausible constraint
•• implication problem
•• implied constraint
•• incoherent constraint set
•• incomplete data
•• inconsistent database instance
•• inference rule
Data,
 Information and Knowledge in the Computer Era 31

•• information
•• knowledge and database management system (KDBMS)
•• knowledge base (kb)
•• knowledge-based management system (KBMS)
•• Logical Data Model (LDM)
•• metadata
•• minimal constraint set
•• negative data
•• NoSQL DBMS
•• object set
•• object set property
•• plausible constraint
•• positive data
•• redundant constraint
•• redundant constraint set
•• Relational Data Model (RDM)
•• semistructured data
•• Structured Query Language (SQL)
•• trivial constraint
•• unique objects assumption

REFERENCES

No finite point has meaning without an infinite reference point.

—Jean-Paul Sartre

Abiteboul, S., Buneman, P., Suciu, D. (2000). Data on the Web. From Relations to Semis-
tructured Data and XML. Morgan Kaufman: San Francisco, CA.
Abiteboul, S., Hull, R., Vianu, V. (1995). Foundations of Databases; Addison-Wesley:
Reading, MA.
Ashdown, L., Kyte, T. (2014). Oracle Database Concepts, 12c Release 1 (12.1). Oracle
Corp. (http://docs.oracle.com/cd/E16655_01/server.121/e17633.pdf).
Astrahan, M. M., et al. (1976). System R: A Relational Approach to Database Manage-
ment. In ACM TODS, 1(2), 97–137.
32 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

Campbell, J., Haderle, D., Parekh, S., Purcell, T. (2012). IBM DB2: The Past, Present and
Future: 30 Years of Superior Innovation. MC Press Online, LLC, Boise, ID.
Chamberlin, D. D., Boyce, R. F. (1974). SEQUEL—A Structured English QUEry Lan-
guage. In SIGMOD Workshop, 1, 249–264.
Chang, C. C. (1981). On the Evaluation of Queries Containing Derived Relations in a Rela-
tional Database. In H. Gallaire, J. Minker, and J. Nicolas, eds., Advances in Database
Theory, Vol. I, Plenum Press.
Chen, P. P. (1976). The entity-relationship model: Toward a unified view of data. ACM
TODS 1(1), 9–36.
Chimenti, D., Gamboa, R., Krishnamurti, R., Naqvi, S., Tsur, S., Zaniolo, C. (1990). The
LDL system prototype. IEEE TKDE 2(1):76–90.
Churcher, C. (2012). Beginning Database Design: From Novice to Professional, 2nd ed.,
Apress Media LLC: New York, NY.
Codd, E. F. (1970). A relational model for large shared data banks. CACM 13(6), 377–387.
Date, C. J. (2003). An Introduction to Database Systems, 8th ed., Addison-Wesley: Read-
ing, MA.
Date, C. J. SQL and Relational Theory: How to Write Accurate SQL Code, 2nd ed.; Theo-
ries in Practice; O’Reilly Media, Inc.: Sebastopol, CA, 2011.
Date, C. J. Database Design & Relational Theory: Normal Forms and All That Jazz; Theo-
ries in Practice; O’Reilly Media, Inc.: Sebastopol, CA, 2012.
Date, C. J. Relational Theory for Computer Professionals: What Relational Databases are
Really All About; Theories in Practice; O’Reilly Media, Inc.: Sebastopol, CA, 2013.
De Moor, O. et al., (2008). QL: Object-Oriented Queries Made Easy. Lammel, R., Visser,
J., Saraiva, J. (Eds.): GTTSE 2007, LNCS 5235: 78–133, Springer-Verlag, Berlin
Heidelberg.
Gallaire, H., Minker, J. (1978). Logic and Databases. Plenum Press, New York, U.S.A.
Garcia-Molina, H., Ullman, J. D., Widom, J. (2014). Database Systems: The Complete
Book, 2nd ed., Pearson Education Ltd.: Harlow, U.K., (Pearson New International
Edition)
Hammer, M., McLeod, D. (1978). The Semantic Data Model: a Modeling Mechanism for
Database Applications. In ACM SIGMOD Int. Conf. on the Manag. of Data.
Hernandez, M. J. (2013). Database Design for Mere Mortals: A Hands-on Guide to Rela-
tional Database Design, 3rd ed., Addison-Wesley: Reading, MA.
Hoberman, S. (2009). Data Modeling Made Simple 2nd Edition. Technics Publications
LLC, Bradley Beach, NJ.
Hoberman, S., Blaha, M., Inmon, B., Simsion, G. (2009). Data Modeling Made Simple:
A Practical Guide for Business and IT Professionals 2nd Edition. Technics Publica-
tions LLC, Bradley Beach, NJ, (Fourth Printing).
Lightstone, S. S., Teorey, T. J., Nadeau, T., Jagadish, H. V. (2011). Database Modeling and
Design: Logical Design, 5th ed., Data Management Systems; Morgan Kaufmann:
Burlington, MA.
Maier, D. (1983). The Theory of Relational Databases. Computer Science Press: Rock-
ville, MD.
Mancas, C. (1985). Introduction to a data model based on the elementary theory of sets, re-
lations and functions (in Romanian). In Proc. of INFO IASI’85, 314–320, A.I. Cuza
University, Iasi, Romania.
Data,
 Information and Knowledge in the Computer Era 33

Mancas, C. (1990). A Deeper Insight into the Mathematical Data Model. Proc. 13th Intl.
Seminar on DBMS, ISDBMS’90, 122–134, Mamaia, Romania.
Mancas, C. (1997). Conceptual data modeling. (in Romanian) Ph.D., Thesis: Politehnica
University, Bucharest, Romania.
Mancas, C. (2002). On Knowledge Representation Using an Elementary Mathematical
Data Model. In Proc. IASTED IKS 2002. Conf. on Inf. and Knowledge Sharing,
206–211, Acta Press, St. Thomas, U.S. Virgin Islands, U.S.A.
Mancas, C., Dragomir S., Crasovschi, L. (2003). On modeling First Order Predicate Cal-
culus using the Elementary Mathematical Data Model in MatBase DBMS. In Proc.
IASTED AI 2003. MIT Conf. on Applied Informatics, 1197–1202, Acta Press, Inns-
bruck, Austria.
Mancas, C., Dragomir S. (2004). MatBase Datalog Subsystem Metacatalog Conceptual
Design. In Proc. IASTED SEA 2004. MIT Conf. on Software Eng. and App., 34–41,
Acta Press, Cambridge, MA.
Mancas, C., Mancas, S. (2005). MatBase E-R Diagrams Subsystem Metacatalog Concep-
tual Design. In Proc. IASTED DBA 2005. Conf. on DB and App., 83–89, Acta Press,
Innsbruck, Austria.
Mancas, C., Mancas, S. (2006). MatBase Relational Import Subsystem. In Proc. IASTED
DBA 2006. Conf. on DB and App., 123–128, Acta Press, Innsbruck, Austria.
Mealy, G. H. (1967). Another Look at Data. In AFIPS Fall Joint Comp. Conf., 31, 525–534.
Naqvi, S., Tsur, S. (1989). A Logical Language for Data and Knowledge Bases. Computer
Science Press: Rockville, MD.
Quillian, H. R. (1968). Semantic memory. In Semantic Inf. Processing, Minskz, M. ed.,
M.I.T. Press.
Ramakrishnan, R., Srivastava, D., Sudarshan, S. (1992). CORAL: Control, Relations and
Logic. In VLDB.
Ramakrishnan, R., Srivastava, D., Sudarshan, S., Sheshadri, P. (1993). Implementation of
the {CORAL} deductive database system. In SIGMOD.
Redmond, E., Wilson, J. R. (2012) Seven Databases in Seven Weeks: A Guide to Modern
Databases and the NoSQL Movement. Pragmatic Bookshelf.
Robinson, I., Webber, J., Eifrem, E. (2013). Graph Databases. O’Reilly Media Inc.,
Sebastopol, CA.
Sarkar, D. (2013) Microsoft SQL Server 2012 with Hadoop. Packt Publishing, Birmingham,
Mumbai.
Simsion, G. C. (2007). Data Modeling: Theory and Practice. Technics Publications LLC,
Bradley Beach, NJ.
Simsion, G. C., Witt, G. C. (2004). Data Modeling Essentials, 3rd Edition. Morgan-Kauf-
mann Publishers, Burlington, MA.
Smith, J. M., Smith, D. P. C. (1977). Database abstractions: Aggregation and generaliza-
tion. ACM TODS 2(2):105–133.
Srivastava, D., Ramakrishnan, R., Sudarshan, S., Sheshadri, P. (1993). CORAL++; Adding
Object-orientation to a Logic Database Language. In Proc. VLDB Dublin, Ireland,
158–170.
Stonebraker, M., Wong, E. A., Kreps, P., Held, G. (1976) The design and implementation
of Ingres. In ACM TODS, 1(3), 189–222.
Tsichritzis, D. C., Lochovsky, F. (1982). Data models. Prentice-Hall: Upper Saddle River, NJ.
34 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

Ullman, J. D. (1988). Principles of Database and Knowledge-Based Systems—Volume 1:

Classical Database Systems. Computer Science Press: Rockville, MD.
Ullman, J. D. (1989). Principles of Database and Knowledge-Based Systems—Volume 2:
The New Technologies. Computer Science Press: Rockville, MD.
Ullman, J. D., Widom, J. (2007). A First Course in Database Systems, 3rd ed. Prentice
Hall: Upper Saddle River, NJ.
CHAPTER 2

THE QUEST FOR DATA

ADEQUACY AND SIMPLICITY:
THE ENTITY-RELATIONSHIP DATA
MODEL (E-RDM)

The most incomprehensible thing about the world is that

it is comprehensible.
—Albert Einstein

CONTENTS

  2.1  Entity and Relationship Type Object Sets..................................... 37

  2.2  Attributes and Surrogate Keys...................................................... 38
  2.3  Entity-Relationship Diagrams (E-RDs)........................................ 40
  2.4 Functional Relationships and the Key Propagation
Principle (KPP)............................................................................. 42
  2.5  Relationship Hierarchies............................................................... 44
  2.6  Higher Arity Non-Functional Relationships................................. 48
  2.7  Restriction Sets............................................................................. 49
  2.8 Case Study: A Public Library (Do We Know Exactly
What a Book Is?).......................................................................... 60
  2.9 The Algorithm for Assisting the Data Analysis and Modeling
Process (A0): An E-R Data Model of the E-RDM....................... 69
2.10  Best Practice Rules....................................................................... 75
2.11 The Math Behind E-RDs and Restriction Sets—The Danger
of “Many-To-Many Relationships” and the Correct E-RD
of E-RDM..................................................................................... 92
36 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

2.12 Conclusion.................................................................................... 98
2.13 Exercises....................................................................................... 99
2.14  Past and Present...........................................................................110
Keywords................................................................................................111
References...............................................................................................112

After studying this chapter you should be able to:

• Concisely and precisely define, and exemplify the following key
terms: entity, relationship (one-to-one, one-to-many, many-to-one,
many-to-many, functional, recursive, non-functional), attribute,
surrogate key, role, Entity-Relationship (E-R) Diagram (E-RD),
their equivalent (Chen, Barker, and Mancas) notations, E-R data
model, the Key Propagation Principle (KPP), relationship hierar-
chy, restriction set, restriction (attribute range, compulsory data,
uniqueness, other types), controlled data redundancy.
• Apply algorithm A1 (assistance of data analysis and conceptual
modeling) and the best practice rules for correctly solving at least
all exercises found at the end of the chapter.
Generally, our customers (be them companies, or our team leaders,
professors, etc.) are providing us a brief informal description of the dbs
they would like us to design and implement for them. These descriptions
are very rarely complete, accurate, clear, and not ambiguous; sometimes
they are even only oral ones.
Starting to implement corresponding dbs is a wrong decision even for
the very small and simple ones of them, just like starting to type code
before analyzing corresponding software applications usage, architect-
ing, and designing them: without data analysis, conceptual, and physical
design, dbs will not be correct, scalable, and optimal.
For each db, data analysis should come up with an E-R data model,
made out of the following data modeling deliverables, all of them agreed
by both the customer and the db designer:
1.  A complete, accurate, clear, not ambiguous, well structured, and
concise written description of the desired db (that may refer to the
documents in 2 and 3 below).
The
 Quest for Data Adequacy and Simplicity 37

2.  A set of corresponding E-R diagrams (E-RDs).

3.  An associated restriction set (that refers to 1 and 2 above).
Moreover, a “contract” (be it juridical or not) between the customer
and the db designer (that should have as its appendixes all of the above
documents) stipulating requirements, deadlines, costs, etc., would be
desirable too.
Obviously, the “contract” is not within the scope of this book, whereas
1 is fairly simple (and we too will provide examples in the sequel), as it is
merely the result of an informal reverse engineering process applied to the
results of 2 and 3; this is why we are only focusing on obtaining the second
and third above components, which are the first and, generally, the most
important steps towards both formalizing the customer db requirements
and needs, as well as understanding undocumented legacy dbs.

2.1  ENTITY AND RELATIONSHIP TYPE OBJECT SETS

I love bringing the inanimate object to life.

—John Lasseter

E-RDM considers two types of object sets:

• Entity: atomic, independent ones (e.g., PEOPLE, EMPLOYEES,
STUDENTS, COURSES, CUSTOMERS, CITIES, COUNTRIES,
PRODUCTS, PRODUCT_TYPES, SERVICES, etc.), graphically
represented by rectangles.
• Relationship: complex, nonindependent, mathematically relation
type (i.e., Cartesian products subsets; for example, MARRIAGES,
STOCKS, COAUTHORS, BIRTH_PLACES, HOME_TOWNS,
CAPITALS, etc.), graphically represented by diamonds. Initially,
their underlying sets were only of entity type. Depending on their
cardinalities (that are labeling corresponding edges), they are classi-
fied into many-to-many, one-to-many, many-to-one, and one-to-one
types (e.g., STOCKS, MARRIAGES, CO-AUTHORS are many-to-
many, BIRTH_PLACES is either one-to-many, if CITIES is repre-
sented at its left side, or many-to-one, if PEOPLE is represented
at its left side, whereas CAPITALS is one-to-one). Cardinalities are
38 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

also used for representing compulsory type restrictions: by using

0 instead of 1, you are asserting that the corresponding property
value might not always be known or applicable (e.g., see BIRTH_
PLACES in Fig. 2.1). Relationship edges should be labeled by roles
(e.g., Husband, Wife, Product, Warehouse, etc.). We extended rela-
tionships by allowing them to be defined on other relationships as
well, thus being possible to build relationship hierarchies too.

2.2  ATTRIBUTES AND SURROGATE KEYS

Painting is concerned with all the 10 attributes of sight, which are:

Darkness, Light, Solidity and Color, Form and Position, Distance and
Propinquity, Motion and Rest.
—Leonardo da Vinci

Both types of these object sets have associated attributes (e.g., Sex, First-
Name, LastName, e-mailAddr, BirthDate, MarriageDate, HireDate, Cus-
tomerName, CityName, ZipCode, Address, CountryName, CountryCode,
CountryPopulation, WarehouseName, ProductTypeName, StockQuantity,
EndDate, etc.), graphically represented by ellipses attached to the corre-
sponding rectangles or diamonds.
Please note that in any such data model, names of the object sets (be
them entity or relationship type) must be unique. Similarly, all attributes
of any given object set should have distinct names (but there may be attri-
butes of different object sets having same names).
Mathematically, attributes are mappings defined on corresponding
associated sets and taking values into corresponding value sets (i.e., sub-
sets of data types; for example, {‘F’, ‘M’}, [01/01/1900, 31/12/2012],
{True, False}, [30, 250], ASCII(32), UNICODE(32), etc.).
In the original E-RDM, attributes may take values into Cartesian prod-
uct sets as well; we do not allow it and restrict their codomains to atomic
sets only.
The E-RDM also generally recommends adding for each object set (be
it of type entity or relationship) a distinguished attribute called surrogate
key, having no other semantics except for numeric unique identification
The
 Quest for Data Adequacy and Simplicity 39

of the elements of the corresponding set. Their names generally contain

either “ID” (most commonly) or “#”; for example, ID_Discipline or Dis-
ciplineID, or Discipline#.
However, we do not recommend adding surrogate keys at this highest
data modeling level, first of all in order not to confuse users (for whom
they are irrelevant), but also in order to simplify business analysts work, as
they may be automatically added later. When they are added, their names
should be underlined.
Please note that same concepts should be correctly modeled either as
(entity-type) object sets or as attributes, depending on the context. For exam-
ple, for people, their eyes or hair colors may be modeled as attributes (Eye-
Color and HairColor) of the PEOPLE object set; in a physics context, where
colors are defined by very many properties (e.g., Kelvin grades temperature,
saturation, hue, etc.) you should abstract an object set COLORS instead.
Generally, whenever several properties are needed to define an object
you should abstract a corresponding object set; if there is only one prop-
erty, then you might not do so, but simply declare that property (e.g.,
Color) as being a property of a corresponding related object set (like Eye-
Color and HairColor above).
However, sometimes, for optimality, it is preferable to abstract object
sets even if they have only one property. For example, in order not to
store several times the names of eye and hair colors, but especially in
order not to impose to users to type them every time (but provide them a
drop-down list from where to choose the desired values), the best model-
ing solution for this subuniverse is to consider the entity-type object sets
PEOPLE, EYE_COLORS and HAIR_COLORS and the many-to-one rela-
tionship-type sets PEOPLE_EYE_COLORS and PEOPLE_HAIR_COL-
ORS between them.
Context dependency is also dictating the needed data modeling refine-
ment level, from which we should similarly decide too whether a given
concept should be modeled as an (entity-type) object set or as a property
(attribute).
For example, in most usual subuniverses where people addresses are
also needed, besides the entity-type object sets PEOPLE and CITIES and
the many-to-one relationship-type one PEOPLE_HOMETOWNS, all that
is needed is an attribute Address of PEOPLE.
40 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

However, again, when street addresses risk to repeat themselves a lot,

like, for example, in a postal or electoral subuniverse, addresses should not
be modeled anymore as (text) attributes, but split on several refinement lev-
els, by abstracting object sets STREETS, BUILDINGS, and APPARTMENTS
(and in big cities also NEIGHBORHOODS and STAIRCASES), as well as
the corresponding many-to-one relationships between them and PEOPLE.

2.3  ENTITY-RELATIONSHIP DIAGRAMS (E-RDS)

Each relationship nurtures a strength or weakness within you.

—Mike Murdock

Figure 2.1 presents an example of an entity-relationship diagram (E-RD).

Many other graphical conventions exist and very many extensions
were added to E-RDM, mainly in order to represent other constraint types.
For example, Oracle uses the so-called “crow’s foot” (or Barker) notation
instead of diamonds and corresponding cardinalities; for example, MAR-
RIAGES and BIRTH-PLACES are represented as in Fig. 2.2.

FIGURE 2.1  An example of an entity-relationship diagram (E-RD).

FIGURE 2.2  An example of an Oracle-style E-RD.

The
 Quest for Data Adequacy and Simplicity 41

Similarly, mainly because, mathematically, both one-to-many, many-

to-one, and one-to-one relations are functions, we prefer a mathematical
type convention and represent these mappings by arrows, double arrows
for one-to-one and onto ones and with reverse arrows not tangent to the
domain set for only one-to-one and not onto ones, like for Capital in
Fig. 2.3.
Note that this notation is not only simpler and familiar since elemen-
tary school, but also more natural; for example, in contrast with Fig. 2.1,
BIRTH-PLACES is replaced by BirthPlace (anybody has a birth place and
only one), STATES_CITIES by State (any city belongs to a state and to
only one), COUNTRIES_STATES by Country (any state belongs to a coun-
try and to only one), and CAPITALS by Capital (any country has a capital
and only one and, dually, no city may simultaneously be the capital of
several countries, but of only one). Also note that one-to-one attributes are
represented similar to one-to-one relationships (e.g., see SSN in Fig. 2.5).
Moreover, in our notation, inclusions16 are represented consistently too
(and do not need names17), like in Fig. 2.4.

FIGURE 2.3  An example of a mathematical-style E-RD.

FIGURE 2.4  Representation of the inclusion EMPLOYEES # PEOPLE.

Computed objects, be them entity or relationship type object sets or

attributes, are represented by dotted rectangles, diamonds, arrows, and
ellipsis, respectively; in Fig. 2.5, *MALES and *FEMALES, as well as
*Age and*HomeCountry are computed objects.
The table in Fig. 2.6 summarizes contrastively the original (Chen),
Barker (Oracle), and our (Mancas) E-RD notations.
16
IS-A relationships from AI.
17
As associated mappings are the corresponding canonical inclusion ones.
42 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

FIGURE 2.5  An example of an E-RD also including a one-to-one attribute and

computed object sets, structural mappings, and attributes.

FIGURE 2.6  Chen, Barker, and Mancas E-RD notational equivalences.

2.4  FUNCTIONAL RELATIONSHIPS AND THE KEY

PROPAGATION PRINCIPLE (KPP)

Failure comes only when we forget our ideals and

objectives and principles.
—Jawaharlal Nehru

The relationship examples from Figs. 2.3 and 2.4 suggest that not all rela-
tionship types need to be associated with object sets: all the one-to-one,
The
 Quest for Data Adequacy and Simplicity 43

one-to-many and many-to-one ones may be modeled instead as properties

of some object sets, as they are functional (i.e., they have at least an “one”
part); only the many-to-many (which are non-functional) ones need add-
ing to the corresponding data models such object sets.
For example, if we are to store all marriages we need the many-to-
many relationship type object set from Fig. 2.1 (as almost anybody may
get married several times during their lifetimes).
Dually, for example, we could also use a set of pairs <city, country>
for storing the countries where cities are located, according to Fig. 2.7,
although this is a many-to-one relationship (any city belonging to only one
country and each country having, generally, several cities):
However, it is simpler and optimal to consider a Country property of
CITIES instead, like in Fig. 2.8.
Obviously, Country should store corresponding values of any property
of COUNTRIES that uniquely identifies corresponding countries; as we
will see in the next chapter, surrogate keys are always the best optimal
choice for it.
Generally, the Key Propagation Principle (KPP) states, at this concep-
tual data modeling level, that functional relationship types may be mod-
eled as object set properties and that it is better to model them like that,
instead of modeling them as object sets. We will deal in details with KPP’s
implementation in the next chapter.
Applying KPP for all functional relationship types is guarantee-
ing minimality of your data models in terms of the total number of

FIGURE 2.7  Functional relationship CITIES_LOCATIONS modeled as a relationship-

type object set.

FIGURE 2.8  CITIES_LOCATIONS modeled as a functional relationship Country.

44 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

needed object sets and mappings, one of the main measures of model
complexity.
From this point of view too, but not only, please note that the cardi-
nalities of relationship types are not absolute, but highly depending on the
context.
For example, if we are only interested in current marriages (and not in
their history too) for a subuniverse where only one marriage is allowed at
a time, then MARRIAGES from Fig. 2.1 is not anymore a many-to-many,
but a one-to-one relationship type.
In such cases, obviously, by applying KPP, we should model MAR-
RIAGES rather as the one-to-one Spouse, like in Fig. 2.9 (which is an
example of a so-called recursive relationship18, which is any functional
relationship between a set and itself); please also note that, in such con-
texts, MarriageDate is not anymore a property of MARRIAGES, but of
PEOPLE.

FIGURE 2.9  An example of a recursive relationship.

2.5  RELATIONSHIP HIERARCHIES

Hierarchies are celestial. In hell all are equal.

—Nicolas Gomez Davila

E-RDM was introduced with very many simplifying restrictions. In an

attempt to make it THE only needed conceptual data model, hundreds of
extensions were later added to it. From our point of view, however, only
two of its restrictions are not justified, either theoretically or practically,
namely:

Mathematically, recursive relationships are autofunctions (functions having same domain and codo-
18

main).
The
 Quest for Data Adequacy and Simplicity 45

• relationship-type sets should be based only on entity-type sets;

• there are only four types of business rules that apply for E-RDs
data instances (namely the RDM ones: uniqueness, attribute value
ranges, their existence, and tuple constraints). Moreover, the orig-
inal E-RDM did not propose any formalism for embedding busi-
ness rules; later extensions embedded primary keys and existence
constraints.
In what follows, we consider examples of our E-RDM main exten-
sions: relationship hierarchies and restriction sets.
For example, consider first any high-school, college, or university sub-
universe. Figure 2.10 presents the best possible data modeling solution for
students’ attendance to classes and their grades, which makes use of a four
level hierarchy of both relationship and entity type object sets.

FIGURE 2.10  An example of a relationship and entity type object sets hierarchy.

Please first note that all relations between a relationship-type object

and any of its underlying object sets need to be labeled by their corre-
sponding roles (the “role” of that object set in that relationship-type object
set)19; in this case, they are Student and Class for ATTENDANCES, Room
and Competence for SCHEDULES, and Teacher and Discipline for COM-
PETENCES. All role names of a relationship-type set should be unique not
19
Formally, roles are the names of the corresponding canonical Cartesian projections.
46 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

only between them: moreover, there may not be an attribute of a relation-

ship-type set having a same name as one of its roles.
Then, note that Schedule stores for each class an element of the SCHED-
ULES relationship-type set, which, in its turn, is built upon the entity-type
set ROOMS and the relationship-type set COMPETENCES. Please also
note that, in order to graphically distinguish between relationships and
their underlying sets in such cases, the normal line for the corresponding
role (here Competence) has to be replaced by an arrow pointing from the
higher level relationship-type set (here SCHEDULES) to its corresponding
underlying relationship-type set (here COMPETENCES).
In this solution, COMPETENCES stores teachers competency for
various disciplines; as SCHEDULES is defined on ROOMS and COM-
PETENCES, it is impossible to schedule classes for teachers that are not
competent for the corresponding disciplines; next, CLASSES are the actual
daily instances of SCHEDULES, factorizing out dates (note that if it were
missing, dates would have to be stored for each corresponding student
attendance); finally, ATTENDANCES takes full advantage of this four
levels hierarchy, only implicitly storing student attendances (to classes
that were scheduled in rooms according to teachers competences in cor-
responding disciplines) and explicitly only the grades they got (if any).
A poorer solution would alternatively use only one flat relationship
STUDENTS_ACTIVITIES of arity 4 (see Fig. 2.11), overloaded with all of
the four semantics above (those of CLASSES and of the three relationships
COMPETENCES, SCHEDULES, and ATTENDANCES), which would

FIGURE 2.11  Alternative flat solution instead of using relationship and entity types
object sets hierarchy.
The
 Quest for Data Adequacy and Simplicity 47

obviously need many additional restrictions in order to reject implausible

instances.
This solution seems to be simpler (as it only needs 5 object sets instead
of 8), but, in fact, it is much more complicated, as it needs the following
additional associated restrictions to be enforced, for any element of STU-
DENTS_ACTIVITIES:
• Teachers should teach corresponding disciplines.
• The disciplines should be scheduled with the corresponding teach-
ers in the corresponding rooms in order for students to be able to
attend corresponding classes.
Without enforcing them, implausible data might be entered; for exam-
ple, teachers of music only could be stored as teaching and giving marks
also for algebra, biology, and physics.
Moreover, disciplines, teachers, rooms, week day, start and end hours,
as well as dates are stored (also meaning that they have to be entered) lots
of times redundantly: for example, assuming that there are 10 teachers,
teaching a total of 16 disciplines, during a 30 weeks year, 5 days a week,
6 h daily, for some 100 students, even if we simplify assuming that each
discipline is taught only once every week for 2 h each, all seven above
cells would need to be filled with data 16 × 30 × 5 × 3 × 100 = 720,000
times each, that is a grand total of 720,000 × 7 = 5,040,000.
For the solution in Fig. 2.7 one would only need to fill 16 lines in COM-
PETENCES (i.e., for teachers and disciplines: 16 × 2 = 32 cells), 3 × 5 = 15
lines in SCHEDULES (i.e., for competences, rooms, weekdays, start and
end hours: 15 × 5 = 75 cells), 30 × 15 = 450 lines in CLASSES (i.e., for
dates and schedules: 2 × 450 = 900 cells), and 450 × 100 = 45,000 lines in
ATTENDANCES (i.e., for students and classes: 2 × 45,000 = 90,000 cells),
that is a grand total of only 32 + 75 + 900 + 90,000 = 91,007, that is only
some 1.8% of the 5,040,000 for STUDENTS_ACTIVITIES!
Even up to 20 years ago, dbs had to be more carefully designed as lim-
ited disk and memory available (both in terms of size, price, and speed)
did not allow for such a waste; never think please that, as disk and mem-
ory nowadays are both affordable and much faster, the flat solution is an
option: always think that you should never ask your end-users to work
more than 50 times more (and, even worse, for repetitive redundant tasks)!
48 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

Otherwise, rather sooner than later, you will lose them, as competition in
this field too is merciless.
Consequently, abstract relationship hierarchies whenever this is pos-
sible in the considered subuniverse of discourse. Please note that this pro-
cess is, essentially, just another type of factorization of data as much as
possible, so that each data is placed on the highest possible generalization
level, in order to avoid storing any duplicates.

2.6  HIGHER ARITY NON-FUNCTIONAL RELATIONSHIPS

The depth of any story is proportionate to the protagonist’s commitment

to their goal, the complexity of the problem, and the grace of the solution.
—Steve House

The above example should not mislead us: properly defined, without
semantic overloading, relationships having arity greater than two may be
successfully used too. For example, in order to store various options for
flying between pairs of airports, the following ternary relationship might
be used, like in Fig. 2.12.
However, we will see in the second volume that (E)MDM includes
both a theorem proving that the only relationships fundamentally needed
for data modeling are the functions (i.e., the arrows in E-RDM), which
means that we could never use relationship-type sets, as well as counter-
examples proving that, generally, it is not a good idea to use relationships
of arity greater than two.

FIGURE 2.12  An example of a ternary relationship.

The
 Quest for Data Adequacy and Simplicity 49

Consequently, we would recommend only usage of binary non-func-

tional (i.e., many-to-many) relationship-type sets, just to draw the atten-
tion on their non-functionality.

2.7  RESTRICTION SETS

The golden rule for every business man is this:

“Put yourself in your customer’s place”.
—Orison Swett Marden

All business rules of any data subuniverse to be modeled, be them explicit

or implicit, should be discovered and added to our data models, in order to
be able to guarantee that only plausible data is stored in the corresponding
dbs.
Obviously, some of them are embedded in E-RDs; for example, any
attribute attached to an object set implies that each element of that object
set may have at most one value for that attribute (in Fig. 2.12, for exam-
ple, each airline has only one name); similarly, each arrow implies that to
each element of the object set from which the arrow starts corresponds
at most one element from the one to where the arrow ends (in Fig. 2.5,
for example, each person has only one home city); moreover, a backward
arrow implies that the dual functional property holds too (in Fig. 2.12, for
example, the backward arrow that links Airline to AIRLINES means that
there may not be several airlines having same name), etc.
Note that we are not even representing compulsory type restrictions
on E-RDs20 as, anyhow, it is either impossible to graphically represent
all restriction types or doing it even only partially complicates them too
much: it is true that a good picture is worth hundreds of words, but it is
also true that, as the number of objects in a picture increases, it is becom-
ing increasingly uninteresting.
Moreover, recall that our customers should agree with our E-RDs, so
let’s keep them simple, only presenting the structure of the corresponding
subuniverses, the results of our data abstraction and aggregation efforts.

20
For example, Barker (Oracle) notation uses dotted half-lines for indicating that no element might
correspond to the corresponding object set.
50 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

Consequently, we prefer attaching restriction sets to every E-RD;

another argument would be that, anyhow, restrictions’ not only formaliza-
tions, but especially informal descriptions should be managed apart, in
order to fully understand and agree with our customers on them too.
There are always explicit business rules that we have to discover. In
order to do it, it is simpler to first search for those of the types that are
provided by the vast majority of the available commercial and open source
DBMSs, which are the following three:
• Attribute ranges (or allowable/permitted values)
• Compulsory attributes (data) (or mandatory values, or not null values)
• Elements (objects) uniqueness
Next, we should add a fourth type for other business rule types, when-
ever needed.
As “business” sounds inappropriate in noncommercial contexts and as
“constraint” is not that prized by most of our customers, I’ve tried several
other words for “business rule” in almost 40 years of experience, out of
which “restriction” was the most widely accepted.
Notationally, it is very convenient to uniquely label all restrictions
associated to the E-RDs of a db (for ease of subsequent references to them)
and to group them by object set and then by their type.

2.7.1  RANGE RESTRICTIONS

Wisdom begins with sacrifice of immediate pleasures

for long-range purposes.
—Louis Finkelstein

The most trivial restriction type is the attributes’ ranges: what should be,
for example, the plausible values for the attributes in Fig. 2.10?
For some of them, the answer is obvious to most humans; for example,
Weekday should probably take values from {1, 2, 3, 4, 5}, as there are no
classes either on Saturdays or Sundays. Note that, even for these attributes,
their ranges are not absolute, but relative to the corresponding context:
for example, under the communism, in all east European countries almost
everybody was working Saturdays too, including students; probably today
The
 Quest for Data Adequacy and Simplicity 51

too there are still countries where students have classes on Saturdays (or
even Sundays).
Similar considerations apply for StartH and EndH (most probably the
former always ranges from 8 to 19 and the latter from 9 to 20).
Even for text-type attributes we should indicate what is the longest
possible corresponding text string; for example, how long could be a per-
son (be it student or teacher) or discipline name? Obviously, it is not plau-
sible that they might have, for example, 4 billion characters each.
Similarly, it is trivial that Date should not take values between, for exam-
ple, January 1st, 4712 BC and December 31st, 9999 (as Oracle is letting us
to store), but only between something like October 1st 2010 and today.
For other attributes, the answer is not that obvious; for example, should
Room# take natural or text values? Probably, as at least some schools are
labeling rooms by using letters too, in order to accommodate them too (for
having as many customers as possible), we would decide to restrict them
to text strings (e.g., of at most 8 or 16 ASCII characters), even if numbers
are stored and processed more conveniently by computers.
Similar considerations apply for Grade (in some countries varying
from F to A+, in others from 1 to 7, or to 10, or to 20 etc.).
The nine digits social security number (SSN) is used for uniquely iden-
tifying U.S. citizens and residents; in Romania, for example, its equivalent
(called personal numeric code), has 13 digits (and embeds birthdate and
sex); other countries have other types of equivalent codes that include let-
ters too (e.g., Greece, Italy, Hong Kong, etc.), while others (e.g., U.K.) do
not have such codes.
Last, but not least, we should always also specify the maximum car-
dinality of each object set. For example, corresponding values might be
1,000 for DISCIPLINES, TEACHERS and ROOMS, 10,000 for COMPE-
TENCES and CLASSES, 100,000 for STUDENTS and SCHEDULES, and
1,000,000,000 for ATTENDANCES.
The restriction on the maximum cardinality of the corresponding set
is the range of its surrogate key: for example, the range of ID_Discipline
would be [1; 1,000].
Specifying maximum cardinality restrictions should be done too, not
only for later implementation considerations, but also for having the most
precise idea possible on the complexity of the considered subuniverse.
52 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

Range restrictions are the basic ones for guaranteeing data plausibility.
Consequently, for any attribute (ellipsis) of any E-RD you should add a
corresponding range restriction.
Unfortunately, there are still very many of those believing that it is
enough to specify as ranges data types: numbers (generally divided in
implementation subtypes: naturals between 0 and 256, integers between
–2,147,483,648 and 2,147,483,647, etc.), calendar dates, etc. Obviously,
this is not generally guaranteeing data plausibility (except for rare cases,
like the Boolean data type): for example, StockQty should never be nega-
tive and should always be less than a maximum value that depends on the
context, but which is never that large as 2,147,483,647 (also see Altitude
from Table 1.1).
To conclude with, to any attribute of any E-RD you should add a sen-
sible range restriction, such that its values are plausible in the correspond-
ing context.
Notationally, each range restriction may be represented by the name
of the corresponding attribute, followed by a column (or by the keywords
“in” or “[“), the corresponding range set, and the (unique) restriction
label (possibly enclosed in parentheses) (see R02, R03, etc.).
Distinction between integer and noninteger ranges should be done by
corresponding minimum and maximum values: for example, [1, 5] is a
subset of the naturals, while [1.0, 5.0] is one of the reals (rationals only
for computers).
ASCII(n), UNICODE(n), etc. represent the sets of strings made out of
characters from the corresponding alphabets and having at most n (natu-
ral) such characters.
Similarly, NAT(n) represents the subset of naturals having at most n
digits, INT(n) the corresponding one of integers, RAT(n.m) the one of
rationals having at most n digits before the decimal dot (or comma in
Europe) and m after it, and RAT+(n.m) its subset of positive values (obvi-
ously, NAT(n) = INT+(n)).
CURRENCY(n) is a subset of RAT+(n.2).
For expressing the particular case of cardinality constraints, we may
use the standard algebraic mappings card, which returns the number of
elements in a set and max, which computes the maximum value of an
expression (see R01, R04, R07, etc.).
The
 Quest for Data Adequacy and Simplicity 53

For example, the following might be (for Romania) the range restric-
tions associated to the E-RD in Fig. 2.10 (where SysDate() is a system
function returning current OS calendar date):

STUDENTS
max(card(STUDENTS)) = 105 (R01)
SSN: [1000101000000, 8991231999999] (R02)
Name: ASCII(255) (R03)
TEACHERS
max(card(TEACHERS)) = 103 (R04)
SSN: [1000101000000, 8991231999999] (R05)
Name: ASCII(255) (R06)
DISCIPLINES
max(card(DISCIPLINES)) = 103 (R07)
Discipline: ASCII(128) (R08)
ROOMS
max(card(ROOMS)) = 103 (R09)
Room#: [1, 104] (R10)
CLASSES
max(card(CLASSES)) = 104 (R11)
Date: [01/10/2010, SysDate()] (R12)
SCHEDULES
max(card(SCHEDULES)) = 105 (R13)
Weekday: [1, 5] (R14)
StartH: [8, 19] (R15)
EndH: [9, 20] (R16)
ATTENDANCES
max(card(ATTENDANCES)) = 109 (R17)
Grade: [1, 10] (R18)
COMPETENCES
max(card(COMPETENCES)) = 104 (R19)

Obviously, as already discussed, the above range restrictions might

have different minimum and maximum values and/or even different
data types for other countries. Of course that if you want to deliver a
universal solution, you should choose the more general data types (e.g.,
54 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

ASCII(2) for Grade) and the corresponding widest intervals (e.g., [1, 7]
for Weekday).
Please note that, generally, besides choosing the right data type set for
every attribute, it is crucial that all of your range restrictions specify as
narrow as possible intervals (explicitly or implicitly of the type [minValue,
maxValue]), such as to actually guarantee that only plausible data will be
accepted for all attributes.
As a counterexample, many RDBMSs (including DB2, MS SQL
Server, Access, and MySQL) chose for the DATE data type ranges starting
in our era and ending on Dec. 31st, 9999 (most probably dreaming that
their products will last unchanged, from this point of view, at least another
7,986 years from now); obviously, we generally need no more than 100
years in the future, but, whenever confronted with mankind history data,
we badly need dates before our era, which are very hard to store and pro-
cess with these RDBMSs.
For example, the famous battle of Actium took place on September
2, 31 BC; the almost equally one of Kadesh, in May 1274 BC; the old-
est recorded one (on the walls of the Hall of Annals from the Temple of
Amun-Re in Karnak, Thebes—today’s Luxor) is the one of Megiddo,
which took place on April 16, 1457 BC (or, after other calculations, 1479
BC or even 1482 BC), etc.
Consequently, a much more intelligent choice for DATE would have
been something like [1/1/–7000, 12/31/2999], as, almost surely, 985 years
from now on are enough from any practical point of view, and, for exam-
ple, the oldest intact bows discovered (in Holmegaard, Denmark) have
some 8000 years, the Goseck Circle astronomic observatory discovered
in Germany has some 7000 years, the oldest discovered (near Ljubljiana,
Slovenia) wheel has some 5200 years, etc.

2.7.2  COMPULSORY DATA RESTRICTIONS

Learning is not compulsory … neither is survival.

—W. Edwards Deming
The
 Quest for Data Adequacy and Simplicity 55

Sometimes we either do not temporarily know the values of some attri-

butes or they do not apply to some objects. Dually, it is obviously useless
to store no data on any object, be it only temporarily.
Consequently, for each object set there should be at least one property
whose values should be known for all of its objects. For example, taking
once more into consideration Fig. 2.10, as we cannot even refer to a dis-
cipline without knowing its name, Discipline’s values should be compul-
sory; similarly, all other attributes, except for Grade, should have all their
values known: for example, how could both students and teachers attend
classes without knowing where (Room#), when (WeekDay, StartH), and
for how long (EndH) to go and remain for?
Grade is an example of attribute whose values should not be manda-
tory, as it is not applicable for all elements of ATTENDANCES: there may
be (lot of) students that do not get any grade during (many) classes.
Similarly, if, for example, we were to add BirthDate to both STU-
DENTS and TEACHERS, in most contexts we should not add correspond-
ing compulsory data restrictions, as temporarily we might not know some
of their values.
Arrows (i.e., many-to-one, one-to-many, and one-to-one relationships)
also should or should not be compulsory, depending on the context.
For example, Schedule from Fig. 2.10 should be mandatory, as no class
may be scheduled at no matter what date without knowing exactly in what
room, starting with what hour, up to what end hour, etc.
All roles should be compulsory too, because no relationship may be
defined without knowing all of its components.
For example, both Student and Class must be mandatory, as otherwise
we might not know what classes attended each students or/and what stu-
dents attended each class.
Similarly, Room, Competence, Teacher, and Discipline should be com-
pulsory too.
Notationally, each compulsory data restriction may be represented by
the name of the corresponding object set, followed by a column, and then
the corresponding property (attribute, arrow, role) name (see R22 and
R23 bellow); for simplicity, all such restrictions for a same object set may
be merged into a single compound one, by factoring out the name of the
56 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

object set and separating the names of the properties by columns (see R20,
R21, R24, R25, R26, and R27).
For example, the following might be the compulsory data restrictions
associated to the E-RD in Fig. 2.10.

STUDENTS: SSN, Name (R20)

TEACHERS: SSN, Name (R21)
DISCIPLINES: Discipline (R22)
ROOMS: Room# (R23)
CLASSES: Date, Schedule (R24)
SCHEDULES: Weekday, StartH, EndH, Room, Competence (R25)
ATTENDANCES: Student, Class (R26)
COMPETENCES: Teacher, Discipline (R27)

2.7.3  UNIQUENESS RESTRICTIONS

Always remember that you are absolutely unique.

Just like everyone else.
—Margaret Mead

Just like for (algebraic) sets, db object sets do not allow for duplicates: it
does not make sense to store data on objects that are not uniquely identifi-
able.
For example, in the subuniverse modeled by the E-RD from Fig. 2.10,
we should be able to uniquely distinguish between any two students,
teachers, disciplines, rooms, etc.
Any time when we do not need to uniquely distinguish between ele-
ments of a same set of objects (like, for example, two kilos of sugar that
we are buying) for which we would like to store data, we should abstract
a corresponding higher conceptual level set instead (e.g., PRODUCT_
TYPES, having “sugar” as one of its elements and a Quantity property,
besides the ProductTypeName that should be unique).
Uniqueness is a constraint that applies either on properties or on sets of
properties. For example, by definition, SSN values are unique, but in order
to uniquely distinguish between world’s states we need both their names
The
 Quest for Data Adequacy and Simplicity 57

and the corresponding country (as there may not be states having same
names in a same country, but, worldwide, there may be states of different
countries having same names: for example, both Spain and Argentina have
states called Cordoba and La Rioja).
Object sets may have several uniqueness constraints; for example, as
in no room may simultaneously either start or end more than one class,
SCHEDULES above has two such triple constraints: one built with Room,
Weekday and StartH, and the other (its dual one) with Room, Weekday and
EndH.
At this conceptual level, however, in order to be sure that object sets
are correctly defined (i.e., from this point of view, they do not allow for
duplicates), it is enough to add to any of them at least one uniqueness con-
straint; of course that anytime we discover even at this stage several ones,
we should add them all to the corresponding uniqueness restriction set.
There are only a couple of exception types among object sets that might
not have associated uniqueness restrictions: subsets, people genealogy-
type sets, and poultry/rabbit/etc. cages type sets.
For example, the subset DRIVERS of EMPLOYEES (that we might
abstract in order to store only for them the otherwise nonapplicable prop-
erties LicenseType, LicenseDate, etc.) might not have any uniqueness
restriction. For subsets, unique identification may be done through the
uniqueness restrictions of the corresponding supersets: all subsets of any
set inherit all of its restrictions.
PEOPLE in a genealogical db might not have uniqueness restrictions
either: for example, there may be several persons having same names on
which we do not know anything else (but their names). For such sets, the
only way to uniquely distinguish between their elements is by the values
of their surrogate keys.
Poultry/rabbit/etc. CAGES in a (farming) db might not need any other
uniqueness than the one provided by surrogate keys: you can use their
system generated numbers as cage labels too (surrogate keys thus getting
semantics too).
Notationally, we may simply specify uniqueness restrictions just like
the compulsory data ones; however, as they are not that obvious, we
should add the corresponding reason in parenthesis after each of them.
58 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

Compound uniqueness restrictions are specified by using the ‘•’ separa-

tor21 between the corresponding property names.
For example, the following are the uniqueness restrictions associated
to the E-RD in Fig. 2.10.

STUDENTS: SSN (there may not be two students having same SSN) (R28)
TEACHERS: SSN (there may not be two teachers having same SSN) (R29)
DISCIPLINES: Discipline (there may not be two disciplines having
same name) (R30)
ROOMS: Room# (there may not be two rooms having same number) (R31)
CLASSES: Date • Schedule (there may not be two classes scheduled at
the same date) (R32)
SCHEDULES:
Room • Weekday • StartH (in no room may simultaneously start more

than one class) (R33)
Room • Weekday • EndH (in no room may simultaneously end more

than one class) (R34)
ATTENDANCES: Student • Class (there is no use to store more than once
the fact that a student attended a class) (R35)
COMPETENCES: Teacher • Discipline (there is no use to store more
than once the fact that a teacher has the competence to teach a discipline) (R36)

2.7.4. OTHER TYPES OF RESTRICTIONS

Progress imposes not only new possibilities for the future,

but new restrictions.
—Norbert Wiener

Even a very simple subuniverse, like, for example, the one in Fig. 2.10
may have other type of business rules than the three above that are gov-
erning it. Some of them are of common sense types (like, for example,
21
Mathematically, ‘•’ is the Cartesian product operator on mappings.
The
 Quest for Data Adequacy and Simplicity 59

nothing may end before it starts, nobody can simultaneously be present in

different places, no resource that can accommodate only one object at a
time may do it simultaneously for several ones, etc.), others may be appli-
cable strictly to that particular subuniverse (e.g., in order to be eligible,
anybody who would like to run for an office should have a minimum age:
in Romania, for example, currently they are 23 for local or national repre-
sentatives and Europarlimentaries, 25 for senators, and 35 for presidents).
As not adding any business rule as a restriction to our data models
allow for storing implausible data, each existing business rule must be
always added.
Notationally, we may specify such restrictions at this level even in
plain English (or any other natural language that is understood by our cus-
tomers and by all other implied persons).
Of course that, when appropriate, we may use simple (i.e., proposi-
tional) logic formulas (like, for example, StartH < EndH), but it is a good
practice also in this case to provide an explanation (enclosed in parenthe-
sis) for each such restriction too.
When such a restriction involves properties of several object sets, it is
better to include them at the end of the restriction set or, at least, at the end
of the subset of the corresponding last object set involved (in order not to
reference properties not yet defined).
For example, the following are additional restrictions associated to the
E-RD in Fig. 2.10.

SCHEDULES:
StartH < EndH (no class may end before it starts) (R37)
No teacher may be simultaneously present in
more than one room. (R38)
No student may be simultaneously present in more than
one room. (R39)
No room may simultaneously host more than one class.22 (R40)
There may not be two people (be them teachers or students)
having same SSN.23 (R41)

22
Note that (R40) implies both (R33) and (R34).
23
Note that (R41) implies both (R28) and (R29).
60 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

2.8  CASE STUDY: A PUBLIC LIBRARY (DO WE KNOW EXACTLY

WHAT A BOOK IS?)

A room without books is like a body without a soul.

—Cicero

As the public libraries subuniverse is familiar to almost everyone, the

only description for it in this case is a very simple one: the final db applica-
tion should assist librarians in managing books and subscribers.
Most of my undergraduate, but also very many graduate students will
come up in some 5′ with the entity-type object sets BOOKS (having prop-
erties Title, ISBN, Price, BooksNo), AUTHORS (FName, LName), and
SUBSCRIBERS (FName, LName, e-mail), the relationship-type one BOR-
ROWS (BorrowDate, ReturnDate), and the ER-D in Fig. 2.13.
First of all I’m generally drawing their attention on the fact that among
authors there are lot of still living persons and that at least some of them
might also be subscribers (writers are among the most assiduous readers!);
consequently, each such author should be stored twice in their db, once
as author and a second time as subscriber (without even being possible
to tell whether or not these two object set elements correspond to a same
person). Moreover, by consolidating AUTHORS and SUBSCRIBERS (e.g.,
into PERSONS), their solution becomes not only conceptually better, but
also simpler and faster in terms of implementations.

FIGURE 2.13  Public library first data analysis iteration E-RD.

The
 Quest for Data Adequacy and Simplicity 61

Secondly, I remind them that there are plenty of books (generally

scientific, technologic, but not only) having more than one author: con-
sequently, Author should be replaced by a relationship-type set (e.g., CO-
AUTHORS); moreover, for retrievals (but also for accurately modeling
this subuniverse), their positions in the coauthor lists should also be stored.
Next, for finishing with “minor” issues, I remind them too that return
dates are generally not the same for all books that you borrow together
and that actual return dates should also be stored for each book, first of all
for keeping track of still not returned, although overdue ones, but not only
(e.g., tracking both best subscribers, those that are never late, and less or
much less accountable ones, as they are frequently returning books late,
very late, etc.).
Finally, I’m closing this first round of comments and suggestions by
asking them the following questions (for which, unfortunately, I’m not
getting correct, generally even satisfactory answers):
• What is, for example, the ISBN of Shakespeare’s King Lear?
• What’s the price of Homer’s Odyssey?
• Is the books number (by which they mean number of copies of a
same book from the library) computable or not?
• How exactly do they ask for a book in a public library (do they
generally know its ISBN)?
• What do they exactly mean by “book”?
Here are then my answers:
• When written, no book has associated ISBN; moreover, generally
there are several editions of a book (even by a same publisher!),
each one having different ISBNs.
• Like any other masterpiece, Homer’s Odyssey is priceless! Even
ordinary books do not have prices when written: only their editions
have prices.
• The number of copies of an edition, as well as the total one for all
editions of any book is computable, so it may be eliminated from
the data model.
• Except for insignificant number of cases, subscribers (and even
librarians) do not know ISBNs: generally, you ask for a book by
title, authors, publishing house, and edition year.
62 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

• They mean at least two things by “book”: written, conceptual works,

on one hand and their (printed or electronic) copies on the other.
Consequently, they are replacing the previous E-RD with the much
better one in Fig. 2.14.
When trying then to establish uniqueness for BOOKS, another issue
arises, as there may be several books having same title (e.g., “Logic”,
“Algebra”, “Euclidean Geometry”, “Poems”, etc.) and I’m suggesting
them to add a FirstAuthor functional relationship (and to consequently
store coauthors, if any, only starting with the second position in coauthor
lists); as there are, however, few exceptions of books having same first
author and title, I’m also suggesting them to also add the year of book
finishing (which is also interesting per se).
Next, focusing on their meaning for COPIES, I’m asking them why
are they breaching first db axiom (“any db should store any fundamental
data only once”) with both ISBN and Price, as there are, generally, several
identical copies of any edition; moreover, which is worse, I’m asking them
how about several editions of a book (by different publishers, but even by
a same one) that have different ISBNs? The answer is that COPIES is still
overloaded with several semantics, so we identify together the entity-type
object set EDITIONS that is also needed. Note that, besides publisher, title,
and year, as these three properties do not uniquely identify editions, simi-
lar to BOOKS, EDITIONS should have an additional functional relation-
ship FirstBook (as no publisher would publish more than one edition per
year of a book under a same edition title).

FIGURE 2.14  Public library second data analysis iteration E-RD.

The
 Quest for Data Adequacy and Simplicity 63

In order to skip redrawing the E-RDs too many times, I’m then also
drawing their attention to other two relevant facts, proving that, if left like
that, EDITIONS is also overloaded semantically:
• Often (possibly regardless of the initial (co)author(s) book vol-
umes structuring, which is of no great interest to libraries), editions
are split into several volumes, each with their own distinct ISBN
and possibly having their own title.
• Sometimes, even if not that often, volumes may contain several books
(e.g., “Complete Shakespeare’s Dramatic and Poetical Works”).
This proves that their initial BOOKS object set was overloaded with the
semantics of five object sets, namely: BOOKS, VOLUMES_CONTENTS,
VOLUMES, EDITIONS, and COPIES.
Finally, I’m also suggesting them to add a Boolean Author? to PEO-
PLE, which will prove very useful for both users and db application
developers.

2.8.1  PUBLIC LIBRARY E-RD

I have always imagined that Paradise will be a kind of library.

—Jorge Luis Borges

The final corresponding E-RD is presented in Fig. 2.15, immediately

followed by its associated restriction set and by its full description.

2.8.2  PUBLIC LIBRARY RESTRICTION SET

Never trust anyone who has not brought a book with them.
—Lemony Snicket

1. PERSONS (The set of book (co)authors and subscribers

of interest to the library.)
a. Cardinality: max(card(PERSONS)) = 109 (RP0)
b. Data ranges:
FName: ASCII(128) (RP1)
LName: ASCII(64) (RP2)
64 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

FIGURE 2.15  Public library E-RD.

Author?: BOOLEAN (RP3)

e-mail: ASCII(255) (RP4)
c. Compulsory data: LName (RP5)
d. Uniqueness: e-mail • FName • LName (there may not be
two persons having same first and last name, as well as
same e-mail) (RP6)
e. Other type restrictions:
FName and e-mail should be compulsory for
subscribers. (RP7)

2. BOOKS (The set of written works of interest to the library.)

a. Cardinality: max(card(BOOKS)) = 106 (RB0)
b. Data ranges:
BTitle: ASCII(255) (RB1)
BYear: [–5000, current year] (RB2)
The
 Quest for Data Adequacy and Simplicity 65

c. Compulsory data: BTitle, FirstAuthor (RB3)

d. Uniqueness: FirstAuthor • BTitle • BYear (no author
writes two books with a same title in a same year) (RB4)

3. CO-AUTHORS = (PERSONS, BOOKS) (The set of

pairs <p, b> storing the fact that book b was (also)
written by person p.)
a. Cardinality: max(card(CO-AUTHORS)) = 2 × 106 (RCA0)
b. Data ranges: PosInList: [2, 16] (p’s position in b’s
co-authors list) (RCA1)
c. Compulsory data: Co-author, Book, PosInList (RCA2)
d. Uniqueness: Co-author • Book (no author should appear
more than once in any book co-authors list) (RCA3)

4. PUBLISHERS (The set of publishers of interest to

the library.)
a. Cardinality: max(card(PUBLISHERS)) = 104 (RPB0)
b. Data ranges: PubName: ASCII(128) (RPB1)
c. Compulsory data: PubName (RPB2)
d. Uniqueness: PubName (there may not be two publishers
having same name) (RPB3)

5. EDITIONS (The set of book editions of interest to

the library.)
a. Cardinality: max(card(EDITIONS)) = 106 (RE0)
b. Data ranges:
ETitle: ASCII(255) (RE1)
EYear: [–2500, current year] (RE2)
c. Compulsory data: Publisher, FirstBook, ETitle (RE3)
d. Uniqueness: Publisher • FirstBook • EYear (no publisher
publishes more than one edition of any given book in any
given year) (RE4)

6. VOLUMES (The set of edition volumes of interest to

the library.)
a. Cardinality: max(card(VOLUMES)) = 2 × 106 (RV0)
66 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

b. Data ranges:
ISBN: ASCII(16) (RV1)
VTitle: ASCII(255) (RV2)
VNo: [1, 255] (RV3)
Price: [0, 100000] (RV4)
c. Compulsory data: Edition, VNo, Price (RV5)
d. Uniqueness: ISBN (by definition, ISBNs are unique for
any volume) (RV6)

7. VOLUME_CONTENTS = (BOOKS, VOLUMES) (The

set of pairs <b, v> storing the fact that book b is
included in volume v.)
a. Cardinality: max(card(VOLUME_CONTENTS))
= 4 × 106 (RVC0)
b. Data ranges: BookPos: [1, 16] (b’s position in v’s table
of contents) (RVC1)
c. Compulsory data: Book, Volume, BookPos (RVC2)
d. Uniqueness: Volume • Book (no book should be included
more than once in any volume) (RVC3)
e. Other type restrictions:
For any edition, its first book should be the first one
published in its first volume. (RVC4)
No edition may contain same book more than once. (RVC5)

8. COPIES (The set of volume copies that the library ever

possessed.)
a. Cardinality: max(card(COPIES)) = 32 × 106 (RC0)
b. Data ranges: InvNo: ASCII(32) (RC1)
c. Compulsory data: InvNo, Volume (RC2)
d. Uniqueness: InvNo: (by definition, inventory numbers
are unique for any copy) (RC3)

9. BORROWS (The set of copy borrows by subscribers.)

a. Cardinality: max(card(BORROWS)) = 1011 (RBR0)
The
 Quest for Data Adequacy and Simplicity 67

b. Data ranges: BorrowDate: [1/6/2000, SysDate()]

(assuming, for example, that first borrow date of interest
is Jan. 6th, 2000) (RBR1)
c. Compulsory data: BorrowDate, Subscriber (RBR2)
d. Uniqueness: BorrowDate • Subscriber
(no subscriber may simultaneously borrow more
than once) (RBR3)

10. BORROW_LISTS = (BORROWS, COPIES) (The set of

pairs <b, c> storing the fact that volume copy c was
(also) borrowed in borrow b.)
a. Cardinality: max(card(BORROW_ LISTS)) = 1012 (RBL0)
b. Data ranges:
DueReturnDate: [1/6/2000, SysDate() + 300]
(assuming, for example, that maximum borrow
period is 300 days) (RBL1)
ActualReturnDate: [1/6/2000, SysDate()] (RBL2)
c. Compulsory data: Borrow, Copy, DueReturnDate (RBL3)
d. Uniqueness: Borrow • Copy (no copy may be
simultaneously borrowed more than once) (RBL4)
e. Other type restrictions:
No copy may be borrowed less than 0 days or
more than 300 days. (RBL5)
No copy may be simultaneously borrowed to
more than one subscriber. (RBL6)
No copy may be returned before it was
borrowed or after 100 years since corresponding
borrow date. (RBL7)

2.8.3  PUBLIC LIBRARY CLEAR, CONCISE AND

COMPREHENSIVE DESCRIPTION

It is what you read when you don’t have to

that determines what you will be when you can’t help it.
—Oscar Wilde
68 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

The db should store data on books (title, writing year, first author,
coauthors, and their order), people (authors and/or library subscrib-
ers) e-mail addresses, first and last names, whether they are authors or
not, as well as (book copies) borrows by subscribers (borrow, due, and
actual return dates).
Books are published by publishing houses, possibly in several
editions, even by same publishers (in different years). Each edition
may contain several volumes, each having a number, a title, a price,
and an ISBN code. Volumes may contain several books.
The library owns several copies (uniquely identified by an inven-
tory code) of any volume and may lend several copies for any bor-
row; not all borrowed copies should be returned at a same date;
maximum lending period is 300 days.
Last names, book titles, first authors, publisher names, edition pub-
lishers, first books and titles, as well as volume numbers and prices,
copy inventory codes, borrow subscribers, dates, and due return dates
are compulsory; for subscribers, first names and e-mail addresses are
compulsory too.
People are uniquely identified by their first and last names, plus
e-mail address; books by their first author, title, and writing year;
publishers by their names; editions by their first book, publisher,
title, and year; volumes by corresponding edition and volume num-
ber, as well as by their ISBN; copies by their inventory number.
There may be at most 1,000,000 persons, books, and editions,
2,000,000 coauthoring and volumes, 10,000 publishers, 4,000,000
books publishing, 32,000,000 copies, 100,000,000,000 borrows, and
1,000,000,000,000 copy borrows of interest. For example, first bor-
row date of interest is 01/06/2000.
No author should appear more than once in any book coauthors list.
For any edition, its first book should be the first one published in
its first volume. No edition may contain same book more than once.
No copy may be borrowed less than 0 days or more than 300 days.
No copy may be simultaneously borrowed to more than one
subscriber.
No copy may be returned before it was borrowed or after 100
years since corresponding borrow date.
The
 Quest for Data Adequacy and Simplicity 69

2.9  THE ALGORITHM FOR ASSISTING THE DATA ANALYSIS AND

MODELING PROCESS (A0): AN E-R DATA MODEL OF THE E-RDM

The secret of getting ahead is getting started. The secret of getting started is
breaking your complex overwhelming tasks into small manageable tasks,
and starting on the first one.
—Mark Twain

The intuition gained with the above case study should make easier the
task of always applying first the algorithm for assisting data analysis and
conceptual modeling processes (A0) presented in Fig. 2.1624.

Algorithm A0 (Data analysis and conceptual modeling process

assistance)
Input: an informal description of a subuniverse of interest (provided
by the customer, team leader, professor, etc.);
Output: a corresponding initial E-R data model;
Strategy: done = false;
while not done repeat
rewrite input description in order to make it crystal-clear, compre-
hensive, concise, structured, with highlighted keywords (that is all
words susceptible of defining and characterizing object sets, their
properties, as well as the rules governing the corresponding subuni-
verse of discourse; for example, customers, contracts, people, cities,
services, e-mails, quantities, addresses, uniquenesses, etc.);
repeat for all keywords in the above description
select keyword
case: object set
if current E-RD is already too complicated then
start another E-R subdiagram;
if object set is linking two other existing object sets then
add a corresponding diamond in the current E-RD (correspond-
ingly labeled);
else add a rectangle in the current E-RD (correspondingly
labeled);

24
Whose complexity is linear in the sum of object set collection, property and restriction sets cardinals
(see Exercise 2.9)
70 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

if current E-R subdiagram is a new but not the first one then
link the new object set with at least another existing one (from
previous E-R subdiagrams);
case: object set property
if the corresponding object set does not exist in the data model
then
add a corresponding object set like above;
if property is a functional relationship then
add corresponding arrow (correspondingly labeled);
else add corresponding ellipse (correspondingly labeled);
case: business rule
add corresponding restriction to the restriction set (note: some-
times it may be the case that, in order to do it, you should first add
new object sets and/or properties);
end select;
end repeat;
meet again customers and ask them about everything which is still
not clear or missing, if any, or whether they agree with your deliver-
ies;
if everything is clear and you got customers’ OK then done = true;
end while;
End algorithm A0;

FIGURE 2.16  Algorithm A0 (Data analysis and modeling process assistance).

Note that such algorithms are sometimes referred as being of type for-
ward engineering (as opposed to their reverse engineering dual ones—see,
for example, Sections 3.5, 4.4, and 4.6).
By applying A0 to our variant of the E-RDM, the corresponding E-R
data metamodel from Fig. 2.17 is obtained.
The
 Quest for Data Adequacy and Simplicity 71

FIGURE 2.17  E-RD of the E-RDM.

2.9.1  E-RD OF E-RDM

He who thinks less is much more wrong.

—Lucian Blaga
Figure 2.17 presents the corresponding E-RD.

2.9.2  AN EXAMPLE OF POSSIBLE RESTRICTION SET

ASSOCIATED TO E-RDM

After having discovered that life is senseless,

the only thing that remains to be done is to give it a sense.
—Lucian Blaga

The following example uses some arbitrary chosen maximum values,

which may actually differ depending on the context (e.g., possible imple-
mentation).

1. OBJECT_SETS (The finite nonempty collection of object sets of

interest in any subuniverse of discourse, partially ordered by set inclusion)
  a. Cardinality: max(card(OBJECT_SETS)) = 1038 (RO1)
72 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

  b. Data ranges:
Name, Description: ASCII(255) (RO2)
MaxCard: NAT(38) (maximum cardinality) (RO3)
  c. Compulsory: Name, MaxCard, Description
(RO4)
  d. Uniqueness: Name (There may not be two object sets
having same names in a same subuniverse of discourse.) (RO5)
  e. Other type restrictions:
There are no other types of object sets than entity and
relationship ones and any object set is either of entity
or of relationship-type.25 (RO6)

2. ENTITY_TYPES # OBJECT_SETS (The finite

nonempty object sets subcollection of atomic sets)

3. NON_FUNCTIONAL_RELATIONSHIP_TYPES # OBJECT_
SETS (The finite object sets subcollection of Cartesian
product type sets)
*Arity: corresponding Cartesian product arity

4. UNDERLYING_SETS = (RelationshipType → NON_FUNC-

TIONAL_RELATIONSHIP_TYPES, ObjectSet → OBJECT_SETS)
(The finite set of pairs of type <r, o>, storing the fact that the non-
functional relationship-type object set r has object set o as one of
its underlying sets)
  a. Cardinality: max(card(OBJECT_SETS)) = 1038 (RU1)
  b. Data ranges: Role: ASCII(255) (the role of o in r) (RU2)
  c. Compulsory: RelationshipType, ObjectSet, Role (RU3)
  d. Uniqueness: RelationshipType • Role (There may not
be two underlying object sets of a same non-functional
relationship-type object set having same roles.) (RU4)
  e. Other type restrictions:
For each non-functional relationship-type object set
there should be at least two underlying object sets. (RU5)
The graph of UNDERLYING_SETS should be acyclic
(that is no non-functional relationship-type object set
25
Mathematically, OBJECT_SETS = ENTITY_TYPES ! NON_FUNCTIONAL_RELATIONSHIP_
TYPES.
The
 Quest for Data Adequacy and Simplicity 73

may be defined on itself, neither directly, nor indirectly);

equivalently: recursive definitions of non-functional
relationship-type object sets is not allowed. (RU6)

5. MAPPINGS (The finite nonempty set of mappings defined on

object sets of interest)
  a. Cardinality: max(card(MAPPINGS)) = 1038 (RM1)
  b. Data ranges:
Name, Description: ASCII(255) (RM2)
 OneToOne?: Boolean (is mapping uniquely identifying
the object set elements?) (RM3)
 Compulsory?: Boolean (RM4) (should mapping have
known all of its values?) (RM4)
  c. Compulsory: Name, DomainObjectSet, OneToOne?,
Compulsory? (RM5)
  d. Uniqueness: Name • DomainObjectSet (There may
not be two mappings defined on a same object set and
having same names.) (RM6)
  e. Other type restrictions:
There are no other types of mappings than attributes and
functional relationship ones and any mapping is either
of attribute or of functional relationship-type.26 (RM7)

6. ATTRIBUTES # MAPPINGS (The finite nonempty

mapping subset of attributes)
  a. Compulsory: RangeRestriction (RA1)

7. FUNCTIONAL_RELATIONSHIPS # MAPPINGS
(The finite mapping subset of functional relationships)
  a. Compulsory: CodomainObjectSet (RF1)

8. DATA_TYPES (The finite nonempty collection of needed

data type sets, partially ordered by set inclusion)
  a. Cardinality: max(card(DATA_TYPES)) = 100 (RD1)
  b. Data ranges: Name: ASCII(255) (RD2)
26
Mathematically, MAPPINGS = FUNCTIONAL_RELATIONSHIPS ! ATTRIBUTES.
74 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

  c. Compulsory: Name (RD3)

  d. Uniqueness: Name (There may not be two data type sets
having same names in a same subuniverse of discourse.) (RD4)

9. RANGES (The finite nonempty collection of needed attribute

data range sets, partially ordered by set inclusion)
  a. Cardinality: max(card(RANGES)) = 104 (RR1)
  b. Data ranges: MinValue, MaxValue: ASCII(255) (RR2)
  c. Compulsory: DataType, MinValue, MaxValue (RR3)
  d. Uniqueness: DataType • MinValue • MaxValue (There
may not be two data range sets of a same data type and
having same minimum and maximum values.) (RR4)

2.9.3  DESCRIPTION OF E-RDM

Reality is the ruin of a fairy tale.

—Lucian Blaga

There are at most 1038 (for example) object sets of interest in any sub-
universe of discourse; some of them may be included in others; they are
characterized by the compulsory properties name (an ASCII string of max-
imum 255 characters, for example), uniquely identifying each object set in
any given subuniverse of discourse, maximum cardinal (a natural having
at most 38 digits, for example) and a description (an ASCII string of maxi-
mum 255 characters, for example).
Object sets may be either atomic (of the so-called “entity” type) or of
Cartesian product type (the so-called “relationship”).
The Cartesian product type ones obviously have arity greater than
one and are nonrecursively defined on other underlying object sets. For
any such relationship-type set, each canonical Cartesian projection has a
compulsory name (of at most, for example, 255 ASCII characters), unique
within the set and called “role”.
Each subuniverse of interest may also need at least one and at most
100 (for example) so-called (programming) “data types”, which are finite
subsets of the naturals, integers, rationals, strings of characters over some
alphabet (generally ASCII or UNICODE), calendar dates, Boolean truth
values, etc. They are uniquely identified by their compulsory name (an
The
 Quest for Data Adequacy and Simplicity 75

ASCII string of maximum 255 characters, for example) and some of them
may be included in other ones.
There are at most 1038 (for example) mappings of interest in any sub-
universe of discourse defined on object sets; they are characterized by the
compulsory properties name (an ASCII string of maximum 255 characters,
for example), uniquely identifying each mapping within those defined on
the same object set, and a description (an ASCII string of maximum 255
characters, for example); it is also compulsory to specify for any map-
ping whether or not its values are compulsory (or may be sometimes left
unspecified)—the so-called “compulsory restriction”—and are uniquely
identifying the elements of its domain object set (i.e., if the mapping is
one-to-one or not)—the so-called “uniqueness restriction”.
Mappings may take values in either object sets (in which case they
are called “functional relationships”, which are formalizing existing links
between object sets) or data ranges (in which case they are called “attri-
butes”, which are formalizing object set properties).
Attribute codomains (the so-called “range restrictions”) are subsets
(called “ranges”) of data types, uniquely identified by the triple made out
of their compulsory data type superset, minimum and maximum values
(both being, for example, ASCII strings of maximum 255 characters);
some ranges may be included into other ones.
In order to keep things simpler, this E-R data metamodel does not
include restrictions of other types and uniquenesses made out of several
mappings (see Exercise 2.6, as well as Section 3.3).

2.10  BEST PRACTICE RULES

The Golden Rule finds no limits of application in business.

—James Cash Penney

Here is a set of best practice rules in data analysis and conceptual modeling.

2.10.0  NO SEMANTIC OVERLOADING

Information overload is a symptom of our desire

to not focus on what’s important.
—Brian Solis
76 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

R-DA-0. No object set, be it of entity or relationship type, should be

semantically overloaded: any object set should have only one simple and
clear semantics. On the other hand, somewhat dually, obeying to this para-
mount best practice rule should not diminish your power of abstraction.
For example, in a medical db, do not abstract sets of PATIENTS and
DOCTORS, but only one set of PEOPLE (also having a Boolean attribute
Doctor?—you will see why in the second volume): MDs may also get
sick, do have the right to also be patients, and if you were choosing the
former solution you should duplicate data for each such MD, without ever
being able to conclude only from the corresponding db instance that any
two persons from these object sets, even if having exactly same data, do
represent a same actual person.
Similarly, do not abstract sets of SUPPLIERS and CUSTOMERS (why
not starving to make at least some of your suppliers your customers too?),
but a set of PARTNERS instead (and, by the way, for example, banks are
also your partners, so there is no need to add a BANKS object set either:
abstract banks too as partners).

2.10.1  NAMING CONVENTIONS

I named my album today. It was like naming a child for me.

Such a hard decision.
—Carly Rae Jepsen

R-DA-1.
a. All software objects (e.g., dbs, object sets, properties, tables, col-
umns, constraints, applications, classes, variables, methods, librar-
ies, etc.) should be named consistently, adequately, in a uniform
manner, and their names should be the same at all involved levels
(data analysis, E-RDs, restriction sets, mathematical, relational,
and implemented db schemas, SQL, high level programming lan-
guages code, as well as both global/external and local/embedded
in-line documentation).
b. Object set names should be plurals and uppercase; property and
restriction names should be singular and using generally lower-
case, except for acronyms and letters starting words. Examples:
The
 Quest for Data Adequacy and Simplicity 77

PEOPLE, CITIES, COUNTRIES, SSN, FName, LName, Birth-

Place, Capital, CitiesKey.
c. Object set and constraint names should be unique in any data
model, whereas property names should be unique within the set of
all properties of the corresponding object set.
d. Do not ever use the ID suffix or prefix for properties relating object
sets (i.e., functional relationships): not only that, for db people,
it should be trivial that their implementation will reference a sur-
rogate key, whereas for customers implementation details should
remain transparent; moreover, it is absolutely unnatural (when ask-
ing somebody in what city they live, we do not ever ask them in
what cityID or IDcity they do!), but it is also a very bad idea for
any object set for which there are several properties relating them
to a same object set—and not only conceptually, but also relation-
ally (as for any table its columns should have distinct names). For
example, BirthPlace, HomeTown, WorkingPlace, etc., all of them
relating PEOPLE to CITIES, are much, much better than CityID1,
CityID2, and CityID3, respectively.

2.10.2  OBJECT SET TYPES

I’m more of an adventurous type than a relationship type.

—Bob Dylan

R-DA-2. Object sets are either atomic (i.e., of “entity” type) or not (i.e., of
non-functional “relationship” type) that is, mathematically, relations (i.e.,
subsets of Cartesian products).
a. For each abstracted object set, you should always first determine
its type in the corresponding context.
b. In different contexts, object sets might have different types.
c. Obviously, if a set is not atomic, then you should also exactly deter-
mine and specify its underlying object sets.
For example, considering a logistic subuniverse where several product
types might be stored in a same warehouse and warehouses may store sev-
eral product types, we should abstract three related object sets:
78 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

FIGURE 2.18  Many-to-many (nonfunctional) STOCKS E-RD fragment.

PRODUCTS (atomic),
•
WAREHOUSES (atomic), and
•
STOCKS (nonatomic, linking the former two atomic ones, which
•
are its underlying sets, through two corresponding roles: Product
and Warehouse).
The attribute StockQty is a property of STOCKS, as there may be, for
example, q1 products of type p1 in warehouse w1, q2 in w2, and so on.
The corresponding E-RD fragment should then be as in Fig. 2.18.
When there is only one warehouse, both WAREHOUSES and STOCKS are
no more needed, and attribute StockQty is associated to PRODUCTS instead.

2.10.3  THE KEY PROPAGATION PRINCIPLE

Change your opinions, keep to your principles;

change your leaves, keep intact your roots.
—Victor Hugo

R-DA-3. Always model functional relationships, according to KPP, as

(relating/referencing) properties, instead of relationship-type object sets.

2.10.4  OBJECT SETS ABSTRACTION

I paint objects as I think them, not as I see them.

—Pablo Picasso

R-DA-4.
a. Objects should be always abstracted as set elements, not as char-
acter strings, numbers, calendar dates, etc., that is object property
The
 Quest for Data Adequacy and Simplicity 79

values (e.g., people, companies, cities, states, countries, invoices,

payments, etc. are all object sets; for example, a city is an object,
element of an object set CITIES, not the character string denoting
its name).
b. Dually, you should never add to your data models sets that either
are not relevant (e.g., subset of calendar dates when invoices were
issued), or should conceptually be merged with other ones (e.g.,
passports set should be merged with the corresponding people
set), or are equipotent with other ones, even if implementation
would then distribute them apart (e.g., HR, salary, competence
data, etc.), and so on.
c. Moreover, whenever sets become obsolete they should either be
dropped from the corresponding data models or, at least, renamed
accordingly.
d. To all of the abstracted object sets there should consequently be
added corresponding entity or relationship-type sets in the E-R
data models conceptually modeling the corresponding subuniverse
of interest.
e. Make full use of abstraction: conceptual data models should have
a minimum possible number of object sets, none of them seman-
tically overloaded though. Always inflate instances and deflate
schemes.
For example, you should not abstract sets NOTEBOOKS, DESK-
TOPS, MONITORS, MOTHERBOARDS, MEMORIES, HARD_
DISKS, etc.; instead, you should abstract sets:
• PRODUCTS
• PRODUCT_TYPES—whose elements include notebook, desk-
top, monitor, motherboard, memory, hard disk, etc.
• CHARACTERISTICS—whose elements include monitor
dimension, CPU speed, hard disk size, hard disk speed, mem-
ory size, speed, etc.
• MEASURE_UNITS, and
• PROD_TYPES_CHARACTERISTICS—storing data on what
characteristics has each product type, as well as the values for
these characteristics as presented in the E-RD from Fig. 2.19.
f. Data models too may contain computed sets (e.g., *CAPITALS
# CITIES, *STATE_CAPITALS # CITIES, *MEN # PEOPLE,
80 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

FIGURE 2.19  Example of using abstraction in data modeling for reducing models
complexity.

*WOMEN # PEOPLE, etc.), provided they are consistently

named, documented, read-only for users, and automatically recom-
puted whenever needed.

2.10.5  OBJECT SET BASIC PROPERTIES

A theory is the more impressive the greater the simplicity of its prem-
ises is, the more different kinds of things it relates, and the more extended
is its area of applicability. Therefore the deep impression which classical
thermodynamics made upon me. It is the only physical theory of univer-
sal content concerning which I am convinced that within the framework
of the applicability of its basic concepts, it will never be overthrown.
—Albert Einstein

R-DA-5. Object sets should be described by all of their relevant properties

of interest, properties that should also make possible unique identification
of their elements.
For example (also see Fig. 2.18), PName (unique) and PPrice could
characterize PRODUCTS, WName (unique)—WAREHOUSES, and Stock-
Qty—STOCKS (obviously, stocks are uniquely identified by the product of
the corresponding Product and Warehouse values), as in Fig. 2.20.
The
 Quest for Data Adequacy and Simplicity 81

FIGURE 2.20  Example of property aggregation for uniquely identifying elements of a

non-functional relationship-type object set.

2.10.6  OBJECT SET PROPERTY TYPES

I bought a two story house. … One story before I bought, and another after.
—(anonymous author)

R-DA-6. Object set properties are mappings defined on the correspond-

ing object sets and taking values either in some finite subset of some data
types (for the basic ones, represented by ellipsis in E-RDs) or in other (but
also possibly the same) object sets (for the functional type relationships—
see R-DA-3 above and R-DA-7 below). As such, we should always ask
ourselves for each of them whether or not, in the corresponding subuni-
verse, they are one-to-one or not.
For example, in R-DA-5 above, PName and WName are taking values
in subsets of the ASCII strings (e.g., having maxim length of 64 for the
former and 16 for the latter) and, being one-to-one, both are linked to
PRODUCTS by arrows, while PPrice and Qty are taking numerical values
(e.g., in [0, 1000]) and are not one-to-one (as several products may have
same price and as several stocks may have same quantity, respectively).
In Fig. 2.21, both CountryName and CityName are taking values, for
example, in ASCII(64); CountryName is one-to-one (as there may not be
two countries having a same name), CityName is not (as there may be sev-
eral cities, even in a same country, but in different counties, having same
names), whereas Capital is one-to-one too (as no city may simultaneously
be the capital of more than one country).

FIGURE 2.21  Example of one-to-one and not one-to-one properties.

82 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

2.10.7  CONTROLLED DATA REDUNDANCY: COMPUTED

DATA

Redundancy is my favorite business strategy.

—Amit Kalantri

R-DA-7.
a. Fundamental dbs should store any fundamental data only once.
Any computable data is redundant; hence, generally, it should not
be stored.
b. Exceptions justified by querying speeding-up needs should not only
be documented as such (starting with prefixing their name with a
special character, like ‘*’, indicating the corresponding expression
for computing it, as well as the reasons why it was added to the
model), but always be made read-only for end-users in any appli-
cation built on that db, and their values be updated only automati-
cally, by trigger-like methods, immediately when needed (which is
referred to as controlled data redundancy).
c. Generally, such exceptions should not be introduced at this level
(the only conceptual one available to customers too), as it may con-
fuse customers (possible exception: when you want to justify to
your customers higher costs for faster solutions), but only starting
with the next conceptual level.
d. Before introducing such exceptions, you should make sure that they
are either critical or that significant querying speed-up advantages
justify additional costs in developing, db size, and corresponding
automatic updates needed time.
e. You might add both computed sets, possibly having associated
properties and computed properties of fundamental sets.
f. Computed restrictions and fundamental objects uncontrolled
redundancy should be banned!
For example, let us consider the E-RD from Fig. 2.22 (see also, for
example, the standard MS Northwind demo db), consisting of object sets:
• CUSTOMERS,
• CITIES,
• STATES, and
The
 Quest for Data Adequacy and Simplicity 83

FIGURE 2.22  E-RD of part of the MS Northwind demo db.

• COUNTRIES,
where:
– every customer has its headquarters in only one city,
– any city belongs to only one state, and
– any state belongs to only one country, and, consequently,
– any headquarter is located in only one state and country.
Both CState and CCountry are computable, hence redundant:
CState = State ° CCity and
CCountry = Country ° CState = Country ° State ° CCity.27
This means that, whenever you also add them to your solution, you
should also add two corresponding additional constraints (CState = State
° CCity and either CCountry = Country ° CState or CCountry = Country
° State ° CCity) to it: otherwise, implausible instances would be possible.
For example, even if, in CITIES, Paris belongs to Ile-de-France, in
STATES Ile-de-France belongs to France, and Tennessee to U.S., in CUS-
TOMERS Sephora, whose headquarters are actually in Paris, may oth-
erwise, according to such a db instance, be located, for example, in the
Greenland’s unknown state from Romania!
Obviously, such a solution is either very costly (and not that much
because of the needed extra db space for storing these two computed col-
umns and time to update their values too—be it automatically, as it should,
or manually, as it shouldn’t—, but especially for enforcing these two extra
needed constraints) or an incorrect one (whenever these two constraints
are not added too).

27
Recall please that “°” is the algebraic symbol for function composition; for example, for any city c,
the country to which c belongs may be computed by composing Country and State: (Country ° State)
(c) = Country(State(c)).
84 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

Normally, all cycles in this E-RD should be broken, by dropping both

CState and CCountry (which may be computed anytime when needed),
implying that no extra constraints are needed anymore, like in Fig. 2.23.
Whenever querying speed is critical or updates to CState and CCoun-
try are much less frequent than queries involving them (and additional
needed db space is affordable), they may be added as computed properties
*CState = State ° CCity and/or *CCountry = Country ° State ° CCity (with
controlled redundancy, that is being read-only to end-users and automati-
cally updated whenever changes occur in CCity, State, and/or Country
corresponding values, by trigger-like methods); note that this solution (see
Fig. 2.24) is as costly as the initial one with the two extra constraints added
too, since plausibility is always guaranteed).
Moreover, in such cases, even if costlier, in fact it is often better that
these two redundant properties store not pointers to the corresponding
object sets, but plain names for customer cities, states, and countries (i.e.,
*CCity = CityName ° CCity, *CState = StateName ° State ° CCity, and
*CCountry = CountryName ° Country ° State ° CCity), like in Fig. 2.25:
with this solution, when querying data about customers, only CUSTOM-
ERS have to be accessed (whereas even in the previous one, from Fig.
2.24, for getting corresponding city, state, and country names too, all four
object sets have to be accessed).

FIGURE 2.23  Resulting E-RD after breaking the cycles in Fig. 2.22.

FIGURE 2.24  Resulting E-RD after adding controlled redundancies to the one in
Fig. 2.23.
The
 Quest for Data Adequacy and Simplicity 85

FIGURE 2.25  Resulting E-RD after optimizing controlled redundancies of the one in
Fig. 2.24.

2.10.8  DATA COMPUTABILITY

If you don’t want to be replaced by a computer, don’t act like one.

—Arno Penzias

R-DA-8. Fundamental dbs should not ever store computable data instead
of corresponding fundamental one.
For example, you should not ever store (for people/living beings/spiri-
tual periods, etc.) Age instead of, say, BirthDate/Year: trivially, Age is com-
putable from BirthDate/Year; moreover, if you would add Age instead of
BirthDate/BirthYear, you are forcing users each day/year to check whether
or not one or more of its values should be incremented and, whenever it is
the case, to do it!

2.10.9  RESTRICTIONS NECESSITY

Computers are like Old Testament gods; lots of rules and no mercy.
—Joseph Campbell

R-DA-9. Any fundamental (i.e., non-implied, non redundant) restriction

governing a subuniverse should be added in all corresponding data mod-
els. Not including such a constraint might lead to implausible db instances.
For example, consider (from an academic subuniverse) object sets (of
entity-type) STUDENTS, COURSES, and (of relationship-type) ENROLL-
MENTS and PREREQUISITES, like in Fig. 2.26.
86 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

FIGURE 2.26  E-RD for a fragment of the student enrollment subuniverse.

PREREQUISITES should be acyclic, as no course should be a pre-

requisite of itself (neither directly, nor indirectly). If this restriction is not
added to the data model of such a subuniverse, then, obviously, no student
may ever be enrolled in any course belonging to a cycle in the prerequi-
sites graph.
For example, suppose that in order to be enrolled for Databases, you
need as prerequisites Computer Programming, Algebra, and Logic; for
being enrolled in Computer Programming you need as a prerequisite
Operating Systems, but Operating Systems would have Databases as a
prerequisite; then, trivially, no student may ever enroll either for Operat-
ing Systems, or Computer Programming, or Databases; moreover, he/she
might never be enrolled for any other courses having any one of these
three as a prerequisite!
Here is another, simpler and somewhat dual example: assume that
SSN is not declared as one-to-one; obviously, implausible instances might
be possible, because any number of people might then have stored same
SSNs, which should not be acceptable.

2.10.10  NO IMPLAUSIBLE RESTRICTIONS

It has been claimed at times that our modern age of technology

facilitates dictatorship.
—Henry A. Wallace

R-DA-10. No implausible (dictatorial, not existing, aberrant) restric-

tions should ever be added to db schemas and/or applications (and not
only: in fact, to anything else in the Universe!). Any such restriction is
not only time consuming, but is aberrantly rejecting possible plausible
instances.
The
 Quest for Data Adequacy and Simplicity 87

For example, in a worldwide context, adding the restriction that city

names, or zip codes, or state telephone prefixes, or vehicle plate codes, etc.
should be (worldwide) unique is aberrant and would prevent, for exam-
ple, storing more than one city having any given name (when, in fact,
for example, besides France’s capital, there are plenty of other cities in
the world named Paris, from Tennessee, Texas, and Kentucky, to Illinois,
Ontario, etc.).

2.10.11  NO RESTRICTIONS ON COMPUTED OBJECTS,

EXCEPT FOR KEYS

Any kind of restrictions put on free speech would have worse conse-
quences than bullying.
—Colleen Martin (aka Lady Starlight)

R-DA-11. No restrictions should ever be added to computed object sets or

properties: as their instances/values must be computed only from plausible
instances, they should also be intrinsically plausible (of course, provided
that their computation expressions are at least plausible too!), except for
computed columns added especially in order to be able to define keys (See
Section 3.2.3.3).
For example, no constraints should be added either on computed proper-
ties of fundamental object sets or on computed (derived, non fundamental)
sets, be them temporary or not, except for prime properties (see Fig. 3.5).

2.10.12  MAXIMUM CARDINALITIES

The more minimal art, the more maximum the explanation.

—Hilton Kramer

R-DA-12. For each object set its maximum possible number of elements
should be added as its first associated restriction.
For example, #Cities values should be between 0 and 1,000 for states/
regions/departments/lands/etc., or 100,000 for countries, or 100,000,000
worldwide, whereas #Countries values should be between 0 and 250.
88 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

Generally, reasonable margins should be provided in order to let room

for foreseeable evolutions (e.g., 16 billion people worldwide by 2300, as
current U.N. forecast is of some 9 billion).

2.10.13  RANGE RESTRICTIONS

I admire how she protects her energy and understands her limitations.
—Terry Tempest Williams

R-DA-13. For each attribute you should associate a corresponding range

restriction, that is, a plausible finite subset of a data type.
For example, FName and LName might both take values in ASCII(64),
the subset of all ASCII strings having length less or equal to 64; BirthDate
and HireDate might have [1/1/1945, SysDate()] , DATE as range restric-
tions; Sex might take values into {‘F’, ‘M’}, Quantity into [0.0, 100.0],
MarriageDate into [1/1/-4000, SysDate() + 1year], Grade into [0, 20],
StartH into [8, 18], etc.
Please note that one of the most typical programming runtime errors
arises when programs assign implausible values to variables (e.g., out of
range subscripts) and that the syntagm “garbage in, garbage out” would
not perhaps be still that actual nowadays if programs were always check-
ing both their input and output values for plausibility. Consequently, do
not ever range restrict properties with implausible sets: for example, let-
ting BirthDate to take any DATE value for EMPLOYEES would allow
end-users to enter implausible data, resulting, for example, in employees
that are either 2,000, 2, or −2,000 years old!

2.10.14  MANDATORY (COMPULSORY) TYPE RESTRICTIONS

Getting to the top is optional. Getting down is mandatory.

—Ed Viesturs

R-DA-14. For each object set, there should be at least one compulsory
property (i.e., one that should not accept NULL values), besides the cor-
responding surrogate key. Do not forget to also ban dirty nulls.
The
 Quest for Data Adequacy and Simplicity 89

For example, SSN, FName, LName, CityName, StateName, Country-

Name, State, Country, Precondition, Course, Sex, and Quantity above
should all be compulsory, while BirthDate, BirthPlace, HomeTown, Work-
ingPlace, and Capital above might allow null values too.
Generally, names, surrogate keys, and roles should always be compulsory.
Dually, it is obvious that it would be useless, incomprehensible, and
implausible to store no data for all of the properties of an element of any
object set (even if it would have an associated surrogate key value).

2.10.15  UNIQUENESS RESTRICTIONS

Feeling unique is no indication of uniqueness.

—Douglas Coupland

R-DA-15.
a. For each object set, except for very rare cases, we should be able to
uniquely identify any of its elements through at least one semantic
(i.e., nonsurrogate) attribute, functional relationship, or aggrega-
tion of attributes and/or functional relationships.
For example, countries should be uniquely identified by their Coun-
tryName, CountryCode, and Capital, states should be uniquely
identified by the product StateName • Country (as no country may
simultaneously have two states having same name), people should
be uniquely identified by their SSN, etc.
b. There are only three exception types to “a.” above:
• Subsets: some of them might not have any semantic key (please
note that they may, however, be uniquely identified by the
semantic uniquenesses of their corresponding objects supersets,
as they inherit them all). For example, DRIVERS # EMPLOY-
EES is not generally having any semantic uniqueness of its own
(but it inherits EMPLOYEES’ ones, for example SSN).
• Object sets for which it makes sense to overload their surro-
gate (syntactic) keys with some semantics (and, consequently,
it does not make sense anymore to duplicate its values in an
identical semantic property). For example, RABBIT_CAGES
90 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

might actually too be uniquely identified by their (generally,

automatically generated number) surrogate key values (e.g.,
cage 1, cage 2, cage 3, etc.) and only by these values.
• Object sets for which we may not have enough data for some
of their elements (like people that lived in ancient times and
are ancestors of people for which we have enough data). How-
ever, note that even in such cases we might add some one-to-
one Notes property (e.g., “ancestor of x from the Xth century
BC who distinguished himself in the battle of y”).
c. Please note that, even if most of the object sets have several seman-
tic uniquenesses (that we should eventually discover and enforce
in the end—see next chapter), in this first stage discovering only
one (generally, the most used one by your customers) is enough;
however, if you already discovered several ones, you should add
them all to the corresponding restriction set.

2.10.16  OTHER RESTRICTION TYPES

The golden rule is that there are no golden rules.

—George Bernard Shaw

R-DA-16. Very frequently, apart from cardinality, range, mandatory, and

uniqueness restrictions, data subuniverses are also governed by restric-
tions of other types as well: obviously, all of them should always be added
to our corresponding models (see R-DA-9 above). Do not ever forget, in
particular, that trees are acyclic, so you should add this constraint to any
autofunction modeling a tree.
For example, in R-DA-6 above, a restriction enforcing that, for each
country, its capital should be one of its cities (and not one of another
country) should be added; in R-DA-9 above, we already seen that PRE-
REQUISITES should be acyclic; in R-DA-13 above, a restriction stating
that minor young ones cannot be employed should be added too: for each
employee, HireDate ≥ BirthDate + 18years; for the above case study, a
restriction should ban first authors from also appearing in the correspond-
ing coauthor lists.
The
 Quest for Data Adequacy and Simplicity 91

2.10.17  OBJECT SETS AND THEIR RESTRICTIONS ORDERING

AND NAMING

To create architecture is to put in order.

Put what in order? Function and objects.
—Le Corbusier

R-DA-17.
a. You should consider object sets in the descending order of their
importance in the corresponding subuniverse.
b. For each object set you should specify its restrictions grouped by
their types (and always order types consistently; above R-DA-12 to
R-DA-16 rule types order is preferable).
c. For other restriction types involving several object sets (e.g., Bor-
rowDate ≤ DueReturnDate and BorrowDate ≤ ActualReturnDate
from the case study above), you should place them either in the
restrictions subset of the last involved object set or at the end of the
restriction set, in a dedicated multiset restrictions section.
d. All restrictions of any data model should be uniquely named
within it.

2.10.18  OBJECT SETS AND THEIR PROPERTIES AND

RESTRICTIONS DOCUMENTATION

Documentation is like sex: when it is good, it is very, very good;

and when it is bad, it is better than nothing.
—Dick Brandon

R-DA-18. Except for cases when (preferably generally, not only in that
particular subuniverse) object sets, properties, and restriction names are
unambiguous and self-explanatory (e.g., PEOPLE, FirstName, SSN, PEO-
PLE_SSNuniqueness, etc.), you should always associate substantial, com-
prehensive, concise, clear, nonambiguous descriptions for both object sets,
properties, and restrictions.
92 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

2.10.19  E-RDS ARCHITECTURE FOR COMPLEX DATA MODELS

The art of simplicity is a puzzle of complexity.

—Douglas Horton

R-DA-19.
a. First establish atomic subdiagrams, only containing one object set
and its basic properties (i.e., one rectangle or one diamond – with
its underlying object sets too – and all associated ellipsis).
b. All subsequent E-RDs should be structural diagrams (i.e., only
containing rectangles, diamonds, and arrows, without ellipsis).
c. Structural E-RDs should be established by (sub)domains of the
universe of discourse (HR, Salaries, Accounting, Finance, Custom-
ers, Orders, etc.).
d. Links between these structural E-RDs are naturally ensured by
commonly shared object sets (e.g., EMPLOYEES will appear both
in HR, Salaries, and Orders).
e. Each E-RD should fit on one page.
f. Rectangles, diamonds, ellipsis, arrows, and lines should not inter-
sect each other.
For example, the E-RD from Fig. 2.15 could be split into nine atomic
subdiagrams and one structural E-RD, as presented in Figs. 2.27–2.37.

2.11  THE MATH BEHIND E-RDS AND RESTRICTION SETS: THE

DANGER OF “MANY-TO-MANY RELATIONSHIPS” AND THE
CORRECT E-RD OF E-RDM

If people do not believe that mathematics is simple,

it is only because they do not realize how complicated life is.
—John Louis von Neumann

2.11.1  THE MATH BEHIND E-RDS AND RESTRICTION SETS

The person who reads too much and uses his brain too little will fall
into lazy habits of thinking.
—Albert Einstein
The
 Quest for Data Adequacy and Simplicity 93

FIGURE 2.27  PERSONS subdiagram.

FIGURE 2.28  BOOKS subdiagram.

FIGURE 2.29  CO-AUTHORS subdiagram.

FIGURE 2.30  VOLUMES subdiagram.

FIGURE 2.31  PUBLISHERS subdiagram.

FIGURE 2.32  EDITIONS subdiagram.

94 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

FIGURE 2.33  VOLUME_CONTENTS subdiagram.

FIGURE 2.34  BORROWS subdiagram.

FIGURE 2.35  COPIES subdiagram.

FIGURE 2.36  BORROW_LISTS subdiagram.

Formally, any E-RD D is a pair D = <S, M >, where S is a finite nonempty

collection of sets and M a finite nonempty set of mappings defined on and
taking values from sets in S; (S, #) is a poset.
S = Ω ! V, where Ω is a nonempty collection of object sets and V is
one of value sets; sets of Ω are explicitly figured (as rectangles or diamonds)
in any E-RD, while those in V are implicitly assumed (never figured).
Ω = E ! R, where E is a nonempty collection of atomic sets (repre-
sented as labeled rectangles and whose elements are called entities) and R
is a (possibly empty) collection of mathematical relations (represented as
labeled diamonds and whose elements are called relationships).
The
 Quest for Data Adequacy and Simplicity 95

FIGURE 2.37  Public library structural E-RD.

Sets in V (generally called data ranges) are generally finite subsets of

the so-called data types from computer programming, like naturals, inte-
gers, rationals, calendar dates, freely generated monoids over some alpha-
bets (usually ASCII and UNICODE), Boolean values, etc., but may also
be user defined, like, for example, the set of the rainbow colors.
M # (S ø NULLS)Ω, where NULLS is a distinguished countable set
of nulls (or null values); M = A ! F, where A # (V ø NULLS)Ω is a
nonempty set of attributes (represented as labeled ellipsis) and F # (Ω ø
NULLS)Ω is a (possible empty) set of functional relationships (or struc-
tural functions, represented as labeled arrows). R ù F = Ø.
Roles are names of canonical Cartesian projections; for example, in
Fig. 2.25, Prerequisite and Course are the names of the corresponding
(first and second) such projections of PREREQUISITES on COURSES.
The only unlabeled arrows (represented in E-RDs as ,→) are canonical
injections; for example, Fig. 2.4 above represents the canonical injection i :
EMPLOYEES → PEOPLE.
Obviously, both canonical Cartesian projections and canonical inclu-
sions are structural functions (functional relationships).
96 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

Surrogate keys are natural injections defined on object sets (not more
than one per set) having no semantics other than uniquely identifying their
elements: ∀O[Ω, there may be only one surrogate key x : O → NAT(n), n
natural; other authors are naming them ID or OID or IDO or #O etc. Obvi-
ously, surrogate keys are attributes.
Any restriction set C associated to an E-RD D is a nonempty set of
closed Horn clauses whose first order predicates only refer to sets and
mappings of D.
Any (attribute) range restriction c[C is, in fact, specifying the codo-
main of a mapping a[A (c = “range a is C ”, where dom(a) = O [ Ω, is
equivalent to a : O → C).
Any (attribute) compulsory restriction c[C is stating that the codo-
main of a mapping m[M is disjoint from NULLS (c = “compulsory m” is
equivalent to codom(m) ù NULLS = Ø).
Any uniqueness restriction c[C is stating that a mapping m[M is one-
to-one or that a mapping product m1 • … • mn, mi[M, ∀i[{1, …, n},
n > 1 natural, is minimally one-to-one. Note that, unfortunately, some
authors and db “experts” wrongly consider one-to-one “relationships” as
also being onto.

2.11.2  THE DANGER OF “MANY-TO-MANY RELATIONSHIPS”

AND THE CORRECT E-RD OF E-RDM

Discovery consists of seeing what everybody has seen and thinking

what nobody has thought.
—Albert Szent–Gyorgyi

From the E-RDM point of view, the relationship-type object set UNDER-
LYING_SETS from Subsection 2.9.2 above is correctly used: any object
set may be the underlying set of several non-functional relationship-type
sets and any non-functional relationship-type set has several (at least two)
underlying object sets, so this is a “typical” so-called many-to-many rela-
tionship-type set.
In fact, modeled as such, it is not even a set, as it should accept dupli-
cates: any non-functional relationship-type set may have a same object
The
 Quest for Data Adequacy and Simplicity 97

set as underlying one as many times as needed—and the most frequently

studied and encountered such type of math relation is the homogeneous
binary ones (i.e., those defined on two occurrences of a same set, like, for
example, PREREQUISITES from Fig. 2.26).
Generally, as it is shown in Section 2.15 of the second volume (dedi-
cated to the math behind (E)MDM), for any math relation the product of
its canonical Cartesian projections should be one-to-one (and, in particu-
lar, for the binary ones, minimally one-to-one).
As a same object set may occur several times among the underlying
sets of a non-functional relationship-type object set, the product Relation-
shipType • ObjectSet is not one-to-one, which means that, mathematically,
UNDERLYING_SETS is ill-defined: the correct data model of it should use
an entity-type set and not a relationship-type one.
Figure 2.38 shows the corresponding correct E-RD for the E-RDM.
From this example (and other similar ones are provided in Section
2.3 of the second volume), it is clear that “many-to-many” relationships
should be used very cautiously if taking the shortest advisable path of
conceptual data modeling and db design presented in this volume, as they
are dangerously imprecise.

FIGURE 2.38  The correct E-RD of the E-RDM.

98 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

Fortunately, as formally proved in Section 2.10 of the second volume,

you can always replace relationship-types object sets with entity-type
ones (and “downgrading” their canonical Cartesian projections to ordi-
nary functional relationships), just as in Fig. 2.38 as compared to the E-RD
from Subsection 2.9.2 (Fig. 2.17).

2.12 CONCLUSION

The effective exploitation of his powers of abstraction

must be regarded as one of the most vital activities of a competent
programmer.
—Edsger W. Dijkstra

Db design should start with a thorough data analysis and modeling, for
correctly identifying all needed object sets, their properties, as well as the
rules governing the corresponding subuniverse of discourse. This process
should result in an E-R data model—that is a set of E-RDs, an associ-
ated restriction set, and an informal but clear, concise, and comprehensive
description of the corresponding subuniverse.
Not only that you should never start creating the tables of a db with-
out a proper data analysis, but, moreover, during data analysis you should
never think in terms of db tables, but in terms of object sets, their proper-
ties, the relationships between them, and the business rules that are gov-
erning the corresponding subuniverse.
Algorithm A0 from Section 2.9, together with the best practice rules
from Section 2.10, provide a framework for assisting this process. As also
seen from the case study presented in Section 2.8, this process is an itera-
tive one. In the end, customers and db designers should agree (preferable
formally) on all three deliverables of this process.
Obviously, this process is greatly favored by abstraction power, knowl-
edge of the subuniverses of discourse semantics and governing rules, com-
munication skills (in order to both find out from customers exactly what
they need and everything relevant you don’t know, but they do, as well
as to adequately present your deliverables as to convince and get from
your customers the final agreements), and experience (i.e., a library of best
solutions to as many subuniverses as possible: contributing to this library
is also one of the main goals of this book).
The
 Quest for Data Adequacy and Simplicity 99

Two golden rules are crucial to the success or failure of data analysis:
• Never overload object sets with several semantics.
• Discover and include in your models not only all needed object
sets, properties of and functional relationships between them, but
also all business rules that are governing the corresponding subuni-
verses of discourse.
Although further refinements are generally needed, as shown in the
second volume, a corresponding relational db scheme and an associated
(possibly empty) set of non-relational constraints (that will have to be
enforced either through db triggers or by db applications) may already be
obtained from these deliverables by applying algorithm A1-7 introduced
in Section 3.3 and also exemplified in Section 3.4.
Dually, when you have to work with an unknown db, for which there is
no such documentation, by using the reverse engineering algorithm REA1-2
introduced and exemplified in Section 3.5 you will benefit a lot especially from
the structural E-RD and restriction set that you will obtain, which is always
providing a very useful overview on both the set of all existing interobject sets
connections, as well as on the business rules governing that subuniverse.
Hopefully, the E-R data model of the E-RDM introduced in Section
2.9 makes understanding of this data model simpler. The math behind this
E-RDM kernel (presented in Section 2.11) is very straightforward, while
providing a deeper understanding of this simple, yet powerful data model.
As many are wrongly assuming that one-to-one relationships are also onto,
as using many-to-many ones proves dangerous (see Subsection 2.11.2) and
not fundamental (see Section 2.10 of the second volume), and as many-to-one
and one-to-many ones may be misleading too (as, in fact, they are both ordi-
nary mappings with only the arrow direction being different between them),
we would strongly recommend to get rid of them all and always think in data
modeling only in terms of atomic object sets and mappings defined on them.

2.13 EXERCISES

In any situation, the best thing you can do is the right thing;
the next best thing you can do is the wrong thing;
the worst thing you can do is nothing.
—Theodore Roosevelt
100 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

2.13.1  SOLVED EXERCISES

He who asks a question may be a fool for five minutes,

but he who never asks a question remains a fool forever.
—Thomas Martin Connelly Jr.

2.0. Apply algorithm A0 and the above best practice rules to design the
E-RDs, restriction set, and description for the following simplified pay-
ments subuniverse: payments to suppliers (names and locations) by cus-
tomers (names and locations) are done for invoices (supplier, customer,
number, date, total amount) either cash, with credit/debit cards (type,
credit/debit, number, expiration month and year, holder, security code,
account), or through bank wires; payment documents are, correspond-
ingly, receipts, bank payment confirmations and orders; for all of them,
supplier, customer, number, date, and total amount are needed; moreover,
the paid total amount of a payment may represent the sum of several par-
tial invoice payments; for bank payments also needed are involved bank
accounts (bank subsidiaries’ addresses and cities, holders, number, IBAN
code, available amount); suppliers may be customers too; there may not be
two cities of a same country having same zip code.
Solution:
Banks may also be suppliers and/or customers, so we abstract them
too as partners; as not all suppliers and/or customers are banks, we add to
PARTNERS a Boolean property Bank? too.
For simplicity, we assume that each bank account has only one holder
and at most one associated card valid in any given time period, and that
any card is valid at least one year.
Obviously, total amounts on both invoices and payment documents are
computable (for payment documents, from corresponding amounts in pay-
ments; for invoices, from corresponding subtotal amounts in invoice lines,
not figured here as it does not matter from the payments point of view);
however, in order to speed-up querying, we have added two computed
*TotAmount properties, one for INVOICES and the other for PAYM_DOCS.
The corresponding description of this subuniverse is as follows:
1. Payments to suppliers by customers are done for invoices (supplier,
customer, number, date, all of them compulsory) either cash, with
The
 Quest for Data Adequacy and Simplicity 101

credit/debit cards (type, credit/debit, number, expiration month,

holder, security code, account, all of them mandatory), or through
bank wires; payment documents are, correspondingly, receipts,
bank payment confirmations, and corresponding orders; for all
of them, debtor, number, date, and type are needed (all of them
mandatory); for bank wires, supplier and customer correspond-
ing accounts (bank subsidiaries, holders, IBAN code, available
amount, all of them mandatory) are needed too; for card payments,
card is needed too; moreover, the paid total amount of a payment
may represent the sum of several partial invoice payments.
2. Suppliers and banks may be customers too; consequently, suppli-
ers, customers, and banks are considered as partners, for which
names, cities, and addresses within cities (all of them mandatory)
are stored; some partners may be subsidiaries of others; obviously,
no partner may be a subsidiary of its own, either directly, or indi-
rectly; only bank type partners may manage bank accounts. There
may not be two partners having same name, city, and address.
There may be at most 100,000 partners of interest.
3. There may not be two invoices issued by a same supplier and hav-
ing same number and date. Self-invoicing is not allowed. There
may be at most 10,000,000 invoices of interest.
4. There may not be two payments issued by a same debtor and hav-
ing same type, number, and date. No payment for an invoice can
be done before the invoice is established or after more than 100
years after corresponding establishing date. The sum of all paid
amounts for different invoices should be equal to the total amount
of the corresponding payment document. The sum of payments for
an invoice may not be greater than the corresponding invoice total
amount. For cash payments, there are no associated bank accounts
or cards. For card ones, card and supplier account are mandatory,
while the debtor’s one should not be stored (as it is the correspond-
ing card account); no payment may be done with an expired card;
there may not be two cards of a same type and having same number.
For bank wire ones, there may be no associated card, but the sup-
pliers’ and the debtors’ accounts are mandatory; no self-payment is
allowed. There may not be two payments for a same invoice on a
102 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

same payment document. There may be at most 100,000,000 pay-

ment documents and details, 1,000,000 cards, and 100 card types
of interest.
5. There may not be two bank accounts having same IBAN code.
There may be at most 1,000,000 bank accounts of interest.
6. There may not be two cities of a same country having same zip
code. City names, corresponding countries, and zip codes are man-
datory. There may be at most 100,000 cities of interest.
7. There may not be two countries having same name. Country name
is mandatory. There may be at most 250 countries of interest.
Figure 2.39 presents the corresponding E-RD. Corresponding restric-
tion set is the following:

1. PARTNERS (The set of all involved partners: suppliers,

customers, banks)

FIGURE 2.39  E-RD for the “payments” subuniverse.

The
 Quest for Data Adequacy and Simplicity 103

a. Cardinality: max(card(PARTNERS)) = 100,000 (P0)

b. Data ranges:
Bank?: BOOLE (P1)
Name: ASCII(128) (P2)
Address: ASCII(255) (P3)
c. Compulsory data: Bank?, Name, Address, Location (P4)
d. Uniqueness: Name • Location • Address (there may not
be two partners having same name, city, and address) (P5)
e. Other type restrictions:
SubsidiaryOf should be acyclic (no partner can be a
subsidiary of itself, either directly or indirectly). (P6)
All and only the subsidiaries of a bank should be of
bank type. (P7)

2. INVOICES (The set of all involved invoices issued by

suppliers to customers)
a. Cardinality: max(card(INVOICES)) = 10,000,000 (I0)
b. Data ranges:
No.: ASCII(16) (I1)
Date: [1/1/2000, SysDate()] (I2)
*TotAmount = Sum(INVOICE_DETAILS.Amount)
(Exercise 2.8) (I3)
c. Compulsory data: No., Date, Supplier, Customer (I4)
d. Uniqueness: No. • Date • Supplier (there may not be
two invoices issued by a same supplier and having
same number and date) (I5)
e. Other type restrictions: Supplier and Customer values
should not ever be equal for a same invoice. (I6)

3. PAYM_DOCS (The set of all involved payment documents

issued by customers according to supplier invoices)
a. Cardinality: max(card(PAYM_DOCS)) = 100,000,000 (PD0)
b. Data ranges:
No.: ASCII(16) (PD1)
Date: [1/1/2000, SysDate()] (PD2)
*TotAmount = Sum(PAYMENTS.Amount) (PD3)
104 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

c. Compulsory data: No., Date, Type, Debtor (PD4)

d. Uniqueness: No. • Date • Debtor • Type (there may not
be two payments issued by a same debtor and having
same type, number, and date) (PD5)
e. Other type restrictions:
No payment for an invoice can be done before the
invoice is established or after more than 100 years after
corresponding establishing date. (PD6)
Whenever Type is ‘c’ (cash), To, From, and Card
should be null. (PD7)
Whenever Type is ‘C’ (card), From should be null,
whereas To and Card should not be null. (PD8)
Whenever Type is ‘w’ (bank wire), Card should be null
and To and From should not be null. (PD9)
To and From values should not ever be equal for a same
payment document. (PD10)

4. PAYMENTS (The set of all involved, possibly partial

and/or consolidated, payments made by customers for their
invoices)
a. Cardinality: max(card(PAYMENTS)) = 100,000,000 (PM0)
b. Data ranges:
Amount: CURRENCY(6) (PM1)
c. Compulsory data: Amount, Invoice, PaymDoc (PM2)
d. Uniqueness: Invoice • PaymDoc (there may not be two
payments for a same invoice on same payment document) (PM3)
e. Other type restrictions:
Sum of all payments for a same invoice should not be
greater than the corresponding invoice total amount. (PM4)

5. PAYM_DOCS_TYPES (The set of all 3 possible payment types:

cash, card, and bank wire)
a. Cardinality: max(card(PAYM_DOCS_TYPES)) = 3 (PDT0)
b. Data ranges:
PDType: {“c”, “C”, “w”} (PDT1)
c. Compulsory data: PDType (PDT2)
The
 Quest for Data Adequacy and Simplicity 105

d. Uniqueness: PDType (there may not be two payment

types having same name) (PDT3)

6. ACCOUNTS (The set of all involved bank accounts hold

by partners)
a. Cardinality: max(card(ACCOUNTS)) = 1,000,000 (A0)
b. Data ranges:
No.: ASCII(32) (A1)
Amount: CURRENCY(32) (available amount) (A2)
c. Compulsory data: BankSubsidiary, Holder, No., Amount (A3)
d. Uniqueness: No. (there may not be two bank accounts
having same IBAN code) (A4)
e. Other type restrictions:
Bank subsidiaries should always be bank type partners. (A5)

7. CARDS (The set of all involved debit/credit cards used

by partners for their payments)
a. Cardinality: max(card(CARDS)) = 1,000,000 (C0)
b. Data ranges:
No.: [1015, 1016 - 1] (C1)
Month: [1, 12] (C2)
Year: [Year(SysDate()) — 2, Year(SysDate()) + 2] (C3)
SecurityCode: [0, 999] (C4)
D/C: {‘D’, ‘C’} (debit or credit card) (C5)
c. Compulsory data: Account, No., Type, Month,
Year, SecurityCode, D/C (C6)
d. Uniqueness: No. • Type • Year (there may not be two cards of
a same type and having same number and expiration year) (C7)
e. Other type restrictions:
No card payment may be done with an expired card. (C8)

8. CARD_TYPES (The set of all involved card types;

e.g. American Express, Visa, Visa Electron, Maestro, etc.)
a. Cardinality: max(card(CARD_TYPES)) = 64 (CT0)
b. Data ranges: CardType: ASCII(16) (CT1)
106 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

c. Compulsory data: CardType (CT2)

d. Uniqueness: CardType (there may not be two card types
having same name) (CT3)

9. CITIES (The set of all involved cities, where partners reside)

a. Cardinality: max(card(CITIES)) = 100,000 (CS0)
b. Data ranges:
City: ASCII(64) (CS1)
ZipCode: ASCII(16) (CS2)
c. Compulsory data: Country, City, ZipCode (CS3)
d. Uniqueness: Country • ZipCode (there may not be two
cities of a same country having same zip code) (CS4)

10. COUNTRIES (The set of all involved countries, where

partners reside)
a. Cardinality: max(card(ACCOUNTS)) = 250 (CN0)
b. Data ranges:
Country: ASCII(128) (CN1)
c. Compulsory data: Country (CN2)
d. Uniqueness: Country (there may not be two countries
having same name) (CN3)

2.13.2  PROPOSED EXERCISES

How you think about a problem is more important than the problem itself
– so always think positively.
—Norman Vincent Peale

Apply algorithm A0 to the following subuniverses:

2.1. Add to the public library db data on people birth dates, phone num-
bers, and addresses, other types of both authors and subscribers (e.g., com-
panies, governments, legislators, nongovernmental organizations, etc.),
books topics, publisher headquarters locations, edition types (e.g., hard
cover, paperback, electronic), volumes’ number of pages and format (e.g.,
B5, A4, A3, etc.), and copies’ status (e.g., brand new, new, satisfactory,
The
 Quest for Data Adequacy and Simplicity 107

deteriorated, etc.). Establish E-R subdiagrams for each object set and one
or several structural ones, and update correspondingly restriction set, and
description of this augmented subuniverse.

2.2. Extend the public library db for also managing journals, maga-
zines, newspapers, movies, music, etc., both in printed and electronic edi-
tions. Update correspondingly E-RDs, restriction set, and description of
this extended subuniverse.

2.3. The subuniverses from Figs. 2.5, 2.12, 2.17, 2.18, 2.23–2.25, as
well as the one mentioned in Section 1.6.

2.4. The following subuniverses:

a. public schools
b. TV stations and their programs
c. Internet web sites and their contents
d. e-commerce sites (e.g., eBay)
e. music and film sites (e.g., YouTube)
f. social networking sites (e.g., Facebook)
g. pictures and albums sites (e.g., Picassa)
h. e-mail clients (e.g., Gmail)

2.5. The following subuniverses:

a. (“supplies”): manufacturers (name, headquarters, plants’ coun-
tries), vendors (name, headquarters, subsidiaries), and products
(name, manufacturer, type, vendor quotations, serial no., manufac-
turer code, vendor code, stocks, characteristics, conditioning); for
countries, only their name and capital are needed; for cities, only
name and country; for characteristics, only name and measurement
unit; for measurement units, only names and immediately upper
ones (e.g., for KB, 1024 KB = 1 MB).
b. (“orders”): customers (types, e.g., people, ltd/Inc./etc. company,
governmental agency, etc., names, contact persons, and locations),
products (name, manufacturer, type, manufacturer code, ven-
dor code, own code, stocks, characteristics, conditioning), orders
108 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

(number, date, deadline, customers, products, quantities, prices,

employees in charge).
c. (“marketing”): products/services (name, type), campaigns (time
interval, types, e.g., internet, TV, radio, banners, etc., sites, manag-
ers, costs, sale increases per periods).
d. (“sales”): products/services (name, type, conditioning), discount
percentages and sold quantities per period and conditioning.
e. (“invoices”): customers (types, e.g., people, ltd/Inc./etc. company,
governmental agency, etc., names, invoice and delivery locations,
and orders), invoices (number, date, and total amount), products/
services (code, name, quantity, and price).
f. (“deliveries”): carriers (names, locations, costs), customers (types,
e.g., people, ltd/Inc./etc. company, governmental agency, etc.,
names, invoices, and delivery locations), invoices (number, date,
total amount), products/services (code, name, quantity, price).
g. (“human resources”): employees and candidates (first and last
name, birth place and date, home and e-mail addresses, sex, par-
ents, spouse, children, education, previous and current jobs, hire
and leaving dates).
h. (“salaries”): employees (first and last name, job, base salary, taxes,
due and actual working hours, bonuses, retention clause, bank
account).
i. (“postal”): continents (names), countries (names, abbreviations,
phone prefixes, state names—e.g. state, department, land, region,
etc.), states (name, country), cities (name, state, subordinated cit-
ies), and zip codes.
j. (“time tables”): departures, stops, and destinations (name, city)
weekdays and hours, per seasons, companies (type, e.g., airline,
railroad, bus, ship, etc., name, country), prices per comfort classes.
k. (“football leagues”): countries (names), cities (names, countries),
stadiums (name, city, owner club), clubs (name, city), matches (home
and host teams, season, date, score, referees, attendees number),
players (names, origin countries, clubs, numbers, positions, start-
ing and ending seasons, buying and selling amounts), goals (match,
player, minute), coaches (names, origin countries, clubs, salaries,
won nonleague competitions), referees (names, origin countries).
The
 Quest for Data Adequacy and Simplicity 109

l. (“school schedules”): disciplines (titles, credits), pre and coreq-

uisites, teachers (first and last names, competences, sex), rooms
(numbers, buildings floors), schedules (semesters, weekdays, dis-
ciplines, teachers, start and end hours, rooms), students (first and
last names, disciplines enrollments, sex).
m. (“movies”): industry people (e.g., directors, actors, composers,
cameramen, sound engineers, etc.) first and last names, birth and
passed away places and dates, movies (title, producers, year, lan-
guage, novel/play/movie adapted, casting, prizes, category—e.g.,
drama, western, musical, etc.).
n. (“music”): composers (first and last names, birth and passed away
places and dates), their works (title, opus and catalog number,
tonality, soloist instruments, parts, lyrics language), interpreters
(types – e.g., person, duo, trio, quartet, choral, chamber/symphonic
orchestra–, first and last names, birth and passed away places
and dates), interpretations (year, works, interpreters, hall/studio),
recordings (publisher, release year, interpretations, media type –
e.g., blu-ray, dvd, cd, vhs–, price).

2.6. Add to the E-R data metamodel presented in Section 2.9 unique-
ness restrictions made out of several mappings, as well as all existing
restrictions of other types.
2.7. Design an E-R data metamodel for the original E-RDM.
2.8. Add to Exercise 2.0 products (unique name and unit price) and
invoice details (invoice, line number, product, unit price, quantity, and
total amount).
2.9. Characterize algorithm A0, proving that its complexity is linear
in the sum of object set collection, property and restriction sets cardinals.
2.10. Analyze the data models proposed by http://www.databas-
eanswers.org/data_models/index.htm and establish E-R data models for
each of them.
2.11. Extend problem 2.0 considering that a bank account can have
several holders and several associated cards simultaneously valid, one for
each such holder.
110 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

2.14  PAST AND PRESENT

Employ your time in improving yourself by other men’s writings,

so that you shall gain easily what others have labored hard for.
—Socrates

If you are depressed, you are living in the past.

If you are anxious, you are living in the future.
If you are at peace, you are living in the present.
—Lao Tzu

The E-RDM introduced by Chen (1976), was the first hugely successful
higher level conceptual data model, not only still in use today, but, very
probably, to be at least as used and revered in the foreseeable future too.
Each year there are dozens of E-RDM international conferences,
mainly proposing new extensions to it (e.g., the International Conferences
on Conceptual Modeling (ER), see, for example, http://www.informatik.
uni-trier.de/~ley/db/conf/er/index.html).
Besides the thousands of papers on it, there are dozens of good books
too; for example, one of the “bibles” in E-RDM is Thalheim (2000).
A recent book in this field is Bagui and Earp (2012).
The math behind E-RDM section 2.10 of this chapter follows the
E-RDM algebraic formalization from Mancas (1985), which also extended
E-RDM with hierarchies of relationships, refined in Mancas (2005).
Even if E-RDM is so simple and with very limited constraint expres-
sive power, it is still heavily used in conceptual data modeling, generally
as a first step, due to its graphical suggestive power.
E-RDs (also abbreviated ERDs) too are ancestors of UML dia-
grams (see http://www.iso.org/iso/iso_catalog/catalog_tc/catalog_detail.
htm?csnumber=32624 and http://www.omg.org/spec/UML/2.4.1/Super-
structure/PDF/), both being types of organograms.
Most of the RDBMS are offering their variants of E-RDs, at least for
visualizing links between db tables. For example, MS Access provides
among its db tools the so-called Relationships window; its MS SQL Server
equivalent is the Database Diagrams; IBM Data Studio (http://www.ibm.
com/developer works/downloads/im/data/index.html) and Oracle SQL
The
 Quest for Data Adequacy and Simplicity 111

Developer Data Modeler (http://www.oracle.com/technetwork/developer-

tools/datamodeler/overview/index.html) empower their users to also cre-
ate and modify db schemas through their E-R-type diagrams.
Dually, there are also E-R data modeling tools that can connect to sev-
eral RDBMSs like, for example, CA Erwin (http://erwin.com/products/
data-modeler).
Algorithm A0 and the E-R data metamodel of the E-RDM (see Section
2.9), the public library case study (Section 2.8), as well as the best practice
rules (Section 2.10) and the above exercises (Section 2.13) come from
Mancas (2001).
All URLs mentioned in this section were last accessed on July
17th, 2014.

KEYWORDS

•• attribute
•• attribute range restriction
•• Barker E-RD notations
•• Chen E-RD notations
•• compulsory data restriction
•• computed data
•• computed objects
•• controlled data redundancy
•• data analysis
•• data computability
•• data modeling
•• E-R data model (E-RDM)
•• E-RD architecture
•• entity
•• Entity-Relationship (E-R)
•• Entity-Relationship Diagram (E-RD)
•• functional relationship
•• implausible restriction
•• Key Propagation Principle (KPP)
112 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

•• Mancas E-RD notations

•• many-to-many relationship
•• many-to-one relationship
•• non-functional relationship
•• object set basic properties
•• object set property types
•• object sets abstraction
•• one-to-many relationship
•• one-to-one relationship
•• Oracle E-RD notations
•• other restriction types
•• plausible restriction
•• recursive relationship
•• relationship
•• relationship hierarchy
•• restriction
•• restriction set
•• role
•• semantic overloading
•• surrogate key
•• uniqueness restriction

REFERENCES

Principles and rules are intended to provide a thinking man

with a frame of reference.
—Carl von Clausewitz

Bagui, S., Earp, R. (2012). Database Design Using Entity-Relationship Diagrams, 2nd
Edition: CRC Press, Boca Raton, FL.
Chen, P. P. (1976). The entity-relationship model: Toward a unified view of data. ACM
TODS 1(1), 9–36.
The
 Quest for Data Adequacy and Simplicity 113

Mancas, C. (1985). A Formal Viewpoint on the Entity-Relationship Data Model: Proc.

6th Int. Conf. on Control Syst. and Comp. Sci., 2, 174–178, Politehnica University,
Bucharest, Romania.
Mancas, C. (2001). Databases for Undergraduates Lecture Notes: Computer Science De-
partment, Ovidius State University, Constanta, Romania.
Mancas, C., Mancas, S. (2005). MatBase E-R Diagrams Subsystem Metacatalog Concep-
tual Design. Proc. IASTED DBA 2005. Conf. on DB and App., 83–89, Acta Press,
Innsbruck, Austria.
Thalheim, B. (2000). Fundamentals of Entity-Relationship Modeling. Springer-Verlag,
Berlin.
This page intentionally left blank
CHAPTER 3

THE QUEST FOR DATA

INDEPENDENCE, MINIMAL
PLAUSIBILITY, AND FORMALIZATION:
THE RELATIONAL DATA MODEL (RDM)

Facts and theories are different things, not rungs in a hierarchy of

increasing certainty. Facts are the world’s data. Theories are
structures of ideas that explain and interpret facts. Facts do not
go away while scientists debate rival theories for explaining them.
Einstein’s theory of gravitation replaced Newton’s, but apples did not
suspend themselves in mid-air pending the outcome.
—Stephen Jay Gould

CONTENTS
  3.1 First Normal Form Tables, Columns, Constraints, Rows,
Instances.......................................................................................118
  3.2 The Five Basic Relational Constraint Types............................... 123
  3.3 The Algorithm for Translating E-R Data Models into
Relational Schemas and Non-Relational Constraint Sets (A1-7):
An RDM Model of the E-RDM.................................................. 146
  3.4 Case Study: The Relational Scheme of the Public Library
Data Model.................................................................................. 152
  3.5 The Reverse Engineering Algorithm for Translating Relational
Schemas into E-R Data Models (REA1-2).................................. 156
  3.6 The Algorithm for Assisting Keys Discovery (A7/8-3).............. 164
116 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

  3.7 RDBMS Metacatalogs. Relational and E-R Data Models

of the RDM................................................................................. 177
  3.8 Relational Schemas Definition. SQL DDL................................. 193
  3.9 Relational Instances Manipulation. SQL DML. Relational
Calculi and Algebra..................................................................... 196
3.10 Higher and the Highest RDM Normal Forms............................. 227
3.11 Best Practice Rules...................................................................... 237
3.12 The Math behind RDM............................................................... 247
3.13  Conclusion.................................................................................. 256
3.14 Exercises..................................................................................... 263
3.15 Past and Present.......................................................................... 313
Keywords............................................................................................... 325
References.............................................................................................. 332

After studying this chapter you should be able to:

• Concisely and precisely define, and exemplify the following key
terms: relational db (rdb) (name, scheme, instance), First Normal
Form (1NF), table (relation) (name, nested, header (scheme), col-
umn (attribute, field), arity, row, instance, cardinal, cursor, record
set, fundamental, computed, temporary, universal), cross tabula-
tion, relation (name, scheme, instance, tuple, arity, cardinal), null
(value, not applicable, temporarily unknown, no information, dirty,
marked), attribute (name, prime, nonprime, domain (range), not
null (mandatory, compulsory, required, total, computed), unique,
concatenation (product), minimally uniqueness), constraint
(dependency) (name, nonrelational, relational, primitive, domain,
(unique) key, referential integrity (typed inclusion), check (tuple),
existence, not null (totality), (partial, transitive) functional, mul-
tivalued (complementary property), join, acyclicity, object, data,
equalities/tuple generating, preservation (representation)), key
(surrogate (syntactic), autonumber, primary, candidate (semantic),
simple, concatenated (composed), superkey, foreign, Propagation
Principle), dangling pointer, metadata catalog (metacatalog), rela-
tional calculus (domain (DRC), QBE, tuple (TRC), SQL, Quel,
The
 Quest for Data Independence, Minimal Plausibility, and Formalization 117

domain independence, type/value domain, CH- completeness

(computable queries, QL), IsNULL predicate), relational algebra
(union, intersection, complementation, Cartesian product, divi-
sion, selection, (total) projection, renaming, inner/outer (left, right,
full) natural/equi/theta/complete/semi/anti join, augmentation, ker-
nel equivalence relation, quotient set, expressive power, lossless
decomposition, completeness), SQL (query, extended, embed-
ded, static, dynamic, DDL, CREATE/ALTER/DROP SCHEMA/
TABLE, CONSTRAINT, NOT NULL, UNIQUE, FOREIGN
KEY, CHECK, GRANT, REVOKE, DML, INSERT, UPDATE,
DELETE, SELECT, aggregation functions (COUNT, SUM, MIN,
MAX, AVG), DISTINCT, DISTINCTROW, TOP, AS, FROM,
INNER/OUTER (LEFT, RIGHT, FULL) JOIN, WHERE, GROUP
BY, HAVING, ORDER BY, ANY, SOME, ALL, (NOT) IN, (NOT)
EXISTS, UNION (ALL), BEGIN/START/COMMIT/ROLLBACK
TRANSACTION, subquery (static, dynamic), standards), view,
stored procedure/function/trigger, transaction, ACIDity, transitive
closure, data manipulation anomalies (insert, update, delete), 2nd
Normal Form (2NF), 3rd Normal Form (3NF), Boyce-Codd Nor-
mal Form (BCNF, 3.5NF), 4th Normal Form (4NF), 5th Normal
Form (5NF), Projection-Join Normal Form (PJNF), Domain-Key
Normal Form (DKNF), Domain-Key Referential Integrity Normal
Form (DKRINF), normalization problem (synthesis, decomposi-
tion), derivation rules ((limited) complete axiomatization, sound-
ness, completeness, finite implication, decidability/undecidability),
chase (tableaux).
• Correctly apply algorithm A1–7 (translation of E-RDs and restric-
tion sets into relational schemas and associated nonrelational con-
straint sets) to any E-R data model.
• Correctly apply algorithm REA1–2 (translation of relational sche-
mas into E-R data models) to any rdb scheme.
• Correctly apply algorithm A7/8–3 (assistance of existing keys vali-
dation and discovery of all keys) to any rdb table scheme.
• Use provided key terms definitions, A1–7, REA1–2, A7/8–3 algo-
rithms, and/or the best practice rules for correctly solving at least
all exercises found at the end of this chapter.
118 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

3.1  FIRST NORMAL FORM TABLES, COLUMNS, CONSTRAINTS,

ROWS, INSTANCES

He is able who thinks he is able.

—The Buddha

Informally, a relational db (rdb) is a finite set of interlinked tables, each

of which has to have a unique name within the rdb and be in the so-called
First Normal Form (1NF)—that is all of their columns should be atomic
(i.e., not subtables; tables that are not in 1NF are called nested) and there
should not be duplicate rows. Both RDM and RDBMSs are generally con-
sidering only 1NF rdbs.28
Note that, formally, atomic cross tabulations like the one in Table 3.1a
are in 1NF; in fact, as cross-tabulations are computable, in RDM they are
never stored: Table 3.1b shows how data from Table 3.1a is stored instead (as
generally, only strictly positive data is stored, while negative one it is not).

TABLE 3.1A  Example of a Cross Tabulation

TABLE 3.1B  RDM 1NF Table Storing Data from Which Cross-Tabulation in
Table 3.1a May Be Computed

28
But there are both theoretical (e.g., O-O extensions of RDM) and implementation (e.g., MS Access
2010 “multiple values” combo and list boxes) approaches of nested table schemes too.
The
 Quest for Data Independence, Minimal Plausibility, and Formalization 119

Also note that RDM’s relations (derived from the homonym sets
algebra ones) may be also represented using graphs or lists, etc., but, for-
tunately, their most common representation uses tables. Historically, in
RDM tables are called relations, columns are called attributes, and rows
are called tuples (or n-tuples).
Table headers (schemas) contain both the corresponding finite set of
column names (which should be unique across any table) and a finite set
of relational constraints. The number of columns of a table (its arity) is a
natural strictly positive number. Table rows (if any) are making up corre-
sponding table (data) instances. Note that, abusively, relation sometimes
designates in RDM only the graph of a relation (i.e., its corresponding
table instance). The number of rows of a table (or its cardinal) is a natural
number. The union of all table headers of an rdb constitutes its scheme; the
union of all table instances of an rdb constitutes its instance.
Figure 3.1 summarizes basic notion equivalences between these three
formalisms’ notational conventions.
For example, in Fig. 3.2, COMPOSERS has arity 5 and cardinality 8;
MUSICAL_WORKS has arity 6 and cardinality 9; TONALITIES has arity 2
and cardinality 6; in total, this rdb instance has cardinality 8 + 9 + 6 = 23.
Parenthesis after tables names, as well as second and third table head-
ers rows contain the associated relational constraints:
Ø in the parenthesis after the table names are listed the key (unique-
ness) constraints, stating that corresponding columns or con-
catenation29 of columns do not accept duplicate values (in this
example: in table COMPOSERS there may not be two rows hav-
ing same values in columns FirstName and LastName; in table
MUSICAL_WORKS there may not be two rows having same val-
ues in columns Composer, Opus and No; in table TONALITIES
there may not be two rows having same values in column Tonal-
ity); consequently, they may be used for unique identification of
corresponding instance rows;
Ø note that each table has a so-called primary key, denoted x in
this example (and almost always in this book: Subsection 3.2.3.1
explains why); primary key names are underlined; unfortunately,
RDM allows to declare as primary any key, which you should never
Formally, f • g denotes the mapping (Cartesian) product of mappings f and g (see Appendix). In
29

RDM, it is usually abbreviated as f g.

120 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

FIGURE 3.1  Equivalences between informal, RDM, and math terminology.

do (see why in Subsection 3.2.3.1); also note that all tables should
have all existing constraints, so all existing keys, which means that,
with a couple of exceptions to be explained and exemplified in
Subsection 3.2.3.1, all tables should have at least one non primary
key as well: for example, if Tonality were not a key in TONALI-
TIES, instances of this table might store, for example, “G major”
twice (or any number of times, thus violating axiom A-DI17 from
Section 5.1.2);
Ø after these parenthesis are listed the check constraints, each of which
is stating that the corresponding formula should hold for every corre-
sponding instance row (in this example, BirthYear + 3 < PassedAway
The
 Quest for Data Independence, Minimal Plausibility, and Formalization 121
122 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

FIGURE 3.2  A classical music rdb example.

Year states that every composer has to live at least 3 years –as it is not
plausible that somebody could compose before being 3–, while Passed-
AwayYear—BirthYear < 120 states that no composer may live longer
than 120 years);
Ø in the first header line are listed the names of the columns (attri-
butes, mappings);
Ø in the second header line are listed both the domain constraints and
the foreign keys:
ü domain constraints specify column range restrictions (in this
example, all x column values are naturals auto-generated by
the system: for COMPOSERS having at most 8 digits, for
MUSICAL_WORKS at most 12, and for TONALITIES at most
2; FirstName and LastName may only be character strings
using the ASCII alphabet, having length at most 32; Passed-
AwayYear may only be a natural number between 1200 and the
current year; etc.);
ü foreign keys specify that corresponding columns should
take values only among the values of other columns (in
this example, there are only two such foreign keys, both in
the MUSICAL_WORKS table: Composer has to take values
only among those of column x in table COMPOSERS, while
Tonality has to take values only among those of column x in
table TONALITIES); foreign keys are interrelating tables or
same table’s instances; if intelligently used, always only like
in the MUSICAL_WORKS table (and everywhere throughout
The
 Quest for Data Independence, Minimal Plausibility, and Formalization 123

this book), namely always referencing numeric, single pri-

mary keys, especially when coupled with combo boxes (drop
down lists), preferably hiding the stored numerical point-
ers to primary keys and replacing them with corresponding
nonprimary key values (e.g., in column Tonality of table
MUSICAL_WORKS with corresponding Tonality values from
table TONALITIES), foreign keys are reducing rdbs size (by
factorizing data according to axiom A-DI10 in Section 5.1.2),
simplifying end-user task (selecting from combo boxes
desired values, instead of repeatedly typing them), and guar-
anteeing referential integrity: any pointer should always point
to an existing row; dually, note that, unfortunately, the RDM
allows for foreign keys to reference any other columns, not
only primary keys, which you should never do; note that, in
order to avoid ambiguity, all referential integrity constraints
specify referenced columns with their full name, consisting
of the name of the table separated by a dot from the name of
the column (e.g., here, COMPOSERS.x and TONALITIES.x);
Ø finally, in the third header line are listed the not-null constraints,
each of which stating that values in the corresponding column
are mandatory (in this rdb example, end-users should always
enter values for LastName and BirthYear of table COMPOS-
ERS, Composer and Opus in table MUSICAL_WORKS, as well
as for Tonality in table TONALITIES, whereas values in all other
not primary keys columns might always miss; for example,
obviously, if we were to declare PassedAwayYear as not-null
too, still alive composers could not ever been stored); note that,
by definition, values for primary keys are always mandatory,
so, for example, the not-null constraints in all three x columns
from Fig. 3.2 might have been omitted, as they are implicitly
assumed; also note that, always, each table should have at least
one not-null non primary key column (even if RDM does not
request it). Conventionally, not known and/or not applicable
values (like, for example, those in column No of table MUSI-
CAL_WORKS above, except for its seventh row) are said to be
null values (nulls); all other (“normal”) values are not null. Note
that nulls are elements of a distinguished countable NULLS set
124 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

that we may freely union with any attribute codomain, except

for primary keys.
Although RDM, which is a purely “syntactic” model (i.e., deprived of
any semantics), does not require it, it is best that tables correspond to object
sets, be them actual (like COMPOSERS and MUSICAL_WORKS) or con-
ceptual (like TONALITIES), their columns to properties of these sets, and
their rows to elements (objects) of them, just like, for example, in Fig. 3.2.
Note that, just like in math, the order of rows in tables does not mat-
ter and that, unlike in math30, the order of columns does not matter either:
for example, in table COMPOSERS above, if we were swapping between
them pairs of columns FirstName and LastName, as well as BirthYear and
PassedAwayYear, both stored data and derivable information would be
exactly the same.
Figure 3.3 summarizes relational constraint type equivalences between
three formalisms notational conventions (also see Fig. 3.1).

FIGURE 3.3  Equivalences between informal, RDM, and math relational constraint
types terminology.
30
where the Cartesian product is not commutative.
The
 Quest for Data Independence, Minimal Plausibility, and Formalization 125

3.2  THE FIVE BASIC RELATIONAL CONSTRAINT TYPES

What we anticipate seldom occurs.

What we least expected usually happens.
—Benjamin Disraeli

The following five relational constraint types are implemented by any

RDBMS: (co)domain, not-null (mandatory, compulsory, required, total-
ity), key (minimal uniqueness), foreign key (referential integrity, typed
inclusion), and check (tuple).

3.2.1  DOMAIN (RANGE) CONSTRAINTS

The shallow consider liberty a release from all law, from every constraint.
The wise man sees in it, on the contrary, the potent Law of Laws.
—Walt Whitman

Domain (range) constraints are those constraints that specify plausible

data ranges for column values.
Programmers are acquainted with assigning data types to their vari-
ables. DBMSs also provide data types for table columns. Declaring
such a type for a column is sometimes enough, for example: BOOLEAN,
NUMBER(n), VARCHAR2(n).
But for most of the time it is not: even if, by definition, all computer
data types are finite, they are generally huge and most of their values are
implausible for actual db instances.
For example, if letting EMPLOYEES.BirthDate to store DATE values,
depending on the DBMS version, users might store birth dates either from
year 100 or in year 10,000, both of them being trivially implausible.
This is why, generally, column values should be restricted to plausible
subsets of type [minValue, maxValue] , dataType. For example, EMPLOY-
EES.BirthDate could be restricted to [1/1/1930, SysDate() — 365 * 18] ,
DATE (as it is implausible to have, for example in 2014, employees older
than 83 and younger than 18).
126 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

3.2.2  NOT-NULL (MANDATORY DATA) CONSTRAINTS

I have three kinds of friends:

those who love me,
those who pay no attention to me,
and those who detest me.
—Nicolas de Chamfort

Very often, it is the case that some values are not applicable or, at least
temporarily, are unknown. In table MUSICAL_WORKS of Fig. 3.2, for
example, there are no values in some rows for the No columns: some
musical works have both an opus and number (within that opus), like, for
example, Chopin’s Nocturne in G major opus 37 no. 2; others do not have
a number (as there is only one musical work within that opus), as it is the
case for all other musical works in Fig. 3.2—for which this number is not
applicable. On the other hand, if we were, for example, to add to COM-
POSERS Vladimir Cosma (born in 1940 and, thanks God, still alive at the
time of finishing this volume), his passing away year, although applicable,
remains unknown.
Conventionally, such values are called nulls. The ANSI report (ANSI/
X3/SPARC, 1975) distinguishes between 14 types of nulls; most of us
consider, however, that there are only three basic types (the rest being
either particular cases of these three or only due to implementation con-
siderations): nonexistent (inapplicable), (temporarily) unknown, and no-
information (the logical or between the first two: it is not known whether
such a value exists or, if it were existing, nothing is known on its nature).
Examples for this last type do not abound; my favorite one is on
Moldavian king Stephen the Great (1433–1504) and the Russian lan-
guage: did he read Russian? As Russians existed at that time, even if it
took another two centuries to become Moldavia’s neighbors, it is possi-
ble that the answer would be affirmative; however, no proof exists for or
against it, and the chances to find at least one in the foreseeable future are
almost negligible.
Consequently, in dbs we have to accommodate with null values too.
However, constraints should also be placed on their usage: what would be
the meaning of a row having only nulls (even if except for its autonumber
The
 Quest for Data Independence, Minimal Plausibility, and Formalization 127

primary key)? Obviously, at least one non-surrogate key column per each
table should not accept nulls: not-null (mandatory, compulsory, required,
totality) constraints are used to specify such columns.
Please note that there is great misunderstanding on null values: some
DBMSs (e.g., MS SQL Server) wrongly assume that there is only one null
value, so columns accepting nulls are not allowed in unique constraints;
others (IBM DB2, Oracle, and even MS Access) assume correctly that
there is a countable infinite number of nulls (so they accept non not null
columns into keys).
One final note on dirty nulls—that is not empty text strings made out of
only non-printable/displayable characters (spaces, for example): obviously
they should be banned too, but, unfortunately, very few DBMSs are doing
it (and sometimes when they may be banned too, as, for example, in MS
Access, this is not the default setting). Consequently, whenever a DBMS
provides this facility too, we should always use it for all columns, regard-
less of the fact that they should accept nulls or not (what would be the sense
to allow storing dirty nulls even in columns that accept nulls?); whenever a
DBMS does not provide this facility, we should enforce dirty nulls banning
through the software applications managing corresponding dbs.

3.2.3  KEY (MINIMAL UNIQUENESS) CONSTRAINTS

The truly important things in life –love, beauty, and one’s own uniqueness–
are constantly being overlooked.
—Pablo Casals

In reality, there are both objects whose uniqueness is interesting and

objects that we do not need to uniquely identify. For example, in the first
category there are people (with their unique e-mail addresses, phone num-
bers, SSNs, etc.), while in the second are the chairs of a room/apartment/
building.
In dbs we should always care for uniqueness, in each table, according
to axiom A-DA04 from 5.1.1: for example, if the db should store detailed
information on each chair, then they have to be uniquely labeled, for exam-
ple with an inventory number, and these numbers stored for each chair in
a CHAIRS table; if not, then we would probably abstract a CHAIR_TYPES
128 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

table and for each of its rows we would store a unique type name (plus
description, total number of chairs per type or, in other tables, per type and
room/apartment/building, if needed).
A key constraint is a statement of the type “C1 • … • Cn key”, where
n > 0 is a natural, Ci are columns of a table T, “•” denotes concatenation31
(which is, generally, omitted) of these columns and key means minimally
unique32—that is it is unique and it does not include any other key (see, for
example, Fig. 3.2). When n = 1, the key is called simple; when n > 1 it is
called concatenated; if a unique column concatenation properly contains a
key (i.e., the included key has smaller than n arity), then it is not a key (as
it is not minimal), but a superkey; note that superkeys are of no actual, but
only theoretical interest.
For example, in Fig. 3.2, Tonality is a simple key of TONALITIES,
while FirstName • LastName and Composer • Opus • No are concatenated
keys of COMPOSERS and MUSICAL_WORKS, respectively. FirstName
• LastName • BirthYear, FirstName • LastName • PassedAwayYear, and
FirstName • LastName • BirthYear • PassedAwayYear are all superkeys33.
FirstName • LastName : COMPOSERS → ASCII(32) × ASCII(32) is,
in this context, one-to-one (as all “old” composers had unique such pairs
and today no composer would dare to take both first and last names of an
“old” one and even if a composer were having, by chance, identical first
and last names as another today’s one, he/she would at least add a surname
in order to differentiate him/herself) and also minimally one-to-one (as
neither of its subproducts are one-to-one: FirstName, as there are several
composers having same first name –for example Franz Schubert and Franz
Liszt– and LastName, as there are several composers having same last
name—for example Johann Strauss and Richard Strauss).
Similarly, Composer • Opus • No : MUSICAL_WORKS → [0,
99999999] × ASCII(16) × [1, 255] is one-to-one too (as, by definition,
musical works are uniquely identified by their composer, opus (catalog)
and number) and also minimally one-to-one (as neither of its subproducts
are one-to-one: Composer • Opus, as there are several works by a same
composer having same opus but different numbers, Composer • No, as
there are several works by a same composer having same number but dif-
31
Mathematically, this means mapping (Cartesian) product.
32
That is minimally one-to-one (see Appendix).
33
Recall that if f : A → B is one-to-one then, for any g : A → C, f • g is one-to-one too (see Appendix).
The
 Quest for Data Independence, Minimal Plausibility, and Formalization 129

ferent opuses, and Opus • No, as there are several works having same opus
and number, but different composers).
Only keys need to be declared, not superkeys34: for example, if instead
of FirstName • LastName you declare FirstName • LastName • BirthYear
as a key, then implausible data might be stored (e.g., a second Antonio
Vivaldi, born in 1927 or any other valid year than 1678, could be stored
too); if FirstName • LastName and FirstName • LastName • BirthYear
are both declared as keys, then no implausible data might be stored, but
updates to COMPOSERS would be slower (and the db would need addi-
tional disk and memory space to store and process the superkey FirstName
• LastName • BirthYear), as the system would enforce the superkey too.

3.2.3.1  Primary Surrogate (Syntactic) Keys

If we have data, let’s look at data.

If all we have are opinions, let’s go with mine.
—Jim Barksdale

Although RDM does not require it, it is a best practice to declare for
each table a primary key: a key that is not accepting nulls and which is
the default referenced one by foreign keys35. Conventionally, its name is
underlined.
Unfortunately, RDM allows users to declare any key as being the
primary one; consequently, it is a widespread bad practice to declare as
primary keys concatenated ones and/or keys containing columns that are
non-numeric. For example, a vast majority of db “experts” would declare
FirstName • LastName as COMPOSERS’ primary key (and would not add
x to this table); consequently, they would need FirstName • LastName
as a foreign key in MUSICAL_WORKS (referencing COMPOSERS) and
declare FirstName • LastName • Opus • No as its primary key (not adding
x to this table either).
There are several disadvantages of this approach:

34
Unfortunately, most RDBMSs (including, for example, Oracle, MS SQL Server and Access, etc.)
allow for enforcement of both keys and superkeys!
35
Moreover, for most RDBMS versions, table instances are stored by default in the order of their
primary keys, if any.
130 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

ü db scheme is more complicated and needs more space, having two

more ASCII columns and a greater arity key (4 instead of 3) for
table MUSICAL_WORKS;
ü any updates to columns FirstName and LastName in COMPOS-
ERS have to be immediately propagated to their homonym ones in
MUSICAL_WORKS;
ü any comparison/join between these two table columns takes much
longer: instead of comparing two (small) numbers (which is done
by the fastest CPU—the arithmetical-logic one), four strings of
ASCII characters (each having 32 ones, which, in the worst case,
means comparing 64 with 64 characters) have to be compared for
each pair of rows;
ü by using a systematic approach of this type, a table CONCERTS_
PROGRAMS, which would store pairs of type <c, w> for each
time that a musical work w is programmed in a concert c, would
need a foreign key of arity 4 referencing MUSICAL_WORKS and
a primary key of arity at least 5 (assuming that the foreign key
to CONCERTS is simple, which is generally not the case in such
approaches); moreover, a table PERFORMANCES storing triples
of type <p, a, i> for each performance p of work w in concert c by
artist a on instrument i would need a foreign key of arity at least
5 referencing CONCERTS and a primary key of arity at least 7
(assuming, again, the very improbable most favorable case when
foreign keys referencing ARTISTS and INSTRUMENTS are sim-
ple); and so on and so forth…
To conclude with, it is always preferable to declare primary keys for all
fundamental tables (derived and temporarily ones generally do not need
keys or any other constraints), but they should be simple numeric (prefer-
ably auto generated) ones, deprived of any semantics (i.e., of the so-called
surrogate or syntactic type), except for very special cases: for example, in
rabbit/poultry/etc. farms, cages could be uniquely identified (and actually
labeled) only by this surrogate primary key (auto generated) values.
Surrogate keys are generally denoted by “ID”, used as such or as a
prefix/suffix of the table name, or by a special character pre/suffix, like
‘#’; I am using x for them (see, for example, Fig. 3.2) for reasons that will
The
 Quest for Data Independence, Minimal Plausibility, and Formalization 131

become clear in Section 3.6 and especially in the second chapter of the
second volume.

3.2.3.2  Semantic (Candidate) Keys

There are only two things in the world: nothing and semantics. 
—Werner Erhard

Note that, unfortunately, there are lot of dbs whose fundamental tables do
not have keys (and, generally, no other constraints either); others always
have only the primary one (generally, of surrogate type). Any table should
have all keys existing in the subuniverse modeled by the corresponding
db, just like it should have any other existing constraint: a table lacking
an existing constraint allows for implausible data storage in its instances.
All nonprimary keys are called semantic (candidate36) keys.
For example, let us consider table COUNTRIES from Fig. 3.4, where all
columns except for Population are keys: x, the surrogate primary one, by
definition; Country, as there may not be two countries having same names;
CountryCode, as there may not be two countries having same codes37;
finally, Capital, as there may not be two countries having same city as their
capitals.
If we were not declaring, for example, Capital as a key too, then
users might have entered 4 (instead of 5) in this column for the fifth row
too, storing the highly implausible fact that Washington D.C. is not only
U.S.A.’s capital, but also Romania’s one.
Please note that uniqueness is not an absolute, but a relative property:
in some contexts a column or concatenation of columns are unique, while
in others (even in a same subuniverse) they are not.
For example, in the subuniverse modeled by Fig. 3.4, where only big
cities are considered, there may not be two cities having same name in a
same state; if we would need to store villages and small cities too, this
product (City • State) would not anymore be unique, as there may be sev-
eral such villages or even small cities of a same state having same name.
36
Unfortunately (from the semantic point of view), in RDM candidate means that any of them might
be chosen as the primary one.
37
See ISO 3166 codes.
132 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

In this case, for example, in most of Europe (where states are called
regions, lands, departments, cantons, etc., depending on countries) there
are three rules:
Ø villages and small cities may be subordinated to bigger ones, gen-
erally called “communes” and there may not be two cities of a
same commune having same name;
Ø there may not be two communes of a same state having same
names;
Ø there may not be two identical zip codes in a same country (where
small cities have one zip code, bigger cities none, but their street
addresses have one too).
Figures 3.4 and 3.5 show the corresponding CITIES tables.
Another, even worse example (as the subuniverse is strictly the same),
is computer file names: almost everybody knows that “there may not be
two files having same name and extension in a folder”, which means that
the triple FileName • FileExt • Folder is a key; in fact, OSs enforce a
supplementary constraint too: there may not be two files in a same folder
having null (no) extensions and same name, which means that for the sub-
set of files without extension FileName • Folder is a key (and FileName •
FileExt • Folder is a superkey).
Consequently, first of all, semantic keys, just like any other type of
fundamental (i.e., not derived) constraints may only be discovered and
declared by humans: there may not ever be any tools, be them hardware,
software, conceptual, etc., able to do such a job. Moreover, especially keys
are not at all easy to discover: besides their relativeness, as the number of
columns increases, the number of their products (theoretically very many
of them being possible keys) increases exponentially. This is why Section
3.6 presents an algorithm for assisting their discovery.
Please note that, dually, again just like for any other constraint type,
you should never assert keys that do not exist in the corresponding sub-
universe: if you were doing it, you would aberrantly prevent users to store
plausible data. For example, if in any table of Figs. 3.4 and 3.5 you were
declaring Population as a key too, then no two corresponding objects
(countries, states, communes, or/and cities) having same population fig-
ures could be stored simultaneously.
The
 Quest for Data Independence, Minimal Plausibility, and Formalization 133

FIGURE 3.4  A small geographical rdb for big cities.

3.2.3.3  Keys Including Computed Columns

Facts do not cease to exist because they are ignored.

—Aldous Huxley
134 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

Even if not that often, uniqueness sometimes involves columns funda-

mentally belonging to different tables, like in the zip codes example above
for small cities only having one such code; another example is provided
by some countries, including Romania, where businesses must have their
headquarters in a city, but their names should be unique within the cor-
responding state (county).
Not only RDM can’t cope with such keys by using only fundamental
columns: not even math can, without computed mappings.
For example, in table CITIES of Fig. 3.5, in order to embed the rule
that small cities only have one zip code, a Zip column has been added; in
order to add the (more general) rule that there may not be two identical zip
codes into a same country, a computed column *Country has been added
too, plus the key *Country • Zip. Obviously, *Country is computable both
mathematically38 and in SQL39.

3.2.3.4  Prime and Nonprime Mappings

You can use all the quantitative data you can get, but you still have to
distrust it and use your own intelligence and judgment.
—Alvin Toffler

RDM also introduced the concepts of prime and nonprime (nonessential)

attributes (columns): if it is part of at least one (unique) key, a column is
said to be prime and otherwise nonprime (nonessential). Obviously, any
single key column is prime.
38
Mathematically, *Country = Country ° State ° Commune, where ° denotes mapping composition
(see Appendix).
39
By the statement:
SELECT CITIES.x, Country
FROM (STATES INNER JOIN COMMUNES
ON STATES.x = COMMUNES.State) INNER JOIN CITIES
ON CITIES.Commune = COMMUNES.x;
whereas, for example, *Country(4) (which is equal to 4) is computable by:
SELECT Country
FROM (STATES INNER JOIN COMMUNES
ON STATES.x = COMMUNES.State) INNER JOIN CITIES
ON CITIES.Commune = COMMUNES.x
WHERE CITIES.x = 4;
The
 Quest for Data Independence, Minimal Plausibility, and Formalization 135

FIGURE 3.5  Tables COMMUNES for big cities and CITIES for small ones.

For example, in Fig. 3.5, columns x from both tables, Commune and State
from COMMUNES, as well as City, Commune, *Country, and Zip from CIT-
IES are prime, whereas columns Population from both tables are nonprime.
Just like RDM itself, defined as such, primeness and nonprimeness are
purely syntactic concepts: they are applied only after the set of table keys
are established.
As it will become clear why in (E)MDM (see the second chapter of the
second volume), we need in fact slightly different semantic definitions for
them, to be applied before the set of all keys for a table is established: a
136 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

column is said to be prime in any given fixed subuniverse, if in that subuni-

verse it is either a key or might be part of a key and nonprime otherwise.
Consequently, in particular, this semantic definition of nonprimeness
(the only interesting one, in fact, as primeness is somewhat the default
one) of a column is: “not only it is not a single key, but it cannot be part of
any concatenated one either” (in that particular subuniverse).
For example, generally, except for very few particular contexts (e.g.,
tables for time zones or colors), very many other numeric properties, like
population, area, temperature, pressure, altitude and their dual (rivers,
lakes, seas, oceans, etc.) depth, length, width, GDP, rating, probabilities,
percentages, time zones, etc., some text ones like color, notes, even some
names (like, for example, states/county ones in different countries), even
some calendar dates (like, for example, discovery, national dates, etc.),
etc., as well as most of the Boolean ones (e.g., Married?, Divorced?,
Widow?, Author?, Holiday?, etc.) are nonprime.
Algorithm A7/8–3 (see Section 3.6), which is assisting validation of
the keys declared for any table and discovery of any other existing ones,
proves that correctly declaring of all nonprime mappings is crucial in sig-
nificantly reducing the otherwise huge number of possible cases to analyze
by this exponential algorithm.
This is why we are using the above-introduced semantic concept of
nonprimeness, instead of the primeness syntactic one.

3.2.3.5  Keys and Null Values

True genius resides in the capacity for evaluation of uncertain,

hazardous, and conflicting information.
—Winston Churchill

As we have already seen, by definition, primary keys do not allow for

null values. Unfortunately, some RDBMSs (e.g., MS SQL Server40) do
not allow for null values in any other key (implicitly thus assuming that
NULLS is not infinite, but only has one value).
Unfortunately, there are only two types of solutions for overcoming
such aberrant restrictions:
40
Which is quite amazing, as MS Access allows for them and there are decades already since Microsoft
bought the SQL Server license from Sybase.
The
 Quest for Data Independence, Minimal Plausibility, and Formalization 137

ü (the “hard” one) enforce such keys through db management appli-

cation code (i.e., trigger-type methods) or
ü (the “soft” one) automatically generate explicit “unknown” rows,
one per each involved table (e.g., if in table COMMUNES from
Fig. 3.5 above STATES should also conceptually accept nulls, then
an “unknown” state should be generated in corresponding table
STATES for each country in COUNTRIES and these “unknown”
states be used in column COMMUNES.State instead of nulls).
Dually, it is obvious that when nulls are accepted by RDBMSs for
nonprimary keys, but they should not be accepted conceptually (like, for
example, having two files with no extensions and same names in a same
folder), there is only one solution: declare as key the corresponding (con-
catenation of) column(s) (in the above example: Folder • FileName • File-
Extension) and enforce subkeys as keys (Folder • FileName, in the above
example) for null values through db management application code (trig-
gers or trigger-type methods).

3.2.4  FOREIGN KEY (REFERENTIAL INTEGRITY)

CONSTRAINTS

Integrity without knowledge is weak and useless,

and knowledge without integrity is dangerous and dreadful.
—Samuel Johnson

Links between tables, as well as those between a table and itself are done
by foreign keys: pointer-type columns whose values should always be
among the values of the corresponding referenced columns. For exam-
ple, from Fig. 3.2, Composer and Tonality in MUSICAL_WORKS, from
Fig. 3.4, Capital in both COUNTRIES and STATES, Country in STATES,
State in CITIES and from Fig. 3.5, State in COMMUNES and Commune
and *Country in CITIES are all foreign keys; for example, by its defini-
tion, values of *Country are always among the values of STATES.Country,
which, in their turn, are always among the values of COUNTRIES.x.
This is why, on the second table header row, instead of a corresponding
domain constraint, for all foreign keys is shown the so-called referential
138 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

integrity constraint41 (or (typed) inclusion dependency), stating that their

values should always be among the values of the corresponding referenced
column.
This does not mean that foreign keys do not have associated domain
constraints: it only means that these constraints are derived ones, from the
corresponding referential integrity constraints, so they do not need to be
explicitly asserted. For example, in Fig. 3.4, from the fact that COUNTRIES.
Capital should always take its values from within CITIES.x values, it fol-
lows that the former’s domain constraint is the same as the one of the latter’s
(i.e., autonumber(16)). However, note that, unfortunately, RDBMSs do not
automatically derive corresponding domain constraints; even worse, you are
sometimes free to oversize foreign key domain constraints: they should only
have same data type as corresponding referenced columns and be able to
accommodate all of its values (i.e., at least, they cannot be undersized).
The referential integrity constraint between foreign key f and corre-
sponding referenced column g it’s generally denoted by f # g42. Obvi-
ously, this is a notational abuse43: in fact, its meaning is Im(f) # Im(g),
where Im is the image operator, which computes the set of values taken
by its operand (i.e., the set of all of the values taken by the correspond-
ing mapping). This is why, didactically, I prefer to denote such referential
integrity by Im(g) instead of, simply, g.
Note that failing to enforce this constraint44 may result in storing the
so-called dangling pointers: values that point to nonexisting values in the
set of values of the referenced columns.
For example, if first data row from table COUNTRIES of Fig. 3.4 above
were containing 7 (or any other number except those from 1 to 6) instead
of 6 in column Capital, the corresponding referential integrity would be
violated (as there is no row in the CITIES tables having 7 in its x column)
and this 7 would consequently be a dangling pointer; in this case, the capi-
tal of the U.S. could not be actually computed from that rdb instance.
41
Note that some RDBMSs use confusing terminology in this respect too: for example, MS Access
table “design” mode considers referential integrity as being only half of this constraint type, the other
half being enforceable through the “limit to list” property of foreign keys.
42
This being the reason why RDM theory also refers to these constraints as being (typed) inclusion
dependencies.
43
Columns being mappings, it is senseless to talk about mapping inclusions.
44
It may be done, for example, through the ALTER TABLE schemeName.tableName ADD
CONSTRAINT constraintName FOREIGN KEY (columnName) REFERENCES table-
Name (colName) statements, see Section 3.8 below.
The
 Quest for Data Independence, Minimal Plausibility, and Formalization 139

Referential integrity constraints might be violated from both sides of

the inclusion mathematical relation: as just seen above, from the left one
by storing in a foreign key column a value which does not belong to the
referenced column, but, dually, from the right one too, by deleting from a
table a row which is referenced by at least one row.
For example, if the fifth COUNTRIES’ row of Fig. 3.4 above (cor-
responding to Romania) were deleted, then table STATES would become
inconsistent, as its fourth line (corresponding to Bucharest) would contain
a dangling pointer value in column Country.
Also note that, as already discussed above, RDM allows for concatenated
foreign keys and, moreover, for foreign keys referencing any columns, not
necessarily being keys. Best practice, however, is to only use single foreign
keys, always referencing primary keys (which should always be surrogate-
type appropriately range restricted numeric ones—see Subsection 3.2.3.1).
As RDM, however, allows for defining of any foreign keys, provisions
are made in SQL for easing referential integrity constraints: foreign keys
definitions may also include the predicates “ON UPDATE | DELETE
[RESTRICT | CASCADE | SET NULL | NO ACTION]”.
If you only use single foreign keys referencing primary autonumber
columns, you do not need any such “ON UPDATE” predicate: as autonum-
bers may not be updated, no such problem arises. On the contrary, when
you use foreign keys referencing nonautonumber columns (which may
also be the case of subset-type tables primary keys—see Subsection
3.2.4.3 below) you should decide whether not to allow referenced col-
umn’s values modification (with “ON UPDATE RESTRICT”, generally
the default value), or to allow them and either ask the system to automati-
cally also update corresponding foreign key values (with “ON UPDATE
CASCADE”), or to update all these latter values by replacing them with
nulls (with “ON UPDATE SET NULL”), but this should be possible (i.e.,
corresponding foreign keys should allow nulls). By choosing in such cases
“ON UPDATE NO ACTION”, dangling pointers might appear.
As the “ON DELETE” predicate has similar behavior for the above
four parameters, cascading deletes is very risky and should therefore be
very cautiously used: for example, let us consider a db storing details
for all phone calls, which are detailed invoice lines for telecom invoices
of worldwide customers, in which all corresponding foreign keys (from
phone calling details to phone calls, from phone calls to invoices, from
140 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

invoices to customers, from customers to their corresponding cities,

from cities to states, from states to countries, and from countries to
continents) have been defined with the predicate “ON DELETE CAS-
CADE”; if the row corresponding to North America is deleted from the
continents table, cascading will also automatically delete the three rows
corresponding to Canada, U.S., and Mexico from the countries tables,
then all some 100 rows of their corresponding states from the states
table, then all probably some 1,000 rows of corresponding cities from
the cities table, then probably some corresponding hundred thousand
rows from the customers table, millions of rows from the invoices and
phone calls ones, as well as some 1 billion rows from the phone call
details one.
Consequently, not only for commercial dbs, “ON DELETE RESTRICT”,
which does not allow deleting referenced rows, is generally the best choice.
Note that various RDBMSs versions are generally offering only some
of the above parameters and that even corresponding default values are
varying.45

3.2.4.1  The Key Propagation Principle

When will the world know that peace and propagation

are the two most delightful things in it?
—Horace, 4th Earl of Orford Walpole

Foreign keys arouse from the following Key Propagation Principle (KPP),
to be explained in what follows by an example (also see Sections 2.4 and
3.12): let us consider the tables COUNTRIES and STATES from Fig. 3.4,
except for the foreign key column Country of STATES. In order to be able
to link data from these tables, we should store for each state the country
to which it belongs—that is the graph of mapping Country : STATES →
COUNTRIES (as each state belongs to one and only one country).
As for any finite mapping f : A → B, the first (brute force type) idea
of storing its graph is a table with two columns: x (from A) and f(x) (from
45
Probably the most famous example is MS Access: when programmatically creating a foreign key,
the default is “ON DELETE RESTRICT”, whereas when interactively creating it, for example, with
its Lookup Wizard (but not only), the default is “ON DELETE NO ACTION”.
The
 Quest for Data Independence, Minimal Plausibility, and Formalization 141

B). Table 3.2 shows this solution for the graph STATES_COUNTRIES of
Country : STATES → COUNTRIES, where, for optimality reasons, State
references STATES.x and Country references COUNTRIES.x.
Both from A and B, keys should be chosen for storing f’s graph (in
order to uniquely identify any element, both from A and B). As all keys
of a set are equivalent (see Exercise 3.42(iv)), for any mapping f, any key
from A could be chosen as x and any key from B could be chosen as f(x).
For optimality reasons, however, the best choice for them is always the
two involved surrogate primary keys.
When we add this table to the rdb scheme in Fig. 3.4 it is almost impos-
sible not to notice that both columns x and State of it are redundant, as they
are duplicating values of column x from STATES. Consequently, an equiv-
alent but much more elegant and optimal solution to this data modeling
problem is simply adding the foreign key Country from STATES_COUN-
TRIES to table STATES, instead of adding the three columns of STATES_
COUNTRIES, out of which two are redundant.
This process is called key propagation for obvious reasons: for any
such mapping (i.e., functional relationships between two tables), accord-
ing to KPP, instead of adding a new table, it is better to propagate in the
domain table46 a key from the codomain one47. This is also why propagated
columns are called foreign keys: generally, they take values from keys of
other (foreign) tables.
Country : STATES → COUNTRIES
Note that “foreign” in the “foreign key” syntagm is a false friend: when
applying KPP to auto-mappings (i.e., to mappings defined on and taking
values from a same set), corresponding foreign keys are referencing a key
from the same table (not from a “foreign” one).
For example, applying KPP to ReportsTo : EMPLOYEES → EMPLOY-
EES, results in the “foreign” key ReportsTo of table EMPLOYEES that
references its x column, as shown by Table 3.3, where, obviously, the first
employee does not report to anybody, the second one reports to the first
one, whereas the third one reports to the second one.
46
Generally, the “many” one: in the above example, STATES. For the one-to-one ones, it is, generally,
the one potentially having less rows than the other, like COUNTRIES (as compared to CITIES) for
Capital (see Fig. 3.4).
47
For those including one “many”, it is the “one” one: in the above example, COUNTRIES. For the
one-to-one ones, it is, generally, the one potentially having more rows than the other, like CITIES (as
compared to COUNTRIES) for Capital (see Fig. 3.4).
142 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

TABLE 3.2  Brute Force Type Graph Storage Implementation of Mapping Country :
STATES → COUNTRIES

TABLE 3.3  An Example of a Foreign Key Which is Actually Not “Foreign” At All

3.2.4.2  Orthogonality of Unique and Foreign Keys

In God we trust. All others must bring data.

—W. Edwards Deming

In fact, the syntagm “foreign key” is a double false friend: not only “for-
eign” is a false friend, but also “key” is one, as, generally, foreign keys are
not also (unique) keys.
For example, let us consider table COUNTRIES from Fig. 3.4 aug-
mented with a foreign key Currency (referencing a table CURRENCIES).
Obviously, Capital is both a key and a foreign key, Country is a key but
not a foreign key, Currency is a foreign key but not a key (as there are, for
The
 Quest for Data Independence, Minimal Plausibility, and Formalization 143

TABLE 3.4  Proof of Orthogonality of Unique and Foreign Key Concepts

example, lot of countries having Euro as currency), whereas Population is

neither a key nor a foreign key. Table 3.4 is summarizing these findings,
which prove that the concepts of unique and foreign keys are orthogonal
to each other.

3.2.4.3  Modeling Set Inclusions

Most of the fundamental ideas of science are essentially simple, and

may, as a rule, be expressed in a language comprehensible to everyone.
—Albert Einstein

Let us consider SUBORDINATED_COUNTRIES # COUNTRIES that is

the subset of countries and territories that are subordinated to other coun-
tries; if we need to store their corresponding subordinations together with
their starting subordination years, the best solutions is to add a SUBORDI-
NATED_COUNTRIES table as shown in Fig. 3.6.
We could have defined its surrogate primary key x as being of type
autonumber too, but then we need an additional foreign key Country refer-
encing COUNTRIES for storing subordinated countries. The solution from
Fig. 3.6 is better, as it consolidates Country into x.
Generally, the best way to represent relationally an inclusion T # S is
by declaring primary surrogate key T.x as being a foreign key referencing
primary surrogate key S.x. T.x may not be an autonumber: it should be of
number type (and maximum cardinality equal to the one of S.x).
Mathematically, any set is a subset of its own (even if, just like the
empty set, an “improper” one), but you should never try to define such
inclusions, which would recursively include a table into itself: while, for
example, MS Access is not letting you define such trivial inclusions (T # T),
144 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

FIGURE 3.6  Example of relational modeling of a set inclusion.

even the current 12c version of Oracle doesn’t have anything against it! It
is true that the only harm done by enforcing such trivial inclusions is that
it is wasting for nothing additional time for enforcing the corresponding
referential integrity constraint for each row insertion into the correspond-
ing table.
Moreover, pay attention to inclusion cycles too: recall from sets alge-
bra that, for any sets S1, S2, …, Sn, n natural, S1 # S2 # … # Sn # S1 
S1 = S2 = … = Sn (i.e., all sets involved in a cycle of inclusions are equal).
First of all, equal sets should be relationally modeled by using only one
table for all of them; but what is much, much worse is that most RDBMSs
(including, for example, MS Access 2010 and Oracle 12c) are allowing
you to define such cycles (even when they only involve two tables!),
which results in not being then able to insert any line in any of the involved
The
 Quest for Data Independence, Minimal Plausibility, and Formalization 145

tables. Consequently, we should always keep table inclusions acyclic—

that is we should never close cycles of included sets.

3.2.5  CHECK (TUPLE) CONSTRAINTS

The more constraints one imposes, the more one frees one’s self.
—Igor Stravinsky

For check constraints there is no generally agreed definition; from most of

RDBMSs implementations, they are first order logic formulas that have to
be satisfied by all rows of a table and have (by notational abuse) a simpli-
fied, propositional logic type form: they are using parenthesis, the logical
not, and, and or operators connecting terms having the form C θ D, where
all such C and D (from all terms) are columns of a same table, and θ is a
standard operator48 (this being the reasons why these constraints are also
called tuple constraints, as they relate columns of a same table that have to
be satisfied by all tuples of that table).
For example, for table COMPOSERS of Fig. 3.2, check constraint
BirthYear + 3 < PassedAwayYear states that, for each row, corresponding
life duration should have been at least 3 years (assuming that no one could
ever compose before he/she is at least 3 years old). Both check constraints
of this table may be consolidated into a single one, by using either the
logical and (as (BirthYear + 3 < PassedAwayYear) and (PassedAwayYear
— BirthYear < 120)) or the logical between … and (as PassedAwayYear —
BirthYear between 3 and 120).
In fact, check/tuple constraints are first order logic formulas with only
one variable universally quantified, which, by notational abuse, is omitted.
For example, the above apparently propositional calculus formula stands
for (;x[COMPOSERS)(PassedAwayYear(x) — BirthYear(x) between 3
and 120).
The limitation to only one table is essential: although syntactically
looking similar, no RDBMS is recognizing (and, consequently, enforcing),

48
The set of db standard operators always include the corresponding math ones, plus some additional
ones, either derived from them, like between ... and, or generally coming from regular expressions
manipulation, as like (which uses, just like in OSs, meta-characters too: ‚%’ or ‚*’ for any character
string, including empty ones, etc.).
146 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

for example, constraints like COUNTRIES.Population ≥ STATES.Popula-

tion (see Fig. 3.4).
Note that you should not get confused because of theory and RDBMS
terminologies: for example, not-null constraints are also considered by
Oracle as being check constraints.

3.3  THE ALGORITHM FOR TRANSLATING E-R DATA MODELS

INTO RELATIONAL SCHEMAS AND NON-RELATIONAL
CONSTRAINT SETS (A1–7). AN RDM MODEL OF THE E-RDM

There are truths that you cannot justly understand without previously
experiencing some errances.
—Lucian Blaga

In order to further refine especially restriction sets, but sometimes also

even E-RDs, conceptual data modeling should not end only with establish-
ing E-RDs and corresponding restriction sets, as it will be proved in the
next volume.
However, when in a hurry, when accuracy of conceptual data modeling
is not that big a concern, one may directly translate E-RDs and restriction
sets into relational db schemas and non-relational db constraint sets by
applying the algorithm presented in Subsection 3.3.1.

3.3.1  THE ALGORITHM FOR TRANSLATING E-R DATA

MODELS INTO RELATIONAL SCHEMAS AND NON-
RELATIONAL CONSTRAINT SETS (A1–7)

An algorithm must be seen to be believed.

—Donald Knuth

The algorithm is shown in Figs. 3.7–3.1149. Note that A1–7 assumes the
following: the corresponding rdb exists and there are no duplicated object
set names or mapping names defined on a same object set.
49
whose complexity is linear in the sum of object sets collection, property and restriction sets cardi-
nals, as proved in Section 3.12 (see Proposition 3.4).
The
 Quest for Data Independence, Minimal Plausibility, and Formalization 147

FIGURE 3.7  Algorithm A1–7 (Translation of E-R data models into corresponding
relational schemas and non-relational constraint sets).

3.3.2  AN RDM MODEL OF THE E-RDM

Learning is a treasure that will follow its owner everywhere.

—(Chinese Proverb)

Section 3.4 is applying the algorithm A1–7 to the public library E-RD and
restriction set designed in Section 2.8.
Solved exercise 3.0 provides another example of its application (to a
subuniverse of invoice payments).
By applying this algorithm to the E-R data metamodel of the E-RDM
presented in Section 2.9 above, the RDM model of the E-RDM shown in
148 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

FIGURE 3.8  Algorithm A1–7’s addForeignKey method.

FIGURE 3.9  Algorithm A1–7’s createTable method.

Fig. 3.12 is obtained (regardless of the fact that it is applied to the E-RD
from Subsection 2.9.1 or to the one from Subsection 2.11.2).
Note that the unlabeled autofunctions of OBJECT_SETS, DATA_
TYPES and RANGES, which are modeling the partial ordering of the ele-
ments of these sets by inclusion, have all been labeled Superset, because
their graphs store for each such element the corresponding smallest super-
set that includes them too.
For example, ASCII(16) , ASCII(32) , ASCII(64) , ASCII(128) ,
ASCII(255), [2, 16] , [1, 16] , [1, 255] , NAT(38) , INT(38) , RAT(38,
2), CURRENCY(38) , RAT(38, 2), [–2500, Year(SysDate())] , INT(38),
[6/1/2011, SysDate()] , [6/1/2011, SysDate() + 300] , DATE, etc.
The
 Quest for Data Independence, Minimal Plausibility, and Formalization 149

FIGURE 3.10  Algorithm A1–7’s completeScheme method.

FIGURE 3.11  Algorithm A1–7’s addColumn method.

This rdb may be used as the kernel of the metadata catalog (see Section
3.7) of a DBMS providing E-RDM to its users.
The example instance provided corresponds to the public library case
study from the Section 2.8 (also see Section 3.4).
Note that there is no need (in this kernel metacatalog) for a table
ENTITY_TYPES, as it would only have an x primary column, which would
duplicate the one of OBJECT_SETS.
The associated non-relational constraint set is made out of the follow-
ing two constraints:
150 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

FIGURE 3.12  A RDM model of the E-RDM and a valid instance of it corresponding to
the public library case study from Section 2.8.

– For each nonfunctional relationship-type object set there

should be at least two underlying object sets. (RU5)
– The graph of UNDERLYING_SETS should be acyclic
(that is no non-functional relationship-type object set may
The
 Quest for Data Independence, Minimal Plausibility, and Formalization 151

FIGURE 3.12  (Continued)

be defined on itself, neither directly, nor indirectly;

equivalently: recursive definitions of non-functional
relationship-type object sets are not allowed). (RU6)
Note that constraints (RO6) and (RM7) need no explicit enforcement,
as they are implicitly enforced by the closed world assumption and the
structure of the rdb.
Also note that any software application built on top of this rdb should
also automatically (re)compute (immediately after inserts, deletes or
updates in column RelationshipType of table UNDERLYING_SETS) the
column *Arity of table NON_FUNCTIONAL_RELATIONSHIP_TYPES.
152 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

FIGURE 3.12  (Continued)

3.4  CASE STUDY: THE RELATIONAL SCHEME OF THE PUBLIC

LIBRARY DATA MODEL

Science is what we understand well enough to explain to a computer.

Art is everything else we do.
—Donald Knuth

As an example, by applying Algorithm A1–7 on the E-RD and its associ-

ated restriction set for a public library (see Sections 2.8.2 and 2.8.3), the
following relational db scheme (see Section 3.4.2) and non-relational con-
straint set (see Section 3.4.1) are obtained:

3.4.1  PUBLIC LIBRARY NON-RELATIONAL CONSTRAINTS SET

A list is only as strong as its weakest link.

—Donald Knuth
The
 Quest for Data Independence, Minimal Plausibility, and Formalization 153

FIGURE 3.12  (Continued)

1. PERSONS
 FName and e-mail should be compulsory for
subscribers (RP7)
154 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

FIGURE 3.12  (Continued)

The
 Quest for Data Independence, Minimal Plausibility, and Formalization 155

2. VOLUMES_CONTENTS = (BOOKS, VOLUMES)

For any edition, its first book should be the first one
published in its first volume. (RVC4)
No edition may contain same book more than once. (RVC5)
3. BORROWS_LISTS = (BORROWS, COPIES)
No copy may be borrowed less than 0 days or more than
300 days. (RBL5)
No copy may be simultaneously borrowed to more than
one subscriber. (RBL6)
No copy may be returned before it was borrowed or after
100 years since corresponding borrow date. (RBL7)
(RP7) might seem as being the conjunction of two not-null constraints
(one on FName and the other on e-mail), but, in fact, it is not, because it
does not apply to all rows of PERSONS, but only to some of them (those
corresponding to subscribers).
(RVC4) and (RVC5) are not relational as they involve columns from
three tables: x and FirstBook (from EDITIONS), x and Edition (from VOL-
UMES), as well as Volume, Book and BookPos (from VOLUME_CON-
TENTS).
Similarly, (RBL5) involves columns x and BorrowDate (from BOR-
ROWS), as well as Borrow and DueReturnDate (from BORROWS_LISTS);
(RBL6) involves x, Subscriber and BorrowDate (from BORROWS), as
well as Borrow, Copy and ActualReturnDate (from BORROWS_LISTS);
(RBL7) involves x and BorrowDate (from BORROWS), as well as Borrow
and ActualReturnDate (from BORROWS_LISTS).

3.4.2  PUBLIC LIBRARY DB RELATIONAL SCHEME AND A

PLAUSIBLE VALID INSTANCE OF IT

The hardest thing is to go to sleep at night, when there are so many urgent
things needing to be done. A huge gap exists between what we know is pos-
sible with today’s machines and what we have so far been able to finish.
—Donald Knuth

Figure 3.13 shows the corresponding public library rdb and a possible
plausible instance of it.
156 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

FIGURE 3.13  Public library rdb and a possible instance of it.

3.5  THE REVERSE ENGINEERING ALGORITHM FOR

TRANSLATING RELATIONAL SCHEMAS INTO E-R DATA MODELS
(REA1–2)

The softest things in the world overcome

the hardest things in the world.
—Lao Tzu

Very often, probably much more often than having opportunities to design
a new db, we are faced with a dual problem, of reverse engineering (RE)
The
 Quest for Data Independence, Minimal Plausibility, and Formalization 157

FIGURE 3.13  (Continued)

158 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

FIGURE 3.13  (Continued)

type: we have to work with (and, so, understand) a db for which there are
no E-RDs and/or restriction set and/or description. This section presents,
in Fig. 3.14, algorithm REA1–2 that outputs these three items of an E-R
data model from any RDM scheme.
The
 Quest for Data Independence, Minimal Plausibility, and Formalization 159

FIGURE 3.14  Algorithm REA1–2 for reverse engineering RDM schemas into
E-R data models.
160 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

In practice, before applying this algorithm, another RE algorithm has

to be applied, in order to get the corresponding RDM scheme from an
actual rdb (or, instead, apply directly the composition of these two algo-
rithms: see, for example, Section 4.4).
Note that, as RDM is a purely syntactic formalism, the E-RDs obtained
by reverse engineering may never contain relationship-type sets, but only
entity-type ones; however, as proved in Section 2.10 of the second vol-
ume, these E-RDs are strictly equivalent with ones also having relation-
ship-type sets.
Also note that if you do not need computed sets or when the number of
views is high, you may want to apply the above algorithm only for tables
and/or some views.
As an example, let’s consider the RDM scheme from Fig. 3.15;
by applying to it the algorithm REA1–2, the E-RD from Fig. 3.16,
the associated restriction set from Fig. 3.17, and the description from
Fig. 3.18 (obtained by E-RD and restrictions reverse engineering) are
obtained.
Note that, in fact, NEIGHBORS, which stores pairs of countries <c,
d> whenever c and d are neighbors, is a non-functional binary math-
ematical relation (i.e., a many-to-many relationship) defined on COUN-
TRIES, so, most probably, when applying algorithm A0 (see Section
2.9), one would rather model it as a diamond. Nothing differs, however,
at the rdb level for this table: when applying algorithm A1–7 above to the
NEIGHBORS diamond exactly the same table as in Fig. 3.15 is obtained
(see Exercise 3.1).
In Volume 2, we will see that, generally, any relationship-type set may
be equivalently modeled with an entity-type one, providing that all of
its roles are included too as corresponding functional relationships. This
means that the only fundamental mathematical relations needed in data
modeling are the functions.50

3.6  THE ALGORITHM FOR ASSISTING KEYS DISCOVERY (A7/8–3)

50
Moreover, this implies that E-RDM and functional data models are strictly equivalent as expressive
powers.
The
 Quest for Data Independence, Minimal Plausibility, and Formalization 161

FIGURE 3.15  An rdb example.

162 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

FIGURE 3.15  (Continued)

FIGURE 3.16  The E-RD corresponding to the rdb example from Fig. 3.15.

Do the difficult things while they are easy and do the great things while
they are small. A journey of a thousand miles must begin with a single step.
—Lao Tzu

Generally, even db architects and designers are designing very few new
dbs and are modifying/using much more other existing ones, which very
rarely have all constraints (and, consequently, all keys as well) enforced.
The
 Quest for Data Independence, Minimal Plausibility, and Formalization 163

FIGURE 3.17  The restriction set associated to the E-RD from Fig. 3.16.
164 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

There may be at most 100 continents, characterized only by their

compulsory names, which may have at most 64 characters; there
may not be two continents having same names.
At most 1,000,000 pairs of neighbor countries are also stored,
only once.
There may be at most 1,000 countries, characterized by their com-
pulsory names (at most 255 characters) and codes (at most 8 char-
acters), each one of them belonging to only one continent (which is
mandatory too); there may not be two countries having same names
or codes in a same state. When known, country capitals are stored
too; no city may simultaneously be the capital of two countries.
There may be at most 100,000,000 states, characterized by their
compulsory names (at most 128 characters) and codes (at most 8
characters), each one of them belonging to only one country (which
is mandatory too); there may not be two states having same names
or codes in a same country.
There may be at most 10,000,000,000 cities, characterized by
their compulsory names (at most 128 characters), each one of them
belonging to only one state (which is mandatory too); there may not
be two cities having same names or codes in a same state.

FIGURE 3.18  The description corresponding to the rdb example from Fig. 3.15.

As we will see in the second chapter of the second volume, keys should
be designed earlier, on higher conceptual levels, not on the RDM one or,
even lower, on the RDBMS managed rdbs one, but better later than never.
It is true that both this chapter, as well as the second chapter of the
second volume are also providing reverse engineering algorithms, which,
given any rdb scheme, are computing corresponding higher conceptual
level schemas, so db designers might first apply them for obtaining an
E-R data model, then apply A1 to translate this model into a(n) (E)MDM
scheme, A2 to A6 (which includes A3/3′) to refine it and then A7, A8, and
the needed algorithm from AF8’, in order to obtain the new rdb scheme
with the keys designed according to A3 or A3′.
The
 Quest for Data Independence, Minimal Plausibility, and Formalization 165

However, in order not to be obliged to apply all of these algorithms

(especially for legacy rdbs, but not only), this section is introducing in Fig.
3.19 A7/8–3, an equivalent algorithm for assisting discovery of all keys in
any table (where “//” are prefixing comments; [x], x real, denotes the inte-
ger part of x –for example, [5/2] = 2, [π] = 3; C(n, j) denotes the (natural)
number of “n choose j”, n and j naturals, n ≥ j,—that is combinations of j
elements taken from n ones: C(n, j) = n!/(j! * (n—j)!), where x! = 1 * 2 *
… * (x–1) * x is the factorial of x, x natural; for example, C(4,2) = 4!/(2! *
2!) = 4 * 3 * 2/(2 * 2) = 6– see Appendix too; and statement numbers are
shown for easing the understanding of the following example of applying
A7/8–3).
Let us apply, for example, Algorithm A7/8–3 to the Table 3.5 (also pre-
sented in Fig. 3.4, here augmented with two new columns for state phone
prefixes and areas):
T = STATES, m = 8, C1 = x, …, C8 = Area, K = { x, Capital, State •
Country }, k = 3.
Note that, in order to simplify things, it is considered that each state has
only one (national) telephone prefix, which is true only for small states/
countries; for cases where there are several such prefixes per state (as in
NY, for example), you can consider TelPrefix as storing the first one of
them.
Results of running Algorithm A7/8–3 for this input step by step are the
following (only “actual” statements being numbered: repeat for, else, end
…, etc. are not):
01. K’ = { x, State • Country, Capital };
02. x is a key, by definition (being a surrogate column);
02. Capital is a key, as there may not be cities that are (simultaneously)
capitals of two (or more) countries; as there are no superkeys of it, K’
remains the same;
02. State • Country is a key, as there may not be two states of a same
country having same names; as there are no superkeys of it, K’ remains
the same;
05. SK = { x, State, Country, Capital };
06. T’ = { StateCode, Population, TelPrefix, Area };
166 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

FIGURE 3.19  Algorithm A7/8–3 (table keys discovery assistance).

51

51
Depending on RDBMS versions, some data types, generally those needing “huge” storage sizes
(for example memos, pictures, etc.) are not actually allowed to be declared as prime.
The
 Quest for Data Independence, Minimal Plausibility, and Formalization 167

FIGURE 3.19  (Continued).

07. StateCode is not one-to-one (as there may be two states hav-
ing same code, but from different countries), so neither T’, nor K’ are
modified;
10. StateCode is prime (as there may not be two states having same
code in a same country), so T’ is not modified;
07. Population is not one-to-one (as there may be two states hav-
ing same population, even in a same country), so neither T’, nor K’ are
modified;
168 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

TABLE 3.5  Input Example Table for Applying Algorithm A7/8–3.

The
 Quest for Data Independence, Minimal Plausibility, and Formalization 169

10. Population is nonprime (as it cannot help, in any context, to

uniquely identify states), so T’ = { State, StateCode, Country, TelPrefix,
Area };
07. TelPrefix is not one-to-one (as there may be two states having same
telephone prefixes, but from different countries), so neither T’, nor K’ are
modified;
10. TelPrefix is prime (as there may not be two states having same tele-
phone prefixes in a same country), so T’ is not modified;
07. Area is not one-to-one (as there may be two states having same
area, even in a same country), so neither T’, nor K’ are modified;
10. Area is nonprime (as it cannot help, in any context, to uniquely
identify states), so T’ = { StateCode, TelPrefix };
11. T’= { State, StateCode, Country, TelPrefix };
12. n = 4;
13. as 4 > 1:
14. l = 3;
15. kmax = C(8, 4) = 8!/(4! * 4!) = 8 * 7 * 6 * 5 * 4 * 3 * 2/(4 * 3 * 2 *
4 * 3 * 2) = 8 * 7 * 6 * 5 /(4 * 3 * 2) = 7 * 2 * 5 = 70;
16. i = 2;
17. allSuperkeys = false;
18. as 2 ≤ 4 and 3 < 70 and true = true:
19. allSuperkeys = true;
20. State • StateCode is not already known to be a key, so:
21. State • StateCode is not a superkey, so:
22. allSuperkeys = false;
23. State • StateCode is not a key, as there may be two states having
same names and codes in different countries
20. State • Country is already known to be a key, so it is skipped
20. State • TelPrefix is not already known to be a key, so:
21. State • TelPrefix is not a superkey, so:
22. allSuperkeys = false;
23. State • TelPrefix is not a key, as there may be two states having
same names and telephone prefixes in different countries
20. StateCode • Country is not already known to be a key, so:
170 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

21. StateCode • Country is not a superkey, so:

22. allSuperkeys = false;
23. StateCode • Country is a key, as there may not be two states having
same code in a same country
24. K’ = { x, State • Country, Capital, StateCode • Country };
25. l = 4;
20. StateCode • TelPrefix is not already known to be a key, so:
21. StateCode • TelPrefix is not a superkey, so:
22. allSuperkeys = false;
23. StateCode • TelPrefix is not a key, as there may be two states having
same codes and telephone prefixes in different countries
20. Country • TelPrefix is not already known to be a key, so:
21. Country • TelPrefix is not a superkey, so:
22. allSuperkeys = false;
23. Country • TelPrefix is a key, as there may not be two states having
same telephone prefixes in a same country
24. K’={x, State • Country, Capital, StateCode • Country, Country •
TelPrefix};
25. l = 5;
26. i = 3;
18. as 3 ≤ 4 and 5 < 70 and true = true:
19. allSuperkeys = true;
20. State • StateCode • Country is not already known to be a key, so:
21. State • StateCode • Country is a superkey (as State • Country is a
key)
20. State • StateCode • TelPrefix is not already known to be a key, so:
21. State • StateCode • TelPrefix is not a superkey, so:
22. allSuperkeys = false;
23. State • StateCode • TelPrefix is not a key, as there may be two states
having same names, codes, and telephone prefixes in different countries
20. State • Country • TelPrefix is not already known to be a key, so:
21. State • Country • TelPrefix is superkey (as State • Country is a key)
20. StateCode • Country • TelPrefix is not already known to be a
key, so:
The
 Quest for Data Independence, Minimal Plausibility, and Formalization 171

21. StateCode • Country • TelPrefix is superkey (as StateCode • Coun-

try is a key)
26. i = 4;
18. as 4 ≤ 4 and 5 < 70 and true = true:
19. allSuperkeys = true;
20. State • StateCode • Country • TelPrefix is not already known to be
a key, so:
21. State • StateCode • Country • TelPrefix is a superkey (as State •
Country is a key)
26. i = 5;
18. as 5 ≤ 4 and 5 < 70 and true = false (because 5 > 4), the while loop
is exited
27. 5 ≠ 0, so:
29. as 5 ≠ 1, execution of A7/8–3 ends.

Corresponding output is K’ = { x, State • Country, Capital, StateCode •

Country, Country • TelPrefix } $ { x, Capital, State • Country } = K, l = 5
≥ 3 = k. Table 3.6 shows the corresponding modified table scheme.
Note the following general considerations on Algorithm A7/8–3:
ü In the best case (happening when K is empty and all m columns
of T are keys: only the m iterations of the second repeat loop are
needed, as, initially, k = 0, so the first repeat loop is not even entered
and, at the end of the second one, n = 0, and the algorithm exits at
statement 18), so A7/8–3 only asks m questions: in this case, it only
has to ask for each of the m columns, whether or not it is a key (as
all answers are yes in this case).
ü In the worst case (happening when K contains all possible com-
binations, so k = 2m — 1, which are then all discarded and no key
is discovered, so l = 0), all possible 2m — 1 + 2m + m — 1 = 2m + 1
+ m — 2 combinations are considered (as there are, first, 2m — 1
questions on initially existing keys, all answered with no, then
m questions on nonprimeness, all answered again with no and
2m — 1 questions on uniqueness, all answered with no too, where
the only two ones that are not needed are corresponding to C(m,
0), as there is no such thing as a mapping product made out of no
172 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

TABLE 3.6  Output Example Table for Applying Algorithm A7/8–3 to Table 3.5
The
 Quest for Data Independence, Minimal Plausibility, and Formalization 173

mappings), which means that the complexity of the algorithm is

2m+1 (i.e., exponential in the number of the table columns).
Consequently, in average, this algorithm needs some 2m + m/2 – 1 steps
(i.e., half of the worst case one).
For example, in the worst case with k = 0, for m = 4 there are 19 com-
binations to consider, for m = 5 –36, for m = 6 –69 (still “bearable”!?),
for m = 7 –134 (already “hard”), for m = 8 —263 (extremely “hard”),
for m = 9 –520 (almost actually intractable), for m = 10 –1033 (already
“hopeless”).
Fortunately, for example, in my more than 35 years of professional
career in this field, both in IT and University teaching, I have never
encountered any example having m > 7 (out of thousands of them).
When k > 0, things are worsening, as in the first loop k additional ques-
tions have to be asked and answered, but they might get a little bit better
then, if at least some initial keys are confirmed, as they will correspond-
ingly reduce the number of asked questions on uniqueness and possibly
also on nonprimeness (if single keys are reconfirmed).
Unfortunately, at least theoretically, they also might get really nasty if,
initially, all or very many of the possible 2m – 1 combinations are wrongly
declared as keys.
For example, in the extreme worse case, in which all possible 2m – 1
combinations are wrongly declared as keys, for m = 4 there are 19 + 16–1
= 34 combinations to consider and for m = 10 there are 1033 + 1024 – 1 =
2056 combinations to consider.
Please note that in step 0 of the algorithm, however, even if all of the
2 – 1 possible combinations were wrongly declared as keys and, conse-
m

quently, A7/8–3 needs to deal with them all, in fact it will ask users only
at most C(m, [m/2]) times whether or not some declared keys are actually
keys, as it will automatically discard the remaining 2m – 1 – C(m, [m/2])
ones as superkeys.
For example, if, initially, K = { x, Capital, State • Country, State •
Capital, Country • Capital, State • Capital • Country }, immediately after
users are confirming that Capital is a key, the algorithm is automatically
discarding superkeys State • Capital, Country • Capital, and State • Capi-
tal • Country.
174 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

Fortunately, first of all, although most RDBMSs stupidly allows for

enforcing both keys and superkeys, the vast majority of currently exist-
ing dbs do not have (except, as always, for exceptions, but they are never
systematical –that is including all possible ones–, but rather accidental)
superkeys enforced52.
Secondly, especially for semantic ones, current dbs rather lack of
keys than having too much of them enforced; thirdly, most of the cur-
rently enforced keys are correct ones, as incorrect ones are noticed rather
sooner than later at least by customers, so they need to be fixed almost
immediately.
As such, from a practical point of view, K’s cardinal and instance are
not that much of an issue, as, generally, the former is small and the latter
is almost valid.
ü Consequently, however, especially for m > 5 and card(SK) = 0, it
is crucial to minimize n by correctly identifying all nonprime col-
umns.
ü The algorithm’s bottom-up strategy is the best possible, as it elim-
inates as soon as possible the maximum number of implausible
combinations, namely all superkeys.
For example, supposing that Capital was not already declared as a key,
then STATES’ would have initially 7 elements (instead of 6), meaning that
in the worst case there would have been 27 – 1 = 127 cases, out of which
120 in the while loop; by discovering (in the first repeat loop) that Capital
is a key, the number of cases drops to 26 = 64, out of which only 57 in the
while loop.
Generally, it is easy to see that each single key discovered (i.e., in the
first repeat loop) is halving the initial number of nonsuperkeys combina-
tions of the worst case.
Next, on level two, by discovering, for example, the key StateCode •
Country, out of the initial C(4, 3) + C(4, 4) = 4 + 1 = 5 combinations to be
analyzed on the following two levels, only C(4, 3)/2 = 2 remain.
Generally, by discovering a first key on any level i, 1 < i < n, it can
be shown that the number of combinations to be analyzed on remaining
52
Is it because user common sense, which cannot even believe that such a stupidity exists and even
if, exceptionally, discovers it, it does not use it, as it knows or at least feels that it would only aber-
rantly worsen their db throughput? Most probably yes, as it is hard to believe that, again, except for
exceptions, they are much wiser than most of the RDBMS designers, who should have prevented any
such stupidity.
The
 Quest for Data Independence, Minimal Plausibility, and Formalization 175

levels (i.e., i + 1, i + 2, …) are decreasing with increasing numbers as n

increases, even if, as i increases, savings are less and less impressive, as
they have less and less time to continue growing (e.g., when i = 2, with
1 for n = 3, from 1 to 0, with 3 for n = 4, from 5 to 2, with 12 for n = 5,
from 16 to 4, with 15 for n = 6, from 42 to 27, with 30 for n = 7, from
98 to 68, etc.; when i = 3, with 1 for n = 4, from 1 to 0, with 4 for n = 5,
from 6 to 2, with 6 for n = 6, from 22 to 16, with 14 for n = 7, from 63 to
49, etc.; when i = 4, with 1 for n = 5, from 1 to 0, with 3 for n = 6, from
7 to 4, with 14 for n = 7, from 29 to 49, etc.; when i = 5, with 1 for n = 6,
from 1 to 0, with 3 for n = 7, from 8 to 5, etc.; when i = 6, with 1 for n =
7, from 1 to 0, etc.).
For example, discovering that the last level combination (which is
made out of all n mappings) is a key does not save anything, whereas dis-
covering a first key on the last but one level (i.e., n — 1) only saves one
combination (the last one above).
Discovering subsequent keys on a same level is still decreasing the
number of combinations to be analyzed on remaining levels, but not that
much as the previous ones (as some combinations have been already dis-
carded by these latter).
ü kmax is computed according to an expression obtained through a
theorem presented in section 3.12.
ü Iteration in the while loop should end not only when the maximum
level n combination was analyzed or when the maximum possible
number of keys has been reached, but also when on the previous
level all combinations where superkeys: trivially, on the current as
well as on all subsequent ones, all combinations would be super-
keys too (as they all include superkeys).
Consequently, Boolean variable allSuperkeys is used to detect when
this latter condition holds.
ü A7/8–3 should be initially run once for each table; moreover, each
time a table scheme changes (be it by adding new columns, drop-
ping existing ones, or only changing their semantics), it should
immediately be rerun for that table.
ü All additional keys found by running A7/8–3 should then be
enforced; dually, all dropped ones should be removed (and, for
both of them, the sooner, the better).
176 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

ü A7/8–3 cannot ever loop indefinitely (i.e., dually, it is always end-

ing after a finite number of steps):
(i) input m and k, as well as n, i, l, kmax, and C(n, i) are always
finite (being naturals);
(ii) the first repeat loop is executed for at most 2m – 1 times (in the
worst case, when card(K) = 2m –1, meaning that all possible
keys and superkeys are initially enforced);
(iii) the second repeat loop is executed for at most m times (in the
worst case, when card(SK) = 0, meaning that no single keys are
initially declared);
(iv) the third repeat loop, the one inside while, is always executed
C(n, i) times;
(v) finally, the while loop is executed at most n – 1 times, as i starts
with 2 and is incremented by 1 in each such execution.
ü A7/8–3 always asks only legitimate questions53 (i.e., nonprimeness
ones only for single nonunique columns and uniqueness ones only
for initially enforced keys and then for nonsuperkeys that are not
known already to be keys).
Moreover, A7/8–3 always asks all such possible questions54 (i.e.,
dually, no possible question on nonprimeness or uniqueness is skipped).
ü A7/8–3 is only assisting its users: it is asking all and only legitimate
questions, as it cannot respond to any of them (so it is only syntac-
tically correct, being both sound and complete, the maximum that
such an algorithm may ever reach).
Consequently, the semantic accuracy of its outputs is solely its corre-
sponding user’s responsibility.
ü A7/8–3 above is optimal relatively to its strategy: it asks the mini-
mum number of questions in each possible case and it is doing its
job with minimum number of statement executions.
However, it is not absolutely minimal: for example, the slight variation
of it labeled A3′ is bringing significant improvement in the case when the
only semantic key of a table is the concatenation of all of its non-surrogate
columns (see second volume of this book).
53
Mathematically, such an algorithm is said to be sound (see Appendix).
54
Mathematically, such an algorithm is said to be complete (see Appendix).
The
 Quest for Data Independence, Minimal Plausibility, and Formalization 177

3.7  RDBMS METACATALOGS. RELATIONAL AND E-R DATA

MODELS OF THE RDM

The purpose of models is not to fit the data, but to sharpen the questions.
—Samuel Karlin

3.7.1  RDBMS METACATALOGS

Everyday life is like programming, I guess.

If you love something you can put beauty into it.
—Donald Knuth

To begin with, this section is essentially presenting a relational model of

the RDM: the schemas (and associated valid instances) of the main tables
used by RDBMSs to store metadata, that is data on hosted dbs—the so
called metadata catalog or metacatalog.
The considered tables store metadata on dbs, their tables, columns,
and associated relational constraints. Actual metacatalogs have very
many other columns for these kernel tables, as well as very many other
tables (e.g., for users, groups of users, access right types, access rights
of users and groups to db objects, views, triggers, stored functions and
procedures, etc.).
Metacatalogs also store metadata on their schemas. Conventionally,
we can consider that each RDBMS comes with a built-in system db (that
we will call here System) storing metadata on its catalog, which is gener-
ally hidden to users, but accessible through read-only views. Besides it,
for simplicity, we will consider in what follows a very simple user rdb
called BigCities, made out of the two tables (COUNTRIES and CITIES)
presented in Fig. 3.20.
Note that both tables have a primary key (denoted x), that COUNTRIES
has two semantic keys (Country, as there may not be two countries having
same name, and Capital, as no city may simultaneously be the capital of
more than one country), whereas CITIES has only one such key (the prod-
uct City • Country, as it is assumed that no country has several big cities
with the same name).
178 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

FIGURE 3.20  BigCities db.

Country (maximum 128 ASCII characters) and city (maximum 64

ASCII characters) names are mandatory, just as the country to which any
city belongs. Capital is a foreign key referencing the CITIES’ primary one
and Country from CITIES is a foreign key referencing the COUNTRIES’
primary one.

3.7.2  A RELATIONAL METAMODEL OF THE RDM

God is a challenge because there is no proof of his existence and

therefore the search must continue.
—Donald Knuth

Table 3.7 shows the metacatalog table DATABASES and its instance for the
two dbs above: System and BigCities. Note that db names (maximum 255
ASCII characters) are mandatory and unique (for any RDBMS).
This table actually has other columns too (e.g., DateCreated, Pass-
word, Status, etc.) and associated constraints.
The
 Quest for Data Independence, Minimal Plausibility, and Formalization 179

TABLE 3.7  The DATABASES Metacatalog Table

TABLE 3.8  The RELATIONS Metacatalog Table

Table 3.8 presents the metacatalog table RELATIONS and its instance
for the two dbs above. Note that table (relation) names (maximum 255
ASCII characters) are mandatory and unique within any db. As any col-
umn belongs to only one table and primary keys are columns, the Pri-
maryKey foreign key (referencing the CONSTRAINTS’ primary one) is a
semantic key too. As primary keys are optional, this column is not manda-
tory. The foreign key Database (referencing DATABASES’ primary one) is
mandatory, as any table belongs to a db. The first seven rows correspond
to the System metacatalog tables, while the last two to the BigCities tables.
This table actually has many other columns too (e.g., DateCreated,
Tablespace, Status, Cardinal, AverageRowLength, Cached, Partitioned,
Temporary, ReadOnly, etc.) and associated constraints.
180 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

TABLE 3.9  The DOMAINS Metacatalog Table

Table 3.9 shows the metacatalog table DOMAINS, which stores both
the RDBMS provided column data types and the users defined ones, as
well as a partial valid instance for it. Note that domain names (maximum
32 ASCII characters) are mandatory and unique (for any RDBMS). This
table has actually other columns too (e.g., Owner, UpperBound, Length,
CharacterSet, etc.) and associated constraints, as well as very many other
rows.
All DOMAINS seven rows are system ones, as BigCities does not con-
tain any user-defined domain. Note that by INT, RAT, and DATE/TIME,
respectively, it is understood corresponding OS/RDBMS representable
subsets of the integers, rationals, and calendar dates, that autonumber is
actually a synonym for INT (the only difference being that its values are
automatically generated by the system and are unique for any primary
key), and that BOOLE is the set {True, False}.
ASCII and UNICODE are subsets of the sets55 of strings built upon
these two character sets and having a maximum (RDBMS dependent)
allowed length.
This table actually has other columns too (e.g., minValue, maxValue,
etc.) and associated constraints as well.
Table 3.10 presents the metacatalog table ATTRIBUTES and its instance
for the two dbs above (System and BigCities). Note that column (attribute)
55
Mathematically, the freely generated monoids over the corresponding alphabets.
The
 Quest for Data Independence, Minimal Plausibility, and Formalization 181

TABLE 3.10  The ATTRIBUTES Metacatalog Table

ATTRIBUTES (x, AttributeName • Relation)
x Attribute- Total? Relation Domain DConstr
Name
auto- ASCII(255) BOOLE Im(RELATIONS.x) Im(DOMAINS.x) ASCII(255)
number
NOT NOT NULL NOT NOT NULL NOT NULL
NULL NULL
0 x yes 0 0
1 DBName yes 0 4 ≤ 255
2 x yes 1 0
3 Relation- yes 1 4 ≤ 255
Name
4 PrimaryKey no 1 2
5 Database yes 1 2
6 x yes 3 0
7 Attribute- yes 3 4 ≤ 255
Name
8 Total? yes 3 1
9 Relation yes 3 2
10 Domain yes 3 2
11 DConstr no 3 4 ≤ 255
12 x yes 2 0
13 Domain- yes 2 4 ≤ 32
Name
14 x yes 4 0
15 Constr- yes 4 4 ≤ 255
Name
16 Constr- yes 4 2
Relation
17 ConstrType yes 4 4 ‘F’, ‘K’, ‘T’
18 *DB no 4 2
19 x yes 5 0
20 Position yes 5 2 between
1–64
21 Key yes 5 2
22 Attribute yes 5 2
182 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

TABLE 3.10  (Continued).

x Attribute- Total? Relation Domain DConstr
Name
23 x yes 6 0
24 ForeignKey yes 6 2
25 FKAttribute yes 6 2
26 Ref- yes 6 2
Attribute
27 Position yes 6 2 between
1–64
28 x yes 7 0
29 Country yes 7 4 ≤ 128
30 Capital no 7 2
31 x yes 8 0
32 Country yes 8 2
33 City yes 8 4 ≤ 64

names (maximum 255 ASCII characters) are mandatory and unique within
any table. The foreign key Relation (referencing RELATIONS’ primary
one) is mandatory, as any column belongs to a table.
The foreign key Domain (referencing DOMAINS’ primary one) is
mandatory too, as any column stores values of a desired data type. Total?
is also mandatory, as for any column the system has to know whether or
not to accept null values for it.
Note that columns Domain, DConstr, and Total? are storing domain
(the first two) and not-null (the third) constraints. Note also that all primary
keys are taking values into the autonumber subset of the integers, while
foreign keys have integer values. The first twenty eight rows correspond
to the System metacatalog tables, while the last six to the BigCities tables.
This table actually has very many other columns too (e.g., Position,
DefaultValue, LowestValue, HighestValue, DistinctValues, CharacterSet,
AverageLength, Hidden, Computed, etc.) and associated constraints.
Table 3.11 shows the metacatalog table CONSTRAINTS, which stores
the relational constraints of the remaining three types (‘K’ for (unique)
keys, ‘F’ for foreign keys, and ‘T’ for tuple/check ones), as well as its
instance for the two dbs above (System and BigCities).
The
 Quest for Data Independence, Minimal Plausibility, and Formalization 183

TABLE 3.11  The CONSTRAINTS Metacatalog Table

CONSTRAINTS (x, ConstrName • *DB)

x ConstrName ConstrRelation Constr *DB

Type
autonumber ASCII(255) Im(RELATIONS.x) {‘F’,’K’,’T’} Im(DATABASES.x)
NOT NOT NULL NOT NULL NOT NULL
NULL
0 #DatabasesPK 0 K 0
1 DatabaseNameK 0 K 0
2 #RelationsPK 1 K 0
3 RelationNameK 1 K 0
4 PrimaryKeyK 1 K 0
5 #AttributesPK 3 K 0
6 AttributesKeyK 3 K 0
7 #DomainsPK 2 K 0
8 DomainNameK 2 K 0
9 #ConstraintsPK 4 K 0
10 ConstraintsK 4 K 0
11 #KeysColsPK 5 K 0
12 KeysCols- 5 K 0
AttrKeyK
13 KeysColsPos- 5 K 0
KeyK
14 #ForeignKeysPK 6 K 0
15 ForeignKeys- 6 K 0
AttrK
16 Foreign- 6 K 0
KeysPosK
17 Relations- 1 F 0
DatabaseFK
18 Relations- 1 F 0
PrimaryKeyFK
19 Attributes- 3 F 0
RelationFK
20 Attributes- 3 F 0
DomainFK
21 Constraints- 4 F 0
RelationFK
22 KeysColsKeyFK 5 F 0
184 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

TABLE 3.11  (Continued).

x ConstrName ConstrRelation Constr *DB
Type
23 KeysCols- 5 F 0
AttributeFK
24 FKColsForeign- 6 F 0
KeyFK
25 FKColsFKAt- 6 F 0
tributeFK
26 FKColsRef- 6 F 0
AttributeFK
27 #CountriesPK 7 K 1
28 Countries- 7 K 1
CountryK
29 Countries- 7 K 1
CapitalK
30 #CityPK 8 K 1
31 CitiesCitiesK 8 K 1
32 Countries- 7 F 1
CapitalFK
33 CitiesCountryFK 8 F 1

Note that constraint names (maximum 255 ASCII characters) are man-
datory and unique within any db. This is why the computed foreign key
*DB (referencing DATABASES’ primary one) is needed; its graph is com-
putable56.
Note that, except for the key, no constraints are needed on it. The
foreign key ConstrRelation (referencing RELATIONS’ primary one) is
mandatory, as any constraint belongs to a table scheme. ConstrType is
mandatory too, for partitioning the table’s instance according to the three
constraint types stored in it. The first twenty seven rows correspond to the
System metacatalog tables, while the last seven to the BigCities tables.
This table actually has many other columns too (e.g., Generated, Status,
Deferrable, Deferred, Validated, Invalid, etc.) and associated constraints.

56
Mathematically, *DB = Database ° ConstrRelation. Relationally:
SELECT CONSTRAINTS.x, Database
FROM CONSTRAINTS INNER JOIN RELATIONS
  ON CONSTRAINTS.ConstrRelation = RELATIONS.x
The
 Quest for Data Independence, Minimal Plausibility, and Formalization 185

TABLE 3.12  The KEYS_COLUMNS Metacatalog Table

KEYS_COLUMNS (x, Attribute • Key, Key • Position)

x Key Attribute Position

autonumber Im(CONSTRAINTS.x) Im(ATTRIBUTES.x) [1, 64]
NOT NULL NOT NULL NOT NULL NOT NULL
0 0 0 1
1 1 1 1
2 2 2 1
3 3 3 1
4 3 5 2
5 4 4 1
6 5 6 1
7 6 7 1
8 6 9 2
9 7 12 1
10 8 13 1
11 9 14 1
12 10 15 1
13 10 18 2
14 11 19 1
15 12 22 1
16 12 21 2
17 13 21 1
18 13 20 2
19 14 23 1
20 15 25 1
21 15 24 2
22 16 24 1
23 16 27 2
24 27 28 1
25 28 29 1
26 29 30 1
27 30 31 1
28 31 33 1
29 31 32 2
186 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

Table 3.12 presents the metacatalog table KEYS_COLUMNS and its

instance for the two dbs above (System and BigCities). Note that column
(attribute) positions (maximum 64 for IBM DB2, 32 for Oracle, 10 for MS
Access, etc.) are mandatory within any key; moreover, in any position of
any key, there may be only one column: hence, Key • Position is a seman-
tic key of this table.
The foreign key ‘Key’ (referencing CONSTRAINTS’ primary one) is
mandatory, as any key column belongs to a key constraint. The foreign key
‘Attribute’ (referencing ATTRIBUTES’ primary one) is mandatory too, as
all columns of any key have to be known.
Note that no key should contain a same column more than once: hence,
Attribute • Key is also a semantic key of this table. For example, its fourth
and fifth rows (x = 3, 4) store the key RelationName • Database (con-
straint 3, named RelationNameK) of system table 1 (RELATIONS), which
is made out of attributes 3 (RelationName) in position 1 and 5 (Database)
in position 2. The first twenty four rows correspond to the System meta-
catalog keys, while the last six to the BigCities ones.
This table too may actually have other columns (e.g., SortOrder) and
associated constraints.
Table 3.13 shows the metacatalog table FOREIGN_KEYS_COLS and
its instance for the two dbs above (System and BigCities).
Note that all of its columns (attributes) are mandatory: the primary
key x, by primary keys definition, and the foreign keys ForeignKey (ref-
erencing CONSTRAINTS’ primary one), FKAttribute, and RefAttribute
(both referencing ATTRIBUTES’ primary one), as the foreign keys, their
columns, and the corresponding referenced columns, respectively, have to
be all known.
Also note that column (attribute) positions (maximum 64 for IBM
DB2, 32 for Oracle, 10 for MS Access, etc.) are mandatory within any for-
eign key; moreover, in any position of any foreign key, there may be only
one column: hence, ForeignKey • Position is a semantic key of this table.
Also note that, in fact, the RDM might not allow for concatenated
foreign keys, as single ones are enough (and we will see in Subsection
3.11.1.3, best practice rule R-FK-1, that it is always better to use single
numeric ones, always referencing single surrogate primary keys, in which
case Position would not be needed anymore). This is why here, where all
foreign keys are single ones, Position is always 1.
The
 Quest for Data Independence, Minimal Plausibility, and Formalization 187

TABLE 3.13  The FOREIGN_KEYS_COLUMNS Metacatalog Table

FOREIGN_KEYS_COLS (x, FKAttribute • ForeignKey, ForeignKey • Position)

x ForeignKey FKAttribute Position RefAttribute

auto- Im(CONSTRAINTS.x) Im(ATTRIBUTES.x) [1, 64] Im(ATTRIBUTES.x)
number
NOT NOT NULL NOT NULL NOT NOT NULL
NULL NULL
0 17 5 1 0
1 18 4 1 14
2 19 9 1 2
3 20 10 1 12
4 21 16 1 2
5 22 21 1 14
6 23 22 1 6
7 24 24 1 14
8 25 25 1 6
9 26 26 1 6
10 32 30 1 31
11 33 32 1 28

No foreign key should contain a same column more than once: hence,
FKAttribute • ForeignKey is also a semantic key of this table. For example, its
first row (x=0) store the foreign key Database (RelationsDatabaseFK, con-
straint 17, attribute 5, position 1) of system table 1 (RELATIONS), which is
referencing attribute 0 (x), the autonumber (domain 0) primary key (#Data-
basesPK, constraint 0) of table 0 (DATABASES). The first ten rows correspond
to the System metacatalog keys, while the last two to the BigCities ones.
This table too may actually have other columns and associated con-
straints as well.
Note that some actual RDBMS metacatalog tables are not even in
BCNF (see Subsection 3.10.5). Moreover, surprisingly, some of them do
not even correctly implement all of the RDM concepts.57

57
For example, Oracle, MS SQL Server, and Access do not enforce either C3 or C11, that
is minimal uniqueness (as they allow defining both keys and superkeys) or foreign keys
acyclicity (as they allow attributes to reference themselves, both directly, except for MS
Access, and indirectly).
188 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

Also note that RDBMS engines have to enforce, besides the relational
constraints stored in CONSTRAINTS (see Table 3.11), many other non-
relational constraints, including the ones shown in Fig. 3.21.
Note that for also storing tuple constraints you should either add at
least a column to table ATTRIBUTES above or another kernel table (see
Exercise 3.75).

3.7.3  AN E-R DATA MODEL OF THE RDM

There are truths so less striking that their discovery is almost creation.
—Lucian Blaga

By applying REA1–2 to the above RDBMS kernel metacatalog, the fol-

lowing E-R data model of the RDM is obtained, in which descriptions of
the corresponding object sets and constraints, as well as table cardinality
examples were added too for didactical reasons:

3.7.3.1  E-RD of the RDM

Friends share all things.

—Pythagoras

Figure 3.22 shows the E-RD of this fragment of the RDM.

3.7.3.2  Associated Restriction Set for the RDM

In fact what I would like to see is thousands of computer scientists let

loose to do whatever they want. That’s what really advances the field.
—Donald Knuth

1. DATABASES (The set of rdbs hosted by a RDBMS)

a. Cardinality: max(card(DATABASES)) = 104 (RDB0)
b. Data ranges: DBName: ASCII(255) (RDB1)
c. Compulsory data: x, DBName (RDB2)
The
 Quest for Data Independence, Minimal Plausibility, and Formalization 189

ü C0: Relation surj (any relation has at least one attribute)

ü C1: (;x[RELATIONS)(PrimaryKey(x) Ó NULLS 
ConstrType(PrimaryKey(x)) = ‘K’) (primary keys are constraints
of type key)
ü C2: (;x[RELATIONS)(;y[KEYS_COLUMNS)
(PrimaryKey(x) = Key(y)  Relation ° Attribute(y) = x) (any
attribute of the primary key of any relation has to be an attri-
bute of that relation)
ü C3: (;x[CONSTRAINTS)(;y1, …, yn [ KEYS_COLUMNS,
n > 0, natural)(ConstrType(x) = ‘K’ ` Key(y1) = … = Key(yn) =
x  Attribute(y1) • … • Attribute(yn) is one-to-one) (keys are
one-to-one)
ü C4: ConstrRelation ° PrimaryKey = 1RELATIONS (the primary key
of any relation is a constraint of that relation)
ü C5: ConstrRelation ° Key = Relation ° Attribute (any attribute
of a key should belong to the relation to which the correspond-
ing key constraint is associated)
ü C6: (;x[KEYS_COLUMNS)(PrimaryKey(Relation(Attribut
e(x))) = Key(x)  Attribute(x) is part of the primary key of the
relation Relation(Attribute(x))) (primary key attributes are key
attributes)
ü C7: (;x[FOREIGN_KEYS_COLS)
(ConstrType(ForeignKey(x)) =‘F’) (referential integrities are
constraints of type foreign key)
ü C8: ConstrRelation ° ForeignKey = Relation ° FKAttribute (any
attribute of a foreign key should belong to the relation to which
the corresponding referential integrity constraint is associated)
ü C9: (;x, y[FOREIGN_KEYS_COLS, x ≠ y) (ForeignKey(x)
= ForeignKey(y)  Relation ° RefAttribute(x) = Relation °
RefAttribute(y)) (all attributes referenced by a foreign key should
belong to a same relation)
ü C10: (;x[FOREIGN_KEYS_COLS)(Im(FKAttribute(x)) #
Im(RefAttribute(x))) (referential integrity enforcement: values

FIGURE 3.21  Non-relational constraints associated to the above metacatalog.

190 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

of foreign keys should always be among those of the corre-

sponding referenced columns)
ü C11: (;x[FOREIGN_KEYS_COLS) ((FKAttribute • RefAttri-
bute)(x) acyclic) (no attribute should reference itself, neither
directly, nor indirectly)
ü C12: (;x[CONSTRAINTS)('y[FOREIGN_KEYS_COLS)
(ConstrType(x) =‘F’  ForeignKey(y)=x) (any foreign key has
at least one attribute)
ü C13: (;x[CONSTRAINTS)('y[KEYS_COLUMNS)(ConstrType(x)
=‘K’  Key(y)=x) (any key has at least one attribute)

FIGURE 3.21  (Continued).

FIGURE 3.22  The E-RD of the RDM.

d. Uniqueness: x, DBName (there may not be two rdbs

hosted on a same RDBMS and having same names) (RDB3)

2. DOMAINS (The set of needed column data types)

a. Cardinality: max(card(DOMAINS)) = 102 (RD0)
b. Data ranges: DomainName: ASCII(32) (RD1)
c. Compulsory data: x, DomainName (RD2)
d. Uniqueness: x, DomainName (there may not be two
domains managed by a same RDBMS and having
same names) (RD3)
The
 Quest for Data Independence, Minimal Plausibility, and Formalization 191

3. RELATIONS (The set of tables of all the rdbs)

a. Cardinality: max(card(RELATIONS)) = 108 (RR0)
b. Data ranges:RelationName: ASCII(255) (RR1)
c. Compulsory data: x, RelationName (RR2)
d. Uniqueness: x, RelationName • Database (there may
not be two tables of a same rdb having same names) (RR3)

4. ATTRIBUTES (The set of columns of all tables)

a. Cardinality: max(card(ATTRIBUTES)) = 1010 (RA0)
b. Data ranges:
  AttributeName, DConstr: ASCII(255) (RA1)
  Total?: Boolean (RA2)
c. Compulsory data: x, AttributeName, Relation,
Domain, Total? (RA3)
d. Uniqueness: x, AttributeName • Relation (there may
not be two columns of a same table having
same names) (RA4)

5. CONSTRAINTS (The set of constraints of all tables)

a. Cardinality: max(card(CONSTRAINTS)) = 1010 (RC0)
b. Data ranges:
  ConstrName: ASCII(255) (RC1)
  ConstrType: {‘F’, ‘K’, ‘T’} (RC2)
c. Compulsory data: x, ConstrName, ConstrType,
ConstrRelation (RC3)
d. Uniqueness: x, ConstrName • *DB (there may not be
two constraints in a same rdb having same names) (RC4)

6. FOREIGN_KEYS_COLS (The set of pairs of type <fk, c>

storing the fact that column c is a member of the foreign key
constraint fk)
a. Cardinality: max(card(FOREIGN_KEYS_COLS)) = 109 (RFK0)
b. Data ranges: Position: [1, 64] (RFK1)
c. Compulsory data: x, ForeignKey, FKAttribute,
RefAttribute, Position (RFK2)
192 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

d. Uniqueness: x,
FKAttribute • ForeignKey (no attribute should take
part more than once in a foreign key), (RFK3)
ForeignKey • Position (no position of a foreign key
may be occupied by more than one attribute). (RFK4)

7. KEYS_COLUMNS (The set of pairs of type <k, c> storing the fact
that column c is a member of the (unique) key constraint k)
a. Cardinality: max(card(KEYS_COLUMNS)) = 109 (RK0)
b. Data ranges: Position: [1, 64] (RK1)
c. Compulsory data: x, Key, Attribute, Position (RK2)
d. Uniqueness: x,
 Attribute • Key (no attribute should take part more than
once in a key), (RK3)
 Key • Position (no position of a key may be occupied
by more than one attribute). (RK4)

All nonrelational constraints from Fig. 3.21 must be added here too.

3.7.3.3  Description of the RDM

There’s ways to amuse yourself while doing things

and that’s how I look at efficiency.
—Donald Knuth

For example (for all the following maximum cardinalities), any RDBMS
manages at most 104 rdbs, which are characterized by their compulsory
unique name (at most 255 ASCII characters), and provides at most 102
data types, which are characterized by their compulsory unique name (at
most 32 ASCII characters).
Any rdbs may include several relations, at most 108 in total, for all
rdbs, which are characterized by their compulsory name (at most 255
ASCII characters), unique within their rdbs.
Any relation should include at least one attribute; attributes are charac-
terized by their compulsory name (at most 255 ASCII characters), unique
within their relation, NOT NULL constraint (does the corresponding column
also accept null values or not?) and data type; optionally, a domain restriction
The
 Quest for Data Independence, Minimal Plausibility, and Formalization 193

(a propositional formula of at most 255 characters) may also be specified for

attributes that should not accept all values of their domains, but only a subset
of them. In total, for all rdbs, there may be at most 101° attributes.
Any relation may have attached constraints, which are characterized by
their compulsory name (at most 255 ASCII characters), unique within their
rdbs, and type (one character, either ‘F’—for foreign keys, ‘K’—for (unique)
keys, or ‘T’—for tuple constraints). In total, for all rdbs, there may be at most
101° constraints, out of which at most 109 foreign and unique keys.
Any foreign or unique key should include at least one and at most 64
attributes; all attributes of a key, be it foreign or unique, should belong to
the relation to which the corresponding constraint belongs; no attribute
may appear more than once in a key (be it foreign or unique).
For any key, the RDBMS should enforce its one-to-oneness —that is
its values should uniquely identify the corresponding relation tuples, in
any moment of time (or, dually, no duplicates should ever be allowed for
the values of a key).
For any foreign key, the RDBMS should enforce its referential integ-
rity: its values should always be among those of the corresponding refer-
enced attributes.
Any attribute of a foreign key references another attribute of a same
(or compatible) domain, acyclically (i.e., no attribute of a foreign key may
recursively reference itself, neither directly, nor indirectly); all attributes
of a foreign key should reference attributes belonging to a same relation.
Any relation may have at most one primary key, which must be a key
composed of some of its attributes.

3.8  RELATIONAL SCHEMAS DEFINITION. SQL DDL

Morphological information has provided the greatest single source of

data in the formulation and development of the theory of evolution
and that even now, when the preponderance of work is experimental,
the basis for interpretation in many areas of study remains the form
and relationships of structures.
—Everett C. Olson
Let us consider in what follows the DDL of ANSI-92-type SQL. For
creating a db named “Classical_Music” with the three tables from Fig.
194 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

FIGURE 3.23  ANSI-92 SQL statements for creating the db scheme of Fig. 3.2.
The
 Quest for Data Independence, Minimal Plausibility, and Formalization 195

3.2, the statements shown in Fig. 3.23 are needed (see algorithm A8 from
Fig. 4.1).
Please note the following:
0. Actual RDBMS implementations syntax varies both as compared to
the ANSI SQL standards and even between their versions. However, for
most of them, “;” is ending statements.
1. Auto-generated numbers (autonumbers)58 were introduced in the
ANSI-2003 SQL standard.
2. “– –” are introducing (i.e., are followed by) comments.
3. As table names should be unique only within db schemas, full refer-
ence to any table should be of the type db_name.table_name (e.g., Clas-
sical_Music.COMPOSERS). When the context is unambiguous, the
db names may be skipped.
4. Except for some aggregation ones (e.g., COUNT, SUM, MAX,
MIN, AVG, obviously computing cardinals, total amounts, maximum,
minimum, and arithmetic average values, respectively), there are no
ANSI-92 SQL standard functions; however, SysDate (returning current
system date) and Year (returning the year part of a calendar date value)
above, as well as many others are provided by any RDBMS (possibly
slightly renamed)59.
5. CREATE TABLE statements are creating corresponding tables and
their columns; for each column, its name is followed by its data type and,
whenever needed, its maximum allowed length (e.g., VARCHAR2(32)
means variable length strings of maximum 32 characters; NUMBER(12,0)
means integer numbers (i.e., having 0 digits after the decimal point) using
maximum 12 digits; INT(12) … IDENTITY(1,1) means an auto gen-
erated integer of maximum 12 digits, whose values start with and are
incremented always by 1 (which is the default60). PRIMARY is declar-
ing the corresponding column as being the corresponding table’s primary
key, while NOT NULL is declaring that the corresponding column is not
accepting null values.
58
As such, for example, Oracle continued not to provide them, to the despair of most of its users, up
until December 2013, when version 12c was released.
59
Note that, for example, Oracle does not allow for SysDate either in table constraints or computed
columns definitions; consequently, corresponding constraints may only be implemented through PL/
SQL (see Subsection 4.3.2).
60
Implementations are sometimes also requiring explicit start and increment values.
196 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

6. ALTER TABLE statements are declaring domain, check, unique

(key) and referential integrity (foreign key) constraints. Note that, on
one hand, these constraints might be declared in the CREATE TABLE
statements too, using the CONSTRAINT clause and that, on the other,
generally, ALTER TABLE statements may also add/modify columns,
indexes, etc.
7. When needed, dbs, tables, columns, constraints, indexes, etc. may be
removed by using the DROP statement; for example, for removing the con-
straint enforcing for column No of table MUSICAL_WORKS above values
only between 1 and 255, you should execute the statement:
ALTER TABLE Classical_Music.MUSICAL_WORKS
DROP CONSTRAINT MUSICAL_WORKS_NO_CHK;
8. In any rdb scheme, all tables and constraints should have distinct
names and within each table all columns should have distinct names.
9. SQL DDL also contains CREATE, ALTER and DROP statements for
managing users and groups of users, as well as GRANT and REVOKE state-
ments for granting to and revoking from them predefined access rights
(e.g., create, alter, drop, read-only querying, insert, update, delete, etc.) to
db objects (e.g., db and tables schemas, instances, etc.), as well as other
object types (especially indexes, used for both enforcing unique con-
straints and speeding up query evaluation).

3.9  RELATIONAL INSTANCES MANIPULATION. SQL DML.

RELATIONAL CALCULI AND ALGEBRA

Not everything that can be counted counts,

and not everything that counts can be counted.
—Albert Einstein

Databases are not used only for data storage, but especially in order to
retrieve and process it (e.g., by computing cardinals, totals, averages,
minimums, maximums, etc.). This is why all computer data models
include data manipulation languages. Querying languages are tools for
formulating, syntax and semantic correctness checking, and computing
answers to all possibly valid queries over any db. Data manipulation lan-
The
 Quest for Data Independence, Minimal Plausibility, and Formalization 197

guages that include them are also providing statements for adding, modi-
fying, and deleting rows in/from tables.
Although data manipulation languages are beyond the scope of this
book, we have to review their basics, as they are needed even in db design,
implementation and, especially, optimization.

3.9.1  INSERTING, UPDATING, AND DELETING DATA

Get the facts first and then you can distort them as much as you please.
—Mark Twain

For example, in order to populate the rdb scheme created by statements in

Fig. 3.23 above with data presented in Fig. 3.2, you need the statements
shown in Fig. 3.24.
Please note the following:
0. Above form of the INSERT statement only inserts one row at a
time, provided that valid data is specified for each column; for example, if
in the last but one statement above the VALUES clause were (8,8, ’83’,
null,1, ’The 2nd piano concerto’, null), its execution would
fail as there is one more value than the number of fields in the MUSI-
CAL_WORKS table; if it were (7,8, ’83’, null,1, ’The 2nd piano
concerto’), its execution would fail as x is a key, so no duplicates on
this column are allowed; if it were (8, null, ’83’, null,1, ’The 2nd
piano concerto’), its execution would fail as column Composer
does not accept nulls; if it were (8,8, ’83’, null,0, ’The 2nd piano
concerto’), its execution would fail as column No only accepts naturals
between 1 and 255.
1. Several lines may be inserted at once in a table by using the alternate
INSERT INTO … SELECT … statement.
2. For modifying existing data, SQL offers the UPDATE statement; for
example, if we would like to add the fact that Don Giovanni starts in D
minor, we could execute: UPDATE MUSICAL_WORKS SET Tonal-
ity=3 WHERE x=4; if we were trying, for example, to execute UPDATE
MUSICAL_WORKS SET Tonality=7 WHERE x=4; instead, it would
fail, as there is no tonality in table TONALITIES having x = 7.
198 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

FIGURE 3.24  ANSI-92-type SQL statements for inserting instances of Fig. 3.2 into the
rdb scheme presented in Fig. 3.23.
The
 Quest for Data Independence, Minimal Plausibility, and Formalization 199

FIGURE 3.24  (Continued).

Caution: if you omit the optional WHERE clause of UPDATE then all
corresponding rows will be updated. For example all rows of table MUSI-
CAL_WORKS would contain 3 in its column Tonality after executing the
statement: UPDATE MUSICAL_WORKS SET Tonality = 3;
3. For removing existing data, SQL offers the DELETE statement; for
example, if we would like to remove Brahms’ 2nd piano concerto, we
could execute: DELETE FROM MUSICAL_WORKS WHERE x=8;
if we were trying to delete any line in TONALITIES, we wouldn’t be
successful as all of them are referenced by at least one line in MUSI-
CAL_WORKS.
Caution: if you omit the optional WHERE clause of DELETE then all
corresponding rows will be deleted. For example all rows of table MUSI-
200 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

CAL_WORKS would be deleted, leaving the table empty, after executing

the statement: DELETE FROM MUSICAL_WORKS;

3.9.2  QUERYING DATA

That is the essence of science: ask an impertinent question,

and you are on the way to the pertinent answer.
—Jacob Bronowski

Any db scheme may be also viewed as a collection of queries, with its

instance being the set of corresponding answers. For example, the db in
Fig. 3.2 embeds the following three queries:
1. “Which are the first and last names, birth and passed away years of
the composers of interest (i.e., those on whom data is stored?)?”
2. “Which are the composer, opus, number, tonality, and title of all of
the musical works of interest?” and
3. “Which are the musical tonality names of interest?”
Especially in logical approaches of data modeling, but not only, it is
considered that each table row is the answer to a query. For example, the
first row of the TONALITIES table above is interpreted as “It is true that
there is a tonality called ‘B-flat major’ whose db x code is 1”, whereas the
last one in table COMPOSERS as “It is true that, between 1833 and 1897,
lived a composer named ‘Johannes Brahms’, whose db x code is 8”.
Based on this embedded queries and answers, explicit queries may be
built and their corresponding answers computed too. For example, consid-
ering the db in Fig. 3.2, the queries:
a. “Who are the composers having same first name?” (none in this
case);
b. “Who were the composers contemporary with J.S. Bach?” (Antonio
Vivaldi and Joseph Haydn);
c. “How many musical works were written in D-sharp minor?” (none
in this case);
d. “Who are the composers who composed at least 2 musical works?”
(Johannes Brahms in this case) etc.
The
 Quest for Data Independence, Minimal Plausibility, and Formalization 201

may be written in ANSI-92-type SQL as follows, respectively (where

“<>“ means “not equal” and “COUNT (*)” means the number of all
involved rows):

Please note the following:

0. SELECT is the only standard SQL querying statement (as there is no
need for any other one).61
1. With only one tiny exception for some RDBMSs62, its two first and
only mandatory clauses are SELECT and FROM, in this order.
2. All expressions in the SELECT clause (separated between them by
commas) will generate corresponding columns (which should have
unique names) in the result, in corresponding order; these expressions
may be constants, columns from the table instances occurring in the
FROM clause, or arbitrary algebraic expressions based on these col-
umns and constants, including some restricted forms of subqueries
(i.e., queries embedded in parenthesis, just like, for example, in c. and
d. above: see 18 to 25 below). Both column (in SELECT clauses) and
table instances (in FROM ones) may be renamed (using the optional AS
61
Some implementations include non-standard ones as well. For example, MS Access also provides
TRANSFORM for computing crosstabulations.
62
For example, in MS Access, SELECT @@IDENTITY; returns the latest auto-number value gener-
ated by the system.
202 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

operator); see, for example, C in a. above. Note that in some RDBSs

the keyword AS is optional (e.g., in a. above, instead of COMPOSERS
AS C, you could/should simply write COMPOSERS C).
3. Set quantifier predicates may be added to the SELECT clause in
order to eliminate duplicates (DISTINCT, DISTINCTROW) or not
(ALL, the default one). Note that DISTINCT removes duplicated
in the end, only before final ordering (if any), while DISTINC-
TROW removes them before evaluating the join operators (see notes
5 to 11). Other predicates may be added too (e.g., TOP/LIMIT n,
requesting only for the first n rows of the answer, n natural).
4. The FROM clause specifies a table expression: an expression whose
operands are table instances (possibly even several ones of a same
table, like, for example, in a. above –where the second instance of
COMPOSERS had to be renamed as C, in order to uniquely dis-
tinguish between them– or named subqueries) and whose operators
are the Cartesian set product (denoted by comma) and/or the join
operators.
5. The join operators are of two types: inner and outer. The inner join
is a Cartesian product followed by a filtering WHERE-type condition.
6. The first optional clause is the WHERE one, which declares a con-
dition (formalized as a Boolean expression) that all corresponding
computed rows should satisfy. For example, in b. above, due to its
WHERE clause, are selected only composers whose birth year is at
most 1750 (J.S. Bach’s passed away year) and whose passed away
year is at least 1685 (J.S. Bach’s birth year) and whose x is not 2 (i.e.,
J.S. Bach will not be selected in the result).
7. Most of the Oracle users do not use or even ignore the existence of
inner joins (as well as the fact that comma in FROM means Cartesian
product): they are explicitly using instead its definition given in 4
above. For example, they would write a. above as:

SELECT COMPOSERS.FirstName, COMPOSERS.LastName,

  C.LastName AS otherLastName
FROM COMPOSERS, COMPOSERS C
WHERE COMPOSERS.FirstName = C.FirstName AND
  COMPOSERS.LastName <> C.LastName
The
 Quest for Data Independence, Minimal Plausibility, and Formalization 203

ORDER BY COMPOSERS.FirstName, COMPOSERS.LastName,

 C.LastName;

Note, however, that this is advisable in Oracle, DB2, and latest versions
of SQL Server only because these RDBMS have internal optimizers that
convert any such Cartesian products followed by WHERE filters into corre-
sponding inner joins (for which computation best strategies are also chosen,
depending on current data instances, indexes, etc.). Generally, this is not at all
advisable for RDBMSs which do not have such optimizing capabilities (e.g.,
MS Access and MySQL), as they will first compute the Cartesian product and
then will retain from it only the subset satisfying the WHERE filter (which is
generally taking much, much more time than the corresponding inner join).
8. Outer joins are of three types: left, right, and full ones.
9. The left (outer) join unions the inner join and all other rows from
its left operand, if any (padded with nulls for corresponding right
operand’s columns). For example, the statement:

SELECT FirstName, LastName, Opus, No,

  TONALITIES.Tonality, Title
FROM (COMPOSERS INNER JOIN MUSICAL_WORKS
  ON COMPOSERS.x = MUSICAL_WORKS.Composer) INNER JOIN
  TONALITIES ON MUSICAL_WORKS.Tonality = TONALITIES.x
ORDER BY FirstName, LastName, TONALITIES.Tonality,
  Opus, No, Title;

computes the result presented in Table 3.14, which does not contain
either Bach’s Well tempered clavier or Mozart’s Don Giovanni, as
they have no tonality; if we also need these two works, then we may
use a left join with TONALITIES instead (Table 3.15 shows its result):

SELECT FirstName, LastName, Opus, No,

  TONALITIES.Tonality, Title
FROM (COMPOSERS INNER JOIN MUSICAL_WORKS
  ON COMPOSERS.x = MUSICAL_WORKS.Composer) LEFT JOIN
  TONALITIES ON MUSICAL_WORKS.Tonality = TONALITIES.x
ORDER BY FirstName, LastName, TONALITIES.Tonality,
  Opus, No, Title;
204 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

TABLE 3.14  Inner Join Result

FirstName LastName Opus No Tonality Title
Antonio Vivaldi 8 RV 269   E major La primavera
Franz Schubert D 759   B minor The “Unfinished”
symphony
Frédéric Chopin 37 2 G major Nocturne
Johannes Brahms 115   B minor The clarinet quintet
Johannes Brahms 83   B-flat The 2nd piano con-
major certo
Joseph Haydn H 1/94   G major The “Surprise” sym-
phony
Ludwig van Beethoven 125   D minor The 9th symphony

TABLE 3.15  Left Join Result

FirstName LastName Opus No Tonality Title
Antonio Vivaldi 8 RV 269   E major La primavera
Franz Schubert D 759   B minor The “Unfinished”
symphony
Frédéric Chopin 37 2 G major Nocturne
Johann Bach BWV The Well-Tempered
Sebastian 846–893 Clavier
Johannes Brahms 115   B minor The clarinet quintet
Johannes Brahms 83   B-flat major The 2nd piano concerto
Joseph Haydn H 1/94   G major The “Surprise” sym-
phony
Ludwig Beethoven 125   D minor The 9th symphony
van
Wolfgang Mozart K 527 Don Giovanni
Amadeus

10. The right (outer) join is the dual of the left one: it unions the inner
join and all other rows from its right operand, if any (padded with
nulls for corresponding left operand’s columns). For example, the
statement:
The
 Quest for Data Independence, Minimal Plausibility, and Formalization 205

SELECT FirstName, LastName, Opus, No,

  TONALITIES.Tonality, Title
FROM TONALITIES RIGHT JOIN (COMPOSERS INNER JOIN
MUSICAL_WORKS
ON COMPOSERS.x = MUSICAL_WORKS.Composer)
ON MUSICAL_WORKS.Tonality = TONALITIES.x
ORDER BY FirstName, LastName, TONALITIES.Tonality,
  Opus, No, Title;
is strictly equivalent to the one above (and it also computes the result
shown in Table 3.15).
11. The full outer join is the union of the left and right ones. For exam-
ple, the statement:
SELECT FirstName, LastName, Opus, No,
TONALITIES.Tonality, Title
FROM (COMPOSERS INNER JOIN MUSICAL_WORKS
ON COMPOSERS.x = MUSICAL_WORKS.Composer) FULL JOIN
TONALITIES ON MUSICAL_WORKS.Tonality = TONALITIES.x
ORDER BY FirstName, LastName, TONALITIES.Tonality
Opus, No, Title;
computes the result shown by Table 3.16.

12. Sometimes we need to compute aggregations (e.g., cardinals, sums,

minimums, maximums, averages, etc.) on all rows of a table or
query, like, for example, in c. above. Other times, we need to com-
pute them by groups (partitions63) of rows, like, for example, in d.
above (where we need to compute the number of musical works per
composer). Partitioning is achieved in SELECT statements by the
second optional clause named GROUP BY.

63
For example, the set of MUSICAL_WORKS rows may be partitioned by Tonality values into six
partitions: one for those that do not have tonality values (here only “The well-tempered clavier” and
“Don Giovanni”) and five for those that have, namely one for each of the five distinct tonalities (three
of them made out of only one row in this example, while the one corresponding to G major, which
comprises Haydn’s “Surprise” symphony and Chopin’s Nocturne and the one corresponding to B
minor, including Schubert’s “Unfinished” symphony and Brahms’ clarinet quintet having two rows
each). Generally, partitions are the classes of an equivalent relation. In this case, the equivalent relation
is “has same tonality as”.
206 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

TABLE 3.16  Full Join Result

FirstName LastName Opus No Tonality Title
Antonio Vivaldi 8 RV 269   E major La primavera
The “Unfinished” sym-
Franz Schubert D 759   B minor
phony
Frédéric Chopin 37 2 G major Nocturne
Johann BWV The Well-Tempered
Bach
Sebastian 846–893 Clavier
Johannes Brahms 115   B minor The clarinet quintet
B-flat
Johannes Brahms 83   The 2nd piano concerto
major
Joseph Haydn H 1/94   G major The “Surprise” symphony
Ludwig van Beethoven 125   D minor The 9th symphony
Wolfgang
Mozart K 527 G minor Don Giovanni
Amadeus

13. The GROUP BY clause may contain several columns from the table
expression of the FROM clause (separated by commas), which are
defining partition hierarchies. For example, if we were to compute
how many works were composed by composers in each tonality,
the following statement partitions each partition corresponding to
a composer into as many subpartitions as the number of tonalities
that he/she used (in this example, all partitions are subpartitioned
in only one subpartition, as there are only one work per composer,
except for the one corresponding to Brahms, which is subparti-
tioned in two, as there are two works by him, each of which was
written in a distinct tonality):

SELECT FirstName, LastName, TONALITIES.Tonality,

 Count(MUSICAL_WORKS.x) AS WorksNoByTonality
FROM (COMPOSERS INNER JOIN MUSICAL_WORKS
  ON COMPOSERS.x = MUSICAL_WORKS.Composer) LEFT JOIN
  TONALITIES ON MUSICAL_WORKS.Tonality = TONALITIES.x
GROUP BY FirstName, LastName, TONALITIES.Tonality
ORDER BY FirstName, LastName, TONALITIES.Tonality;

The result of this query is shown in Table 3.17:

The
 Quest for Data Independence, Minimal Plausibility, and Formalization 207

TABLE 3.17  Multiple Grouping Result

FirstName LastName Tonality WorksNoByTonality
Antonio Vivaldi E major 1
Franz Schubert B minor 1
Frédéric Chopin G major 1
Johann Sebastian Bach   1
Johannes Brahms B minor 1
Johannes Brahms B-flat major 1
Joseph Haydn G major 1
Ludwig van Beethoven D minor 1
Wolfgang Amadeus Mozart   1

14. The GROUP BY clause has an associated “golden rule”: whenever

a SELECT statement contains such a clause, its SELECT clause
may contain columns as such only from the GROUP BY clause;
any other column that might appear in its SELECT clause should
be an argument of an aggregation function. For example, in the
statement:
SELECT Tonality, Count(x) AS WorksNoByTonality
FROM MUSICAL_WORKS
GROUP BY Tonality
ORDER BY Tonality;
x (or any other column from MUSICAL_WORKS except for
Tonality) may not appear as such in its SELECT clause. Indeed,
suppose that we would forget to apply Count to x (or try to con-
vince ourselves that it is not possible not to apply to x an aggregate
function); any RDBMS would reject it syntactically, as senseless:
for example, from the groups containing two rows, which x value
should be selected and why?
15. The third optional clause, HAVING, is similar to the WHERE one,
except for the moment when it is applied: WHERE is applied before,
while HAVING is applied after grouping. Consequently, HAVING
could do the job of WHERE too, but WHERE cannot do HAVING’s
job: in d. above, for example, if we were trying to replace the sub-
query with SELECT Composer FROM MUSICAL_WORKS
208 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

WHERE COUNT(x) > 1 GROUP BY Composer, the query

would be rejected.
16. Although WHERE might generally not be used at all, its usage saves
enormous amount of time (and temporary disk and memory space
as well); for example, let us consider a slight variation of d. above,
when only composers that wrote (no matter how many) musical
works in D minor are of interest; the following statement would
compute this result correctly:
SELECT FirstName, LastName FROM COMPOSERS WHERE x IN
(SELECT Composer FROM MUSICAL_WORKS
GROUP BY Composer, Tonality
HAVING Tonality = 3)
  ORDER BY FirstName, LastName;
However, this is not an optimal solution: when evaluating the sub-
query, the RDBMS will partition the whole MUSICAL_WORKS instance
by Composer and then by Tonality –obtaining 9 partitions–, and will elim-
inate all of them, but one—as only Beethoven’s 9th symphony is in D
minor.
The equivalent subquery:
SELECT DISTINCT Composer FROM MUSICAL_WORKS
WHERE Tonality = 3
runs much faster (and with much fewer memory needed) to produce
same results: from the MUSICAL_WORKS instance, the RDBMS will
select only the composers of those written in D minor (in this case, only
one Beethoven’s occurrence) and will eliminate duplicates, if any, without
even needing grouping!
Consequently, you should always use WHERE for all conditions that
may be evaluated before grouping (i.e., all that do not contain aggregate
functions) and HAVING only for those that may not be evaluated before
grouping, but only after it (i.e., only those that do contain aggregate func-
tions on groups).
For example, in order to compute the set of composers that lived in
the nineteenth century and who composed at least two musical works
(together with the total number of their musical works), the following
statement is one of the best SQL solutions possible (its result is, obviously,
<“Johannes”, “Brahms”, 2>):
The
 Quest for Data Independence, Minimal Plausibility, and Formalization 209

SELECT FirstName, LastName, Count(MUSICAL_WORKS.x) AS

TotalWorksNo
FROM COMPOSERS INNER JOIN MUSICAL_WORKS
ON COMPOSERS.x = MUSICAL_WORKS.Composer
WHERE BirthAwayYear Between 1800 AND 1899 OR
PassedAwayYear Between 1800 AND 1899
GROUP BY FirstName, LastName
HAVING Count(MUSICAL_WORKS.x) > 1
ORDER BY Count(MUSICAL_WORKS.x) DESC, FirstName,
  LastName;
17. The fourth optional clause, ORDER BY, accepts a comma sepa-
rated set of pairs <e, so>, where e are algebraic expressions based
on the columns of the FROM table expression and so are the sort
order predicates ASC(ENDING), the default one (which may con-
sequently be omitted) and DESC(ENDING). For example, if we
would like to get the results of b. above first in descending order of
their passed away year and only then in the ascending one of their
first and then last names, the following statement should be run
instead:
SELECT FirstName, LastName FROM COMPOSERS
WHERE BirthYear <= 1750 AND PAssedAwayYear >= 1685
AND x <> 2
ORDER BY PAssedAwayYear DESC, FirstName,
LastName;
18. Subqueries are SELECT statements embedded (syntactically,
between parenthesis) into other SELECT, INSERT, UPDATE, and
DELETE statements (see, for example, c. and d. above); they may
not contain the ORDER BY clause. There are four types of sub-
query usages:
• (SELECTstatement)
• comparison [ANY | ALL | SOME] (SELECTstatement)
• expression [NOT] IN (SELECTstatement)
• [NOT] EXISTS (SELECTstatement)
where SELECTstatement is a subquery, comparison is an expres-
sion followed by a comparison operator (comparing the expres-
sion result to the subquery one) and expression is an expression
210 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

whose values are to be searched among the ones of the corre-

sponding subquery evaluation result.
19. First above usage type is allowed only in SELECT (provided their
result is only one value), FROM and GROUP BY clauses; the other
three ones may be used only in WHERE and HAVING clauses. Col-
umns defined by subqueries in the SELECT and FROM clauses may
not be used in the host SELECT statement GROUP BY clause.
Subqueries used in FROM clauses must be named by using the AS
operator (in order to use them in ON subclauses, as well as to unam-
biguously refer to their computed columns).
20. When subqueries of the second and third type above are return-
ing multicolumns row sets, the corresponding expressions should
contain same arity and data type rows (syntactically enclosed in
parenthesis and comma separated).
21. If the second type above is used without predicates (ANY, ALL or
SOME) –that is only with =, <>, <, <=, >, or >= –, then the subquery
should return only one value; if a predicate is used too, then the
subquery may return any number of columns and rows.
22. ANY and SOME are both equivalent to the existential quantifier,
whereas ALL is equivalent to the universal quantifier. Note that <
ANY means “less than maximum”; > ANY, “more than minimum”;
< ALL, “less than minimum”; and > ALL, “more than maximum”.
23. IN stands for “included in” (also expressible as = ANY), whereas
NOT IN for “not included in” (equivalent to <> ANY).
24.[NOT] EXISTS are comparisons of the subquery evaluation result
to the empty set.
25. Subqueries using synonyms of table instances from the host query
are dynamic and need reevaluation for each row in the host query
instance. The rest are static and need evaluation only once, before
evaluating the host query.
26. Unions may be computed by using the UNION operator that can
merge both table instances and/or query results. Note that for
unions, on one hand, result column names are given by their first
operand, while ordering can only be done at the last one. Also note
The
 Quest for Data Independence, Minimal Plausibility, and Formalization 211

that, by default, UNION eliminates duplicates; you can ask for pre-
serving them by using UNION ALL instead.
27. Not all query results are updatable: for example, those obtained
through DISTINCT, GROUP BY, and UNION are not.
28. One might also ask queries whose answer cannot be computed only
based on a db because either there is no such stored data or queries
are incorrectly or ambiguously formulated.
For example, considering Fig. 3.2, the answer to the query “Which is
the tonality in which Mozart wrote his Requiem?” (SELECT Tonality
FROM MUSICAL_WORKS WHERE Composer = 4 and Title =
‘Requiem’;) is the empty set (to be interpreted as “There is no data on
a work by Mozart called Requiem”); the answer to the query “In which
year was first publicly performed Brahms’ 2nd piano concerto?” (SELECT
FirstPublicPerformanceYear FROM MUSICAL_WORKS WHERE
Composer = 8 and Title = ‘The 2nd piano concerto’;)
is a syntactic error (to be interpreted as “this db does not store musical
works first public performance year”); the answer to the query “Who are
the composers whose birth year is greater than their last name?” (SELECT
FirstName, LastName FROM COMPOSERS WHERE BirthY-
ear > LastName;) is a semantic error (to be interpreted as “there is no
meaning in comparing birth years, which are naturals, with last names,
which are text strings”).
Note that the following two subsections closing 3.9.2, devoted to rela-
tional calculus and algebra, may be skipped by all those not interested in
the theoretical foundations of querying. However, especially the second
one (on the relational algebra) could be interesting in order to understand
how the SELECT SQL statement is internally evaluated by RDBMSes.
Moreover, this second subsection is also providing intuition on the math
formalization (presented in section 3.12) of the GROUP BY clause.

3.9.2.1  Relational Calculus

Programming is one of the most difficult branches of applied mathematics;

the poorer mathematicians had better remain pure mathematicians.
—Edsger W. Dijkstra
212 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

SQL is based on relational calculus (RC): a family of languages based on

first-order predicate calculus. As their operands are tables, which are bi-
dimensional objects, there are two such subfamilies: tuple RC (TRC), from
which SQL stems64, whose variables are rows (tuples), and domain RC
(DRC), from which Query-by-Example (QBE)65 stems, whose variables
are columns (domains).
The first RC major simplification as compared to first order logic is
lack of mapping symbols, except for the standard Boolean ones and a
handful of so-called aggregate ones (whose set depend on the DBMSs
implementations, but generally include COUNT, for cardinal, SUM, MAX,
MIN, and AVG, for arithmetic average). Moreover, except for standard
math operators (as, for example, the comparison ones: =, ≠, <, ≤, >, ≥), all
other predicate symbols (those that in 1OL correspond to math relations)
are interpreted as db table instances.
Second major simplification is that there are no propositions66.
Third difference, a minor syntactic one, lies in notations (which does
not affect at all semantics) that are slightly modified because RDM, as we
have already seen, replaces for relations’ underlying sets the math posi-
tional (not necessarily distinct) name lists with unordered collections of
unique names ones.
Being subsets of 1OL, RC languages only contain declarative (non
imperative), nonprocedural expressions, stating results’ properties (with-
out any indication whatsoever on how these results are to be computed).
This is their most important source of power, compactness, and elegance:
for example, a SQL SELECT statement of a dozen of lines might be
equivalent to a Java method of some 120 statements or, sometimes, even
much more.

64
SQL is the most frequently used, but there are other equivalent languages too as, for example, Quel,
ISBL, UnQL, YQL, as well as subsets and/or extensions of them designed and used for particular do-
mains, as DMX (for data mining), MQL and SMARTS (for chemistry), MDX (for OLAP dbs), TMQL
(for topic maps), XQuery (for XML files), etc.
65
QBE is the only successful DRC language, which was imitated by nearly all RDBMSs (e.g., MS
calls it Query Design mode in Access and Query Designer in SQL Server, pretending that they are
Visual QBEs). Its success is due to its graphical nature, simplicity, and intuitiveness.
66
In dbs, constraints are propositions, while queries are open formulas. The db instance subsets for
which queries are true are considered to be their interpretation (meaning).
The
 Quest for Data Independence, Minimal Plausibility, and Formalization 213

CRT and CRD are equivalent as expressive powers67, with CRT for-
mulas being slightly simpler. This allows, for example, translating QBE in
SQL (and vice-versa too).

3.9.2.2  Relational Algebra

Arithmetic! Algebra! Geometry! Grandiose trinity! Luminous triangle!

Whoever has not know you is senseless.
—Comte de Lautréamont

The relational algebra (RA) is an extension of the algebra of sets (SA) with
five fundamental and several derived operators; RA operands are table
instances.
Recall that fundamental algebra of sets operators are the union, inter-
section, complementation, and Cartesian product (denoted ø, ∩, C, ×,
respectively). Derived ones include, for example, difference (or relative
complement: \, —) and direct sum (!). Moreover, besides equality, inclu-
sions (#, ,, ., $, ÷) play a significant role.
Fundamental RA proper initial (i.e., added from the first RDM specifi-
cation) operators are selection, projection and renaming.
Selection is a “horizontal splitting” operator: it is selecting from a RA
expression E instance only those rows that satisfy a logic formula F (nota-
tion: σF(E)). SQL implements it in its clauses WHERE and HAVING. Trivi-
ally, if all E’s rows satisfy F, then no selection is actually performed, the
result being equal to E’s instance; dually, if no E’s row is satisfying F, then
the result of the selection is the empty instance (of the E’s scheme).
Note that selection compositions may be consolidated into only one
selection, by using the logical and operator: σF(σG(E)) = σF ` G(E).
Projection is a dual “vertical splitting” operator: it is retaining from
a RA expression E corresponding instance only desired explicitly stated
columns included in its not empty selection set (also called target list)
S (notation: πS(E)). SQL implements it in its clause SELECT. Trivially, if
S = E, no projection is actually performed: πE(E) = E; in order to simplify
writing in this case (i.e., not being obliged to mention all E’s columns
67
In the rdbs querying context, a language expressive power means the class of queries that the lan-
guage can express.
214 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

C1, …, Cn if we want them in this db order), SQL offers a corresponding

abbreviation meta-character, generally ‘*’: instead of SELECT C1, …,
Cn FROM E, we can simply write SELECT * FROM E.68 Dually, if S =
Ø (which is only theoretically possible: RDBMSs only accept target lists
with at least one column), then the result is the empty set (both as scheme
and instance).
Note that, theoretically, π affects not only E’s scheme (by dropping
some of its columns), but also its instance, by dropping, after projection,
all duplicates (as instances should be sets). Obviously, if E contains a key,
then no duplicates are possible. In order for users not to lose data by lack of
knowledge or enough care, in fact, RDBMSs are not automatically remov-
ing duplicates after projection: it is to the user to explicitly ask for it by
specifying the predicates DISTINCT or DISTINCTROW in the SELECT
clause (see note 3 in Subsection 3.9.2).
Also note that projection compositions may be consolidated into only
one projection, by considering only the last target list: pS2(pS1(E)) = pS2(E).
Finally, as for some queries we might need either several instances of
a same expression, or to name calculated columns or subquery results,
or to link dynamic subqueries with their host queries, or to rename result
columns, RA also includes a renaming operator, generally denoted by ρB ←
A
(E), meaning that all occurrences of A in E are renamed as B. SQL imple-
ments renaming with the AS operator (that can be used both in SELECT,
FROM, WHERE and HAVING clauses) and refers to new names as aliases
of the old corresponding ones (which might also be implicitly generated
by the system). As renaming is associative, ρB1 ← A1(…ρBn ← An (E)…) might
be simply written as ρB1, …, Bn ← A1, …, An(E). Note that SQL programmers are
also declaring corresponding AS defined aliases simply for using shorter
table and/or column names (than the dbs ones).
Note that projection and renaming are performed by RDBMSs in vir-
tual no time, in internal memory, together with selections (that are anyhow
taking, in average, only some 5% of the total evaluation time of typical
queries, especially when indexes on the involved columns are present—
see the 2nd volume of this book).

68
Unfortunately, perhaps the most frequently used statement in MySQL (be it under PHP or whatever
other high level programming platform), as its users do not generally care of projecting desired expres-
sions on only actually needed columns.
The
 Quest for Data Independence, Minimal Plausibility, and Formalization 215

Two other proper relational fundamental operators have been added

lately (when SQL was designed): ordering and grouping. Note that as their
notation is not “standardized”, we will use in what follows the SQL cor-
responding ones.
Users are generally not happy at all with unordered results, so that an
ORDER BY operator has been added to RA as well, implemented by the
SQL ORDER BY clause (see note 17 above).
Note that ordering is generally an expensive operator (even in the
presence of appropriate indexes): consequently, SQL does not allow, for
example, ordering of subqueries or union operators results; for us, as pro-
grammers, it is also important to keep in mind to almost always order
results (i.e., to left them unordered only when ordering is not a must), but
only as much as needed (i.e., not also ordering on columns that need not
to be ordered).
Finally, as very often aggregate functions need to be applied for parti-
tions rather than on the whole instance of an RA expression, the kernel and
quotient set operators have been considered too (and implemented by the
GROUP BY SQL clause—see notes 12 to 14 above), in order to compute
desired partitions.
Note that SELECT S FROM E WHERE F GROUP BY C1, …, Cn
HAVING G ORDER BY L is first computing E’s scheme and instance,
then applies F, then partitions this subinstance into the equivalence classes
of the equivalence relation ker(C1 • … • Cn)69, applies G on this quotient
set70, and, finally, orders the result according to L, so that it is translated
into the following equivalent RA expression:
ORDER_BYL(πS(σG(σF(E)/ker(C1 • … • Cn)))).
For example, the SQL statement from 16 above:
  SELECT FirstName, LastName FROM COMPOSERS WHERE x IN
(SELECT Composer FROM MUSICAL_WORKS
GROUP BY Composer, Tonality
HAVING Tonality = 3)
  ORDER BY FirstName, LastName;

69
ker is the kernel (nucleus) operator on mappings (see Appendix), which places in a same equivalence
class all elements of the operand mapping’s domain for which the mapping computes a same value.
70
A quotient set is the set of partitions generated by an equivalence relation.
216 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

is automatically translated by RDBMSs into the following equivalent RA

expressions (the first one evaluating the subquery and the second one eval-
uating the final result):
sq: πComposer(sTonality = 3(MUSICAL_WORKS/ker(Composer • Tonality)));
ORDER_BYFirstName, LastName(πFirstName, LastName(sx [ sq(COMPOSERS))).
Generally, SELECT S FROM E WHERE F GROUP BY G HAVING H
is translated by RDBMSs into the RA expression pS(σH(σF(E)/ker(G))).
Note that the grouping operator is also an expensive one (even in
the presence of appropriate indexes): consequently, we should not ever
group on columns that are not needed and we should always filter out all
possible rows before grouping, by fully using the power of the WHERE
clauses.
As we’ve seen in notes 5 to 11 above, a family of seven join derived
RA operators are included too: the inner ones because not all RDBMSs
have optimizers able to replace composition of selections and Cartesian
products by corresponding inner joins (which are then evaluated much
faster, depending on the context, either by processing them in main mem-
ory—for small instances—or by exploiting existing indexes, or even by
building temporary ones, etc.) and the outer ones for simplifying both SQL
programming and internal evaluations.
For example, the a. SQL statement above:
SELECT COMPOSERS.FirstName, COMPOSERS.LastName,
C.LastName AS otherLastName
FROM COMPOSERS INNER JOIN COMPOSERS C
ON COMPOSERS.FirstName = C.FirstName AND
COMPOSERS.LastName <> C.LastName
ORDER BY COMPOSERS.FirstName, COMPOSERS.LastName,
C.LastName;
is translated by any RDBMS into the following equivalent RA expression
(note that, generally, inner joins are denoted in RA by the “bowtie” sign
|><|):
ORDER_BYFirstName, LastName, otherLastName(πFirstName, LastName, otherLastName(ρFirstName,
LastName, otherLastName ← COMPOSERS.FirstName, COMPOSERS.LastName, C.LastName
(COMPOSERS
|><|COMPOSERS.FirstName = C.FirstName ` COMPOSERS.LastName ≠ C.LastNameρC ←
COMPOSERS
(COMPOSERS)))).
However, the SQL statement from 7 above:
The
 Quest for Data Independence, Minimal Plausibility, and Formalization 217

SELECT COMPOSERS.FirstName, COMPOSERS.LastName,

C.LastName AS otherLastName
FROM COMPOSERS, COMPOSERS C
WHERE COMPOSERS.FirstName = C.FirstName AND
COMPOSERS.LastName <> C.LastName
ORDER BY COMPOSERS.FirstName,
COMPOSERS.LastName, C.LastName;
is translated by any RDBMS having an optimizer that translates σF(E1 ×
E2) into the equivalent E1 |><|F E2 exactly like above, while those RDBMS
not having such a facility are translated it into the following equivalent RA
expression (which is taking much more time and space to evaluate, both
of them increasing to the square of increases in the COMPOSERS instance
cardinal):
ORDER_BYFirstName, LastName, otherLastName(πFirstName, LastName, otherLastName(ρFirstName,
(σ
LastName, otherLastName ← COMPOSERS.FirstName, COMPOSERS.LastName, C.LastName COMPOSERS.

FirstName = C.FirstName ` COMPOSERS.LastName ≠ C.LastName

(COMPOSERS × ρC ←
COMPOSERS
(COMPOSERS)))).
If F = E1.A = E2.A, then the join is called natural (and F may be omit-
ted); in our opinion, you should always avoid it at least from the following
two reasons:
ü the most important is that this unfortunate practice of naming identi-
cally two join columns from two tables is firstly obscuring semantics
of such foreign keys; for example, if E1.A = CITIES.CityID and E2.A
= PEOPLE.CityID how could anyone guess what stores PEOPLE.
CityID? Are they birth places or home towns or etc.? Secondly, when
two foreign keys both referencing same table are needed in a same
table, only one join may remain natural and semantic confusion
increases; for example, if in the above example both birth places and
home towns have to be stored for people, which one of PEOPLE.
CityID and PEOPLE.CityID1 is storing what? Isn’t it much better to
ignore natural joins and call them instead PEOPLE.BirthPlace and
PEOPLE.HomeTown?
ü moreover, if two tables have more than one pair of identically named
columns, most QBE implementations will automatically generate
natural joins for all of them: if you don’t actually need all of them
and if you do not notice or forget do delete the unwanted ones, your
218 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

query will be wrong; for example, if E1 and E2 both have columns A

and B, then the automatically generated join condition is F = E1.A =
E2.A ` E1.B = E2.B.
If F = E1.A = E2.B, then the join is called equijoin; trivially, natural
joins are particular cases of equijoins (when B = A).
If F = E1.A θ E2.B, where θ [ {≠, <, ≤, >, >}, then the join is called
θ-join.
All RDBMSs are conceptually71 translating the SQL left join statement
from 9 above:
SELECT FirstName, LastName, Opus, No,
TONALITIES.Tonality, Title
FROM (COMPOSERS INNER JOIN MUSICAL_WORKS
ON COMPOSERS.x = MUSICAL_WORKS.Composer) LEFT JOIN
TONALITIES ON MUSICAL_WORKS.Tonality = TONALITIES.x
ORDER BY FirstName, LastName, TONALITIES.Tonality,
Opus, No, Title;
into the following equivalent RA expression:
ORDER_BYFirstName, LastName, TONALITIES.Tonality, Opus, No, Title(πFirstName, LastName,
TONALITIES.Tonality, Opus, No, Title
((COMPOSERS |><|COMPOSERS.x = MUSICAL_WORKS.Composer
MUSICAL_WORKS) |><|TONALITIES.x = MUSICAL_WORKS.TonalityTONALITIES) ø
πFirstName, LastName, rTonality ← NULL, Opus, No, Title (sTonality Ó Im(TONALITIES.x)(COMPOSERS
|><|COMPOSERS.x = MUSICAL_WORKS.Composer MUSICAL_WORKS)).
Note that many books and even research papers are wrongly present-
ing inner joins as being fundamental operators, whereas they are in fact
derived ones: E1 |><|F E2 = σF(E1 × E2), for any RA expressions E1, E2 and
any filter F.
A row of an inner join operand which is taking part in the result is
called joinable (with respect to that join—that is with its filter F). If all
rows of both operands are joinable, then the corresponding inner join is
called a complete one. For example, the join from the a. SQL statement
above is a complete one, whereas the first SQL statement from 9 above is
not, as both the fourth row of MUSICAL_WORKS (because Don Giovanni
does not have an associated tonality) and the sixth one from table TONAL-
ITIES (as there is no musical work written in G minor) are not joinable.

71
Actually, the best of them will optimize it (as union is the most expensive operator) by embedding
the union within the inner join.
The
 Quest for Data Independence, Minimal Plausibility, and Formalization 219

There is also a dual notion to the full join, which is also important
in higher RDM so-called normal forms: a table is said to have a loss-
less(–join) decomposition (LD) with respect to some of its projections if
its instance can be exactly reconstructed from these projections (i.e., if
pX1(r) |><| … |><| pXn(r) = r, where r is the instance of a table R and X1, …,
Xn are (concatenations of) its columns).72
Joins are dual to projections, as they are “vertical composition” opera-
tors (building up tables with columns from two tables, through the embed-
ded Cartesian product).
The nature of this duality is much more profound than graphical
appearances; it can be shown, for inner joins, that:
ü the instance of the projection of a join between n RA expressions
over the scheme of any of its operands is a subset of that operand’s
instance;
ü above inclusions are equalities when joins are full;
ü the composed operator defined above is idempotent (i.e., repeat-
edly applying same projection of a join over the scheme of one
of its operands yields same result as applying it only once) and,
dually, that:
ü any RA expression instance is a subset of the instance of the join of
any of its scheme full projections (i.e., projections whose union is
the expression’s scheme);
ü for any full projections of a RA expression the composition between
the corresponding join and projections is idempotent too.
When F is missing, joins are equal to the corresponding Cartesian
products: E1 |><|Ø E2 = E1 × E2.
Joins of same type are associative (consequently, they may also be con-
sidered as n-ary operators, n natural) and inner as well as the full outer
ones are commutative too.
However, composition of several types of joins is sometimes ambigu-
ous, so they are not allowed: for example, E1 L|><|F E2 |><|G E3 L|><|H E4,
where L|><| denotes a left join, is ambiguous as it is not associative—(E1
L
|><|F E2) |><|G (E3 L|><|H E4) ≠ E1 L|><|F (E2 |><|G (E3 L|><|H E4)).

72
Note that, in this context, “lossless” is a false-friend: lossy joins are not “loosing” stored data, but,
on the contrary, computing additional, not stored data.
220 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

Also note that joins (even if heavily optimized) are the third more
time consuming (after unions and Cartesian products) operators (taking
some 95% of the average time for typical queries not having groupings
and orderings, and, as approximate orders of magnitude, some 66% of
the ones having groupings and orderings, which are spending only some
10% for grouping, 5% for filtering before grouping, 1% after and 18% for
ordering). Outer joins (and especially the full ones) are more expensive
than inner ones.73, 74, 75
As, most frequently, RA expressions only use the select, project, and
join operators (and much rarely renaming, Cartesian product, union, etc.)
they are also called SPJ expressions.

TABLE 3.24  Equivalences between RA and SQL

RA SQL
RøS SELECT * FROM R UNION SELECT * FROM S73
R∩S SELECT * FROM R INNER JOIN S74
or:
SELECT * FROM R WHERE R.* IN (SELECT * FROM S)
R—S SELECT R.* FROM R LEFT JOIN S ON R.X = S.Y
WHERE S.Y IS NULL
or:
SELECT * FROM R WHERE R.* NOT IN (SELECT * FROM S)
R×S SELECT R.*, S.* FROM R, S
ρy ← x(R) SELECT x AS y FROM R
πx1, …, xn(R) SELECT x1, …, xn FROM R
σF(R) SELECT * FROM R WHERE F
R |><|F S SELECT R.*, S.* FROM R INNER JOIN S ON F
or:
SELECT R.*, S.* FROM R, S WHERE F75
πS(σH(σF(E)/ SELECT S FROM E WHERE F
ker(G)
)) GROUP BY G HAVING H

73
R and S must have same number of columns, pairwise compatible.
74
R and S have to be joinable on at least one column.
75
Beware that, if the corresponding RDBMS does not have a query optimizer able to detect and re-
place the Cartesian product followed by a selection with a corresponding join, then this alternative so-
lution is very inneficiently computed and may take hours or even days to complete for large instances.
The
 Quest for Data Independence, Minimal Plausibility, and Formalization 221

Other derived, less frequently used RA operators are left and right semi-
joins (joins projected on their corresponding operands: the left one is T |><
U = πT(T |><| U), whereas the right one is T ><| U = πU(T |><| U)) and left
and right antijoins (or antisemijoins: T |> U = T—T |>< U, T <| U = U—T
><| U), as well as division (T ÷ U = πC1, …, Cn(T)—πC1, …, Cn(πC1, …, Cn(T) ×
U—T), where C1, …, Cn are T’s columns which are not joined to U’s ones).
Note that RA’s expressive power is considered the minimum accept-
able standard for any relational querying language: any language that has
at least its expressive power is called complete. Also note that RA and RCs
are incomparable from this viewpoint: while RCs are, generally, much
more expressive, they cannot, however, express unions. In order for them
to have at least RA’s expressive power (and thus becoming comparable)
the union operator has to be added to RCs too. Also note that RA expres-
sions can compute all possible table instances only containing data stored
in corresponding rdb instances.
Table 3.24 shows the equivalences between RA and SQL.

3.9.3  VIEWS, STORED PROCEDURES, FUNCTIONS AND

TRIGGERS. TRANSACTIONS AND ACIDITY

Make everything as simple as possible, but not simpler.

—Albert Einstein

Most RDBMSs provide saving queries (under unique names per db) as
views, which are considered being computed tables; the rationale is that
their parsing, compiling, and optimization is done only once and, each
time that their results are needed, they may be referred to in any FROM
clause just like an ordinary table. Note that, generally, views do not accept
parameters76 and that their names should be distinct not only from any
other view, but also from any other table of the db. Often, hierarchies of
views are declared, saved, and used later too.
Similarly, any parameterized sequences of extended SQL manipula-
tion statements may be saved too (under unique names per db) as stored
76
However, for example, MS Access ones (called queries) do accept them, as any symbol which is not
a constant, a table, or column name is considered as being a parameter.
222 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

procedures, functions, and triggers: stored procedures are the equivalent of

subroutines, while triggers are event-driven methods (mainly for enforcing
non-relational constraints, but also for any other event-driven tasks, like, for
example, generating unique numbers for surrogate keys). Procedures and
functions may be consolidated into packages, the equivalent of high-level
PLs libraries.
Both DDL and DML statements are executed ACIDly, where ACID is
the acronym for Atomic, Consistent, Isolated, and Durable, that is their
execution is reliable because it is:
ü Atomic—that is indivisible, “all of nothing”: either all concerned
data table rows or none are processed;
ü Consistent—that is preserving corresponding db consistency
(dually, modifications made to the scheme or instance must not
violate any db constraint);
ü Isolated—that is independent of any other concurrent statements
which might be executed simultaneously (more precisely, serial-
ized: as if concurrent statements would have been executed seri-
ally, in one of the possible serialization orders);
ü Durable—that is their scheme or instance modifications are saved
on nonvolatile memory and remain stored there (even if the sys-
tem/db crash, power fails, etc. immediately after their execution
terminates).
Very often, a sequence of SQL statements (be it in-line or stored in a
procedure, function or trigger) should also be executed ACIDly: for exam-
ple, withdrawing money from a bank account in order to credit another
one should execute ACIDly. To achieve it, such sequences are embedded
into explicit transactions: SQL code enclosed into a START (BEGIN)
TRAN(SACTION) and a corresponding COMMIT or ROLLBACK dedicated
SQL statements.
Any transaction should end with both a COMMIT (for making all trans-
action modifications durable in case of success) and a ROLLBACK (for
discarding all transaction modifications in case of failure).
Note that transactions may be nested on any number of levels and that
DDL statements are implicit transactions (i.e., they do not need explicit
COMMIT).
The
 Quest for Data Independence, Minimal Plausibility, and Formalization 223

3.9.4  RELATIONAL LANGUAGES LIMITATIONS. EMBEDDED

AND EXTENDED SQL

Raise your quality standards as high as you can live with,

avoid wasting your time on routine problems, and always try to
work as closely as possible at the boundary of your abilities.
Do this because it is the only way of discovering how that boundary
should be moved forward.
—Edsger Dijkstra

First relational languages (RLs) limitation, a minor one, is the lack of func-
tion symbols, except for the aggregate ones. Practically, RDBMSes pro-
vide lot of library functions that empower their SQLs.
Second RLs limitation, a major one, is their incapacity to compute,
for example, transitive closures of binary homogenous relations (i.e., for
example, the set of all ancestors and/or descendants of somebody, or the
set of all files belonging to an OS logical drive, or the set of all employees
subordinated, directly or indirectly, to any given employee, etc.)77.
Consequently, two types of solutions were devised: embedded and
extended SQL.
The vast majority of high-level PLs are embedding SQL, that is include
at least one statement that accepts as parameters text strings containing
SQL statements and/or clauses or views and/or stored procedure names
and their actual parameters that are passed (generally, through middle-
wares like ADO78) to SQL engines; if they are queries, their results are then
passed back from these engines to the host high-level PL in accordingly
structured memory buffers called data (or record) sets.
The dual approach consists of extending SQL with high-level PL con-
structs (variable declarations, if, while, etc.79): IBM DB2 SQL PL, Oracle
PL/SQL, Sybase and MS T-SQL, etc. are examples of extended SQLs.
77
In fact, RLs limitations are much more severe: they cannot express any unbounded computation
(i.e., computations on unbounded data), as they are not Turing complete (or computationally univer-
sal) languages.
78
The acronym for ActiveX Data Objects, a MS COM-type middleware between PLs and OLE DB
(the acronym for Object Linking and Embedding DataBase, a MS application programming interface
(API) providing uniform manner access to data stored in a variety of formats).
79
Moreover, in order to also compute transitive closures (more generally, in fact, basic linear and even
nonlinear and mutual recursion), SQL was extended by powerful RDBMSs with a “WITH RECUR-
SIVE” statement (see first Chapter of the 2nd volume of this book).
224 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

Note that, unfortunately, there is no standard extended SQL; dually,

fortunately, middlewares like ADO are generally successful in automati-
cally performing most of the needed translations between different SQL
idioms.
Explicit embedded SQL is executed slower, just like dynamic SQL (see
next subsection); only embedded calls to views and/or stored procedures
are executed almost as fast as corresponding static SQL (see next subsec-
tion), as ADO is very fast, adding almost negligible overhead.
Extended SQLs generally do not provide the full power of high-level
PLs: there are no graphic user interfaces (GUI), networking, web, etc. capa-
bilities. Fortunately, their performances are, generally, the best possible.
Consequently, the majority of well-architectured, designed and devel-
oped db (software) applications use a mixture of both embedded and
extended SQL. A major challenge in this area is to optimally divide their
intermediate business logic (BL) levels between extended SQL and host
high-level PL code: the best approach is to maximize application speed
(which is not always easy to achieve).

3.9.5  STATIC VERSUS DYNAMIC SQL

Simplicity is a great virtue but it requires hard work to achieve it and

education to appreciate it.
And to make matters worse: complexity sells better.
—Edsger Dijkstra

Static SQL is code created before execution and which is not changing
during execution, be it inline or as views and/or parameterized stored pro-
cedures.
Dynamic SQL is SQL code created dynamically (be it from scratch or
using static SQL that is modified), during its execution, either inline or
in stored procedures, be it in host high-level PLs embedding SQL or in
extended SQLs.
Static SQL has no disadvantage and the following speed advantages:
ü Only at creation/replacement time, its code is parsed and, if ok, then
translated into internal RA expressions, which are then optimized,
The
 Quest for Data Independence, Minimal Plausibility, and Formalization 225

and for which, finally, the system builds and saves best execution
plans possible, depending on current instance, indexes and, option-
ally, on DBA directions too. Note that optimization is done not only
locally, for each statement, but globally too, for each views hierar-
chy, transaction, and stored procedure.
ü When executed (which is generally very frequent when com-
pared to creation/replacement), there is no need for any of the
above steps to be redone: the system just executes directly the
corresponding execution plan (possibly parameterized with actual
parameter values).
Dynamic SQL has no significant advantage, except for increasing
developers’ productivity, and the following speed disadvantage: for each
execution, each dynamic statement has to be parsed, translated into inter-
nal RA expressions, which are then optimized, and for which, finally, the
system builds (but does not save) and executes best execution plans pos-
sible. Optimization may be done only locally, for each statement, but not
globally too, either for inline code or for stored procedures.
Any dynamic SQL code may be equivalently rewritten statically.
It is true that, sometimes, dynamic SQL is easier to develop than its
equivalent static one; as such, it could be a temporary solution when dead-
lines are tight. However, keep in mind that, generally, there are no other
more permanent solutions than the temporary ones! Normally, once dead-
lines are met successfully, dynamic SQL should be replaced by equivalent
static one.
Unfortunately, there are always new deadlines, almost all the time
tighter, plus fatigue, which are not at all encouraging rewriting SQL code.
This leads to a PHP+MySQL-type architecture, which is extremely slow,
not using RDBMS’s powerful views and static parameterized stored pro-
cedure facilities.
Here is an example of equivalent static and dynamic SQL, for querying
some data for which end-users might ask for any combination of n condi-
tions c1, …, cn, n > 0 natural; obviously, there are 2n combinations:
CASE c1 IS NULL AND … AND cn IS NULL:
SELECT tl FROM te;
CASE c1 IS NOT NULL AND c2 IS NULL AND … AND cn IS NULL:
SELECT tl FROM te WHERE c1;
226 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

…
CASE c1 IS NOT NULL AND … AND cn IS NOT NULL:
SELECT tl FROM te WHERE c1 AND … AND cn;
In order to reduce complexity from exponential (2n) to linear (n), most
programmers, unfortunately, write the following type of dynamic embed-
ded SQL code:
sqlStr:= “SELECT tl FROM te”;
where:= false;
for i:= 1, n begin
if not IsNull(ci) then begin
if not where then begin
sqlStr:= sqlStr & “ WHERE ”;
where:= true;
end;
sqlStr:= sqlStr & “ AND “ & ci;
end;
end;
execute sqlStr;
The following static mixed extended and embedded solution is equiva-
lent and has same linear complexity:
CREATE OR REPLACE PROCEDURE p (
c1 VARCHAR DEFAULT NULL, …, cn VARCHAR DEFAULT
NULL)IS
BEGIN
 SELECT tl FROM te WHERE
  ((c1 IS NOT NULL AND c1) OR c1 IS NULL) AND … AND
  ((cn IS NOT NULL AND cn) OR cn IS NULL);
END;
This is executed only once, for creating parameterized stored proce-
dure p; the only necessary embedded SQL statement for calling it is of
the type:
execute “p(“ & c1 & ” , “ & … & ” , “ & cn & “);”;
Whenever a condition ci is null, corresponding WHERE line reduces to
“AND true”, which is neither applying ci, nor tampering with the rest of
the conditions; whenever condition ci is not null, corresponding WHERE
The
 Quest for Data Independence, Minimal Plausibility, and Formalization 227

line reduces to “AND ci”, which is applying ci, so that exactly same desired
functionality is achieved also in only n steps.

3.10  HIGHER AND THE HIGHEST RDM NORMAL FORMS

All forms of distorted thinking must be corrected.

—John Bradshaw

Readers who are not interested in why db design should not be done
in RDM could skip this subsection, except for 3.10.1, without losing
anything.
There is a hierarchy of relational normal forms (NFs), each of which is
essentially trying to eliminate particular cases of the so-called data manip-
ulation anomalies: insertion, update, and deletion ones.

3.10.1  DATA MANIPULATION ANOMALIES

Experience is the teacher of all things.

—Julius Caesar

A table presents an insert anomaly if one cannot insert in it a line, except

for adding nulls to some of its columns. For example, Table 3.18 has such
an anomaly: no customer data may be stored unless that customer places
at least one order.
A table presents an update anomaly if one should update same data on
several rows. For example, Table 3.18 has several such anomalies too: for
example, customer having id 1 has different names and its order number 1
has different dates (even for a same order, for both of them!).
A table presents a delete anomaly if, by deleting data on some facts,
data on some other ones is deleted too. For example, Table 3.18 has such
an anomaly too: by deleting its last line, the only one storing data on orders
of customer “Some Customer”, data on this customer is lost too.
228 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

TABLE 3.18  A Table Having an Insertion, a Deletion, and Several Update Anomalies
Customer- Customer- OrderNo OrderDate OrderedItems
ID Name
NAT(4) ASCII(64) NAT(4) [1/1/2010, SysDate()] Im(PRODUCTS.x)
NOT NOT NULL NOT NOT NULL NOT NULL
NULL NULL
1 Some Cola 1 3/15/2013 1
1 Someother 1 3/16/2013 2
Cola
2 Some 2 3/16/2013 1
Customer

3.10.2  FUNCTIONAL DEPENDENCIES

Nothing is a waste of time if you use the experience wisely.

—Auguste Rodin

Among other NFs, the 2nd, the 3rd, and the Boyce-Codd one, situated at
the base of the relational NFs hierarchy are based on a constraint called
functional dependency.
Let A and B be any two columns of a table T; B is said to be functionally
dependent on A (or, equivalently A is said to functionally determine B) if
and only if, by definition, to any value of A there is only one corresponding
value of B. Symbolically, any such assertion, which is called a functional
dependency (FD) or, abbreviated, dependency, is denoted by A → B (or,
sometimes, dually: B ← A). A is called the left hand and B the right hand
of A → B.
For example, in Table 3.6, x and Capital are each functionally deter-
mining any other column (including themselves): for example, both x →
Capital and Capital → x, as well as Capital → Population hold.
These notions are immediately extendable to column concatenations:
let X and Y be any such column products of a table T; Y is said to be
functional dependent on X if for any rows having same values for X, cor-
responding values for Y are the same.
For example, in Table 3.6, Country • State → Population • Area.
The
 Quest for Data Independence, Minimal Plausibility, and Formalization 229

Obviously, X → Y where Y # X is trivial, as it holds in all instances

of any table T including X and Y, from any rdb80. For example, Country •
State → Country, Country • State → State, and Country • State → Country
• State, are all trivial.
Note that, generally, if X • Y = T, then X → Y holds in T if and only
if X is a key of T (so, key dependencies are a particular case of the func-
tional ones).
Also note that the implication problem for FDs may be solved linearly.

3.10.3  THE 2ND NORMAL FORM

Experience is a good school, but the fees are high.

—Heinrich Heine

A FD X → A is said to be partial if X is part of a key and A is non-prime

(i.e., it is not part of a key). A table T is in 2nd NF (2NF) if it is in 1NF and
does not have any partial functional dependency.
For example, all above tables are in 2NF, while Table 3.18 is in 1NF,
but not in 2NF, because of the partial FD OrderNo → OrderDate.

3.10.4  THE 3RD NORMAL FORM

Experience is simply the name we give to our mistakes.

—Oscar Wilde

Z is said to be transitively dependent on X if it functionally depends on X

both directly and indirectly, that is there are at least three FDs of the type
X → Y, Y → Z, and X → Z, with Y not being unique. Moreover, X → Z is
called transitive. A table T is in 3rd NF (3NF) if it is in 2NF and does not
have any transitive functional dependency.
For example, all above tables, except for 3.18 are in 3NF, while Table
3.19 is in 2NF, but not in 3NF, because of the transitive FD Country →

80
Moreover, trivially, Ø → Ø and X → Ø are trivial FDs too, completely uninteresting consequently,
whereas FDs of type Ø → Y, which implies that Y is a constant function, should never be asserted: on
the contrary, such completely uninteresting constant Ys should be eliminated from all schemes.
230 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

TABLE 3.19  An Example of a Table in 2NF, but Not in 3NF

COUNTRIES (x, Country)
x Country Currency CurrencySymbol
autonumber(3) ASCII(255) Im(CURRENCIES.x) ASCII(4)
NOT NULL NOT NULL NOT NULL NOT NULL
1 U.S.A. 1 USD
2 U.K. 2 BP
3 Germany 3 Euro
4 France 3 Euro

CurrencySymbol (as there also hold Country → Currency and Currency

→ CurrencySymbol).
Note that, theoretically, experts and, for example, Oracle too considers
3NF as the minimum acceptable form of an rdb.

3.10.5  THE BOYCE-CODD NORMAL FORM

A pessimist sees the difficulty in every opportunity;

an optimist sees the opportunity in every difficulty.
—Winston Churchill

A table T is in Boyce-Codd NF (BCNF) (or 3.5NF) if all of its nontrivial

functional dependencies contain a key in their left hands.
For example, all above tables are in BCNF, while Table 3.20 is in 3NF,
but not in BCNF, because of the FD City → ZipCode.

TABLE 3.20  An Example of a Table in 3NF, but Not in BCNF

SMALL_CITIES_ADDRESSES (x, ZipCode)
x City Street ZipCode
autonumber(8) Im(CITIES.x) ASCII(128) ASCII(16)
NOT NULL NOT NULL NOT NULL NOT NULL
1 4 Edouard Vaillant 95870
2 4 Emile Zola 95870
3 4 Henri Barbusse 95870
The
 Quest for Data Independence, Minimal Plausibility, and Formalization 231

Any BCNF table is also in 3NF, but BCNF is not always achievable:
for example, the set of FDs {A • B → C, C → B} cannot be represented by
a BCNF scheme.

3.10.6  MULTIVALUED DEPENDENCIES AND THE 4TH

NORMAL FORM

Knowledge rests not upon truth alone, but upon error also.
—Carl Jung

Table 3.21 is in BCNF (as both of its FDs, Student • Course • Club → x
and x → Student • Course • Club, have keys as their left hand sides), but
exhibits, however, bad design issues.
No FDs are governing this table; unfortunately, two unrelated between
them enrollment types (courses and clubs) are each related, however, to
students. RDM calls this a multivalued dependency (MVD) and denotes it
by Student →→ Course • Club.
Generally, a multivalued dependency X →→ Y • Z, where X, Y, and Z
are (concatenations of) columns of a table T, is defined as a constraint (of
“tuple generating” type) asserting that whenever two rows of T have same
values a for X, for which Y • Z’s values are <b, c> and <d, e> respectively,
then there should also exist other two rows in the same T’s instance having
a values for X, but <b, e> and <d, c> values for Y • Z (i.e., all combina-
tions of Y and Z values should exist for each common X value). Note that
a MVD X →→ U of T is trivial if U # X or if T = X • U.
A table is said to be in the 4th NF (4NF) if any nontrivial MVD has a
superkey as its left-hand side.
Table 3.21 is in BCNF (as it has no FDs), but not in 4NF (as Student is
not a key); this is why it stores that much redundant data for each student
in both Course and Club.
Note that:
ü any table in 4NF is in BCNF too;
ü tables that are not in 4NF have their semantics overloaded with 2
distinct related ones;
232 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

TABLE 3.21  An Example of a Table in BCNF, but Not in 4NF

STUDENTS_COURSES_AND_CLUBS_ENROLLMENTS (x, Student • Course • Club)

x Student Course Club

autonumber(8) Im(STUDENTS.x) Im(COURSES.x) Im(CLUBS.x)
NOT NULL NOT NULL NOT NULL NOT NULL
1 1 1 1
2 1 1 2
3 1 2 1
4 1 2 2

ü FDs are particular cases of MVDs; dually, if X → Y holds, then X

→→ Y holds too (but the reverse is not true, which means that FDs
are stronger than MVDs);
ü MVDs have the complementary property: if X →→ U holds, then
X →→ T — U holds too (i.e., MVDs always exist in complemen-
tary pairs);
ü a table in 4NF cannot have more than one pair of complementary
nontrivial MVDs;
ü generally, a table scheme is in BCNF or 4NF, respectively, if any
of its FDs or MVDs d, respectively, is implied by at least one of its
keys;
ü the implication problem for FDs and MVDs is a hard one.

3.10.7  JOIN DEPENDENCIES, THE 5TH, AND THE

PROJECTION-JOIN NORMAL FORMS

The words of truth are always paradoxical.

—Lao Tzu

In fact, MVDs are only the simplest particular case (for n = 2) of a yet
weaker constraint type called join dependency (JD), denoted |><| [X1, X2,
…, Xn], where n > 1 natural, X1, X2, …, Xn , T and X1X2…Xn = T (i.e., all
The
 Quest for Data Independence, Minimal Plausibility, and Formalization 233

Xi are (concatenations of) columns of a table T and any column of T is

belonging to at least one Xi): T is said to satisfy JD |><| [X1, X2, …, Xn] if it
has a lossless decomposition over X1, X2, …, Xn. A JD is trivial if there is
an i, 1 ≤ i ≤ n, such that Xi = T. Note that X →→ Y holds in T if and only
if JD |><| [X • Y, X • (T — Y)] holds too (which proves that MVDs are just
particular cases of JDs).
A table T is said to be in the 5th (5NF) normal form if, for all of its non-
trivial JDs, each corresponding projection includes a key of the original
table. Any T in 5NF is also in 4NF.
A table T is said to be in the projection-join normal form (PJNF) if
all of its nontrivial JDs are implied by the set of its keys (i.e., for any JD
|><| [X1, X2, …, Xn], n > 1 natural, any Xi, 1 ≤ i ≤ n, is a superkey). Any T
in PJNF is also in 5NF. Dually, any 3NF table only having single keys is
also in PJNF.
A table in 4NF is also in 5NF if it cannot be losslessly decomposed
into three or more smaller relations: for example, Table 3.22 is in 4NF,
but not in 5NF, as it should satisfy JD |><| [x • SSN, Job, Skill], with
neither Job, nor Skill being keys, whereas all its four projections on x •
SSN (EMPLOYEES), x • Job (EMPLOYEES_JOBS), x • Skill (EMPLOY-
EES_SKILLS), and Job • Skill (JOBS_SKILLS) are in 5NF.
Note that, generally, all tables that are not in 5NF have their semantics
overloaded with n distinct related ones, n > 2 natural being the arity of
their longest JD.
Also note that the implication problem for JDs is a hard one too.

TABLE 3.22  An Example of a Table in 4NF, but Not in 5NF

EMPLOYEES (x, SSN)

x SSN Job Skill

autonumber(4) NAT(9) Im(JOB_TYPES.x) Im(SKILL_TYPES.x)
NOT NULL NOT NULL NOT NULL NOT NULL
1 123456789 1 2
2 987654321 2 1
234 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

3.10.8  THE DOMAIN-KEY NORMAL FORM

Experience teaches only the teachable.

—Aldous Huxley

Before being able to define the so-called “ultimate” RDM NF, the Domain-
Key one (DKNF), as it has been shown that any table in DKNF is in 5NF
too, we need to investigate interactions between domain and other type
of constraints and also formally define compatibility between rows and
instances and their associated anomalies.
Trivially, empty domains only allow for corresponding empty instances,
which trivially satisfy any FD, MVD, and JD. If the domain of a column
C of a table T only has one value, then any FD X → C holds, for any (con-
catenation of) column(s) X of T. Moreover, no JD having n components,
n > 2 natural, may hold in any table not having at least n values for each of
the involved columns. This is why, theoretically, RDM considers infinite
domains.
It can be shown that, for any table, if all of its involved columns may
take at least two values, then any of its FDs and MVDs is implied only by
FDs and MVDs (not by its domain constraints) and that if they may take
at least n values, n > 2 natural, then any of its JDs with n components is
implied only by its keys.
Domain and key constraints are called primitive constraints.
By definition, a table T having an associated constraint set CS is said
to be in DKNF if any constraint c of CS is implied by the primitive con-
straints belonging to CS+.
Row t is said to be compatible with table T if, together with T’s
instance, it satisfies T’s primitive constraints (i.e., all of its values satisfy
corresponding domain constraints and none of them is duplicating values
of T’s keys).
A table T is said to have an insertion anomaly whenever there is a
consistent instance of and a compatible row with it, but the union of that
instance and that compatible row is not a consistent instance of T; dually, T
is said to have a deletion anomaly whenever there is a consistent instance
and a row of it such that when removing that row its remaining instance
is inconsistent.
The
 Quest for Data Independence, Minimal Plausibility, and Formalization 235

DKNF is characterized as being anomaly free (i.e., T is in DKNF if and

only if it has no insertion or deletion anomalies).
For example, Table 3.23 (also see Fig. 3.4), with CS = PCs ø FDs ø
RI, where PCs = {x [ [0, 103], Country [ ASCII(128), CountryCode [
ASCII(8), Capital [ [0, 1016], Population [ [105, 2×109], x key, Coun-
try key, CountryCode key, Capital key} contains its primitive constraints,
FDs = {x → Country • CountryCode • Capital • Population, Country →
x • CountryCode • Capital • Population, CountryCode → x • Country •
Capital • Population, and Capital → x • Country • CountryCode • Popu-
lation} its functional dependencies, and RI = {Capital # CITIES.x} its
referential integrity constraint is in 5NF (as there are no MVDs or JDs and
all four FDs have keys as their left hand side, so it is in BCNF), but not in
DKNF, as its referential integrity constraint is not implied by its primitive
ones: for example, tuple <7, “Japan”, “JP”, 7, 128,000,000> is compatible
with this table, but it cannot be added to it, because it would violate this
referential integrity constraint, as 7 does not belong to Im(CITIES.x) (see
Fig. 3.4).
Note that most of the published examples of 5NF tables not in DKNF
are in 5NF, in fact, only because some FD (or MVD, very rarely JD) is
purposely “hidden”81—that is, although it is obvious and, consequently,
should have been asserted in the first place, it is initially omitted; when
added, generally, the corresponding scheme is not even in 2NF. Such
“examples” are not only unconvincing at all, but, in fact, counterexamples
of how not to give wrong examples: not initially including a constraint in
a table’s associated constraint set is not contradicting DKNF definition!
Consequently, such tables are, in fact, according to their initial constraint
set in DKNF too: only after adding the “hidden” constraint they are not
in DKNF anymore, but not because they violate its definition, but simply
because they are not even in 2/3/4NF.
Unfortunately, counterexamples like the above ones (i.e., the ones
obtained by also considering inclusion dependencies associated to for-
eign keys) are dismissed by some, as they wrongly consider referential
integrity as being domain constraints instead: for them, RI is always
empty and, for example, Capital # CITIES.x above is replaced by the

81
Unfortunately, published true counter-examples of tables in 5NF but not in DKNF are very, very
rare; a fortunate exception is the one given by Wikipedia.
236 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

TABLE 3.23  An Example of a Table in 5NF, but Not in DKNF

COUNTRIES (x, Country, CountryCode, Capital)

x Country CountryCode Capital Population

autonumber(3) ASCII(128) ASCII(8) Im(CITIES.x) [103, 2*109]
NOT NULL NOT NULL NOT NULL
1 U.S.A. USA 6 313,000,000
2 U.K. UK 2 63,200,000
3 Germany D 3 82,000,000
4 France F 1 65,650,000
5 Romania RO 4 19,050,000
6 Canada C

“domain constraint” Capital [ [1, 6], so Table 3.23 would be in DKNF

in their opinion.
Obviously, this is not true, as inclusion dependencies are not particular
types of domain constraints (e.g., Capital [ [1, 6] may be equivalent to
Capital # CITIES.x only for some periods of time, but extremely rarely
forever): on the contrary, they could rather be thought of as dynamic exten-
sions of the domain constraints, which are static (as both involved map-
ping images may and generally are changing in time, most of the times
even very frequently).
In other words, as we’ve already seen in Subsection 3.2.4 above,
domain constraints for foreign keys are derived from the corresponding
referential integrity ones, not vice-versa.
Please also note that DKNF does not mention FDs, MVDs, JDs or
any other constraint types, but only domain and key ones, so its asso-
ciated implication problem is relatively simple to solve, as keys are
particular types of FDs and domain constraints do not contribute to
implication.
Unfortunately, DKNF neglection of referential integrities too makes
it unacceptably vulnerable, as it does not ban dangling pointers. Conse-
quently, we defined a higher NF, the Domain-Key Referential Integrity one
(DKRINF), which is presented in the second chapter of the second volume
of this book and which adds to DKNF referential integrities too.
The
 Quest for Data Independence, Minimal Plausibility, and Formalization 237

3.11  BEST PRACTICE RULES

The U.S. states that allow for citizens’ initiatives tend to have fewer laws
and lower taxes that the ones that don’t. But the beauty of the system is
that it encourages the spread of best practices.
—Daniel Hannan

At the RDM level, to the db axioms presented in section 5.1 bellow cor-
respond the following best practice rules (which are divided into two cat-
egories, corresponding to the DDL and DML, respectively).

3.11.1  RELATIONAL SCHEMAS

Knowledge and error flow from the same mental sources,

only success can tell the one from the other.
—Ernst Mach

3.11.1.1 Tables

Einstein’s results again turned the tables and now very few philoso-
phers or scientists still think that scientific knowledge is, or can be,
proven knowledge.
—Imre Lakatos
R-T-1. (Table creation, update, and drop)
1. For any dynamic fundamental data object set having at least one
dynamic property a corresponding table should be added in the
corresponding relational scheme. For example, even for a set of
configuration data always having only one element, a correspond-
ing table (which will always have only one line and one column)
should be added.
2. On the contrary, for (not only small instances) static sets (i.e., sets
whose elements should never be updated; for example, people title
names, rainbow colors, payment methods, etc.), corresponding
static enumerated sets or ranges may be used instead in the cor-
responding “foreign-like” keys.
238 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

3. Sets having only one calendar date property should be relationally

implemented as columns (e.g., instead of a table BILLING_DATES
and possibly several foreign keys referencing it, no table should be
added, but the corresponding foreign keys have to be downgraded
to normal, date type, BillingDate columns).
4. When updating table schemas having nonvoid instances, adding
new compulsory columns (or adding NOT NULL constraints to
existing ones) will not be accepted by some RDBMS (including
Oracle); obvious standard solution to this problem is saving cor-
responding instance into a temporary table, deleting it, adding(/
modifying) column, reimporting instance back (also providing val-
ues for the altered column whenever needed), and finally dropping
the temporary table. Always remember in such cases (and, gen-
erally, in all cases having a potential risk of data loss) Murphy’s
Law “If something may go wrong, it will!”: backup involved data
before deletion/update and embed corresponding SQL statements
in a transaction that should commit only when things went ok and
rollback otherwise.

R-T-2. (Relational set inclusion implementation)

1. Any first pair <subset, set> of objects (let them be denoted as R
# S, and having primary keys #R and #S, respectively) should be
relationally implemented with a pair of corresponding tables as fol-
lows: #R should also be declared as a foreign key referencing #S
(note that #R cannot be declared as an autonumber!); if S is not a
subset, then primary key #S should be declared as an autonumber;
otherwise, it should be declared only as an integer.
2. Any other subsequent pair <subset, other_set> (let them be denoted
as R # T, and having primary keys #R and #T, respectively) should
be relationally implemented with a pair of corresponding tables as
follows: an one-to-one not null column labeled T should be added
to table R as a foreign key referencing #T; if T is not a subset, then
primary key #T should be declared as an autonumber; otherwise, it
should be declared only as an integer.
The
 Quest for Data Independence, Minimal Plausibility, and Formalization 239

R-T-3. (DKRINF—Domain-Key Referential Integrity Normal Form)

Any fundamental db table scheme should be at least in the DKRINF.
Please note that RDM does not provide any mean to achieve DKRINF,
but there are algorithmically approaches that translate semantic NF sche-
mas of higher level data models (e.g., (E)MDM) into DKRINF ones (see
Algorithm A7, in the second volume of this book).

3.11.1.2 Keys

Angels are like diamonds. They can’t be made, you have to find them.
Each one is unique.
—Jaclyn Smith
R-K-1.(Primary keys)
Recall that, by definition, primary keys are unique and do not accept
NULLs.
1. Any fundamental table should have a primary key (implementing
the corresponding surrogate key).
2. Any primary key should be of integer type (range restricted accord-
ing to the maximum possible corresponding instances cardinal-
ity—see R-DA-12), and except for subsets (see R-T-2) they should
be of the autonumber type.
3. By exception, not referenced tables might have concatenated pri-
mary keys instead, provided that all of their columns are of type
integer. You should never make such exceptions for referenced
tables: as soon as such a table is referenced, the former primary
key should be downgraded to a non primary (semantic) one, and a
sole surrogate primary key should be added to it.
4. Derived/computed tables may have no keys (be them primary or not).

R-K-2.(Semantic keys)
1. Any fundamental table should have all of the corresponding
semantic keys. With extremely rare exceptions (see the rabbit
cages example in R-DA-15 above), any nonsubset table should
consequently have at least one semantic key: any of its lines should
differ in at least one non primary key column (note that there is an
infinite number of NULLs, all of them distinct!).
240 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

2. Only subset tables (see R-T-2 above), poultry/rabbit cage-type

tables, and derived/computed ones might not have semantic keys
(but they may have such keys too, as well).

3.11.1.3  Foreign Keys

All of life is a foreign country.

—Jack Kerouac
R-FK-1.(Foreign keys)
1. Any foreign key should always reference the primary key of the
corresponding table. Hence, it should be a sole integer column:
dually, you should never use concatenated columns or non integer
columns as foreign keys.
2. Moreover, their definitions should match exactly the ones of the
corresponding referenced primary keys (see R-K-1 above).
This rule is not simply about minimum db space, but mainly for pro-
cessing speed:
Ø numbers are processed by the fastest logic subprocessors, the arith-
metic-logic ones (and the smaller the number, the fastest the speed:
for not huge numbers, only one simple CPU instruction is needed,
for example, in comparing two such numbers for, let’s say, a join),
whereas:
Ø character strings need the slowest processors (and a loop whose
number of steps is directly proportional to the string lengths);
Ø moreover, nearly a thousand natural numbers, each of them between
0 and nearly 4.3 billion, are read from the hard disk (the slowest
common storage device) with only one read operation (from a typi-
cal index file),
Ø while reading a thousand strings of, let’s say, 200 ASCII characters
needs 6 such read operations…

3.11.2  DATA MANIPULATION

An expert is a person who has made all the mistakes that can be made in
a very narrow field.
—Niels Bohr
The
 Quest for Data Independence, Minimal Plausibility, and Formalization 241

3.11.2.1  General Rules

You have to learn the rules of the game. And then you have to play
better than anyone else.
—Albert Einstein
R-G-0.(Think in sets of elements, not elements of sets)
Whenever you design queries, think in terms of sets of elements, not ele-
ments of sets.
For example, by using either extended or embedded SQL, you can
always process rows of tables one by one, just like in ordinary sequential
programming; however, especially as you are expected to always design
the fastest possible solutions, do not ever forget that SQL is optimized for
performing in parallel any (parameterized) data processing tasks on any
number of rows and not just one.
Consequently, for example, whenever not compulsory (which is
extremely rare), use pure SQL and forget about cursors.

R-G-1.(Use parallelism)
Whenever possible, be it for querying and/or updating, use parallel pro-
gramming. For example:
Ø whenever available, partition large tables instances (e.g., by months/
years/etc.) and use parallel queries on several such partitions;
Ø whenever available, use parallel-enabled functions (including
pipelined table ones), which allows them to be used safely in slave
sessions of parallel DML evaluations;
Ø process several queries in parallel by declaring and opening
multiple explicit cursors, especially when corresponding data is
stored on different disks and the usage environment is a low-
concurrency one.

R-G-2.(Minimize I/Os)
Always factorize I/O operations, for keeping them to the minimum pos-
sible. For example:
Ø always use only one UPDATE SQL statement to update several col-
umns simultaneously, instead of several such statements—one per
column; for example, replace:
242 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

UPDATE T SET A = v;
UPDATE T SET B = w;
by:
UPDATE T SET A = v, B = w;
Ø always update only needed rows; for example, replace:
UPDATE T SET A = TRUE WHERE B = w;
by:
UPDATE T SET A = TRUE
WHERE B = w AND A = FALSE;
Ø when selecting data, always select only needed columns, from only
needed tables, and eliminate all unneeded rows as soon as pos-
sible; for example, if, in the end, you only need male person names,
(even if you are for the time being, for example, only a PHP +
MySQL developer) replace:
SELECT * FROM PERSONS;
by:
SELECT Name FROM PERSONS WHERE Sex = ‘M’;
Ø use cache and, generally, internal memory always when available;
for example, small static tables, functions, and queries should be
cached whenever they are frequently used.

R-G-3.(Reuse db connections)
1. Avoid continually creating and releasing db connections.
2. In web-based or multi-tiered applications where application serv-
ers multiplex user connections, connection pooling should be used
to ensure that db connections are not reestablished for each request.
3. When using connection pooling always set a connection wait time-
out to prevent frequent attempts to reconnect to the db if there are
no connections available in the pool.
4. Provide applications with connect-time failover for high-availabil-
ity and client load balancing.

R-G-4.(Avoid dynamic SQL)

Whenever possible, avoid dynamic SQL and use static one instead; when
it is absolutely needed, prefer the native dynamic one and keep dynamicity
to the minimum possible.
The
 Quest for Data Independence, Minimal Plausibility, and Formalization 243

Besides poor execution speed, note that successfully executing a

dynamic SQL statement also involves an automatic commit performed by
the system, which may have undesired side effects on concurrency and
data integrity.

R-G-5.(Minimize undo segments querying)

From the application design phase, always keep querying in parallel with
updating same data to the absolute minimum possible (in order to mini-
mize undo segments querying).

R-G-6.(Unit testing)
Do not ever deliver any query, view, stored procedure, function, or trigger
without, at least, completely and thoroughly unit testing it.
All subsequent tests should be run only with nearly perfectly function-
ing units.

3.11.2.2  Data Modifications

Our creation is the modification of relationship.

—Rabindranath Tagore
R-M-0.(Backup)
Before modifying data values, especially for massive updates and deletes,
always backup the corresponding table instances (so that if anything goes
wrong you may restore previous values).
Alternatively, you can disable automatic committing and issue explicit
COMMITs only after checking that everything went as desired.
Moreover, regular backup (either full or incremental) of any db is crucial to:
Ø avoid data loss and
Ø minimize db recovery time.

R-M-1.(Test SELECTs first)

Before modifying data values by using SELECT subqueries/clauses, espe-
cially for massive modifications, firstly test corresponding SELECT state-
ment results, for checking whether or not they correctly compute the set of
desired rows (be it for insertion, update, or deletion).
244 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

For example, before executing statement DELETE FROM T WHERE

x IN (SELECT …);, first execute its subquery and make sure that it
returns all and only the rows that you want to delete from table T.

R-M-2.(Use rollback transactions)

Even when “nothing can go wrong”, embed data modification statements
in transactions having not only COMMITs, but ROLLBACKs too: not even
RDBMSs are perfect, communications may fail, power outages may
occur, etc.!

R-M-3.(Caution with the ON DELETE CASCADE “atomic bombs”)

Use the ON DELETE CASCADE predicate with same caution as when
manipulating atomic bombs: it is a potential data “mass-extinction”
weapon!

R-M-4.(Rule out dirty nulls)

Whenever available, instruct DBMSs to ban dirty nulls (not null, but
unprintable characters); when such a feature is not available, add it to your
dbs through triggers or trigger-type methods.

R-M-5.(Immediately purge unneeded old data)

Do not forget that, generally, when executing SQL DELETE statements,
RDBMSs do not physically, but only logically delete (in fact, they mark
for deletion) corresponding deleted data: immediately when it becomes
obsolete, purge it explicitly.

3.11.2.3 Querying

Sometimes it is not enough to do our best; we must do what is required.

—Winston Churchill

R-Q-0.(Data and processing minimality)

Always use only needed data and operations: for example,
• do not select more data than needed;
• do not use more table instances than needed;
The
 Quest for Data Independence, Minimal Plausibility, and Formalization 245

• filter data as soon as possible;

• do not group data when not needed;
• do not order data until the end and never beyond a key.
Moreover,
• decompose your complex queries into the smallest possible sub-
queries,
• program each of them only using primary surrogate keys and cor-
responding foreign keys, and
• get semantic needed data only in the last step.

R-Q-1.(WHERE-HAVING rule)
Always use the WHERE clause to eliminate all unneeded rows before
grouping (if any); always use the HAVING clause only for conditions
involving application of aggregation functions on groups.
Dually, never place on HAVING a filter that can be placed on WHERE.

R-Q-2.(WHERE rules)
a. In any adjacent conditions pair, the first condition should eliminate
more rows than the second one.
b. Do first inner and then outer joins.
c. Join tables in their increasing number of rows order.
d. Do not prevent index usage by applying unneeded functions to the
corresponding columns; take also care of implicit conversions.
e. Not only in Oracle, if you use filtered Cartesian products instead of
explicit joins, always place non join filters before join conditions.

R-Q-3.(Cautiously use static subqueries)

Subqueries allow for much more elegant and close to mathematical logic
queries, but, unfortunately, due to lack of adequate RDBMSs optimiza-
tions, even the static (uncorrelated) ones are generally less efficient than
corresponding equivalent subqueryless queries: consequently, use them
cautiously.

R-Q-4.(Avoid dynamic subqueries)

Use dynamic (correlated) subqueries if and only if there is no other
equivalent solution or speed is not critical (recall that, unlike for static
246 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

(uncorrelated) ones, for which subqueries are evaluated only once, before
evaluating the corresponding query, the dynamic ones need to be reevalu-
ated for each row in the corresponding query).

R-Q-5.(Always sort intelligently)

Never present unordered results (except for cases when explicitly asked
to do so). Always order intelligently, in the best possible meaningful
sorting order. Do not use column positions for sorting, but only column
names: code maintenance and debugging are much easier when using
names.

3.11.2.4  Extended SQL

Once you have a firefighter in your family,

your family and the families from his crew become one big extended family.
—David Leary
R-E-1.(Use named constants)
On one hand, do not use hardcoded, but named constants: your code
becomes much easier to understand and maintain. On the other, whenever
possible, do not use constant functions instead of constants: your code
becomes slower.
For example, instead of:
if obType = 1 then … // 1 is a hardcoded constant
first declare, for example, the named integer constant Orders:
Orders constant int := 1;
and then use it:
if obType = Orders then …
Others, unfortunately always use constant functions instead:
FUNCTION Get_Orders_ID(b boolean default true)
RETURN int;
…
FUNCTION Get_Orders_ID(b boolean default true)
RETURN int is
BEGIN
return 1;
The
 Quest for Data Independence, Minimal Plausibility, and Formalization 247

END Get_Orders_ID;
…
if obType = Get_Orders_ID then …
It is true that, for example, unfortunately, in Oracle, you can use named
constants only in PL/SQL, but not also in SQL (like, for example, DDL
statements, views, triggers, etc.), which needs constant functions instead;
however, this is not a reason not to use constants in PL/SQL.

3.12  THE MATH BEHIND RDM

Difficulties mastered are opportunities won.

—Winston Churchill

Definition 3.0 The scheme of a rdb is a triple < S, F, C >, where:

Ø S is a finite nonempty collection of atomic sets (i.e., they are not
Cartesian products or power sets, etc.) having unique names (across
the rdb);
ü (S, #) is a poset;
ü S = R !V !CS ! {NULLS, Ø}, where:
• R is a finite nonempty collection of finite (fundamental
object) sets called relations;
• V is a finite nonempty collection of (value) sets called
domains;
• CS is a finite collection of finite (object) sets computed
from those in R (by using RA set operators);
• NULLS is a distinguished countable set whose elements
are called nulls;
Ø F # (R !V ! CS ! {NULLS})R !CS is a finite nonempty set of
functions called (relation) attributes;
ü ;R[R !CS, let f1, …, fk, k > 0 natural, be all F functions
defined on R (note that they all must have distinct names);
• the product f1 • … • fk : R → codom(f1) × … × codom(fk) is
called a relation scheme (of R);
• its graph is called the corresponding relation instance (con-
tent) or sometimes, by notational abuse, simply relation;
248 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

• the union of all relation schemas is called the db scheme;

• the union of all relation instances is called the db
instance (content) or sometimes, by notational abuse,
simply database;
ü F = A !M !CF, where:
• A # (V ! {NULLS})R !CS is a finite nonempty set of
(fundamental) functions;
• M # (R !CS ! {NULLS})R !CS is a finite set of (struc-
tural fundamental) functions; ;f[M, f : R → R’ (R, R’[R
!CS), and ;R’ attribute g : R’ → codom(g), there is a
mapping fk = g ° f : R → codom(g), called a foreign key (of
R, referencing g in R’); trivially, Im(fk) # Im(g), which in
RDM is called a referential integrity constraint (or a typed
inclusion dependency) and is abusively denoted by fk # g;
note that both g and fk may be function products too and
that R and R’ need not be distinct; also note that, unfor-
tunately82, all functions in M are implemented in RDM
relations by a corresponding foreign key;
• CF # (R !V !CS ! {NULLS})R !CS is a finite set
of computed functions from those of A !M (by using
function algebra and RA function operators);
Ø C is a finite nonempty set of closed first order logical formulas
called constraints (or dependencies) whose predicates are either
standard mathematical or existence ones associated to the sets in
R and whose names must be unique (across the rdb);
ü C = D ! E ! K ! T, where:
• D is a finite nonempty set of domain constraints of the
type codom(f) = V[V ! {NULLS}, ;f[F, generally
denoted f # V; note that, unfortunately, domain con-
straints have to be explicitly asserted for foreign keys too
(although they could have been automatically inferred
from the corresponding referential integrity constraints)
and, even worse, they may be oversized (e.g., codom(fk)
. codom(g), see foreign keys definition above); also note

82
Unfortunately especially as g needs not be one-to-one, which would have always guaranteed that fk
is not only existing, but is also unique.
The
 Quest for Data Independence, Minimal Plausibility, and Formalization 249

that, unfortunately, V needs not to be finite83, which is not

guaranteeing data plausibility;
• E is a finite set of existence constraints denoted f |— g
(where f and g [ F, are distinct of each other and have a
same domain R), which means: (;x[R)(f(x)ÓNULLS 
g(x)ÓNULLS). Note that, in fact, no RDBMS is providing
such constraint type, but only its particularization Ø |— f,
meaning codom(f) ∩ NULLS = Ø, which are called NOT
NULL (or required values) constraints. Also note that,
unfortunately, both RDM and RDBMSs allow for empty
sets of NOT NULL constraints, thus allowing for empty
rows in tables, which are, trivially, senseless.
• K is a finite set of key constraints, denoted generally f key,
f[F, and meaning that f is minimally one-to-one. Note
that f may be a function product and that, unfortunately,
this set may also be empty (both in RDM and RDBMSs),
thus allowing for table instances containing from redun-
dant and confusing duplicates to only a same row dupli-
cated billions of times, and that (for function products),
although RDM makes the difference between one-to-one
(called superkeys) and minimally one-to-one (keys) ones,
RDBMSs do not: you can freely declare any combination
of keys and superkeys (which is trivially absurd, as, on
one hand, they could easily prevent superkeys declara-
tions and, on the other, they are losing both disk space and
data manipulation speed with every enforced superkey). A
mapping which is either one-to-one or is a member of at
least one minimally one-to-one mapping product is said to
be prime (and otherwise nonprime).
• T is a finite set of tuple (or check) constraints, for which there
is neither a RDM commonly agreed precise definition, nor
a RDBMS industry standard. Generally, they are restricted
83
Even if, in actual RDBMS implementations, all value sets (generally called, just like in computer
programming, data types) are finite, they are too large: for example, Oracle allows calendar dates
starting with 4712 BC and up to 9999, while MS Access, between years 100 and 9999; if you only
declare that employees birth date is a calendar date (which is, unfortunately, not only possible, but the
de facto standard practice), then end-users might store data about employees being born some nearly
7,000 years ago or to be born in some 7,000 years from now on!
250 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

forms of first order logic propositions, denoted as zero-th

order logic ones: they only mention functions defined on a
same relation and use only one universally quantified vari-
able (which allows for discarding it: for example, instead
of ((;x[PEOPLE)(BirthDay(x) ≤ DeathDay(x)), Birth-
Day ≤ DeathDay is written instead). Note that, except for
referential integrities, there are no other interrelational,
but only intrarelational constraints (which is natural for
the domain, existence, and key types ones, but constitutes
a severe limitation for the tuple ones: for example, Birth-
Day ≤ MarriageDay ≤ DivorceDay ≤ DeathDay, that is, for
example, (;x[PEOPLE) (;z[Marriages) (Husband(z)
= x  BirthDay(x) ≤ MarriageDay(z) ≤ DivorceDay(z)
≤ DeathDay(x)), is not assertable either in RDM or in
RDBMSs, as it needs two universally quantified variables
and functions defined on two sets).

Definition 3.1 (First Normal Form) A relation scheme for which all
function members are atomic, that is they are not either function products
or taking values from Cartesian products or power sets, etc., is said to be
in the First Normal Form (1NF). An rdb scheme whose relations are all in
1NF is in 1NF too.

Definition 3.2 (function product chains) A function product chain (or

simply chain) of length n, n > 1, natural, made up of subsets of a function
product Πnj=1Sj is a sequence M1 , M2 , … , Mn-1 , Πnj=1Sj such that |Mi|
= i, for i = 1, 2, …, n – 1.
Note that, trivially, the number of chains of length n that can be con-
structed in Πnj=1Sj, n > 0, natural, is n!

Proposition 3.0 (Maximum number of keys) For any relation scheme

ô = f1 • … • fn : O → codom(f1) × … × codom(fn), n > 0 natural, ;O[R
!CS, the maximum number of keys is C(n, [n/2]).
Proof:
Let X , ô, |X|= r, r < n; the number of chains of length n that include X
and have the form M1 , …, Mr-1 , X , Mr+1 ,…. , ô is equal to r!(n–r)!.
The
 Quest for Data Independence, Minimal Plausibility, and Formalization 251

Let 1 ≤ i, j ≤ n, i ≠ j; by definition, Xi and Xj might simultaneously be

keys if Xi ÷ Xj and Xj ÷ Xi. Obviously, if Xi and Xj are not #-comparable,
then any chain containing Xi is different of any chain containing Xj.
Denoting by p the number of possible simultaneous keys and by |Mi|=ni,

it follows that this

implies that:

Conversely, by considering ô subsets family having m = [n/2] ele-

ments, it can be easily proved that max(p) ≥ C(n, [n/2]). Consequently, it
follows that max(p) = C(n, [n/2]). Q.E.D.

Proposition 3.1 (Algorithm A7/8–3 characterization) A7/8–3 has the

following properties:
(i) its complexity is O(2m);
(ii) it is complete (i.e., it is generating all possible keys);
(iii) it is sound (i.e., it is generating and outputting only possible keys);
(iv) it is relatively optimal (i.e., it generates the minimum possible
number of questions relative to its strategy).
Proof:
(i) (Complexity) Obviously, as it has three embedded loops, all of
them finite (hence, it never loops infinitely) and the first one may
take at most 2m – 1 steps (as all superkeys may have been wrongly
declared as keys), the second one may take at most C(m, [m/2])
steps (as the maximum possible number of wrong keys may have
been previously declared), while the third one may take at most
2m – 1 – m steps (when no keys are found). Consequently, in the
worst case scenario, the total number of steps is 2m – 1 + C(m,
[m/2]) + 2m – 1 – m = 2m + 1 + C(m, [m/2]) – m – 2.
(ii) (Completeness) All possible keys are considered, as the algorithm
is generating in its third loop all 2m – 1 non-void combinations of
any set of mappings having cardinal m or all of the maximum C(m,
[m/2]) possible simultaneous keys.
(iii) (Soundness) Moreover, as each one of them is first checked to
ensure that it is not a superkey and only then db designer is asked
252 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

whether it is a key or not, it follows that the algorithm is computing

only possible remaining keys.
(iv) (Optimality) In all three loops, each key is eliminating the maxi-
mum possible number of further questions to ask db designers and
no possible key is processed more than once. Q.E.D.

Proposition 3.2 (Key Propagation Principle formalization) Let R and

R’ be any (object) sets (not necessarily distinct), f : R → R’ any function
between them and a : R’ → A be any one-to-one function or function (Car-
tesian) product of R’;
(i) '!fk : R → A such that fk = a ° f (with fk called a foreign key of R
referencing a of R’ and being considered as a key “propagated”
from R’ to R):

(ii) Im(fk) # Im(a) (called the associated referential integrity constraint)
(iii) card(Im(fk)) = card(Im(f))
(iv) if f is one-to-one, then fk is one-to-one too.
Proof:
(i) Let fk : R → A such that, ;x[R, fk(x) = a(f(x)); trivially, fk is totally
defined: ;x[R, 'y[Im(f) # R’ such that y = f(x) and 'z[Im(a) #
A, such that z = a(y) = a(f(x)) = fk(x); moreover, fk is well-defined:
;x, y[R such that x = y, according to a ° f definition, a(f(x)) =
a(f(y))  fk(x) = fk(y).
Obviously, fk = a ° f: ;x[R, let y = a(f(x)) [ Im(a) # A; according
to fk definition, fk(x) = y  a(f(x)) = fk(x)  a ° f = fk.
Moreover, fk is the only function having these properties: let us
suppose that 'fk’: R → A such that a ° f = fk’  ;x[R, a(f(x)) =
fk’(x); according to fk definition, fk(x) = a(f(x))  fk(x) = fk’(x) 
fk ≡ fk’.
(ii) Trivially, ;y[Im(fk), 'x[R such that y = fk(x); as f is totally
defined, it follows that 'z[R’ such that z = f(x); as a is totally
defined, it follows that 'u[ Im(a) such that u = a(z) = a(f(x)) =
fk(x) = y.
The
 Quest for Data Independence, Minimal Plausibility, and Formalization 253

(iii) Trivially, ;y[R’ \ Im(f), ¬('x[R such that y = f(x))  ¬('z[A

such that a(y) = z[Im(a°f))  card(Im(fk)) = card(Im(a°f))
≤ card(Im(f)); conversely, let us suppose that card(Im(fk)) <
card(Im(f)); it follows that 'x, y[R such that x ≠ y and fk(x) = fk(y),
but f(x) ≠ f(y); as a is one-to-one, it follows that a(f(x)) ≠ a(f(y))
 fk(x) ≠ fk(y)  ¬(card(Im(fk)) < card(Im(f)))  card(Im(fk)) =
card(Im(f)).
(iv) Let f and a be one-to-one and suppose that fk it is not  'x, y[R
such that x ≠ y, but fk(x) = fk(y); as f is one-to-one, f(x) ≠ f(y); as a
is one-to-one too, fk(x) =a(f(x)) ≠ a(f(y)) = fk(y)  fk(x) ≠ fk(y).
Q.E.D.

Note that, trivially, even if a is not one-to-one, then fk still exists, it is the
only function satisfying 3.2(i) above, and its associated referential integrity
constraint still holds (as a’s one-to-oneness is not needed in their proofs).
Unfortunately, this allows RDM to define foreign keys referencing any
columns or column products, be them one-to-one or not.
Obviously, referencing a column or column product which is not one-
to-one has two disadvantages (also see Exercise 3.43) and no advantage;
let us consider, for example, a = CityName : CITIES → ASCII(256), which
is not generally one-to-one, as there may be cities having same names
even inside a same country; suppose, for example, that CityName(1) =
CityName(2); trivially, any foreign key fk referencing CityName has the
following two disadvantages:
• card(Im(fk)) ≤ card(Im(f)), that is fk is not always unambiguously
identifying the elements of R’ computed by f; for example, if f =
BirthPlace : PEOPLE → CITIES, BirthPlace(1) = 1, BirthPlace(2)
= 2, then, although, in fact, people 1 and 2 were born in different cit-
ies, the corresponding foreign key fk = CityName ° BirthPlace would
mislead us by indicating that both of them were born in a same city
(as, in fact, it does not compute unique cities, but their names, which
are the same in this case)
• fk does not preserve anymore f’s one-to-oneness; for example, if
f = Capital : COUNTRIES → CITIES, which is obviously one-to-
one (as no city may simultaneously be a capital of two or more
countries), Capital(1) = 1, Capital(2) = 2, then, although, in fact,
254 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

countries 1 and 2 have as capitals different cities, the correspond-

ing foreign key fk = CityName ° Capital would mislead us by
indicating that both of them have as capital a same city (for rea-
sons similar to the above ones).
Consequently, you should never use foreign keys referencing not one-
to-one columns or column products; moreover, as all keys of any set are
equivalent (see Exercise 3.42(iv)), for minimality reasons (both as needed
disk storing space and, especially, querying speed when joining R and R’
for finding all x[R such that either fk(x) = a(f(x)) or fk(x) ≠ a(f(x))) the best
solution is to always reference minimal single keys of R’; trivially, when
not erroneously oversized, always the best such choice is the surrogate
(primary) key of R’.
Moreover, (ii) above proves that referential integrities are implied by
the KPP, so that it is not necessary to assert them explicitly.

Definition 3.3 (Functional dependencies) Let f : S → A and g : S → B

be any two attributes of S; f is said to functionally determine g (or g is said
to be functionally dependent on f, or that there is a functional dependency
between f and g, denoted f → g) iff ker(f) # ker(g).

Proposition 3.3 (Functional dependencies characterization) f → g 

'! fd : Im(f) → Im(g) such that g = fd ° f:

Proof:
() Let us define fd : Im(f) → Im(g) such that ;y[Im(f), fd(y) = z,
where y = f(x) and z = g(x), and firstly prove that fd is well-defined:
(fd totally defined) According to mapping image definition, ;y[Im(f),
'x[S, y = f(x); as g is totally defined, ;x[S, 'z[Im(g), z = g(x) 
;y[Im(f), fd(y) = z.
(fd functional) Let t, u[Im(f), t = u; by fd definition, ;x, y[S such
that fd(t) = v, where t=f(x), v=g(x), and fd(u)=w, where u=f(y) and w=g(y);
t=u f(x)=f(y)  (x, y)[ker(f); as ker(f) # ker(g), it follows that (x, y)
[ker(g) g(x) = g(y) v = w  fd(t) = fd(u).
The
 Quest for Data Independence, Minimal Plausibility, and Formalization 255

Let us also show that g = fd °f; ;x[S, denote z = g(x)[Im(g) and y = f(x)
[Im(f); according to fd definition, fd(y) = z  fd(f(x)) = g(x)  fd °f = g.
Finally, we prove that fd is unique: assume that 'h : Im(f) → Im(g)
such that h °f = g; it follows that, ;x[Im(f), 'y[S, x = f(y) such that h(f(y))
= g(y); according to fd definition, fd(x) = g(y) fd(x) = h(x) fd ≡h.
() Assume ¬(f →g); ¬(ker(f) #ker(g))  '(x, y) [ ker(f) such that
(x, y) Óker(g), that is '(x, y)[S2, f(x) = f(y) and g(x) ≠g(y); as fd functional,
f(x) = f(y) fd(f(x)) = fd(f(y)), which, by fd definition, implies that g(x)
= g(y); this trivially means that the assumption is false and, consequently,
f →g. Q.E.D.

Note that the functional dependency relation is a (partial) preorder (see

Exercise 3.2) and that it is neither symmetric, nor antisymmetric.

Proposition 3.4 (Algorithm A1–7 complexity) Let the total number of

distinct fundamental rectangles (entity-type sets) be e and the correspond-
ing ones for fundamental diamonds (relationship-type sets) be d, for roles
(canonical Cartesian projections) be r, for arrows (functional relation-
ships) be f, for ellipsis (attributes) be a, for restrictions (constraints) be c,
and for computed sets (views) be v (e, d, r, f, a, c, v naturals); A1–7 has
complexity O(e + d + r + f + a + c + v).
Proof:
Obviously, A1–7 only processes once any rectangle, diamond, role,
arrow, ellipsis, and constraint; its main method (Fig. 3.7) has three loops:
the first one is executed e times, the third one v times, and the second one
is executed d + r times, with each of the e + d times calling method Crea-
teTable once and with each of the r times calling method addForeignKey
once; method CreateTable (Fig. 3.8) also calls method addForeignKey in
a loop for a total number of i times, where i is the total number of canoni-
cal inclusions (i natural, i ≤ f); if there is no surrogate key for the current
object set, then CreateTable executes a supplementary step for it, in order
to automatically add a primary key to the corresponding table; methods
addForeignKey (Fig. 3.9) and addColumn (Fig. 3.11) always perform only
one step; finally, method completeScheme (Fig.3.10) has six loops: the
first one is performed in total a times; the second one for f – i times; the
last four ones, together, for c times; trivially, all of the above loops are
256 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

finite (hence, A1–7 never loops infinitely). Consequently, in the worst case
scenario (when no surrogate key is declared in the input E-R data model),
the total number of steps is 2*(e + d) + r + f + a + c + v. Q.E.D.
Proposition 3.5 (Algorithm REA1–2 complexity) Let t be the total num-
ber of tables and views of S and the corresponding ones for foreign keys be
f, for non foreign keys be a, for keys be k, for NOT NULL constraints be
n, and for tuple (check) constraints be c (t, f, a, k, n, c naturals); REA1–2
has complexity O(t + 2*a + f + k + n + c).
Proof:
Obviously, REA1-2 only processes once any table, view, column (be it
foreign key or not), and constraint (be it domain, key, NOT NULL, or tuple/
check); it has one outer loop, which is executed t times, and which embeds
other four ones: in total, the first one is executed a times (for creating the
ellipses), the second one f times (for the arrows), the third one k times (for
the keys), and the last one a + n + c times (for adding the corresponding
domain, NOT NULL, and tuple/check constraints, respectively); trivially,
all of the above loops are finite (hence, REA1-2 never loops infinitely). Con-
sequently, the total number of steps is t + 2*a + f + k + n + c. Q.E.D.

3.13 CONCLUSION

I think and think for months and years. Ninety-nine times,

the conclusion is false.
The hundredth time I am right.
— Albert Einstein

3.13.1  RDM PROS

We chose it because we deal with huge amounts of data.

Besides, it sounds really cool.
— Larry Page

RDM introduction had as a declared goal guaranteeing data independency

from the physical, implementation level and outcoming restrictions of pre-
vious data models.
The
 Quest for Data Independence, Minimal Plausibility, and Formalization 257

Its most frequently used predecessors, the hierarchical and network

data models, include explicit references to implementation characteristics,
like pointers, which are almost excluding any potential user who is not a
computer programmer. Moreover, these models lack formal definitions,
hence precision, consistency, and rigor.
However, RDM’s success is not based only on math formalization and
the high data independency level from any possible physical implementa-
tion, but also on the simplicity of the sole used data structure (the relation),
on its natural and expressive representation as a table, and, above all, on
the power and elegance of the relational query languages.
RDM is a cornerstone in conceptual data modeling and querying at
least for the following four reasons:
1. The vast majority of both commercial and open source existing
DBMS are based on it for already some 35 years and even today,
with the advent of NoSQL ones, RDM remains the best for storing
and high speed processing of homogeneous data, even for the fore-
seeable future.
2. Any higher-level conceptual homogeneous data model has to
include an algorithm for translating its schemas into RDM ones,
as the most natural way to store discrete homogeneous data is
table-like.
3. The vast majority of data query languages, be them relational or
not, are based (externally) on the relational calculus (i.e., are SQL-
like and/or QBE-like) and (internally) on the relational algebra,
SQL being a de facto db standard.
4. The RDM mathematic theory is still open and fundaments or at
least inspires the majority of research endeavors in other data and
knowledge models, as well as the general discrete models theory,
logic, and even imperative procedural programming.
RDM theory has two major parts: the constraint (dependency) and the
query ones. Although it can almost entirely be considered as a (finite) spe-
cialization of mathematical logic (with its two parts corresponding to the
closed and, respectively, open first order formulae), its roots come also
from the seminaïve algebraic theory of sets, relations and functions, as
well as from the computational complexity, computer programming, for-
mal languages, etc.
258 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

Besides the five basic type ones implemented by nearly all RDBMSs,
dependency theory formalized a plethora of integrity constraints, out of
which only the most “famous” ones are presented in Section 3.10 (some
others are briefly discussed in the last section of this chapter).
Relational languages used by programmers are of logic type, being
declarative and nonprocedural; internally, they are translated by RDBMSs
into optimized relational algebra expressions. Their main limitation (not
being able to compute transitive closures) is overcome by recursive
extensions that are presented in the first chapter of the second volume of
this book.
RDM is also important because of its very interesting interactions with
logic and object-oriented programming, as well as expert systems and AI.
Moreover, I consider that it is a must to include RDM in any db book,
even if the literature in this field is that abundant, at least for stating your
point of view on its most important results, its limits, and how to over-
come them.
Hopefully, besides this point of view, this chapter also provides a
rich collection of correct relational solutions for interesting subuniverses
of discourse, as well as some of our results in the field (especially the
algorithms A1–7, A7/8–3 and REA1–2, the best practice rules, the for-
malization of the SQL GROUP BY clause with quotient sets computed
by the kernel of mapping products equivalence relations, the sections 3.7
–including its RDM and E-R data models of the RDM–, and 3.12 –mainly
including the formalization of the KPP and the proof that any table having
n columns may have at most C(n, [n/2]) keys–, and the subsections 3.2.3.3,
3.2.3.4, 3.2.4.2, 3.2.4.3, and 3.3.2—the relational model of the E-RDM);
the Domain-Key-Referential-Integrity Normal Form (DKRINF) is charac-
terized in the second volume of this book.

3.13.2  RDM CONS

Truth is not defined by how many people believe something.

Ask. Question. Think. Decide for yourself.
—Jonathan Lockwood Huie
The
 Quest for Data Independence, Minimal Plausibility, and Formalization 259

The main disadvantage of RDM is that it is purely syntactic, while, for

example, both domain and key constraints, the only two constraint types
needed for defining DKNF, its highest normal form, as well as the exis-
tence (hence not null) and tuple ones are intrinsically semantic.
Consequently, no algorithms could ever design domain or key or exis-
tence or tuple constraints: the best things that may be designed for discov-
ering and fine tuning them are assistance algorithms, like, for example, A0
(for domain, existence, and tuple ones) and A7/8–3 (for keys).
Moreover, all other RDM constraints are semantic, just like any other
possible constraints, in any other context, except for the trivial and implied
ones (like, for example, the referential integrities, which are not at all
interesting for db design, but, on the contrary, have to be eliminated): for
example, no algorithm could ever decide whether or not a column should
accept nulls in a given context (mandatory type constraints also being rela-
tive: for example, Capital may accept nulls in commercial dbs, but not in
U.N. or geographical ones) or whether or not net salaries should always be
less than or equal to corresponding brut ones.
Even much more convincing is that no algorithm may ever decide that
people birth places are cities, or that folders are files, or that “reports to”
are employees (or anything else), etc. Same goes for FDs, MVDs, JDs,
etc., too.
Second argument is that there are two RDM dual approaches for db
design, both essentially unsuccessful: synthesis (bottom-up) and decom-
position (top-down). Both start with a constraint set (generally made up of
FDs, so keys too, as a particular case, MVDs, JDs, etc.) and:
Ø synthesis—with the set of all needed columns (i.e., unary tables or
vectors), while
Ø decomposition—with a single “universal” table, made out of all
needed columns,
both of them trying to obtain corresponding table schemas, in the desired
NF, preferably also being lossless and preserving dependencies (i.e., all
initial dependencies being retrievable in the final db scheme, no matter in
which table).
Currently, decomposition is the main strategy used for achieving
desired NFs. For example, column CurrencySymbol from Table 3.19
should be moved to a table CURRENCIES; column ZipCode from Table
260 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

3.20 should be moved to a table SMALL_CITIES; Table 3.21 should be

decomposed into tables STUDENTS_COURSES_ENROLLMENTS hav-
ing columns x, Student, Course and keys x and Student • Course and STU-
DENTS_CLUBS_ENROLLMENTS having columns x, Student, Club and
keys x and Student • Club; Table 3.22 should be decomposed into tables
EMPLOYEES having columns x, SSN and keys x and SSN, JOBS_HIS-
TORY having columns x, Employee, Job and keys x and Employee • Job,
EMPLOYEES_SKILLS having columns x, Employee, Skill and keys x and
Employee • Skill, and JOBS_SKILLS having columns x, Job, Skill and keys
x and Job • Skill.
It has been shown that, unfortunately, for both these strategies, obtain-
able results are not at all encouraging, as:
ü it is not always possible to obtain lossless and dependency preserv-
ing results;
ü even when this is achievable, sometimes it is nondeterministic (i.e.,
there are several equivalent solutions, out of which the rdb design-
ers should then choose the “best semantically fitted” one; more-
over, being purely syntactic, RDM cannot provide any guidelines
to do it);
ü the lossless decomposition property does not generally prevent
storing aberrant data;84
ü generally, given an arbitrary set of RDM dependencies (con-
straints), one cannot always obtain any desired RDM NF;
ü even when obtaining of a desired NF is possible, sometimes it takes
much too long to obtain it85, too much to be practically reasonable
to wait for86 and, to further complicate things, sometimes this pro-
cess is also nondeterministic.
ü there is no algorithm for obtaining DKNF, the highest relational
normal form.
Thirdly, both bottom-up and top-down rdb design strategies are essen-
tially trying to compute object sets (i.e., tables, in this paradigm) from
84
Fact that, paradoxically, was and perhaps is still considered by some as an advantage.
85
As some of such algorithms, like, for example, finding tables of an rdb that violate BCNF, are NP-
complete.
86
For example, assuming that an algorithm only needs 2n seconds to decompose a n = 3,000 columns
universal relation (which is only an average actual complex one), this means an unconceivable period
of time: for example, if n = 30, this means some 34.5 years; when powering 34.5 with 100, you have
to start counting in light years already!
The
 Quest for Data Independence, Minimal Plausibility, and Formalization 261

mappings’ domains and math relations’ underlying sets, which is trivially

absurd: for centuries already and for all foreseeable centuries to come,
both relations and mappings (their particular case) are defined on such sets
(and taking values from such sets) and not vice-versa. For example, from
the Altitude mapping values (see Table 3.1), one cannot ever infer that its
domain is the Carpathians, or the Rockies, or the Earth’s, or the Moon’s
mountains, or, even worse, any other object type which is measurable in
meters …
Fourthly, as DKRINF is proving it in the second volume of this book,
DKNF, the “ultimate” NF, is too weak even in relational terms, as it does
not take also into account referential integrity ((typed) inclusion dependencies
(INDs)). Moreover, not even DKRINF, even if it also takes into account INDs,
is not satisfactory either, as it also allows for storing implausible data when
non-relational type constraints are also considered (which is compulsory).
For example, if to the db in Fig. 3.4 we associate the constraint set
CS = PCs ø FDs ø NRC, where:
ü PCs = {COUNTRIES.x [ [0, 999], COUNTRIES.Country [
ASCII(128), CountryCode [ ASCII(8), COUNTRIES.Capital [
[0, 1016], COUNTRIES.Population [ [105, 2×109], COUNTRIES.x
key, COUNTRIES.Country key, CountryCode key, COUNTRIES.
Capital key, COUNTRIES.Capital # CITIES.x, STATES.x [
[0, 108], STATES.State [ ASCII(128), StateCode [ ASCII(8),
STATES.Country [ [0, 999], STATES.Capital [ [0, 1016], STATES.
Population [ [105, 5×107], STATES.x key, STATES.Country •
StateCode key, STATES.Capital key, STATES.Capital # CITIES.x,
CITIES.x [ [0, 1016], CITIES.State [ [0, 108], City [ ASCII(128),
CITIES.Population [ [105, 3×107], CITIES.x key, CITIES.State •
City key, CITIES.State # STATES.x} contains its extended primi-
tive constraints (i.e., also including INDs),
ü FDs = {COUNTRIES.x → COUNTRIES.Country • CountryCode •
COUNTRIES.Capital • COUNTRIES.Population,
COUNTRIES.Country → COUNTRIES.x • CountryCode •
COUNTRIES.Capital • COUNTRIES.Population,
CountryCode → COUNTRIES.x • COUNTRIES.Country •
COUNTRIES.Capital • COUNTRIES.Population,
262 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

COUNTRIES.Capital → COUNTRIES.x • COUNTRIES.Country •

CountryCode • COUNTRIES.Population,
STATES.x → STATES.State • STATES.Country • StateCode •
STATES.Capital • STATES.Population,
STATES.Country • STATES.State → STATES.x • StateCode •
STATES.Capital • STATES.Population,
STATES.Capital → STATES.x • STATES.State • STATES.Country •
StateCode • STATES.Population,
CITIES.x → CITIES.State • CITIES.City • CITIES.Population,
CITIES.City • CITIES.State → CITIES.x • CITIES.Population}
its functional dependencies, and
ü NRC = {STATES.Country ° CITIES.State ° COUNTRIES.Capital =
1COUNTRIES, CITIES.State ° STATES.Capital = 1STATES} includes only
the non-relational constraints “no country may have as its capital a
city of another country” (or, dually, “no city may be the capital of a
country to which it does not belong”) and “no state may have as its
capital a city of another state” (or, dually, “no city may be the capi-
tal of a state to which it does not belong”), this db is not anymore
in DKRINF, as its nonrelational constraints are not implied by its
extended primitive ones.
For example, row <7, “Japan”, “JP”, 5, 128,000,000> is compatible
with table COUNTRIES, but it cannot be added to it, because it would
violate the first non-relational constraint, as 5 in CITIES (see Fig. 3.4)
is New York, which does not belong to Japan, but to the U.S.; similarly,
even if, previously, in table CITIES a row <7, “Buffalo”, 6, 262,000>
were inserted, row <7, “Constanta”, “CT”, 5, 630,000>, although com-
patible with STATES, could not be inserted into this table, as the second
non-relational constraint would be violated (trivially, Buffalo, a city of
the U.S. New York State, cannot be the capital of the Romanian county
Constanta).
Finally (fifthly), as the example above proves it and as this will hope-
fully become crystal clear in the second volume of this book, RDM is
much too poor as constraint types, allowing storage of implausible data
even for very simple DKRINF and DKNF dbs: much higher, semantic
type data models, providing much more powerful constraint types than
The
 Quest for Data Independence, Minimal Plausibility, and Formalization 263

RDM ones are needed for db design, in order to fully guarantee data
plausibility.
For (just another) example, RDM is not able to accept even very com-
mon constraints such as “population of a country should be greater or
equal than the sum of populations of its states” and “population of a state
should be greater or equal than the sum of populations of its cities”, which
are very similar to check (tuple) ones, except for the fact that they involve
columns from different tables.
Consequently, db design should always be done at a higher, semantic
level and RDM should only be used as an abstraction layer above RDBMS
versions.
On the contrary, querying should always be done relationally, as it is
not only powerful, simple, and elegant, but also benefits from very fast
processing.

3.14 EXERCISES

Exercise to stimulate, not to annihilate. The world hasn’t formed in a day,

and neither were we.
Set small goals and build upon them.
—Lee Haney

3.14.1  SOLVED EXERCISES

Problems are only opportunities in work clothes.

–Henry J. Kaiser

3.0. (a) Apply algorithm A1–7 to the E-R data model of Exercise 2.0 above
and provide a valid instance of at least two rows per table.
(b) Apply algorithm A7/8-3 to the above obtained relational scheme in
order to discover and add to it any other existing keys (hint: skip verifying
existing keys as we just correctly identify all of them).
264 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

Ø SubsidiaryOf should be acyclic.    (P6)

Ø All and only the subsidiaries of a bank should be
of bank type.    (P7)
Ø No payment for an invoice can be done before the
invoice is established or after more than 100 years
after corresponding establishing date.   (PD6)
Ø Whenever Type is ‘c’ (cash), To, From, and Card should
be null.   (PD7)
Ø Whenever Type is ‘C’ (card), From should be null,
whereas To and Card should not be null.   (PD8)
Ø Whenever Type is ‘w’ (bank wire), Card should be null
and To and From should not be null.   (PD9)
Ø Sum of all payments for a same invoice should not be
greater than the corresponding invoice total amount.   (PM4)
Ø  Bank subsidiaries should always be bank type partners.   (A5)
Ø  No card payment may be done with an expired card.     (C8)

FIGURE 3.25  Non-relational constraint set from Exercise 2.0.

Solution:
(a) The associated non-relational constraint set (that will have to be
enforced either through db triggers or by db applications) is pre-
sented in Fig.3.25.
Obviously, acyclicity (P6) is not a relational type constraint.
Each one of (PD6), (PD7), (PM4), and (C8) involve columns from
more than one table.
(P7), (PD8), (PD9), and (A5) apply to only some rows of the corre-
sponding tables.
Corresponding relational scheme, together with a test plausible and
valid instance are presented in Fig. 3.26.
(b)
(i) Table COUNTRIES
m = 2; K’ = K = {x, Country}; T’ = Ø; n = 0; l = k = 2;

(ii) Table CITIES

m = 3; K’ = K = {x, Country • ZipCode}; T’ = {City, Country, Zip-
Code}; n = 3; l = k = 2; kmax = C(3, 1) = 3;
The
 Quest for Data Independence, Minimal Plausibility, and Formalization 265

FIGURE 3.26  Relational Scheme from Exercise 2.0 and a Valid Test Instance.

i = 1 (C(3, 1) = 3)
1. City is not a key, as there may be cities having same names but dif-
ferent zip codes in a same country or even same zip codes too but from
different countries;
2. ZipCode is not a key, as there may be cities having same zip codes
but different names in a same country or even same names too but in dif-
ferent countries;
266 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

FIGURE 3.26  (Continued).

3. Country is not a key, as there may be several cities of a same country,

but having different zip codes and/or names;
i = 2 (C(3, 2) = 3)
1. Country • ZipCode is already known to be a key;
2. City • ZipCode is not a key, as there may be cities having same name
and zip codes, but from different countries;
The
 Quest for Data Independence, Minimal Plausibility, and Formalization 267

FIGURE 3.26  (Continued).

3. City • Country is not a key, as there may be cities from a same coun-
try having same name, but having different zip codes;
i = 3 (C(3, 3) = 1)
268 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

City • Country • ZipCode is a superkey;

To conclude with, K’ = K = {x, Country • ZipCode}; l = 2.

(iii) Table CARD_TYPES

m = 2; K’ = K = {x, CardType}; T’ = Ø; n = 0; l = k = 2;

(iv) Table PARTNERS

m = 6; K’ = K = {x, Name • Location • Address}; SK = {x}; T’ = {Name,
Location, Address, SubsidiaryOf, Bank?}; n = 5; l = 2; kmax = C(5, 2) = 10;
i = 1 (C(5, 1) = 5)
1. Name is not a key, as there may be several partners having same
names, but different types and/or being located in different cities
and/or addresses and/or being subsidiaries of different partners;
2. Location is not a key, as there may be several partners being located
in same cities, but having different names and/or types and/or being
located at different address in that city and/or being subsidiaries of
different partners;
3. Address is not a key, as there may be several partners having same
address within their cities, but different names and/or types and/or
being located in different cities and/or being subsidiaries of differ-
ent partners;
4. SubsidiaryOf is not a key, as there may be several partners being
subsidiaries of a same partner or not being subsidiaries, but having
different names and/or types and/or being located in different cities
and/or addresses;
5. Bank? is not a key, as there may be several partners having same
type, but different names and/or being located in different cities
and/or addresses and/or being subsidiaries of different partners;
i = 2 (C(5, 2) = 10)
1-3. Name • Location, Name • Address, and Location • Address are
subproducts of Name • Location • Address, so they cannot be keys;
4. Name • SubsidiaryOf is not a key, as there may be partners having
same names and being subsidiaries of a same partner or not being
subsidiaries, but being located in different cities and/or different
addresses and/or having different types;
The
 Quest for Data Independence, Minimal Plausibility, and Formalization 269

5. Name • Bank? is not a key, as there may be partners having same

names and type, but being located in different cities and/or differ-
ent addresses and/or being subsidiaries of other partners;
6. Location • SubsidiaryOf is not a key, as there may be partners
located in same cities and being subsidiaries of a same partner or
not being subsidiaries, but being located at different addresses and/
or having different names and/or types;
7. Location • Bank? is not a key, as there may be partners located in
same cities and having same type, but being located at different
addresses and/or having different names and/or being subsidiaries
of other partners;
8. Address • SubsidiaryOf is not a key, as there may be partners
located at same addresses within their cities and being subsidiaries
of a same partner or not being subsidiaries, but being located in
different cities and/or having different names and/or types;
9. Address • Bank? is not a key, as there may be partners located at
same addresses within their cities and having same type, but being
located in different cities and/or having different names and/or
being subsidiaries of other partners;
10. SubsidiaryOf • Bank? is not a key, as there may be partners having
same type and being subsidiaries of a same partner or not being
subsidiaries, but being located in different cities and/or at different
addresses and/or having different names;
i = 3 (C(5, 3) = 10)
1. Name • Location • Address is already known to be a key;
2. Name • Location • SubsidiaryOf is not a key, as there may be part-
ners having same names, located in same cities, and being subsid-
iaries of a same partner or not being subsidiaries, but being located
at different addresses and/or having different types;
3. Name • Location • Bank? is not a key, as there may be partners
having same names and types, and being located in same cities, but
being located at different addresses and/or being subsidiaries of
other partners;
4. Name • Address • SubsidiaryOf is not a key, as there may be part-
ners having same names, located at same addresses within their
cities, and being subsidiaries of a same partner or not being subsid-
270 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

iaries, but being located in different cities and/or having different

types;
5. Name • Address • Bank? is not a key, as there may be partners hav-
ing same names and types, and located at same addresses within
their cities, but being located in different cities and/or being sub-
sidiaries of other partners;
6. Name • SubsidiaryOf • Bank? is not a key, as there may be partners
having same names and types, and being subsidiaries of a same
partner or not being subsidiaries, but being located in different cit-
ies and/or at different addresses;
7. Location • Address • SubsidiaryOf is not a key, as there may be
partners located in a same city at a same address and being sub-
sidiaries of a same partner or not being subsidiaries, but having
different types and/or names;
8. Location • Address • Bank? is not a key, as there may be partners
located in a same city at a same addresses and having same types,
but being subsidiaries of different partners or not being subsidiar-
ies and/or having different names;
9. Location • SubsidiaryOf • Bank? is not a key, as there may be part-
ners having same types, located in same cities, and being subsidiar-
ies of a same partner or not being subsidiaries, but having different
names and/or being located at different addresses;
10. Address • SubsidiaryOf • Bank? is not a key, as there may be partners
having same types, located at same addresses within their cities, and
being subsidiaries of a same partner or not being subsidiaries, but
having different names and/or being located in different cities;
i = 4 (C(5, 4) = 5)
1-2. Name • Location • Address • SubsidiaryOf and Name • Location •
Address • Bank? are superkeys;
3. Name • Location • SubsidiaryOf • Bank? is not a key, as there may
be several partners having same names and types, located in a same
city, and being subsidiaries of a same partner or not being subsid-
iaries, but being located at different addresses within the city;
4. Name • Address • SubsidiaryOf • Bank? is not a key, as there may
be several partners having same names and types, located at same
addresses within their cities, and being subsidiaries of a same part-
ner or not being subsidiaries, but being located in different cities;
The
 Quest for Data Independence, Minimal Plausibility, and Formalization 271

5. Location • Address • SubsidiaryOf • Bank? is a key, as there may

not be several partners located in a same city, at a same address,
having same type, and being subsidiaries of a same partner or not
being subsidiaries, but having different names;
K’ = {x, Name • Location • Address, Location • Address • SubsidiaryOf
• Bank?}; l = 3;
i = 5 (C(5, 5) = 1)
Name • Location • Address • SubsidiaryOf • Bank? is a superkey;
To conclude with, K’ = {x, Name • Location • Address, Location •
Address • SubsidiaryOf • Bank?}; l = 3.

(v) Table ACCOUNTS

m = 5; K’ = K = {x, BankSubsidiary • No.}; T’ = {BankSubsidiary, No.,
Holder, Amount}; n = 4; l = k = 2;
Amount is not a key (as there may be several accounts having same
amount, but managed by different bank subsidiaries or even by a same one,
but having different numbers) and, moreover, it is not prime, as it cannot
even help uniquely distinguishing accounts together with other columns; it
follows that T’ = {BankSubsidiary, No., Holder}; n = 3; kmax = C(3, 1) = 3;
i = 1 (C(3, 1) = 3)
1. BankSubsidiary is not a key, as any bank subsidiary manages sev-
eral accounts;
2. No. is not a key, as there may be several accounts having same
numbers, but managed by different bank subsidiaries;
3. Holder is not a key, as anybody may have several accounts man-
aged by different bank subsidiaries, and even by a same one, but
having different numbers;
i = 2 (C(3, 2) = 3)
1. BankSubsidiary • No. is already known to be a key;
2. BankSubsidiary • Holder is not a key, as anybody may hold several
accounts at a same bank subsidiary, but with different numbers;
3. No. • Holder is not a key, as anybody may hold several accounts
having same numbers, but at different bank subsidiaries;
i = 3 (C(3, 3) = 1)
BankSubsidiary • No. • Holder is a superkey;
To conclude with, K’ = K = {x, BankSubsidiary • No.}; l = 2.
272 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

(vi) Table CARDS

m = 8; K’ = K = {x, Type • No. • Year}; T’ = {Type, No., Month, Year,
SecurityCode, D/C, Account}; n = 7; l = k = 2;
Month is not a key, as there are several cards having same expiration
month, but of different types and/or having different numbers and/or expi-
ration years; moreover, Month is not prime, as cards have at least one
year validity, and hence it cannot even help uniquely distinguishing cards
together with other columns.
SecurityCode is not a key (as there may be several cards having same
security code, but managed by different bank subsidiaries or even by a
same one, but having different numbers and/or types, etc.) and, more-
over, it is not prime, as it cannot even help uniquely distinguishing cards
together with other columns.
D/C is not a key (as there may be several debit or credit cards, but
managed by different bank subsidiaries or even by a same one, but hav-
ing different numbers and/or types, etc.) and, moreover, it is not prime,
as it cannot even help uniquely distinguishing cards together with other
columns.
It follows that T’ = {Type, No., Year, Account}; n = 4; kmax = C(4, 2) = 6;
i = 1 (C(4, 1) = 4)
1. Type is not a key, as there are several cards of any type;
2. No. is not a key, as there are several cards having same numbers,
but of different types;
3. Year is not a key, as there are several cards having same expiration
year, but of different types and/or having different numbers;
4. Account is not a key, as there may be several cards having same
bank account, but having different expiration years;
i = 2 (C(4, 2) = 6)
1-3. Type • No., Type • Year, and No. • Year, are subproducts of the key
Type • No. • Year;
4. Type • Account is not a key, as there are several cards having same
type and associated to a same bank account, but having different
expiration years;
5. No. • Account is not a key, as there are several cards having same
number and associated to a same bank account, but having differ-
ent expiration years;
The
 Quest for Data Independence, Minimal Plausibility, and Formalization 273

6. Year • Account is a key, as, according to the simplifying assumption

that there may be no several cards associated to a same account and
simultaneously valid, there may not be several cards having same
expiration year and associated to a same bank account;
K’ = {x, Type • No. • Year, Year • Account}; l = 3;
i = 3 (C(4, 3) = 4)
1. Type • No. • Year is already known as a key;
2-3. Type • Year • Account and No. • Year • Account are superkeys;
4. Type • No. • Account is not a key, as there may be cards of a same
type, having a same number, and being associated to a same bank
account, but having different expiration years;
i = 4 (C(4, 4) = 1)
Type • No. • Year • Account is a superkey;
To conclude with, K’ = {x, Type • No. • Year, Year • Account}; l = 3.

(vii) Table INVOICES

m = 6; K’ = K = {x, No. • Date • Supplier}; T’ = {No., Date, Supplier,
Customer, *TotAmount}; n = 5; l = k = 2;
*TotAmount is not a key (as there may be invoices having same total
amount) and non prime either, as it cannot even help uniquely distinguish-
ing cards together with other columns.
It follows that T’ = {No., Date, Supplier, Customer}; n = 4; kmax = C(4,
2) = 6;
i = 1 (C(4, 1) = 4)
1-3. No., Date, and Supplier, are not keys as they are members of the
key No. • Date • Supplier;
4. Customer is not a key, as there may be several invoices for a same
customer (established even by a same supplier, and even at a same
date, but with different numbers);
i = 2 (C(4, 2) = 6)
1-3. No. • Date, No. • Supplier, and Date • Supplier are subproducts of
the key No. • Date • Supplier;
4. No. • Customer is not a key, as there may be several invoices for a
same customer having a same number (established even by a same
supplier, but at different dates);
274 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

5. Date • Customer is not a key, as there may be several invoices for

a same customer having a same date (established even by a same
supplier, but with different numbers);
6. Supplier • Customer is not a key, as there may be several invoices
for a same customer established by a same supplier, but with dif-
ferent numbers and/or dates;
i = 3 (C(4, 3) = 4)
1. No. • Date • Supplier is already known to be a key;
2. No. • Date • Customer is not a key, as there may be several invoices
for a same customer having a same number and date, but estab-
lished by different suppliers;
3. No. • Supplier • Customer is not a key, as there may be several
invoices for a same customer established by a same supplier, with
a same number, but at different dates;
4. Date • Supplier • Customer is not a key, as there may be several
invoices for a same customer established by a same supplier, at a
same date, but with different numbers;
i = 4 (C(4, 4) = 1)
No. • Date • Supplier • Customer is a superkey;
To conclude with, K’ = K = {x, No. • Date • Supplier}; l = 2.

(viii) Table PAYM_DOCS

m = 9; K’ = K = {x, No. • Date • Debtor • Type}; T’ = {No., Date,
Debtor, Type, From, To, Card, *TotAmount}; n = 8; l = k = 2;
*TotAmount is not a key (as there may even be payment documents
of a same debtor, having same number and amount, but different dates)
and non prime either, as it cannot even help uniquely distinguishing cards
together with other columns.
It follows that T’ = {No., Date, Debtor, Type, From, To, Card}; n = 7;
kmax = C(7, 3) = 35;
i = 1 (C(7, 1) = 7)
1-4. No., Date, Debtor, and Type are not keys, as they are members of
the key No. • Date • Debtor • Type;
5. From is not a key, as there may be several payments from a same
bank account;
The
 Quest for Data Independence, Minimal Plausibility, and Formalization 275

6. To is not a key, as there may be several payments to a same bank

account;
7. Card is not a key, as there may be several payments done with a
same card;
i = 2 (C(7, 2) = 21)
1-6. No. • Date, No. • Debtor, No. • Type, Date • Debtor, Date • Type,
and Debtor • Type are not keys, as they are subproducts of the key
No. • Date • Debtor • Type;
7. No. • From is not a key, as there may be several payments from a
same bank account with a same number, but having different dates
and/or debtors, etc.;
8. No. • To is not a key, as there may be several payments to a same
bank account with a same number, but having different dates and/
or debtors, etc.;
9. No. • Card is not a key, as there may be several payments done with
a same card and having a same number, but different dates and/or
debtors, etc.;
10. Date • From is not a key, as there may be several payments from
a same bank account at a same date, but having different numbers
and/or types, etc.;
11. Date • To is not a key, as there may be several payments to a same
bank account at a same date, but having different numbers and/or
types, etc.;
12. Date • Card is not a key, as there may be several payments done
with a same card at a same date, but having different numbers and/
or destination bank accounts, etc.;
13. Debtor • From is not a key, as there may be several payments from
a same bank account done by a same debtor, but having different
numbers and/or types, etc.;
14. Debtor • To is not a key, as there may be several payments to a
same bank account done by a same debtor, but having different
numbers and/or types, etc.;
15. Debtor • Card is not a key, as there may be several payments done
with a same card by a same debtor, but having different numbers
and/or destination bank accounts, etc.;
276 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

16. Type • From is not a key, as there may be several payments of

a same type and from a same bank account, but having different
numbers and/or dates, etc.;
17. Type • To is not a key, as there may be several payments of a same
type and to a same bank account, but having different numbers and/
or dates, etc.;
18. Type • Card is not a key, as there may be several payments done
with a same card, but having different numbers and/or destination
bank accounts, etc.;
19. From • To is not a key, as there may be several payments from a
same bank account to a same bank account, but having different
numbers and/or dates, etc.;
20. From • Card is not a key, as there may be several payments done
with a same card, but having different numbers and/or destination
bank accounts, etc.;
21. To • Card is not a key, as there may be several payments done with
a same card to a same bank account, but having different numbers
and/or dates, etc.;
i = 3 (C(7, 3) = 35)
1-4. No. • Date • Debtor, No. • Date • Type, No.• Debtor • Type, and
Date • Debtor • Type are not keys, as they are subproducts of the
key No. • Date • Debtor • Type;
5. No. • Date • From is a key, as there may not be several payments
from a same bank account with a same number and date;
K’ = {x, No. • Date • Debtor • Type, No. • Date • From}; l = 3;
6. No. • Date • To is a key, as there may not be several payments to a
same bank account with a same number and date;
K’ = {x, No. • Date • Debtor • Type, No. • Date • From, No. • Date •
To}; l = 4;
7. No. • Date • Card is a key, as there may not be several payments
done with a same card at a same date and with a same number;
K’ = {x, No. • Date • Debtor • Type, No. • Date • From, No. • Date • To,
No. • Date • Card}; l = 5;
8. No. • Debtor • From is not a key, as there may be several pay-
ments done by a same debtor from a same bank account with a
same number, but having different dates and/or destination bank
accounts, etc.;
The
 Quest for Data Independence, Minimal Plausibility, and Formalization 277

9. No. • Debtor • To is not a key, as there may be several payments

done by a same debtor to a same bank account with a same number,
but having different dates and/or source bank accounts, etc.;
10. No. • Debtor • Card is not a key, as there may be several payments
done by a same debtor with a same card with a same number, but
at different dates;
11. No. • Type • From is not a key, as there may be several payments
having a same type, number, and from a same bank account, but
different dates and/or destination bank accounts, etc.;
12. No. • Type • To is not a key, as there may be several payments hav-
ing a same type, number, and to a same bank account, but different
dates and/or source bank accounts, etc.;
13. No. • Type • Card is not a key, as there may be several payments
having a same type and number, done with a same card, but at dif-
ferent dates;
14. No. • From • To is not a key, as there may be several payments
having a same number, from a same bank account to a same bank
account, but at different dates, etc.;
15. No. • From • Card is not a key, as there may be several payments
having a same number, from a same bank account, done with a
same card, but at different dates;
16. No. • To • Card is not a key, as there may be several payments hav-
ing a same number, to a same bank account, done with a same card,
but at different dates;
17. Date • Debtor • From is not a key, as there may be several pay-
ments done by a same debtor from a same bank account at a
same date, but having different numbers and/or destination bank
accounts, etc.;
18. Date • Debtor • To is not a key, as there may be several payments
done by a same debtor to a same bank account at a same date,
but having different numbers and/or source or destination bank
accounts, etc.;
19. Date • Debtor • Card is not a key, as there may be several payments
having a same date, done with a same card by a same debtor, but
having different numbers and/or destination bank accounts;
278 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

20. Date • Type • From is not a key, as there may be several payments
of a same type, from a same bank account, and at a same date, but
having different numbers and/or destination bank accounts, etc.;
21. Date • Type • To is not a key, as there may be several payments of a
same type, to a same bank account, and at a same date, but having
different numbers and/or source bank accounts, etc.;
22. Date • Type • Card is not a key, as there may be several payments
having a same date, done with a same card, but having different
numbers and/or destination bank accounts;
23. Date • From • To is not a key, as there may be several payments
done at a same date, from a same bank account to a same bank
account, but with different numbers, etc.;
24. Date • From • Card is not a key, as there may be several payments
done at a same date, from a same bank account, with a same card,
but with different numbers and/or to different destination bank
accounts, etc.;
25. Date • To • Card is not a key, as there may be several payments
done at a same date, to a same bank account, with a same card, but
with different numbers, etc.;
26. Debtor • Type • From is not a key, as there may be several pay-
ments of a same type, from a same bank account, done by a same
debtor, but having different numbers and/or dates and/or destina-
tion bank accounts;
27. Debtor • Type • To is not a key, as there may be several payments
of a same type, to a same bank account, done by a same debtor,
but having different numbers and/or dates and/or source bank
accounts;
28. Debtor • Type • Card is not a key, as there may be several payments
done by a same debtor with a same card, but having different num-
bers and/or dates and/or destination bank accounts;
29. Debtor • From • To is not a key, as there may be several payments
from a same bank account to a same bank account, done by a same
debtor, but having different numbers and/or dates;
30. Debtor • From • Card is not a key, as there may be several pay-
ments done by a same debtor with a same card, but having different
numbers and/or dates and/or destination bank accounts;
The
 Quest for Data Independence, Minimal Plausibility, and Formalization 279

31. Debtor • To • Card is not a key, as there may be several payments

done by a same debtor with a same card to a same bank account,
but having different numbers and/or dates;
32. Type • From • To is not a key, as there may be several payments of
a same type, from a same bank account to a same bank account, but
with different numbers and/or dates, etc.;
33. Type • From • Card is not a key, as there may be several payments
done from a same bank account, with a same card, but with dif-
ferent numbers and/or dates and/or to different destination bank
accounts;
34. Type • To • Card is not a key, as there may be several payments
done to a same bank account, with a same card, but with different
numbers and/or dates;
35. From • To • Card is not a key, as there may be several payments
done to a same bank account, with a same card, but with different
numbers and/or dates;
i = 4 (C(7, 4) = 35)
1. No. • Date • Debtor • Type is already known to be a key;
2-10. No. • Date • Debtor • From, No. • Date • Debtor • To, No. • Date
• Debtor • Card, No. • Date • Type • From, No. • Date • Type • To,
No. • Date • Type • Card, No. • Date • From • To, No. • Date • From
• Card, and No. • Date • To • Card are superkeys;
11. No. • Debtor • Type • From, is not a key, as there may be several
payments of a same type done by a same debtor from a same bank
account with a same number, but having different dates and/or des-
tination bank accounts;
12. No. • Debtor • Type • To, is not a key, as there may be several pay-
ments of a same type done by a same debtor to a same bank account
with a same number, but having different dates and/or source bank
accounts;
13. No. • Debtor • Type • Card, is not a key, as there may be several
payments done by a same debtor with a same card and having a
same number, but different dates and/or destination bank accounts;
14. No. • Debtor • From • To, is not a key, as there may be several pay-
ments done by a same debtor from a same bank account to a same
bank account with a same number, but having different dates;
280 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

15. No. • Debtor • From • Card, is not a key, as there may be several
payments done by a same debtor with a same card and having a
same number, but different dates and/or destination bank accounts;
16. No. • Debtor • To • Card, is not a key, as there may be several
payments done by a same debtor with a same card to a same bank
account and having a same number, but different dates;
17. No. • Type • From • To, is not a key, as there may be several pay-
ments of a same type from a same bank account to a same bank
account with a same number, but having different dates;
18. No. • Type • From • Card, is not a key, as there may be several
payments done with a same card and having a same number, but
different dates and/or destination bank accounts;
19. No. • Type • To • Card, is not a key, as there may be several pay-
ments done with a same card to a same bank account and having a
same number, but different dates;
20. No. • From • To • Card, is not a key, as there may be several pay-
ments done with a same card to a same bank account and having a
same number, but different dates;
21. Date • Debtor • Type • From, is not a key, as there may be several
payments of a same type done by a same debtor from a same bank
account at a same date, but having different numbers and/or desti-
nation bank accounts;
22. Date • Debtor • Type • To, is not a key, as there may be several
payments of a same type done by a same debtor to a same bank
account at a same date, but having different numbers and/or source
bank accounts;
23. Date • Debtor • Type • Card, is not a key, as there may be several
payments done by a same debtor with a same card at a same date,
but having different numbers and/or destination bank accounts;
24. Date • Debtor • From • To, is not a key, as there may be several
payments done by a same debtor from a same bank account to a
same bank account at a same date, but having different numbers;
25. Date • Debtor • From • Card, is not a key, as there may be several
payments done by a same debtor with a same card at a same date,
but having different numbers and/or destination bank accounts;
The
 Quest for Data Independence, Minimal Plausibility, and Formalization 281

26. Date • Debtor • To • Card, is not a key, as there may be several

payments done by a same debtor with a same card to a same bank
account at a same date, but having different numbers;
27. Date • Type • From • To, is not a key, as there may be several pay-
ments of a same type from a same bank account to a same bank
account at a same date, but having different numbers;
28. Date • Type • From • Card, is not a key, as there may be several
payments done with a same card at a same date, but having differ-
ent numbers and/or destination bank accounts;
29. Date • Type • To • Card, is not a key, as there may be several pay-
ments done with a same card at a same date to a same bank account,
but having different numbers;
30. Date • From • To • Card, is not a key, as there may be several
payments done with a same card at a same date to a same bank
account, but having different numbers
31. Debtor • Type • From • To, is not a key, as there may be several pay-
ments of a same type from a same bank account to a same bank account
done by a same debtor, but having different numbers and/or dates;
32. Debtor • Type • From • Card, is not a key, as there may be several
payments done by a same debtor with a same card, but having dif-
ferent numbers and/or dates and/or destination bank accounts;
33. Debtor • Type • To • Card, is not a key, as there may be several
payments done by a same debtor with a same card to a same bank
account, but having different numbers and/or dates;
34. Debtor • From • To • Card, is not a key, as there may be several
payments done by a same debtor with a same card to a same bank
account, but having different numbers and/or dates;
35. Type • From • To • Card, is not a key, as there may be several pay-
ments done with a same card to a same bank account, but having
different numbers and/or dates;
i = 5 (C(7, 5) = 21)
1-10. No. • Date • Debtor • Type • From, No. • Date • Debtor • Type •
To, No. • Date • Debtor • Type • Card, No. • Date • Debtor • From
• To, No. • Date • Debtor • From • Card, No. • Date • Debtor • To
• Card, No. • Date • Type • From • To, No. • Date • Type • From •
282 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

Card, No. • Date • Type • To • Card, and No. • Date • From • To •

Card are superkeys;
11. No. • Debtor • Type • From • To is not a key, as there may be several
payments of a same type done by a same debtor from a same bank
account to a same bank account and having a same number, but
different dates;
12. No. • Debtor • Type • From • Card is not a key, as there may be
several payments by a same debtor with a same card and having a
same number, but different dates and/or destination bank accounts;
13. No. • Debtor • Type • To • Card is not a key, as there may be sev-
eral payments by a same debtor with a same card to a same bank
account and having a same number, but different dates;
14. No. • Debtor • From • To • Card is not a key, as there may be sev-
eral payments by a same debtor with a same card to a same bank
account and having a same number, but different dates;
15. No. • Type • From • To • Card is not a key, as there may be several
payments with a same card to a same bank account and having a
same number, but different dates;
16. Date • Debtor • Type • From • To is not a key, as there may be sev-
eral payments of a same type done by a same debtor from a same
bank account to a same bank account and having a same date, but
different numbers;
17. Date • Debtor • Type • From • Card is not a key, as there may be
several payments by a same debtor with a same card and having a
same date, but different numbers and/or destination bank accounts;
18. Date • Debtor • Type • To • Card is not a key, as there may be sev-
eral payments by a same debtor with a same card to a same bank
account and having a same date, but different numbers;
19. Date • Debtor • From • To • Card is not a key, as there may be sev-
eral payments by a same debtor with a same card to a same bank
account and having a same date, but different numbers;
20. Date • Type • From • To • Card is not a key, as there may be several
payments with a same card to a same bank account and having a
same date, but different numbers;
The
 Quest for Data Independence, Minimal Plausibility, and Formalization 283

21. Debtor • Type • From • To • Card is not a key, as there may be sev-
eral payments done by a same debtor with a same card to a same
bank account, but having different numbers and/or dates;
i = 6 (C(7, 6) = 7)
1-5. No. • Date • Debtor • Type • From • To, No. • Date • Debtor • Type
• From • Card, No. • Date • Debtor • Type • To • Card, No. • Date
• Debtor • From • To • Card, and No. • Date • Type • From • To •
Card are superkeys;
6. No. • Debtor • Type • From • To • Card is not a key, as there may
be several payments done by a same debtor with a same card to a
same bank account and having a same number, but different dates;
7. Date • Debtor • Type • From • To • Card is not a key, as there may
be several payments done by a same debtor with a same card to a
same bank account and having a same date, but different numbers;
i = 7 (C(7, 7) = 1)
No. • Date • Debtor • Type • From • To • Card is a superkey;
To conclude with, K’ = {x, No. • Date • Debtor • Type, No. • Date •
From, No. • Date • To, No. • Date • Card}; l = 5.

(ix) Table PAYMENTS

m = 4; K’ = K = {x, Invoice • PaymDoc}; T’ = {Invoice, PaymDoc,
Amount}; n = 3; l = k = 2;
Amount is not a key (as there may even be payment documents includ-
ing partial or total payments for invoices having same amount) and non
prime either, as it cannot even help uniquely distinguishing payments
together with other columns.
It follows that T’ = {Invoice, PaymDoc}; n = 2; kmax = C(2, 1) = 2;
i = 1 (C(2, 1) = 2)
1-2. Invoice and PaymDoc are not keys as they are members of the key
Invoice • PaymDoc;
i = 2 (C(2, 2) = 1)
Invoice • PaymDoc is already known to be a key;
To conclude with, K’ = K = {x, Invoice • PaymDoc}; l = 2.
In conclusion, the following 5 newly discovered keys should be added
to the tables from Figure 3.26 above:
284 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

FIGURE 3.27  Rdb for Exercise 3.1.

- Location • Address • SubsidiaryOf • Bank? to PARTNERS;

- Year • Account to CARDS;
- No. • Date • From, No. • Date • To, and No. • Date • Card to
PAYM_DOCS.
3.1. (An actual example of basic test for hiring SQL developers) Con-
sider a RDBMS that does not provide autonumbering and an rdb having
the following two tables and associated instances (see Fig. 3.27).
The
 Quest for Data Independence, Minimal Plausibility, and Formalization 285

(a) Design, develop, and test ANSI-92 standard SQL statements for
creating and populating these two tables.
(b) Consider that triggers exist for enforcing irreflexivity and asymme-
try of the NEIGHBORS binary relation (i.e., rejecting any attempts
to store in columns Country and Neighbors pairs of type <x, x>,
as well as <y, x>, whenever <x, y> already is stored) and design,
develop, and test SQL statements for computing:
1. The number of neighbors of Romania. What is the corresponding
result?
2. The number of neighbors of Moldova. What is the corresponding
result?
3. The set of countries without neighbors, in ascending order of their
names. What is the corresponding result?
4. The set of countries having at least k neighbors (k natural), in
descending order of the number of their neighbors and then ascend-
ing on their names. What is the corresponding result for k = 3?
5. The set of countries that do not appear in column Country of table
NEIGHBORS, in ascending order of their names, without using
subqueries. What is the corresponding result?
6. The set of pairs <country name, neighbors number>, in descending
order of neighbors number and then ascending on country name,
also including countries that do not have neighbors. What is the
corresponding result?
7. Adding the fact that Ukraine is also neighbor to Russia, without
knowing anything on the two table instances, except for the fact
that Ukraine exists in COUNTRIES, while Russia does not. What
lines will be added according to the current test instance?
8. Changing to uppercase the names of the countries having the prop-
erty that they are neighbors to at least one neighbor of one of their
neighbors (i.e., country x will have its name modified if it is neigh-
bor both to countries y and z, where z is a neighbor of y). What are
the countries whose names will be modified?
9. Discarding all neighborhood data for all countries whose names
start with the letter ‘G’. What lines will disappear?
286 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

10. Adding to table COUNTRIES a column labeled Code for storing

the international country codes (those also used on vehicle plates;
e.g. USA, RO, MD, UK, etc.), having at most 8 characters, and
whose values are mandatory and distinct between them (i.e., stor-
ing two countries with the same code must not be allowed); add
then corresponding code values, in uppercase, by extracting only
the first 2 letters of each country name. How will the extended
table look like, sorted in ascending order by country names, taking
into account all modifications that were done to it by 1 to 9 above?

Solution:
(a)
CREATE TABLE COUNTRIES (
   X INT(3) PRIMARY,
   COUNTRY VARCHAR(128 BYTE) NOT NULL UNIQUE
);
Insert into COUNTRIES (X, COUNTRY) values
(1, ’Romania’);
Insert into COUNTRIES (X, COUNTRY) values
(2, ’Moldova’);
Insert into COUNTRIES (X, COUNTRY) values
(3, ’Serbia’);
Insert into COUNTRIES (X, COUNTRY) values
(4, ’Bulgaria’);
Insert into COUNTRIES (X, COUNTRY) values
(5, ’Hungary’);
Insert into COUNTRIES (X, COUNTRY) values
(6, ’Ukraine’);
Insert into COUNTRIES (X, COUNTRY) values
(7, ’Greece’);
Insert into COUNTRIES (X, COUNTRY) values
(8, ’Malta’);
CREATE TABLE NEIGHBORS (
   X INT(4) PRIMARY,
   COUNTRY INT(3) NOT NULL,
   NEIGHBOR INT(3) NOT NULL,
The
 Quest for Data Independence, Minimal Plausibility, and Formalization 287

  CONSTRAINT NEIGHBORS_Key UNIQUE (COUNTRY, NEIGH-

BOR),
 CONSTRAINT NEIGHBORS_COUNTRY_FK FOREIGN KEY
(COUNTRY) REFERENCES COUNTRIES,
 CONSTRAINT NEIGHBORS_NEIGHBOR_FK FOREIGN KEY
(NEIGHBOR) REFERENCES COUNTRIES
);
Insert into NEIGHBORS (X, COUNTRY, NEIGHBOR)
values (1,1,2);
Insert into NEIGHBORS (X, COUNTRY, NEIGHBOR)
values (2,1,3);
Insert into NEIGHBORS (X, COUNTRY, NEIGHBOR)
values (3,1,4);
Insert into NEIGHBORS (X, COUNTRY, NEIGHBOR)
values (4,1,5);
Insert into NEIGHBORS (X, COUNTRY, NEIGHBOR)
values (5,1,6);
Insert into NEIGHBORS (X, COUNTRY, NEIGHBOR)
values (6,2,6);
Insert into NEIGHBORS (X, COUNTRY, NEIGHBOR)
values (7,3,4);
Insert into NEIGHBORS (X, COUNTRY, NEIGHBOR)
values (8,3,5);
Insert into NEIGHBORS (X, COUNTRY, NEIGHBOR)
values (9,4,7);
(b)
1. SELECT COUNT(*) AS NEIGHBORS# FROM NEIGHBORS
WHERE COUNTRY = 1;
    Result: 5
2. SELECT COUNT(*) AS N# FROM NEIGHBORS
    WHERE COUNTRY = 2 OR NEIGHBOR = 2;
    Result: 2
3. SELECT COUNTRY FROM COUNTRIES WHERE X NOT IN
    (SELECT DISTINCT COUNTRY FROM NEIGHBORS
    UNION
    SELECT DISTINCT NEIGHBOR FROM NEIGHBORS)
288 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

    ORDER BY COUNTRY;
    Result: Malta
4. SELECT COUNTRIES.COUNTRY, NEIGHBORS#
    FROM COUNTRIES INNER JOIN
    (SELECT COUNTRY, SUM (N#) AS NEIGHBORS# FROM
        (SELECT COUNTRY, COUNT(*) AS N#
        FROM NEIGHBORS GROUP BY COUNTRY
        UNION ALL
        SELECT NEIGHBOR, COUNT(*) AS N#
        FROM NEIGHBORS GROUP BY NEIGHBOR)
    GROUP BY COUNTRY
    HAVING SUM (N#) > :K — 1) S
   ON COUNTRIES.X = S.COUNTRY
   ORDER BY NEIGHBORS# DESC, COUNTRIES.COUNTRY;
Result:
COUNTRY NEIGHBORS#
Romania 5
Bulgaria 3
Serbia 3
5.  SELECT COUNTRIES.COUNTRY
FROM COUNTRIES LEFT JOIN NEIGHBORS
ON COUNTRIES.X = NEIGHBORS.COUNTRY
WHERE NEIGHBORS.COUNTRY IS NULL
ORDER BY COUNTRIES.COUNTRY;

Result: Greece
Hungary
Malta
Ukraine
6. SELECT COUNTRIES.COUNTRY, SUM (N#) AS NEIGHBORS#
  FROM COUNTRIES INNER JOIN
  (SELECT COUNTRY, COUNT(*) AS N# FROM NEIGHBORS
GROUP BY COUNTRY
  UNION ALL
  SELECT NEIGHBOR, COUNT(*) AS N# FROM NEIGHBORS
  GROUP BY NEIGHBOR) S
The
 Quest for Data Independence, Minimal Plausibility, and Formalization 289

  ON COUNTRIES.X = S.COUNTRY

GROUP BY COUNTRIES.COUNTRY
UNION
SELECT COUNTRY, 0 FROM COUNTRIES WHERE X NOT IN
  (SELECT DISTINCT COUNTRY FROM NEIGHBORS
  UNION
  SELECT DISTINCT NEIGHBOR FROM NEIGHBORS)
ORDER BY NEIGHBORS# DESC, COUNTRY;
Result:
COUNTRY NEIGHBORS#
Romania 5
Bulgaria 3
Serbia 3
Hungary 2
Moldova 2
Ukraine 2
Greece 1
Malta 0
7. INSERT INTO COUNTRIES
SELECT max(X) + 1, ‘Russia’ FROM COUNTRIES;
Result: line < 9, Russia> is appended to table COUNTRIES

CREATE TABLE T AS
SELECT X FROM COUNTRIES
WHERE COUNTRY IN (‘Ukraine’, ’Russia’);
INSERT INTO NEIGHBORS
SELECT max(NEIGHBORS.X) + 1, min(T.X), max(T.X)
FROM NEIGHBORS, T;
Result: line < 10, 6, 9 > is appended to table NEIGHBORS

DROP TABLE T;

Note that, for example, the equivalent, much more elegant statement:

INSERT INTO NEIGHBORS(X, Country, Neighbor)

SELECT max(X) + 1, (SELECT X FROM COUNTRIES WHERE
290 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

Country = ‘Ukraine’), (SELECT X

FROM COUNTRIES WHERE Country = ‘Russia’)
FROM NEIGHBORS;

is accepted by MS Access, but not by Oracle (which is not able to notice

that X is a primary key, so at most only one value is returned by each of
the subqueries)!
8. UPDATE COUNTRIES SET COUNTRY = UPPER(COUNTRY)
  WHERE X IN
  (SELECT DISTINCT A.COUNTRY FROM
(SELECT COUNTRY, NEIGHBOR FROM NEIGHBORS
UNION
SELECT NEIGHBOR, COUNTRY FROM NEIGHBORS)
  A INNER JOIN (SELECT DISTINCT A.COUNTRY,
B.NEIGHBOR FROM
  (SELECT COUNTRY, NEIGHBOR FROM NEIGHBORS
 UNION
  SELECT NEIGHBOR, COUNTRY FROM NEIGHBORS)
  A INNER JOIN
  (SELECT COUNTRY, NEIGHBOR FROM NEIGHBORS
 UNION
  SELECT NEIGHBOR, COUNTRY FROM NEIGHBORS)
  B ON A.NEIGHBOR = B.COUNTRY AND
A.COUNTRY <> B.NEIGHBOR) B
  ON A.NEIGHBOR = B.COUNTRY AND
A.COUNTRY = B.NEIGHBOR);
Result: Romania, Moldova, Serbia, Bulgaria, Hungary, and
Ukraine will be changed to ROMANIA, MOLDOVA, SERBIA, BUL-
GARIA, HUNGARY, and UKRAINE, respectively
9. DELETE FROM NEIGHBORS WHERE COUNTRY IN
(SELECT X FROM COUNTRIES WHERE COUNTRY LIKE ‘G%’)
OR NEIGHBOR IN
(SELECT X FROM COUNTRIES WHERE COUNTRY LIKE ‘G%’);
Result: line < 9, 4, 7 > will be deleted from NEIGHBORS
10. CREATE TABLE C AS SELECT * FROM COUNTRIES;
CREATE TABLE N AS SELECT * FROM NEIGHBORS;
The
 Quest for Data Independence, Minimal Plausibility, and Formalization 291

DELETE FROM NEIGHBORS;

DELETE FROM COUNTRIES;
ALTER TABLE COUNTRIES
ADD (CODE VARCHAR(8) NOT NULL UNIQUE);
INSERT INTO COUNTRIES
SELECT X, COUNTRY, UPPER(SUBSTR(COUNTRY,1,2))
FROM C;
INSERT INTO NEIGHBORS SELECT * FROM N;
DROP TABLE T;
DROP TABLE N;
Result:
X COUNTRY CODE
4 BULGARIA BU
7 Greece GR
5 HUNGARY HU
8 Malta MA
2 MOLDOVA MO
1 ROMANIA RO
9 Russia RU
3 SERBIA SE
6 UKRAINE UK

3.2. Prove that the FD math relation is a partial preorder.

Proof:
Let R, A, B, C be any sets, f : R → A, g : R → B, and h : R → C be
any attributes of R with g functionally dependent on f and h functionally
dependent on g, that is f → g and g → h hold in any R’s instance, so, by
definition, ker(f) # ker(g) and ker(g) # ker(h);
(partiality) trivially, by the FD definition, not even two functions
defined on a same set need to be functionally dependent;
(reflexivity) trivially, from the set inclusion reflexivity, ;f, ker(f) #
ker(f)  f → f;
(transitivity) obviously, from the set inclusion transitivity, (;f, g, h)
(ker(f) # ker(g) ` ker(g) # ker(h)  ker(f) # ker(h))  f → h.
Q.E.D.
292 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

3.3. A relation satisfies X → Y  it satisfies X → A, ;A[Y.

Proof:
Let R, B = C1 × … × Cn (n > 1, natural), D be any sets, with R and Ci
atomic, while D might be a Cartesian product too, Y : R → B, X : R → D,
Ai : R → Ci, for any 1 ≤ i ≤ n, and Y = A1 • … • An be any attributes and
attribute products of R;
() By definition, X → Y  ker(X) # ker(Y); trivially, as ker(f•g) =
ker(f) ∩ ker(g), for any function product f•g, and as set intersection is associa-
tive, ker(Y) = ker(A1) ∩ … ∩ ker(An); obviously, as ker(A1) ∩ … ∩ ker(An) #
ker(Ai), for any 1 ≤ i ≤ n, it follows that ker(X) # ker(Ai), so X → A, ;A [ Y.
() By definition, X → A, ;A[Y  ker(X) # ker(Ai), for any 1 ≤ i ≤
n, which obviously means that ker(X) # ker(A1) ∩ … ∩ ker(An); as ker(Y)
= ker(A1) ∩ … ∩ ker(An), it follows that ker(X) # ker(Y), so X → Y.
Q.E.D.

3.4. Consider a table r(U) and X1, X2, X , U such that X1 ø X2 = U and
X = X1 ∩ X2; r(U) decomposition over X1, X2 is lossless  r(U) satisfies
MVD X →→ X1 (or, equivalently, X →→ X2).
Proof:
() Let r = πX1(r) |><| πX2(r) and t1, t2 [ r such that t1[X] = t2[X]; it has
to be shown that 't[r such that t[X1] = t1[X1] and t[X2] = t2[X2], which
would confirm the fact that r satisfies X →→ X1. Let t1′ = t[X1] and t2′ =
t[X2]; it follows that t1′[πX1(r), and t2′[πX2(r). By join definition, t is a
tuple of πX1(r) |><| πX2(r) that comes out of t1′ and t2′. Finally, as r = πX1(r)
|><| πX2(r), it follows that t[r, and that r satisfies X →→ X1.
() Let r be an instance satisfying X →→ X1 and t[πX1(r)|><|πX2(r);
for proving this implication, it has to be shown that t[r (because r #
πX1(r) |><| πX2(r) always). From the join and projection operators defini-
tions, 't1,t2[r such that t[X1] = t1[X1] and t[X2] = t2[X2]; from the MVD def-
inition, as X = X1 ∩ X2, so t1[X] = t2[X], 't’[r such that t’[X1] = t1[X1] and
t’[X2] = t2[X2]. This obviously means that t = t’, which proves that t [ r.
Q.E.D.

3.5.* (DKNF characterization theorem)

A table is in DKNF if and only if it does not have insertion or deletion
anomalies.
The
 Quest for Data Independence, Minimal Plausibility, and Formalization 293

Proof:
() Let R be a DKNF table, r a valid instance of R, and t a tuple
compatible with r; then, r ø {t} satisfies all constraints of R as, from the
compatibility definition, primitive constraints are satisfied, and, from the
DKNF definition, primitive constraints imply all other constraints.
Similarly, for any t[r, r — {t} satisfies all constraints of R as, obvi-
ously, deleting a tuple cannot violate any primitive constraint, and, from
the DKNF definition, primitive constraints imply all other constraints.
Consequently, R does not have update anomalies.
() Let R be a table free of update anomalies; let us assume that R is
not in DKNF; as R is not in DKNF, there is at least one constraint c that is
not implied by the primitive constraints; this means that there is an instance
r of R in which c does not hold, although all primitive constraints are hold-
ing; as the constraint set associated to R is coherent, it follows that there is
at least an instance r* of R that is valid (and, consequently, c holds in it too);
let us build r from r* in the following two steps: first, deleting from r* all
rows, one by one, and secondly inserting in r*, again one by one, all rows
of r (which is obviously possible, as both instances are finite, and all con-
straints on R are static); as r* is valid and r is not, then, during this building
process, there has to be a tuple t and two instances r’ and r” such that:
1. r’ valid
2. r” invalid and
3. r” = r’ ø {t} or r” = r’ — {t}.
Let us consider both above possible cases:
– if r” = r’ ø {t}, then r’ , r; as r satisfies the primitive constraints,
t is compatible with r’; as r” is invalid, it follows that R has an
insertion anomaly;
– if r” = r’ — {t}, then r’ , r*, so r’ is valid; as r” is not valid, it
follows that R has a deletion anomaly.
Trivially, as in any above possible cases the assumption that R does not
have update anomalies is violated, it follows that R is in DKNF.
Q.E.D.

3.6. Prove that C5 from 3.7 (ConstrRelation ° Key = Relation ° Attribute)

implies (;x, y[KEYS_COLUMNS, x ≠ y) (Key(x) = Key(y)  Relation °
294 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

FIGURE 3.28  E-RD for the rdb in Figure 3.2.

There may be at most 108 composers characterized by their com-

pulsory last names (at most 32 characters) and birth years (between
1200 and three years ago), as well as optional first names (at most 32
characters) and passed away years (between 1200 and current year).
There may not be two composers having same first and last names.
Nobody can compose before he/she is three or after his/her death.
Composers may not live more than 120 years.
There may be at most 100 tonalities characterized by their com-
pulsory names (at most 13 characters). There may not be two tonali-
ties having same names.
There may be at most 1012 musical works characterized by their
compulsory composer and opus number (at most 16 characters), as
well as optional number within opus (a natural between 1 and 255),
title (at most 255 characters), and tonality. There may not be two
musical works by a same composer having same opuses and numbers.

FIGURE 3.29  Description of the rdb scheme in Figure 3.2.

Attribute(x) = Relation ° Attribute(y)) (all attributes of a key should belong

to a same relation) and that the converse is not true.
Proof:
() Relation ° Key = Relation ° Attribute  ;x,y[KEYS_COLUMNS,
x ≠ y, Relation(Key(x)) = Relation(Attribute(x)) ` Relation(Key(y)) =
Relation(Attribute(y)); let’s consider that Key(y) = Key(x); as Relation
is a function, it follows that Relation(Key(y)) = Relation(Key(x)) and,
The
 Quest for Data Independence, Minimal Plausibility, and Formalization 295

1. COMPOSERS
a. Range restrictions:
- x (cardinality): 108 (RC1)
- FirstName, LastName: ASCII(32) (RC2)
- BirthYear: [1200, currentYear() — 3] (RC3)
- PassedAwayYear: [1200, currentYear()] (RC4)
b. Compulsory data: x, LastName, BirthYear (RC5)
c. Uniquenesses:
- x (surrogate key) (RC6)
- FirstName • LastName (there may not be two
composers having same first and last names) (RC7)
d. Other restrictions:
- Nobody can compose before he/she is three or
after his/her death. (RC8)
- Composers may not live more than 120 years. (RC9)

2. MUSICAL_WORKS
a. Range restrictions:
- x (cardinality): 1012 (RM1)
- Opus: ASCII(16) (RM2)
- Title: ASCII(255) (RM3)
- No: [1, 255] (RM4)
b. Compulsory data: x, Composer, Opus (RM5)
c. Uniquenesses:
- x (surrogate key) (RM6)
- Composer • Opus • No (there may not be two
musical works of a same composer having same
opus and number) (RM7)
3. TONALITIES
a. Range restrictions:
- x (cardinality): 100 (RT1)
- Tonality: ASCII(13) (RT2)
b. Compulsory data: x, Tonality (RT3)

FIGURE 3.30  Restriction set for the rdb in Figure 3.2.

296 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

c. Uniquenesses:
- x (surrogate key) (RT4)
- Tonality (there may not be two tonalities having
same names) (RT5)

FIGURE 3.30  (Continued)

by equality transitivity, that Relation(Attribute(x)) = Relation(Key(x)) =

Relation(Key(y)) = Relation(Attribute(y))  Relation ° Attribute(x) =
Relation ° Attribute(y)
() Let us consider ;x,y[KEYS_COLUMNS, x ≠ y, such that Key(x)
= Key(y); it follows that Relation ° Attribute(x) = Relation ° Attribute(y);
obviously, as Relation is a function, it follows that Relation(Key(y))
= Relation(Key(x)); let us denote u = Relation ° Attribute(x) = Rela-
tion ° Attribute(y) and v = Relation(Key(y)) = Relation(Key(x)),
u,v[RELATIONS; trivially, it does not follow that u = v for any x and y, so
the converse is not true. Q.E.D.

3.7. Apply algorithm REA1–2 to the rdb from Fig. 3.2.

Solution:
Figure 3.28 presents the corresponding E-RD, Fig. 3.29 its associated
restriction set, and Fig. 3.30 its full description.

3.8. Apply algorithm A7/8–3 to table CITIES from Fig. 3.5.

Solution:
T = CITIES, m = 6, C1 = x, C2 = City, C3 = Commune, C4 = *Country,
C5 = Zip, C6 = Population,
K = {x, City • Commune, *Country • Zip}, k = 3;
K’ = {x, City • Commune, *Country • Zip};
- x is a surrogate autonumber key;
- City • Commune is a key, as there may not be two cities of a same
commune having same names;
The
 Quest for Data Independence, Minimal Plausibility, and Formalization 297

- *Country • Zip is a key, as there may not be two identical zip codes
in a same country;
- City and Commune are prime, but are not keys as they are part of the
key City • Commune;
- *Country and Zip are prime, but are not keys as they are part of the
key *Country • Zip;
SK = { x, City, Commune, *Country, Zip };
T’ = { Population };
- Population is not a key, as there may be two cities having same
population; moreover, it is nonprime, as population cannot help with
uniquely identifying cities.
T’ = Ø;
T’ = { City, Commune, *Country, Zip };
n = 4;
- as 4 > 1:
l = 3;
kmax = C(4, 2) = 4!/(2! * 2!) = 4 * 3 * 2/(2 * 2) = 3 * 2 = 6;
i = 2;
allSuperkeys = false;
- as 2 ≤ 4 and 3 < 6 and true = true:
allSuperkeys = true;
- City • *Country is not a key, as there may be two cities of a same
country (but of different communes) having same names;
- City • Zip is not a key, as there may be two cities (but of different
countries) having same zip codes;
- Commune • *Country is not a key, as, generally, communes have
several cities;
- Commune • Zip is not a key, as there may be two cities of a same com-
mune having same zip codes;
allSuperkeys = false;
i = 3;
- as 3 ≤ 4 and 3 < 6 and true = true:
allSuperkeys = true;
- as all products of arity 3 are superkeys, the algorithm halts without
discovering any new keys, so that K’ = {x, City • Commune, *Country
• Zip}.
298 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

3.14.2  PROPOSED EXERCISES

Some problems are so complex that you have to be highly intelligent

and well-informed just to be undecided about them.
—Laurence J. Peter
3.9. Apply to the following E-R data models the algorithm A1–7 (for
obtaining corresponding relational schemas and nonrelational constraint
sets) and provide a plausible valid instance of at least two rows for each
table:
a. the one from Fig.2.10 and restriction set specified in Subsections
2.7.1 to 2.7.4;
b. the ones obtained in Exercises 2.1 to 2.5 above.
3.10. Apply the algorithm REA1–2 to the following rdbs:
a. the ones from Figs. 3.4 and 3.5;
b. the one presented in section 3.7.
3.11. Apply algorithm A7/8–3 to the Tables 3.8 to 3.13 and 3.18,
3.19, 3.20, 3.21, and 3.22, as well as to all tables from Figs. 3.13 and 3.26.
3.12. Design and develop needed standard ANSI-92 SQL statements
for creating and populating the tables shown in Figs. 3.4 and 3.5, as well as
those from subsection 3.7 (Tables from 3.7 to 3.13). You can do it directly
in the SQL of a RDBMS version of your choice, so as to be able to also
test them.
3.13. Design, develop, and test SQL queries for the ones from section
1.6 (on the corresponding rdb obtained from Exercise 3.9.b).
3.14. Design, develop, and test SQL statements for computing the fol-
lowing sets:
a. from Fig. 3.2:
– <First and LastName, Title, TonalityName, Opus, No> for all
musical works, in the ascending order of the first and last names,
then tonality, opus, and number.
– <First and LastName, LifeYears> for all composers that lived
at least k years, k natural parameter, in descending order of the
numbers of years they lived and then ascending on first and last
names.
– <First and LastName, WorksNo, TonalitiesNo> for all composers
that wrote in at least k tonalities and for each such tonality at least
The
 Quest for Data Independence, Minimal Plausibility, and Formalization 299

n works, k and n natural parameters, in descending order of the

numbers of tonalities, of works, and then ascending on first and
last names; give (and compare their execution times after adding
at least 100 other works) two solutions: one with and one without
subqueries.
b. from Fig. 3.4:
– <CountryName, CapitalName, Population, StatesNo, CitiesNo>
for all countries, including those for which the capital is not
known, in the descending order of population, cities and states
numbers, and then ascending on country names.
– < CityName, StateName, CountryName, Population> for all cit-
ies that are both country and state capitals, in the descending
order of population and then ascending on country name.
c. from Fig. 3.5:
– <CountryName, StateName, CommuneName, CityName, Popu-
lation> for all cities and communes, in the descending order of
their population and then ascending on country, state, commune,
and city names.
– <CountryName, StateName, CommunesNo, CitiesNo, TotCities-
Popul.> for all states of all countries, including those that have
no communes and/or cities, in the descending order of the total
city populations, number of communes, then cities, and then
ascending on country and state name.
d. from Table 3.3: <FirstandLastNames, SSN, CoordEmploy-
eesNo> for all managers, in descending order of the number of
corresponding coordinated employees and then ascending on
their first and last names.
e. from Fig. 3.6: <CountryName, CapitalName, SubordCoun-
triesNo, MinSubordYear>, including countries with not known
capitals, in descending order of the number of subordinated
countries and then ascending on the minimum subordination
year and, finally, country name.

3.15. Design, develop, and test SQL statements for:

a. adding all remaining U.S. states data in the STATES table of Fig.
3.4.
300 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

b. replacing country names from the COUNTRIES table of Fig. 3.4

with their corresponding uppercase names.
c. deleting from table TONALITIES of Fig. 3.2 all lines for which
there is not at least one musical work composed in those tonalities.
3.16. Design, develop, and test SQL statements for replacing the
instance of a temporary table CAPITALS(CountryName, StateName,
CapitalName) with the corresponding data from the tables in Fig. 3.4, in
ascending order of countries, states, and capital names, only for cities that
are both country and state capitals.
3.17. Design and develop needed standard ANSI-92 SQL statements
for the following queries on the metacatalog rdb presented in section 3.7
and compute the corresponding results (alternatively, you can do it directly
in the SQL of a RDBMS version of your choice, so that to may also test
them):
a. Compute the set of quadruples <db name, no. of relations, total
no. of attributes, total no. of constraints>, in the descending
order of the numbers of relations, then on the one of constraints,
then of attributes, and then ascending on the db names.
b. Compute the set of quadruples <db name, relation name, no. of
attributes per relation, no. of constraints per relation> only for
those dbs that have at least 3 relations, in the descending order
of the numbers of attributes, then on the one of constraints, and
then ascending on the db and relation names.
c. Compute the set of quintuples <db name, relation name, attribute
name, no. of keys containing current attribute, no. of foreign
keys containing current attribute> only for those relations that
have at least n keys and m foreign keys (n and m natural param-
eters), in the descending order of the numbers of keys, then on
the one of foreign keys, and then ascending on the db, relation,
and attribute names.
d. Compute the set of quintuples <db name, relation name, attribute
name, no. of keys containing current attribute, no. of foreign
keys containing current attribute> only for the prime attributes,
in the descending order of the numbers of keys, then on the one
of foreign keys, and then ascending on the db, relation, and attri-
bute names.
The
 Quest for Data Independence, Minimal Plausibility, and Formalization 301

e. Compute the set of quintuples <db name, relation name, con-

straint name, constraint type, no. of attributes of the current con-
straint> only for those relations that have at least n constraints
(n natural parameter), in the descending order of the numbers of
attributes, and then ascending on the db, relation, and constraint
types and names.
3.18. Prove that the join operator is commutative and associative.
3.19. Compute minimum and maximum cardinalities for the joins of n
tables, n > 1, natural, based on its operands’ cardinalities.
3.20. Prove that the natural join of two tables having identical schemas
is equal to their intersection.
3.21. Is the full outer join equal to the Cartesian product? Why? Please
exemplify.
3.22. Design, develop, and test equivalent LEFT, RIGHT, and FULL
OUTER JOIN SELECT statements by using only INNER JOIN and
UNION operators.
3.23. Provide necessary and sufficient conditions for SELECT state-
ments to always compute same results regardless of the rdb instance when
they only differ by DISTINCTROW instead of DISTINCT. Give an exam-
ple and a counterexample.
3.24. a. Design, develop, and test a parameterized SQL statements fam-
ily template for computing the difference between instances
of two tables having identical schemas, by using the IsNull
predicate.
b. Do same thing as above for the symmetrical difference.
3.25. Prove that division is a sort of Cartesian product inverse, as, for
any tables t1 and t2, (t1 × t2) ÷ t2 = t1.
3.26. (a) Try to design, develop, and test a SELECT having in its
WHERE clause a parameterized IN predicate. Why do you
think it does not execute? (hint: revise RDBMSs data types).
(b) Design, develop, and test, however, a pure SQL (ANSI-92)
solution to this problem (hint: use a temporary table for stor-
ing actual parameter values).
3.27. a. Add to Table 3.3 at least 10 new rows such as to store in
ReportsTo a hierarchy of at least 5 levels.
b. Design, develop, and test a SQL query for computing the set of all
“leaf” employees (i.e., those to whom nobody is reporting to).
302 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

c. Design, develop, and test a SQL query for computing the set of all
pairs <“leaf” employee, employee to whom the “leaf” employee
reports to>.
d. Design, develop, and test a SQL query for computing the set of
all similar pairs for the first and the second hierarchical levels
(hint: if 1 reports to 2 and 2 reports to 3, then the answer should
include the set {<1,2>, <2,3>, <1,3>}).
e.* Design, develop, and test a high level programming language
program (e.g., C#, Java, C++, VBA, etc.) that generates the
SQL query for computing the set of all such pairs up to the k-th
hierarchical level, k being a natural parameter; test the result-
ing SQL output for k = 4.
f.* Design, develop, and test for k = 4 and k greater than the
ReportsTo tree height a high level programming language pro-
gram (e.g., C#, Java, C++, VBA, etc.) with embedded SQL to
compute same thing, by using a temporary table for storing
the result; if k ≤ 0, then compute the subset of all employees
reporting to nobody (hint: pad second column with nulls); if k
greater than the highest hierarchical level h, stop immediately
after adding level h pairs, that is, when the transitive closure
of ReportsTo was computed (hint: add to the temporary table a
Level numeric column; when k > 0, initialize it with the result
of query (c) above, all of them having 1 for Level; loop until k
or until no more lines are added to the temporary result table;
in each step i, increase Level values by 1 and add to the tempo-
rary tables the pairs of the current level, which can be obtained
by a join between the rows of the temporary table having the
Level values equal to i – 1 and Table 3.3).
g.** Compute the transitive closure of ReportsTo in extended SQL
(e.g., in T-SQL, PL/SQL, etc.); establish a top of execution
speeds for computing it with solutions (e), (f), and (g), when
Table 3.3 has at least 100 rows and 16 levels.
3.28. Consider the subuniverse of a simple OS network file manage-
ment system that is storing only file names, extensions, sizes, folders
(which are files too, having size 0!), virtual, logic, and physical drives, as
well as computers.
The
 Quest for Data Independence, Minimal Plausibility, and Formalization 303

a. Using algorithms A0, A1–7 and A7/8–3, as well as the best prac-
tice rules from sections 2.10 and 3.11 design and implement its
DKNF rdb.
b. Populate it with at least 2 computers, each of which having at
least 2 physical, 4 logic, and 2 virtual drives; each logic drive
should have at least 16 non folder files and 8 folders structured
in hierarchies of at least 4 levels.
c.* Design, develop, and test a program for computing the total
size of all files of a subtree of folders, the parameter being the
path of its root folder (hint: see (f) and (g) of exercise 3.27
above, as this is also a transitive closure type problem).
d.* Design, develop, and test a program for computing the set
<computer name, physical drive name, logic drive letter, full
folder path, file name and extension, size> for all computers
having at least k logic drives, each of which having at least
m folders with at least n files each, each such file having size
at least s, where k, m, n, and s are natural parameters, in the
ascending order of computer names, logic drive letters, full
folder paths, then descending on size and ascending on file
names and extensions (hint: also see (f) and (g) of exercise
3.27 above, as computing full paths is also a transitive closure
type problem).
3.29.** Prove that transitive closures are not the only meaningful
queries that complete relational languages cannot express, although their
results contain only rdb stored data87 (hint: consider Table 3.3 and try to
design SQL statements for computing pairs of employees reporting to a
same person such that the result does not contain any pair (f, g) if it already
contains the pair (g, f). Design, develop, and test a program in a high level
programming language at your choice, which embeds SQL, for computing
this desired result. Generalize!).
3.30. Translate all remaining SQL queries from subsection 3.9.2,
as well as those from exercises 3.12 to 3.14 into corresponding RA
expressions.

And, moreover, they are invariant with respect to any automorphism of its instance (where automor-
87

phisms are autofunctions renaming values such that the corresponding set remains the same).
304 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

3.31. Prove that the RA selection and projection operators are mono-
tonic. Extend monotonicity definition to n-ary operators, n > 1, natural,
and prove that the RA join operator is monotonic too.
3.32. Consider any m tables t1(X1), …, tm(Xm), m ≥ 1, natural (where Xj-s
are their schemas); prove that the projection operator is idempotent, as:
πXj(|><|k=1 m (πXk(|><|i=1 m ti))) = πXj(|><|i=1 m ti), ;j, 1≤ j ≤ m.
3.33. Consider same tables as above; prove that there is a duality
between projection and join, as:
(a) πXj(|><|i=1 m ti) # tj, ;j, 1≤ j ≤ m.
(b) πXj(|><|i=1 m ti) = tj, ;j, 1≤ j ≤ m  t1, …, tm have a full join.
3.34. Consider table t(X) and X1, …, Xm, m ≥ 1, natural, attribute sets
such that øj=1 mXj=X; prove the following somewhat dual results of the
above ones:
(a) |><|j=1 m (πXj(t)) $ t.
(b) |><|k=1 m (πXk(|><|j=1 m(πXj(t)))) = |><|j=1 m (πXj(t)).
3.35. Prove that any RA expression may be transformed into an
equivalent one in which selections have only atomic conditions, projec-
tions eliminate only one attribute, and the rest of the needed operators are
unions, renaming, differences, and Cartesian products.
3.36. Prove that any RA expression may be transformed into an equiva-
lent one whose constant tables are all defined over only one attribute and
contain only one row.
3.37. Prove that the proposition corresponding to no-information null
values (i.e., R-k(t[A1], ..., t[Ak–1], t[Ak+1], ..., t[An])) ) is equivalent to the
disjunction of the formulas associated to the other two types of null val-
ues: x(R(t[A1], ..., t[Ak–1], x, t[Ak+1], ..., t[An])) ~ (x(R(t[A1], ..., t[Ak–1], x,
t[Ak+1], ..., t[An])) = R–k(t[A1], ..., t[Ak–1], t[Ak+1], ..., t[An])).
3.38. Prove that ;Y , X, card(X) = card(πY(X))  'K # Y, K key.
3.39. Apply algorithm A7/8–3 to the following table schemas:
a. RULERS (x, Title, Name, Sex, Dynasty, Father, Mother, BirthDay,
BirthPlace, PassedAwayDay, PassedAwayPlace, BurrialPlace,
KilledBy, Nationality, Notes)
b. REIGNS (x, Ruler, Country, FromDate, ToDate)
c. BATTLES (x, BattleYear, BattleSite, Ruler, Opponent, RuledArmy,
EnnemyArmy, Victory?, Notes)
d. MARRIAGES (x, Husband, Wife, WeddingDate, DivorceDate)
The
 Quest for Data Independence, Minimal Plausibility, and Formalization 305

e. MONUMENTS_PICTURES (x, Monument, JPEGPicture, Picture-

Size, PictureDescription)
3.40. Let R be any rdb table and c1, …, cn its columns, n > 1, natural, 1
≤ i, j ≤ n; if 'j such that, ;i, ci → cj, cj is minimal; dually, if 'i such that,
;j, ci → cj, ci is maximal; prove that:
(i) c is minimal  card(Im(c)) = 1
(ii) c is maximal  c is one-to-one.
3.41. Let S be any set and f, g, h, f1, …, fn any functions defined on it,
n > 1, natural; let K and K’ be any function (Cartesian) products defined
on S; prove that:
(i) ker(f•g) = ker(f) ∩ ker(g)
(ii) f1•…•fn → fi, ;i[{1, …, n}
(iii) g → fi, ;i[{1, …, n}  g → f1•…•fn.
(iv) K is a key  K → f, for any f
(v) K is a key and K’ → K  K’ is one-to-one.
3.42. Let R be any rdb relation and f, g, h some of its columns; if f → g
and g → f then f is functionally equivalent to g88, denoted f ↔ g; prove that:
(i) the functional equivalence (FED) “↔“ is an equivalence relation
(ii) all minimal columns of a relation are functional equivalent
(iii) all maximal columns of a relation are functional equivalent
(iv) all keys of a relation are functional equivalent
(v) (f•g)•h ↔ f•(g•h)
(vi) f•g ↔ g•f
(vii) f•f ↔ f.
3.43. Prove that when discarding a’s one-to-oneness from Proposition
3.2, 3.2(i) and 3.2(ii) still hold, while 3.2(iv) does not hold anymore, and
3.2(iii) becomes card(Im(fk)) ≤ card(Im(f)).
3.44. Prove that FD X → Y is trivial  X $ Y.
3.45. Let R(U) be a table and n = card(U), n > 1, natural; prove that the
number of trivial FDs over R(U) is 3n — 2n.
3.46. Prove that in any instance of a table T(U), if X • Y = U, then X →
Y  X is a key.
3.47. Prove that any table satisfies FD X → Y if and only if it satisfies
X → A, ↔A [ Y.
88
Note that, trivially, the kernels of two equivalent functions are equal and any function is equivalent
to itself.
306 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

3.48. Let t(U) be a table and X1, X2, X , U such that X1 • X2 = U and
X = X1 ∩ X2; t(U) has a lossless decomposition over X1, X2 if it satisfies at
least one of the following FDs: X → X1 or X → X2.
3.49. Prove that if instance i of a table satisfies X→ Y, Z # X and W $
Y, then i satisfies Z → W as well.
3.50. Prove that if instance i of a table satisfies X → Y and Z→ W, Z #
Y, then i satisfies X → YW as well.
3.51. Prove that (unikey FD schemas characterization theorem): a
scheme [R, F], such that R(U) and F = {X1 → Y1, …, Xk → Yk}, k > 0,
natural, has only one key if and only if U — Z1•…•Zk is a superkey, where
Zi = Yi — Xi, 1 ≤ i ≤ k.
3.52. Prove that BCNF  3NF  2NF  1NF (any scheme in BCNF
is in 3NF too, any one in 3NF is in 2NF too, and any one in 2NF is in
1NF too).
3.53. By definition, a table scheme S with FDs is in (3,3)-NF if for any
logic proposition F, a nontrivial FD X → A holds in σF(s) for any valid
instance s of S if and only if X is a superkey of S; prove that (3,3)-NF 
BCNF (a scheme in (3,3)-NF is in BCNF too).
3.54.* A scheme [R, F] includes an elementary key K if there is at least
an attribute A[R such that ¬'K’ , K with K’ → A [ F+; [R, F] is in the
elementary keys NF (EKNF) if for any nontrivial FD X → A [ F, either X
is an elementary key or A is part of such a key. Prove that EKNF is strictly
stronger than 3NF, but strictly weaker than BCNF.
3.55. Prove that (MVD complementary property) if R(U) satisfies MVD
X →→ Y, then it satisfies MVD X →→ U — XY too.
3.56. Prove that (trivial MVDs characterization theorem) X →→ U of
T is trivial if and only if U # X or T = X • U.
3.57. Consider table t(U) and XY # U; prove that:
(a) t satisfies FD X → Y  t satisfies MVD X →→ Y
(b) the reverse is not true.
3.58. Prove that a relation may satisfy MVD X →→ YZ, although it
violates FD X → Y.
3.59. Prove that X →→ Y ` X →→ Z  X →→ Y — Z (and conse-
quently, due to complementarity, it also implies X →→ Y ∩ Z).
3.60. Prove that:
(a) 4NF  BCNF (a scheme in 4NF is in BCNF too)
The
 Quest for Data Independence, Minimal Plausibility, and Formalization 307

(b) above implication would not be true were 4NF only referring
MVDs from Γ instead of Γ +.
3.61. Prove that a table in 4NF cannot have more than one pair of
complementary nontrivial MVDs.
3.62. (i) What is the truth value of the following sentence: “Any table
with two columns is in PJNF”?
(ii) Propose and prove a proposition that provides necessary and
sufficient conditions for a two columns table to be in PJNF.
3.63. Prove that 5NF  4NF (a scheme in 5NF is in 4NF too).
3.64. Prove that:
(i) a scheme [R, F] is in BCNF (4NF, respectively) if for any FD
(MVD, respectively) d[F there is only one key dependency
implying d.
(ii) a similar result for PJNF (i.e., d is a JD) is not true.
3.65.* Prove that, generally, given a set F of FDs, it is not possible
to find a set M of MVDs such that an instance satisfies F if and only if it
satisfies M.
3.66.* Let R(U) be a relation scheme, D its associated domain constraint
set, K its key dependency set, and M an associated set of FDs and MVDs;
(i) if, for any attribute A[U, card(DA) ≥ 2, then, for any FD or
MVD d of M, d is implied by D ø M if and only if it is implied
by M;
(ii) Prove that in the absence of card(DA) ≥ 2, (a) above does not
hold anymore;
(iii) if, for any attribute A[U, card(DA) ≥ n, n > 1, natural, then, for
any JD j over U having n components (i.e., of form |><| [X1, X2,
…, Xn], where Xi # U, ↔i, 1≤ i ≤ n), j is implied by D ø K if
and only if it is implied by K.
3.67. (DKNF is the highest RDM normal form theorem) Consider a
scheme [R, Γ ] such that all domain constraints allow at least k distinct
values, where natural k > 1 is the components number of the JD having
the greatest number of components from all JDs of a cover J of the JDs
subset of Γ +, and, when this subset is empty, there are at least two possible
distinct values for any attribute of any table from R; prove that:
(i) DKNF  PJNF (any scheme in DKNF is in PJNF too)
(ii) DKNF  (3,3)-NF (any scheme in DKNF is in (3,3)-NF too).
308 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

3.68. Give an example of an incoherent constraint set made out only

from coherent constraints.
3.69. Prove that a constraint set Γ does not imply a constraint Γ if and
only if there exists a db instance that satisfies Γ, but not Γ.
3.70.* Consider any two constraint sets Γ and Δ; prove that:
(i) Γ # Γ +  (any constraint set is included in its
closure)
(ii) (Γ ) = Γ
+ + +
(the closure operator is idempotent)
(iii) Γ # Δ  Γ + # Δ+ (the closure operator is monotonic)
(iv) Γ + ø Δ+ ≠ (Γ ø Δ)+ (the closure operator is not additive)
3.71. Prove that C9 from Fig. 3.21 (ConstrRelation ° ForeignKey = Rela-
tion ° FkAttribute) implies the constraint (;x,y[FOREIGN_KEYS_COLS,
x≠y) (ForeignKey(x) = ForeignKey(y)  ConstrRelation ° FkAttribute(x)
= ConstrRelation ° FkAttribute(y)) (all attributes of a foreign key should
belong to a same relation) and that the converse is not true.
3.72. Prove that the referential integrity constraint (C10 from Fig.
3.21: (;x[FOREIGN_KEYS_COLUMNS)(Im(FKAttribute(x)) #
Im(RefAttribute(x)))) implies the constraint (;x[FOREIGN_KEYS_COL-
UMNS)(Im(Domain ° FKAttribute(x)) # Im(Domain ° RefAttribute(x)))
(foreign keys’ domains should always be included in the corresponding
referenced attributes’ domains) and that the converse is not true.
3.73*. Considering the relations from Fig. 3.21, prove the following
proposition: (;x[KEYS_COLUMNS)(;y[FOREIGN_KEYS_COLS)(Pri
maryKey(Relation(Attribute(x))) = Key(x) ` Attribute(x) = FKAttribute(y)
 Relation(Attribute(x)) # Relation(RefAttribute(y))) (a primary key of
a relation is a foreign key if and only if that relation corresponds to a set
that is included in the set to which corresponds the relation referenced by
the primary key).
3.74. Prove that the metacatalog scheme presented in Section 3.7 is in
DKNF.
3.75. Design a DKNF extension to the metacatalog scheme presented
in Section 3.7 for also storing tuple constraints. Provide a nonvoid, valid,
and plausible instance for it.
3.76. Consider the scheme STUDENTS(x, Course, Credits, Student,
Sex, Age). Identify its FDs and keys. Design an equivalent BCNF scheme.
Is a DKNF equivalent one obtainable too? Justify your answer.
The
 Quest for Data Independence, Minimal Plausibility, and Formalization 309

3.77. Consider the scheme STUDENTS(#Student, #Lab, #PC, Student,

Lab, PC) having following FDs: #Student → Student, #Student → Lab,
#Student → #Lab, #Student → PC, #Lab→ PC, #Lab → Lab, #PC → PC,
#PC → #Lab, and Lab → #Lab. Analyze if it has anomalies and design an
equivalent DKNF one.
3.78. Consider a “technological” db for storing data on items (spare
parts, submodules, modules, products, etc.) that are made out of other
items. The technological item structures are of tree type.
a. Design and implement in any DBMS of your choice a correspond-
ing DKNF db, only containing item code, description, price, and
structure. Provide a valid instance with at least 4 tree levels.
b. Design and test for this instance SPJR expressions and correspond-
ing SQL statements for computing the subsets of:
(i). structures (<item code and description, immediately superior
item code and description>) for all items having at least three
embedded levels;
(ii). items (<code, description, price>) for which price is at most
10% of the price of its immediately superior item;
3.79. Consider a HR subuniverse interested in employees’ first and last
name, sex, birth date and place, home address, hire and end working dates,
and working place (subsidiary), as well as subsidiaries’ name, address,
and subordination (e.g., Microsoft Romania srl, from Bucharest, Romania,
is subordinated to Microsoft Central and Eastern Europe GmbH, from
Munich, Germany, which is subordinated to Microsoft Corporation, from
Redmond, U.S.A.).
a. Design and implement in any DBMS of your choice a correspond-
ing DKNF db and provide a valid instance with at least four rows
per table.
b. Design and test for this instance SPJR expressions and correspond-
ing SQL statements for computing the subsets of:
(i) stored subordinations (working place name, direct subsidiary
name), for all working places coordinating at least three levels
of subsidiaries;
(ii) employees that worked in their birth city at a working place
directly subordinated to an entity having its head-quarters in
the same city;
310 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

3.80. Consider the set of courses offered by a University; each of them

has a (possibly empty) set of prerequisite ones (i.e., courses that have to
be completed prior to enrollment to the current one); only titles are of
interest for courses. For each course several books may be recommended
and some books may be even recommended for several courses; only the
title and the (co)authors are of interest for books. For students, only first,
middle, and last name, birth date and place, completion degrees, and cur-
rent enrollments are of interest.
a. Design a DKRINF db scheme for this subuniverse and provide a
plausible instance of it with at least four rows per table.
b. Design and test for this instance SPJR expressions and correspond-
ing SQL statements for computing the subsets of:
(i) courses (title, precondition title) having at least two preconditions;
(ii) courses (title) for which at least four students are enrolled;
(iii) students (first, middle, and last name, birth date and place) that
enrolled to courses for which books containing “Mathematical
Logic” in their titles are recommended;
(iv) completions (student first, middle, and last name, birth date and
place, course title) for which course title is identical or included
in the title of at least one recommended book for that course.
3.81.
a. Refine the rdb scheme from Fig. 3.5, by adding a STREETS table to
store street names (which, generally, are repeatedly found in doz-
ens, if not hundreds of cities) and all needed foreign keys, such as
to obtain a DKNF scheme. Provide a plausible instance of it with
at least two rows per city from CITIES.
b. Apply the algorithm A7/8–3 for finding all corresponding STREETS
keys.
c. Design and test for this instance SPJR expressions and correspond-
ing SQL statements for computing the subsets of:
(i) countries having at least two states each of which having at
least two cities each having at least 1,000,000 inhabitants;
(ii) states (name, country name, capital city name) to which capital
cities having at least 1,000,000 inhabitants belong to;
(iii) Romania’s states (name, vehicle plate code, telephone prefix,
state capital city name) whose names are different from their
capital city name;
The
 Quest for Data Independence, Minimal Plausibility, and Formalization 311

(iv) countries (name, vehicle plates code, telephone prefix, capital

city name) having same name as their capital city name;
(v) cities (country name, capital city name, state name, state capi-
tal city name) that are country capitals, but not state capitals;
(vi) streets (country name, state name, city name, street name) that
have names existing in at least two cities, in the ascending
order of street, country, state, and city names.
3.82. Build a directed graph having as nodes all RDM concepts from this
chapter, including exercises, and with arrows (implicitly) labeled “directly
based on” (hint: some definitions are leaves, whereas propositions, theo-
rems, corollaries, and the rest of the definitions are nonterminal nodes).
a. Which are its sets of minimal and maximal elements?
b.* Design the corresponding DKNF db scheme and provide its full
instance (hint: extend the RDM metamodel from Subsection 3.7.2).
3.83. Consider the following table scheme: PEOPLE (x, FirstName,
LastName, Sex, BirthDate, Mother, Father), where Mother and Father
are acyclic autofunctions (i.e., foreign keys referencing PEOPLE.x whose
column values are not allowed storing cycles, i.e. the fact that somebody
is his/her own ancestor/descendant).
a. Provide at least three different valid instances, each having at least
4 rows per table.
b. Design and test for every instance a SPJR expression and its corre-
sponding SQL statement for computing the set <x, First&LastName,
Sex, BirthDate, xSibling, SiblingFirst&LastName, SiblingBirth-
Date, xCommonGrandFather, xCommonGrandMother>.
3.84. Consider the table FILES from Fig. 3.31.
a. Apply algorithm A7/8–3 for discovering all of its semantic keys.
b.* Discover all of its other (both relational and non-relational)
constraints.
c.** Design and develop in any high-level event-driven programming
language of your choice (e.g., Java, C#, VBA, etc.) needed methods
for enforcing all found non-relational constraints.
d. Design the SQL statements (both with and without subqueries) for
computing the set of all files from the root folder, its subfolders,
and their subsubfolders; what is the corresponding result?
e. Same as (c) above, but also adding the fourth folder hierarchical
level of files.
312 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

x FileName FileExt Folder FileSize Folder?

0 0 Y
1 Bootmgr 0 399 N
2 Swapfile sys 0 266,144 N
3 Windows 0 0 Y
4 Explorer exe 3 2,336 N
5 System32 3 0 Y
6 Shutdown exe 5 35 N
7 WinMetadata 5 0 Y
8 Windows.Security winmd 7 45 N

FIGURE 3.31  Some files and folders from a MS Win8 computer.

f. Design and develop a parameterized method in any high-level pro-

gramming language of your choice (e.g., Java, C#, VBA, etc.) that
accepts as parameters a table name T, one of its columns name C,
and a natural value k, and generates and stores in a (e.g., .txt or
.sql) file the corresponding ANSI-92 SQL standard statements for
computing the k-th level C’s partial closure. Run this method for
T = “FILES”, C = “Folder”, and k = 7. Run the corresponding gen-
erated SQL statement over table FILES’ instance.
3.85. Consider the RDM model of E-RDM presented in Subsection
3.3.2.
a. Apply the algorithm A7/8–3 to each of its tables, thus proving that
all keys are correctly declared and that there are no other keys.
b.* Automatically (re)compute *Arity values and enforce the non-rela-
tional constraints (RU5) and (RU6) (hint: use either extended or
embedded SQL).
3.86.* Prove that constraint C11 from Fig. 3.21 implies the follow-
ing constraint: C14: (;x[KEYS_COLUMNS)(;y[FOREIGN_KEYS_
COLS)(PrimaryKey(Relation(Attribute(x))) = Key(x) ` Attribute(x) =
FKAttribute(y)  Relation(Attribute(x)) ≠ Relation(RefAttribute(y))) (a
primary key of a relation must not reference that relation, that is trivial
inclusions R # R should be banned).
3.87. Consider the constraint set from Fig. 3.21 (Subsection 3.7.2).
a. Prove that it is coherent.
The
 Quest for Data Independence, Minimal Plausibility, and Formalization 313

b.* Prove that it does not imply constraint C: (;x[CONSTRAINTS)

(;y1, …, yn [ KEYS_COLUMNS, n > 0, natural)(ConstrType(x) =
‘K’ ` Key(y1) = … = Key(yn) = x  Attribute(y1) • … • Attribute(yn)
is onto) (keys are onto)
c.** Prove that it is minimal.
d. Add a constraint to it such that it becomes incoherent.
3.88. Consider the http://www.databaseanswers.org/data_models/index.
htm data models and:
a. apply algorithm A7/8–3 to all of their tables (in order to discover
all of their keys);
b. apply algorithm REA1–2 to all models (in order to infer their cor-
responding E-R data models);
c. analyze each such E-R data model and correct it if necessary;
d. for each corrected E-R data model, apply algorithm A1–7 and com-
pare its results with the corresponding initial data model.
3.89. Consider the extension from problem 2.11 and modify accord-
ingly the solution of problem 3.0.

3.15  PAST AND PRESENT

Einstein’s space is no closer to reality than Van Gogh’s sky. The

glory of science is not in a truth more absolute than the truth of Bach or
Tolstoy, but in the act of creation itself. The scientist’s discoveries impose
his own order on chaos, as the composer or painter imposes his; an
order that always refers to limited aspects of reality, and is based on the
observer’s frame of reference, which differs from period to period as a
Rembrandt nude differs from a nude by Manet.
—Arthur Koestler

There are very many books on RDM. Not only from my point of view, the
current “Bible” in this domain still remains (Abiteboul et al., 1995).
Chris Date’s books (the latest being Date (2013, 2012, 2011, 2003)), are
a must too. Other latest not to miss ones are Garcia-Molina et al. (2014),
Hernandez (2013), Churcher (2012), Lightstone et al. (2011), and Halpin
and Morgan (2008). Jeffrey D. Ullman’s ones (Ullman (1988, 1989), Ull-
man and Widom (2007)) are still of very great value.
314 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

3.15.1  DEPENDENCY (CONSTRAINT) THEORY

Although the big word on the left is ‘compassion’, the big agenda on
the left is dependency.
—Thomas Sowell

3.15.1.1  Dependencies and Normal Forms

Form follows function.

—Louis Henri Sullivan

RDM and RA were introduced by Edgar F. Codd (Codd, 1970), to

whom we all are especially indebted, among others, for introducing and
studying RC (Codd, 1971a), keys, FDs, and the first three NFs (Codd,
1971b, 1972, 1974)), as well as for extending RDM with semantic con-
cepts too (Codd, 1979) and RA, by including a three valued logic and a
so-called nulls substitution principle89 (Codd, 1975). Although BCNF was
also introduced in Codd (1974), it was actually first presented in Heath
(1971).
The much more elegant definition of FDs used in this book is due to
Val (Breazu-)Tannen (Breazu and Mancas, 1980 and 1981).
MVDs were independently introduced by Zaniolo (1976), Fagin
(1977), and Delobel (1978), and then generalized by Rissanen (1977),
Nicolas (1978), and Aho et al. (1979).
JDs were introduced by Rissanen (1979).
We are all especially indebted to Ronald Fagin too, who, besides
MVDs, introduced 4NF (Fagin, 1977), 5NF (Fagin, 1979), DKNF and the
primitive (domain and key) constraints (Fagin, 1981).
Anomalies formal study is due to Bernstein and Goodman (1980),
LeDoux and Parker (1982), and, especially, to Chan (1989).
(3,3)-NF and EKNF (see Exercises 3.53 and 3.54, respectively) were
proposed by Smith (1978) and Zaniolo (1982), respectively.

89
which simply states that unknown nulls may take on values only in a finitely
restricted attribute domain
The
 Quest for Data Independence, Minimal Plausibility, and Formalization 315

Orlowska and Zhang (1992) proved that the PJNF (generally thought
of as being equivalent to 5NF) is stronger than 5NF. Date and Fagin (1992)
proved that any 3NF table only having single keys is also in PJNF.
Unknown nulls were introduced by Grant (1977); Vassiliou (1979)
and Vassiliou (1980a) consider both unknown and nonexistent ones, in
the framework of denotational semantics; no-information ones were intro-
duced by Zaniolo (1977) and then studied by lot of researchers for more
than a decade.
Lipski (1979) and Lipski (1981) proposed a first null values gener-
alization, by using partially specified values (nonvoid subsets of attri-
bute domains from which nulls may take values). Wong (1982) advanced
another generalization, based on assigning probabilities to each value of
any domain: 1 to specified ones, 0 to all others, 0 too for all values corre-
sponding to a nonexistent null, 1/n to all unknown nulls for a domain hav-
ing n values90, 1/k to partially specified nulls for a subset having k values,
and again 0 for all of the rest of the values.
Korth and Ullman (1980), Maier (1980), Sagiv (1981), Imielinski and
Lipski (1981) proposed and studied marked nulls: numbered null values
that allow for storing the fact that certain ones should coincide.
Interaction between FDs and nulls was studied by (Vassiliou, 1980b),
(Honeyman, 1982) (who considered FDs as being interrelational con-
straints), Lien (1982), Atzeni and Morfuni (1984) and Atzeni and Morfuni
(1986).
Existence constraints are due to Maier (1980a). Other constraints on
nulls were proposed by Maier (1980a) (disjunctive existence constraints),
then studied in depth by Goldstein (1981) and Atzeni and Morfuni (1986)
and Sciore (1981) (objects).
Although referential integrity was already well-known in the 70s and
despite their crucial role in any RDBMS, inclusion dependencies were
first considered much later, starting with Casanova (1981) and Casanova
and Fagin (1984). IND interaction with other dependency types is studied,
for example, in Casanova and Vidal (1983).
Tuple (check) constraints were very rarely formally studied. Mancas et al.
(2003) introduced a generalization of tuple (check) constraints called object
constraints, which also presents and studies a first order object calculus,
90
When domains are infinite, distributions should be considered instead.
316 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

strictly more powerful than the relational ones and less powerful than the
predicates one, designed to express both domain and object constraints.
More general constraint classes were introduced and studied by Fagin
(1982) (where also studied is domain independence—see Subsection
3.15.2), Yannakakis and Papadimitriou (1982), Beeri and Vardi (1984a),
and Beeri and Vardi (1984b).
General dependency theory was addressed by very many papers: for
example, Fagin and Vardi (1986), Vardi (1987), Kanellakis (1991), and
Thalheim (1991); Nicolas (1978), Gallaire and Minker (1978), and Nico-
las (1982) approaches it from the first order logic perspective; Imielinski and
Lipski (1983) had an interesting and singular point of view in this field too.
The set of all known dependencies may be partitioned in the following
two classes (Beeri and Vardi (1984a) and Beeri and Vardi (1984b)):

FIGURE 3.32  RDM normal forms and corresponding dependency types’ hierarchy.
The
 Quest for Data Independence, Minimal Plausibility, and Formalization 317

1. equalities generating dependencies (which includes FDs) that are

always equalizing column values on certain rows;
2. tuples generating dependencies (which includes MVDs and JDs)
that are always adding new rows to tables’ instances.
Figure 3.32 synoptically shows NFs hierarchy and corresponding con-
straint types.
Lot of effort in RDM’s theory was deployed for solving the normaliza-
tion problem: given a class of constraints (generally FD, or FD and MVD),
design an as fast as possible algorithm that accepts any set of constraints
of that class and outputs a db scheme in a desired NF and, if possible, also
displays other desired properties.
Besides LD, another desirable property, called constraint preservation
(CP) or representation, was first introduced in RDM for FDs (Beeri and
Rissanen, 1980): given a relation scheme R with an associated set F of
constraints and a decomposition R1, …, Rn of it with corresponding associ-
ated constraint sets F1, …, Fn, the decomposition is said to preserve depen-
dencies if (F1 ø…ø Fn)+= F+.
Two equivalent strategy types were considered towards this goal: syn-
thesis (bottom-up), which assumes no relations, but only a set of attri-
butes (which is suited only for FDs and not for MVDs) and decomposition
(top-down), which assumes either several relations or even only one (said
“universal”).
Beeri and Honeyman (1981) includes a polynomial algorithm for FD
preservation testing, which is fundamental for the top-down approach.
Table 3.25 contains the main normalization algorithms (where S/D indi-
cates the strategy type –S(ynthesis) or D(ecomposition), NF –the output

TABLE 3.25  Main Normalization Algorithms

No. Authors S/D NF Class O(?) Prop.
1 (Bernstein, 1976) S 3 FD P CP
2 (Biskup et al., 1979) S 3 FD P LD+CP
3 (Fagin, 1977) D 4 FD+MVD E LD
4 (Tsou and Fischer, 1982) D 4 FD+MVD P LD
5 (Ullman, 1982) D 4 FD+MVD E LD
6 (Zaniolo and Melkanoff, 1981) D 3 FD+MVD E LD+CP
318 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

NF, Class —the considered constraints class, O(?) stores the algorithm
complexity –P(olynomial) or E(xponential)–, and Prop. the normalization
properties —C(onstraint)P(reservation) and L(ossless)D(ecomposition)):
Initially, normalization aimed to provide automatic design of rdb
schemas. At least starting with the 90s (see, for example, Atzeni and de
Antonellis (1993)), more and more researchers are convinced that such an
approach is not adequate and that it is preferable to design db schemas in
a higher, semantic data model that provides an algorithm for translating
then its schemas into the desired relational NFs. In this context, RDM nor-
malization theory only provides milestones and characterizes the targets of
these translating algorithms.

3.15.1.2  The RDM Constraints Implication Problem

Whatever may happen to you was prepared for you from all eternity;
and the implication of causes was from eternity spinning
the thread of your being.
—Marcus Aurelius

3.15.1.2.1  Derivation Rules

Omnis cellula e cellula. (Every cell is derived from another cell.)

—François-Vincent Raspall

Derivation rules was the first used technique for studying RDM constraints
implication problem: Armstrong (1974) introduced a sound and complete
set of inference rules for FDs. Of course that, generally, besides being a
“nice” theoretical result per se, the existence of such a set is very useful
for solving the corresponding implication problem.
Note, however, that, surprisingly, for decades, the existence of such
a sound and complete set of inference rules for any constraint class (the
so-called complete axiomatization of the corresponding class) was consid-
ered more important than the implication problem decidability, although
the existence or nonexistence of complete axiomatizations and the decid-
ability (generally, the complexity) of the corresponding implication prob-
lem are orthogonal.
The
 Quest for Data Independence, Minimal Plausibility, and Formalization 319

For example, for some constraint classes, there is no limited complete

axiomatization (i.e., a complete one in which each rule has to have at most
a given maximum number of antecedents), even if the corresponding impli-
cation problem is decidable: see (Kanellakis et al., 1983) for the unary
inclusion dependencies class and (Casanova et al., 1984) for FD ø IND.
Dually, there are, for example, two classes for which limited com-
plete axiomatization exist, but the corresponding implication problem is
undecidable: the encapsulated implicational dependencies (Chandra et al.
(1981) and Beeri and Vardi (1984a)) and the typed template dependencies
(Sadri and Ullman (1982) and Vardi (1984)).
FDs’ implication problem was “classically” studied in Beeri and Ber-
nstein (1979) and Maier (1980b), and, alternatively, in the propositional
logic framework, in Sagiv et al. (1981), Cosmadakis and Kanellakis
(1985), and Cosmadakis et al. (1986).
A sound and complete set of inference rules for FD ø IND finite impli-
cation was proposed in Mitchell (1983b). A seminal paper on INDs theory
is Casanova et al. (1984), which not only introduces their inference rules,
but it is also proved that FD ø IND finite and unrestricted implications
axiomatization is impossible. Related results are presented in Laver et al.
(1983) and Johnson and Klug (1984) (where finite and unrestricted impli-
cations are generalized, also used in FD ø IND chasing, see below, and for
characterizing inclusion between conjunctive queries).
The undecidability of FD ø IND finite implication was also independently
proved in Mitchell (1983a) and Chandra and Vardi (1985) (which is also pro-
viding an elegant proof of the fact that decidability of FD ø IND unrestricted
implication cannot be generalized, because the finite corresponding implication
is undecidable; the proof uses reduction to the undecidable problems of monoid
words (Post, 1947) and finite monoid words (Gurevich, 1966)).
This negative result, corroborated with the fact that IND implication
problem complexity is PSPACE-complete, lead to restricting to only
unary INDs, for which, unioned with FDs, the decision problem for both
implication types is in PTIME (see, Cosmadakis et al. (1990), where addi-
tional results on unary INDs unioned with other, more general dependency
classes are provided too).
Similar to FD ø IND, MVDs do interact with FDs too: the axiomat-
ization of FD ø MVD is due to Beeri et al. (1977), which also solved the
MVD implication problem, by using inference rules.
320 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

After one decade of unfruitful research, Petrov (1989) proved that

there is no complete axiomatization for JDs. However, such axiomatiza-
tions exist for larger constraint sets that include JDs (e.g., Sciore, 1982),
but their implication problem is not decidable.

3.15.1.2.2  The Chase

Gratitude is the most exquisite form of courtesy.

—Jacques Maritain

The second technique used for studying the RDM constraints implication
problem was the chasing tableaux algorithms (or, simply, chase): algo-
rithms that start with a tableaux (i.e., a table whose instance contains not
only constants, but variables too, thus abstracting several possible (sub)
instances) and a constraint set, try to build a counterexample (of the fact
that every set is included in its closure) tableaux and force it to satisfy all
given constraints, in order to obtain a tableaux which satisfies the given set
of constraints and is as similar as possible with the original one.
Note that a given constraint is implied by the given set of constraints iff
it holds in the obtained tableaux too. Also note that chase’s appeal comes
not only from its very intuitive nature, but also from the fact that it is
somewhat parameterizable with respect to the considered constraint class.
A chase foremother was already prefigured in Aho et al. (1979a); chase
was introduced in Maier et al. (1979) (that developed ideas from Aho et al.
(1979a) and Aho et al. (1979b)), which also uses it both for solving logic
implication problems and tableaux queries, as well as for studying the FD
and MVD implication problem.
Chang and Lee (1973) and Beeri and Vardi (1980) compare the chase
with the paramodulation resolution theorem proofing technique. Beeri and
Vardi (1984b) extends chase use to the more general data dependencies.
Typed INDs are studied using the chase and equational theories in Cos-
madakis and Kanellakis (1986).
Abiteboul et al. (1995) used chase both for JDs and JD ø FD, proving
that they are NP-complete and that the implication problem JDs |— MVD
is NP-hard.
The
 Quest for Data Independence, Minimal Plausibility, and Formalization 321

3.15.2  RELATIONAL LANGUAGES THEORY. SQL, QUEL, QBE

Blessings flow toward gratitude.

—Jonathan Lockwood Huie

As we have already seen, E. F. Codd introduced both RA (Codd, 70) and

RC (Codd, 71a). Final results of RA’s expressive power are due to (Pare-
daens, 1978); similar ones were obtained for RC by Bancilhon (1978).
Codd (71a) proposed the TRC language ALPHA, which has an embed-
ded values domain declarations form that laid the foundation for the RDBM-
Ses one’s. Pirotte (1978) made the distinction between TRC and DRC, while
providing a detailed analysis of various RC forms, including issues occur-
ring when transforming a generic applied calculus into a query language.
One year before RDM’s introduction, Di Paola (1969) proposed the
notion of domain independence, in a broader context. Unfortunately, in
db theory, this notion91 was assimilated very slowly, although undesir-
able expressions whose result might have been the whole domain were
detected as soon as TRC and DRC introduction, which immediately lead
to the notions of type (Codd, 71a) and value domain (Bancilhon, 1978).
Fagin (1982) finally recognized the importance of domain indepen-
dence as a semantic notion (in the context of formulae describing integ-
rity constraints); its undecidability was inferred from previous results (Di
Paola, 1969; Vardi, 1981), whereas domains’ restraint, safety, evaluabil-
ity, permission, and interpretation limitations were considered as suffi-
cient conditions for it.
The equivalence between RA and CR was proved first by Codd (1972);
alternative proofs were then proposed by Ullman (1982) and Maier (1983).
This equivalence determined E. F. Codd to also introduce the notion of com-
pleteness (Codd, 1972); note that, before, for ALPHA already, he felt like
needing more than completeness and, consequently, introduced tuple and
aggregation functions. Completeness was subsequently characterized by
Paredaens (1978) and Bancilhon (1978). Klug (1982) proved the equivalence
between RA and CR, when both are extended with aggregation functions.
A more general completeness notion (CH-completeness) was intro-
duced by Chandra and Harel (1980), called computable queries, which
An expression is domain independent iff it represents a same query, regardless of its subjacent do-
91

mains; a language is domain independent iff all of its expressions are domain independent.
322 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

moves the focus from query results for a given db instance to queries
viewed as functions: a language L is CH-complete iff, for any db scheme
and generic query Q, there exists a L-expression E such that, for any cor-
responding db instance i, Q(i) = E(i).
Moreover, this paper also proposed QL, a canonical CH-complete lan-
guage: informally, QL includes RA extended with variables over relations,
an assignment statement, and a looping one equipped with a test for reach-
ing the empty set, the kernel of today’s extended SQLs.
Related notions over generic queries were studied by Hull (1986).
Note that, as we have already discussed, actual commercial SQL
and Quel are generally extended in the CH-completeness sense92 and/or
embedded in high-level host programming languages93 that provide full
computing power.
Aho and Ullman (1979) proved that transitive closures cannot be
expressed in RA.
Chandra (1988) is an excellent overview of query languages, present-
ing hierarchies of such languages based on expressive power, complexity,
and programming primitives; moreover, it provides an impressive bibliog-
raphy, including mathematical logic previously obtained results (in larger
contexts) that apply to rdb querying as well.
There are three approaches for dealing with nulls in rdb querying:
1. based on derived RA operators: for example, augmentation (unary,
computing a relation over the operand’s attributes, plus other ones,
and whose tuples are obtained from those of the operand’s, but
extending them with nulls) and external (natural) join (or full outer
join) from Gallaire and Minker (1978), and total projection (the
usual projection followed by elimination of all tuples containing
nulls) from Kandzia (1979);
2. based on constraints on nulls: see Imielinski and Lipski (1981,
1983, 1984), (Abiteboul et al., 1987), and (Grahne, 1989);
3. based on the IsNULL predicate, which does not originate from the
db research, but from the RDBMS industry.
SQL stemmed from IBM’s SEQUEL (“Structured English QUEry
Language”), introduced by Chamberlin and Boyce (1974), while work-
92
See, for example, IBM DB2 SQL PL, Oracle PL/SQL, MS T-SQL, etc.
93
See, for example, IBM embedded SQL for Java, C, C++, Cobol, etc., MS VBA and .NET languages, etc.
The
 Quest for Data Independence, Minimal Plausibility, and Formalization 323

ing for the world’s first RDBMS prototype, System R (Astrahan et al.,
1976; Chamberlin et al., 1981). SQL’s formal semantics is provided by
Negri et al. (1991). SQL is the relational both de facto and de jure (ISO/
IEC, 1991; ISO/IEC, 1992) standard language; its O-O extension (called
O2SQL) was proposed in Cattell (1997).
Starting with the SQL-99 standard (ISO/IEC, 2003), which extended
SQL, among others, with recursion, both linear, nonlinear, and mutual
recursion are allowed, but recursive subqueries and aggregation are disal-
lowed. The current standard is (ISO/IEC, 2011).
Lot of good books on SQL programming are available, from those
offered by the RDBMS providers (e.g., Zamil et al. (2004), Lorentz et al.
(2013), Moore et al. (2013)), part of their online documentation, and up
to the commercial ones (e.g., Rockoff (2011), Ben-Gan (2012), Berkowitz
(2008), Murach (2012)).
A similar TRC-type language, Quel, was introduced for the RDBMS
INGRES (Stonebraker et al., 1976).
QBE is due to another IBM researcher, Moshe Zloof (Zloof, 1977).
Ozsoyoglu and Wang (1993) presents all of its important extensions, as
well as all of its flavors. QBE’s exceptional graphic power is exploited
by several RDBMSs: for example, IBM embeds QBE into its DB2 QMF
(Query Management Facility), Microsoft provides it as its queries’ Design
View, in both Access and SQL Server, and also in Excel, as MS Query,
PostgreSQL offers its OBELisQ, etc.
Selinger et al. (1979) describes SystemR’s query evaluation optimi-
zations. Ullman (1982) presents a widely used today heuristic optimiza-
tion algorithm in six steps. An exhaustive discussion of query evaluation
optimizations is presented in Graefe (1993). Sun et al. (1993) contains
both a proposal of an estimation technique based on series approximating
instances values distribution and the regressions analysis for estimating
queries with joins results cardinality, as well as an impressive list of refer-
ences dealing with query evaluation plans’ cost estimations.
For a couple of decades, db instances modifications were almost
neglected, as it was believed that they may be trivially formalized with
three simple operators (insert, update, and delete of tuples satisfying desired
criteria) that might be reducible to basic set operators. Abiteboul and Vianu
(1988a, 1988b, 1989, 1990) proved that there are, however, depth prob-
324 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

lems in this direction too, which opened a new study field still active. An
overview of the fundamentals in this area is provided by (Abiteboul, 1988).

3.15.3 MISCELLANEA

Philosophy of science without history of science is empty;

history of science without philosophy of science is blind.
—Imre Lakatos

This chapter is mainly an abstract of Mancas (2007) and Mancas (2002a).

There is lot of algorithms for translating E-RDs into relational schemas
(e.g., see the one presented in Levene and Loizou (1999), which is also
citing other 7 ones). A1–7 (presented in Subsection 3.3.1), which is also
taking into considerations associated restriction sets and informal descrip-
tions, as well as the RDM model of the E-RDM come from Mancas (2001).
There are also other reverse engineering algorithms for translating rela-
tional schemas into E-RDs (e.g., Chiang et al. (1994) presents a methodol-
ogy for extracting an extended E-R model from a rdb, through a combination
of data schema and instance analysis, which derives a semantically richer
and more comprehensible for maintenance and design purposes model than
the original rdb). The simple REA1–2 (presented in Section 3.6), which is,
however, taking into considerations associated restriction sets and informal
descriptions too, comes also from Mancas (2001).
The keys discovery assistance algorithm A7/8–3 (see Section 3.7),
together with proposition 3.1, were initially proposed in the framework
of the (E)MDM, first for relationship structural keys (see second chapter
of the second volume) (Mancas and Dragomir, 2003) and then for all pos-
sible keys (see A3 and A3′ in the second chapter of the second volume)
(Mancas and Crasovschi, 2003). Its current English version for the RDM
comes from (Mancas, 2008).
Proposition 3.0, on which A3, A3’, and A7/8–3 heavily rely, was pub-
lished earlier (Mancas, 1990). Its simplified proof from (Mancas and Cra-
sovschi, 2003) was inspired by the one from Lubell (1966).
The RDBMS metacatalog kernel scheme and the corresponding relational
metamodel of the RDM (see Section 3.7) were published first in Mancas
(2002b); the E-R data metamodel of the RDM comes from Mancas (2001).
The
 Quest for Data Independence, Minimal Plausibility, and Formalization 325

Note that the relational one from (Mancas, 2002b) is slightly simpler
than this one (coming from Mancas (2008)), as it was in fact designed for
MatBase, which only uses single surrogate primary keys and single for-
eign keys always referencing primary keys.
DKRINF and the best practice rules (see Section 3.11) come from
Mancas (2001).
Definition 3.3 and proposition 3.3, as well as Exercises 3.2, 3.40, 3.41
and 3.42 are due to Val Tannen (Breazu and Mancas, 1980, 1981, and
1982).
The key propagation principle formalization (proposition 3.2), comes
from Mancas (2001) and was first published outside Romania in Mancas
and Crasovschi (2003), in the (E)MDM framework.
For the theory of NP-completeness, see, for example, Garey and John-
son (1979). For the complexity theory, see, for example, the excellent
(Calude, 1988).
Exercises 3.0, 3.1, 3.6 to 3.17, 3.26 to 3.30, 3.39, 3.43, 3.68, and 3.71 to
3.87 are original (most of them are from Mancas (2001), Mancas (2002),
Mancas (2007), and Mancas (2008)). All other exercises are from other
books (mainly from Abiteboul et al. (1995) and Atzeni and de Antonellis
(1993)).
Cons on relational db design (see also, for example, Atzeni and de
Antonellis (1993)) were first presented in Kent (1979). The ones from
Subsection 3.13.2 come also from Mancas (1985, 1997, 2007).
For ISO 3166 country (and their subdivisions) codes international stan-
dard, see http://www.iso.org/iso/home/standards/country_codes.htm.
All URLs mentioned in this section were last accessed on
August 15th, 2014.

KEYWORDS

•• 2nd Normal Form (2NF)

•• 3rd Normal Form (3NF)
•• 4th Normal Form (4NF)
•• 5th Normal Form (5NF)
•• ACID
326 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

•• acyclicity constraint
•• aggregation function
•• ALL
•• ALTER TABLE
•• anti join
•• ANY
•• arity
•• AS
•• attribute
•• attribute concatenation
•• attribute domain
•• attribute name
•• attribute product
•• augmentation
•• autonumber
•• AVG
•• BEGIN TRANS
•• Boyce-Codd Normal Form (BCNF, 3.5NF)
•• candidate key
•• cardinal
•• Cartesian product
•• CH-completeness
•• chase
•• CHECK
•• check constraint
•• column
•• COMMIT
•• complementary property
•• complementation
•• complete axiomatization
•• complete join
•• completeness
•• composed key
The
 Quest for Data Independence, Minimal Plausibility, and Formalization 327

•• compulsory attribute
•• computable queries
•• computed attribute
•• computed relation
•• concatenated key
•• constraint
•• constraint name
•• constraint preservation
•• constraint representation
•• COUNT
•• CREATE TABLE
•• cross tabulation
•• cursor
•• dangling pointer
•• data definition language (DDL)
•• data dependency
•• data manipulation anomalies
•• data manipulation language (DML)
•• data set
•• decidability
•• decomposition
•• DELETE
•• delete anomaly
•• dependency
•• derivation rules
•• dirty null
•• DISTINCT
•• DISTINCTROW
•• Division
•• Domain-Key Normal Form (DKNF)
•• Domain-Key Referential Integrity Normal Form (DKRINF)
•• domain constraint
•• domain independence
328 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

•• domain relational calculus (DRC)

•• DROP CONSTRAINT
•• DROP SCHEMA
•• DROP TABLE
•• dynamic SQL
•• dynamic subquery
•• embedded SQL
•• equalities constraint
•• equijoin
•• existence constraint
•• EXISTS
•• expressive power
•• extended SQL
•• field
•• finite implication
•• First Normal Form (1NF)
•• FOREIGN KEY
•• foreign key
•• FROM
•• full join
•• functional dependency
•• function kernel (ker)
•• fundamental relation
•• GRANT
•• GROUP BY
•• HAVING
•• header
•• IN
•• inner join
•• INSERT
•• insert anomaly
•• instance
•• intersection
The
 Quest for Data Independence, Minimal Plausibility, and Formalization 329

•• IsNULL predicate
•• join
•• join dependency
•• kernel equivalence relation
•• key
•• key constraint
•• Key Propagation Principle (KPP)
•• left join
•• LIMIT
•• limited complete axiomatization
•• lossless decomposition
•• mandatory attribute
•• marked null
•• MAX
•• metacatalog
•• metadata catalog
•• MIN
•• minimally one-to-one
•• minimally uniqueness
•• multivalued dependency
•• natural join
•• nested relation
•• no information null
•• non-relational constraint
•• non prime attribute
•• normalization
•• normalization problem
•• not applicable null
•• NOT EXISTS
•• NOT IN
•• NOT NULL
•• not null
•• not null constraint
330 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

•• null value
•• object constraint
•• ORDER BY
•• outer join
•• partial functional dependency
•• primary key
•• prime attribute
•• primitive constraint
•• projection
•• Projection-Join Normal Form (PJNF)
•• QBE
•• QL
•• Quel
•• query
•• quotient set
•• range
•• rdb instance
•• rdb name
•• rdb scheme
•• record set
•• referential integrity
•• relation
•• relational algebra (RA)
•• relational calculus (RC)
•• relational completeness
•• relational constraint
•• relational db (rdb)
•• relation name
•• renaming
•• required attribute
•• REVOKE
•• right join
•• ROLLBACK
The
 Quest for Data Independence, Minimal Plausibility, and Formalization 331

•• row
•• scheme
•• SELECT
•• Selection
•• semantic key
•• semi join
•• simple key
•• SOME
•• soundness
•• SQL
•• static SQL
•• static subquery
•• stored function
•• stored procedure
•• subquery
•• SUM
•• superkey
•• surrogate key
•• syntactic key
•• synthesis
•• table
•• tableaux
•• table name
•• temporarily unknown null
•• temporary relation
•• theta join
•• TOP
•• total attribute
•• totality constraint
•• total projection
•• transaction
•• transitive closure
•• transitive functional dependency
332 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

•• trigger
•• tuple
•• tuple constraint
•• tuple relational calculus (TRC)
•• tuples generating constraint
•• typed inclusion constraint
•• type domain
•• undecidability
•• union
•• UNION
•• UNION ALL
•• UNIQUE
•• unique attribute
•• uniqueness constraint
•• universal relation
•• UPDATE
•• update anomaly
•• value domain
•• view
•• WHERE

REFERENCES

I have a lot of cultural references that have amassed in my brain like

shrapnel over the years that are meaningful to me.
—John Hodgman

Abiteboul, S. (1988). Updates, a new frontier. ICDT’88 2nd Intl. Conf. on DB Theory,
1–18, Springer-Verlag, Berlin, Germany.
Abiteboul, S., Hull, R., Vianu, V. (1995). Foundations of Databases; Addison-Wesley:
Reading, MA.
Abiteboul, S., Kanellakis, P. C., Grahne, G. (1987). On the representation and querying of
possible worlds. ACM SIGMOD Intl. Conf. on Manag. of Data, 34–48.
Abiteboul, S., Vianu, V. (1988a). Equivalence and optimization of relational transactions.
JACM, 35(1), 70–120.
The
 Quest for Data Independence, Minimal Plausibility, and Formalization 333

Abiteboul, S., Vianu, V. (1988b). Procedural and declarative database update languages.
7th ACM SIGACT SIGMOD SIGART Symp. Princ. DB Syst., 240–250.
Abiteboul, S., Vianu, V. (1989). A transaction-based approach to relational database speci-
fication. JACM, 36(4), 758–789.
Abiteboul, S., Vianu, V. (1990). Procedural languages for database queries and updates. J.
Comp. Syst. Sci., 41(2), 181–229.
Aho, A. V., Beeri C., Ullman J. D. (1979). The theory of joins in relational databases. ACM
TODS 4(3), 297–314.
Aho, A. V., Sagiv, Y., Ullman, J. D. (1979a). Efficient optimization of a class of relational
expressions. ACM TODS 4(3), 435–454.
Aho, A. V., Sagiv, Y., Ullman, J. D. (1979b). Equivalence of relational expressions. J.
Comp. Syst. Sci., 8(2), 218–246.
Aho, A. V., Ullman, J. D. (1979). Universality of data retrieval languages. 6th ACM Symp.
on Princ. of Progr. Lang., 110–117.
Allison, C., Berkowitz, N. (2008). SQL for Microsoft Access, 2nd edition: Wordware Pub-
lishing, Inc. Plano, TX.
ANSI/X3/SPARC Study Group on Database Management Systems. (1975). Interim Report
75–02–08. FDT-Bulletin ACM SIGMOD, 7(2).
Armstrong, W. W. (1974). Dependency structure of database relationships. IFIP Congress,
580–583.
Astrahan, M. M., et al. (1976). System R: A Relational Approach to Database Manage-
ment. ACM TODS, 1(2), 97–137.
Atzeni, P., de Antonellis, V. (1993). Relational Database Theory. Benjamin/ Cummings:
Redwood, CA.
Atzeni, P., Morfuni, N. M. (1984). Functional dependencies in relations with null values.
Inf. Processing Letters, 18(4), 233–238.
Atzeni, P., Morfuni, N. M. (1986). Functional dependencies and constraints on null values
in database relations. Information and Control, 70(1), 1–31.
Bancilhon, F. (1978). On the completeness of query languages for relational databases.
LNCS Math. Found. of Comp. Sci., 64, 112–124, Springer-Verlag: Berlin, Germany.
Beeri, C., Bernstein, P. A. (1979). Computational problems related to the design of normal
form relational schemas. ACM TODS 4(1), 30–59.
Beeri, C., Fagin, R., Howard, J. H. (1977). A complete axiomatization for functional and
multivalued dependencies. ACM SIGMOD Intl. Symp. Manag. Data, 47–61.
Beeri, C., Honeyman, P. (1981). Preserving functional dependencies. SIAM Journal on
Computing, 10(3), 647–656.
Beeri, C., Rissanen, J. (1980). Faithful representation of relational database schemata. Re-
port RJ 2722. IBM Research. San Jose.
Beeri, C., Vardi, M. Y. (1980). A proof procedure for data dependencies (preliminary re-
port). Tech. Rep., Hebrew Univ.: Jerusalem, Israel.
Beeri, C., Vardi, M. Y. (1984a). Formal systems for tuple and equality-generating depen-
dencies. SIAM J. Computing, 13(1), 76–98.
Beeri, C., Vardi, M. Y. (1984b). A proof procedure for data dependencies. JACM 31(4),
718–741.
Ben-Gan, I. (2012). Microsoft SQL Server 2012. T-SQL Fundamentals (Developer Refer-
ence). O’Reilly Media, Inc.: Sebastopol, CA.
334 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

Bernstein, P. A. (1976). Synthesizing third normal form relations from functional depen-
dencies. ACM TODS, 1(4), 277–298.
Bernstein, P. A., Goodman N. (1980). What does Boyce-Codd form do?. 6th Intl. Conf. on
VLDB, 245–259, Montreal, Canada.
Biskup, J., Dayal, U., Bernstein, P. A. (1979). Synthesizing independent database schemes.
ACM SIGMOD Intl. Conf. on Manag. of Data, 143–151.
Breazu, V., Mancas, C. (1980). On a Functional Data Model in the database mathematical
theory. Comparison to the Relational Data Model (in Romanian). Unpublished com-
munication at the 6th Symposium on Informatics for Management CONDINF’80,
Cluj-Napoca, Romania.
Breazu, V., Mancas, C. (1981). On the description power of a Functional Database Model.
Proc. 4th Intl. Conf. on Control Systems and Computer Science, 4:223–226, Po-
litehnica University, Bucharest, Romania.
Breazu, V., Mancas, C. (1982). Normal forms for schemas with functional dependencies in the
Relational Database Model (in Romanian). Proc. 12ve Conf. on Electronics, Telecom-
munications and Computers, 152–1566, Politehnica University, Bucharest, Romania.
Calude, C. (1988). Theories of Computational Complexity. North Holland: Amsterdam,
Holland.
Casanova, M. A. (1981). The theory of functional and subset dependencies over relational
expressions. Tech. Rep., 3/81, Rio de Janeiro, Brazilia.
Casanova, M. A., Fagin R., Papadimitriou C. H. (1984). Inclusion dependencies and their
interaction with functional dependencies. J. Comp. Syst. Sci., 28(1), 29–59.
Casanova, M. A., Vidal, V. M. P. (1983). Towards a sound view integration technology.
ACM SIGACT SIGMOD Symp. Princ. DB Syst., 36–47.
Cattell, R. G. G., ed. (1994). The Object Database Standard: ODMB-93. Morgan Kaufman:
Los Altos, CA.
Chamberlin, D. D., Boyce, R. F. (1974). SEQUEL—A Structured English QUEry Lan-
guage. SIGMOD Workshop, 1, 249–264.
Chamberlin, D.D., et al. (1981). A History and Evaluation of System R. CACM, 24(10),
632–646.
Chan, E. P. F. (1989). A design theory for solving the anomalies problem. SIAM J. Com-
puting, 18(3), 429–448.
Chandra, A. K. (1988). Theory of database queries. 7th ACM SIGACT SIGMOD SIGART
Symp. Princ. DB Syst., 1–9.
Chandra, A. K., Harel, D. (1980). Computable queries for relational databases. J. Comp.
Syst. Sci., 21, 333–347.
Chandra, A. K., Lewis, H. R., Makowsky, J. A. (1981). Embedded implicational dependen-
cies and their inference problem. 13th ACM SIGACT Symp. on Theory of Compu-
ting, 342–354.
Chandra, A. K., Vardi, M. Y. (1985). The implication problem for functional and inclusion
dependencies is undecidable. SIAM J. Computing, 14(3), 671–677.
Chang, C. L., Lee, R. M. (1973). Symbolic Logic and Mechanical Theorem Proving. Aca-
demic Press: New York, NY.
Chiang, R. H. L., Barron, T. M., Storey, V. C. (1994). Reverse engineering of relational da-
tabases: Extraction of an EER model from a relational database. Data & Knowledge
Engineering 12(2): 107–142, Elsevier B.V.
The
 Quest for Data Independence, Minimal Plausibility, and Formalization 335

Churcher, C. (2012). Beginning Database Design: From Novice to Professional, 2nd ed.,
Apress Media LLC: New York, NY.
Codd, E. F. (1970). A relational model for large shared data banks. CACM, 13(6), 377–387.
Codd, E. F. (1971a). A database sublanguage founded on the relational calculus. SIGFI-
DET’71, Proceedings of the ACM SIGFIDET (now SIGMOD) Workshop on Data
Description, Access, and Control, San Diego, CA, November 11–12, (1971); ACM:
New York, NY, 35–68.
Codd, E. F. (1971b). Further normalization of the data base relational model; Research
Report RJ909; IBM: San Jose, CA.
Codd, E. F. (1972). Relational completeness of data base sublanguages; Research Report
RJ987; IBM: San Jose, CA.
Codd, E. F. (1974). Recent investigations into relational database systems; Research Re-
port RJ1385; IBM: San Jose, CA.
Codd, E. F. (1975). Understanding Relations. FDT (ACM SIGMOD Records), Vol. 7, No.
3–4, 23–28.
Codd, E. F. (1979). Extending the database relational model to capture more meaning.
ACM TODS, 4(4), 397–434.
Cosmadakis, S., Kanellakis, P. C. (1985). Equational theories and database constraints.
ACM SIGACT Symp. on the Theory of Computing, 73–284.
Cosmadakis, S., Kanellakis, P. C. (1986). Functional and inclusion dependencies: A graph
theoretic approach. Adv. in Computing Res., 3, 164–185, JAI Press.
Cosmadakis, S., Kanellakis, P. C., Spyratos, N. (1986). Partition semantics for relations. J.
Comp. Syst. Sci., 32(2), 203–233.
Cosmadakis, S., Kanellakis, P. C., Vardi, M. Y. (1990). Polynomial-time implication prob-
lems for unary inclusion dependencies. JACM, 37(1), 15–46.
Date, C. J. (2003). An Introduction to Database Systems, 8th ed., Addison-Wesley: Read-
ing, MA.
Date, C. J. (2011). SQL and Relational Theory: How to Write Accurate SQL Code, 2nd ed.,
Theories in Practice; O’Reilly Media, Inc.: Sebastopol, CA.
Date, C. J. (2012). Database Design & Relational Theory: Normal Forms and All That
Jazz; Theories in Practice; O’Reilly Media, Inc.: Sebastopol, CA.
Date, C. J. (2013). Relational Theory for Computer Professionals: What Relational Data-
bases are Really All About; Theories in Practice; O’Reilly Media, Inc.: Sebastopol,
CA.
Date, C. J., Fagin, R. (1992). Simple conditions for guaranteeing higher normal forms in
relational databases. ACM TODS 17(3), 465–476.
Delobel, C. (1978). Normalization and hierarchical dependencies in the relational data
model. ACM TODS 3(3), 201–222.
Di Paola, R. (1969). The recursive unsolvability of the decision problem for the class of
definite formulas. JACM, 16(2), 324–327.
Fagin, R. (1977). Multivalued dependencies and a new normal form for relational data-
bases. ACM TODS 2(3), 226–278.
Fagin, R. (1979). Normal forms and relational database operators. ACM SIGMOD Intl.
Conf. on Manag. of Data, 123–134.
Fagin, R. (1981). A normal form for relational databases that is based on domains and keys.
ACM TODS 6(3), 387–415.
336 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

Fagin, R. (1982). Horn clauses and database dependencies. JACM, 29(4), 952–983.
Fagin, R., Vardi, M. Y. (1986). The theory of data dependencies: A survey. Math. of Infor-
mation Processing: Proc. of Symp. In Applied Math., 34, 19–71, American Math.
Soc., Providence, RI.
Gallaire, H., Minker, J. (1978). Logic and Databases. Plenum Press: New York, NY.
Garcia-Molina, H., Ullman, J.D., Widom, J. (2014). Database Systems: The Complete
Book, 2nd ed., Pearson Education Ltd.: Harlow, U.K., (Pearson New International
Edition)
Garey, M. R., Johnson, D. S. Victor Klee. ed. (1979). Computers and Intractability: A
Guide to the Theory of NP-Completeness. A Series of Books in the Mathematical
Sciences. W. H. Freeman and Co.
Goldstein, B. S. (1981). Constraints on null values in relational databases. 7th Intl. Conf.
on VLDB, 101–111.
Graefe, G. (1993). Query evaluation techniques for large databases. ACM Computing Sur-
veys, 25(2), 73–170.
Grahne, G. (1989). Horn tables—an efficient tool for handling incomplete information in
databases. 8th ACM SIGACT SIGMOD SIGART Symp. Princ. DB Syst., 75–82.
Grant, J. (1977). Null values in a relational database. Inf. Processing Letters, 6(5), 156–159.
Groff, J. R. (1990). Using SQL. Osborne McGraw-Hill, New York, NY.
Gurevich, Y. (1966). The word problem for certain classes of semigroups. Algebra and
Logic, 5, 25–35.
Halpin, T., Morgan, T. (2008). Information Modeling and Relational Databases, 2nd Edi-
tion. Morgan Kauffman: Burlington, MA.
Heath, I. (1971). Unacceptable File Operations in a Relational Database. SIGFIDET’71,
Proceedings of the ACM SIGFIDET (now SIGMOD) Workshop on Data Descrip-
tion, Access, and Control, San Diego, CA, November 11–12, ACM: New York, NY,
19–33.
Hernandez, M. J. (2013). Database Design for Mere Mortals: A Hands-on Guide to Rela-
tional Database Design, 3rd ed., Addison-Wesley: Reading, MA.
Honeyman, P. (1982). Testing satisfaction of functional dependencies. JACM, 29(3), 668–
677.
Hull, R. B. (1986). Relative information capacity of simple relational schemata. SIAM J.
Computing, 15(3), 856–886.
Imielinski, T., Lipski, W. (1981). On representing incomplete information in relational da-
tabases. 7th Intl. Conf. on VLDB, 388–397, Cannes, France.
Imielinski, T., Lipski, W. (1983). Incomplete information and dependencies in relational
databases. ACM SIGMOD Intl. Conf. on Manag. of Data, 178–184.
Imielinski, T., Lipski, W. (1984). Incomplete information in relational databases. JACM,
761–791.
ISO/IEC. Database language SQL (SQL-2011). 9075: 2011, 2011.
ISO/IEC. Database language SQL (SQL-99). 9075–2: 1999, 2003.
ISO/IEC. Database language SQL (SQL3). JTC1/SC21 N 6931, 1992.
ISO/IEC. Database language SQL. JTC1/SC21 N 5739, 1991.
Johnson, D. S., Klug, A. (1984). Testing containment of conjunctive queries under func-
tional and inclusion dependencies. J. Comp. Syst. Sci., 28, 167–189.
The
 Quest for Data Independence, Minimal Plausibility, and Formalization 337

Kandzia, P., Klein, H. (1979). On equivalence of relational databases in connection with

normalization. Workshop on Formal Bases for DB, ONERA-CERT: Toulouse,
France.
Kanellakis, P. C. (1991). Elements of relational database theory. Handbook of Theoretical
Comp. Sci., 1074–1156, Elsevier: Amsterdam, Holland.
Kanellakis, P. C., Cosmadakis, S., Vardi, M. Y. (1983). Unary inclusion dependencies have
polynomial-time inference problems. 15th ACM SIGACT Symp. on Theory of Com-
puting, 264–277.
Kent, W. (1979). Limitations of record-based information models. ACM TODS 4(1).
Klug, A. (1982). Equivalence of relational algebra and relational calculus query languages
having aggregate functions. JACM, 29(3), 699–717.
Korth, H. F., Ullman, J. D. (1980). System/U: A system based on the universal relation as-
sumption. Stonybrook, NY.
Laver, K., Mendelzon, A. O., Graham, M. H. (1983). Functional dependencies on cyclic
database schemes. ACM SIGMOD Intl. Symp. Manag. Data, 79–91.
LeDoux, C. H., Parker, D. S. (1982). Reflections on Boyce-Codd normal form. 8th Intl.
Conf. on VLDB, 131–141, Mexico City, Mexico.
Levene, M., Loizou, G. (1999). A Guided Tour of Relational Databases and Beyond:
Springer Verlag, London, U.K.
Lien, Y. E. (1982). On the equivalence of database models. JACM, 29(2), 333–362.
Lightstone, S. S., Teorey, T. J., Nadeau, T., Jagadish, H. V. (2011). Database Modeling and
Design: Logical Design, 5th ed., Data Management Systems; Morgan Kaufmann:
Burlington, MA.
Lipski, W. (1979). On semantic issues connected with incomplete information databases.
ACM TODS 4(3), 262–296.
Lipski, W. (1981). On databases with incomplete information. JACM, 28(1), 41–70.
Lorentz, D., Roeser, M. B., et al. (2013). SQL Language Reference 11 g Release 2: Oracle
Corp.
Lubell, D. (1966). A short proof of Sperner’s lemma. J. Combinatorial Theory, 1, 299.
Maier, D. (1980a). Discarding the universal relation assumption: Preliminary report. Sto-
nybrook, NY.
Maier, D. (1983). The Theory of Relational Databases. Computer Science Press: Rock-
ville, MD.
Maier, D., Mendelzon, A. O., Sagiv Y. (1979). Testing implications of data dependencies.
ACM TODS 4(4), 455–468.
Maier D. (1980b). Minimum covers in the relational database model. JACM, 27(4), 664–
674.
Mancas, C. (1985). Some db design counterexamples proving that data modeling should
not be approached relationally. Proc. INFO IAŞI’85, 303–313, A.I. Cuza University,
Iaşi, Romania (in Romanian).
Mancas, C. (1990). A Deeper Insight into the Mathematical Data Model. 13th Intl. Seminar
on DBMS, ISDBMS, 122–134, Mamaia, Romania.
Mancas, C. (1997). Conceptual Data Modeling. PhD Thesis: Politehnica University, Bu-
charest, Romania (in Romanian).
Mancas, C. (2001). Databases for Undergraduates Lecture Notes: Computer Science
Dept., Ovidius State University, Constanta, Romania.
338 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

Mancas, C. (2002a). Programming in SQL ANSI-92 with applications in MS JetSQL 4:

Ovidius University Press, Constanta, Romania (in Romanian).
Mancas, C. (2002b). On Modeling the Relational Domain-Key Normal Form Using an
Elementary Mathematical Data Model. Proc. IASTED SEA 2002 MIT Conf. on Soft-
ware Eng. and App., 767–772, Acta Press, Cambridge, MA.
Mancas, C. (2007). Theoretical foundations of the Relational Data Model: Ovidius Univer-
sity Press, Constanta, Romania (in Romanian).
Mancas, C., Crasovschi L. (2003). An Optimal Algorithm for Computer-Aided Design of
Key Type Constraints. Proc. 1st Balkan Inf. Techn. BIT 2003 Conf., 574–584, Aris-
totle University Press, Thessaloniki, Greece.
Mancas, C., Dragomir, S., Crasovschi, L. (2003). On modeling First Order Predicate Cal-
culus using the Elementary Mathematical Data Model in MatBase DBMS. Proc.
IASTED AI 2003 MIT Conf. on Applied Informatics, 1197–1202, Acta Press, Inns-
bruck, Austria.
Mancas, C., Dragomir S. (2003). An Optimal Algorithm for Structural Keys Design. Proc.
IASTED SEA 2003 Conf. on Software Eng. and App., 328–334, Acta Press, Marina
del Rey, CA.
Mancas C. (2008). Database Management Systems for Undergraduates Lecture Notes: En-
gineering in Foreign Languages Dept., Politehnica University, Bucharest, Romania.
Mitchell, J. C. (1983a). The implication problem for functional and inclusion dependen-
cies. Information and Control, 56(1), 154–173.
Mitchell, J. C. (1983b). Inference rules for functional and inclusion dependencies. ACM
SIGACT SIGMOD Symp. Princ. DB Syst., 58–69.
Moore, S., et al. (2013). PL/SQL Language Reference 11 g Release 2: Oracle Corp.
Murach, J. (2012). Murach’s MySQL. Mike Murach & Associates, Inc.: Fresno, CA.
Negri, M., Pelagatti, S., Sbattella, L. (1991). Formal semantics of SQL queries. ACM
TODS 16(3), 513–535.
Nicolas, J.-M. (1978). First order logic formalization for functional, multivalued, and mu-
tual dependencies. ACM SIGMOD Intl. Symp. Manag. Data, 40–46.
Nicolas, J.-M. (1982). Logic for improving integrity checking in relational databases. Acta
Informatica, 18(3), 227–253.
Orlowska, M., Zhan, Y. (1992). Understanding the 5th normal form (5NF). Australian
Computer Science Communications 14(1), 631–639.
Ozsoyoglu, G., Wang, H. (1993). A survey of QBE languages. Computer, 26.
Paredaens, J. (1978). On the expressive power of the relational algebra. Inf. Processing
Letters, 7(2).
Petrov, S. V. (1989). Finite axiomatization of languages for representation of system prop-
erties. Inf. Sci., 47, 339–372.
Pirotte, A. (1978). High-level database languages. Logic and Databases, 409–435, Plenum
Press: New York, NY.
Post, E. L. (1947). Recursive unsolvability of a problem of Thue. J. of Symb. Logic, 12,
1–11.
Rissanen, J. (1977). Independent components of relations. ACM TODS, 2(4), 317–325.
Rissanen, J. (1979). Theory of joins for relational databases—a tutorial survey. LNCS
Math. Found. of Comp. Sci., 64, 537–551, Springer-Verlag, Berlin, Germany.
Rockoff, L. (2011). The Language of SQL: Course Technology, Boston, MA.
The
 Quest for Data Independence, Minimal Plausibility, and Formalization 339

Sadri, F., Ullman, J. D. (1982). Template dependencies: A large class of dependencies in

the relational model and its complete axiomatization. JACM, 29(2), 363–372.
Sagiv, Y. (1981). Can we use the universal instance assumption without using nulls? ACM
SIGMOD Intl. Conf. Manag. Data, 108–120.
Sagiv, Y., Delobel, C., Parker, D. S., Fagin, R. (1981). An equivalence between relational
database dependencies and a fragment of propositional logic. JACM, 28, 435–453.
Sciore, E. A complete axiomatization of full join dependencies. JACM, 29(2), 373–393,
(1982).
Sciore, E. The Universal Instance and Database Design. Ph.D. dissertation, Princeton
Univ. (Dept. Of EECS), (1980).
Selinger, P., Astrahan, M. M., Chamberlin, D. D., Lorie, R. A., Price, T. G. (1979). Ac-
cess path selection in a relational database management system. ACM SIGMOD Intl.
Symp. on Manag. Data, 23–34.
Smith, J. M. (1978). A normal form for abstract syntax. 4th Intl. Conf. on VLDB, 156–162,
Berlin, Germany.
Stonebraker, M., Wong, E. A., Kreps, P., Held, G. (1976). The design and implementation
of Ingres. ACM TODS, 1(3), 189–222.
Sun, W., Ling, Y., Rishe, N., Deng, Y. (1993). An instant and accurate size estimation
method for joins and selection in a retrieval-intensive environment. ACM SIGMOD
Intl. Symp. Manag. Data, 79–88.
Thalheim, B. (1991). Dependencies in Relational Databases. Teubner Verlagsgesellschaft:
Sttutgart, Germany.
Tsou, D.-M., Fischer, P. C. (1982). Decomposition of a relation scheme into Boyce-Codd
normal form. SIGACT News, 14(3), 23–29.
Ullman, J. D. (1982). Principles of Database Systems, 2nd ed. Computer Science Press:
Rockville, MD.
Ullman, J. D. (1988). Principles of Database and Knowledge-Based Systems—Volume 1:
Classical Database Systems. Computer Science Press: Rockville, MD.
Ullman, J. D. (1989). Principles of Database and Knowledge-Based Systems—Volume 2:
The New Technologies. Computer Science Press: Rockville, MD.
Ullman, J. D., Widom, J. (2007). A First Course in Database Systems, 3rd ed. Prentice
Hall: Upper Saddle River, NJ.
Vardi, M. Y. (1981). The decision problem for database dependencies. Inf. Processing Let-
ters, 12(5), 251–254.
Vardi, M. Y. (1984). The implication and finite implication problems for typed template
dependencies. J. Comp. Syst. Sci., 28(1), 3–28.
Vardi, M. Y. (1987). Fundamentals of dependence theory. Trends in Theoretical Comp. Sci.,
171–224, Computer Science Press: Rockville, MD.
Vassiliou, Y. (1979). Null values in database management: A denotational semantics ap-
proach. ACM SIGMOD Intl. Conf. Manag. Data, 162–169.
Vassiliou, Y. (1980a). Functional dependencies and incomplete information. 6th Intl. Conf.
on VLDB, 260–269, Montreal, Canada.
Vassiliou, Y. (1980b). A Formal Treatment of Imperfect Information in Database Manage-
ment. Ph.D. thesis (2nd Ed.): University of Toronto.
Wong, E. (1982). A statistical approach to incomplete information in database systems.
ACM TODS, 7(3), 470–488.
340 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

Yannakakis, M., Papadimitriou, C. H. (1982). Algebraic dependencies. J. Comp. Syst. Sci.,

21(1), 2–41.
Zamil, J., et al. (2004). DB2 SQL PL: Essential Guide for DB2 UDB on Linux, Unix, Win-
dows, i5/OS, and z/OS, 2nd edition: IBM Press, Indianapolis, IN.
Zaniolo, C. (1976). Analysis and Design of Relational Schemata for Database Systems.
Ph.D. thesis, Eng-7669, UCLA, Los Angeles, U.S.A.
Zaniolo, C. (1977). Relational views in a database system; support for queries. IEEE Intl.
Conf. on Comp. Software App., 267–275.
Zaniolo, C. (1982). A new normal form for the design of relational database schemas. ACM
TODS 7(3), 489–499.
Zaniolo, C., Melkanoff, M. A. (1981). On the design of relational database schemata. ACM
TODS, 6(1), 1–47.
Zloof, M. M. (1977). Query-by-example: A Database Language. IBM Syst. Journal, 16(4),
324–343.
CHAPTER 4

RELATIONAL SCHEMAS
IMPLEMENTATION AND REVERSE
ENGINEERING

If God were romantic, He would have created other gods.

As He is classic, He created the world.
—Lucian Blaga

CONTENTS

4.1 The Algorithm for Translating Relational Schemas into

SQL DDL ANSI-92 Scripts (A8)................................................... 347
4.2 Relevant Differences Between IBM DB2, Oracle Database
and MySQL, Microsoft SQL Server and Access............................ 348
4.3 Case Study: Implementing the Public Library rdb into
DB2, Oracle, MySQL, SQL Server, and Access............................ 425
4.4 The Reverse Engineering Algorithm for Translating
Access 2013 rdb Schemas into SQL ANSI DDL Scripts
(REA2013A0), A Member of REAF0’........................................... 473
4.5 The Algorithms for Translating E-R Data Models into rdbs
and Associated Non-Relational Constraint Sets (AF1-8).............. 488
4.6 The Reverse Engineering Family of Algorithms for
Translating rdb Schemas into E-R Data Models (REAF0-2)........ 490
4.7 Case Study: Reverse Engineering of An Access Stocks
db Scheme into both An ANSI Standard SQL DDL Script
and An E-R Data Model................................................................ 497
4.8 Best Practice Rules....................................................................... 529
4.9 The Math Behind the Algorithms Presented In This Chapter....... 534
342 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

4.10 Conclusion.................................................................................. 534
4.11 Exercises..................................................................................... 535
4.12 Past and Present.......................................................................... 577
Keywords............................................................................................... 578
References.............................................................................................. 579

After studying this chapter you should be able to:

Ø Concisely and precisely define, and exemplify the following key
terms: database (creation, reverse engineering), relational scheme,
tablespace, table (fundamental, temporary –both local and global),
column (prime, non-prime, computed, virtual), key (uniqueness,
primary, foreign, superkey), constraint ((co-)domain, not null, ref-
erential integrity, minimal uniqueness, check, tuple, relational, non-
relational), index file (unique, not unique), E-R data model, trigger,
data (event, compulsory, range, instance), data type (SQL ANSI
standard, DB2 10.5, Oracle 12c, MySQL 5.7, SQL Server 2014,
Access 2013, built-in, extended, user defined, strongly/weakly
typed, geospatial), identifier (ordinary, delimited), type hierarchies
(root, super, sub, target, and recursive –direct and indirect), number
(infinity, NaN –both quiet and signaling–, subnormal, sequence),
array (ordinary, associative), view (materialized, query), identifier
(ordinary, delimited, global transaction –GTID).
Ø Correctly choose the data types you need for any RDBMS ver-
sion considered in this book (IBM DB2, Oracle Database and
MySQL, Microsoft SQL Server and Access) among those that
they offer (ANCHOR, ANYDATA, ANYDATASET, ANYTYPE,
ATTACHMENT, AUTO_INCREMENT, BFILE, BIGINT, BIT,
BINARY, BINARY_DOUBLE, BINARY_FLOAT, BLOB,
BOOLEAN, BYTE, CHAR, CLOB, COUNTER, CURRENCY,
DATE, DATETIME2, DATETIMEOFFSET, DBCLOB, DBU-
RIType, DECFLOAT, DECIMAL, DISTINCT, DOUBLE, DOU-
BLEPRECISION, ENUM, FLOAT, GEOGRAPHY, GEOMETRY,
GRAPHIC, GUID, HIERARCHYID, HTTPURIType, IDENTITY,
INT, INTEGER, INTERVAL DAY TO SECOND, INTERVAL
YEAR TO MONTH, LOGICAL, LONG, LONGBLOB, LONG
Relational
 Schemas Implementation and Reverse Engineering 343

RAW, LONGTEXT, MEDIUMBLOB, MEDIUMINT, MEDI-

UMTEXT, MONEY, MULTILINESTRING, MULTIPLE BYTE,
MULTIPOINT, MULTIPOLYGON, NCHAR, NCLOB, NESTED
TABLE, NEWID, NUMBER, NVARCHAR, NVARCHAR2,
OBJECT, POINT, POLYGON, RAW, REAL, REF, REFERENCE,
ROW BEGIN, ROW CHANGE TIMESTAMP, ROW END,
ROWGUIDCOL, ROWID, ROWVERSION, SDO_GEOMETRY,
SDO_GEORASTER, SDO_TOPO_GEOMETRY, SERIAL, SET,
SHORT, SHORT TEXT, SI_AverageColor, SI_Color, SI_Colo-
rHistogram, SI_FeatureList, SI_PositionalColor, SI_StillIm-
age, SI_Texture, SINGLE, SMALLDATETIME, SMALLINT,
SMALLMONEY, SQL_VARIANT, STRUCTURED, TIME-
STAMP, TIMESTAMP WITH LOCAL TIME ZONE, TIME-
STAMP WITH TIME ZONE, TINYBLOB, TINYINT, TINYTEXT,
TEXT, UNIQUEIDENTIFIER, URIFactory, URIType, UROWID,
VARBINARY, VARCHAR, VARCHAR2, VARGRAPHIC, VAR-
RAYS, XMLType, YESNO).
Ø Understand and apply algorithm A8 (translating relational schemes
into standard ANSI SQL DDL scripts for creating rdbs), which is
equivalent to understanding exactly how RDBMSes that provide
translation of the rdbs they manage into SQL DDL scripts are gen-
erating these scripts.
Ø Understand, explain, and use the knowledge on the differences and
similarities between the SQL ANSI standards and IBM DB2 10.5,
Oracle Database 12c and MySQL 5.7, MS SQL Server 2014 and
Access 2013 SQL idioms.
Ø Understand the five SQL DDL scripts for creating the public library
case study rdb in all above idioms.
Ø Design and use algorithms of the family AF8’, for translating stan-
dard ANSI SQL DDL scripts into any RDBMS version SQL DDL
idiom.
Ø Design and use algorithms of the family REAF0, to reverse engi-
neer any rdbs in order to obtain corresponding standard ANSI SQL
DDL scripts.
Ø Understand and apply algorithm REA0 (reverse engineering of
SQL DDL scripts, in order to obtain the corresponding relational
344 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

schemes), which is equivalent to understanding exactly how

RDBMSes create rdbs when executing SQL DDL scripts.
Ø Design and use algorithms of the family REAF0–2, to reverse engi-
neer any rdbs in order to obtain corresponding E-R data models.
Ø Understand and apply the best practice rules for correctly creating
and populating rdbs, as well as for reverse engineering them.
Ø Better understand the most usual forms of the DDL and DML
SQL statements ALTER SESSION, ALTER TABLE, COM-
MENT, CREATE DATABASE, CREATE FUNCTION, CRE-
ATE SEQUENCE, CREATE TABLE, CREATE TRIGGER,
CREATE TYPE, CREATE USER, DROP, INSERT INTO,
SELECT, UNION ALL, and of their clauses and subclauses
CONSTRAINT, CHECK, DELETE RESTRICT, DETER-
MINISTIC, FOREIGN KEY, GROUP BY, NOT NULL, ON
DELETE CASCADE (CASCADE DELETE), ON UPDATE
CASCADE (CASCADE UPDATE), ORDER BY, REFER-
ENCES, UNIQUE (UNIQUE INDEX), UPDATE OF, UPDATE
RESTRICT, and VALUES, in all of the five RDBMS versions
considered in this book, as well some of their non-standard
extensions (DESCRIBE, DO, EXPLAIN, FLUSH, LOAD
DATA INFILE, PIVOT, REPLACE, RESET, SET, SHOW,
TRANSFORM, UNPIVOT).
Ø Get acquainted with some data events (AFTER UPDATE, AS
ROW BEGIN, AS ROW CHANGE, AS ROW END, BEFORE
INSERT, BEFORE UPDATE, VALIDATED, VALIDATING) and
associated programming techniques.
Ø Understand the solved exercises and solve all of the proposed ones.
Ø Better understand the relationships between the algorithms and
family of algorithms A1-7, A2014SQLS1-8, A7/8-3, A8, AF1-8,
AF8’, REA0, REA1, REA2, REA2013A0-2, REAF0’, REAF0-1,
and REAF0-2.
Ø Understand the meaning of a plethora of DBMS jargon terms and
abbreviations (@@DBTS, @@IDENTITY, ALL_OBJECTS,
Allow Zero Length, BasicFile LOB, BITEMPORAL, BOOL, BUSI-
NESS_TIME, CAST, CCSID, CODEUNITS16, CODEUNITS32,
CONCAT, COUNT, CurDate, CurTime, CURRENT DATE,
Relational
 Schemas Implementation and Reverse Engineering 345

CURRENT_DATE, CURRENT LOCALE, CURRENT TIME,

CURRENT_TIME, CURRENT TIMESTAMP, CURRENT_
TIMESTAMP, CURRENT TIMEZONE, CURRENT_TIME-
ZONE, DANGLING, DATALINK, DATE_ADD, DATE_SUB,
DBCS, DBMS_XMLSCHEMA, DBTimeZone, DB2SECURITY-
LABEL, DB2SQLSTATE, DBA_OBJECTS, DEREF, Design
View, DETERMINED BY, dummy date, EBCDIC, Enforce Ref-
erential Integrity, Extract, FOR BIT DATA, FOR EACH ROW,
GENERATE_UNIQUE, GetDate, GetUTCDate, grbit, HEX,
Indexed, LAST_INSERT_ID, LC_TIME, LIMIT, Limit to List,
Local_Timestamp, Lookup, Lookup Wizard, MAX_STRING_
SIZE, MAXDB, MBCS, MIXED DATA, MIXED DECP, Mixed
Based Replication (MBR), MSysRelationships, Now, OID, ON
COMMIT PRESERVE ROWS, PARAMETERS, Query Design,
QUOTED_IDENTIFIER, REF, REGEXP, Required, Row Based
Replication (RBR), SecureFile LOB, SBCS, Statement Based Rep-
lication (SBR), STDDT, SessionTimeZone, STRAIGHT_JOIN,
SUM, SysDate, SysDateTime, SysDateTimeOffset, SysTime-
stamp, SysUTCDateTime, SYSCAT, SYSDUMMY, SYSFUN,
SYSIBM, SYSPROC, SYSTEM_TIME, szColumn, szObject,
szReferencedColumn, szReferencedObject, szRelationship, Time,
TOP, TRANSACTION START ID, TRANSLATE, TRIM, UDT,
Universal Naming Convention (UNC), USER_OBJECTS, UTF-
16, UTF-8, Validation Rule, Validation Text, WTDDT, Yes (No
Duplicates)).
Any relational scheme may be at least partially implemented as an
rdb, by using any RDBMS version. Generally, their objects are grouped in
tablespaces (typically, there are several system ones, for the metacalogs,
the temporary tables, etc. and at least a user one).
Uniqueness is implemented through unique index files, also used for
speeding-up data search. Index files are using several data structures (e.g.,
trees, bitmaps, hash tables, linked lists, etc.) to essentially store pairs of the
type <data value, pointer to the location where that data value is stored>.
Non-unique index files are used to speed-up data search (that should be
balanced with the cost of maintaining them when data is modified—see
Chapter 4 of the second volume).
346 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

Sometimes, due to limitations in RDBMS versions implemented

SQL, in order to implement domain and tuple constraints (but also non-
relational ones), it is necessary to also use their extended SQL triggers:
special stored procedures associated to a table and a (or several) data
manipulation associated event(s) that are automatically called by the
system whenever the corresponding data event(s) occur(s). These data
events (events for short), which may have different names for different
platforms94, are generally associated to moments in time corresponding
to immediately before and after inserting, updating, or deleting a row, to
user’s clicks, double-clicks, etc.
Although SQL is standardized by both ANSI and ISO, due to histori-
cal and commercial competition reasons, its actual implementations differ
among RDBMSes, especially at the DDL level (while at the DML one
differences are minor). This is why, on one hand, section 4.1 presents only
the algorithm A8, for translating RDM schemes into SQL ANSI-92 scripts
(ignoring any RDBMS implementation differences), Section 4.2 is focus-
ing on the differences among the five RDBMSs chosen to illustrate the
approach proposed by this book, section 4.3 shows actual SQL DDL scripts
(as well as additional MS Access GUI techniques needed for enforcing
domain and tuple/check constraints) for implementing the public library
case study RDM scheme obtained in Section 3.4, and, on the other, in
order to actually implement any RDM scheme in any given RDBMS ver-
sion, you need the additional family of algorithms AF8’, which translates
ANSI standard SQL DDL scripts into RDBMSs corresponding ones.
In fact, algorithm A8 also embeds the ANSI-2003 standard of auto-
generated (autonumber) columns; moreover, as this is the only way to
implement autonumbering in Oracle versions prior to 12c and tuple/check
constraints in MySQL (but also in Oracle for the ones that need nondeter-
ministic functions), the scripts that implement the public library rdb from
Subsections 4.3.2 and 4.3.3 also use triggers, which were standardized
only in the SQL ANSI-99.
As shown in Section 4.2, most of the considered RDBMS versions also
provide either other higher SQL ANSI standards or/and own nonstandard
facilities; exercise 4.4 proposes to readers designing corresponding exten-
sions to algorithm A8.
94
For example, even for products of a same manufacturer: in .NET, MS refers as Validating and
Validated the events that in Access are known (just like in Oracle) as BeforeUpdate and AfterUpdate,
respectively.
Relational
 Schemas Implementation and Reverse Engineering 347

As they are very easy to design and even implement, especially after
studying Sections 4.2 and 4.3, AF8′ algorithms are left to the reader (see
Exercise 4.3). Similarly, the dual REAF0′ algorithms are very simple,
except for Access, which is dealt with in section 4.4; consequently, they
are left to the reader too (see Exercise 4.5). Same thing goes for the dual of
A8, the reverse engineering algorithm REA0 (see Exercise 4.6).
Please note that A8 only considers single foreign keys referencing (single)
primary surrogate keys, as you should always use. If, however, you also use
the totally not recommended concatenated foreign keys provided too by the
RDM, then you need a slight variation of this algorithm (see Exercise 4.7).
Section 4.5 discusses the family of algorithms AF1–8, probably the
most important set of forward “shortcuts” for actually implementing E-R
data models into rdbs as fast as advisable, and presents, as an example, one
of its members (for MS SQL Server dbs).
Section 4.6 discusses the REAF0–2 family, dual to the above one,
and at least as important, especially for decrypting the architecture and
the semantics of legacy (but not only), undocumented (or poorly docu-
mented) rdbs, and presents too, as an example, one of its members (for
MS Access dbs).
Section 4.7 presents (as a rdb reverse engineering case study) the
results of applying for an Access db (Access again, as it is the only one
out of the five RDBMSes considered in this book for which reverse engi-
neering is not that obvious) the corresponding algorithms of the families
REAF0′ (composed with REA0) and REAF0–2.
Finally, this chapter ends like all other ones of this book with sections
dedicated to the corresponding best practice rules, the math behind, con-
clusion, exercises, and past, present and references.

4.1  THE ALGORITHM FOR TRANSLATING RELATIONAL

SCHEMAS INTO SQL DDL ANSI-92 SCRIPTS (A8)

Thinking things through is hard work and it sometimes seems safer to

follow the crowd. That blind adherence to such group thinking is,
in the long run, far more dangerous than
independently thinking things through.
—Thomas Watson, Jr.
348 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

Figure 4.1 presents algorithm A8, for translating relational schemas into
SQL DDL ANSI standard corresponding scripts. For the column domains
D’, see Table 4.1, which is presenting the SQL ANSI-92 data types (with
the addition of the autonumbers, collections, and XML that were intro-
duced in later ANSI standards).
Variants of the algorithm A8 (which are targeting as output language
their own SQL idioms instead of the ANSI standard ones) are implemented
by all RDBMS versions that provide automatic generation of SQL DDL
scripts of the rdbs they manage (which includes DB2 10.5, Oracle 12c,
MySQL 5.7, and SQL Server 2014).

4.2  RELEVANT DIFFERENCES BETWEEN IBM DB2, ORACLE

DATABASE AND MYSQL, MICROSOFT SQL SERVER AND
ACCESS

Technology, which is generally believed to kill the imagination free-

dom, the beauty, and the soul, has, among its latest consequences,
the chance to bring to the Universe the fairytale,
the poetry, and the immortality.
—Lucian Blaga

Although most of both commercial and open source RDBMSes adhere at

least to the pure ANSI-92 SQL standard, there are, however, significant
differences between them especially in the data types they provide, but
also in naming conventions, and sometimes, even in SQL syntax.
This is why this section is presenting the most relevant of these differ-
ences, in order to help with implementing ANSI standard SQL scripts in
each of these 5 flagship RDBMSs (and fully understand the differences
between their implementations of the library db case study presented in
Section 4.3).
For those in a hurry, at least the tables at the end of each Data Types
subsections below should be of interest, as they summarize what data type
is best to choose, depending on your actual needs and target RDBMS.
In all data type definitions that follow, square brackets are embracing
optional items; for example, DECIMAL[(n[,m])] means that both n
Relational
 Schemas Implementation and Reverse Engineering 349

FIGURE 4.1  Algorithm A8 (Translation of RDM Schemas into SQL ANSI-92 Scripts).
350 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

TABLE 4.1  The SQL ANSI Standard Data Types

Data Type Description

CHARACTER(n) Character string, fixed length n. A string of text in an imple-

menter-defined format. The size argument is a single nonnega-
tive integer that refers to the maximum length of the string.
Values for this type must be enclosed in single quotes.

VARCHAR(n) Variable length character string, maximum length n.

BINARY(n) Fixed length binary string, maximum length n.

BOOLEAN Stores truth values—either TRUE or FALSE.

VARBINARY(n) Variable length binary string, maximum length n.

INTEGER(p) Integer numerical, precision p.

SMALLINT Integer numerical precision 5.
INTEGER Integer numerical, precision 10. It is a number without deci-
mal point with no digits to the right of the decimal point, that
is, with a scale of 0.
BIGINT Integer numerical, precision 19.

IDENTITY(s, i) Integer numerical auto-generated, starting value s, increment

value i (added in SQL2003).

DECIMAL(p, s) Exact numerical, precision p, scale s. A decimal number is a

number that can have a decimal point in it. The size argument
has two parts: precision and scale. The scale cannot exceed
the precision. Precision comes first, and a comma must sepa-
rate from the scale argument. Mantissa’s length is p – s.

NUMERIC(p, s) Exact numerical, precision p, scale s. (Same as DECIMAL).

FLOAT(p) Approximate numerical, precision p. A floating number in

base 10 exponential notation. The size argument for this type
consists of a single number specifying the minimum precision.
REAL Approximate numerical, mantissa precision 7.
FLOAT Approximate numerical, mantissa precision 16.
DOUBLE Approximate numerical, mantissa precision 16.
PRECISION
DATE, TIME, Composed of a number of integer fields, representing an abso-
TIMESTAMP lute point in time, depending on subtype.
Relational
 Schemas Implementation and Reverse Engineering 351

TABLE 4.1  (Continued)

Data Type Description

INTERVAL Composed of a number of integer fields, representing a period
of time, depending on the type of interval.
COLLECTION  ARRAY (added in SQL99) is a set-length and ordered col-
(ARRAY, lection of elements, MULTISET (added in SQL2003) is a
MULTISET) variable-length and unordered collection of elements. Both the
elements must be of a predefined data type.
XML Stores XML data (added in SQL99). It can be used wherever a
SQL data type is allowed, such as a column of a table.

and m parameters are optional, so all of (and only) the following expres-
sions are valid: DECIMAL, DECIMAL(n), and DECIMAL(n, m).

4.2.1 IBM DB2 10.5

Good design is good business.

—Thomas J. Watson

DB2 10.5 is a very robust, well-performing and fully featured RDBMS,

also available in the cloud and supporting both SQL and XQuery (DB2 has
native implementation of XML data storage, where XML data is stored as
XML (not as relational or CLOB data) for faster access using XQuery).
DB2 10.5 can also be accessed through ODBC, ADO, and OLEDB and
has APIs for Java, .NET CLI, Python, Perl, PHP, Ruby, C++, C, Delphi,
REXX, PL/I, COBOL, RPG, FORTRAN, and many other programming
languages. DB2 also supports integration into the Eclipse and Visual Stu-
dio integrated development environments.
In the clustered DBMS arena, where databases can grow to many tera-
bytes, IBM offers two approaches (that compete with Oracle Real Appli-
cation Clusters (RAC)): DB2 pureScale and DB2 Database Partitioning
Feature (DPF). DB2 pureScale is a shared-disk db cluster solution that is
ideal for high-capacity OLTP workloads. DB2 DPF lets you partition your
db across multiple servers or within a large SMP server, which is ideal for
OLAP workloads.
352 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

DB2 DPF is sold as part of the IBM InfoSphere Warehouse, which is

the DB2’s name when sold in data warehouse environments. It includes
several BI features, such as ETL, data mining, OLAP acceleration, and
in-line analytics.
As market share, according to Gartner (probably the most objective
source in the field), until 2013, DB2 was slightly surpassed only by Ora-
cle, which lead it by some 33% at 30%.95 In 2013, Oracle rose at more
than 40%, whereas DB2 was for the first time surpassed by the MS SQL
Server (mainly due to its native bridge to MS Excel, but also to the one to
Hadoop, as well as to the MS Azure cloud). Corresponding data for 2014
will certainly be very interesting too.
Besides several commercial versions, there is also a free to download
DB2 Express-C version (for 15 TB dbs maximum that should kill any cur-
rent competition, which only allows for some 10 GB!) that can also be
redistributed with your products; it allows for 16 GB RAM (again remark-
able, as competition currently allows for only 1 GB) and 2 CPU cores.
DB2 10.5 was the base for all SQL ANSI standards, also including sup-
port for temporal dbs.
DB2 10.5 runs on Linux, Unix, MS Windows, zOS (mainframes),
iSeries (formerly OS/400) and VM/VSE OSs. Its Express-C free edition
runs on MS Windows (x86/x64), Linux (x86, x64, Power Systems), Solaris,
and Mac OS X.

4.2.1.1  Identifiers and Literals

To be successful, you have to have your heart in your business and

your business in your heart.
—Thomas J. Watson

DB2 object names may generally have at most 128 bytes (i.e., between 32
and 128 characters, depending on the chosen alphabet and encoding) and
are not case sensitive: you can freely use both upper and lower case (and
DB2 stores them exactly as you defined them, if they were enclosed in
95
Third place, with some 17%, was occupied by the MS SQL Server; all other competitors, including
the very popular open source MySQL that was leading the pack, are fighting to survive in or emerge
from the rest of some 20%; surprisingly, at that time, the only NoSQL DBMS ranked among the top
ten as popularity was MongoDB; again as popularity, Access ranked 7th, only two seats behind DB2.
Relational
 Schemas Implementation and Reverse Engineering 353

double quotes, case in which you should always refer to them alike, or in
uppercase, if they were not delimited with double quotes).
The so-called ordinary identifiers should start with a letter and can
contain only letters (with all lowercase ones automatically translated to
uppercase), digits, and ‘_’. The delimited identifiers, which should be
declared and used enclosed in double quotes, can use almost anything,
including lowercase letters and even leading spaces.
Date and time constants are enclosed in single quotation marks. Text
strings are enclosed in double quotation marks. You can use double quo-
tation marks within text strings by doubling them; for example, the text
string “““quote”””’s value is “quote”.

4.2.1.2  Databases and Tables

If you want to achieve excellence, you can get there today.

As of this second, quit doing less-than-excellent work.
—Thomas J. Watson

DB2 provides the SQL CREATE DATABASE statement too, but, mainly
due to security and physical disk space management, this is typically done
only by DBAs, who are generally using for this task too the GUI of the
IBM Data Studio tool. However, you can do it too in its Express editions,
as there you are the DBA too.
Computed tables are called views and can be created with the CRE-
ATE VIEW … AS SELECT … SQL DDL statement; they can be created
graphically through the IBM DB2 Query Management Facility’s (QMF)
QBE-type GUI too. Unfortunately, QMF (which also provides analytics)
currently has only a commercial Enterprise edition.
DB2 was the first RDBMS to introduce in-memory tables, very useful
especially for OLAP processing (but not only) as it speeds up data analysis
even more than 25 times, as part of the BLU technologies (a development
code name standing for “big data, lightning fast, ultra-easy”), which is a
bundle of novel techniques for columnar processing, data deduplication,
parallel vector processing, and data compression.
Temporary tables, either local or global are created in user temporary
tablespaces with the CREATE [GLOBAL] TEMPORARY TABLE … SQL
354 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

DDL statement. The local ones exist during the current user session or the
lifetime of the procedure in which they are declared. The global ones have
persistent definitions, but data is not persistent (as it is automatically deleted
by the system at the end of the db sessions) and they generate no redo or
rollback information. If, however, you would like their instances be persistent
too, you can add the ON COMMIT PRESERVE ROWS clause at the end of
the above statement.
Moreover, DB2 also provides temporary tables which only exist in a
FROM or WHERE clause during the execution of a single SQL statement, as
well as temporary tables that are part (and exist only during the execution)
of a WITH SQL statement.
You can associate indexes, triggers, statistics, and views to temporary
tables too. Dually, for temporary tables you cannot define referential integ-
rity, or specify default values for their columns, or have columns declared
as a ROWID or LOB (V6), or create such tables LIKE a declare global
temporary table, or issue LOCK TABLE statements on them, or use DB2
parallelism, etc.
DB2 also provides table hierarchies, in which subtables may inherit the
privileges of their supertables.

4.2.1.3  Data Types

All the problems of the world could be settled easily if men were only
willing to think.
The trouble is that men very often resort to all sorts of devices in
order not to think, because thinking is such hard work.
—Thomas J. Watson

4.2.1.3.1  Numeric Data Types

The great accomplishments of man have resulted from the

transmission of ideas and enthusiasm.
—Thomas J. Watson
DB2 has a BOOLEAN data type.
There are 7 (signed) numeric data types:
Relational
 Schemas Implementation and Reverse Engineering 355

Ø SMALLINT = [–32,768; 32,767];

Ø INTEGER (or INT) = [–2,147,483,648; 2,147,483,647];
Ø BIGINT = [–9,223,372,036,854,775,808;
9,223,372,036,854,775,807];
Ø DECIMAL (or DEC)[(n[, m])] = [–1031+1; 1031–1]; 0 ≤ n ≤
31, 0 ≤ m ≤ 31, 1 ≤ n + m ≤ 31, default values: n = 5, m = 0; almost
equivalent is the NUMERIC (or NUM): the only difference between
them is in the internal storage mode on AS/400 systems;
Ø REAL = [–3.4028234663852886 × 1038; 3.4028234663852886 ×
1038]; precision is 7 digits after the decimal point;
Ø DOUBLE = [–1.7976931348623158 × 10308; 1.7976931348623158
× 10308]; precision is 15 digits after the decimal point;
A virtual type FLOAT(n) can also be used: 0 < n < 25 means REAL,
24 < n < 54 means DOUBLE.
Ø DECFLOAT(n) = [–9.999999999999999999999999999999999 ×
106144; 9.999999999999999999999999999999999 × 106144]; n [
{16, 34}.
Decimal floating-point numbers (DECFLOAT) can have a precision of
16 or 34 and an exponent range of 10–383 to 10384 or 10–6143 to 106144, respec-
tively: the minimum exponent Emin for DECFLOAT values are −383 for
DECFLOAT(16) and −6,143 for DECFLOAT(34); the maximum exponent
Emax values are 384 for DECFLOAT(16) and 6,144 for DECFLOAT(34).
Decimal floating-point supports both exact representations and approx-
imations of real numbers and so is not considered either an exact or an
approximate numeric type.
In addition to finite numbers, DECFLOAT numbers are able to repre-
sent the following named decimal floating-point special values:
ü Infinity (a value that represents a number whose magnitude is infi-
nitely large);
ü Quiet NaN (a value that represents undefined results and that does
not cause an invalid number warning);
ü Signaling NaN (a value that represents undefined results and
that causes an invalid number warning if used in any numeric
operation).
When a number has one of these special values, its coefficient and
exponent are undefined.
356 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

It is possible to have positive or negative infinity. The sign of NaN

values has no meaning for arithmetic operations.
Nonzero numbers whose adjusted exponents are less than Emin are
called subnormal numbers. These subnormal numbers are accepted as oper-
ands for all operations and can result from any operation. For a subnormal
result, the minimum values of the exponent become Emin—(precision—1),
called Etiny, where precision is the working precision. If necessary, the
result is rounded to ensure that the exponent is no smaller than Etiny. If the
result becomes inexact during rounding, an underflow warning is returned.
A subnormal result does not always return the underflow warning.
When a number underflows to zero during a calculation, its exponent
will be Etiny. The maximum value of the exponent is unaffected. The
maximum value of the exponent for subnormal numbers is the same as the
minimum value of the exponent that can arise during operations that do
not result in subnormal numbers. This occurs when the length of the coef-
ficient in decimal digits is equal to the precision.
Autonumbering is obtained by declaring the corresponding column as
being GENERATED ALWAYS AS IDENTITY. A table can have only
one identity column. The IDENTITY keyword can only be specified for
columns of exact numeric or user-defined DISTINCT types (see Subsec-
tion 4.2.1.2.3) for which the source type is an exact numeric type, with a
scale of zero. An identity column is implicitly NOT NULL. Obviously, an
identity column cannot have a DEFAULT clause.
Only SMALLINT, INTEGER, BIGINT, DECIMAL with a scale of zero,
and a DISTINCT type based on one of these types are exact numeric types.
REFERENCE types (see Subsection 4.2.1.2.3 below), even if represented
by an exact built-in numeric type, cannot be defined as identity columns.
Explicit start, stop, (no) minimum, (no) maximum, increment, and (no)
caching values, as well as (no) order direction and (no) cycling can also
be specified for IDENTITY columns. Implicitly, they are 1 for start and
increment, corresponding minimum and maximum data type values, 20
for cache, ascending for order, and no for cycling.
There are 3 DATE/TIME data types:
Ø DATE = [1/1/1; 12/31/9999];
Ø TIME = [00:00:00; 24:00:00];
Ø TIMESTAMP[(fsp)] = [1/1/1 00:00:00.000000000000;
12/31/9999 00:00:00.000000000000]; 0 ≤ fsp ≤ 12, default 6.
Relational
 Schemas Implementation and Reverse Engineering 357

You can get the current system date, time, and timestamp directly
from the registers CURRENT DATE (or CURRENT_DATE), CURRENT
TIME (or CURRENT_TIME), and CURRENT TIMESTAMP (or CUR-
RENT_TIMESTAMP), respectively. For getting the current system locale
(by default, the “en_US” one, for English U.S.), you can use the CUR-
RENT LOCALE LC_TIME register. Similarly, you can get the local offset
with respect to the current UTC time from the CURRENT TIMEZONE (or
CURRENT_TIMEZONE) register. You can get the year part of a date or
timestamp d with the function Year(d).
There is no CURRENCY data type: during data integrations, DB2
translates such data types into DECIMAL.

4.2.1.3.2  String Data Types

Whenever an individual or a business decides that success has been

attained, progress stops.
—Thomas J. Watson

There are seven (signed) numeric data types in DB2:

Ø CHAR[(n)] = ASCII/UNICODE(n), fixed length, 1 ≤ n ≤ 254, for
ASCII (OCTET), and 1 ≤ n ≤ 63, for UNICODE (CODEUNITS32,
that is UTF-32); default n value: 1; synonym: CHARACTER;
Ø VARCHAR(n) = ASCII/UNICODE(n), variable length, 1 ≤ n ≤
32,672, for ASCII (OCTET) and 1 ≤ n ≤ 8,168, for UNICODE
(CODEUNITS32, that is UTF-32); synonym: CHAR (or CHARAC-
TER) VARYING;
Ø CLOB[(n)] = ASCII/UNICODE(n), variable length, 1 ≤ n ≤
2,147,483,647, for ASCII (OCTET), and 1 ≤ n ≤ 536,870,911, for UNI-
CODE (CODEUNITS32, that is UTF-32); default n value: 1 M (one
megabyte); synonym: CHAR (or CHARACTER) LARGE OBJECT;
Ø GRAPHIC[(n)]: equivalent of CHAR, but with encodings of 2
bytes per character (UTF-16); corresponding maximum values for
n are 127 and 63;
Ø VARGRAPHIC(n): equivalent of VARCHAR, but with encodings of
2 bytes per character (UTF-16); corresponding maximum values
for n are 16,336 and 8,168, respectively;
358 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

Ø DBCLOB[(n)]: equivalent of CLOB, but with encodings of 2

bytes per character (UTF-16); corresponding maximum values for
n are 1,073,741,823 and 536,870,911;
Ø BLOB[(n)]: variable length binary strings of up to 2 GB (1 ≤ n
≤ 2,147,483,647 bytes), not associated with a code page; default n
value: 1 M (one megabyte).
BLOB, CLOB, and DBCLOB sizes (i.e., their parameter n) can be
expressed either in bytes, kilobytes (K), megabytes (M), or gigabytes (G),
specified by n’s suffix (nothing, ‘K’, ‘M’, or ‘G’, respectively).
NULL-terminated character strings that are found in C are handled dif-
ferently, depending on the standards level of the precompile option. Each
character string is further defined as one of:
ü BIT data (binary string data not associated with a code page);
ü Single-byte character set (SBCS) data (data in which every charac-
ter is represented by a single byte);
ü Mixed data (data that might contain a mixture of characters from a
single-byte character set and a multibyte character set (MBCS)).
If you need to store UNICODE data, please note the following eight
facts:
ü In the corresponding CREATE DATABASE, CREATE TABLESPACE,
or/and CREATE TABLE statements, specify the CCSID UNICODE
clause.
ü By default, the encoding scheme of a table is the same as the encod-
ing scheme of its tablespace. Also by default, the encoding scheme
of a tablespace is the same as the encoding scheme of its db. You
can override the encoding scheme with the CCSID clause in the
CREATE TABLESPACE or CREATE TABLE statement. How-
ever, all tables within a tablespace must have the same CCSID.
ü In the CREATE TABLE statements, for each column definition,
specify the appropriate data type, subtype, and length value.
ü Use one of the following data types:
• For UTF-8 data: CHAR, VARCHAR, or CLOB.
• For UTF-16 data: GRAPHIC, VARGRAPHIC, or DBCLOB.
• For binary data: BINARY, VARBINARY, or BLOB.
ü For character columns, optionally specify one the following subtypes,
by adding the FOR subtype DATA clause to the column definition:
Relational
 Schemas Implementation and Reverse Engineering 359

• SBCS: columns contain only those UTF-8 characters that are

stored as 1 byte (the first 128 ones in the UNICODE code
page); data stored in SBCS columns of UNICODE tables has
CCSID 367.
• MIXED (default value: if you do not specify a subtype, DB2
assumes by default the FOR MIXED DATA one): columns
contain any UTF-8 data that is more than 1 byte; character
data in UNICODE tables is stored as mixed data by default,
even if your subsystem is defined with a MIXED DECP value
of NO; data stored in MIXED columns of UNICODE tables has
CCSID 1208.
• BIT: columns contain BIT (i.e., binary, not string) data; CCSID
66534 is associated with FOR BIT DATA columns.
ü Do not use FOR BIT DATA columns for the sole purpose of
handling international data: only use FOR BIT DATA columns
if you have a specific reason, such as encryption; otherwise, this
data type can cause problems. For example, if you have a string of
length 10 and put it in a FOR BIT DATA column of length 12,
DB2 pads the string with two blanks. The hexadecimal value that
is used for those blanks is system specific. For example, X’40’ is
used for ASCII/EBCDIC and X’20’ is used for UNICODE. These
different hexadecimal values can potentially cause problems when
you convert this data.
ü To determine the appropriate length value, follow these instructions:
• For UTF-8 data, allocate three times the column size that you
would allocate for a non UNICODE table. For example, if
you use CHAR(10) for a name column in an ASCII/EBCDIC
table, use VARCHAR(30) for the same column in a UNICODE
table. Such columns can contain 30 bytes or ten 3-byte char-
acters. Use VARCHAR instead of CHAR, because the length 30
is greater than 18 (where 18 is traditionally the length from
which VARCHAR should be used instead of CHAR.) This esti-
mate allows for the worst-case expansion of UTF-8 data. The
worst case for SBCS data is that 1 byte in ASCII/EBCDIC
expands to 3 bytes in UTF-8. For mixed data, such as Chinese,
Japanese, or Korean characters, the same worst-case scenario
360 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

applies: you might have 2, 3, and 4-byte characters, depending

on the encoding, that expand to a four-byte UTF-8 character
in the worst case; however, because these characters use more
than one byte in ASCII or EBCDIC, the worst-case expansion
in UTF-8 is still three times the original size.
s For UTF-16 data, allocate two times the column size that
you would allocate for a non UNICODE table, and use the
GRAPHIC or VARGRAPHIC data types. For example, if you
use CHAR(10) for a name column in an ASCII/EBCDIC table,
use VARGRAPHIC(10) for the same column in a UNICODE
table: CHAR(10) is 10 bytes long and VARGRAPHIC(10) is 20
bytes long, or the equivalent of 10 two-byte characters.
ü DB2 associates certain CCSIDs with columns, depending on
the data type that you specify. Table 4.2 summarizes the pos-
sible column data types in UNICODE tables and their associated
CCSIDs:

TABLE 4.2  CCSIDs Associated by DB2 with Columns in UNICODE Tables

Column data type Associated CCSID Format in which data is stored
CHAR 1208 UTF-8
CHAR FOR SBCS DATA 367 7-bit ASCII
CHAR FOR MIXED DATA 1208 UTF-8
CHAR FOR BIT DATA 66534 NA
VARCHAR 1208 UTF-8
VARCHAR FOR SBCS DATA 367 7-bit ASCII
VARCHAR FOR MIXED 1208 UTF-8
DATA
VARCHAR FOR BIT DATA 66534 NA
CLOB 1208 UTF-8
CLOB FOR SBCS DATA 367 7-bit ASCII
CLOB FOR MIXED DATA 1208 UTF-8
GRAPHIC 1200 UTF-16
VARGRAPHIC 1200 UTF-16
DBCLOB 1200 UTF-16
Relational
 Schemas Implementation and Reverse Engineering 361

BLOBs are generally used to store non-traditional data such as pictures,

voice, or mixed media, but they can also hold structured data for process-
ing by user-defined data types and functions. Character strings of the FOR
BIT DATA subtype may be used for similar purposes.
In UNICODE dbs, when using CODEUNITS16 or CODEUNITS32,
CHAR and GRAPHIC may also be referred as NCHAR, VARCHAR and
VARGRAPHIC as NVARCHAR, whereas CLOB and DBCLOB as NCLOB
(all of them having the prefix ‘N’ from National character sets).
CLOBs are used to store large SBCS or mixed (SBCS and MBCS)
character-based data (such as documents written with a single character
set) and, therefore, has an SBCS or mixed code page that is associated
with it.
The double-byte character large object (DBCLOB) data type is used to
store large double-byte character set (DBCS) based data (such as docu-
ments written with a single character set) and, therefore, has a DBCS code
page that is associated with it.

4.2.1.3.3  Other Data Types

Every time we’ve moved ahead in IBM, it was because someone was
willing to take a chance,
put his head on the block, and try something new.
—Thomas J. Watson

There are also six other DB2 data types: one built-in (XML), one extended
(ANCHOR), and four user-defined (DISTINCT, STRUCTURED, REFER-
ENCE, and ARRAY).
Ø XML stores well-formed XML documents; such values share the
first 5 restrictions of the CLOB, DBCLOB, and BLOB data types
above.
Ø ANCHOR is a data type based on another SQL object such as a
column, global variable, SQL variable or parameter, or a table
or view row. A data type defined using an anchored type defini-
tion maintains a dependency on the object to which it is anchored.
Any change in the data type of the anchor object will impact the
362 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

anchored data type. If anchored to the row of a table or view, the

anchored data type is ROW96, with the fields defined by the columns
of the anchor table or view.
User-defined data types (UDT) are derived from existing ones, for
extending the built-in ones and create your own customized data types:
Ø DISTINCT shares its internal representation with an existing built-
in data type (its “source” type) and includes qualified identifiers.
If the schema name is not used to qualify the distinct type name when
used in other than the CREATE TYPE (DISTINCT), DROP, or COMMENT
statements, the SQL path is searched in sequence for the first schema with
a distinct type that matches.
Distinct types that are sourced on LOB types are subject to the same
restrictions as their source type.
Not all built-in data types can be used to define DISTINCT types: the
source types cannot be BOOLEAN, XML, ARRAY, CURSOR, or ROW.
The specified distinct type cannot have any data type constraints and
the source type cannot be an ANCHORed data type.
A distinct type is defined to use either strong or weak typing rules, with
the strong ones being the default.
A strongly typed DISTINCT data type (STDDT) is considered to be
a separate and incompatible type for most operations. For example, you
would use it to define a picture, text, or audio type, as each of these types
has its own different semantics, but all of them use the built-in data type
BLOB for their internal representation.
The AUDIO following example is a STDDT:
CREATE TYPE AUDIO AS BLOB(1 M);
This definition allows the creation of functions that are written spe-
cifically for AUDIO, and assures that these functions are not applied to
values of any other data type (e.g., pictures or text ones): STDDTs support
strong typing by ensuring that only those functions and operators that are
explicitly defined on the distinct type can be applied to its instances. For
this reason, a STDDT does not automatically acquire the functions and

96
ROW is a user-defined data type that can be used only within the IBM’s extended SQL Procedural
Language (SQL PL) applications; it is a structure composed of multiple fields, each with its own name
and data type, which can be used to store the column values of a row in a result set or other similarly
formatted data.
Relational
 Schemas Implementation and Reverse Engineering 363

operators of its source type, because these functions and operators might
not be meaningful.
For example, a LENGTH function could be defined to support a param-
eter with the data type AUDIO that returns the length of the object in sec-
onds instead of bytes.
A weakly typed DISTINCT data type (WTDDT) is considered to be
the same as its source type for all operations, except when it applies con-
straints on values during assignments, casts, and function resolutions.
The NATURALS following example is a WTDDT:
CREATE TYPE NATURALS AS INTEGER
WITH WEAK TYPE RULES CHECK (VALUE >= 0);
Weak typing means, in this case, that except for accepting only posi-
tive integer values, NATURALS operates in the same way as its underlying
INTEGER data type.
WTDDTs can be used as alternative methods of referring to built-in
data types within application code: the ability to define constraints on the
values that are associated with them provides a method for checking val-
ues during assignments and casts.
There are two distinguished built-in system DISTINCT types that can
be used as table column ones:
ü SYSPROC.DB2SECURITYLABEL must be used to define the row
security label column of a protected table; its underlying data type
is VARCHAR(128) FOR BIT DATA; a table can have at most
one column of type DB2SECURITYLABEL; for such a column,
NOT NULL WITH DEFAULT is implicit and cannot be explicitly
specified; the default value is the session authorization ID’s secu-
rity label for write access.
ü SYSPROC.DB2SQLSTATE is used to store DB2 errors, warnings,
and information codes; its underlying data type is INTEGER.
Using DISTINCT types provides benefits in the following areas:
ü Extensibility: increasing the set of data types available to support
your applications.
ü Flexibility: any semantics and behavior can be specified for these
data types by using user-defined functions to augment the diversity
of the data types available in the system.
364 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

ü Consistent and inherited behavior: strong typing guarantees that

only functions defined on your distinct type can be applied to
instances of the distinct type; weak typing ensures that the distinct
type behaves the same way as its underlying data type and so can
use all the same functions and methods available to that underly-
ing type.
ü Encapsulation: using a WTDDT makes it possible to define data
type constraints in one location for all usages within application
code for that distinct type.
ü Performance: distinct types are highly integrated into the db man-
ager; because they are internally represented the same way as
the built-in ones, they share the same efficient code that is used
to implement components such as built-in functions, comparison
operators, and indexes for built-in data types.
Ø STRUCTURED contains a sequence of named attributes, each of
which has a data type and also includes a set of method specifica-
tions. A structured type can be used as the type of a table, view, or
column. When used as a type for a table or view, that table or view
is called a typed one.
For typed tables and views, the names and data types of the attributes
of the structured type become the names and data types of the columns of
that typed table or view. Rows of the typed table or view can be thought of
as a representation of instances of the structured type. When used as a data
type for a column, the column contains values of that structured type (or
values of any of the subtypes for that type, as defined immediately below).
Methods are used to retrieve or manipulate attributes of a structured col-
umn object.
A supertype is a structured type for which other structured types, called
subtypes, are defined. A subtype inherits all the attributes and methods of
its supertype and can have additional attributes and methods defined. The
set of structured types that is related to a common supertype is called a
type hierarchy and the type that does not have any supertype is called the
root type of the type hierarchy.
The term subtype applies to a user-defined structured type and all user-
defined structured types that are below it in the type hierarchy. Therefore,
a subtype of a structured type T is T and all structured types below it in
Relational
 Schemas Implementation and Reverse Engineering 365

the hierarchy (i.e., when the hierarchy is a tree, the subtree rooted in T).
A proper subtype of a structured type T is a structured type below T in the
type hierarchy.
Recursive type definitions in type hierarchies are subject to some
restrictions, for which it is necessary to develop a shorthand way of refer-
ring to the specific type of recursive definitions that are allowed; the fol-
lowing definitions are used:
ü Directly usage: a type A is said to directly use another type B, if and
only if one of the following statements is true:
1. Type A has an attribute of type B.
2. Type B is a subtype of A or a supertype of A.
ü Indirectly usage: A type A is said to indirectly use a type B, if one
of the following statements is true:
1. Type A directly uses type B.
2. Type A directly uses some type C and type C indirectly uses
type B.
A type cannot be defined so that one of its attribute types directly or
indirectly uses itself (i.e., STRUCTURED type definitions should be acy-
clic). If it is, however, necessary to have such a configuration, consider
using a REFERENCE (see below) as the attribute. For example, with struc-
tured type attributes, there cannot be an instance of “employee” with an
attribute of “manager” when “manager” is of type “employee”. There can,
however, be an attribute of “manager” with a type of REF(employee).
A type cannot be dropped if certain other objects use the type, either
directly or indirectly. For example, a type cannot be dropped if a table or
view column makes direct or indirect use of the type.
Ø REFERENCE (or REF) is a companion type to a structured type.
Similar to a DISTINCT type, a reference one is a scalar type that
shares a common representation with one of the built-in data types.
This same representation is shared for all types in the type hier-
archy. The reference type representation is defined when the root
type of a type hierarchy is created. When using a reference type, a
structured type is specified as a parameter of the type. This param-
eter is called the target type of the reference.
Ø ARRAY is a data type that is defined as an array with elements of
another data type. An array is a structure that contains an ordered
366 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

collection of data elements in which each element can be referenced

by its index value in the collection. All elements in an array have
the same data type. The cardinality of an array is the number of its
elements. There are two types of arrays: ordinary and associative.
An ordinary array has a defined upper bound on the number of ele-
ments, known as the maximum cardinality. Each element in the array is
referenced by its ordinal position as the index value. If n is the number of
elements in an ordinary array, the ordinal position associated with each
element is an integer value greater than or equal to 1 and less than or equal
to n. Every ordinary array type has an INTEGER index.
The maximum cardinality of an ordinary array is not related to its physi-
cal representation, unlike the ones of some PLs such as C. Instead, the maxi-
mum cardinality is used by the system at run time to ensure that subscripts
are within bounds. The amount of memory required to represent an ordinary
array value is not proportional to the maximum cardinality of its type either.
An associative array has no specific upper bound on the number of
elements (i.e., no defined maximum cardinality). Each element is refer-
enced by its associated index value. The data type of the index value can
be either INTEGER or VARCHAR, but is the same one for the entire array.
The amount of memory required to represent an associative array value is
usually proportional to its cardinality.
When an array is being referenced, all of the values in the array are
stored in main memory. Therefore, arrays that contain a large amount of
data will consume large amounts of main memory.
The ARRAY type is not supported for multirow insert, update, or delete.
Optionally, you can buy a license for IBM DB2 Data Links Manager,
with which, by using the data type DATALINK, you can connect DB2 db
table columns to Windows, AIX, and/or Solaris OSs managed files.
Optionally too, you can buy the IBM DB2 Spatial Extender, which
provides geospatial data, as well as a complimentary geobrowser.

4.2.1.3.4  Computed Columns

Machines might give us more time to think but will

never do our thinking for us.
—Thomas Watson, Jr.
Relational
 Schemas Implementation and Reverse Engineering 367

Computed columns can be defined with the clause GENERATED ALWAYS

AS (generation-expression).
If the expression for a GENERATED ALWAYS column includes a user-
defined external function, changing the executable for the function (such
that the results change for given arguments) can result in inconsistent data.
This can be avoided by using the SET INTEGRITY statement to force the
generation of new values.
The generation-expression cannot contain any of the following:
ü Subqueries
ü XMLQUERY or XMLEXISTS expressions
ü Column functions
ü Dereference operations or DEREF functions
ü User-defined or built-in functions that are nondeterministic
ü User-defined functions using the EXTERNAL ACTION option
ü User-defined functions that are not defined with NO SQL
ü Host variables or parameter markers
ü Special registers and built-in functions that depend on the value of
a special register
ü Global variables
ü References to columns defined later in the column list
ü References to other generated columns
ü References to columns of type XML
The data type for the column is based on the result data type of the
generation-expression. A CAST specification can be used to force a par-
ticular data type and to provide a scope (for a reference type only). If the
data type is specified, values are assigned to the column according to the
appropriate assignment rules.
A generated column is implicitly considered nullable, unless the NOT
NULL constraint is used too. The data type of a generated column and the
result data type of the generation-expression must have equality defined
between them: this excludes columns and generation expressions of types
LOB, XML, STRUCTURED, and DISTINCT based on any of these types.
For defining temporal tables (that may be SYSTEM_TIME, BUSI-
NESS_TIME, or BI-TEMPORAL, which is the combination of the for-
mer two ones), DB2 provides the following four system generated column
types:
368 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

Ø FOR EACH ROW ON UPDATE AS ROW CHANGE TIME-

STAMP: generated values are assigned by the db manager in
each row that is inserted and for any row in which any column is
updated. The value that is generated for such a column is a time-
stamp that corresponds to the insert or update time for that row. If
multiple rows are inserted or updated with a single statement, the
value of the ROW CHANGE TIMESTAMP column might be dif-
ferent for each row.
If the data type is specified, it must be TIMESTAMP or TIMESTAMP(6).
Ø AS ROW BEGIN: generated values are assigned by the db man-
ager whenever a row is inserted into the table or any column in the
row is updated.
Values are generated using a reading from the time-of-day clock during
execution of the first data change statement in the transaction that requires
a value to be assigned to the ROW BEGIN or TRANSACTION START
ID (see the fourth type below) column in the table, or a row in a system-
period temporal table is deleted.
For a system-period temporal table, the db manager ensures unique-
ness of the generated values for a ROW BEGIN column across transac-
tions. The timestamp value might be adjusted to ensure that rows inserted
into an associated history table have the end timestamp value greater than
the begin timestamp one. This can happen when a conflicting transaction
is updating the same row in the system-period temporal table. The db con-
figuration parameter systime_period_adj must be set to Yes for this adjust-
ment to the timestamp value to occur.
If multiple rows are inserted or updated within a single SQL transaction
and an adjustment is not needed, the values for the ROW BEGIN column
are the same for all the rows and are unique from the values generated for
the column for other transactions. A ROW BEGIN column is required as
the begin column of a SYSTEM_TIME period, which is the intended use
for this type of generated columns.
Ø AS ROW END: generated values are assigned by the db man-
ager whenever a row is inserted or any column in the row is
updated. The assigned value is TIMESTAMP ‘9999–12–30–
00.00.00.000000000000’.
Relational
 Schemas Implementation and Reverse Engineering 369

A ROW END column is required as the second column of a SYS-

TEM_TIME period, which is the intended use for this type of generated
columns.
Ø AS TRANSACTION START ID: generated values are assigned
by the db manager whenever a row is inserted into the table or any
column in the row is updated. Such values are unique timestamp
ones per transaction or the null value.
The null value is assigned to the TRANSACTION START ID col-
umns if they are nullable and if there is a ROW BEGIN column in the
table for which the value did not need to be adjusted; otherwise, the value
is generated using a reading of the time-of-day clock during execution of
the first data change statement in the transaction that requires a value to be
assigned to a ROW BEGIN or TRANSACTION START ID column in the
table, or a row in a system-period temporal table is deleted.
If multiple rows are inserted or updated within a single SQL transac-
tion, the values for the TRANSACTION START ID column are the same
for all the rows and are unique from the values generated for the column
for other transactions.
A TRANSACTION START ID column is required for a system-
period temporal table, which is the intended use for this type of generated
columns.
These four system generated column types share the following com-
mon restrictions and properties:
ü tables can only have one column of each such type (ROW CHANGE
TIMESTAMP, ROW BEGIN, ROW END, and TRANSACTION
START ID);
ü they cannot have a DEFAULT clause;
ü except for the TRANSACTION START ID, columns of all other
three types must be defined as NOT NULL;
ü except for the ROW CHANGE TIMESTAMP, columns of all other
three types are not updatable;
ü except for the ROW CHANGE TIMESTAMP, if the data type of
the columns of all other three types is not specified, columns
are defined as TIMESTAMP(12); otherwise, it must be TIME-
STAMP(12).
370 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

4.2.1.3.5  Data Types Usage Overview

Wisdom is the power to put our time and our knowledge to the proper use.
—Thomas J. Watson

Table 4.3 shows how to implement in DB2 your most frequently needed
data types.
To conclude this subsection, Table 4.4 presents the rest of the data
types provided by DB2.

4.2.1.4  Other DB2 10.5 Particularities

IBM’s future is in the hands of its people. Our future is unlimited.

—Thomas Watson, Jr.

The closest to GUIDs (or UUIDs) unique values are those obtainable by
calls to the GENERATE_UNIQUE() function; as they are numeric, not
string ones, in order to obtain their string GUID-like equivalent also use
the HEX(), CHAR(), and TRIM() functions, as follows:
SELECT TRIM(CHAR(HEX(GENERATE_UNIQUE())))
FROM SYSIBM.SYSDUMMY1;
However, you do not need GUIDs in DB2: data replication is managed
by the dedicated powerful tool IBM InfoSphere Data Replication.
DB2 also provides an improved syntax of the INSERT … VALUES
SQL DML statement that allows for inserting as many rows as desired with
only one such statement.
The functions in the SYSFUN schema taking a VARCHAR as an argu-
ment will not accept VARCHARs greater than 4,000 bytes long as an argu-
ment. However, many of these functions also have an alternative signature
accepting a CLOB(1 M). For these functions, the user can explicitly cast
the greater than 4,000 VARCHAR strings into CLOBs and then recast the
result back into VARCHARs of the required length.
Special restrictions apply to expressions that result in a CLOB, DBCLOB,
and BLOB data types, as well as to structured type columns; such expres-
sions and columns are not allowed in:
Relational
 Schemas Implementation and Reverse Engineering 371

TABLE 4.3  Most Frequently Needed DB2 Data Types

Data type DB2 10.5 implementation
AUTONUMBER(n), n < 5 SMALLINT GENERATED ALWAYS AS
IDENTITY
AUTONUMBER(n), n [ [5; 9] INT GENERATED ALWAYS AS
IDENTITY
AUTONUMBER(n), n [ [10; 18] BIGINT GENERATED ALWAYS AS
IDENTITY
AUTONUMBER(n), n [ [19; 31] DECIMAL (n) GENERATED ALWAYS
AS IDENTITY
BOOLEAN BOOLEAN
NAT(n), INT(n), n < 5 SMALLINT
NAT(n), INT(n), n [ [5; 9] INTEGER (INT)
NAT(n), INT(n), n [ [10; 18] BIGINT
RAT(n, m), NAT(p), INT(p), DECIMAL (DEC, NUMERIC)
CURRENCY(n), n < 32, p [ [19; 31]
RAT(n, m), n < 39 (fastest, but only 7 REAL
digits scale precision)
RAT(n, m), n [ [39; 308] (faster, but only DOUBLE
15 digits scale precision)
RAT(n, m), NAT(n), INT(n), DECFLOAT(16)
CURRENCY(n), n [ [32; 385]
RAT(n, m), NAT(n), INT(n), DECFLOAT(34)
CURRENCY(n), n [ [386; 6,145]
DATE/TIME [1/1/1; 12/31/9999] DATE (no time), TIMESTAMP (date and
time, user defined scale of maximum 12
digits, default 6)
TIME [00:00:00; 24:00:00] TIME
ASCII(n), n ≤ 254, fixed length CHAR(n) FOR SBCS DATA
UNICODE(n), n ≤ 63, fixed length CHAR(n) [FOR MIXED DATA]
ASCII(n), n ≤ 32,767, variable length VARCHAR(n) FOR SBCS DATA
UNICODE(n), n ≤ 8,168, variable length VARCHAR(n) [FOR MIXED DATA]
ASCII(n), n [ [32,768; 2,147,483,647], CLOB(n) FOR SBCS DATA
variable length (up to 2 GB)
UNICODE(n), n [ [8,169; 536,870,911], CLOB(n) [FOR MIXED DATA]
variable length (up to 0.5 GB)
User defined data types DISTINCT
Computed column GENERATED ALWAYS AS
372 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

TABLE 4.4  Other DB2 Data Types

Data type DB2 10.5 implementation
BLOBs (pictures, voice, mixed media, structured BLOB(n)
user-defined) up to 2 GB
equivalent of CHAR, but with encodings of at least GRAPHIC(n)
2 bytes per character; corresponding maximum
values for n are 127 and 63
equivalent of VARCHAR, but with encodings VARGRAPHIC(n)
of at least 2 bytes per character; corresponding
maximum values for n are 16,336 and 8,168,
respectively
equivalent of CLOB, but with encodings of at least DBCLOB(n)
2 bytes per character; corresponding maximum
values for n are 1,073,741,823 and 536,870,911
File handlers DATALINK (only with IBM DB2
Data Links Manager)
Homogeneous vectors ARRAYS
XML (well-formatted) files XML
Dependencies on SQL objects ANCHOR
Temporal tables (dbs) ROW CHANGE TIMESTAMP, ROW
BEGIN, ROW END, and TRANS-
ACTION START ID
GPS coordinates, imagery data, geometric de- Only with the data types provided
scriptions of 3-D objects. by the IBM DB2 Spatial Extender
Nested (not 1NF) tables (object-relational) STRUCTURED, REFERENCE

ü SELECT lists preceded by the DISTINCT clause

ü GROUP BY and ORDER BY clauses
ü A subselect of a set operator other than UNION ALL
ü Basic, quantified, BETWEEN, and IN predicates
ü Aggregate functions
ü VARGRAPHIC, TRANSLATE, and date/time scalar functions
ü The pattern operand in a LIKE predicate, or the search string oper-
and in a POSSTR function
ü The string representation of a date time value.
A ROW CHANGE TIMESTAMP column cannot be used as part of a
primary key.
Relational
 Schemas Implementation and Reverse Engineering 373

Although the LONG VARCHAR and LONG VARGRAPHIC data types

continue to be supported, do not use them anymore, as they are deprecated
and might be removed in a future release.
DB2 also includes as check (tuple) constraints the functional depen-
dencies: column-name DETERMINED BY column-name or (col-
umn-name, …) DETERMINED BY (column-name, …) defines a
functional dependency between columns.
The parent set of columns contains the identified columns that immedi-
ately precede the DETERMINED BY clause. The child set of columns con-
tains the identified columns that immediately follow the DETERMINED
BY clause.
All of the restrictions on the search conditions apply to both parent and
child set columns, and only simple column references are allowed in the
set of columns.
The same column must not be identified more than once in the func-
tional dependency. The data type of the column must not be a LOB data
type, a DISTINCT type based on a LOB data type, an XML data type, or
a STRUCTURED type.
No column in the child set of columns can be a nullable column.
Functional dependencies are not enforced by the db manager dur-
ing normal operations such as insert, update, delete, or set integrity.
They might be used during query rewrite to optimize queries. Incorrect
results might be returned if the integrity of a functional dependency is
not maintained.

4.2.2 ORACLE DATABASE 12C

When you innovate, you’ve got to be prepared for people

telling you that you are nuts.
—Larry Ellison

Oracle 12c is a robust, well-performing and fully featured RDBMS,

designed for the cloud (the first one in this respect, as Oracle Corp. claims).
Standard components include db language and storage server, developer
and ETL tools, schedulers, network encryption, and basic replication.
374 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

There are dozens of other components too, including OLAP, BI, spatial
and graph data, and the new multitenant architecture, which allows user
to centrally manage any number of dbs. Oracle 12c can also be accessed
through ODBC, ADO, and OLEDB.
Besides several commercial versions, there is also a free to download
Oracle Express (XE) one (for 11 GB dbs maximum), which can also be
redistributed with your products. It allows for 1 GB RAM and 1 CPU core.
Oracle 12c partially adheres to the SQL 2011 ANSI standard (which
includes the 1999 object-relational extensions, recursive SELECT queries,
triggers, and support for procedural and control-of-flow statements in its
extended SQL PL/SQL language, as well as the 2003 and 2006 SQL/XML),
plus its own extensions (that can be detected with its FIPS Flagger, provided
you execute the SQL statement ALTER SESSION SET FLAGGER = ENTRY).
Oracle 12c runs on (z)Linux, MS Windows, HP-UX, AIX, Solaris, and
OpenVMS OSs.

4.2.2.1  Identifiers and Literals

I am so disturbed by kids who spend all day playing videogames.

—Larry Ellison

Oracle object names may have at most 30 characters, are not case sensitive
(and Oracle stores them in uppercase), except for diagrams, and, among
other reserved characters, do not accept ‘-’ (that you have to replace with
‘_’). Apart letters and numbers, you may only use ‘$’, ‘#’, ‘@’, and ‘_’.
Only for parameters you can use any of them as the first one: for the rest of
the identifiers, the first one should be a letter. However, you may use any
number of ASCII not allowed characters in names, including for the first
one, provided you embed them in double quotation marks.
Text strings have to be embedded in apostrophes (single quotation
marks). Calendar date values have to be entered as the result of the system
conversion function To_Date applied to their text string values and system
calendar date template (e.g., in the U.S., ‘MM/DD/YYYY’).
You can use apostrophes within text strings by doubling them; for
example, the text string ‘‘‘quote”’’s value is ‘quote’.
Relational
 Schemas Implementation and Reverse Engineering 375

Comments in Oracle SQL start immediately after “--” and end at the
end of the line; alternatively, you can embed any number of lines in a
single comment by embracing it in “/*” (for start) and “*/” (for end).

4.2.2.2  Databases and Tables

They don’t call it the Internet anymore, they call it cloud computing.
I’m no longer resisting the name. Call it what you want.
—Larry Ellison

In Oracle, dbs are curiously named users.97 Oracle provides a SQL CRE-
ATE USER statement too, but, similarly to DB2, mainly due to security
and physical disk space management, this is typically done only by DBAs,
who are generally using for this task too the GUI of the Oracle Enterprise
Console tool. However, in the free Express edition, you can do it too, as
you are also the DBA.
Computed tables are called views and can be created with the CREATE
VIEW … AS SELECT … SQL DDL statement; they cannot be created
through the Oracle SQL Developer’s GUI too, as it doesn’t have a QBE-
type facility.
You can also define in-memory tables, very useful for OLAP process-
ing, but only for some additional $23,000 per CPU.
Temporary tables, either local or global are created in user temporary
tablespaces with the CREATE [GLOBAL] TEMPORARY TABLE … SQL
DDL statement. The local ones exist during the current user session or
the lifetime of the procedure in which they are declared. The global ones
have persistent definitions, but data is not persistent (as it is automatically
deleted by the system at the end of the db sessions) and they generate no
redo or rollback information. If, however, you would like their instances
be persistent too, you can add the ON COMMIT PRESERVE ROWS clause
at the end of the above statement.
You can associate indexes, triggers, statistics, and views to temporary
tables too.

97
Reminiscence of its beginnings, when they did not anticipate that a user might need to manage sev-
eral dbs: this is another great counterexample of how not to name objects, especially conceptual ones.
376 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

4.2.2.3  Data Types

I have had all of the disadvantages required for success.

—Larry Ellison

4.2.2.3.1  Numeric Data Types

Number rules the universe.

—Pythagoras

Unfortunately, first of all, Oracle did not provide autonumbering up to its

12c version. The standard workaround for each table in which you need
an autonumber primary key is to define it as a NUMBER one, to declare
a sequence and a trigger associated to the event BEFORE INSERT of
that table, which uses the sequence for automatically assigning its current
value to the primary key for the currently to be inserted row (see examples
in Subsection 4.3.2 below).
An Oracle sequence is a special type (namely, arithmetic progression)
of stored parameterized numeric function (parameters: minimum, maxi-
mum, and ratio); any call to it returns the current value in the progression,
if any (otherwise it returns a system error), starting with the minimum
value (for the first call) and up to the maximum one.
There is no data type for replication either: replication (as well as het-
erogeneous data integration with various other platforms) is dealt with the
powerful (and expensive) Oracle GoldenGate add-on.
There is no Boolean data type in Oracle either: instead, you have to use
numeric or string data types (of at least 1 byte) and your own codifications
for True and False (preferably defined as a user data type).
There is no CURRENCY data type either: you have to use the numeric
types instead.
There are four numeric data types:
Ø NUMBER[(p[, s])] = [–1038; 1038] , RAT(38). Precision p [ [1;
38], default 38 and scale s [ [–84; 127], default 0;
Ø BINARY_FLOAT = [–3.40282×1038; 3.40282×1038];
Relational
 Schemas Implementation and Reverse Engineering 377

Ø BINARY_DOUBLE = [–1.79769313486231×10308;
1.79769313486231×1038];
Both BINARY_FLOAT and BINARY_DOUBLE support the distin-
guished values infinity and NaN (Not a Number).
Ø FLOAT[(p)] = [–8.5×1037; 8.5×1037]. Precision p [ [1; 126], default
126 (binary); scale is interpreted from the data; used internally,
when converting ANSI FLOAT values; not recommended to be
used explicitly too, due to truncations; use instead the more robust
NUMBER, BINARY_FLOAT, and BINARY_DOUBLE data types.
There are 4 DATE/TIME data types (DATE, TIMESTAMP, TIMESTAMP
WITH TIME ZONE, and TIMESTAMP WITH LOCAL TIME ZONE), plus
2 additional interval (INTERVAL YEAR TO MONTH and INTERVAL
DAY TO SECOND) data types:
Ø DATE = [1/1/4712 BC; 12/31/9999];
Ø TIMESTAMP [(fractional_seconds_precision)]; extension of DATE
with fractional second precision between 0 and 9 digits (default: 6);
Ø TIMESTAMP [(fractional_seconds_precision)] WITH TIME ZONE;
variant of TIMESTAMP that includes in its values a time zone
region name or a time zone offset (i.e., the difference (in hours
and minutes) between local time and UTC — formerly Greenwich
Mean Time).
Ø TIMESTAMP [(fractional_seconds_precision)] WITH LOCAL
TIME ZONE; variant of TIMESTAMP WITH TIME ZONE: it differs
in that data stored in the db is normalized to the db time zone, and
the time zone information is not stored as part of the corresponding
columns data. When a user retrieves such data, Oracle returns it in
the user’s local session time zone. This data type is useful for date
information that is always to be displayed in the time zone of the
client system, in two-tier applications;
Ø INTERVAL YEAR [(year_precision)] TO MONTH = [–4712/1;
9999/12]; year_precision is the number of digits in the YEAR field
and has the default value 2;
Ø INTERVAL DAY [(day_precision)] TO SECOND [(fractional_
seconds_precision)] = [0 0:0:0; 9 23:59:59.999999999]; day_pre-
cision is the number of digits in the DAY field, and accepts values
between 0 and 9, the default being 2; fractional_seconds_precision
378 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

is the number of digits in the fractional part of the SECOND field,

and accepts values between 0 and 9, the default being 6.
All of the above DATE/TIME data types use the following 10 fields:
ü YEAR = [–4712; 9999] — {0};
ü MONTH = [1; 12];
ü DAY = [1; 31];
ü HOUR = [0; 23];
ü MINUTE = [0; 59];
ü SECOND = [0; 59.9(n)]; 9(n) is the precision of fractional seconds;
ü TIMEZONE_HOUR = [–12; 14];
ü TIMEZONE_MINUTE = [0; 59];
ü TIMEZONE_REGION: The result of SELECT TZNAME FROM
V$TIMEZONE_NAMES;
ü TIMEZONE_ABBR: The result of SELECT TZABBREV FROM
V$TIMEZONE_NAMES;.
You can obtain the current system date and/or time, in various formats,
with the functions SysDate, SysTimestamp, Current_Date, Cur-
rent_Timestamp, and Local_Timestamp. Function EXTRACT
extracts any desired field from a DATE/TIME value; for example, in order
to extract years, you should use EXTRACT(year FROM date_value);
functions DBTimeZone and SessionTimeZone return the current
corresponding time zone offsets.

4.2.2.3.2  String Data Types

There is geometry in the humming of the strings.

There is music in the spacing of the spheres.
—Pythagoras
Oracle provides the following 11 string data types:
Ø VARCHAR2(n [BYTE | CHAR]) = ASCII(n), 0 ≤ n ≤ 4,00098;
n should be at least 1; implicitly, n is in bytes, but you can specify
it in characters (with CHAR);
98
4,000 is the standard default maximum value (for MAX_STRING_SIZE = STANDARD);
you can set MAX_STRING_SIZE = EXTENDED, in which case VARCHAR2, as well as
NVARCHAR2, and ROW have maximum length 32,767. For such extended data types, however, restric-
tions apply. Moreover, such columns are stored out-of-line, leveraging Oracle’s LOB technology. In
tablespaces managed with ASSM, extended data type columns are stored as SecureFiles LOBs. Other-
wise, they are internally stored as BasicFiles LOBs.
Relational
 Schemas Implementation and Reverse Engineering 379

Ø NVARCHAR2(n) = UNICODE(n), 0 ≤ n ≤ 4,00099;

Ø CHAR[(n [BYTE | CHAR])] = ASCII(n), 1 ≤ n ≤ 2,000; fixed
size variant of VARCHAR2 (right-padded with blanks if shorter
than n); default n is 1;
Ø NCHAR[(n)] = UNICODE(n/2), 1 ≤ n ≤ 2,000; fixed size (in bytes)
variant of NVARCHAR2 (right-padded with blanks if shorter than
n); default size is 1; when the AL16UTF16 encoding is used (2
bytes per character), maximum 1000 characters can be stored; for
the UTF8 one (3 bytes per character), only 666 may be stored;
Ø RAW(n) = binary data of at most n bytes, 0 ≤ n ≤ 2,000100;
Ø LONG RAW: binary data of at most n bytes, 0 ≤ n ≤ 232–1; Oracle
strongly recommends that you convert LONG RAW columns to
binary LOB (BLOB) columns (see below), as LOB columns are
subject to far fewer restrictions than LONG ones;
Ø ROWID: Base 64 string representing the unique address of a row
in its table; this data type is primarily for values returned by the
ROWID pseudocolumn: you can define columns having this data
type, but Oracle does not guarantee that they contain valid row ids;
Ø UROWID[(n)]: Base 64 string representing the logical address of a
row of an index-organized table; the optional n is the size of a col-
umn of type UROWID; its maximum size and default are 4,000 bytes;
Ø BLOB: binary large object; maximum size is (4 GB–1) * (db
block size);
Ø CLOB: ASCII large object; maximum size (bytes) is (4 GB–1) *
(db block size);
Ø NCLOB: UNICODE large object; maximum size (bytes) is (4
GB–1) * (db block size);
Do not use the LONG data type (provided only for backward compat-
ibility) anymore: use the LOB ones instead.
The rows in heap-organized tables that are native to Oracle have row
addresses called rowids. You can examine a rowid row address by query-
ing the pseudocolumn ROWID. Values of this pseudocolumn are strings

99
4,000 is the standard default maximum value, but you can set it to 32,767 (see above footnote).
2,000 is the standard default maximum value, but you can set it to 32,767 (see above two footnotes).
100

However, extended RAW values are stored as out-of-line LOBs only if their size is greater than 4,000
bytes; otherwise, they are stored inline.
380 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

representing the address of each row. These strings have the data type
ROWID.
Rowids contain the following information:
ü The data block of the data file containing the row; the length of this
string depends on your operating system.
ü The row in the data block.
ü The database file containing the row; the first data file has the
number 1; the length of this string depends on your OS.
ü The data object number, which is an identification number assigned
to every db segment. You can retrieve the data object number from
the data dictionary views USER_OBJECTS, DBA_OBJECTS, and
ALL_OBJECTS. Objects that share the same segment (clustered
tables in the same cluster, for example) have the same object number.
Rowids are stored as base 64 values that can contain the characters
A-Z, a-z, 0–9, and the plus sign (+) and forward slash (/). Rowids are
not available directly: you can use the supplied package DBMS_ROWID to
interpret rowid contents. The package functions extract and provide infor-
mation on the four rowid elements listed above.
The rows of some tables have addresses that are not physical or perma-
nent or were not generated by Oracle. For example, the row addresses of
index-organized tables are stored in index leaves, which can move. Row-
ids of foreign tables (such as DB2 tables accessed through a gateway) are
not standard Oracle rowids.
Oracle uses universal rowids (urowids) to store the addresses of index-
organized and foreign tables. Index-organized tables have logical urowids
and foreign tables have foreign urowids. Both types of urowid are stored
in the ROWID pseudocolumn (as are the physical rowids of heap-orga-
nized tables).
Oracle creates logical rowids based on the primary key of the table.
The logical rowids do not change as long as the primary key does not
change. The ROWID pseudocolumn of an index-organized table has a
data type of UROWID. You can access this pseudocolumn as you would
the ROWID pseudocolumn of a heap-organized table (i.e., by using a
SELECT…ROWID statement). If you want to store the rowids of an index-
organized table, then you can define a column of type UROWID for the
table and retrieve the value of the ROWID pseudocolumn into that column.
Relational
 Schemas Implementation and Reverse Engineering 381

4.2.2.3.3  Other Data Types

Most men and women, by birth or nature, lack the means to advance
in wealth or power, but all have the ability to advance in knowledge.
—Pythagoras

Oracle provides the following additional 23 data types too:

Ø BFILE: locators to large binary files stored outside the db; enables
byte stream I/O access to external LOBs residing on the db server;
maximum size is 4 GB; when deleting such a value from the db, the
corresponding OS file is not deleted;
Ø VARRAYS: homogeneous vectors (i.e., all of their elements are of
a same data type) of variable size that should be declared before
using them (as table columns, but also as an object type attribute,
or a PL/SQL variable, parameter, or function return type); stored
either in-line (as part of the row data) or out of line (in a LOB),
depending on its size; however, if you specify separate storage
characteristics for a varray, then Oracle stores it out of line, regard-
less of its size;
Ø OBJECT: user defined object-relational abstract data types, mod-
eling real-world entities (e.g., purchase orders, addresses, apart-
ments, etc.) that application programs deal with; any object type is
a schema object with the following three components:
ü name, which identifies the object type uniquely within that
schema;
ü attributes, which are Oracle built-in or other user-defined
types modeling the structure of the real-world entity;
ü methods, which are db stored functions or procedures written
in PL/SQL, or written in some PL (e.g., C or Java) and stored
externally, implementing the operations that the application
can perform on that real-world entity;
Ø REF: container for an object identifier101 (pointer to an object102);

101
An object identifier (OID) uniquely identifies an Oracle object and enables you to reference it from
other objects or from db table columns.
102
When a REF value points to a nonexistent object, it is said to be “dangling”; a dangling REF is dif-
ferent from a null REF; to determine whether a REF is dangling or not, use the condition IS [NOT]
DANGLING.
382 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

Ø NESTED TABLE: whenever a nested table appears as the type of

a column in a relational table or as an attribute of the underlying
object type of an object table, Oracle stores all of the nested table
data in a single table, which it associates with the enclosing rela-
tional or object table; you can use both built-in and user-defined
data types for defining nested table columns;
Ø XMLType: stores XML data103; you can choose to store it in a
CLOB column, as binary XML (stored internally as a CLOB), or
object relationally; you can also register the schema (using the
DBMS_XMLSCHEMA package) and create a table or column con-
forming to it: in this case, Oracle stores the XML data in underlying
object-relational columns by default, but you can specify storage in
a CLOB or binary XML column even for schema-based data; que-
ries and DML on XMLType columns operate the same regardless
of the storage mechanism;
Ø URIType: object supertype with three subtypes (DBURIType,
HTTPURIType, and XDBURIType, which are related by an
inheritance hierarchy); you can create columns of this type and
store in them any of its subtypes instances; Oracle provides a
URIFactory package, which can create and return instances of
the various subtypes of the URITypes; the package analyzes the
URL string, identifies the type of URL (HTTP, DBURI, etc.), and
creates an instance of the corresponding subtype;
ü DBURIType: stores DBURIRef values, which reference data
inside or outside the db and access it consistently; DBURIRef
values use an XPath-like representation to reference data104:
if you imagine the db as an XML tree, then you would see the
tables, rows, and columns as elements in the XML document
and the DBURIRef is an XPath expression over this virtual
XML document; using this model, you can reference, for exam-
ple, data stored in CLOB columns and expose them as URLs to

You can also create tables and views of XMLType.

103

For example, in order to reference the Salary value of an employee having x = 205, stored in an
104

HR XML file under the tags EMPLOYEES/ROW, the corresponding value is: /HR/EMPLOYEES/
ROW[x=205]/Salary
Relational
 Schemas Implementation and Reverse Engineering 383

the external world; to create a DBURI instance, the URL must

begin with the prefix /oradb;105
ü HTTPURIType: stores URLs to external web pages or files;
Oracle accesses these files using HTTP;
ü XDBURIType: expose documents in the XML database hier-
archy as URIs that can be embedded in any URIType column
in a table; XDBURIType consists of a URL, which comprises
the hierarchical name of the XML document to which it refers
and an optional fragment representing the XPath syntax; the
fragment is separated from the URL part by a pound sign (#)106.
The following 3 spatial data types are available only if the Oracle Spa-
tial and Graph is installed107:
Ø SDO_GEOMETRY: the geometric description of a spatial object is
stored in a single row, in a single column of object type SDO_
GEOMETRY, in a user-defined table; any table that has a column of
type SDO_GEOMETRY must have a primary key; such tables are
called geometry tables;
Ø SDO_TOPO_GEOMETRY: describes a topology geometry, which
is stored in a single row, in a single column of object type SDO_
TOPO_GEOMETRY, in a user-defined geometry table;
Ø SDO_GEORASTER: in the GeoRaster object-relational model,
a raster grid or image object is stored in a single row, in a single
column of object type SDO_GEORASTER, in a user-defined table;
such tables are called GeoRaster tables.
The following 7 media data types provide compliance with the ISO-
IEC 13249–5 Still Image standard, commonly referred to as SQL/MM
StillImage:
Ø SI_StillImage: represents digital images with inherent image
characteristics such as height, width, and format;
Ø SI_Color: encapsulates color values;
105
For example, URIFactory.getURI(‘/oradb/HR/EMPLOYEES’) would create a DBU-
RIType instance, URIFactory.getUri(‘/sys/ schema’) would create an XDBURIType
instance, etc.
106
For example, /home/oe/doc.xml#/orders/order_item
107
Oracle Spatial and Graph is designed to make spatial data management easier and more natural to
users of location-enabled, geographic information system (GIS), and geoimaging applications. After
spatial data is stored in an Oracle db, you can easily manipulate, retrieve, and relate it to all other data
stored in that db.
384 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

Ø SI_AverageColor: characterizes an image by its average

color;
Ø SI_ColorHistogram: characterizes an image by the relative
frequencies of the colors exhibited by samples of the raw image;
Ø SI_PositionalColor: given an image divided into n by m
rectangles, the SI_PositionalColor object type represents the fea-
ture that characterizes an image by the n by m most significant
colors of the rectangles;
Ø SI_Texture: characterizes an image by the size of repeating
items (coarseness), brightness variations (contrast), and predomi-
nant direction (directionality);
Ø SI_FeatureList: list containing up to four of the image features
represented by the preceding object types (SI_AverageColor,
SI_Color-Histogram, SI_PositionalColor, and SI_
Texture), where each feature is associated with a feature weight.
The following three ANY data types provide highly flexible modeling
of procedure parameters and table columns where the actual type is not
known. These data types let you dynamically encapsulate and access type
descriptions, data instances, and sets of data instances of any other SQL
type. These types have both OCI and PL/SQL interfaces for construction
and access.
Ø ANYTYPE: type descriptions of any named SQL type or unnamed
transient type;
Ø ANYDATA: instances of a given data type, containing both data and a
description of its actual type; allows storing heterogeneous values in
a single column; values can be both built-in and user-defined types;
Ø ANYDATASET: descriptions of a given data type plus a set of data
instances of that type; can be used as a procedure parameter data
type where such flexibility is needed; values can be both built-in
and user-defined types.

4.2.2.3.4  Computed Columns

Above the cloud with its shadow is the star with its light.
—Pythagoras
Relational
 Schemas Implementation and Reverse Engineering 385

Oracle computed columns are referred to as virtual and are defined using
the predicate AS. Their data type may be explicitly stated or implicitly
derived by Oracle from the corresponding expression. Here are the main
characteristics of and restrictions on virtual columns:
• Indexes defined against virtual columns are equivalent to function-
based indexes.
• Virtual columns can be referenced in the WHERE clause of updates
and deletes, but they cannot be manipulated by SQL DML state-
ments.
• Tables containing virtual columns can still be eligible for result
caching.
• Functions in expressions must be deterministic at the time of table
creation, but can subsequently be recompiled and made nondeter-
ministic without invalidating the corresponding virtual columns. In
such cases the following steps must be taken after the function is
recompiled:
ü Constraints on the virtual column must be disabled and reen-
abled.
ü Indexes on the virtual column must be rebuilt.
ü Materialized views that access the virtual column must be fully
refreshed.
ü The result cache must be flushed if cached queries have
accessed the virtual column.
ü Table statistics must be regathered.
• Virtual columns are not supported for index-organized, external,
object, cluster, or temporary tables.
• Expressions used in virtual column definitions have the following
restrictions:
ü cannot refer to another virtual column by name;
ü can only refer to columns defined in the same table;
ü if they refer to a deterministic user-defined function, they can-
not be used as a partitioning key column;
ü their output must be a scalar value: it cannot return an Ora-
cle supplied data type, a user-defined type, or LOB or LONG
RAW.
386 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

4.2.2.3.5  Data Types Usage Overview

Do not say a little in many words, but a great deal in few!

—Pythagoras

Table 4.5 shows how to implement in Oracle your most frequently needed
data types.
To conclude this subsection, Table 4.6 presents the rest of the data
types provided by Oracle.

4.2.2.4  Other Oracle 12c Particularities

Be careful about virtual relationships with artificially intelligent

pieces of software.
—Larry Ellison

Beware that Oracle triggers are very slow. For example, as Oracle does
not accept using the SysDate system function (which returns the current
system calendar date and time) in constraints, although such constraints
might be enforced through triggers, we prefer to use instead additional
virtual columns and associated functions, as they are four times faster than
equivalent triggers.
For example, in order to detect whether or not the values that users
would like to store for years is greater than the current year, we first declare
a Year_in_future function (based on SysDate, see Subsection 4.3.2); then,
this function is used to define virtual columns in all tables that need such
a domain or tuple constraint, for rejecting storing implausible data in the
associated actual columns (see examples in Subsection 4.3.2).
Note that, fortunately, this type of solution is possible only due to the
existence of a bug in Oracle: virtual columns can be based only on deter-
ministic functions, whereas, in fact, Year_in_future is not deterministic,
because is based on SysDate; the helping bug is that, however, Oracle
accepts the false DETERMINISTIC definition keyword for it. Obviously,
same thing is true for the other two such needed functions, Day_in_future
and Day_in_far_future.
Relational
 Schemas Implementation and Reverse Engineering 387

TABLE 4.5  Most Frequently Needed Oracle Data Types

Data type Oracle 12c implementation
AUTONUMBER(n), n < 39 (no NUMBER(n) GENERATED [ ALWAYS | BY
replication) DEFAULT [ ON NULL ] ] AS IDENTITY
[(identity_options)]
(For all previous versions:
NUMBER(n) and associated sequence and BEFORE
INSERT trigger)
AUTONUMBER(n), n [ [39; same as above, but using BINARY_DOUBLE in-
308] (no replication) stead of NUMBER
AUTONUMBER(n) for replica- NUMBER(n) or BINARY_DOUBLE and the Oracle
tion GoldenGate add-on
BOOLEAN NUMBER(1) or CHAR and your own corresponding
codifications
NAT(n), INT(n), n < 39 NUMBER(n)
RAT(n, m), CURRENCY(n), n NUMBER(n, m)
< 39
RAT(n, m), INT(n), BINARY_DOUBLE
CURRENCY(n), n [ [39; 308]
DATE/TIME [1/1/4712 BC; DATE (no scale, no time zone), TIMESTAMP (user
31/12/9999] defined scale of maximum 9 digits, no time zone),
TIMESTAMP WITH TIME ZONE (similar to
TIMESTAMP, but including time zone with respect
to UTC too), TIMESTAMP WITH LOCAL TIME
ZONE (similar to TIMESTAMP, but of local time
zone), depending on desired scale and time zone
(not) awareness
ASCII(n), n ≤ 2,000, fixed length CHAR(n)
UNICODE(n), n ≤ 2,000, fixed NCHAR(n)
length
ASCII(n), n ≤ 32,767, variable VARCHAR2(n)
length
UNICODE(n), n ≤ 32,767, vari- NVARCHAR2(n)
able length
ASCII(n), n [ [32,768; (4 CLOB
GB—1) * (db block size)]
UNICODE(n), n [ [32,768; (4 NCLOB
GB—1) * (db block size)]
User defined data type REF (TYPE)
Computed column AS
388 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

TABLE 4.6  Other Oracle Data Types

Data type Oracle 12c implementation
DATE/TIME intervals (only years and month) INTERVAL YEAR [(year_preci-
[–4712/1; 9999/12] sion)] TO MONTH
DATE/TIME intervals (only days, hours, min- INTERVAL DAY [(day_preci-
utes, and seconds, including fractions of seconds) sion)] TO SECOND [(fractional_
[0 0:0:0; 9 23:59:59.999999999] seconds_precision)]
Binary objects of up to 32767bytes RAW(n)
BLOBs up to (4 GB—1) * (db block size) BLOB
File handlers BFILE
Homogeneous vectors VARRAYS
XML (well-formatted) files XMLType
URLs of documents of a XML db XDBURIType
URLs of (sub)objects of a rdb (e.g., CLOB frag- DBURIType
ments)
URLs to external Web pages or files HTTPURIType
Union of XDBURIType, HTTPURIType, and URIType
DBURIType
GPS coordinates, imagery data, etc. SDO_GEORASTER
Geometric descriptions of 3-D objects SDO_GEOMETRY
Topologic descriptions of 3-D objects SDO_TOPO_GEOMETRY
Images data SI_StillImage
Image colors data SI_Color
Image average colors data SI_AverageColor
Raw image relative frequencies of the colors SI_ColorHistogram
exhibited by samples
Image n by m rectangles most significant colors SI_PositionalColor
Image size of repeating items (coarseness), SI_Texture
brightness variations (contrast), and predominant
direction (directionality)
Lists of weighted values of types SI_Aver- SI_FeatureList
ageColor, SI_ColorHistogram, SI_Po-
sitionalColor, and SI_Texture
Abstract data types OBJECT
Nested (not 1NF) tables (object-relational) NESTED TABLE
Type descriptions of any named SQL type or un- ANYTYPE
named transient type
Instances of pairs <data, description of corre- ANYDATA
sponding actual type>
Descriptions of a given data type plus a set of ANYDATASET
data instances of that type
Relational
 Schemas Implementation and Reverse Engineering 389

LOB columns cannot be used either in keys (be them primary or not),
or in SELECT DISTINCT, joins, GROUP BY or ORDER BY clauses,
or in the UPDATE OF clauses of AFTER UPDATE triggers. You cannot
define VARRAYs or ANYDATAs using LOB columns either.
However, you can specify a LOB attribute of an object type column in
a SELECT…DISTINCT statement, a query that uses the UNION or a MINUS
set operator, if the object type of the column has a MAP or ORDER function
defined on it.
LOB columns have full transactional support; however, you cannot
save a LOB locator in a PL/SQL or OCI variable in one transaction and
then use it in another transaction or session.
Binary file LOBs (pointed by the BFILE data type) do not participate
in transactions and are not recoverable: the underlying OS should provide
their integrity and durability. DBAs must ensure that the corresponding
external file exists and that Oracle processes have OS read permissions
on them: BFILE enables only read-only support of such large binary files
(you cannot modify or replicate such a file: Oracle provides only APIs to
access file data). The primary interfaces that can be used to access file data
are the DBMS_LOB package and the OCI.

4.2.3 ORACLE MYSQL 5.7

Everyone thought the acquisition strategy was extremely risky

because no one had ever done it successfully.
In other words, it was innovative.
—Larry Ellison

MySQL 5.7 is robust, well-performing, and well-featured, although it is a

free product. Since it has been acquired by Oracle, there are also exten-
sions to it, as well as a much more powerful Enterprise Edition that are
commercial products.
MySQL 5.7 handles terabyte-sized dbs (but the code can also be com-
piled in a reduced version suitable for hand-held and embedded devices),
which are also accessible through ODBC and ADO.
Generally, MySQL 5.7 adheres to the pure SQL’99 ANSI standard,
with both higher standards (like the geometry spatial data types, the
390 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

high-availability db clustering using the NDBCLUSTER storage engine,

the XML functions that support most of the W3C XPath standard) and
nonstandard extensions (like the STRAIGHT_JOIN operator, which
is an INNER one, but forcing the read of its left operand first, the DO
statement, which is a slightly faster SELECT as it does not return any
results, the FLUSH, RESET, SET, SHOW, LOAD DATA INFILE, and
EXPLAIN SELECT/DELETE/INSERT/REPLACE/UPDATE statements,
the RENAME, CHANGE, and IGNORE clauses of the ALTER TABLE state-
ment, the REPLACE statement that can be used instead of a DELETE fol-
lowed by an INSERT, the capability of inserting several lines with one
INSERT … VALUES and of dropping several tables with one DROP, the
REGEXP and NOT REGEXP extended regular expression operators, the
CONCAT() and CHAR() functions with only one or more than two argu-
ments, the ORDER BY and LIMIT clauses of the UPDATE and DELETE
statements, the DELAYED clause of the INSERT and REPLACE state-
ments, the LOW_PRIORITY clause of the INSERT, REPLACE, DELETE,
and UPDATE statements, etc.).
MySQL 5.7 runs on the Unix/Linux, MS Windows, Apple Mac OS X,
Solaris, OpenSolaris, and FreeBSD OSs.

4.2.3.1  Identifiers and Literals

Nothing is particularly hard if you divide it into small jobs.

—Henry Ford

MySQL object names may have at most 64 UNICODE BMP characters,

generally are not case sensitive for tables (depending on the host OS: in
Linux, they are) and all other object types, except for triggers; you can
freely use both upper and lower case (and MySQL stores them in UTF-8
UNICODE, exactly as you defined them), but always consistently in a
same statement.
The only other reserved characters for names are the ASCII NULL
(U+0000) and the supplementary characters (U+10000 and higher). If they
are quoted, you may use the full UNICODE BMP, except for NULL, as
well as naturals; otherwise, you can only use the ASCII basic Latin letters,
digits, ‘$’, ‘_’, and the extended characters between U+0080 and U+FFFF.
Relational
 Schemas Implementation and Reverse Engineering 391

Date and time constants are generally enclosed in single quotation

marks, but they can also be used simply as numbers wherever the context
is clear to MySQL (e.g., 20150714 is interpreted as July 14, 2015).
Text strings are enclosed in either apostrophes (single) or double quo-
tation marks. You can use quotation marks within text strings by prefixing
them with the escape character ‘\’ or by doubling them; for example, the
text string “““quote”””’s value is “quote”.

4.2.3.2  Databases and Tables

The only real security that a man can have in this world is
a reserve of knowledge, experience and ability.
—Henry Ford

You can define temporary tables with CREATE TEMPORARY TABLE…

Computed tables are called views and can be created with the CRE-
ATE VIEW … AS SELECT … SQL DDL statement; they cannot be
created through the MySQL Workbench’s GUI too, as it doesn’t have a
QBE-type facility. You cannot use temporary tables within views. Views
do not accept parameters either.

4.2.3.3  Data Types

Thinking is the hardest work there is, which is probably the reason
why so few engage in it.
—Henry Ford

4.2.3.3.1  Numeric Data Types

All is Number.
—Pythagoras

BOOL and BOOLEAN are aliases for TINYINT.

392 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

MySQL provides the following 9 numeric data types (if not otherwise
specified, n = 1 by default):
Ø TINYINT[(n)] [UNSIGNED] = [–128; 127] (signed) or [0; 255]
(unsigned);
Ø SMALLINT [(n)] [UNSIGNED] = [–32,768; 32,767] (signed) or
[0; 65535] (unsigned);
Ø MEDIUMINT [(n)] [UNSIGNED] = [–8,388,608; 8,388,607]
(signed) or [0; 16,777,215] (unsigned);
Ø INT [(n)] [UNSIGNED] = [–2,147,483,648; 2,147,483,647]
(signed) or [0; 4,294,967,295] (unsigned); synonym: INTEGER;
Ø BIGINT [(n)] [UNSIGNED] = [–9,223,372,036,854,775,808;
9,223,372,036,854,775,807](signed)or[0;18,446,744,073,709,551,615]
(unsigned); synonym: INT8;
Ø DECIMAL [(n[, m])] [UNSIGNED] = [–1065; 1065–1] (signed) or
[0; 1065–1] (unsigned), 0 ≤ n ≤ 65, 0 ≤ m ≤ 30; by default, n = 10, m
= 0; precision for arithmetic calculations is 65 digits; synonyms:
NUMERIC, FIXED;
Ø FLOAT [(n[, m])] [UNSIGNED] = [–3.402823466 × 1038;
−3.402823466 × 1038] (signed) or [0; −3.402823466 × 1038]
(unsigned), 0 ≤ n, m ≤ 38 (3.402823466 × 1038 is only theoretical:
actually, depending on the host hardware and software platform, it
can be less than that); no default: if n and/or m are not specified,
their actual values depend on the host hardware platform; precision
for arithmetic calculations is some 7 scale digits; unfortunately,
sometimes, you might get unexpected results, because calculations
are never actually done with FLOAT;
Ø DOUBLE [(n[, m])] [UNSIGNED] = [–1.7976931348623157
× 10308; 1.7976931348623157 × 10308] (signed) or [0;
1.7976931348623157 × 10308] (unsigned), 0 ≤ n, m ≤ 308
(1.7976931348623157 × 10308 is only theoretical: actually, depend-
ing on the host hardware and software platform, it can be less than
that); no default: if n and/or m are not specified, their actual values
depend on the host hardware platform; precision for arithmetic cal-
culations is some 15 scale digits; synonyms: DOUBLE PRECI-
SION, REAL;
Relational
 Schemas Implementation and Reverse Engineering 393

Ø FLOAT (n) [UNSIGNED] = (iif n > 53, unexpected results, (iif 0

≤ n ≤ 24 FLOAT [UNSIGNED], DOUBLE [UNSIGNED])); pro-
vided only for ODBC compatibility.
Actually, all numeric calculations are done with either BIGINT or
DOUBLE; consequently, you should not use unsigned big integers larger
than 9,223,372,036,854,775,807 (63 bits), except with bit functions. If
you do that, some of the last digits in the result may be wrong because of
rounding errors when converting a BIGINT value to a DOUBLE.
However, MySQL correctly handles BIGINT in the following 4 cases:
ü When using integers to store large unsigned values in a BIGINT
column.
ü In MIN(col_name) or MAX(col_name), where col_name refers to a
BIGINT column.
ü When both operands are integers.
ü When storing it using a string: in such cases, MySQL performs a
string-to-number conversion that involves no intermediate double-
precision representation.
Note that when the result of arithmetic operations on two integers
would be outside the BIGINT range, unexpected values within this range
are obtained instead.
Autonumber surrogate keys can be obtained from integer data types by
adding the AUTO_INCREMENT reserved word in their definition.
SERIAL is an alias for BIGINT UNSIGNED NOT NULL AUTO_
INCREMENT UNIQUE.
SERIAL DEFAULT VALUE in the definition of an integer column is
an alias for NOT NULL AUTO_INCREMENT UNIQUE.
You can get the latest auto-increment value that was generated by the
system with SELECT LAST_INSERT_ID();
There are 5 DATE/TIME data types (date format is ‘YYYY-MM-DD’;
the optional, default 0 fsp value –that can range from 0 to 6 digits– speci-
fies the fractional seconds precision):
Ø DATE = [1000–01–01; 9999–12–31];
Ø DATETIME[(fsp)] = [1000–01–01 00:00:00; 9999–12–31
23:59:59.999999];
Ø TIMESTAMP[(fsp)] = [1970–01–01 00:00:01.000000 UTC;
2038–01–19 03:14:07.999999 UTC];
394 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

Ø TIME[(fsp)] = [–838:59:59; 838:59:59];

Ø YEAR[(4)] = [1000; 9999].
Automatic initialization and updating to the current date and time
for DATETIME columns can be specified using the DEFAULT and ON
UPDATE column definition clauses.
The way the server handles TIMESTAMP definitions depends on the
value of the explicit_defaults_for_timestamp system vari-
able; by default, this variable is disabled and the server handles TIME-
STAMP as follows:
ü Unless specified otherwise, the first TIMESTAMP column in a
table is defined to be automatically set to the date and time of the
most recent modification, if not explicitly assigned a value. This
makes TIMESTAMP useful for recording timestamps of INSERT
and UPDATE operations.
ü You can also set any TIMESTAMP column to the current date and
time by assigning it a NULL value, unless it has been defined with
the NULL attribute to permit NULL values.
ü Automatic initialization and updating to the current date and
time can be specified using DEFAULT CURRENT_TIMESTAMP
and ON UPDATE CURRENT_TIMESTAMP column definition
clauses. By default, the first TIMESTAMP column has these prop-
erties, as previously noted. However, any TIMESTAMP column in
a table can be defined to have these properties.
If explicit_defaults_for_timestamp is enabled, there is no
automatic assignment of the DEFAULT CURRENT_TIMESTAMP or ON
UPDATE CURRENT_TIMESTAMP attributes to any TIMESTAMP col-
umn. They must be included explicitly in the column definition. Also, any
TIMESTAMP not explicitly declared as NOT NULL permits NULL values.
If the MySQL server is run with the MAXDB SQL mode enabled, TIME-
STAMP is identical with DATETIME; as a result, such columns use the
DATETIME display format, have the same range of values, and there is no
automatic initialization or updating to the current date and time.
Do not use two digits years (or the legacy data type YEAR(2) either),
as they are ambiguous; if, however, you do it, note that MySQL interprets
two-digit year values using these rules:
Relational
 Schemas Implementation and Reverse Engineering 395

ü Year values in the range 70–99 are converted to 1970–1999.

ü Year values in the range 00–69 are converted to 2000–2069.
MySQL allows storing dates where the day or month and day are zero
in DATE or DATETIME columns, which is very useful for applications
that need to store dates for which you may not know the exact month and/
or date. Obviously, if you store such date values, you should not expect to
get correct results from functions such as DATE_SUB() or DATE_ADD()
(that require complete dates). Moreover, MySQL also allows storing
“zero” values (‘0000–00–00’) as “dummy dates”, which is in some cases
more convenient than using NULL values and uses less data and index
space. To disallow zero year, month, and day parts in dates, enable strict
SQL mode.
You can get the current system date and time by calling the functions
Now(), CurDate(), and CurTime(). You can extract the year from a date d
with the function EXTRACT(YEAR FROM d).
There are three types of replication formats: Statement Based Replica-
tion (SBR), which replicates entire SQL statements, Row Based Replica-
tion (RBR), which replicates only the changed rows, and Mixed Based
Replication (MBR), which combines the above two ones.
SBR of AUTO_INCREMENT and TIMESTAMP values is done cor-
rectly, subject to a couple of exceptions (see bugs #45670, #45677, and
#11754117).
MySQL recommends using the new RBR with transaction-based rep-
lication using global transaction identifiers (GTIDs). When using GTIDs,
each transaction can be identified and tracked as it is committed on the
originating server and applied by any slaves. RBR with GTID replication
has only the following three restrictions:
ü updates to tables using nontransactional storage engines such as
MyISAM cannot be made in the same statement or transaction
as updates to tables using transactional storage engines such as
InnoDB;
ü CREATE TABLE … SELECT statements are not supported;
ü CREATE/DROP TEMPORARY TABLE statements are supported
only outside of any transaction and with autocommit = 1.
396 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

4.2.3.3.2  String Data Types

Wisdom thoroughly learned will never be forgotten.

—Pythagoras

There are 14 types of strings in MySQL (using, for CHAR, VARCHAR, and
TEXT ones either ASCII or UNICODE encodings, at your choice, depend-
ing on their CHARSET attribute and/or SET NAME and/or SET CHAR-
ACTER SET statements execution or the default, which is ASCII; except
for CHAR and VARCHAR, for which n is in characters, n is in bytes):
Ø CHAR(n) = UNICODE(n), 0 ≤ n ≤ 30, fixed length (right padding);
Ø VARCHAR(n) = UNICODE(n), 0 ≤ n ≤ 65,535;
Ø TINYTEXT = UNICODE(n), 0 ≤ n ≤ 255;
Ø TEXT = UNICODE(n), 0 ≤ n ≤ 65,535;
Ø MEDIUMTEXT = UNICODE(n), 0 ≤ n ≤ 16,777,215;
Ø LONGTEXT = UNICODE(n), 0 ≤ n ≤ 4,294,967,295;
Ø BINARY(n): equivalent to CHAR(n), except that there is no char-
acter set, n is in bytes (not characters) and padding is done with
binary 0 (not space);
Ø VARBINARY(n): equivalent to VARCHAR(n), except that there is
no character set and n is in bytes (not characters);
Ø TINYBLOB: binary equivalent of TINYTEXT;
Ø BLOB: binary equivalent of TEXT;
Ø MEDIUMBLOB: binary equivalent of MEDIUMTEXT;
Ø LONGBLOB: binary equivalent of LONGTEXT;
Ø ENUM: static sets of (maximum under 3,000) string literals (that
are stored in numeric coding, but showed as defined); only one of
them is stored per table cell; (example: ENUM(‘red’, ‘orange’,
‘yellow’, ‘green’, ‘blue’, ‘indigo’, ‘violet’) is stored
as TINYINT values 1, 2, 3, 4, 5, 6, and 7, respectively);
Ø SET: static sets of (maximum 64) string literals (that are stored in
numeric coding, but showed as defined); none or any combination
of them may be stored per table cell.
BLOB and TEXT data types differ from VARBINARY and VARCHAR,
respectively, in the following respects:
Relational
 Schemas Implementation and Reverse Engineering 397

ü For indexes on BLOB and TEXT columns, you must specify an

index prefix length. For CHAR and VARCHAR, a prefix length is
optional.
ü BLOB and TEXT columns cannot have DEFAULT values.
ü Because BLOB and TEXT values can be extremely long, there are
some constraints in using them:
• Only the first max_sort_length bytes of such columns
are used when grouping and sorting; the default value of max_
sort_length is 1,024; you can make more bytes significant
in sorting or grouping by increasing the value of max_sort_
length at server startup or runtime; any client can change
the value of its session max_sort_length variable with the
SQL statement SET max_sort_length = n;
• Instances of BLOB or TEXT columns in the result of a query that
is processed using a temporary table causes MySQL to use a table
on disk rather than the internal memory (because the MEMORY
storage engine does not support these data types); obviously, use
of disks incurs performance penalties, so include BLOB or TEXT
columns in the query result only if they are really needed (e.g.,
avoid using SELECT * when you don’t need all columns).
• The maximum size of a BLOB or TEXT object is determined
by its type, but the largest value you can actually transmit
between clients and servers is determined by the amount of
available memory and the size of the communications buffers;
you can change the message buffer size by changing the value
of the max_allowed_packet variable, but you must do so
for both the server and your client program (e.g., both mysql
and mysqldump enable you to change the client-side max_
allowed_packet value).
A table can have no more than 255 unique element list definitions
among its ENUM and SET columns (that are considered as a group).

4.2.3.3.3  Other Data Types

Quality means doing it right when no one is looking.

—Henry Ford
398 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

Similarly to Oracle, MySQL 5.7 also provides 8 extended (planar) geom-

etry (geospatial) spatial data types, for both single values (GEOMETRY)
and value sets (GEOMETRYCOLLECTION):
Ø GEOMETRY: geometry values of any type;
Ø POINT: geometry values of type point (0-dimensional);
Ø LINESTRING: geometry values of type line (1-dimensional);
Ø POLYGON: geometry values of type polygon (2-dimensional);
Ø GEOMETRYCOLLECTION: collections of geometry values of any
type;
Ø MULTIPOINT: collections of geometry values of type point;
Ø MULTILINESTRING: collections of geometry values of type line;
Ø MULTIPOLYGON: collections of geometry values of type polygon.

4.2.3.3.4  Computed Columns

Failure is simply the opportunity to begin again, this time more intelligently.
—Henry Ford

There are no computed columns in MySQL. If you need materialized ones,

the only solution is to define columns of the corresponding type, to design
and implement all needed triggers for computing the corresponding val-
ues whenever needed, and to make sure that db users do not write in such
columns.
You can obviously simply compute with SQL SELECT statements,
whenever needed, not stored desired calculable data.

4.2.3.3.5. Data Types Usage Overview

The oldest, shortest words—”yes” and “no”—are those which

require the most thought.
—Pythagoras

Table 4.7 shows how to implement in MySQL your most frequently needed
data types. To conclude this subsection, Table 4.8 presents the rest of the
data types provided by MySQL.
Relational
 Schemas Implementation and Reverse Engineering 399

TABLE 4.7  Most Frequently Needed MySQL Data Types

Data type MySQL 5.7 implementation
AUTONUMBER(n), n [ {1, 2} TINYINT(n) AUTO_INCREMENT
AUTONUMBER(n), n [ {3, 4} SMALLINT(n) AUTO_INCREMENT
AUTONUMBER(n), n [ {5, 6} MEDIUMINT(n) AUTO_INCREMENT
AUTONUMBER(n), n [ [7, 9] INT(n) AUTO_INCREMENT
AUTONUMBER(n), n [ [10, 19] SERIAL or BIGINT(n) UNSIGNED AUTO_
INCREMENT
AUTONUMBER(n), n [ [20, 65] DECIMAL(n) AUTO_INCREMENT
AUTONUMBER(n), n [ [66, 308] DOUBLE(n) AUTO_INCREMENT
BOOLEAN BOOL, BOOLEAN
NAT(n), n [ {1, 2} TINYINT(n) UNSIGNED
NAT(n), n [ {3, 4} SMALLINT(n) UNSIGNED
NAT(n), n [ {5, 6} MEDIUMINT(n) UNSIGNED
NAT(n), n [ {7, 8, 9} INT(n) UNSIGNED
NAT(n), n [ [10; 19] BIGINT(n) UNSIGNED
NAT(n), n [ [20; 65] DECIMAL(n) UNSIGNED
NAT(n), n [ [66; 308] DOUBLE(n) UNSIGNED
INT(n), n [ {1, 2} TINYINT(n)
INT(n), n [ {3, 4} SMALLINT(n)
INT(n), n [ {5, 6} MEDIUMINT(n)
INT(n), n [ [7, 9] INT(n)
INT(n), n [ [10; 19] BIGINT(n)
INT(n), n [ [20; 65] DECIMAL(n)
INT(n), n [ [66; 308] DOUBLE(n)
RAT(n, m), n < 67 DECIMAL(n, m)
RAT(n, m), CURRENCY(n), n [ DOUBLE(n, m)
[67, 308]
CURRENCY(n), n < 66 DECIMAL(n, 2)
CURRENCY(n), n [ [66, 308] DOUBLE(n, 2)
DATE/TIME DATE (no scale) or DATETIME[(fsp)]
(1000–01–01; 9999–12–31) (user defined scale of maximum 6 digits, default
0)
400 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

TABLE 4.7  (Continued).

Data type MySQL 5.7 implementation
DATE/TIME TIMESTAMP[(fsp)] (user defined scale of
(1970–01–01 00:00:01.000000 maximum 6 digits, default 0)
UTC;
2038–01–19 03:14:07.999999 UTC)
TIME TIME[(fsp)] (user defined scale of maximum
(–838:59:59; 838:59:59) 6 digits, default 0)

YEAR (1000; 9999) YEAR(4)

ASCII(n), n ≤ 30, fixed length CHAR(n) CHARSET CP850 (or HP8,
LATIN2, etc.)
UNICODE(n), n ≤ 30, fixed length CHAR(n) CHARSET UCS2 (or UTF8, UTF16,
UTF32, UTF8 MB4, UTF16LE)
ASCII/UNICODE(n), n ≤ 65,535; VARCHAR(n)[CHARSET see above]
variable length
ASCII/UNICODE(n), n [ [65,536; MEDIUMTEXT [CHARSET see above]
16,777,215]
ASCII/UNICODE(n), n [ [16, 77,216; LONGTEXT [CHARSET see above]
4,294,967,295]
Static sets of strings (only one of ENUM
them is stored per table cell)
Static sets of strings (none or any SET
combination of them may be stored
per table cell; not 1NF tables
anymore)

4.2.3.4  Other MySQL 5.7 Particularities

There are no big problems, there are just a lot of little problems.
—Henry Ford

In some cases, MySQL silently changes column specifications (data types,

and/or data type attributes, and/or index specifications) from those given
in CREATE TABLE or ALTER TABLE SQL DDL statements. All changes
are subject to the internal row-size limit of 65,535 bytes, which may cause
some attempts at data type changes to fail; for example:
Relational
 Schemas Implementation and Reverse Engineering 401

TABLE 4.8  Other MySQL Data Types

Data type MySQL 5.7 implementation
Binary objects of up to 30 bytes, fixed length BINARY(n)
Binary objects between 30 bytes and 64 KB VARBINARY(n)
BLOBs between 64 KB and 16 MB MEDIUMBLOB
BLOBs between 16 MB and 4 GB LONGBLOB
Geometry values of type point POINT
Geometry values of type line LINESTRING
Geometry values of type polygon POLYGON
Geometry values of any type (union of POINT, GEOMETRY
LINESTRING, and POLYGON above)
Geometry value sets of type point MULTIPOINT
Geometry value sets of type line MULTILINESTRING
Geometry value sets of type polygon MULTIPOLYGON
Geometry value sets of any type (union of GEOMETRYCOLLECTION
MULTIPOINT, MULTILINE-STRING, and
MULTIPOLYGON)

ü Columns that are part of a PRIMARY KEY are made NOT NULL,
even if not declared that way.
ü Trailing spaces are automatically deleted from ENUM and SET
member values when the table is created.
ü Certain data types used by other SQL db vendors are mapped to
corresponding MySQL types.
ü When you include a USING clause to specify an index type that
is not permitted for a given storage engine, but there is another
index type available that the engine can use without affecting query
results, the engine uses the available type.
ü If strict SQL mode is not enabled, a VARCHAR column with a
length specification greater than 65,535 is converted to TEXT, and
a VARBINARY column with a length specification greater than
65,535 is converted to BLOB. Otherwise, an error occurs in either
of these cases.
ü Specifying the CHARACTER SET binary attribute for a character
data type causes the column to be created as the corresponding
402 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

binary data type: CHAR becomes BINARY, VARCHAR becomes

VARBINARY, and TEXT becomes BLOB. For the ENUM and SET
data types, this does not occur: they are created as declared.
ü Certain other data type changes can occur if you compress a table
using myisampack.
To see whether MySQL used a data type other than the one you speci-
fied, issue a DESCRIBE or SHOW CREATE TABLE statement after creat-
ing or altering tables.
MySQL, just like DB2, also provides an improved syntax of the
INSERT … VALUES SQL DML statement that allows for inserting as
many rows as desired with only one such statement; for example, the fol-
lowing statement inserts (the first) 2 lines in table TEST_UNSIGNED_
DECIMAL (and rejects the third as only unsigned decimal numbers are
valid for this table):
INSERT INTO TEST_UNSIGNED_DECIMAL
VALUES (9.37), (+1.23), (−4.56);
Unfortunately, MySQL does not enforce check constraints: con-
sequently, you have to enforce them through BEFORE INSERT and
UPDATE triggers.

4.2.4 MS SQL SERVER 2014

The first rule of any technology used in a business is that automation

applied to an efficient operation will magnify the efficiency.
The second is that automation applied to an inefficient operation will
magnify the inefficiency.
—Bill Gates

SQL Server 2014 is a robust, well-performing and fully featured RDBMS

too. Standard components include db language and storage server, devel-
oper and ETL tools, schedulers, and replication. Other components include
OLAP, reporting, and parallel computation. Components run as Windows
NT Services. It can also be accessed through ODBC, ADO, and OLEDB.
Both MS Visual Studio.NET and Access may easily be used to develop db
applications based on SQL Server dbs.
Relational
 Schemas Implementation and Reverse Engineering 403

Besides several commercial versions, there is also a free to download

SQL Server Express version (for 10 GB dbs maximum), which can also
be redistributed with your products; it allows for 1 GB RAM and 4 CPU
cores.
SQL Server 2014 partially adheres to the SQL 2008 ANSI standard
(which includes the 1999 recursive SELECT queries, triggers, and support
for procedural and control-of-flow statements in its extended SQL T-SQL
language, as well as the 2003 and 2006 SQL/XML and identity columns),
plus its own extensions (such as its PIVOT and UNPIVOT clauses of the
SELECT statement, for (un)crosstabulations).
SQL Server main limitation is that it only runs on MS Windows and
Mac OS X OSs.

4.2.4.1  Identifiers and Literals

I really had a lot of dreams when I was a kid, and I think a great deal
of that grew out of the fact that
I had a chance to read a lot.
—Bill Gates

SQL Server object names may have at most 128 characters (116 for tem-
porary tables), are not case sensitive, but you can freely use both upper
and lower case (and SQL Server stores them exactly as you defined
them). They should start with a letter or one of the special characters
‘@’, ‘#’, and ‘_’.
Those starting with ‘@’ are considered as local variables; moreover,
‘@’ cannot be used in any other name position.
Those starting with ‘#’ are considered as temporary objects, only exist-
ing during the current work session; all those starting with “##” are global,
that is accessible to all concurrent users, while the others are local, acces-
sible only to the current user.
Money constants are prefixed (after sign, if any) with ‘$’ or some other
over 30 currency symbols available (including ‘€’, ‘£’, ‘¥’, etc.).
The only other reserved character for names is the space. However,
you may use any number of spaces in names provided you embed them
404 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

in either square brackets or double quotation marks (if portability is not a

concern, always prefer square brackets, in order to avoid confusions with
text strings).
Text strings, as well as date and time constants are enclosed in apostro-
phes (or in double quotation marks, if the QUOTED_IDENTIFIER option
is set to OFF). You can use apostrophes within text strings by doubling
them; for example, the text string ‘“quote”’’s value is ‘quote’.

4.2.4.2  Databases and Tables

If GM had kept up with technology like the computer industry has,

we would all be driving $25 cars that got 1,000 MPG.
—Bill Gates

Computed tables are called views and can be created with the CREATE
VIEW … AS SELECT … SQL DDL statement; they can also be created
and stored through the SQL Server Management Studio’s QBE-type Query
Design GUI too.
You can define temporary tables as variables, but they only exist dur-
ing the execution of the corresponding script (batch, transaction) and
you cannot use the SELECT … INTO SQL statement for populating
them.
You can also define in-memory tables, very useful for OLTP processing.
Temporary tables, either local (names prefixed by ‘#’) or global (names
prefixed by “##”) exist during the current user session or the lifetime of the
procedure in which they are declared.

4.2.4.3  Data Types

Success is a lousy teacher. It seduces smart people

into thinking they can’t lose.
—Bill Gates
Relational
 Schemas Implementation and Reverse Engineering 405

4.2.4.3.1. Numeric Data Types

Governments will always play a huge part in solving big problems.

They set public policy and are uniquely able to provide the resources
to make sure solutions reach everyone who needs them. They also fund
basic research, which is a crucial component of the innovation
that improves life for everyone.
—Bill Gates

The Boolean type is called BIT.

There are only four subsets of the integers:
Ø TINYINT = [0; 255];
Ø SMALLINT = [–32,768; 32,767];
Ø INT = [–2,147,483,648; 2,147,483,647];
Ø BIGINT = [–9,223,372,036,854,775,808;
9,223,372,036,854,775,807].
In order to define an autonumber column, you have to choose for it an
integer data type and add to its declaration the keyword IDENTITY.
If you need replication, you have to use instead the UNIQUEIDENTI-
FIER data type, which stores GUIDs (16-byte binary); such values are
unique worldwide, as they are obtained from the corresponding (world-
wide unique) network card number and the current system timestamp.
You can obtain new values for such a column through the T-SQL function
NEWID, used either as the default value (recommended choice) of the col-
umn or the corresponding inserted value; for example:
CREATE TABLE T
x UNIQUEIDENTIFIER PRIMARY KEY DEFAULT NEWID(),
TextCol VARCHAR(10));
INSERT INTO T(TextCol) VALUES (‘abc’);
or:
CREATE TABLE T (x UNIQUEIDENTIFIER PRIMARY KEY,
TextCol VARCHAR(10));
INSERT INTO T VALUES (NEWID(),’abc’);
Note that the IDENTITY property is not available for such columns,
so that you should declare them as keys (be them primary or not), in order
for the SQL Server to prevent storing duplicates for them, as there may be
406 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

several UNIQUEIDENTIFIER columns in any table. Hopefully, only one

of them may also have the ROWGUIDCOL property, used for replication.108
The UNIQUEIDENTIFIER data type has a major disadvantage: it is
relatively large compared to numeric data types, such as 4-byte integers,
which means that indexes built on them are slower. Consequently, always
use integers and the IDENTITY property when global uniqueness is not
necessary.
There are only three subsets of the rationals (both of them also stored
in binary formats):
Ø DECIMAL (DEC, NUMERIC) = [–1038 + 1; 1038 — 1];
Ø REAL (FLOAT24) = [– 3.402823 × 1038; 3.402823 × 1038];
Ø  FLOAT (DOUBLEPRECISION, FLOAT(53)) =
[– 1.79769313486232 × 10308; 1.79769313486232 × 10308].
The default (fixed) precision for DECIMAL is 18. In T-SQL, any
numeric constant including a decimal point is considered to be decimal;
for example, 123.45 is of type DECIMAL(5,2), where 5 is the precision
(total number of digits) and 2 is the scale (total number of digits after the
decimal point).
REAL and FLOAT are approximate (not fixed) precision numeric data
types. By default, FLOAT’s precision is 15 (with a mantissa of 53 digits);
you can specify any mantissa m, 1 ≤ m ≤ 53, as FLOAT(m), but for any
m ≥ 24, it is treated like the default 53, while for any m < 25 it is treated
like 24, which has a precision of only 7 digits. SQL Server uses roundup
to approximate such values. Avoid using float or real columns in WHERE
clause search conditions, especially with the = and <> operators; it is best
to limit them to only > and < comparisons.
There are two CURRENCY data types:
Ø SMALLMONEY = [– 214,748.3648; 214,748.3647];
Ø MONEY = [–922,337,203,685,477.5808;
922,337,203,685,477.5807].
Both of them have precision of a ten thousands (of the monetary unit).
Some argue that there are significant rounding errors (especially when
dividing money to money, which makes sense only when establishing
currency rates, but which is avoidable with proper conversions) when
Do not get confused, however: the ROWGUIDCOL property indicates that the values of that column
108

uniquely identify the corresponding table rows, but does not do anything to enforce uniqueness.
Relational
 Schemas Implementation and Reverse Engineering 407

using these two data types; moreover, as MONEY is one byte less than
a large DECIMAL, with up to 19 digits of precision and as most real-
world monetary calculations (up to $9.99 M) can fit in a DECIMAL(9,2),
which requires just five bytes, you can save size, worry less about round-
ing errors, and make your code more portable by using DECIMAL instead
of MONEY.
There are several DATE/TIME data types (except for DATETIMEOFF-
SET, all others date/time types ignore time zones):
Ø DATE = [01/01/1; 31/12/9999], with 1 day accuracy, 10 digits pre-
cision, no scale;
Ø DATETIME2 = [01/01/1; 31/12/9999], with 100 nanoseconds
accuracy, precision and (user defined) scale of maximum 7 digits;
Ø DATETIME = [01/01/1753; 31/12/9999], with accuracy rounded
to increments of .000, .003, or .007 seconds and fixed scale of 3
digits;
Ø SMALLDATETIME = [01/01/1900; 06/06/2079], with 1 min accu-
racy and seconds always :00; 23:59:59 is rounded to 00:00:00 of
the next day;
Ø TIME = [00:00:00; 23:59:59.9999999], with 100 nanoseconds
accuracy, precision and (user defined) scale of maximum 16 and 7
digits, respectively;
Ø DATETIMEOFFSET = [01/01/1; 31/12/9999], with 100 nanosec-
onds accuracy, precision and (user defined) scale of maximum 7
digits, and from −14:00 to +14:59 user defined offset (depending
on desired time zone);
Ø ROWVERSION (TIMESTAMP) is an automatically generated rela-
tive date and time stamp used for row version-stamping (avoid
using TIMESTAMP, which will soon be not supported anymore).
A nonnullable ROWVERSION column is semantically equivalent to a
BINARY(8) one (see Section 4.2.4.3.2). A nullable ROWVERSION col-
umn is semantically equivalent to a VARBINARY(8) one (see Section
4.2.4.3.2).
A table can have only one ROWVERSION column. Every time that a
row of such a table is modified or inserted, the incremented db rowver-
sion value is inserted into its ROWVERSION column. To get the current
408 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

rowversion value for the current db (as a VARBINARY!), use SELECT

@@DBTS.
You can add a ROWVERSION column to a table to help maintain the
integrity of the db when multiple users are updating rows simultaneously.
You may also want to know how many rows and exactly which ones were
updated, without requerying the table.
You can use the ROWVERSION column of a table to easily determine
whether any value in its rows has changed since the last time rows were
read: if any change is made to a row, its rowversion value is updated; if
no change was made to a row, its rowversion value is the same as when it
was previously read.
ROWVERSION columns are very poor candidate for prime attributes
and especially for primary key ones: any update made to a row changes the
rowversion value; if a ROWVERSION column is (part of) a primary key,
the old key value is no longer valid, and foreign keys referencing the old
value are no longer valid; moreover, if a table is referenced in a dynamic
cursor, all updates change the position of the rows in the cursor and if the
ROWVERSION column is in an index key, all updates to the data row also
generate updates of the index.
Duplicate rowversion values can be generated by using the SELECT
INTO statement in which a ROWVERSION column is in the SELECT list.
MS does not recommend using ROWVERSION in this manner.
You can obtain the system date and time from the function SysDa-
teTime(), a DATETIME2(7) one (or SysDateTimeOffset(), a DATE-
TIMEOFFSET(7) one, if you need time zones to be also included or
SysUTCDateTime(), also a DATETIME2(7), if you need the correspond-
ing UTC date and time).
If only a DATETIME is needed, you can get it from the functions Cur-
rent_Timestamp() (or GetDate()) and GetUTCDate().
The year from a date can be obtained with the Year() function.

4.2.4.3.2  String Data Types

Good engineering and good business are one in the same.

—Bill Gates
Relational
 Schemas Implementation and Reverse Engineering 409

There are nine types of strings:

Ø CHAR[(n)]—fixed length ASCII strings of maximum 8000 charac-
ters (default 1);
Ø NCHAR[(n)]—fixed length UNICODE strings of maximum 4000
characters (default 1);
Ø VARCHAR[(n)]—variable length ASCII strings of maximum 8000
characters (default 1);
Ø VARCHAR(max)—variable length ASCII strings of more than
8000 and less than 231–1 characters (2 GB storage maximum);
Ø NVARCHAR[(n)]—variable length UNICODE strings of maximum
4000 characters (default 1);
Ø NVARCHAR(max)—variable length UNICODE strings of more
than 4000 and less than 230–1 characters (2 GB storage maximum);
Ø BINARY[(n)]—fixed length binary strings of maximum 8000 bytes
(default 1);
Ø VARBINARY[(n)]—variable length binary strings of maximum
8000 characters (default 1);
Ø VARBINARY(max)—variable length binary strings of more than
8000 and less than 231–1 characters (2 GB storage maximum).
Avoid using the TEXT, NTEXT, and IMAGE data types, as they will not
be supported anymore in the near future: use VARCHAR, NVARCHAR, and
VARBINARY instead, respectively.

4.2.4.3.3  Other Data Types

I believe in innovation and that the way you get innovation is

you fund research and you learn the basic facts.
—Bill Gates

The following other six data types are provided too:

Ø HIERARCHYID
Ø XML
Ø GEOGRAPHY
Ø GEOMETRY
Ø SQL_VARIANT
Ø User-defined data types
410 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

The HIERARCHYID data type is a variable length, system data type,

for representing positions in trees. However, a column of type HIER-
ARCHYID does not automatically represent a tree: it is up to the applica-
tion to generate and assign HIERARCHYID values, in such a way that the
desired relationship between rows is reflected in the corresponding values.
Values for HIERARCHYID have the following properties:
ü are extremely compact: the average number of bits that are required
to represent a node in a tree with n nodes depends on the average
fan-out (i.e., the average number of children per node); for small
fan outs (0–7), the size is about 6*logAn bits, where A is the aver-
age fan-out; for example, a node in an organizational hierarchy of
100,000 people with an average fan-out of 6 levels takes about 38
bits; this is rounded up to 40 bits, or 5 bytes, for storage;
ü comparison is in depth-first order: given two HIERARCHYID val-
ues v and w, v < w means that v comes before w in a depth-first
traversal of the tree; indexes on HIERARCHYID data type columns
are in depth-first order, and nodes close to each other in a depth-
first traversal are stored near each other; for example, all children
of a record are stored adjacent to that record;
ü support for arbitrary insertions and deletions: by using the GetDe-
scendant() method, it is always possible to generate a sibling to the
right of any given node, to the left of any given node, or between
any two siblings; the comparison property is maintained when an
arbitrary number of nodes is inserted or deleted from the tree; most
insertions and deletions preserve the compactness property; how-
ever, insertions between two nodes will produce HIERARCHYID
values with a slightly less compact representation;
ü the encoding used in the HIERARCHYID type is limited to 892
bytes; consequently, nodes which have too many levels in their
representation to fit into 892 bytes cannot be represented by the
HIERARCHYID type.
The HIERARCHYID type logically encodes information about a single
node in a tree, by encoding the path from the root of the tree to that node.
Such a path is logically represented as a sequence of node labels of all
children orderly visited after the root.
Relational
 Schemas Implementation and Reverse Engineering 411

Just like for tree-like file management systems, a slash starts the rep-
resentation and a path that only visits the root is represented by a single
slash; for levels underneath the root, each label is encoded as a sequence of
integers separated by dots; comparison between children is performed by
comparing the integer sequences separated by dots, in dictionary order; each
level is followed by a slash (i.e., slash separates parents from their children).
For example, the following are valid HIERARCHYID paths of lengths
1, 2, 2, 3, and 3 levels respectively:
• /
• /1/
• /0.3/7/
• /1/3/
• /0.1/0.2/
Nodes can be inserted in any location. For example, nodes inserted
after /1/2/ but before /1/3/ can be represented as /1/2.5/. Nodes inserted
before 0 have logical representations as negative numbers. For example,
a node that comes before /1/1/ can be represented as /1/-1/. Nodes cannot
have leading zeros. For example, /1/1.1/ is valid, but /1/1.01/ it is not. To
prevent errors, insert nodes by using the GetDescendant() method.
Columns of type HIERARCHYID can be used on any replicated table.
The requirements for your application depend on whether replication is
a directional or bidirectional one, and on the versions of SQL Server that
are used.
XML has the following syntax:
XML([content | document] xml_schema_collection),
where:
ü content restricts the XML instance to be a well-formed XML
fragment; XML data can contain multiple zero or more elements at
the top level; text nodes are also allowed at the top level (and this
is the default behavior);
ü document restricts the XML instance to be a well-formed XML
document; XML data must have one and only one root element;
text nodes are not allowed at the top level;
ü xml_schema_collection is the name of an XML schema col-
lection; to create a typed XML column or variable, you can option-
ally specify the XML schema collection name.
412 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

The stored representation of XML data type instances size cannot exceed
2 GB. The content and document facets apply only to typed XML.
GEOGRAPHY represents data in a round-earth coordinate system and
stores it ellipsoidally (round-earth), using GPS latitude and longitude
coordinates.
GEOMETRY is a planar spatial data type, representing data in a Euclid-
ean (flat) coordinate system.
SQL_VARIANT is a data type that may store values of any SQL
Server-supported data type, except for the following 14 ones: DATE-
TIMEOFFSET, GEOGRAPHY, GEOMETRY, HIERARCHYID, IMAGE,
NTEXT, NVARCHAR(MAX), ROWVERSION, SQL_VARIANT, TEXT,
USER_DEFINED, VARBINARY(MAX), VARCHAR(MAX), and XML.
SQL_VARIANT can have a maximum length of 8,016 bytes. This
includes both the base type information and the base type value. The maxi-
mum length of the actual base type value is 8,000 bytes.
A SQL_VARIANT data type value must first be cast to its base data
type one, before participating in operations such as addition and sub-
traction. SQL_VARIANT cannot have another SQL_VARIANT as its
base type.
SQL_VARIANT can be assigned a default value. This data type can
also have NULL as its underlying value, but the NULL values will not
have an associated base type.
A unique, primary, or foreign key may include columns of type SQL_
VARIANT, but the total length of the data values that make up the key of
a specific row should not be more than the maximum length of an index,
which is 900 bytes.
A table can have any number of SQL_VARIANT columns.109
Finally, SQL Server’s users can define their own data types. Such types
should be created with the sp_addtype table definitions. They are based
on the system data types and are recommended to be used when several
tables must store the same type of data in a column and you must ensure
that these columns have exactly the same data type, length, and nullability.
When a user-defined data type is created, you must supply the follow-
ing three parameters: new data type name (unique across the db), system
data type upon which the new data type is based, and nullability (i.e.,
109
However, SQL_VARIANT cannot be used in either CONTAINSTABLE or FREETEXTTABLE.
Relational
 Schemas Implementation and Reverse Engineering 413

whether the data type allows null values or not). When nullability is not
explicitly defined, it will be assigned based on the ANSI null default set-
ting for the db or connection.
User-defined data types created in the model db exist in all new user-
defined dbs; if a data type is created in a user-defined db, then it exists only
in that db.
For example, a user-defined data type called Address could be cre-
ated based on the varchar data type in the model db (the “supreme” model)
as follows:
USE model
EXEC sp_addtype Address, ‘VARCHAR(128)’, ‘NOT
NULL’
User-defined data types are not supported in table variables.

4.2.4.3.4  Computed Columns

If your culture doesn’t like geeks, you are in real trouble.

—Bill Gates

You can also add computed columns to tables: columns that are not physi-
cally stored in the table, unless the column is marked PERSISTED. A
computed column expression can use data from other columns of its table
to calculate a value for that column.
A computed column cannot be used as a DEFAULT or FOREIGN KEY
constraint definition or with a NOT NULL constraint definition. However,
if the computed column value is defined by a deterministic expression and
the data type of the result is allowed in index columns, a computed column
can be used as a key column in an index or as part of any PRIMARY KEY
or UNIQUE constraint.
For example, if the table has integer columns A and B, the computed
column C = A + B may be indexed, but computed column D = A
+ DATEPART(dd, GETDATE()) cannot be indexed, because the value
might change in subsequent invocations.
Computed columns can be created either through the GUI of the SQL
Server Management Studio or with the AS predicate of SQL. For example,
414 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

the following SQL DDL statement creates a table PRODUCTS having the
primary key autonumber column x, a column Product for storing prod-
uct names, a QtyAvailable column for storing corresponding quantities in
stock, a UnitPrice one, and a virtual computed column InventoryValue that
calculates, for each row, the corresponding value of the product between
QtyAvailable and UnitPrice:
CREATE TABLE PRODUCTS (
x INT IDENTITY (1,1) PRIMARY KEY,
Product VARCHAR(64) NOT NULL,
QtyAvailable SMALLINT NOT NULL
CHECK (QtyAvailable BETWEEN 0 AND 100),
UnitPrice SMALLMONEY NOT NULL
CHECK (UnitPrice BETWEEN 0 AND 10000),
InventoryValue AS QtyAvailable * UnitPrice);

4.2.4.3.5  Data Types Usage Overview

Treatment without prevention is simply unsustainable.

—Bill Gates

Table 4.9 shows how to implement in SQL Server your most frequently
needed data types. To conclude this subsection, Table 4.10 presents the
rest of the data types provided by the SQL Server.

4.2.4.4  Other SQL Server 2014 Particularities

Information technology and business are becoming inextricably interwoven.

I don’t think anybody can talk meaningfully about one without talking
about the other.
—Bill Gates

SQL Server stores nulls in indexes and considers that there is only one
null value; consequently, it does not accept in keys columns that may
store nulls.
Relational
 Schemas Implementation and Reverse Engineering 415

TABLE 4.9  Most Frequently Needed SQL Server Data Types

Data type SQL Server 2014 implementation
AUTONUMBER(n), n < 3 (no TINYINT IDENTITY
replication)
AUTONUMBER(n), n [ {3, 4} (no SMALLINT IDENTITY
replication)
AUTONUMBER(n), n [ [5; 9] (no INT IDENTITY
replication)
AUTONUMBER(n), n [ [10; 18] BIGINT IDENTITY
(no replication)
AUTONUMBER(n), n [ [19; 38] DECIMAL (n, 0) IDENTITY
(no replication)
AUTONUMBER(n), n [ [39; 308] FLOAT (n) and an associated BEFORE IN-
(no replication) SERT trigger
AUTONUMBER(n), n < 38 (for UNIQUEIDENTIFIER and NEWID()
replication)
BOOLEAN BIT
NAT(n), n < 3 TINYINT
NAT(n), n [ {3, 4} SMALLINT
NAT(n), INT(n), n [ [5; 9] INT
NAT(n), INT(n), n [ [10; 18] BIGINT
NAT(n), INT(n), n [ [19; 38] DECIMAL (n, 0)
NAT(n), INT(n), n [ [39; 308] FLOAT(53)
INT(n), n < 5 SMALLINT
RAT(n, m), n < 39 DECIMAL (n, m)
RAT(n, m), INT(n), CURRENCY(n), FLOAT(53)
n [ [39; 308]
CURRENCY(n), n < 8 DECIMAL (n, 2), SMALLMONEY
CURRENCY(n), n [ [9; 38] DECIMAL (n, 2), MONEY
DATE/TIME DATE (1 day accuracy, no scale, no time zone),
(01/01/1; 31/12/9999) DATETIME2 (100 ns accuracy, user defined
scale of maximum 7 digits, no time zone), DA-
TETIMEOFFSET (similar to DATETIME2, but
user defined time zone), depending on desired
accuracy, scale, and time zone (not) awareness
DATE/TIME DATETIME2, if accuracy rounded to incre-
(01/01/1753; 31/12/9999) ments of .000, .003, or .007 seconds and fixed
scale of 3 digits are enough
416 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

TABLE 4.9  (Continued).

Data type SQL Server 2014 implementation
DATE/TIME SMALLDATETIME, if seconds do not matter
(01/01/1900; 06/06/2079) and 1 min accuracy is enough

ASCII(n), n ≤ 8,000, fixed length CHAR(n)

UNICODE(n), n ≤ 4,000, fixed NCHAR(n)
length
ASCII(n), n ≤ 8,000, variable length VARCHAR(n)
UNICODE(n), n ≤ 4,000, variable NVARCHAR(n)
length
ASCII(n), n [ [8,001; 231 – 1] VARCHAR(MAX)
UNICODE(n), n [ [4,001; 2 – 1]
30 NVARCHAR(MAX)
User defined data type EXEC sp_addtype
Computed column AS (PERSISTED)

The simplest workaround for it is to augment the ranges of these string-

type columns with as many explicit “unknownX” values as needed, and
add the corresponding NOT NULL constraints.
For example, in the CITIES table from Fig. 3.5, if you would like to
store data for small cities without actually knowing their zip code, while
enforcing the key *Country • Zip, you may use in its Zip column values as
“unknown1”, “unknown2”, etc.
Sometimes, fortunately, you need only one “unknown” value per range.
For example, in order to allow storing in the COMMUNES table of the same
Fig. 3.5 data for communes without actually knowing their state, while
enforcing the key State • Country, you may add in the STATES table from Fig.
3.4 one row per country having “unknown” in its State column. Similarly, in
order to be able to enforce its Capital key while being able to store data even
on states without actually knowing their capitals, you may add to the CITIES
table of Fig. 3.4 one row per state having “unknown” in its City column.
Unfortunately, for numeric columns similar solutions are interfering
with the corresponding domain constraints, which is unacceptable. For
example, in table CO-AUTHORS from Fig. 3.13, in order to store data for
coauthors without actually knowing their positions in the corresponding
book coauthor lists, while enforcing the key Book • PosInList (that you
have to discover when solving Exercise 3.11, as, for any book, there may
Relational
 Schemas Implementation and Reverse Engineering 417

TABLE 4.10  Other SQL Server Data Types

SQL Server 2014
Data type implementation
Binary objects of up to 8,000 bytes, fixed length BINARY(n)
Binary objects of up to 8,000 bytes, variable length VARBINARY(n)
BLOBs up to 2 GB VARBINARY(MAX)
Insert and update transactions’ unique ids ROWVERSION
Tree nodes HIERARCHYID
XML (well-formatted) files XML
GPS coordinates GEOGRAPHY
Planar Euclidean coordinates GEOMETRY
Union of all numeric (except for DATETIMEOFFSET) SQL_VARIANT
and string (except for the MAX ones) data types

not be more than one coauthor in any of its positions), a similar to the
above solution would imply relaxing the domain constraint PosInList [
[2; 16], which is unacceptable (as then you were not able to distinguish
between plausible and not plausible PosInList values).
Fortunately, in such cases no workaround is needed: for example, it
doesn’t actually matter whether values entered in PosInList are correct or
not, as long as they are plausible. For example, even if, to begin with, a
user would not exactly know whether Prof. Dan Suciu is the second or the
third coauthor of the book Data on the Web. From Relations to Semistruc-
tured Data and XML, he/she might enter him as the second one and then
enter Prof. Peter Buneman as its third one, which can be corrected later, by
interchanging their list positions.

4.2.5  MS ACCESS 2013

Technology is just a tool. In terms of getting the kids working together

and motivating them, the teacher is the most important.
—Bill Gates

Access 2013 is a robust, well-performing, and well-featured commercial

product, well suited for small businesses.
418 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

Generally, Access 2013 adheres to the pure SQL ANSI 1992 standard,
with some extensions (both standard, as the identity autonumber columns,
and nonstandard, as the very powerful Lookup feature for foreign keys or
the TRANSFORM SQL statement).
Access dbs are limited to 2 GB (but can be linked between them in any
number, so that from one db you can access a huge number of other dbs,
and not only Access ones) and are also accessible through ODBC, ADO,
and OLEDB. Besides its rdb engine, Access comes equipped with the high
level PL VBA, which embeds SQL, ADO, and DAO, with which you can
develop db applications. By also using the MS SharePoint 2013 server or
the Office 365 website as hosts, you can also develop web-based Access
2013 applications.
There are no free Access versions, but there is a freely download-
able Access 2013 Runtime rdb engine that you can distribute together
with your Access 2013 desktop applications to users that do not own full
Access versions.
Access main limitation is that it only runs on MS Windows and Mac
OS X OSs.

4.2.5.1  Identifiers and Literals

I believe that if you show people the problems and you show them the
solutions they will be moved to act.
—Bill Gates

Access object names may have at most 255 characters, are not case sensi-
tive, but you can freely use both upper and lower case (and Access stores
them exactly as you defined them).
Among other reserved characters, they do not accept ‘?’ and ‘#’.
However, you may use any reserved characters in their names (as well as
reserved words), provided you embed them in square brackets.
Text strings have to be embedded in double quotation marks, whereas
date and time constants within sharp (numeric: ‘#’) signs. You can use
double quotation marks within text strings by doubling them; for example,
the text string “““quote”””’s value is “quote”.
Relational
 Schemas Implementation and Reverse Engineering 419

4.2.5.1  Databases and Tables

Your most unhappy customers are your greatest source of learning.

—Bill Gates

As Access manages only one rdb at a time110, its SQL does not include a
CREATE DATABASE statement. The simplest way to create an Access
rdb is to use its GUI for opening a new blank db, selecting the desired path,
giving it the desired name, and saving it.
Computed tables (views) are called queries, which cannot be material-
ized (only their definition is stored): they are evaluated each time when
invoked. They can be created in SQL only through VBA and ADOX with
the CREATE VIEW … AS SELECT … statement; they can also be cre-
ated and stored through the Access’ GUI QBE-type Design View too.
You can create temporary tables too (which exists only during the
work session in which they were created; at the end of these sessions,
they are automatically deleted by the system) with CREATE TEMPO-
RARY TABLE …

4.2.5.2  Data Types

Innovation is moving at a scarily fast pace.

—Bill Gates

4.2.5.2.1  Numeric Data Types

Software is a great combination between artistry and engineering.

—Bill Gates

The Boolean type is called BIT (YESNO, LOGICAL, LOGICAL1).

There are only three subsets of the integers (all of them binary-stored):
110
In fact, through table linking, from any one such rdb you can work with data from as many other
rdbs (not necessarily MS Access or SQL Server ones, for which there is native support: any OLEDB
or ODBC data source for which there is a compatible driver is accepted); you can even have read-only
access to the linked table schemas.
420 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

Ø BYTE (INTEGER1) = [0; 255];

Ø SHORT (SMALLINT, INTEGER2) = [–32,768; 32,767];
Ø LONG (INTEGER, INT, INTEGER4) = [–2,147,483,648;
2,147,483,647].
Autonumbers may be of either the incremental, random, or replica-
tion type. When you don’t need unicity across several platforms, choose
for your surrogate key columns the COUNTER (AUTOINCREMENT) inte-
ger data type (which is an autonumbering LONG, either incremental –the
default– or random, at your choice; for the incremental one, by default,
both the initial value and the increment ones are 1; you can set them to
any other values either through GUI or with a DDL SQL statement of
the form ALTER TABLE table name ALTER COLUMN column name
COUNTER([seed], [increment]). You can switch to the random type only
through the GUI or through VBA and DAO/ADO). For replication pur-
poses, you might use the GUID (16 bytes binary value) data type. How-
ever, you have no support at all for automatic generation of unique GUID
values: you should use custom VBA methods to do it.
Moreover, Access is the only well-known RDBMS that provides
(through his GUI) a remarkable Lookup Wizard for attaching to LONG-
type foreign keys combo-boxes that may hide actual pointers (preferably to
surrogate primary key values) and display instead corresponding semanti-
cally relevant data, computed with attached SELECT SQL statements or
statically enumerated matrices. Lookups can also be attached/modified/
deleted to/from columns manually (through the Lookup tab in its GUI table
Design View) or programmatically (through VBA and ADOX or/and DAO).
Unfortunately, corresponding semantically relevant data may be anything
desired, so it is the db designer task to choose a (super)key for it, in order
for the users to be able to uniquely identify each referenced row.
There are only two subsets of the rationals (both of them also stored in
binary formats):
Ø SINGLE (IEEESINGLE, REAL, FLOAT4) = [– 3.402823 × 1038;
3.402823 × 1038];
Ø DOUBLE (IEEEDOUBLE, NUMBER, NUMERIC, FLOAT, FLOAT8)
= [– 1.79769313486232 × 10308; 1.79769313486232 × 10308].
There is a CURRENCY (MONEY) data type, stored on 8 bytes (just like
DOUBLE and DATE/TIME), which can store rationals with up to 15 digits
Relational
 Schemas Implementation and Reverse Engineering 421

before and 4 after the decimal point. By default, USD is the used currency,
but it can be changed to any other world one (either through Access option
settings or/and through the host OS’ ones).
The DATETIME (DATE, TIME, TIMESTAMP) data type is a subset
of DOUBLE and can store any second since January 1st, 100 and up to
December 31st, 9999. You can obtain the system date and time from the
function Now(), only the date from the function Date() (only the time from
Time()), and the year from a date with the Year() function.

4.2.5.2.2  String Data Types

The advance of technology is based on making it fit in so that you

don’t really even notice it, so it’s part of everyday life.
—Bill Gates

There are two types of strings (both of them stored using corresponding
actual (variable) lengths and UTF-8, for both UNICODE and ASCII):
Ø TEXT (VARCHAR, CHAR, STRING, ALPHANUMERIC) = UNI-
CODE(255);
Ø LONGTEXT (LONGCHAR, MEMO, NOTE) = UNICODE(4096).
You can search within LONGTEXT values, but you cannot group or
order by them.

4.2.5.2.3  Other Data Types

The Internet is becoming the town square for the global village of tomorrow.
—Bill Gates

There are other two data types stored in binary format as well, namely:
Ø BINARY (VARBINARY); Length(BINARY) [ [0; 510 B];
Ø LONGBINARY (OLEOBJECT, GENERAL); Length(LONGBINARY)
[ [0; 1 GB].
LONGBINARY is used only for storing OLE objects (embedded pic-
tures, audio, video, or other Binary Large OBjects (BLOBs) not over 1 GB
each).
422 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

You can also define HYPERLINK columns (stored as LONGTEXTs)

for storing (absolute) hyperlinks to local or remote files (of the Universal
Naming Convention (UNC) type \\server\share\path\filename.extension),
URLs (including file transfer type ones, be it through browser download-
ing or the File Transfer Protocol (FTP)), e-mail addresses, and/or (rela-
tive) Access’ data access page addresses.
Hyperlinks are stored in the following format: displaytext#address#
subaddress#screentip, where displaytext is the text that you want to be
displayed underlined, address (the only mandatory part) is the address of
the file, URL, etc., subaddress is a location within the document found at
address, and screentip is the text that is displayed when the mouse hovers
over the hyperlink.
In fact, you can use Access’ GUI to add, edit, and delete hyperlinks,
which you invoke by right-clicking the corresponding table cell and then
choosing the desired action.
You can define ATTACHMENT-type columns too, for storing lists of
files that users desire to be attached to table rows. A maximum of 2 GB of
data (which is the maximum size for Access dbs) can be attached per db.
Size of any individual attached file cannot exceed 256 MB. Only files of
the following types can be attached (and the ones which are not natively
compressed are compressed by Access):
Ø MS Office (.docx, .xlsx, .accdb, .pptx, etc.);
Ø text and zipped (.txt, .log, .zip, etc.);
Ø image (.bmp, .jpg, .jpeg, .jpe, .gif, .tiff, .tif, .ico, .png, .dib, .rle,
.exif, .wmf, .emf).
Some 50 file types are rejected (among which executable and tempo-
rarily ones, some interpretable ones, etc.).
Names of attached files cannot exceed 255 UNICODE characters
(including extensions) and cannot contain the ones reserved by the NTFS
file system.
Finally, besides ATTACHMENT, there are also 8 other Access data
types that can store lists (namely, MULTIPLE BYTE, DECIMAL, DOU-
BLE, GUID, INT16, INT32, SINGLE, and TEXT), all of them obviously
violating the 1NF.
Relational
 Schemas Implementation and Reverse Engineering 423

4.2.5.2.4  Computed Columns

Until we’re educating every kid in a fantastic way,

until every inner city is cleaned up, there is no shortage of things to do.
—Bill Gates

You can also add computed columns to tables, but only through Access’
GUI. Moreover, you can also add a Total row for any table and specify for
each of its columns with what aggregation function would you like Access
to compute the corresponding total (the only available function for text-
type ones being COUNT).

4.2.5.2.5  Data Types Usage Overview

Research shows that there is only half as much variation

in student achievement between schools
as there is among classrooms in the same school.
If you want your child to get the best education possible, it is actually
more important to get him assigned to a great teacher than to a great school.
—Bill Gates

Table 4.11 shows how to implement in Access your most frequently needed
data types. To conclude this subsection, Table 4.12 presents the rest of the
data types provided by Access.

4.2.5.3  Other Access 2013 Particularities

It’s fine to celebrate success, but it is more important to

heed the lessons of failure.
—Bill Gates

The ANSI SQL standard wildcards ‘_’ (any character) and ‘%’ (any char-
acter string, including the empty one) are different in Access: ‘?’ and ‘*’,
respectively.
424 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

TABLE 4.11  Most Frequently Needed Access Data Types

Data type Access 2013 implementation
AUTONUMBER(n), n < 10 (no COUNTER (AUTOINCREMENT)
replication)
AUTONUMBER(n), n [ [10 ;38] SINGLE (IEEESINGLE, REAL, FLOAT4) and
(no replication) an associated VBA BEFORE INSERT trigger
AUTONUMBER(n), n [ [39; 308] DOUBLE (IEEEDOUBLE, NUMBER, NUMERIC,
(no replication) FLOAT, FLOAT8) and an associated VBA BE-
FORE INSERT trigger
AUTONUMBER(n), n < 38 (for GUID and an associated VBA BEFORE IN-
replication) SERT trigger
BOOLEAN BIT (YESNO, LOGICAL, LOGICAL1)
NAT(n), n < 3 BYTE (INTEGER1)
NAT(n), n [ {3, 4} SHORT (SMALLINT, INTEGER2)
NAT(n), INT(n), n [ [5; 9] LONG (INTEGER, INT, INTEGER4)
NAT(n), n [ [10; 38] SINGLE (IEEESINGLE, REAL, FLOAT4)
NAT(n), n [ [39; 308] DOUBLE (IEEEDOUBLE, NUMBER, NUMERIC,
FLOAT, FLOAT8)
INT(n), n < 5 SHORT (SMALLINT, INTEGER2)
INT(n), n [ [10; 38] SINGLE (IEEESINGLE, REAL, FLOAT4)
RAT(n, m), n < 39 SINGLE (IEEESINGLE, REAL, FLOAT4)
RAT(n, m), INT(n), DOUBLE (IEEEDOUBLE, NUMBER, NUMERIC,
CURRENCY(n), n [ [39; 308] FLOAT, FLOAT8)
CURRENCY(n), n < 15 CURRENCY (MONEY)
CURRENCY(n), n [ [15; 38] SINGLE (IEEESINGLE, REAL, FLOAT4)
DATE/TIME DATETIME (DATE, TIME, TIMESTAMP)
(01/01/100; 31/12/9999)
ASCII/UNICODE(n), n < 256 TEXT (VARCHAR, CHAR, STRING, ALPHANU-
MERIC)
ASCII/UNICODE(n), n [ [256; LONGTEXT (LONGCHAR, MEMO, NOTE)
4,096]
Static sets of strings (only one of Combo-box value list Lookup, with Allow Mul-
them is stored per table cell) tiple Values = No (only through GUI or VBA)
Static sets of strings (none or Combo-box value list Lookup, with Allow Mul-
any combination of them may be tiple Values = Yes (only through GUI or VBA)
stored per table cell; not 1NF tables
anymore)
Computed column CALCULATED (only through GUI or VBA)
Relational
 Schemas Implementation and Reverse Engineering 425

TABLE 4.12  Other Access Data Types

Data type Access 2013 implementation
Binary objects of up to 510 bytes BINARY (VARBINARY)
BLOBs (including OLE) up to 1 LONGBINARY (OLEOBJECT, GENERAL)
GB
File/URL/e-mail address HYPERLINK
Embedded file lists ATTACHMENT
List of various data type values MULTIPLE BYTE, DECIMAL, DOUBLE, GUID,
INT16, INT32, SINGLE, and TEXT

In the logical atoms of the form expr [NOT] BETWEEN value1

AND value2, Access also accepts value1 values greater than value2.
Access accepts grouping and ordering by expressions too.
You can declare and use in Access PARAMETERS for queries.
Besides SELECT, you can also query in Access by using a TRANS-
FORM statement, which computes crosstabulations (just like with the MS
Excel pivoting).
The predicate DISTINCT cannot be used with the SQL aggregate func-
tions (e.g., instead of directly computing SELECT SUM(DISTINCT c)
FROM T; you have first to define a query Q: SELECT DISTINCT c AS
e FROM T; and then to compute the desired result on it, with SELECT
SUM(e) FROM Q;).
There is no LIMIT TO n ROWS clause; you can use instead the TOP n
predicate.
Check (tuple) constraints can be enforced only through GUI or/and
VBA and ADO. Same thing applies for triggers and (explicit) transactions.
You cannot attach comments to Access SQL code.

4.3  CASE STUDY: IMPLEMENTING THE PUBLIC LIBRARY RDB

INTO DB2, ORACLE, MYSQL, SQL SERVER, AND ACCESS

The world plays between a chaos full of symmetries and

a cosmos full of asymmetries.
—Lucian Blaga
426 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

In fact, generally, almost nobody is using SQL for creating or populating

tables, except for the temporary ones, as this is done much easier through
the RDBMS GUIs. However, this section is needed not only for com-
pleteness, but also because, seldom enough, we are forced in this field
to also dynamically create table schemes and manipulate their instances
through SQL.
Please note that, in fact, all powerful RDBMSs like, for example,
DB2, Oracle, SQL Server, are adding to the below requests much more
other default settings for the actual, physical implementation of tables and
indexes (also see the chapter on optimizations from the second volume of
this book). For example, here is the complete SQL statement for the cre-
ation of the BOOKS table in Oracle (see subsection 4.3.1 for the concise
statements needed in fact):

CREATE TABLE "LIBRARY_DB"."BOOKS"(

"X" NUMBER(16,0) NOT NULL ENABLE,
"FIRST_AUTHOR" NUMBER(6,0) NOT NULL ENABLE,
"BTITLE" VARCHAR2(255 BYTE) NOT NULL ENABLE,
"BYEAR" NUMBER(4,0),
"BYEAR_IN_PAST_IND" VARCHAR2(1 BYTE) GENERATED
ALWAYS AS
(CAST("LIBRARY_DB"."YEAR_IN_FUTURE" ("BYEAR")
AS VARCHAR2(1)))
VIRTUAL VISIBLE,
CONSTRAINT "BOOKS_PK" PRIMARY KEY ("X")
USING INDEX PCTFREE 10 INITRANS 2 MAXTRANS 255
COMPUTE
STATISTICS STORAGE(INITIAL 65536 NEXT 1048576
MINEXTENTS 1
MAXEXTENTS 2147483645 PCTINCREASE 0 FREELISTS
1 FREELIST GROUPS
1 BUFFER_POOL DEFAULT FLASH_CACHE DEFAULT
CELL_FLASH_CACHE
DEFAULT) TABLESPACE "USERS" ENABLE,
CONSTRAINT "BOOKS_UK1" UNIQUE ("FIRST_AUTHOR",
"BTITLE", "BYEAR")
Relational
 Schemas Implementation and Reverse Engineering 427

USING INDEX PCTFREE 10 INITRANS 2 MAXTRANS 255

COMPUTE
STATISTICS STORAGE(INITIAL 65536 NEXT 1048576
MINEXTENTS 1
MAXEXTENTS 2147483645 PCTINCREASE 0 FREELISTS
1 FREELIST GROUPS
1 BUFFER_POOL DEFAULT FLASH_CACHE DEFAULT
CELL_FLASH_CACHE
DEFAULT) TABLESPACE "USERS" ENABLE,
CONSTRAINT "BOOKS_CHK1" CHECK (BYEAR >= −2500)
ENABLE,
CONSTRAINT "BOOKS_CHK2" CHECK (BYEAR_IN_PAST_IND
= ‘Y’) ENABLE,
CONSTRAINT "BOOKS_PERSONS_FK1" FOREIGN KEY
("FIRST_AUTHOR")
REFERENCES "LIBRARY_DB"."PERSONS" ("X") ENABLE)
SEGMENT CREATION IMMEDIATE PCTFREE 10 PCTUSED 40
INITRANS 1
MAXTRANS 255 NOCOMPRESS LOGGING STORAGE(INITIAL
65536 NEXT
1048576 MINEXTENTS 1 MAXEXTENTS 2147483645
PCTINCREASE 0
FREELISTS 1 FREELIST GROUPS 1 BUFFER_POOL
DEFAULT FLASH_CACHE
DEFAULT CELL_FLASH_CACHE DEFAULT)
TABLESPACE "USERS";
CREATE OR REPLACE TRIGGER "LIBRARY_DB"."BOOKS_AUTON"
before insert on BOOKS for each row
begin
if (:new.X is null) then
select BOOKS_SEQ.nextval into :new.X from
dual;
end if;
end;
ALTER TRIGGER "LIBRARY_DB"."BOOKS_AUTON" ENABLE;
428 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

The subsections of this section are presenting how to create and popu-
late rdbs corresponding to the public library relations from Fig. 3.13 (Sub-
section 3.4.2) in IBM DB2 10, Oracle Database 12c and MySQL 5.7, MS
SQL Server 2014 and Access 2013, by applying algorithm A8 and taking
into consideration the peculiarities of these RDBMS versions described in
Section 4.2.
Due to differences between these RDBMS versions, table and column
names slightly differ: for example, because Oracle does not accept ‘-’ in
object names, table CO-AUTHORS has been renamed as CO_AUTHORS
and because Oracle stores all such names only using uppercase, column
DueReturnDate has been renamed as DUE_RETURN_DATE.

4.3.1 IBM DB2 10.5 PUBLIC LIBRARY RDB

If you only read the books that everyone else is reading,

you can only think what everyone else is thinking.
—Haruki Murakami

CREATE DATABASE Library_DB;

-- DDL statements for creating the Library_DB

tables

CREATE TABLE PERSONS (

x INT GENERATED ALWAYS AS IDENTITY,
FName VARCHAR(128),
LName VARCHAR(64) NOT NULL,
“e-mail” VARCHAR(255),
“Author?” BOOLEAN,
PRIMARY KEY (x),
CONSTRAINT PKey UNIQUE (FName, LName,
“e-mail”));

CREATE TABLE BOOKS (

Relational
 Schemas Implementation and Reverse Engineering 429

x INT GENERATED ALWAYS AS IDENTITY,

First_Author INT NOT NULL,
BTitle VARCHAR(255) NOT NULL,
BYear INT NULL CHECK(BYear BETWEEN −5000 AND
Year(CURRENT DATE)),
PRIMARY KEY (x),
CONSTRAINT BKey UNIQUE (First_Author, BTitle,
BYear),
CONSTRAINT fkFA FOREIGN KEY (First_Author)
REFERENCES PERSONS (x));

CREATE TABLE CO_AUTHORS (

x INT GENERATED ALWAYS AS IDENTITY,
Co_Author INT NOT NULL,
Book INT NOT NULL,
Pos_In_List SMALLINT NOT NULL CHECK
(Pos_In_List > 1 AND Pos_In_List < 17),
PRIMARY KEY (x),
CONSTRAINT CAAKey UNIQUE (Co_Author, Book),
CONSTRAINT CAPKey UNIQUE (Book, Pos_In_List),
CONSTRAINT fkP FOREIGN KEY (Co_Author)
REFERENCES PERSONS (x),
CONSTRAINT fkB FOREIGN KEY (Book) REFERENCES
BOOKS (x));

CREATE TABLE PUBLISHERS (

x SMALLINT GENERATED ALWAYS AS IDENTITY,
Pub_Name VARCHAR(128) NOT NULL UNIQUE,
PRIMARY KEY (x));

CREATE TABLE EDITIONS (

x INT GENERATED ALWAYS AS IDENTITY,
Publisher INT NOT NULL,
First_Book INT NOT NULL,
ETitle VARCHAR(255) NOT NULL,
EYear INT CHECK (EYear BETWEEN −2500 AND
430 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

Year(CURRENT DATE)),
PRIMARY KEY (x),
CONSTRAINT fkPub FOREIGN KEY (Publisher)
REFERENCES PUBLISHERS,
CONSTRAINT fkBk FOREIGN KEY (First_Book)
REFERENCES BOOKS,
CONSTRAINT EKey UNIQUE (First_Book, Publisher,
EYear));

CREATE TABLE VOLUMES (

x INT GENERATED ALWAYS AS IDENTITY,
Edition INT NOT NULL,
VNo SMALLINT NOT NULL CHECK
(VNo BETWEEN 1 AND 255),
VTitle VARCHAR(255),
ISBN VARCHAR(16) UNIQUE,
Price DECIMAL(8,2) NOT NULL CHECK
(Price BETWEEN 0 AND 100000),
PRIMARY KEY (x),
CONSTRAINT fkE FOREIGN KEY (Edition)
REFERENCES EDITIONS,
CONSTRAINT VNKey UNIQUE (Edition, VNo),
CONSTRAINT VTKey UNIQUE (Edition, VTitle));

CREATE TABLE VOLUMES_CONTENTS (

x INT GENERATED ALWAYS AS IDENTITY,
Volume INT NOT NULL,
Book INT NOT NULL,
Book_Pos SMALLINT NOT NULL CHECK
(Book_Pos BETWEEN 1 AND 16),
PRIMARY KEY (x),
CONSTRAINT fkV FOREIGN KEY (Volume)
REFERENCES VOLUMES,
CONSTRAINT fkVB FOREIGN KEY (Book)
REFERENCES BOOKS,
CONSTRAINT VCBKey UNIQUE (Volume, Book),
Relational
 Schemas Implementation and Reverse Engineering 431

CONSTRAINT VCPKey UNIQUE (Volume, Book_Pos));

CREATE TABLE COPIES (
x INT GENERATED ALWAYS AS IDENTITY,
Volume INT NOT NULL,
Inv_No VARCHAR(32) NOT NULL UNIQUE,
PRIMARY KEY (x),
CONSTRAINT fkS FOREIGN KEY (Volume)
REFERENCES VOLUMES);

CREATE TABLE BORROWS (

x BIGINT GENERATED ALWAYS AS IDENTITY,
Subscriber INT NOT NULL,
Borrow_Date DATE NOT NULL CHECK (Borrow_Date
BETWEEN ‘6/1/2000’ AND CURRENT DATE),
PRIMARY KEY (x),
CONSTRAINT fkSubscriber FOREIGN KEY(Subscriber)
REFERENCES PERSONS,
CONSTRAINT BRKey UNIQUE (Borrow_Date,
Subscriber));

CREATE TABLE BORROWS_LISTS (

x BIGINT GENERATED ALWAYS AS IDENTITY,
Borrow BIGINT NOT NULL,
Copy INT NOT NULL,
Due_Return_Date DATE NOT NULL CHECK (
Due_Return_Date BETWEEN ‘6/1/2000’ AND
CURRENT DATE + 300 DAYS),
Actual_Return_Date DATE CHECK (
Actual_Return_Date BETWEEN ‘6/1/2000’ AND
CURRENT DATE),
PRIMARY KEY (x),
CONSTRAINT fkBw FOREIGN KEY (Borrow)
REFERENCES BORROWS,
CONSTRAINT fkCy FOREIGN KEY (Copy)
REFERENCES COPIES,
CONSTRAINT BLKey UNIQUE (Copy, Borrow));
432 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

-- DML statements for populating the Library_DB db

NSERT INTO PERSONS (FName, LName, “e-mail”,

I
“Author?”)
VALUES (NULL, ‘Homer’, NULL, true),
(‘William’, ‘Shakespeare’, NULL, true),
(‘Peter’, ‘Buneman’, ‘[email protected]’,
true),
(‘Serge’, ‘Abiteboul’, NULL, true),
(‘Dan’, ‘Suciu’, ‘[email protected]’,
true);

NSERT INTO BOOKS(First_Author, BTitle, BYear)

I
VALUES (1, ‘Odyssey’, −700),
(2, ‘As You Like It’, 1600),
(4, ‘Data on the Web: From Relations to
Semi-structured Data and XML’, 1999);

NSERT INTO CO_AUTHORS(Book, Co_Author,

I
Pos_In_List)
VALUES (3, 3, 2), (3, 5, 3);

INSERT INTO PUBLISHERS(Pub_Name)

VALUES (‘Apple Academic Press’),

(‘Springer Verlag’),
(‘Morgan Kaufmann’), (‘Penguin Books’),
(‘Washington Square Press’);

INSERT INTO EDITIONS(Publisher, First_Book,

ETitle, EYear) VALUES
(4, 1, ‘The Odyssey translated by Robert
Fagles’, 2012),
(5, 2, ‘As You Like It’, 2011),
(3, 3, ‘Data on the Web: From Relations to
Semi-structured Data and XML’, 1999);
Relational
 Schemas Implementation and Reverse Engineering 433

INSERT INTO VOLUMES(Edition, VNo, VTitle, ISBN,

Price)
VALUES (1, 1, NULL, ‘0–670–82162–4’, 12.95),
(2, 1, NULL, ‘978–1613821114’, 9.99),
(3, 1, NULL, ‘978–1558606227’, 74.95);

INSERT INTO VOLUMES_CONTENTS(Volume, Book,

Book_Pos)
VALUES (1, 1, 1), (2, 2, 1), (3, 3, 1);

INSERT INTO COPIES(Inv_No, Volume) VALUES (‘H-O-

1’, 1),
(‘H-O-2’, 1), (‘S-AYL1–1’, 2), (‘ABS-W-1’, 3),
(‘ABS-W-2’, 3);

INSERT INTO BORROWS(Subscriber, Borrow_Date)

VALUES (5, ‘10/29/2012’);

INSERT INTO BORROWS_LISTS(Borrow, Copy,

Due_Return_Date, Actual_Return_Date)
VALUES (1, 1, ‘11/29/2012’, ‘11/23/2012’),
(1, 3, ‘12/29/2012’, NULL);

4.3.2  ORACLE DATABASE 12C PUBLIC LIBRARY RDB

I find television very educating.

Every time somebody turns on the set, I go into the other room and
read a book.
—Groucho Marx

Note that, as no desired tablespace is asked, the default user one (gener-
ally called USERS) is used by Oracle to create tables and needed indexes
into. As 12c is, for the time being, not so widely spread, the following
code does not use its newest autonumber feature, but the “classic” trigger
method needed for all of its previous versions.
434 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

-- DDL statements for creating the Library_DB

tables
CREATE SEQUENCE PERSONS_SEQ MINVALUE 1 MAXVALUE
999999
INCREMENT BY 1 NOCYCLE;
CREATE SEQUENCE BOOKS_SEQ MINVALUE 1 MAXVALUE
999999
INCREMENT BY 1 NOCYCLE;
CREATE SEQUENCE CO_AUTHORS_SEQ MINVALUE 1
MAXVALUE 9999999 INCREMENT BY 1 NOCYCLE;
CREATE SEQUENCE PUBLISHERS_SEQ MINVALUE 1
MAXVALUE 9999 INCREMENT BY 1 NOCYCLE;
CREATE SEQUENCE EDITIONS_SEQ MINVALUE 1
MAXVALUE 999999 INCREMENT BY 1 NOCYCLE;
CREATE SEQUENCE VOLUMES_SEQ MINVALUE 1
MAXVALUE 9999999 INCREMENT BY 1 NOCYCLE;
CREATE SEQUENCE VOLUMES_CONTENTS_SEQ MINVALUE 1
MAXVALUE 9999999 INCREMENT BY 1 NOCYCLE;
CREATE SEQUENCE COPIES_SEQ MINVALUE 1
MAXVALUE 99999999 INCREMENT BY 1 NOCYCLE;
CREATE SEQUENCE BORROWS_SEQ MINVALUE 1
MAXVALUE 99999999999 INCREMENT BY 1 NOCYCLE;
CREATE SEQUENCE BORROWS_LISTS_SEQ MINVALUE 1
MAXVALUE 999999999999 INCREMENT BY 1 NOCYCLE;

CREATE FUNCTION Year_in_future (year NUMBER)

RETURN VARCHAR2 DETERMINISTIC IS
begin
return case when year > extract(year from
sysdate) then ‘N’ else ‘Y’ end;
end Year_in_future;

CREATE FUNCTION Day_in_future (day varchar2)

RETURN VARCHAR2 DETERMINISTIC IS
Relational
 Schemas Implementation and Reverse Engineering 435

begin
return case when to_date(day, ‘MM/DD/YYYY’) >
sysdate then ‘N’ else ‘Y’ end;
end Day_in_future;
CREATE FUNCTION Day_in_far_future (day varchar2)
RETURN VARCHAR2
DETERMINISTIC IS
begin
return case when to_date(day, ‘MM/DD/YYYY’) >
sysdate + 300
then ‘N’ else ‘Y’
end;
end Day_in_far_future;

CREATE TABLE PERSONS (

X NUMBER(6,0) PRIMARY KEY,
FNAME VARCHAR2(128 BYTE),
LNAME VARCHAR2(64 BYTE) NOT NULL,
E_MAIL VARCHAR2(255 BYTE),
CONSTRAINT PKEY UNIQUE (E_MAIL, FNAME,
LNAME));
CREATE TRIGGER PERSONS_AUTON before insert on
PERSONS for each row
begin
if (:new.X is null) then
select PERSONS_SEQ.nextval into :new.X from
dual;
end if;
end;

CREATE TABLE BOOKS (

X NUMBER(16,0) PRIMARY KEY,
FIRST_AUTHOR NUMBER(6,0) NOT NULL,
BTITLE VARCHAR2(255 BYTE) NOT NULL,
BYEAR NUMBER(4,0) CHECK (BYEAR >= −5000),
436 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

CONSTRAINT BKEY UNIQUE (FIRST_AUTHOR, BTITLE,

BYEAR),
CONSTRAINT FK_FA FOREIGN KEY (FIRST_AUTHOR)
REFERENCES PERSONS);

ALTER TABLE BOOKS ADD (BYEAR_IN_PAST_IND AS

(cast(Year_in_future(BYEAR) AS VARCHAR2(1))));
ALTER TABLE BOOKS ADD CONSTRAINT BOOKS_CHK2 CHECK
(BYEAR_IN_PAST_IND = ‘Y’);

CREATE TRIGGER BOOKS_AUTON before insert on BOOKS

for each row
begin
if (:new.X is null) then
select BOOKS_SEQ.nextval into :new.X from
dual;
end if;
end;

CREATE TABLE CO_AUTHORS (

X NUMBER(7,0) PRIMARY KEY,
POS_IN_LIST NUMBER(2,0) NOT NULL CHECK (
POS_IN_LIST BETWEEN 2 AND 16),
CO_AUTHOR NUMBER(6,0) NOT NULL,
BOOK NUMBER(6,0) NOT NULL,
CONSTRAINT CO_A_KEY UNIQUE (BOOK, CO_AUTHOR),
CONSTRAINT FK_BOOK FOREIGN KEY (BOOK)
REFERENCES BOOKS,
CONSTRAINT FK_CO_AUTHOR FOREIGN KEY(CO_AUTHOR)
REFERENCES PERSONS);

CREATE TRIGGER CO_AUTHORS_AUTON before insert on

CO_AUTHORS
for each row
Relational
 Schemas Implementation and Reverse Engineering 437

begin
if (:new.X is null) then
select CO_AUTHORS_SEQ.nextval into :new.X
from dual;
end if;
end;

CREATE TABLE PUBLISHERS (

X NUMBER(4,0) PRIMARY KEY,
PUB_NAME VARCHAR2(128 BYTE) NOT NULL UNIQUE);
CREATE TRIGGER PUBLISHERS_AUTON before insert on
PUBLISHERS
for each row
begin
if (:new.X is null) then
select PUBLISHERS_SEQ.nextval into :new.X from
dual;
end if;
end;

CREATE TABLE EDITIONS (

X NUMBER(7,0) PRIMARY KEY,
PUBLISHER NUMBER(4,0) NOT NULL,
FIRST_BOOK NUMBER(6,0) NOT NULL,
ETITLE VARCHAR2(255 BYTE) NOT NULL,
EYEAR NUMBER(4,0) CHECK (EYEAR >= −2500),
CONSTRAINT E_KEY UNIQUE (PUBLISHER, FIRST_BOOK,
EYEAR),
CONSTRAINT FK_FIRST_BOOK FOREIGN KEY
(FIRST_BOOK) REFERENCES BOOKS,
CONSTRAINT FK_PUBLISHER FOREIGN KEY (PUBLISHER)
REFERENCES PUBLISHERS);

ALTER TABLE EDITIONS ADD (EYEAR_IN_PAST_IND AS

(cast(Year_in_future(EYEAR) AS VARCHAR2(1))));
438 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

ALTER TABLE EDITIONS ADD CONSTRAINT EDITIONS_CHK2

CHECK
(EYEAR_IN_PAST_IND = ‘Y’);

CREATE TRIGGER EDITIONS_AUTON before insert on

EDITIONS for each row
begin
if (:new.X is null) then
select EDITIONS_SEQ.nextval into :new.X from
dual;
end if;
end;

CREATE TABLE VOLUMES (

X NUMBER(7,0) PRIMARY KEY,
EDITION NUMBER(6,0) NOT NULL,
VNO NUMBER(3,0) NOT NULL CHECK (VNO BETWEEN 1
AND 255),
VTITLE VARCHAR2(255 BYTE),
ISBN VARCHAR2(16 BYTE) UNIQUE,
PRICE NUMBER(8,2) NOT NULL CHECK (Price
BETWEEN 0 AND 100000),
CONSTRAINT FK_EDITION FOREIGN KEY (EDITION)
REFERENCES EDITIONS);

CREATE TRIGGER VOLUMES_AUTON before insert on

VOLUMES for each row
begin
if (:new.X is null) then
select VOLUMES_SEQ.nextval into :new.X from
dual;
end if;
end;

CREATE TABLE VOLUMES_CONTENTS (

X NUMBER(7,0) PRIMARY KEY,
Relational
 Schemas Implementation and Reverse Engineering 439

VOLUME NUMBER(7,0) NOT NULL,

BOOK NUMBER(6,0) NOT NULL,
BOOK_POS NUMBER(3,0) NOT NULL CHECK (BOOK_POS
BETWEEN 1 AND 16),
CONSTRAINT VC_KEY UNIQUE (VOLUME, BOOK),
CONSTRAINT VOLUMES_FK_BOOKS FOREIGN KEY (BOOK)
REFERENCES BOOKS,
CONSTRAINT FK_VOLUME FOREIGN KEY (VOLUME)
REFERENCES VOLUMES);

CREATE TRIGGER VOLUMES_CONTENTS_AUTON before

insert on
VOLUMES_CONTENTS for each row
begin
if (:new.X is null) then
select VOLUMES_CONTENTS_SEQ.nextval into
:new.X from dual;
end if;
end;

CREATE TABLE COPIES (

X NUMBER(8,0) PRIMARY KEY,
INV_NO VARCHAR2(32 BYTE) NOT NULL UNIQUE,
VOLUME NUMBER(7,0) NOT NULL,
CONSTRAINT FK_COPIES_VOLUME FOREIGN KEY (VOLUME)
REFERENCES VOLUMES);

CREATE TRIGGER COPIES_AUTON before insert on

COPIES for each row
begin
if (:new.X is null) then
select COPIES_SEQ.nextval into :new.X from
dual;
end if;
end;
440 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

CREATE TABLE BORROWS (

X NUMBER(11,0) PRIMARY KEY,
SUBSCRIBER NUMBER(6,0) NOT NULL,
BORROW_DATE DATE NOT NULL CHECK (BORROW_DATE >=
TO_DATE(‘6/1/2000’, ‘MM/DD/YYYY’)),
CONSTRAINT BORROWS_KEY UNIQUE (BORROW_DATE,
SUBSCRIBER),
CONSTRAINT FK_SUBSCRIBER FOREIGN KEY
(SUBSCRIBER) REFERENCES PERSONS);

CREATE TRIGGER BORROWS_AUTON before insert on

BORROWS for each row
begin
if (:new.X is null) then
select BORROWS_SEQ.nextval into :new.X from
dual;
end if;
end;

ALTER TABLE BORROWS ADD (BDATE_IN_PAST_IND AS

(cast(Day_in_future(TO_CHAR(BORROW_DATE, ‘MM/
DD/YYY’)) AS VARCHAR2(1))));

ALTER TABLE BORROWS ADD CONSTRAINT BORROWS_CHK2

CHECK (BDATE_IN_PAST_IND = ‘Y’);

CREATE TABLE BORROWS_LISTS (

X NUMBER(12,0) PRIMARY KEY,
BORROW NUMBER(11,0) NOT NULL,
COPY NUMBER(8,0) NOT NULL,
DUE_RETURN_DATE DATE NOT NULL CHECK (
DUE_RETURN_DATE >= TO_DATE(‘6/1/2000’,
‘MM/DD/YYYY’)),
ACTUAL_RETURN_DATE DATE CHECK (
ACTUAL_RETURN_DATE >= TO_DATE(‘6/1/2000’,
‘MM/DD/YYYY’)),
Relational
 Schemas Implementation and Reverse Engineering 441

CONSTRAINT BLKEY UNIQUE (BORROW, COPY),

CONSTRAINT FK_BORROW FOREIGN KEY (BORROW)
REFERENCES BORROWS,
CONSTRAINT FK_COPY FOREIGN KEY (COPY)
REFERENCES COPIES);

ALTER TABLE BORROWS_LISTS ADD (ARDATE_IN_PAST_IND

AS
(cast(Day_in_future(To_CHAR(ACTUAL_RETURN_DATE,
‘MM/DD/YYY’)) AS VARCHAR2(1))));

ALTER TABLE BORROWS_LISTS ADD CONSTRAINT

BORROWS_LISTS_CHK3 CHECK
(ARDATE_IN_PAST_IND = ‘Y’);

ALTER TABLE BORROWS_LISTS ADD (DRDATE_IN_PAST_IND

AS
(cast(Day_in_far_future(To_CHAR(DUE_RETURN_DATE,
‘MM/DD/YYY’)) AS VARCHAR2(1))));

ALTER TABLE BORROWS_LISTS ADD CONSTRAINT

BORROWS_LISTS_CHK4 CHECK
(DRDATE_IN_PAST_IND = ‘Y’);

CREATE TRIGGER BORROWS_LISTS_AUTON before insert

on BORROWS_LISTS
for each row
begin
if (:new.X is null) then
select BORROWS_LISTS_SEQ.nextval into :new.X
from dual;
end if;
end;

-- DML statements for populating the Library_DB db

442 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

INSERT INTO PERSONS(LName) VALUES (‘Homer’);

INSERT INTO PERSONS(FName, LName) VALUES (
‘William’, ‘Shakespeare’);
INSERT INTO PERSONS(FName, LName, e_mail) VALUES
(‘Peter’, ‘Buneman’, ‘[email protected]’);
INSERT INTO PERSONS(FName, LName) VALUES
(‘Serge’, ‘Abiteboul’);
INSERT INTO PERSONS(FName, LName, e_mail) VALUES
(‘Dan’, ‘Suciu’, ‘[email protected]’);

INSERT INTO BOOKS(First_Author, BTitle, BYear)

VALUES (1, ‘Odyssey’, −700);
INSERT INTO BOOKS(First_Author, BTitle, BYear)
VALUES (2, ‘As You Like It’, 1600);
INSERT INTO BOOKS(First_Author, BTitle, BYear)
VALUES (4, ‘Data on the Web: From Relations to
Semi-structured Data and XML’, 1999);

INSERT INTO CO_AUTHORS(Book, Co_Author,

Pos_In_List)
VALUES (3, 3, 2);
INSERT INTO CO_AUTHORS(Book, Co_Author,
Pos_In_List)
VALUES (3, 5, 3);

INSERT INTO PUBLISHERS(Pub_Name) VALUES

(‘Apple Academic Press’);
INSERT INTO PUBLISHERS(Pub_Name) VALUES
(‘Springer Verlag’);
INSERT INTO PUBLISHERS(Pub_Name) VALUES (
‘Morgan Kaufmann’);
INSERT INTO PUBLISHERS(Pub_Name) VALUES (
‘Penguin Books’);
INSERT INTO PUBLISHERS(Pub_Name) VALUES (
‘Washington Square Press’);
Relational
 Schemas Implementation and Reverse Engineering 443

INSERT INTO EDITIONS(Publisher, First_Book,

ETitle, EYear)
VALUES (4, 1, ‘The Odyssey translated by
Robert Fagles’, 2012);
INSERT INTO EDITIONS(Publisher, First_Book,
ETitle, EYear)
VALUES (5, 2, ‘As You Like It’, 2011);
INSERT INTO EDITIONS(Publisher, First_Book,
ETitle, EYear)
VALUES (3, 3, ‘Data on the Web: From Relations
to Semi-structured Data and XML’, 1999);

INSERT INTO VOLUMES(Edition, VNo, VTitle, ISBN,

Price) VALUES
(1, 1, NULL, ‘0–670–82162–4’, 12.95);
INSERT INTO VOLUMES(Edition, VNo, VTitle, ISBN,
Price) VALUES
(2, 1, NULL, ‘978–1613821114’, 9.99);
INSERT INTO VOLUMES(Edition, VNo, VTitle, ISBN,
Price) VALUES
(3, 1, NULL, ‘978–1558606227’, 74.95);

INSERT INTO VOLUMES_CONTENTS(Volume, Book,

Book_Pos) VALUES
(1, 1, 1);
INSERT INTO VOLUMES_CONTENTS(Volume, Book,
Book_Pos) VALUES
(2, 2, 1);
INSERT INTO VOLUMES_CONTENTS(Volume, Book,
Book_Pos) VALUES
(3, 3, 1);

INSERT INTO COPIES(Inv_No, Volume) VALUES (

‘H-O-1’, 1);
INSERT INTO COPIES(Inv_No, Volume) VALUES (
‘H-O-2’, 1);
444 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

INSERT INTO COPIES(Inv_No, Volume) VALUES (

‘S-AYL1–1’, 2);
INSERT INTO COPIES(Inv_No, Volume) VALUES (
‘ABS-W-1’, 3);
INSERT INTO COPIES(Inv_No, Volume) VALUES (
‘ABS-W-2’, 3);

INSERT INTO BORROWS(Subscriber, Borrow_Date) VALUES

(5, TO_DATE(‘10/29/2012’, ‘MM/DD/YYYY’));
INSERT INTO BORROWS_LISTS(Borrow, Copy,
Due_Return_Date, Actual_Return_Date)
VALUES (1, 1, TO_DATE(‘11/29/2012’, ‘MM/DD/
YYYY’), TO_DATE(‘11/23/2012’, ‘MM/DD/YYYY’));
INSERT INTO BORROWS_LISTS(Borrow, Copy,
Due_Return_Date) VALUES
(1, 3, TO_DATE(‘12/29/2012’, ‘MM/DD/YYYY’));

4.3.3  ORACLE MYSQL 5.7 PUBLIC LIBRARY RDB

There is no friend as loyal as a book.

—Ernest Hemingway

CREATE DATABASE LibraryDB;

-- DDL statements for creating the LibraryDB
tables
CREATE TABLE PERSONS (
x MEDIUMINT AUTO_INCREMENT PRIMARY KEY,
FName VARCHAR(128),
LName VARCHAR(64) NOT NULL,
‘e-mail’ VARCHAR(255),
‘Author?’ BOOLEAN,
UNIQUE INDEX PKey (FName, LName, ‘e-mail’));

CREATE TABLE BOOKS (

x MEDIUMINT AUTO_INCREMENT PRIMARY KEY,
Relational
 Schemas Implementation and Reverse Engineering 445

FirstAuthor MEDIUMINT NOT NULL,

BTitle VARCHAR(255) NOT NULL,
BYear INT,
UNIQUE INDEX BKey (FirstAuthor, BTitle, BYear),
CONSTRAINT fkFA FOREIGN KEY (FirstAuthor)
REFERENCES PERSONS (x));

DELIMITER |
CREATE FUNCTION checkBYear(y INTEGER, message
VARCHAR(256))
RETURNS INTEGER DETERMINISTIC
BEGIN
IF NOT y IS NULL AND (y < −5000 OR
y > EXTRACT(Year FROM
CurDate())) THEN
SIGNAL SQLSTATE ‘ERR0R’
SET MESSAGE_TEXT := message;
END IF;
RETURN y;
END;
|
DELIMITER $$
CREATE TRIGGER CHCKI_BYear BEFORE INSERT ON BOOKS
FOR EACH ROW
BEGIN
SET @dummy := checkBYear(NEW.Byear,
CONCAT(NEW.Byear, ‘ is not a year between −5000
and ‘, EXTRACT(Year FROM
CurDate()),’!’));
END;
$$

DELIMITER $$
CREATE TRIGGER CHCKU_BYear BEFORE UPDATE ON BOOKS
FOR EACH ROW
BEGIN
446 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

SET @dummy := checkBYear(NEW.Byear,

CONCAT(NEW.Byear, ‘ is not a year between −5000
and ‘, EXTRACT(Year FROM CurDate()),’!’));
END;
$$

CREATE TABLE CO_AUTHORS (

x MEDIUMINT AUTO_INCREMENT PRIMARY KEY,
Co_Author MEDIUMINT NOT NULL,
Book MEDIUMINT NOT NULL,
PosInList TINYINT UNSIGNED NOT NULL,
UNIQUE INDEX CAAKey (Co_Author, Book),
UNIQUE INDEX CAPKey (Book, PosInList),
CONSTRAINT fkP FOREIGN KEY (Co_Author)
REFERENCES PERSONS (x),
CONSTRAINT fkB FOREIGN KEY (Book) REFERENCES
BOOKS (x));

DELIMITER |
CREATE FUNCTION checkPosInList(p INTEGER, message
VARCHAR(256))
RETURNS INTEGER DETERMINISTIC
BEGIN
IF p < 2 OR p > 16 THEN
SIGNAL SQLSTATE ‘ERROR’;
SET MESSAGE_TEXT := message;
END IF;
RETURN p;
END;
|

DELIMITER $$
CREATE TRIGGER CHCKI_PosInList BEFORE INSERT ON
CO_AUTHORS
FOR EACH ROW
BEGIN
Relational
 Schemas Implementation and Reverse Engineering 447

SET @dummy := checkPosInList(NEW.PosInList,

CONCAT(NEW.PosInList, ‘ is not a position
between 2 and 16!’));
END;
$$

DELIMITER $$
CREATE TRIGGER CHCKU_PosInList BEFORE UPDATE ON
CO_AUTHORS
FOR EACH ROW
BEGIN
SET @dummy := checkPosInList(NEW.PosInList,
 CONCAT(NEW.PosInList, ‘ is not a position
between 2 and 16!’));
END;
$$

CREATE TABLE PUBLISHERS (

x SMALLINT AUTO_INCREMENT PRIMARY KEY,
PubName VARCHAR(128) NOT NULL UNIQUE);

CREATE TABLE EDITIONS (

x INT AUTO_INCREMENT PRIMARY KEY,
Publisher SMALLINT NOT NULL,
FirstBook MEDIUMINT NOT NULL,
ETitle VARCHAR(255) NOT NULL,
EYear INT,
CONSTRAINT fkPub FOREIGN KEY (Publisher)
REFERENCES PUBLISHERS(x),
CONSTRAINT fkBk FOREIGN KEY (FirstBook)
REFERENCES BOOKS(x),
UNIQUE INDEX EKey (FirstBook, Publisher,
EYear));
DELIMITER |
448 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

CREATE FUNCTION checkEYear(y INTEGER, message

VARCHAR(256))
RETURNS INTEGER DETERMINISTIC
BEGIN
IF NOT y IS NULL AND (y < −2500 OR
y > EXTRACT(Year FROM
CurDate())) THEN
SIGNAL SQLSTATE ‘ERROR’;
SET MESSAGE_TEXT := message;
END IF;
RETURN y;
END;
|
DELIMITER $$
CREATE TRIGGER CHCKI_EYear BEFORE INSERT ON
EDITIONS FOR EACH ROW
BEGIN
SET @dummy := checkEYear(NEW.Eyear,
CONCAT(NEW.Eyear, ‘ is not a year between −2500
and ‘, EXTRACT(Year FROM CurDate()),’!’));
END;
$$

DELIMITER $$
CREATE TRIGGER CHCKU_EYear BEFORE UPDATE ON
EDITIONS FOR EACH ROW
BEGIN
SET @dummy:= checkBYear(NEW.Eyear, CONCAT(NEW.
Eyear, ‘ is not a year between −2500 and ‘,
EXTRACT(Year FROM CurDate()),’!’));
END;
$$

CREATE TABLE VOLUMES (

x INT AUTO_INCREMENT PRIMARY KEY,
Edition INT NOT NULL,
VNo TINYINT UNSIGNED NOT NULL,
Relational
 Schemas Implementation and Reverse Engineering 449

VTitle VARCHAR(255),
ISBN VARCHAR(16) UNIQUE,
Price DECIMAL(8,2) NOT NULL,
CONSTRAINT fkE FOREIGN KEY (Edition)
REFERENCES EDITIONS(x),
UNIQUE INDEX (Edition, VNo),
UNIQUE INDEX (Edition, VTitle));

DELIMITER |
CREATE FUNCTION checkVNo_Price(n INTEGER, p
DECIMAL, messageN VARCHAR(256),
messageP VARCHAR(256))
RETURNS INTEGER DETERMINISTIC
BEGIN
IF n < 1 THEN
SET @r := 1;
SIGNAL SQLSTATE ‘ERROR’;
SET MESSAGE_TEXT := messageN;
ELSEIF p < 0 OR p > 100000 THEN
SET @r := 2;
SIGNAL SQLSTATE ‘ERROR’;
SET MESSAGE_TEXT := messageP;
ELSE
SET @r := 0;
END IF;
RETURN @r;
END;
|

DELIMITER $$
CREATE TRIGGER CHCKI_VNo_Price BEFORE INSERT ON
VOLUMES FOR EACH ROW
BEGIN
SET @dummy := checkVNo_Price(NEW.VNo, NEW.
Price, CONCAT(NEW.VNo, ‘ is not a volume number
450 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

greater than 0!’), CONCAT(NEW.Price, ‘ is not a

price between 0 and 100,000!’));
END;
$$

DELIMITER $$
CREATE TRIGGER CHCKU_VNo_Price BEFORE UPDATE ON
VOLUMES FOR EACH ROW
BEGIN
SET @dummy := checkVNo_Price(NEW.VNo, NEW.
Price, CONCAT(NEW.VNo, ‘ is not a volume number
greater than 0!’),
CONCAT(NEW.Price, ‘ is not a price between
0 and 100,000!’));
END;
$$

CREATE TABLE VOLUMES_CONTENTS (

x INT AUTO_INCREMENT PRIMARY KEY,
Volume INT NOT NULL,
Book MEDIUMINT NOT NULL,
BookPos TINYINT UNSIGNED NOT NULL,
CONSTRAINT fkV FOREIGN KEY (Volume)
REFERENCES VOLUMES(x),
CONSTRAINT fkVB FOREIGN KEY (Book)
REFERENCES BOOKS(x),
UNIQUE INDEX (Volume, Book),
UNIQUE INDEX (Volume, BookPos));
DELIMITER |

CREATE FUNCTION checkBookPos(p INTEGER, message

VARCHAR(256))
RETURNS INTEGER DETERMINISTIC
BEGIN
IF p < 1 OR p > 16 THEN
SIGNAL SQLSTATE ‘ERROR’;
Relational
 Schemas Implementation and Reverse Engineering 451

SET MESSAGE_TEXT := message;

END IF;
RETURN p;
END;
|

DELIMITER $$
CREATE TRIGGER CHCKI_BookPos BEFORE INSERT ON
VOLUMES_CONTENTS
FOR EACH ROW
BEGIN
SET @dummy := checkBookPos(NEW.BookPos,
CONCAT(NEW.BookPos, ‘ is not a position
between 1 and 16!’));
END;
$$

DELIMITER $$
CREATE TRIGGER CHCKU_BookPos BEFORE UPDATE ON
VOLUMES_CONTENTS
FOR EACH ROW
BEGIN
SET @dummy: = checkBookPos(NEW.BookPos,
CONCAT(NEW.BookPos, ‘ is not a position
between 1 and 16!’));
END;
$$

CREATE TABLE COPIES (

x INT UNSIGNED AUTO_INCREMENT PRIMARY KEY,
Volume INT NOT NULL,
InvNo VARCHAR(32) NOT NULL UNIQUE,
CONSTRAINT fkS FOREIGN KEY (Volume)
REFERENCES VOLUMES(x));

CREATE TABLE BORROWS (

452 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

x BIGINT UNSIGNED AUTO_INCREMENT PRIMARY KEY,

Subscriber MEDIUMINT NOT NULL,
BorrowDate DATE NOT NULL,
CONSTRAINT fkSubscriber FOREIGN KEY (Subscriber)
REFERENCES
PERSONS(x),
UNIQUE INDEX (BorrowDate, Subscriber));
DELIMITER |

CREATE FUNCTION checkBorrowDate(d DATE, message

VARCHAR(256))
RETURNS DATE DETERMINISTIC
BEGIN
IF d < ‘2000–01–06’ OR d > CurDate() THEN
SIGNAL SQLSTATE ‘ERROR’;
SET MESSAGE_TEXT := message;
END IF;
RETURN d;
END;
|
DELIMITER $$

CREATE TRIGGER CHCKI_BorrowDate BEFORE INSERT ON

BORROWS
FOR EACH ROW
BEGIN
SET @dummy := checkBorrowDate(NEW.BorrowDate,
CONCAT(NEW.BorrowDate, ‘ is not a date
between 2000–01–06 and ‘, CurDate(),’!’));
END;
$$
DELIMITER $$

CREATE TRIGGER CHCKU_BorrowDate BEFORE UPDATE ON

BORROWS
FOR EACH ROW
Relational
 Schemas Implementation and Reverse Engineering 453

BEGIN
SET @dummy := checkBorrowDate(NEW.BorrowDate,
CONCAT(NEW.BorrowDate, ‘ is not a date
between 2000–01–06 and ‘, CurDate(),’!’));
END;
$$

CREATE TABLE BORROWS_LISTS (

x BIGINT UNSIGNED AUTO_INCREMENT PRIMARY KEY,
Borrow BIGINT UNSIGNED NOT NULL,
Copy INT UNSIGNED NOT NULL,
DueReturnDate DATE NOT NULL,
ActualReturnDate DATE,
CONSTRAINT fkBw FOREIGN KEY (Borrow)
REFERENCES BORROWS(x),
CONSTRAINT fkCy FOREIGN KEY (Copy)
REFERENCES COPIES(x),
UNIQUE INDEX (Copy, Borrow));
DELIMITER |

CREATE FUNCTION checkDue_ActualReturnDates(d

DATE, a DATE, messageD VARCHAR(256), messageA
VARCHAR(256))
RETURNS INTEGER DETERMINISTIC
BEGIN
IF d < ‘2000–01–06’ OR d > DATE_ADD(CurDate(),
INTERVAL 300 DAY)
THEN
SET @r := 1;
SIGNAL SQLSTATE ‘ERROR’;
SET MESSAGE_TEXT := messageD;
ELSEIF a < ‘2000–01–06’ OR a > CurDate() THEN
SET @r := 2;
SIGNAL SQLSTATE ‘ERROR’;
SET MESSAGE_TEXT := messageA;
ELSE
454 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

SET @r := 0;
END IF;
RETURN @r;
END;
|

DELIMITER $$
CREATE TRIGGER CHCKI_Due_ActualReturnDates BEFORE
INSERT ON BORROWS_LISTS FOR EACH ROW
BEGIN
SET @dummy := checkDue_ActualReturnDates(NEW.
DueReturnDate,
NEW.ActualReturnDate,
CONCAT(NEW.DueReturnDate,
‘ is not a date between 2000–01–06 and ‘,
DATE_ADD(CurDate(),
INTERVAL 300 DAY), ‘!’),
CONCAT(NEW.ActualReturnDate,
‘is not a date between 2000–01–06 and
‘, CurDate(),’!’));
END;
$$
DELIMITER $$

CREATE TRIGGER CHCKU_Due_ActualReturnDates BEFORE

UPDATE
ON BORROWS_LISTS FOR EACH ROW
BEGIN
SET @dummy := checkDue_ActualReturnDates(NEW.
DueReturnDate,
NEW.ActualReturnDate,
CONCAT(NEW.DueReturnDate,
‘ is not a date between 2000–01–06 and ‘,
DATE_ADD(CurDate(),
INTERVAL 300 DAY), ‘!’),
CONCAT(NEW.ActualReturnDate,
Relational
 Schemas Implementation and Reverse Engineering 455

‘ is not a date between 2000–01–06 and ‘,

CurDate(),’!’));
END;
$$

-- DML statements for populating the LibraryDB db

INSERT INTO PERSONS (FName, LName, “e-mail”,

“Author?”)
VALUES (NULL, ‘Homer’, NULL, true),
(‘William’, ‘Shakespeare’, NULL, true),
(‘Peter’, ‘Buneman’, ‘[email protected]’,
true),
(‘Serge’, ‘Abiteboul’, NULL, true),
(‘Dan’, ‘Suciu’, ‘[email protected]’,
true);

INSERT INTO BOOKS(First_Author, BTitle, BYear)

VALUES (1, ‘Odyssey’, −700),
(2, ‘As You Like It’, 1600),
(4, ‘Data on the Web: From Relations to
Semi-structured Data and XML’, 1999);

INSERT INTO CO_AUTHORS(Book, Co_Author,

Pos_In_List)
VALUES (3, 3, 2), (3, 5, 3);

INSERT INTO PUBLISHERS(Pub_Name)

VALUES (‘Apple Academic Press’), (‘Springer
Verlag’),
(‘Morgan Kaufmann’), (‘Penguin Books’),
(‘Washington Square Press’);

INSERT INTO EDITIONS(Publisher, First_Book,

ETitle, EYear) VALUES
(4, 1, ‘The Odyssey translated by Robert
Fagles’, 2012),
456 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

(5, 2, ‘As You Like It’, 2011),

(3, 3, ‘Data on the Web: From Relations to
Semi-structured Data and XML’, 1999);

INSERT INTO VOLUMES(Edition, VNo, VTitle, ISBN,

Price)
VALUES (1, 1, NULL, ‘0–670–82162–4’, 12.95),
(2, 1, NULL, ‘978–1613821114’, 9.99),
(3, 1, NULL, ‘978–1558606227’, 74.95);

INSERT INTO VOLUMES_CONTENTS(Volume, Book,

Book_Pos)
VALUES (1, 1, 1), (2, 2, 1), (3, 3, 1);

INSERT INTO COPIES(Inv_No, Volume) VALUES (‘H-O-

1’, 1), (‘H-O-2’, 1), (‘S-AYL1–1’, 2), (‘ABS-W-
1’, 3), (‘ABS-W-2’, 3);

INSERT INTO BORROWS(Subscriber, Borrow_Date)

VALUES (5, ‘10/29/2012’);

INSERT INTO BORROWS_LISTS(Borrow, Copy,

Due_Return_Date, Actual_Return_Date)
VALUES (1, 1, ‘11/29/2012’, ‘11/23/2012’),
(1, 3, ‘12/29/2012’, NULL);

4.3.4  MS SQL SERVER 2014 PUBLIC LIBRARY RDB

Books are a uniquely portable magic.

—Stephen King

CREATE DATABASE LibraryDB;

-- DDL statements for creating the Library_DB tables

CREATE TABLE PERSONS (

x INT IDENTITY PRIMARY KEY,
Relational
 Schemas Implementation and Reverse Engineering 457

FName VARCHAR(128),
LName VARCHAR(64) NOT NULL,
[e-mail] VARCHAR(255),
[Author?] BIT,
CONSTRAINT PKey UNIQUE ([e-mail], LName,
FName));

CREATE TABLE BOOKS (

x INT IDENTITY PRIMARY KEY,
FirstAuthor INT NOT NULL,
BTitle VARCHAR(255) NOT NULL,
BYear INT CHECK (BYEAR BETWEEN −5000 AND
Year(SysDateTime())),
CONSTRAINT fkFA FOREIGN KEY (FirstAuthor)
REFERENCES PERSONS,
CONSTRAINT BKey UNIQUE (BTitle, FirstAuthor,
BYear));

CREATE TABLE [CO-AUTHORS] (

x INT IDENTITY PRIMARY KEY,
[Co-Author] INT NOT NULL,
Book INT NOT NULL,
PosInList TINYINT NOT NULL
CHECK (PosInList BETWEEN 2 AND 16),
CONSTRAINT fkCA FOREIGN KEY ([Co-Author])
REFERENCES PERSONS,
CONSTRAINT fkB FOREIGN KEY (Book)
REFERENCES BOOKS,
CONSTRAINT CAAKey UNIQUE (Book, [Co-Author]),
CONSTRAINT CAPKey UNIQUE (Book, PosInList));

CREATE TABLE PUBLISHERS (

x INT IDENTITY PRIMARY KEY,
PubName VARCHAR(128) NOT NULL UNIQUE);

CREATE TABLE EDITIONS (

x INT IDENTITY PRIMARY KEY,
458 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

Publisher INT NOT NULL,

FirstBook INT NOT NULL,
ETitle VARCHAR(255) NOT NULL,
EYear INT CHECK (EYEAR BETWEEN −2500 AND
Year(SysDateTime())),
CONSTRAINT fkPub FOREIGN KEY (Publisher)
REFERENCES PUBLISHERS,
CONSTRAINT fkBk FOREIGN KEY (FirstBook)
REFERENCES BOOKS,
CONSTRAINT EKey UNIQUE (FirstBook, Publisher,
EYear));

CREATE TABLE VOLUMES (

x INT IDENTITY PRIMARY KEY,
Edition INT NOT NULL,
VNo TINYINT NOT NULL CHECK (VNo BETWEEN 1 AND
255),
VTitle VARCHAR(255),
ISBN VARCHAR(16) UNIQUE,
Price DECIMAL(8,2) NOT NULL CHECK (Price BETWEEN
0 AND 100000),
CONSTRAINT fkE FOREIGN KEY (Edition)
REFERENCES EDITIONS,
CONSTRAINT VNKey UNIQUE (Edition, VNo),
CONSTRAINT VTKey UNIQUE (Edition, VTitle));
CREATE TABLE VOLUMES_CONTENTS (
x INT IDENTITY PRIMARY KEY,
Volume INT NOT NULL,
Book INT NOT NULL,
BookPos TINYINT NOT NULL CHECK (BookPos BETWEEN 1
AND 255),
CONSTRAINT fkV FOREIGN KEY (Volume)
REFERENCES VOLUMES,
CONSTRAINT fkVB FOREIGN KEY (Book)
REFERENCES BOOKS,
CONSTRAINT VCBKey UNIQUE (Volume, Book),
Relational
 Schemas Implementation and Reverse Engineering 459

CONSTRAINT VCPKey UNIQUE (Volume, BookPos));

CREATE TABLE COPIES (
x INT IDENTITY PRIMARY KEY,
Volume INT NOT NULL,
InvNo VARCHAR(32) NOT NULL UNIQUE,
CONSTRAINT fkS FOREIGN KEY (Volume)
REFERENCES VOLUMES);
CREATE TABLE BORROWS (
x BIGINT IDENTITY PRIMARY KEY,
Subscriber INT NOT NULL,
BorrowDate DATETIME NOT NULL CHECK (BorrowDate
BETWEEN
‘6/1/2000’ AND SysDateTime()),
CONSTRAINT fkSubscriber FOREIGN KEY
(Subscriber) REFERENCES
PERSONS,
CONSTRAINT BRKey UNIQUE (BorrowDate,
Subscriber));
CREATE TABLE BORROWS_LISTS (
x BIGINT IDENTITY PRIMARY KEY,
Borrow BIGINT NOT NULL,
Copy INT NOT NULL,
DueReturnDate DATETIME NOT NULL CHECK
(DueReturnDate BETWEEN ‘6/1/2000’ AND
DATEADD (DAY, 300, SysDateTime())),
ActualReturnDate DATETIME CHECK (ActualReturnDate
BETWEEN
‘6/1/2000’ AND SysDateTime()),
CONSTRAINT fkBw FOREIGN KEY (Borrow) REFERENCES
BORROWS,
CONSTRAINT fkCy FOREIGN KEY (Copy) REFERENCES
COPIES,
CONSTRAINT BLKey UNIQUE (Copy, Borrow));

-- DML statements for populating the LibraryDB db

460 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

INSERT INTO PERSONS(LName) VALUES (‘Homer’);

INSERT INTO PERSONS(FName, LName) VALUES
(‘William’, ‘Shakespeare’);
INSERT INTO PERSONS(FName, LName, [e-mail])
VALUES (‘Peter’, ‘Buneman’, ‘[email protected]’);
INSERT INTO PERSONS(FName, LName) VALUES
(‘Serge’, ‘Abiteboul’);
INSERT INTO PERSONS(FName, LName, [e-mail])
VALUES (‘Dan’, ‘Suciu’, ‘[email protected]’);
INSERT INTO BOOKS(FirstAuthor, BTitle, BYear)
VALUES (1, ‘Odyssey’, −700);
INSERT INTO BOOKS(FirstAuthor, BTitle, BYear)
VALUES (2, ‘As You Like It’, 1600);
INSERT INTO BOOKS(FirstAuthor, BTitle, BYear)
VALUES (4, ‘Data on
the Web: From Relations to Semi-structured
Data and XML’, 1999);
INSERT INTO [CO-AUTHORS](Book, [Co-Author],
PosInList)
VALUES (3, 3, 2);
INSERT INTO [CO-AUTHORS](Book, [Co-Author],
PosInList)
VALUES (3, 5, 3);
INSERT INTO PUBLISHERS(PubName) VALUES (‘Apple
Academic Press’);
INSERT INTO PUBLISHERS(PubName) VALUES (‘Springer
Verlag’);
INSERT INTO PUBLISHERS(PubName) VALUES (‘Morgan
Kaufmann’);
INSERT INTO PUBLISHERS(PubName) VALUES (‘Penguin
Books’);
INSERT INTO PUBLISHERS(PubName) VALUES (‘Washing-
ton Square Press’);
INSERT INTO EDITIONS(Publisher, FirstBook,
ETitle, EYear)
Relational
 Schemas Implementation and Reverse Engineering 461

VALUES (4, 1, ‘The Odyssey translated by Robert

Fagles’, 2012);
INSERT INTO EDITIONS(Publisher, FirstBook,
ETitle, EYear)
VALUES (5, 2, ‘As You Like It’, 2011);
INSERT INTO EDITIONS(Publisher, FirstBook,
ETitle, EYear)
VALUES (3, 3, ‘Data on the Web: From Relations to
Semi-structured Data and XML’, 1999);
INSERT INTO VOLUMES(Edition, VNo, VTitle, ISBN,
Price)
VALUES (1, 1, NULL, ‘0–670–82162–4’, 12.95);
INSERT INTO VOLUMES(Edition, VNo, VTitle, ISBN,
Price)
VALUES (2, 1, NULL, ‘978–1613821114’, 9.99);
INSERT INTO VOLUMES(Edition, VNo, VTitle, ISBN,
Price)
VALUES (3, 1, NULL, ‘978–1558606227’, 74.95);
INSERT INTO VOLUMES_CONTENTS(Volume, Book,
BookPos)
VALUES (1, 1, 1);
INSERT INTO VOLUMES_CONTENTS(Volume, Book,
BookPos)
VALUES (2, 2, 1);
INSERT INTO VOLUMES_CONTENTS(Volume, Book,
BookPos)
VALUES (3, 3, 1);
INSERT INTO COPIES(InvNo, Volume) VALUES (‘H-O-
1’, 1);
INSERT INTO COPIES(InvNo, Volume) VALUES (‘H-O-
2’, 1);
INSERT INTO COPIES(InvNo, Volume) VALUES (‘S-
AYL1–1’, 2);
INSERT INTO COPIES(InvNo, Volume) VALUES (‘ABS-W-
1’, 3);
462 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

INSERT INTO COPIES(InvNo, Volume) VALUES (‘ABS-W-

2’, 3);
INSERT INTO BORROWS(Subscriber, BorrowDate)
VALUES (5, ‘10/29/2012’);
INSERT INTO BORROWS_LISTS(Borrow, Copy,
DueReturnDate,
ActualReturnDate) VALUES (1, 1, ‘11/29/2012’,
‘11/23/2012’);
INSERT INTO BORROWS_LISTS(Borrow, Copy,
DueReturnDate)
VALUES (1, 3, ‘12/29/2012’);

4.3.5  MS ACCESS 2013 PUBLIC LIBRARY RDB

If one cannot enjoy reading a book over and over again,

there is no use in reading it at all.
—Oscar Wilde

Especially as Access is not translating in SQL for its users the schemas of
the tables it manages, in order to create tables through SQL you need to
save any CREATE TABLE statement in a separate query and to run then
each of them.
For example, in order to create table PERSONS, you should click on
the Query Design icon of the Create ribbon menu, then click on the Close
button of the Show Table pop-up window (without selecting any table or
view), then click on the SQL View icon of the File ribbon menu, type the
corresponding DDL SQL statement in the newly open Query1 tab (see Fig.
4.2), then close it and save it (changing its default Query1 name in, for
example, createBooks), and, finally, executing this DDL query by double-
clicking on its name.
Here is the DDL SQL needed for creating all the tables of the
LibraryDB db:

CREATE TABLE PERSONS (

x COUNTER PRIMARY KEY,
FName VARCHAR(128),
Relational
 Schemas Implementation and Reverse Engineering 463

FIGURE 4.2  Creating a SQL DDL query in Access.

LName VARCHAR(64) NOT NULL,

[e-mail] VARCHAR(255),
[Author?] BIT,
CONSTRAINT PKey UNIQUE ([e-mail], LName, FName));

CREATE TABLE BOOKS (

x COUNTER PRIMARY KEY,
FirstAuthor LONG NOT NULL,
BTitle VARCHAR(255) NOT NULL,
BYear INT,
CONSTRAINT fkFA FOREIGN KEY (FirstAuthor)
REFERENCES PERSONS,
CONSTRAINT BKey UNIQUE (BTitle, FirstAuthor,
BYear));

CREATE TABLE [CO-AUTHORS] (

x COUNTER PRIMARY KEY,
[Co-Author] LONG NOT NULL,
Book LONG NOT NULL,
PosInList BYTE NOT NULL,
CONSTRAINT fkP FOREIGN KEY ([Co-Author])
REFERENCES PERSONS,
CONSTRAINT fkB FOREIGN KEY (Book) REFERENCES BOOKS,
464 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

CONSTRAINT CAAKey UNIQUE (Book, [Co-Author]),

CONSTRAINT CAPKey UNIQUE (Book, PosInList));

CREATE TABLE PUBLISHERS (

x COUNTER PRIMARY KEY,
PubName VARCHAR(128) NOT NULL UNIQUE);

CREATE TABLE EDITIONS (

x COUNTER PRIMARY KEY,
Publisher LONG NOT NULL,
FirstBook LONG NOT NULL,
ETitle VARCHAR(255) NOT NULL,
EYear INT,
CONSTRAINT fkPub FOREIGN KEY (Publisher)
REFERENCES PUBLISHERS,
CONSTRAINT fkBk FOREIGN KEY (FirstBook)
REFERENCES BOOKS,
CONSTRAINT EKey UNIQUE (FirstBook, Publisher,
EYear));

CREATE TABLE VOLUMES (

x COUNTER PRIMARY KEY,
Edition LONG NOT NULL,
VNo BYTE NOT NULL,
VTitle VARCHAR(255),
ISBN VARCHAR(16) UNIQUE,
Price CURRENCY NOT NULL,
CONSTRAINT fkE FOREIGN KEY (Edition)
REFERENCES EDITIONS,
CONSTRAINT VNKey UNIQUE (Edition, VNo),
CONSTRAINT VTKey UNIQUE (Edition, VTitle));
CREATE TABLE VOLUMES_CONTENTS (
x COUNTER PRIMARY KEY,
Volume LONG NOT NULL,
Book LONG NOT NULL,
Relational
 Schemas Implementation and Reverse Engineering 465

BookPos BYTE NOT NULL,

CONSTRAINT fkV FOREIGN KEY (Volume)
REFERENCES VOLUMES,
CONSTRAINT fkVB FOREIGN KEY (Book)
REFERENCES BOOKS,
CONSTRAINT VCBKey UNIQUE (Volume, Book),
CONSTRAINT VCPKey UNIQUE (Volume, BookPos));

CREATE TABLE COPIES (

x COUNTER PRIMARY KEY,
Volume LONG NOT NULL,
InvNo VARCHAR(32) NOT NULL UNIQUE,
CONSTRAINT fkS FOREIGN KEY (Volume)
REFERENCES VOLUMES);

CREATE TABLE BORROWS (

x COUNTER PRIMARY KEY,
Subscriber LONG NOT NULL,
BorrowDate DATE NOT NULL,
CONSTRAINT fkS FOREIGN KEY (Subscriber)
REFERENCES PERSONS,
CONSTRAINT BRKey UNIQUE (BorrowDate,
Subscriber));

CREATE TABLE BORROWS_LISTS (

x COUNTER PRIMARY KEY,
Borrow LONG NOT NULL,
Copy LONG NOT NULL,
DueReturnDate DATE NOT NULL,
ActualReturnDate DATE,
CONSTRAINT fkBw FOREIGN KEY (Borrow)
REFERENCES BORROWS,
CONSTRAINT fkCy FOREIGN KEY (Copy)
REFERENCES COPIES,
CONSTRAINT BLKey UNIQUE (Copy, Borrow));
466 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

The easiest way to add the check (tuple) and domain constraints is
through the Access GUI, by using the Validation Rule and Validation
Text properties of the corresponding table and columns, respectively (in
the corresponding tables’ Design View). Figures 4.3–4.11 show how to
enforce all the LibraryDB domain constraints.
Here is the DML SQL needed for populating the tables (that should be
stored one by one in distinct queries, just like for the DDL ones (see Fig.
4.2), as Access is only manipulating single statement SQL scripts):

INSERT INTO PERSONS(LName) VALUES ("Homer");

INSERT INTO PERSONS(FName, LName) VALUES
("William", "Shakespeare");
INSERT INTO PERSONS(FName, LName, [e-mail])
VALUES ("Peter", "Buneman", "[email protected]");
INSERT INTO PERSONS(FName, LName) VALUES
("Serge", "Abiteboul");
INSERT INTO PERSONS(FName, LName, [e-mail])
VALUES ('Dan', 'Suciu', '[email protected]');

FIGURE 4.3  Adding BYear’s domain constraint through Access GUI.

Relational
 Schemas Implementation and Reverse Engineering 467

FIGURE 4.4  Adding PosInList’s domain constraint through Access GUI.

FIGURE 4.5  Adding EYear’s domain constraint through Access GUI.

468 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

FIGURE 4.6  Adding VNo’s domain constraint through Access GUI.

FIGURE 4.7  Adding Price’s domain constraint through Access GUI.

Relational
 Schemas Implementation and Reverse Engineering 469

FIGURE 4.8  Adding BookPos’ domain constraint through Access GUI.

FIGURE 4.9  Adding BorrowDate’s domain constraint through Access GUI.

470 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

FIGURE 4.10  Adding DueReturnDate’s domain constraint through Access GUI.

FIGURE 4.11  Adding ActualReturnDate’s domain constraint through Access GUI.

Relational
 Schemas Implementation and Reverse Engineering 471

INSERT INTO BOOKS(FirstAuthor, BTitle, BYear)

VALUES (1, "Odyssey", −700);

INSERT INTO BOOKS(FirstAuthor, BTitle, BYear)

VALUES (2, "As You Like It", 1600);
INSERT INTO BOOKS(FirstAuthor, BTitle, BYear)
VALUES (4, "Data on the Web: From Relations to
Semi-structured
Data and XML", 1999);

INSERT INTO [CO-AUTHORS](Book, [Co-Author],

PosInList)
VALUES (3, 3, 2);
INSERT INTO [CO-AUTHORS](Book, [Co-Author],
PosInList)
VALUES (3, 5, 3);

INSERT INTO PUBLISHERS(PubName) VALUES ("Apple

Academic Press");
INSERT INTO PUBLISHERS(PubName) VALUES ("Springer
Verlag");
INSERT INTO PUBLISHERS(PubName) VALUES ("Morgan
Kaufmann");
INSERT INTO PUBLISHERS(PubName) VALUES ("Penguin
Books");
INSERT INTO PUBLISHERS(PubName) VALUES
("Washington Square Press");

INSERT INTO EDITIONS(Publisher, FirstBook,

ETitle, EYear)
VALUES (4, 1, "The Odyssey translated by Robert
Fagles", 2012);
INSERT INTO EDITIONS(Publisher, FirstBook,
ETitle, EYear)
VALUES (5, 2, "As You Like It", 2011);
INSERT INTO EDITIONS(Publisher, FirstBook,
ETitle, EYear)
472 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

VALUES (3, 3, "Data on the Web: From Relations to

Semi-structured Data and XML", 1999);

INSERT INTO VOLUMES(Edition, VNo, VTitle, ISBN,

Price)
VALUES (1, 1, NULL, "0–670–82162–4", 12.95);
INSERT INTO VOLUMES(Edition, VNo, VTitle, ISBN,
Price)
VALUES (2, 1, NULL, "978–1613821114", 9.99);
INSERT INTO VOLUMES(Edition, VNo, VTitle, ISBN,
Price)
VALUES (3, 1, NULL, "978–1558606227", 74.95);

INSERT INTO VOLUMES_CONTENTS(Volume, Book,

BookPos)
VALUES (1, 1, 1);
INSERT INTO VOLUMES_CONTENTS(Volume, Book,
BookPos)
VALUES (2, 2, 1);
INSERT INTO VOLUMES_CONTENTS(Volume, Book,
BookPos)
VALUES (3, 3, 1);

INSERT INTO COPIES(InvNo, Volume) VALUES ("H-O-

1", 1);
INSERT INTO COPIES(InvNo, Volume) VALUES ("H-O-
2", 1);
INSERT INTO COPIES(InvNo, Volume) VALUES ("S-
AYL1–1", 2);
INSERT INTO COPIES(InvNo, Volume) VALUES ("ABS-W-
1", 3);
INSERT INTO COPIES(InvNo, Volume) VALUES ("ABS-W-
2", 3);

INSERT INTO BORROWS(Subscriber, BorrowDate)

VALUES (5, #10/29/2012#);
Relational
 Schemas Implementation and Reverse Engineering 473

INSERT INTO BORROWS_LISTS(Borrow, Copy,

DueReturnDate, ActualReturnDate) VALUES
(1, 1, #11/29/2012#, #11/23/2012#);
INSERT INTO BORROWS_LISTS(Borrow, Copy,
DueReturnDate) VALUES (1, 3, #12/29/2012#);

4.4  THE REVERSE ENGINEERING ALGORITHM FOR

TRANSLATING ACCESS 2013 RDB SCHEMAS INTO SQL ANSI
DDL SCRIPTS (REA2013A0), A MEMBER OF REAF0'

Let’s reduce to the minimum our fixations!

—Lucian Blaga

There are very few exceptions among RDBMSs that do not provide facili-
ties for generating SQL DDL scripts to create the tables of the dbs they
are managing; the only one among the five ones considered in this book
is MS Access.
This is why we chose Access 2013 as an example for designing
algorithms of the reverse engineering family REAF0′. Obviously, as,
for example, all other four RDBMS versions considered in this book
have DDL GUI interfaces too, the algorithm presented in this subsec-
tion (let’s call it REA2013A0, although it also works for many previ-
ous Access versions) may be an example for designing own similar
algorithms for them too, although it is trivially simpler to first use their
facilities for generating DDL scripts in their own SQL idioms and then
apply much simpler algorithms to translate these scripts into ANSI
standard ones.
Not only because it is a reverse engineering algorithm, but especially
in order to first gain experience, we are starting with how to actually do it,
before summarizing what has generally to be done in the corresponding
algorithm.
In order to simplify things as much as possible, the figures that fol-
low are captured from the Access 2013 LibraryDB rdb (see Subsection
4.3.5).
474 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

For opening an Access db, just double-click on its name; note that if
you only have a Runtime Access version, you can only manipulate Access
created db applications and data (if any): you need a full Access license in
order to also modify db schemas and applications of such dbs.
When double-clicking the LibraryDB.accdb file on a computer hav-
ing a full Access 2013 license installed, the window shown in Fig. 4.12
is opened. In the db Navigation left pane, under the All Tables group, you
see all the tables belonging to the corresponding db. In order to open the
scheme of anyone of them, right-click on its name and then click on the
Design View option; then, in order to get more screen space, you can mini-
mize the Navigation left pane by clicking on its upper rightmost button
(the “«” one; to reopen anytime this pane, just click on the same button,
which, when this pane is minimized, displays “»”).
Figure 4.13 shows the scheme of an Access table: its name (in this case
BORROWS_LISTS) is displayed in the newly opened tab header; its col-
umn names are displayed in the Field Name column; their corresponding
data types are shown in the middle Data Type column; the third Descrip-
tion (optional) column is intended to store the semantics associated to
each column.
The columns that are making up the primary key of the table (if any)
have to their left a yellow small key icon: in Fig. 4.13, this is displayed
only to the left of the x column.
You can see the data subtypes and the domain constraints associated
to each column in the General left subpane under the scheme table; for
example, from Fig. 4.13, you can see that the Autonumber data type is
by default a subtype of the LONG one and, if you open the corresponding
Field Size combo-box, that the only other alternative for it is the Repli-
cation ID one (see Fig. 4.14).
If you open the New Values combo-box, you can see that Access gener-
ates autonumbers either incrementally (which is the default) or randomly
(see Fig. 4.15). From the Access documentation, you can see that both the
starting and the increment values for its incremental autonumbers are 1.
Finally for single primary keys, just like for any other single key, you
can see in Fig. 4.13 that the Indexed property is set to Yes (No Duplicates),
which means that a unique index is built on this column and enforces its
one-to-oneness.
Relational
 Schemas Implementation and Reverse Engineering 475

FIGURE 4.12  Access 2013 Navigation All Tables pane.

FIGURE 4.13  The scheme (Design View) of an Access 2013 table.

476 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

Up to this point, after consulting Tables 4.1 and 4.9 above (where we
can see that LONG is NAT(9) in Access), we may infer that algorithm
REA2013A0 should output for this table the following DDL statement:

CREATE TABLE BORROWS_LISTS (

 x INTEGER(9) GENERATED ALWAYS AS IDENTITY(1,1)
PRIMARY KEY
);

If the cursor is then moved on the second (Borrow) line of the table
from Fig. 4.3, we can see (Fig. 4.16) that this column is a LONG that does
not accept null values (because the Required property is set to Yes). Conse-
quently, the output of REA2013A0 should now be the following:

CREATE TABLE BORROWS_LISTS (

 x INTEGER(9) GENERATED ALWAYS AS IDENTITY(1,1)
PRIMARY KEY,
  Borrow INTEGER(9) NOT NULL
);

When moving the cursor on the third line (Copy, see Fig. 4.17), we can
see that this column is similar to the previous one, so that the output of
REA2013A0 should now be the following:

CREATE TABLE BORROWS_LISTS (

 x INTEGER(9) GENERATED ALWAYS AS IDENTITY(1,1)
PRIMARY KEY,
  Borrow INTEGER(9) NOT NULL,
  Copy INTEGER(9) NOT NULL
);
When moving the cursor on the fourth line (DueReturnDate, see Fig.
4.18), we can see that this column is a DATE/TIME one that does not
accept nulls either and which has an associated domain constraint, as its
Validation Rule property is not empty.
Relational
 Schemas Implementation and Reverse Engineering 477

FIGURE 4.14  The two types of Access autonumbers (LONG and REPLICATION).

FIGURE 4.15. The two types of Access autonumbering (Increment and Random).
478 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

FIGURE 4.16  An example of Access LONG required column.

FIGURE 4.17  Another example of Access LONG required column.

Relational
 Schemas Implementation and Reverse Engineering 479

FIGURE 4.18  An example of Access domain constraint (Validation Rule).

Consequently, with the same SysDate() convention used throughout

this book111 and also knowing that in Access any integer added to a DATE
value is expressed in days, the output of REA2013A0 should now be the
following:

CREATE TABLE BORROWS_LISTS (

x INTEGER(9) GENERATED ALWAYS AS IDENTITY(1,1)
PRIMARY KEY,
Borrow INTEGER(9) NOT NULL,
Copy INTEGER(9) NOT NULL,
DueReturnDate DATE NOT NULL CHECK (DueReturnDate
BETWEEN
‘01/06/2000’ AND SysDate() + 300 DAYS)
);
When moving the cursor on the fifth line (ActualReturnDate, see Fig-
ure 4.19), we can see that this column is similar to the fourth one, except
that it accepts nulls (as its Required property is set to No) and that it has

111
Unfortunately, there is no standard at least for the most frequently needed functions in dbs, although
this would be of great help for everybody.
480 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

another associated domain constraint; the output of REA2013A0 should

now be the following:

CREATE TABLE BORROWS_LISTS (

  x INTEGER(9) GENERATED ALWAYS AS IDENTITY(1,1)
PRIMARY KEY,
  Borrow INTEGER(9) NOT NULL,
  Copy INTEGER(9) NOT NULL,
  DueReturnDate DATE NOT NULL CHECK (DueReturnDate
BETWEEN
  ‘01/06/2000’ AND SysDate() + 300 DAYS),
  ActualReturnDate DATE CHECK (ActualReturnDate
BETWEEN
  ‘01/06/2000’ AND SysDate())
);

As inspection of all table’s columns is finished, we can proceed by

examining the semantic (candidate, unique) keys of the table; in Access,
they can be browsed (and/or modified) in the popup window that opens

FIGURE 4.19  Another example of an Access domain constraint.

Relational
 Schemas Implementation and Reverse Engineering 481

when clicking on the Indexes icon of the ribbon (see Fig. 4.20). Note that
all indexes are defined in this window, but only those that have the prop-
erty Unique set to Yes are enforcing keys (the other ones are of no interest
in this first volume: they will be discussed in the fourth chapter of the
second one).
In this example, there are only two indexes, named blKey and Prima-
ryKey (according to the Index Name column from Fig. 4.20), both enforc-
ing keys, the first one being made out of the columns Borrow and Copy
(in this order), and the second one, which corresponds to the primary key
of the table, only of the column x (according to the column Field Name).
Consequently, the output of REA2013A0 should now be the following:

CREATE TABLE BORROWS_LISTS (

 x INTEGER(9) GENERATED ALWAYS AS IDENTITY(1,1)
PRIMARY KEY,
 Borrow INTEGER(9) NOT NULL,
 Copy INTEGER(9) NOT NULL,
 DueReturnDate DATE NOT NULL CHECK (DueReturnDate
BETWEEN
  ‘01/06/2000’ AND SysDate() + 300 DAYS),
 ActualReturnDate DATE CHECK (ActualReturnDate
BETWEEN
  ‘01/06/2000’ AND SysDate()),
  CONSTRAINT blKey UNIQUE (Borrow, Copy)
);

As inspection of all table keys is done, you should close the Indexes
windows. Next step is to discover the referential integrity constraints
associated with this table: the simplest way to start with is to browse the
instance of the Access metacatalog table MSysRelationships, which stores
all the referential integrity constraints enforced in the db.
By default, all metacatalog tables are hidden; in order to be able to see
in the Navigation pane the system tables too, right-click the pane (e.g., to
the right of its All Tables header) and then click on Navigation Options.
Figure 4.21 shows the corresponding popup window that opens and in
which you should check the Show System Objects check-box (bottom left).
When you then click on its OK button (bottom right), this window closes
482 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

FIGURE 4.20  An example of Access (unique) keys (and indexes).

FIGURE 4.21  Access Navigation Options window.

Relational
 Schemas Implementation and Reverse Engineering 483

and you can see in the Navigation pane the system tables too (with their
icons and names, prefixed by MSys, slightly dimmed, see Fig. 4.22).
When you double-click on the MSysRelationships table icon, its
instance opens (see Fig. 4.23); right-click on the name of the szObject col-
umn and then click on the Sort A to Z option; for better viewing the table
instance, double-click the dividing lines between columns to size them
according to their content.
Column szRelationship stores the names of the corresponding referen-
tial integrity constraints; column grbit stores their subtypes (4352 for ON
UPDATE CASCADE and ON DELETE CASCADE; 4096 for ON UPDATE
RESTRICT and ON DELETE CASCADE; 256 for ON UPDATE CAS-
CADE and ON DELETE RESTRICT; and 0 for the default ON UPDATE
RESTRICT and ON DELETE RESTRICT).
Note that whenever the referenced column is an autonumber one112,
cascading updates does not make sense –as values of automatically gen-
erated numbers cannot ever be changed–, so that Access ignores in such
cases the ON UPDATE CASCADE setting); column szColumn stores the
names of the foreign keys; column szObject stores the corresponding table
names; column szReferencedColumn stores the columns that are refer-
enced by the foreign keys; column szReferencedObject stores the corre-
sponding table names.
For example, from the third and the fourth lines of the table instance
from Fig. 4.23 we can find out that table BORROWS_LISTS has two such
referential integrities (both of them of the default subtype, as grbit is 0,
named COPIESBORROWS_LISTS and BORROWSBORROWS_LISTS,
respectively), the first one associated to its foreign key Copy (which refer-
ences column x of table COPIES) and the second one to its foreign key
Borrow (which references column x of table BORROWS).
Consequently, the output of REA2013A0 should now be the following:

CREATE TABLE BORROWS_LISTS (

 x INTEGER(9) GENERATED ALWAYS AS IDENTITY(1,1)
PRIMARY KEY,
112
As it should always be, except for the cases when the corresponding table represents a subset and
its primary key is also a foreign key referencing the primary key of the superset corresponding table
(as the primary key of the subset table cannot be an autonumber, but only a LONG number referenc-
ing an autonumber); normally, only for such exceptions, you should always enforce the ON UPDATE
CASCADE option for all foreign keys referencing the primary keys of all tables that implement subsets.
484 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

FIGURE 4.22  Access Navigation Pane also showing its system tables.

FIGURE 4.23  Access MSysRelationships metacatalog table.

 
Borrow INTEGER(9) NOT NULL,
 
Copy INTEGER(9) NOT NULL,
 
DueReturnDate DATE NOT NULL CHECK (DueReturnDate
BETWEEN
Relational
 Schemas Implementation and Reverse Engineering 485

  ‘01/06/2000’ AND SysDate() + 300 DAYS),

 ActualReturnDate DATE CHECK (ActualReturnDate
BETWEEN
  ‘01/06/2000’ AND SysDate()),
  CONSTRAINT blKey UNIQUE (Borrow, Copy),
  CONSTRAINT COPIESBORROWS_LISTS FOREIGN KEY
(Copy) REFERENCES
 COPIES(x) ON UPDATE RESTRICT ON DELETE RESTRICT,
  CONSTRAINT BORROWSBORROWS_LISTS FOREIGN KEY
(Borrow) REFERENCES
 BORROWS(x) ON UPDATE RESTRICT ON DELETE RESTRICT
);

As inspection of referential integrity constraints is done for this table,

you can close the MSysRelationships table (without saving its sorting and
column resizing’s, as we only have read-only rights on the system tables)
and finally check whether the table scheme also includes tuple (check)
constraints.
Tuple (check) constraints are browsable/editable in Access in the Vali-
dation Rule property of the Table Properties Property Sheet window that
you can open in a right subpane by right-clicking the left upper corner of
the table header from the Design View window of the corresponding db
table (immediately at the left of the Field Name column header—see Figs.
4.13–4.19) and then click on the Properties option. Figure 4.24 shows this
Table Properties Property Sheet window for the BORROWS_LISTS table.
As the Validation Rule for the BORROWS_LISTS table is empty, the
algorithm REA2013A0 would add no tuple constraint to the above CRE-
ATE TABLE statement.
If a tuple constraint ActualReturnDate ≤ DueReturnDate were also
been part of this table scheme113, then the corresponding Validation Rule
property would have look like in Fig. 4.25 and REA2013A0 would have
also added a corresponding tuple constraint of the form:
CONSTRAINT BORROWS_LISTS_CHK1 CHECK
   (ActualReturnDate <= DueReturnDate)

113
But it is not, as this would prevent storing data that a book copy has actually been returned, when-
ever its actual return date were greater than the corresponding due one.
486 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

As everything on the BORROWS_LISTS table scheme has been found

out, you should close its Design View window and open the one for a next
table. After all table schemas are inspected, you can close the db and/or
Access.
Figure 4.26 presents the REA2013A0 algorithm (designed for all cur-
rently used Access versions and a member of the REAF0′ family of algo-
rithms for translating rdb schemes into corresponding ANSI SQL DDL
scripts).
As it can be seen, algorithm REA2013A0 is assuming, as it should
always be the case, that all tables have a primary key and that all primary
keys are singletons and that they are either autonumbers (for all tables
implementing object sets that are not subsets of other ones) or (for tables
implementing subsets) integers referencing autonumbers; Exercise 4.8

FIGURE 4.24  Access Table Properties Property Sheet window (with no tuple
constraint).

FIGURE 4.25  Access Table Properties Property Sheet window (with a tuple constraint).
Relational
 Schemas Implementation and Reverse Engineering 487

FIGURE 4.26  Algorithm REA2013A0 (Translation of (all current versions of) Access
db schemas into corresponding SQL ANSI DDL scripts).
488 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

invites readers to consider all possible exceptions to these best practice

rules too.
Moreover, REA2013A0 is also assuming that there is no foreign keys
circle-type cycle made out from at least two tables (i.e., no cycle in the
corresponding db E-RD graph has no node in which two of its arrows meet
sharp point with sharp point); exercise 4.9 invites readers to consider such
cases too.

4.5  THE ALGORITHMS FOR TRANSLATING E-R DATA MODELS

INTO RDBS AND ASSOCIATED NONRELATIONAL CONSTRAINT
SETS (AF1–8)

Traditional science is all about finding shortcuts.

—Rudy Rucker

As already seen in Section 1.7, the algorithms of the AF1–8 family are
“shortcut” ones composed out of the following (family of) algorithms (in
the order of their application): A1–7, A7/8–3, A8, and AF8′ (i.e., any of
its members g viewed as a mapping can be obtained as g = f °A8 °A7/8–3
°A1–7, where f is a member of AF8’, which means that, for any E-RDM
data model m, the corresponding rdb g(m) is computable as g(m) =
f(A8(A7/8–3(A1–7(m))))).
Of course that the challenge and beauty of any shortcut is to cut as
short as advisable (i.e., without jeopardizing the journey), in order to gain
time. To begin with in this case, as we have already seen in the previous
chapter, it is not at all advisable to skip A7/8–3. Fortunately, as A7/8–3
can also be applied directly on rdb tables, we can apply it after applying
AF1–8, so one step is already gained (and, actually, g = f °A8 °A1–7, g(m)
= f(A8(A1–7(m)))).
What can next be also done to gain more time is to design AF1–8 as
an extended A1–7 (as output) from RDM directly to RDBMS versions,
by skipping completely both the RDM and the ANSI standard SQL DDL
scripts. The price to be paid is only losing the portability of your imple-
mentations, but, in general, this is very rarely needed today, so it is gener-
ally worthwhile to do it.
Relational
 Schemas Implementation and Reverse Engineering 489

As most of the currently available RDBMS versions also have a GUI

for SQL DDL statements too, two versions of AF1–8 algorithms can gen-
erally be designed for any RDBMS version: one which generates ANSI
standard SQL DDL scripts and one which guides you step by step on how
to create corresponding rdbs through the corresponding GUI.
In this section, for exemplifying AF1–8, we chose one that outputs
SQL DDL scripts, not only because, generally, all RDBMS vendors are
advertising their GUIs in order for their products to be as much attractive
as possible, or because the vast majority of the books published in this field
and even YouTube is full with similar approaches, but mainly because this
process should be completely automatized and thus made transparent to
RDBMS users (and the only way to automatize it is to internally generate
SQL DDL scripts): one day, even if unfortunately not that soon, not only
a very few RDBMSes (including the most recent prototype versions of
MatBase), as it is the case now, will provide to their users only higher level
data models and completely hide the RDM one, but all of them will do it.
In particular, we have chosen as target RDBMS version the MS SQL
Server 2014, so that this section is introducing (as g) the A2014SQLS1–8
algorithm, a member of the AF1–8 family that can be also used for all
currently available versions of the SQL Server, not only the 2014 one, in
order to translate E-RDs and associated restrictions lists directly into SQL
Server DDL scripts for creating corresponding rdbs and associated non-
relational constraint sets. Of course that this algorithm will also help the
readers of this book to better understand how was obtained the DDL script
from Subsection 4.3.4.
Figures 4.27–4.34 show the algorithm A2014SQLS1–8. Just like A1–7,
this algorithm assumes that there are no duplicated object set names or map-
ping names defined on a same object set; moreover, it assumes that every table
has a single numeric surrogate primary key, as it should always be the case,
and that static small set elements are string values that do not contain single
quotes (see Exercise 4.14). For example, by applying this algorithm to the E-R
data model of the public library case study from Subsections 2.8.1 and 2.8.2,
the DDL script from Subsection 4.3.4 is obtained (see Exercise 4.10.a).
Exercise 4.11 invites readers to design similar algorithms for DB2
10.5, Oracle 12c, MySQL 5.7, Access 2013, as well as other RDBMS ver-
sions. Exercise 4.12 is similar, but the requested outputs are corresponding
490 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

FIGURE 4.27  Algorithm A2014SQLS1–8 (Translation of E-R data models into

corresponding SQL Server rdb schemas and non-relational constraint sets).

instructions on how to implement the rdbs through the GUIs instead of

SQL DDL scripts.

4.6  THE REVERSE ENGINEERING FAMILY OF ALGORITHMS FOR

TRANSLATING RDB SCHEMAS INTO E-R DATA MODELS (REAF0–2)

Poet’s road is always towards the sources.

—Lucian Blaga
As already seen in Section 1.7, the algorithms of the REAF0–2 family are
“shortcut” ones composed out of the following (family of) algorithms (in the
Relational
 Schemas Implementation and Reverse Engineering 491

method createSQLSTable(R)
if there is no table R in rdb M and R has attributes or is referencing or it is
not referencing, but it is not static and with small cardinality then
  add to S.sql line “CREATE TABLE [” & R & “] (”;
  if R is a subset of S then
AddSQLSForeignKey(R, x, S, “PK”);
loop for all other sets T with R # T
AddSQLSForeignKey(R, xT, T, “NNU”);
end loop;
else
N = chooseSQLSnumericDT(maxcard(R));
add to S.sql line “[” & x & “]” & N & “ IDENTITY PRIMARY KEY”;
end if; completeSQLScheme(R); add to S.sql line “);”;
  end if;
end method createSQLSTable;
FIGURE 4.28  Algorithm A2014SQLS1–8’s createSQLSTable method.

method addSQLSForeignKey(R, A, S, o)
if S is a table then
N = chooseSQLSnumericDT(maxcard(S), 0);
s = s & “, [” & A & “] ” & N;
else
l = 0;
loop for all e [ S
if length(e) > l then l = length(e);
end loop;
s = s & “[” & A & “] VARCHAR(” & l & “) CHECK ([” & A & “] IN (”;
first = true;
loop for all e [ S
if first then first = false; else s = s & “, ”; end if; s = s & “‘“ & e & “‘“;
end loop;
end if;
if o [ {“NN”, “NNU”} then s = s & “ NOT NULL ”; if o = “NNU” then s =
s & “ UNIQUE ”;
elseif o = “PK” then s = s & “ PRIMARY KEY ”; end if;
s = s & “, CONSTRAINT [” & R & “-” & A & “-FK] FOREIGN KEY
([” & A & “]) REFERENCES [” & S & “]”;
add to S.sql line s;
end method addForeignKey;
FIGURE 4.29  Algorithm A2014SQLS1–8’s createSQLSForeignKey method.
492 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

method completeSQLScheme(R)
loop for all arrows A leaving R and heading to object set S, except for set
inclusion-type ones
if A is compulsory then addForeignSQLSKey(R, A, S, “NN”);
else addForeignSQLSKey(R, A, S,); end if;
end loop (for all arrows);
loop for all ellipses e connected to R, except for the surrogate key x
if e is compulsory then addSQLSColumn(R, e, “NN”); else
addSQLSColumn(R, e,); end if;
end loop (for all ellipses);
loop for all uniqueness restrictions u associated to R
s = “, CONSTRAINT [” & u & “] UNIQUE (”; first = true;
loop for all mappings m making up u
if first then first = false; else s = s & “, ”; end if; s = s & “[” & m &
“]”;
end loop;
add to S.sql line s & “)”;
end loop (for all uniqueness restrictions);
i = 1;
loop for all tuple-type restrictions t associated to R
add to S.sql line “, CONSTRAINT CHK_” & R & i & “ CHECK (” &
t & “)”; i = i + 1;
end loop (for all other tuple-type restrictions);
end method completeSQLScheme;

FIGURE 4.30  Algorithm A2014SQLS1–8’s completeSQLScheme method.

method addSQLSColumn(R, e, o)
s = “, ” & e & “ ” & chooseSQLSDT(R, e);
if o [ {“NN”, “NNU”} then s = s & “ NOT NULL ”;
if o = “NNU” then s = s & “ UNIQUE ”;
if e is computed according to expression E then s = s & “ AS ” & E;
add to S.sql line s;
end method addSQLSColumn;
FIGURE 4.31  Algorithm A2014SQLS1–8’s addSQLSColumn method.
Relational
 Schemas Implementation and Reverse Engineering 493

method chooseSQLSnumericDT(n, m)
if m = 0 then
select case n
case < 3: return “ TINYINT”;
case [ {3, 4}: return “ SMALLINT”;
case [ [5; 9]: return “ INT”;
case [ [10; 18]: return “ BIGINT”;
case [ [19; 38]: return “ DECIMAL(” & n & “, 0) ”;
case [ [39; 308]: return “ FLOAT(53)”;
else return “ INTEGER OVERFLOW”;
end select;
else
select case n
case < 39: return “ DECIMAL(” & n & “, “ & m & “)”;
case [ [39; 308]: return “FLOAT(53)”;
else return “RATIONAL OVERFLOW”;
end select;
end if;
end method chooseSQLSnumericDT;

FIGURE 4.32  Algorithm A2014SQLS1–8’s chooseSQLSnumericDT method.

method chooseSQLStextDT(t, p)
if t = ASCII
if p ≤ 8000 then return “ VARCHAR(” & p & “)”;
elseif p [ [8001; 231 – 1] then return “ VARCHAR(MAX) ”;
else return “ASCII OVERFLOW”; end if;
else
if p ≤ 4000 then return “ NVARCHAR(” & p & “)”;
elseif p [ [4001; 23° – 1] then return “ VARCHAR(MAX) ”;
else return “UNICODE OVERFLOW”; end if;
end if;
end method chooseSQLSDT;

FIGURE 4.33  Algorithm A2014SQLS1–8’s chooseSQLStextDT method.

494 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

FIGURE 4.34  Algorithm A2014SQLS1–8’s chooseSQLSDT method.

order of their application): REAF0’, REA0, REA1, and REA2 (or REAF0–1
and REA2, as REAF0–1 is the composition of REAF0’, REA0, and REA1).
In other words, any of its members g viewed as a mapping can be
obtained as g = f ° REA0 ° REA1 ° REA2, where f is a member of REAF0’,
which means that, for any rdbs r, the corresponding E-R data model g(r)
is computable as g(r) = f(REA0(REA1(REA2(r))))) (or, equivalently, g =
REAF0–1 ° REA2, g(r) = REAF0–1(REA2(r))).
You should first of all please note that, as (E)MDM is not tackled in
this first volume, neither are REA1, REA2, nor REAF0–1.
Relational
 Schemas Implementation and Reverse Engineering 495

FIGURE 4.35  Algorithm REA2013A0–2 (Translation of (all current versions of) Access
db schemas into corresponding E-R data models).
496 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

Consequently, we are forced to come up in this volume with algorithms

of this family simply by extending, from the output point of view, the com-
position of REA0 and REAF0′ from standard ANSI SQL DDL scripts all
the way upwards to E-R data models.
Consequently, mainly in order to use the experience and intuition
gained in section 4.4 above, we chose to exemplify this family of reverse
engineering algorithms also with its member for MS Access dbs. Figures
4.35–4.37 present the corresponding algorithm, called REA2013A0–2. All
other members of this family are left to the readers (see Exercise 4.15).
As it can be seen, just like REA2013A0, algorithm REA2013A0–2 is
assuming, as it should always be the case, that all tables have a primary
key, that all primary keys are singletons, and that they are either autonum-
bers (for all tables implementing object sets that are not subsets of other
ones) or (for tables implementing subsets) integers referencing autonum-
bers; Exercise 4.8 invites readers to consider all possible exceptions to
these best practice rules too.
Moreover, REA2013A0–2 is also assuming that there is no foreign keys
circle-type cycle made out from at least two tables (i.e., no cycle in the
corresponding db E-RD graph has no node in which two of its arrows meet
sharp point with sharp point); exercise 4.9 invites readers to consider such
cases too.

method primaryKey(C, t, D)
underline C;
if t = ‘e’ then
draw the one-to-oneness arrow on the edge from ellipse C to its
rectangle;
write in RS the maximum cardinality restriction for T, labeled Ti,
equal to D’s cardinal;
i = i + 1;
else
make the arrow labeled C a double one;
end if;
end primaryKey;
FIGURE 4.36  Algorithm REA2013A0–2′s method primaryKey.
Relational
 Schemas Implementation and Reverse Engineering 497

method addSemanticsAndClose
investigate the resulted E-RD and RS, and, based on your gained
insights, document in RS the semantics of all object sets, properties,
and restrictions, and write in D a first clear, concise, and as much as
possible exhaustive description of rdb S;
save and close .txt files E-RD, RS, and D, which make up the E-R data
model M;
close Access;
end addSemanticsAndClose;
FIGURE 4.37  Algorithm REA2013A0–2′s method addSemanticsAndClose.

4.7  CASE STUDY: REVERSE ENGINEERING OF AN ACCESS

STOCKS DB SCHEME INTO BOTH AN ANSI STANDARD SQL DDL
SCRIPT AND AN E-R DATA MODEL

Physics is mathematical not because we know so much about

the physical world, but because we know so little;
it is only its mathematical properties that we can discover.
—Bertrand Russell

Figures 4.38–4.74 present a Stocks Access db (see section 4.4 above on

how such screen captures are generally obtainable and why). Figure 4.75
shows the corresponding ANSI standard SQL DDL script for creating this
db obtained by applying the algorithm REA2013A0 to this rdb. Figures
4.76 to 4.88 contain the corresponding E-R data model obtained by apply-
ing the algorithm REA2013A0–2 to it.

4.7.1 THE STOCKS DB

If stock market experts were so expert, they would be buying stock,

not selling advice.
—Norman Ralph Augustine
498 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

In Fig. 4.38 we can see that this db is made out of the following 10 tables:
Colors, Customers, Entries, EntryItems, EntryTypes, OrderItems, Orders,
Products, Stocks, and Warehouses (as Switchboard Items is the standard
Access table for storing the application menu, which is a table-driven one,
and the 6 dimmed tables –whose names start with MSys– are making up
the standard Access metadata catalog).
Figure 4.39 shows the metadata stored by Access on the referential integ-
rity constraints of this db in its table MSysRelationships. We can see that
there are the following 12 foreign keys (as the other two ones belong to and
are referencing metacatalog tables: their names are prefixed by “MSys”):
Ø fk1068REF548: Entries.EntryType references EntryTypes.#EntryType
Ø fk1065REF545: Entries.Customer references Customers.#Customer

FIGURE 4.38  The Stocks db Navigation pane.

Relational
 Schemas Implementation and Reverse Engineering 499

FIGURE 4.39  The Stocks db MSysRelationships table instance.

Ø fk1075REF554: EntryItems.Warehouse references Warehouses.

#Warehouse
Ø fk1073REF551: EntryItems.Product references Products.#Product
Ø EntriesEntryItems: EntryItems.Entry references Entries.#Entry
Ø OrdersOrderItems: OrderItems.Order references Orders.#Order
Ø fk1084REF554: OrderItems.Warehouse references Warehouses.
#Warehouse
Ø fk1082REF551: OrderItems.Product references Products.#Product
Ø fk1087REF545: Orders.Customer references Customers.#Customer
Ø fk1093REF544: Products.Color references Colors.#Color
Ø fk1084REF554: Stocks.Warehouse references Warehouses.#Warehouse
Ø fk1101REF551: Stocks.Product references Products.#Product
Figure 4.40 presents the scheme of table Colors; it can be seen that
it has two columns: a surrogate autonumber primary key #Color and a
string Color one, which may have at most 64 characters (Field Size), does
not accept nulls (Required = Yes), and is one-to-one (Indexed = Yes (No
Duplicates)).
Similarly, Fig. 4.41 presents the scheme of table Customers that also
has two columns: a surrogate autonumber primary key #Customer and
a string Customer one, which may have at most 64 characters, does not
accept nulls, and is one-to-one.
Figures 4.42–4.46 present the seven columns of table Entries: #Entry
(surrogate autonumber primary key), Customer (long integer, no nulls,
500 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

FIGURE 4.40  The Stocks db table Colors scheme.

FIGURE 4.41  The Stocks db table Customers scheme.

Relational
 Schemas Implementation and Reverse Engineering 501

FIGURE 4.42  The Stocks db table Entries column Customer definition.

FIGURE 4.43  The Stocks db table Entries column EntryNo.

502 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

FIGURE 4.44  The Stocks db table Entries column EntryDate.

FIGURE 4.45  The Stocks db table Entries column Stocked.

Relational
 Schemas Implementation and Reverse Engineering 503

FIGURE 4.46  The Stocks db table Entries column EntryType.

FIGURE 4.47  The Stocks db table Entries indexes.

indexed), EntryNo (string of at most 64 characters, no nulls, one-to-one),

EntryDate (date between Jan. 3rd, 2000 and the current system date and
time, which is also the default one, no nulls), Stocked (Boolean), and
EntryType (long integer, default 1, no nulls). As seen in Fig. 4.47, this
504 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

table does not have other keys (as its third index, CustomerIdx, is not a
unique one).
Figures 4.48–4.51 present the five columns of table EntryItems: #Entry-
Item (surrogate autonumber primary key), Entry (long integer, no nulls,
indexed), Product (long integer, no nulls, indexed), Qty (double rational
strictly positive and at most 1000, scale 2, not null), and Warehouse (long
integer, no nulls). As seen in Fig. 4.52, this table has the key EntryItem-
Key, made up of Entry and Product (as its other two indexes, EIProductIdx
and EIEntryIdx, are not unique).
Figure 4.53 presents the two columns of table EntryTypes: #EntryType
(surrogate autonumber primary key) and EntryType (string of at most 64
characters, no nulls, one-to-one).
Figures 4.54–4.57 present the five columns of table OrderItems:
#OrderItem (surrogate autonumber primary key), Order (long integer,
no nulls, indexed), Product (long integer, no nulls, indexed), Qty (double
rational strictly positive and at most 1000, scale 2, not null), and Ware-
house (long integer, no nulls). As seen in Fig. 4.58, this table has the key
OrderItemKey, made up of Order and Product (as its other two indexes,
OIProductIdx and OIOrderIdx, are not unique).

FIGURE 4.48  The Stocks db table EntryItems column Entry.

Relational
 Schemas Implementation and Reverse Engineering 505

FIGURE 4.49  The Stocks db table EntryItems column Product.

FIGURE 4.50  The Stocks db table EntryItems column Qty.

506 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

FIGURE 4.51  The Stocks db table EntryItems column Warehouse.

FIGURE 4.52  The Stocks db table EntryItems keys.

Relational
 Schemas Implementation and Reverse Engineering 507

FIGURE 4.53  The Stocks db table EntryTypes scheme.

FIGURE 4.54  The Stocks db table OrderItems column Order.

508 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

FIGURE 4.55  The Stocks db table OrderItems column Product.

FIGURE 4.56  The Stocks db table OrderItems column Qty.

Relational
 Schemas Implementation and Reverse Engineering 509

FIGURE 4.57  The Stocks db table OrderItems column Warehouse.

FIGURE 4.58  The Stocks db table OrderItems keys.

510 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

Figures 4.59–4.62 present the five columns of table Orders: #Order

(surrogate autonumber primary key), Customer (long integer, no nulls,
indexed), OrderNo (string of at most 64 characters, no nulls, one-to-one),
OrderDate (date between Jan. 3rd, 2000 and the current system date and
time, which is also the default one, no nulls), and Delivery (Boolean). As
seen in Fig. 4.63, this table does not have other keys (as its third index,
CustomerIdx, is not a unique one).
Figures 4.64–4.66 present the four columns of table Products: #Prod-
uct (surrogate autonumber primary key), ProductCode (string of at most
64 characters, no nulls), Color (long integer, no nulls), and Product (string
of at most 64 characters, no nulls. indexed). As seen in Fig. 4.67, this table
also has the key ProductKey made of ProductCode and Color (as its third
index, ProductsProduct, is not unique).
Figures 4.68–4.72 present the six columns of table Stocks: #Stock
(surrogate autonumber primary key), Warehouse (long integer, no nulls,
indexed), Product (long integer, no nulls, indexed), Stock (positive dou-
ble, scale 2, no nulls), InitialStock (positive double, scale 2, default 0, no

FIGURE 4.59  The Stocks db table Orders column Customer.

Relational
 Schemas Implementation and Reverse Engineering 511

FIGURE 4.60  The Stocks db table Orders column OrderNo.

FIGURE 4.61  The Stocks db table Orders column OrderDate.

512 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

FIGURE 4.62  The Stocks db table Orders column Delivery.

FIGURE 4.63  The Stocks db table Orders keys.

Relational
 Schemas Implementation and Reverse Engineering 513

FIGURE 4.64  The Stocks db table Products column ProductCode.

nulls), and InitialStockDate (date between Jan. 3rd, 2000 and the current
system date and time, no nulls). As seen in Fig. 4.73, this table also has the
key StocksKey made of Warehouse and Product (as its other two indexes,
#Product and #Warehouse, are not unique).
Figure 4.74 presents the two columns of table Warehouses: #Ware-
house (surrogate autonumber primary key) and Warehouse (string of at
most 64 characters, no nulls, one-to-one).
514 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

FIGURE 4.65  The Stocks db table Products column Color.

FIGURE 4.66  The Stocks db table Products column Product.

Relational
 Schemas Implementation and Reverse Engineering 515

FIGURE 4.67  The Stocks db table Products keys.

FIGURE 4.68  The Stocks db table Stocks column Warehouse.

516 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

FIGURE 4.69  The Stocks db table Stocks column Product.

FIGURE 4.70  The Stocks db table Stocks column Stock.

Relational
 Schemas Implementation and Reverse Engineering 517

FIGURE 4.71  The Stocks db table Stocks column InitialStock.

FIGURE 4.72  The Stocks db table Stocks column InitialStockDate.

518 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

FIGURE 4.73  The Stocks db table Stocks keys.

FIGURE 4.74  The Stocks db table Warehouses scheme.

Relational
 Schemas Implementation and Reverse Engineering 519

4.7.2  THE CORRESPONDING ANSI STANDARD SQL DDL

SCRIPT

The illiterate of the twenty-first century will not be those who cannot
read and write, but those who cannot learn, unlearn, and relearn.
—Alvin Toffler

Figure 4.75 shows the result of applying the reverse engineering algorithm
REA2013A0 to the above rdb. Note that for all foreign keys the columns
they reference are omitted, as all such columns are primary keys. Also
note that, due to their corresponding domain (check) constraints, the preci-
sion of the Qty columns has been downsized to only 6.

4.7.3  THE CORRESPONDING E-R DATA MODEL

Simplicity is prerequisite for reliability.

—Edsger Dijkstra

Figures 4.76–4.88 show the result of applying the reverse engineering

algorithm REA2013A0–2 to the Stocks rdb, with a slight variation (see
Exercise 4.17): a separate structural E-RD was drawn.
The first ten of them (4.76 to 4.85) present detailed E-RDs per table
(object set). Figure 4.86 shows the structural E-RD of this subuniverse.
Figure 4.87 is dedicated to the associated set of restrictions.
Finally, Fig. 4.88 contains an informal description of this subuniverse
as reversed engineered from the E-RDs and associated restrictions.
Generally, you can get more insight into the semantics of tables, col-
umns, and constraints by also inspecting the db instance, which allows
refining the description component of the E-RD model: see, for example,
Exercise 4.18.a.
Moreover, after reverse engineering dbs, you can not only discover
flaws in their design and/or implementation that need corrections (see, for
example, Exercise 4.18.c.), but also get ideas on what extensions to them
would be of real help to both their users and applications programmers:
see, for example, Exercise 4.19.
520 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

Note that the range for Stock and InitialStock was computed as follows:
according to Table 4.11, the DOUBLE Access data type may store decimal
numbers of up to 308 digits; as these two columns need 2 of them for the
scale, the maximum possible mantissa is 306; moreover, given that they
both accept only positive values, RAT+ is their base set (instead of RAT,
the set of all rationals).

CREATE DATABASE STOCKS;

CREATE TABLE Colors (
#Color INTEGER(9) GENERATED ALWAYS AS
IDENTITY(1,1) PRIMARY KEY,
Color VARCHAR(64) NOT NULL UNIQUE
);
CREATE TABLE Customers (
#Customer INTEGER(9) GENERATED ALWAYS AS
IDENTITY(1,1) PRIMARY KEY,
Customer VARCHAR(64) NOT NULL UNIQUE
);
CREATE TABLE EntryTypes (
#EntryType INTEGER(9) GENERATED ALWAYS AS
IDENTITY(1,1) PRIMARY KEY,
EntryType VARCHAR(64) NOT NULL UNIQUE
);
CREATE TABLE Warehouses (
#Warehouse INTEGER(9) GENERATED ALWAYS AS
IDENTITY(1,1) PRIMARY KEY,
Warehouse VARCHAR(64) NOT NULL UNIQUE
);
CREATE TABLE Products (
#Product INTEGER(9) GENERATED ALWAYS AS
IDENTITY(1,1) PRIMARY KEY,
ProductCode VARCHAR(64) NOT NULL,
Color INTEGER(9) NOT NULL,
Product VARCHAR(64) NOT NULL,

FIGURE 4.75  The Stocks db ANSI standard SQL DDL creation script.
Relational
 Schemas Implementation and Reverse Engineering 521

CONSTRAINT ProductKey UNIQUE (ProductCode,

Color),
CONSTRAINT fk1093REF544 FOREIGN KEY (Color)
REFERENCES Colors
);
CREATE TABLE Stocks (
#Stock INTEGER(9) GENERATED ALWAYS AS
IDENTITY(1,1) PRIMARY KEY,
Warehouse INTEGER(9) NOT NULL,
Product INTEGER(9) NOT NULL,
Stock DECIMAL(40,2) NOT NULL CHECK (Stock >=
0),
InitialStock DECIMAL(40,2) NOT NULL CHECK
(InitialStock >= 0),
InitialStockDate DATE NOT NULL CHECK
(InitialStockDate BETWEEN 2000–01–03 AND
SysDate()),
CONSTRAINT StocksKey UNIQUE (Warehouse,
Product),
CONSTRAINT fk1101REF551 FOREIGN KEY (Product)
REFERENCES Products,
CONSTRAINT fk1103REF554 FOREIGN KEY
(Warehouse) REFERENCES Warehouses
);
CREATE TABLE Entries (
#Entry INTEGER(9) GENERATED ALWAYS AS
IDENTITY(1,1) PRIMARY KEY,
Customer INTEGER(9) NOT NULL,
EntryNo VARCHAR(64) NOT NULL UNIQUE,
EntryDate DATE NOT NULL CHECK (EntryDate
BETWEEN 2000–01–03 AND SysDate()),
Stocked BOOLEAN,
EntryType INTEGER(9) NOT NULL,
CONSTRAINT fk1065REF545 FOREIGN KEY
(Customer) REFERENCES Customers,
FIGURE 4.75  (Continued).
522 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

CONSTRAINT fk1068REF548 FOREIGN KEY

(EntryType) REFERENCES EntryTypes
);
CREATE TABLE EntryItems (
#EntryItem INTEGER(9) GENERATED ALWAYS AS
IDENTITY(1,1) PRIMARY KEY,
Entry INTEGER(9) NOT NULL,
Product INTEGER(9) NOT NULL,
Qty DECIMAL(6,2) NOT NULL CHECK (Qty > 0
AND Qty <= 1000),
Warehouse INTEGER(9) NOT NULL,
CONSTRAINT EntryItemKey UNIQUE (Entry,
Product, Warehouse),
CONSTRAINT EntriesEntryItems FOREIGN KEY
(Entry) REFERENCES Entries,
CONSTRAINT fk1073REF551 FOREIGN KEY
(Product) REFERENCES Products,
CONSTRAINT fk1075REF554 FOREIGN KEY
(Warehouse) REFERENCES Warehouses
);
CREATE TABLE Orders (
#Order INTEGER(9) GENERATED ALWAYS AS
IDENTITY(1,1) PRIMARY KEY,
Customer INTEGER(9) NOT NULL,
OrderNo VARCHAR(64) NOT NULL UNIQUE,
OrderDate DATE NOT NULL CHECK (OrderDate
BETWEEN 2000–01–03 AND SysDate()),
Delivery BOOLEAN,
CONSTRAINT fk1087REF545 FOREIGN KEY
(Customer) REFERENCES Customers
);

CREATE TABLE OrderItems (

#OrderItem INTEGER(9) GENERATED ALWAYS AS
IDENTITY(1,1) PRIMARY KEY,
FIGURE 4.75  (Continued).
Relational
 Schemas Implementation and Reverse Engineering 523

Order INTEGER(9) NOT NULL,

Product INTEGER(9) NOT NULL,
Qty DECIMAL(6,2) NOT NULL CHECK (Qty > 0
AND Qty <= 1000),
Warehouse INTEGER(9) NOT NULL,
CONSTRAINT OrderItemKey UNIQUE (Order,
Product, Warehouse),
CONSTRAINT OrdersOrderItems FOREIGN KEY
(Order) REFERENCES Orders,
CONSTRAINT fk1082REF551 FOREIGN KEY
(Product) REFERENCES Products,
CONSTRAINT fk1084REF554 FOREIGN KEY
(Warehouse) REFERENCES Warehouses
);

FIGURE 4.75  (Continued).

FIGURE 4.76  Colors E-RD. FIGURE 4.77  Customers E-RD.

FIGURE 4.78  EntryTypes E-RD. FIGURE 4.79  Warehouses E-RD.

524 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

FIGURE 4.80  Stocks E-RD. FIGURE 4.81  Products E-RD.

FIGURE 4.82  Entries E-RD. FIGURE 4.83  EntryItems E-RD.

FIGURE 4.84  Orders E-RD. FIGURE 4.85  OrderItems E-RD.

FIGURE 4.86  Stocks structural E-RD.

Relational
 Schemas Implementation and Reverse Engineering 525

Customers (The set of customers and suppliers of interest.)

a. Cardinality: max(card(CUSTOMERS)) = 1,000,000,000   (RC0)
b. Data ranges:
#Customer: INT(9)     (RC1)
Customer: UNICODE(64)     (RC2)
c. Compulsory data: Customer    (RC3)
d. Uniqueness: Customer (no two customers may have
same names)     (RC4)
Colors (The set of product colors.)
a. Cardinality: max(card(Colors)) = 1,000,000,000     (RK0)
b. Data ranges:
#Color: INT(9)     (RK1)
Color: UNICODE(64)     (RK2)
c. Compulsory data: Color    (RK3)
d. Uniqueness: Color (there may not be two colors with
same name)     (RK4)
Products (The set of products of interest.)
a. Cardinality: max(card(Products)) = 1,000,000,000     (RP0)
b. Data ranges:
#Product: INT(9)     (RP1)
Product: UNICODE(64)    (RP2)
ProductCode: UNICODE(64)    (RP3)
c. Compulsory data: Product, ProductCode, Color    (RP4)
d. Uniqueness: Product • Color (there may not be two
products having same codes and colors)     (RP5)
Warehouses (The set of warehouses of interest.)
a. Cardinality: max(card(Warehouses)) = 1,000,000,000     (RW0)
b. Data ranges:
#Warehouse: INT(9)     (RW1)
Warehouse: UNICODE(64)     (RW2)
c. Compulsory data: Warehouse    (RW3)
d. Uniqueness: Warehouse (no 2 warehouses may have
same names)     (RW4)

FIGURE 4.87  Stocks associated restriction set.

526 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

Stocks (The set of pairs <w, p> meaning that warehouse

w stores product p.)
a. Cardinality: max(card(Stocks)) = 1,000,000,000     (RS0)
b. Data ranges:
#Stock: INT(9)     (RS1)
Stock, InitialStock: RAT+(306, 2)    (RS2)
InitialStockDate: [2000–01–03, SysDate()]    (RS3)
c. Compulsory data: Product, Warehouse, Stock,
InitialStock, InitialStockDate    (RS4)
d. Uniqueness: Product • Warehouse (any valid Stocks
instance should be a set: any pair <w, p> should be
stored only once)     (RS5)

EntryTypes (The set of product colors.)

a. Cardinality: max(card(EntryTypes)) = 1,000,000,000     (RT0)
b. Data ranges:
#EntryType: INT(9)     (RT1)
EntryType: UNICODE(64)     (RT2)
c. Compulsory data: EntryType    (RT3)
d. Uniqueness: EntryType (there may not be two entry
types having same name)     (RT4)
Entries (The set of product entries into warehouses.)
a. Cardinality: max(card(Entries)) = 1,000,000,000     (RE0)
b. Data ranges:
#Entry: INT(9)     (RE1)
EntryNo: UNICODE(64)     (RE2)
EntryDate: [2000–01–03, SysDate()]     (RE3)
Stocked: BOOLE (was entry p operated in Stocks?)    (RE4)
c. Compulsory data: EntryType, EntryNo, EntryDate,
Customer    (RE5)
d. Uniqueness: EntryNo (there may not be two entries
having same number)     (RE6)
EntryItems (The set of triples <e, w, p> meaning that, from
entry e, in warehouse w, products p were stored too.)

FIGURE 4.75  (Continued).

Relational
 Schemas Implementation and Reverse Engineering 527

a. Cardinality: max(card(EntryItems)) = 1,000,000,000     (RI0)

b. Data ranges:
#EntryItems: INT(9)     (RI1)
Qty: [0, 1000] , RAT(4, 2)     (RI2)
c. Compulsory data: Entry, Product, Warehouse, Qty    (RI3)
d. Uniqueness: Entry • Product • Warehouse (any valid
EntryItems instance should be a set: any triple <e, w, p>
should be stored only once)     (RI4)
Orders (The set of customer orders.)
a. Cardinality: max(card(Orders)) = 1,000,000,000     (RO0)
b. Data ranges:
#Order: INT(9)     (RO1)
OrderNo: UNICODE(64)     (RO2)
OrderDate: [2000–01–03, SysDate()]    (RO3)
Delivered: BOOLE (was order o delivered?)    (RO4)
c. Compulsory data: OrderNo, OrderDate, Customer    (RO5)
d. Uniqueness: OrderNo (no two orders may have same
numbers)     (RO6)
OrderItems (The set of triples <o, w, p> meaning that, for
order o, from warehouse w, products p were delivered too.)
a. Cardinality: max(card(OrderItems)) = 1,000,000,000   (ROI0)
b. Data ranges:
#OrderItems: INT(9)     (ROI1)
Qty: [0, 1000] , RAT(4, 2)     (ROI2)
c. Compulsory data: Order, Product, Warehouse, Qty    (ROI3)
d. Uniqueness: Order • Product • Warehouse (any valid
OrderItems instance should be a set: any triple <o, w, p>
should be stored only once)     (ROI4)
FIGURE 4.75  (Continued).
528 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

A company trades products characterized by compulsory codes,

names, and colors (all three of them having at most 64 UNICODE char-
acters). There may not be two distinct products having same code and
color. There may not be two colors having same name. Products are
bought from and sold to suppliers/customers whose compulsory names
have at most 64 UNICODE characters. There may not be two such part-
ners having same names. Products are stored into warehouses whose
compulsory names have at most 64 UNICODE characters. There may
not be two warehouses having same names.
In any warehouse, stocks for any product are compulsory and posi-
tive, having two scale digits; moreover, there should have been initial
stocks (possibly 0, also compulsory and positive, two scale digits) in
any warehouse, for any product stocked in that warehouse, at a com-
pulsory date between January 3rd, 2000 and the current system date
(probably prior to the first corresponding stock entry that was recorded
in the db).
The company buys products from (compulsory) suppliers/custom-
ers, according to entry documents having compulsory types (at most
64 UNICODE characters) and unique numbers, at compulsory dates
between January 3rd, 2000 and the current system date, and stores
them in (compulsory) warehouses in (compulsory) strictly positive,
two scale digits, and maximum 1000 unit quantities per entry, ware-
house, and product. There may not be two entry types having same
name.
The company sells products to (compulsory) suppliers/customers,
according to order documents having compulsory unique numbers, at
compulsory dates between January 3rd, 2000 and the current system
date, and delivers them from (compulsory) warehouses in (compulsory)
strictly positive, two scale digits, and maximum 1000 unit quantities
per order, warehouse, and product.
There may be at most 109 elements in each of the following sets:
colors, entry types, warehouses, customers, products, stocks, stock
entries, stock entry items, orders, and order items.
FIGURE 4.88  Stocks db description.
Relational
 Schemas Implementation and Reverse Engineering 529

4.8  BEST PRACTICE RULES

The laws of nature are but the mathematical thoughts of God.

—Euclid

4.8.1 DATABASES

We are what we repeatedly do.

Excellence, then, is not an act, but a habit.
—Aristotle
R-I-D-0.(Do not add any objects to system tablespaces)
Never place your objects in system tablespaces, but only in user ones:
otherwise, you increase data contention and slow both the server and your
db applications.

R-I-D-1.(Safely keep up to date DDL scripts)

Always save SQL DDL scripts in a secure location; every time you
make changes to db objects, be sure to script and check them into some
version-control software repository. This is not a vital part of your db doc-
umentation, but also makes it simpler to replicate your db (e.g., for new
customers) and compare it to other similar ones (including previous ver-
sions of your own).

R-I-D-2.(No UNICODE identifiers)

Do not use UNICODE identifiers, even when they are allowed: it com-
plicates your applications whenever conversions are needed.

4.8.2 TABLES

When you realize you’ve made a mistake,

take immediate steps to correct it.
—Dalai Lama
R-I-T-1.(Concatenated keys column ordering)
For concatenated keys (generally, for all such indexes, be them unique
or not), the order of their columns should be given by the cardinality of
their corresponding duplicate values: any column should have more dupli-
cates than its predecessor and fewer than its successor.
530 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

For example, as in a CITIES table (see, for example, Fig. 3.5) there are
much more distinct ZipCode values (rare duplicates being possible only
between countries) than *Country name ones (as, generally, there are very
many cities in a country), the corresponding key should be defined in the
order <ZipCode, *Country> and not vice versa.

R-I-T-2.(Don’t use triggers to enforce relational constraints)

Triggers should mainly be used for enforcing (event-driven) non-rela-
tional constraints and/or to automate db maintenance and auditing. As they
are slow, never use them to enforce relational constraints.
For example, you might use triggers to supplement declarative referen-
tial integrity too, but never to replace it.

R-I-T-3.(Always use same data types in check constraints)

Always make sure that check constraint expressions compare same
data types: otherwise, in the best case, implicit conversions are needed,
which is adding to the time spent for constraint enforcing; moreover, in
such cases you also risk unpredictable results sometimes.

R-I-T-4.(Place BLOB data from frequently queried tables in subtables)

BLOB data columns should not be defined in frequently queried tables,
because of performance issues: it is better to vertically split such tables,
keeping all non-BLOB columns in the frequently queried one and the
BLOB ones in a separate (sub)table, implemented as if it were correspond-
ing to a subset of it.
For example, split table PEOPLE having columns x, FirstName, Last-
Name, Father, Mother, Sex, BirthPlace, BirthDate, Notes, where Notes
is its only BLOB column, into tables PEOPLE having same columns as
above except for Notes and table PEOPLENotes having columns x and
Notes, with x referencing the primary key x of PEOPLE.

4.8.3 COLUMNS

Anyone who stops learning is old, whether at 20 or 80.

Anyone who keeps learning stays young.
—Henry Ford
Relational
 Schemas Implementation and Reverse Engineering 531

R-I-C-0.(Always use correct data types)

Using incorrect data types might decrease the efficiency of the RDBMS
optimizers, hurt performance, and cause applications to perform unneces-
sary data conversions, which are very costly whenever repeated for thou-
sands of times.
For example, don’t ever use strings to store dates, times, Booleans, or
numbers.

R-I-C-1.(Always use the smallest data types needed)

Always use the smallest data type necessary to implement the required
functionality: you are not only saving disk space, but you speed up pro-
cessing too.
For example, do not use NUMERIC or FLOAT if you only need inte-
gers, do not use BIGINT if you only need SMALLINT, etc.

R-I-C-2.(Always use (VAR)BINARY instead of (VAR)CHAR BIT)

Although in DB2 you can also specify the subtype BIT for CHAR and
VARCHAR columns that contain BIT data, for better performance always
use the BINARY or VARBINARY data types instead.

R-I-C-3.(Always use VARCHAR(n) instead of CHAR(n) for n > 17)

For better processing speed, use VARCHAR (variable length strings)
instead of CHAR (fixed length strings) for all strings having at least 18
characters, unless you know that the entire column is to always be filled
(equivalently: always avoid storing many blanks).
For example, in DB2, if you store a timestamp in character form (not as
the DB2 TIMESTAMP data type), you need a column with some number
of characters: in Version 9, that number would be 26 characters; in ASCII,
EBCDIC, or UTF-8, that column is 26 bytes; in UTF-16, that column is
52 bytes. Because the timestamp is always the same size, using a varying-
length column does not save storage. However, suppose that you have a
name field that is in ASCII or EBCDIC and allows for names of 26 char-
acters (just as in ASCII SBCS or EBCDIC SBCS, but in UTF-8 you need
78 bytes, and in UTF-16 52 bytes). In this case, you want to use a varying-
length column, because the name field is likely to have many blanks and
you should not want to store them.
532 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

R-I-C-4.(AS/400 DB2 DECIMAL versus NUMERIC)

In DB2 on AS/400, if you are going to perform arithmetic operations
with a number, it is more efficient to define it as DECIMAL; if a number is
going to be used mostly for display purposes, it is more efficient to define
it as NUMERIC.

R-I-C-5.(DB2, COBOL, and PL/I best data types matching)

If your application is written in COBOL or PL/I, store your data in
UTF-16 and use the GRAPHIC and VARGRAPHIC data types, in order for
the UNICODE format in your application to match the one in your DB2
db: this setup avoids conversion costs.

4.8.4 CONSTRAINTS

The tools we use have a profound (and devious) influence on our

thinking habits, and, therefore, on our thinking abilities.
—Edsger W. Dijkstra

R-I-P-0.(Always use all constraint types provided by the target DBMS)

In order to guarantee as much as possible db instances’ plausibility,
always use all constraint types provided by the host DBMS. All following
five best practice rules from this category are particular cases of this one
(for the five main relational constraint types).
For example, use MS Access’ Allow Zero Length = No setting for
rejecting dirty nulls for any text type column.
Moreover, always enforce through the host DBMS in your dbs all con-
straints from your E-R data models that may be enforced by that DBMS:
enforcing them only in applications built on top a db leaves that db defense-
less against direct data manipulations, which circumvents the applications.

R-I-P-1.(Always enforce all (co-)domain constraints)

You should always enforce for any column its associated (co-)domain
constraint.
For example, there are very few types of columns that do not need such
constraints (e.g. Boolean, Hyperlink); for text ones, you need to enforce at
least its maximum possible length and sometimes also its plausible subset
Relational
 Schemas Implementation and Reverse Engineering 533

(e.g., for PEOPLE or CREATURES Sex you only need one character and
only a few letters out of the ASCII alphabet); for numbers and dates, only
a subset of a corresponding data type (e.g. [0; 1,000.00] for Stock or Price,
[1/1/2010, SysDate()] for PaymentDate and OrderDate, etc.).

R-I-P-2.(Always enforce all not null constraints, at least one per table
additional to the corresponding surrogate primary key one)
You should always enforce for any column that should always be filled
with data its associated not null constraint: any table should have at least
one such semantic column (i.e. not only the syntactic surrogate key one).
For example, there may be no countries, cities, companies, books,
songs, movies, etc. without names (titles), no order, invoice, bill, delivery
lines without product, price, and quantity, etc.

R-I-P-3.(Always enforce for all foreign keys their corresponding refer-

ential integrity constraints)
Unfortunately, some DBMS GUIs (e.g. MS Access) let you define for-
eign keys without or with only partial referential integrity enforcement:
do not ever forget to fully enforce them in order to ban dangling pointers
from your db instances.
For example, in MS Access you might not leave or turn to No the
combo-boxes Limit To List property (thus allowing users to store any-
thing else than the values provided by the combo-box) or do not check or
uncheck the Enforce Referential Integrity check-box (thus allowing users
to delete rows even if they are referenced by any number of rows).

R-I-P-4.(Always enforce all minimal uniqueness constraints and never

superkeys)
Unfortunately, most DBMSes let you both not defining any (unique)
key or enforcing superkeys for any db table; you should enforce for any
fundamental table all of its (unique) keys and no superkey.
For example, for table COUNTRIES from Fig. 3.4, besides its primary
key x, you should also enforce its other 3 semantic (candidate) keys (Coun-
try, CountryCode, and Capital) and no superkey (e.g. neither Country •
CountryCode, or Country • Capital, or CountryCode • Capital, or Country
• CountryCode • Capital, or one of their permutations).
534 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

R-I-P-5.(Always enforce all tuple/check constraints)

You should use DBMSes for also enforcing any tuple/check constraints
existing in your corresponding restriction set.
For example, table PEOPLE should have enforced the constraint
BirthDate ≤ PassedAwayDate, table SCHEDULES should have enforced
the constraint StartHour ≤ EndHour, etc.

4.9  THE MATH BEHIND THE ALGORITHMS PRESENTED

IN THIS CHAPTER

In most sciences one generation tears down what another has

built and what one has established another undoes. 
In mathematics alone each generation adds a new story to the old structure.
—Hermann Hankel

There is no math behind this chapter, except for the compulsory propo-
sitions characterizing the algorithms that it introduces. As they are very
simple and similar, for example, to Propositions 3.1 and 4.1 (see Exercise
4.1), they are all left to the reader.
Thus, Exercise 4.1 characterizes A8, while Exercise 4.20 invites read-
ers to characterize all of the other algorithms presented in this chapter, as
well as all of the extensions to them and all of the similar ones designed
by the readers as proposed by the other exercises from subsection 4.11.

4.10 CONCLUSION

Machines should work; people should think.

—IBM Pollyanna Principle

New dbs or extensions to existing ones can be implemented either from

relational schemes or directly from E-R data models. Whenever they are
directly implemented by skipping the RDM level, algorithm A7/8–3 should
immediately be applied on all newly created or modified tables. Skipping
the RDM level only affects portability, which is lost in such cases, but may
not be needed sometimes.
Relational
 Schemas Implementation and Reverse Engineering 535

Database implementation can be done algorithmically, either through

running corresponding SQL DDL scripts or by using the GUIs of the
RDBMSs (which are generally helping by being able to generate at any
time corresponding SQL DDL scripts).
This chapter provides such algorithms for both implementation paths
that are linear in the size of the input RDM schemes and E-R data models,
respectively.
Moreover, it also provides dual reverse engineering algorithms of same
complexity, for understanding semantics of legacy, not enough and/or up
to date documented rdbs.
SQL statements’ syntactical differences between RDBMSs are minor;
corresponding differences between additionally provided functions (other
than the aggregate ones) and, especially, those between available data
types, extended SQLs, and implementation limitations are much more
important and need very carefully design, implementation, and optimiza-
tion techniques.
For example, even very commonly used built-in data types like integer,
date/time, ASCII, and UNICODE strings or functions for getting current
system date/time and their components have distinct syntax and seman-
tics; moreover, the limitations of the type no autonumbering, no check
constraints, no nondeterministic functions in check constraints, no null
values in indexed columns, etc. are imposing workarounds using extended
SQL stored triggers, functions, and procedures.
Dually, for example, XML, enumerated small static sets, file handlers,
attachments, hyperlinks, geospatial data, (even recursive) user defined
type hierarchies, and all other provided additional data types are easing
the tasks of architecting, designing, and implementing rdbs on the cor-
responding RDBMSs.
RDBMS manufacturers are also providing free editions of their prod-
ucts, but care must be exercised when choosing the best fitted one for your
needs, as various severe limitations apply to them.

4.11 EXERCISES

It is exercise alone that supports the spirits, and keeps the mind in vigor.
—Cicero
536 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

4.11.1  SOLVED EXERCISES

The test and the use of man’s education is that he finds

pleasure in the exercise of his mind.
—Jacques Barzun

4.0. Apply algorithm A2014SQLS1–8 to the E-R data model obtained

in the Exercise 2.0 and add to the resulting script the DML statements
needed in order to populate the db with the instance from Figure 3.26.
Solution:
a. The corresponding SQL DDL script is the following:

CREATE DATABASE PAYMENTS;

CREATE TABLE COUNTRIES(
x smallint IDENTITY PRIMARY KEY,
Country varchar(128) NOT NULL UNIQUE
);
CREATE TABLE CITIES(
x int IDENTITY PRIMARY KEY,
Country smallint NOT NULL,
City varchar(64) NOT NULL,
ZipCode varchar(16) NOT NULL,
CONSTRAINT key_CITIES UNIQUE (ZipCode, Country),
CONSTRAINT FK_CITIES_Country FOREIGN
KEY(Country) REFERENCES
COUNTRIES
);
CREATE TABLE CARD_TYPES(
x smallint IDENTITY PRIMARY KEY,
CardType varchar(16) NOT NULL UNIQUE
);
CREATE TABLE PARTNERS(
x int IDENTITY PRIMARY KEY,
Name varchar(128) NOT NULL,
[Bank?] bit,
Location int NOT NULL,
Relational
 Schemas Implementation and Reverse Engineering 537

Address varchar(255) NOT NULL,

SubsidiaryOf int,
CONSTRAINT key_PARTNERS UNIQUE (Name, Address,
Location),
CONSTRAINT key_PARTNERS_Subsid UNIQUE
(SubsidiaryOf, Address, Location, [Bank?]),
CONSTRAINT FK_PARTNERS_Location FOREIGN KEY
(Location) REFERENCES
CITIES,
CONSTRAINT FK_PARTNERS_SubsidiaryOf FOREIGN KEY
(SubsidiaryOf)
REFERENCES PARTNERS
);
CREATE TABLE ACCOUNTS(
x int IDENTITY PRIMARY KEY,
Holder int NOT NULL,
BankSubsidiary int NOT NULL,
[No.] varchar(32) NOT NULL UNIQUE,
Amount decimal(32, 2) NOT NULL CHECK (Amount
>= 0),
CONSTRAINT KEY_ACCOUNTS UNIQUE ([No.],
BankSubsidiary),
CONSTRAINT FK_ACCOUNTS_BankSubsidiary FOREIGN
KEY(BankSubsidiary)
REFERENCES PARTNERS,
CONSTRAINT FK_ACCOUNTS_Holder FOREIGN
KEY(Holder) REFERENCES
PARTNERS
);
CREATE TABLE CARDS(
x int IDENTITY PRIMARY KEY,
Type smallint NOT NULL,
[No.] char(16) NOT NULL CHECK ([No.] BETWEEN
‘1000000000000000’
AND ‘9999999999999999’),
538 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

SecurityCode char(3) NOT NULL CHECK (

SecurityCode BETWEEN ‘000’ and ‘999’),
Account int NOT NULL,
[Month] smallint NOT NULL CHECK ([Month]
BETWEEN 1 AND 12),
[Year] smallint NOT NULL CHECK ([Year] BETWEEN
Year(GetDate()) –
2 AND Year(GetDate()) + 2),
[D/C] char(1) NOT NULL CHECK ([D/C] IN (‘D’,
‘C’)),
CONSTRAINT key_CARDS UNIQUE ([No.], Type, [Year]),
CONSTRAINT key_CARDS_Account UNIQUE (Account,
[Year]),
CONSTRAINT FK_CARDS_Account FOREIGN KEY(Account)
REFERENCES
ACCOUNTS,
CONSTRAINT FK_CARDS_Type FOREIGN KEY(Type)
REFERENCES CARD_TYPES
);
CREATE TABLE INVOICES(
x int IDENTITY PRIMARY KEY,
Supplier int NOT NULL,
[No.] varchar(16) NOT NULL,
[Date] datetime NOT NULL CHECK ([Date] BETWEEN
‘01/01/2000’ AND getdate())
CONSTRAINT def_inv_date DEFAULT getdate(),
Customer int NOT NULL,
[*TotAmount] decimal(18,2),
CONSTRAINT CHK_INVOICES_AntiCommut CHECK (Supplier
<> Customer),
CONSTRAINT key_INVOICES UNIQUE (Supplier, [No.],
[Date]),
CONSTRAINT FK_INVOICES_Supplier FOREIGN
KEY(Supplier) REFERENCES
PARTNERS,
Relational
 Schemas Implementation and Reverse Engineering 539

CONSTRAINT FK_INVOICES_Customer FOREIGN

KEY(Customer) REFERENCES
PARTNERS
);
CREATE TABLE PAYM_DOCS(
x int IDENTITY PRIMARY KEY,
Debtor int NOT NULL,
[From] int,
[To] int,
Card int,
[No.] varchar(16) NOT NULL,
[Date] datetime NOT NULL CHECK ([Date] BETWEEN
‘01/01/2000’ AND getdate())
CONSTRAINT def_paym_docs_date DEFAULT getdate(),
Type char(1) NOT NULL CHECK (Type IN (‘c’,
‘C’, ‘w’)),
[*TotAmount] decimal(18,2),
CONSTRAINT CHK_PAYM_DOCS_AntiCommut CHECK
([From] <> [To]),
CONSTRAINT key_PAYM_DOCS UNIQUE (Debtor, Type,
[No.], [Date]),
CONSTRAINT key_PAYM_DOCS_From UNIQUE ([No.],
[Date], [From]),
CONSTRAINT key_PAYM_DOCS_To UNIQUE ([No.],
[Date], [To]),
CONSTRAINT key_PAYM_DOCS_Card UNIQUE ([No.],
[Date], Card),
CONSTRAINT FK_PAYM_DOCS_Debtor FOREIGN
KEY(Debtor) REFERENCES
PARTNERS,
CONSTRAINT FK_PAYM_DOCS_From FOREIGN KEY([From])
REFERENCES
PARTNERS,
CONSTRAINT FK_PAYM_DOCS_To FOREIGN KEY([To])
REFERENCES PARTNERS,
540 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

CONSTRAINT FK_PAYM_DOCS_Card FOREIGN KEY(Card)

REFERENCES CARDS
);
CREATE TABLE PAYMENTS (
x int IDENTITY PRIMARY KEY,
PaymDoc int NOT NULL,
Invoice int NOT NULL,
Amount decimal(8,2) NOT NULL CHECK (Amount > 0),
CONSTRAINT key_PAYMENTS UNIQUE (Invoice,
PaymDoc),
CONSTRAINT FK_PAYMENTS_PaymDoc FOREIGN
KEY(PaymDoc) REFERENCES
PAYM_DOCS,
CONSTRAINT FK_PAYMENTS_Invoice FOREIGN
KEY(Invoice) REFERENCES
INVOICES
);
b. The corresponding non-relational constraints set is the following
(for their enforcement see exercise 3.0 of the second volume of this
book):
Ø SubsidiaryOf should be acyclic. (P6)
Ø All and only the subsidiaries of a bank should be
of bank type. (P7)
Ø No payment for an invoice can be done before the
invoice is established or after more than 100 years
after corresponding establishing date. (PD6)
Ø Whenever Type is ‘c’ (cash), To, From, and Card
should be null. (PD7)
Ø Whenever Type is ‘C’ (card), From should be null,
whereas To and Card should not be null. (PD8)
Ø Whenever Type is ‘w’ (bank wire), Card should be
null and To and From should not be null. (PD9)
Ø Sum of all payments for a same invoice should not be
greater than the corresponding invoice total amount. (PM4)
Ø Bank subsidiaries should always be bank type
partners. (A5)
Relational
 Schemas Implementation and Reverse Engineering 541

Ø No card payment may be done with an expired card. (C8)

c. The needed DML statements are the following (where function

CAST(v, t) converts value v to data type t):

INSERT COUNTRIES (Country) VALUES (‘France’);

INSERT COUNTRIES (Country) VALUES (‘Romania’);
INSERT COUNTRIES (Country) VALUES (‘Holland’);
INSERT CITIES (Country, City, ZipCode) VALUES (1,
‘Chouilly’, ‘51530’);
INSERT CITIES (Country, City, ZipCode) VALUES (3,
‘Amsterdam’, ‘10’);
INSERT CITIES (Country, City, ZipCode) VALUES (2,
‘Bucharest’, ‘0’);
INSERT PARTNERS(Name, [Bank?], Location, Address,
SubsidiaryOf)
VALUES (‘Champagne Nicolas Feuillate’, NULL, 1,
‘CD 40 A «Plumecoq», 51530’, NULL);
INSERT PARTNERS(Name, [Bank?], Location, Address,
SubsidiaryOf)
VALUES (‘ING Commercial Banking’, 1, 3,
‘Amsterdamse Poort Building Bijlmerplein 8881102 MG’,
NULL);
INSERT PARTNERS(Name, [Bank?], Location, Address,
SubsidiaryOf)
VALUES (‘ING Bank N.V. Amsterdam Romania’, 1, 2,
‘Bd. Iancu de Hunedoara nr. 48, Sector 1, 011745’,
2);
INSERT INVOICES (Supplier, [No.], [Date],
Customer, [*TotAmount])
VALUES (1, ‘12’, ‘11/21/2012’, 2, CAST(1000.00 AS
Decimal(18, 2)));
INSERT INVOICES (Supplier, [No.], [Date],
Customer, [*TotAmount])
VALUES (1, ‘13’, ‘11/21/2012’, 3, CAST(200.00 AS
Decimal(18, 2)));
542 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

INSERT INVOICES (Supplier, [No.], [Date],

Customer, [*TotAmount])
VALUES (1, ‘123’, ‘12/21/2012’, 3, CAST(300.00 AS
Decimal(18, 2)));
INSERT ACCOUNTS (Holder, BankSubsidiary, [No.],
Amount)
VALUES (1, 2, ‘HO01INGB0000999900111111’,
CAST(10000000.00 AS Decimal(32, 2)));
INSERT ACCOUNTS (Holder, BankSubsidiary, [No.],
Amount)
VALUES (2, 2, ‘HO01INGB0000999900000000’,
CAST(99999999999.00 AS Decimal(32, 2)));
INSERT ACCOUNTS (Holder, BankSubsidiary, [No.],
Amount)
VALUES (3, 3, ‘RO81INGB0000999900123456’,
CAST(100000.00 AS Decimal(32, 2)));
INSERT CARD_TYPES (CardType) VALUES (‘American
Express’);
INSERT CARD_TYPES (CardType) VALUES (‘Visa’);
INSERT CARD_TYPES (CardType) VALUES
(‘Visa Electron’);
INSERT CARD_TYPES (CardType) VALUES (‘Maestro’);
INSERT CARDS (Type, [No.], SecurityCode, Account,
[Month], [Year], [D/C])
VALUES (2, ‘1234567898765432’, ‘123’, 1, 1, 2015,
‘C’);
INSERT PAYM_DOCS (Debtor, [From], [To], [No.],
[Date], Type, [*TotAmount])
VALUES (2, 2, 1, ‘210’, ‘11/30/2012’, ‘w’,
CAST(500.00 AS Decimal(32, 2)));
INSERT PAYM_DOCS (Debtor, [From], [To], [No.],
[Date], Type, [*TotAmount])
VALUES (2, 2, 1, ‘320’, ‘12/30/2012’, ‘w’,
CAST(500.00 AS Decimal(32, 2)));
INSERT PAYM_DOCS (Debtor, [From], [To], [No.],
[Date], Type, [*TotAmount])
Relational
 Schemas Implementation and Reverse Engineering 543

VALUES (3, 3, 1, ‘101’, ‘12/30/2012’, ‘w’,

CAST(500.00 AS Decimal(32, 2)));
INSERT PAYMENTS (PaymDoc, Invoice, Amount)
VALUES (1, 1, CAST(500.00 AS Decimal(32, 2)));
INSERT PAYMENTS (PaymDoc, Invoice, Amount)
VALUES (2, 1, CAST(500.00 AS Decimal(32, 2)));
INSERT PAYMENTS (PaymDoc, Invoice, Amount)
VALUES (3, 2, CAST(200.00 AS Decimal(32, 2)));
INSERT PAYMENTS (PaymDoc, Invoice, Amount)
VALUES (3, 3, CAST(300.00 AS Decimal(32, 2)));

4.1. Prove the following Proposition 4.1 (Algorithm A8 characteriza-

tion): A8 has the following properties:
(i) its complexity is O(n), where n = t + c + r + c’, where t is the
number of input tables, c is the total one of their columns, r is the
total one of their relational constraints, and c’ is the total number
of column occurrences that make up their concatenated keys;
(ii) it is complete (i.e., it is generating DDL creation statements for all
tables, columns, and associated relational constraints);
(iii) it is sound (i.e., it is generating and outputting only necessary DDL
statements for implementing the input rdb scheme).
Proof:
(i) (Complexity) Obviously, as it has six embedded loops, all of them
finite (hence, it never loops infinitely) and the first one takes t steps,
the second one c steps plus part of the r ones –those for the single
keys and the not nulls–, the third one takes another part –those for
the concatenated keys, which embeds the fourth one that takes c’
steps–, and the fifth and sixth ones with the rest of r, for the refer-
ential integrities and tuple/check constraints, respectively. Conse-
quently, the total number of steps is n = t + c + r + c’.
(ii) (Completeness) Obviously, all tables, columns, and relational con-
straints are considered.
(iii) (Soundness) Trivially, no other SQL statements than those needed
are generated. Q.E.D.
544 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

4.2.
a. Apply algorithm REA2013A0–2 to the MS Access Northwind
demo db.
b. Apply algorithm A7/8–3 to the table Purchase Order Details of the
MS Access Northwind demo db and add all discovered semantic
keys to its scheme.
Solution:
a. If you did not already downloaded the MS Access Northwind db,
first do the following:
1. Open MS Access 2013.
2. In its top screen Search for Online Templates text-box type
Northwind and then press Enter: the window shown in Fig.
4.89 will open.
3. Click on its Northwind icon: the window shown in Fig. 4.90
will open.
4. Choose the desired name for the Northwind db and the folder of
your PC where you want it downloaded, and then click on the Cre-
ate tile: the window shown in Fig. 4.91 will open.
5. Click on the Enable Content button from its SECURITY WARN-
ING row: the window shown in Fig. 4.92 will open.
6. Either leave the default demo employee “Andrew Cencini” or
select any other one and then click on the Login button: the win-
dow shown in Fig. 4.93 will open.
7. Close the Home tab, open the Navigation Pane, select its All
Access Objects container, and open its Tables subset one: the win-
dow shown in Fig. 4.94 will open.
As it can be seen from Fig. 4.94, the Northwind db has the follow-
ing 20 tables: Customers, Employee Privileges, Employees, Inventory
Transaction Types, Inventory Transactions, Invoices, Order Details,
Order Details Status, Orders, Orders Status, Orders Tax Status, Priv-
ileges, Products, Purchase Order Details, Purchase Order Status,
Purchase Orders, Sales Reports, Shippers, Strings, and Suppliers.
8. Enable showing of the Access system metacatalog tables too (see
subsection 4.4, Figs. 4.21 and 4.22): the window shown in Fig.
4.94 will then look like in Fig. 4.95.
9. Double-click on the MSysRelationships table, close the Navigation
Pane, resize MSysRelationships columns so that all data is fully
Relational
 Schemas Implementation and Reverse Engineering 545

FIGURE 4.89  Access 2013 Northwind demo db download selection window.

FIGURE 4.90  Access 2013 Northwind demo db download window.

546 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

FIGURE 4.91  Access 2013 Northwind demo db Startup Screen.

FIGURE 4.92  Access 2013 Northwind demo db Login Dialog window.

Relational
 Schemas Implementation and Reverse Engineering 547

FIGURE 4.93  Access 2013 Northwind demo db application Home tab.

FIGURE 4.94  Access 2013 Northwind demo db tables.

548 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

FIGURE 4.95  Access 2013 Northwind demo db tables, including the system ones.

FIGURE 4.96  Access 2013 Northwind demo db MSysRelationships table instance.

Relational
 Schemas Implementation and Reverse Engineering 549

displayed, and then sort in ascending order this table instance on

the data of its column szObject (right-click on its name and then
click on the Sort A to Z option): the window shown in Fig. 4.96 will
open.
As it can be seen from Fig. 4.96, the Northwind db has the following 21
referential integrity constraints (as the ones of the MSysNavPaneGroups
and MSysNavPaneGroupToObjects tables are system ones):
Ø New_EmployeePriviligesforEmployees: column Employee ID of
table Employee Privileges references column ID of table Employees;
Ø New_EmployeePriviligesLookup: column Privilege ID of table
Employee Privileges references column Privilege ID of table
Privileges;
Ø New_OrdersOnInventoryTransactions: column Customer Order
ID of table Inventory Transactions references column Order ID of
table Orders;
Ø New_PuchaseOrdersonInventoryTransactions: column Purchase
Order ID of table Inventory Transactions references column Pur-
chase Order ID of table Purchase Orders;
Ø New_ProductOnInventoryTransaction: column Product ID of table
Inventory Transactions references column ID of table Products;
Ø New_TransactionTypesOnInventoryTransactiosn: column Trans-
action Type of table Inventory Transactions references column ID
of table Inventory Transaction Types;
Ø New_OrderInvoice: column Order ID of table Invoices references
column Order ID of table Orders;
Ø New_OrderDetails: column Order ID of table Order Details refer-
ences column Order ID of table Orders;
Ø New_OrderStatusLookup: column Status ID of table Order Details
references column Status ID of table Order Details Status;
Ø New_ProductsOnOrders: column Product ID of table Order
Details references column ID of table Products;
Ø New_EmployeesOnOrders: column Employee ID of table Orders
references column ID of table Employees;
Ø New_ShipperOnOrder: column Shipper ID of table Orders refer-
ences column ID of table Shippers;
Ø New_CustomerOnOrders: column Customer ID of table Orders
references column ID of table Customers;
550 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

Ø New_TaxStatusOnOrders: column Tax Status of table Orders refer-

ences column ID of table Orders Tax Status;
Ø New_OrderStatus: column Status ID of table Orders references
column Status ID of table Orders Status;
Ø New_ProductOnPurchaseOrderDetails: column Product ID of table
Purchase Order Details references column ID of table Products;
Ø New_PurchaseOrderDeatilsOnPurchaseOrder: column Purchase
Order ID of table Purchase Order Details references column Pur-
chase Order ID of table Purchase Orders;
Ø New_InventoryTransactionsOnPurchaseOrders: column Inventory
ID of table Purchase Order Details references column Transaction
ID of table Inventory Transactions;
Ø New_PurchaseOrderStatusLookup: column Status ID of table Pur-
chase Orders references column Status ID of table Purchase Order
Status;
Ø New_EmployeesOnPurchaseOrder: column Created By of table
Purchase Orders references column ID of table Employees;
Ø New_SuppliersOnPurchaseOrder: column Supplier ID of table
Purchase Orders references column ID of table Suppliers.
Next steps require to open in Design View, one by one, all of the 20
tables, browse through all of their 177 columns (in total), and inspect all of
their 67 non-primary indexes (see, as similar examples, Sections 4.4 and
4.6). In order to save corresponding editorial space, we apologize to the
readers for not having included these additional 244 print screens (which
would need another 122 pages!) and invite them to do it as homework.
Figures 4.97–4.117 present the corresponding E-R data model obtained
after applying REA20130–2 to this Northwind db, followed by the associ-
ated restrictions set and subuniverse description.
Note that table Products, because of its multivalued column Supplier
IDs (represented in Fig. 4.116 as a double-edged “multivalued” arrow), is
not in 1NF.
Please note that, very probably, in fact, there are other 4 referential
integrity constraints missing from this db, as, in Order Details, Purchase
Order ID and Inventory ID are, in fact, foreign keys referencing Purchase
Orders and Inventory Transactions, respectively, just as in Purchase
Relational
 Schemas Implementation and Reverse Engineering 551

Orders both Approved By and Submitted By are, in fact, foreign keys ref-
erencing Employees.
This is why all these 4 columns are represented in the corresponding
E-RDs above both as ellipses (as they are) and as arrows (as they should
actually be).
The corresponding associated restrictions set is the following:
1. Customers (The set of customers of interest)
a. Cardinality:max(card(Customers)) = 109 (ID) (C0)
b. Data ranges:
Company, Last Name, First Name, E-mail Address,
Job Title, City, State/Province,
Country/Region: UNICODE(50) (C1)
Business Phone, Home Phone, Mobile Phone,
Fax Number: UNICODE(25) (C2)
Address, Notes: UNICODE(4096) (C3)
ZIP/Postal Code: UNICODE(15) (C4)
Web Page: HYPERLINK (C5)
Attachments: ATTACHMENT (C6)
c. Compulsory data: ID (C7)
d. Uniqueness: ID (trivial: there may not be two
customers having same autogenerated id numbers)    (C8)
2. Employees (The set of employees of interest)
a. Cardinality:max(card(Employees)) = 109 (ID) (E0)
b. Data ranges:
Company, Last Name, First Name, E-mail Address,
Job Title, City, State/Province,
Country/Region: UNICODE(50) (E1)
Business Phone, Home Phone, Mobile Phone,
Fax Number: UNICODE(25) (E2)
Address, Notes: UNICODE(4096) (E3)
ZIP/Postal Code: UNICODE(15) (E4)
Web Page: HYPERLINK (E5)
Attachments: ATTACHMENT (E6)
c. Compulsory data: ID (E7)
d. Uniqueness: ID (trivial: there may not be two
employees having same autogenerated id numbers) (E8)
552 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

FIGURE 4.97  Access 2013 Northwind Customers E-RD.

FIGURE 4.98  Access 2013 Northwind Employees E-RD.

FIGURE 4.99  Access 2013 Northwind Shippers E-RD.

Relational
 Schemas Implementation and Reverse Engineering 553

FIGURE 4.100  Access 2013 Northwind Suppliers E-RD.

FIGURE 4.101  Access 2013 Northwind Inventory Transaction Types E-RD.

FIGURE 4.102  Access 2013 Northwind Privileges E-RD.

FIGURE 4.103  Access 2013 Northwind Order Details Status E-RD.

FIGURE 4.104  Access 2013 Northwind Orders Status E-RD.

554 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

FIGURE 4.105  Access 2013 Northwind Privileges E-RD.

FIGURE 4.106  Access 2013 Northwind Strings E-RD.

FIGURE 4.107  Access 2013 Northwind Inventory Transactions E-RD.

FIGURE 4.108  Access 2013 Northwind Invoices E-RD.

FIGURE 4.109  Access 2013 Northwind Orders E-RD.

Relational
 Schemas Implementation and Reverse Engineering 555

114

115

FIGURE 4.110  Access 2013 Northwind Order Details E-RD.

116

FIGURE 4.111  Access 2013 Northwind Products E-RD.

FIGURE 4.112  Access 2013 Northwind Purchase Orders E-RD.

FIGURE 4.113  Access 2013 Northwind Purchase Order Details E-RD.

114, 115, 116

114
In fact, very probably, Purchase Order ID should be a foreign key referencing Purchase Orders,
that is an arrow in Fig. 4.116, not an ellipse.
115
In fact, very probably, Inventory ID should be a foreign key referencing Inventory Transactions,
that is an arrow in Fig. 4.116, not an ellipse.
116
Multivalued attribute, taking values in the powerset of Suppliers.
556 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

FIGURE 4.114  Access 2013 Northwind Purchase Order Status E-RD.

FIGURE 4.115  Access 2013 Northwind Sales Reports E-RD.

FIGURE 4.116  Access 2013 Northwind main structural E-RD.

Relational
 Schemas Implementation and Reverse Engineering 557

FIGURE 4.117  Access 2013 Northwind additional structural E-RD.

3. Suppliers (The set of suppliers of interest)

a. Cardinality:max(card(Suppliers)) = 109 (ID) (S0)
b. Data ranges:
Company, Last Name, First Name, E-mail Address,
Job Title, City, State/Province,
Country/Region: UNICODE(50) (S1)
Business Phone, Home Phone, Mobile Phone,
Fax Number: UNICODE(25) (S2)
Address, Notes: UNICODE(4096) (S3)
ZIP/Postal Code: UNICODE(15) (S4)
Web Page: HYPERLINK (S5)
Attachments: ATTACHMENT (S6)
c. Compulsory data: ID (S7)
d. Uniqueness: ID (trivial: there may not be two suppliers
having same autogenerated id numbers) (S8)
4. Shippers (The set of shippers of interest)
a. Cardinality:max(card(Shippers)) = 109 (ID) (H0)
b. Data ranges:
Company, Last Name, First Name, E-mail Address,
Job Title, City, State/Province,
Country/Region: UNICODE(50) (H1)
558 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

Business Phone, Home Phone, Mobile Phone,

Fax Number: UNICODE(25) (H2)
Address, Notes: UNICODE(4096) (H3)
ZIP/Postal Code: UNICODE(15) (H4)
Web Page: HYPERLINK (H5)
Attachments: ATTACHMENT (H6)
c. Compulsory data: ID (H7)
d. Uniqueness: ID (trivial: there may not be two
shippers having same autogenerated id numbers) (H8)
5. Inventory Transaction Types (The set of inventory
transaction types of interest)
a. C ardinality:max(card(Inventory Transaction
Types)) = 109 (ID) (T0)
b. Data ranges:
Type Name: UNICODE(50) (T1)
c. Compulsory data: ID, Type Name (T2)
d. Uniqueness: ID (trivial: there may not be two inventory
transaction types having same autogenerated id numbers) (T3)
6. Order Details Status (The set of order details status of interest)
a. C ardinality:max(card(Order Details Status)) = 109
(Status ID) (ODS0)
b. Data ranges:
Status Name: UNICODE(50) (ODS1)
c. Compulsory data: Status ID, Status Name (ODS2)
d. Uniqueness: Status ID (trivial: there may not be two
order details status having same autogenerated id
numbers) (ODS3)
7. Orders Status (The set of orders status of interest)
a. C ardinality:max(card(Orders Status)) = 109
(Status ID) (OS0)
b. Data ranges:
Status Name: UNICODE(50) (OS1)
c. Compulsory data: Status ID, Status Name (OS2)
d. Uniqueness: Status ID (trivial: there may not be two
orders status having same autogenerated id numbers) (OS3)
Relational
 Schemas Implementation and Reverse Engineering 559

8. Orders Tax Status (The set of orders tax status of interest)

a. Cardinality:max(card(Orders Tax Status)) = 109 (ID) (OTS0)
b. Data ranges:
Tax Status Name: UNICODE(50) (OTS1)
c. Compulsory data: Status ID, Tax Status Name (OTS2)
d. Uniqueness: Status ID (trivial: there may not be two
orders tax status having same autogenerated id numbers) (OTS3)
9. Purchase Order Status (The set of purchase order status
of interest)
a. C ardinality:max(card(Purchase Orders Status)) = 109
(Status ID) (POS0)
b. Data ranges:
Status: UNICODE(50) (POS1)
c. Compulsory data: Status ID (POS2)
d. Uniqueness: Status ID (trivial: there may not be two
purchase order status having same autogenerated id
numbers) (POS3)
10. Privileges (The set of types of privileges of interest that
employees may have on the MS Access 2013 Northwind
db application objects)
a. Cardinality:max(card(Privileges)) = 109 (Privilege ID) (R0)
b. Data ranges:
Privilege Name: UNICODE(50) (R1)
c. Compulsory data: Privilege ID (R2)
d. Uniqueness: Privilege ID (trivial: there may not be two
types of privileges having same autogenerated id numbers) (R3)
11. Categories (The set of product categories of interest)
a. Cardinality:max(card(Categories)) = 128 (G0)
b. Data ranges:
Category: {“Baked Goods & Mixes”, “Beverages”,
“Candy”, “Canned Fruit & Vegetables”,
“Canned Meat”, “Cereal”, “Chips”, “Snacks”, “Condiments”,
“Dairy Products”,
“Dried Fruit & Nuts”, “Grains”, “Jams”, “Preserves”,
“Oil”, “Pasta”, “Sauces”, “Soups”} (G1)
c. Compulsory data: Category (G2)
560 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

d. Uniqueness: Category (there may not be two product

categories having same names) (G3)
12. PaymMethods (The set of purchase orders payment
methods of interest)
a. Cardinality:max(card(PaymMethods)) = 8 (M0)
b. Data ranges:
Payment Method: {“Credit card”, “Check”, “Cash”} (M1)
c. Compulsory data: Payment Method (M2)
d. Uniqueness: Payment Method (there may not be two
payment methods having same names) (M3)
13. Employee Privileges (The set of pairs <e, p> storing the
fact that employee e has privilege p on the objects of the
MS Access 2013 Northwind db application objects.)
a. Cardinality:max(card(Employee Privileges)) = 130,000  (EP0)
c. Compulsory data: Employee ID, Privilege ID (EP1)
d. Uniqueness: Employee ID • Privilege ID (there is no
use in storing more than once any pair <e, p>) (EP2)
14. Invoices (The set of all issued invoices by the Nortwhind
Traders to their customers since the beginning of the
twentieth century.)
a. Cardinality:max(card(Invoices)) = 109 (Invoice ID) (I0)
b. Data ranges:
Invoice Date, Due Date: [1/1/1900, 12/31/9999] (I1)
Tax, Shipping, Amount Due: CURRENCY(14) (I2)
c. Compulsory data: Invoice ID (I3)
d. Uniqueness: Invoice ID (trivial: there may not be two
invoices having same autogenerated id numbers) (I4)
15. Orders (The set of all orders placed to the Nortwhind
Traders by their customers since the beginning of the
twentieth century.)
a. Cardinality:max(card(Orders)) = 109 (Order ID) (O0)
b. Data ranges:
Order Date, Shipped Date, Paid Date:
[1/1/1900, 12/31/9999] (O1)
Shipping Fee, Taxes: CURRENCY(14) (O2)
Relational
 Schemas Implementation and Reverse Engineering 561

 Ship Name, Ship City, Ship State/Province,

Ship ZIP/Postal Code, Ship Country/Region:
UNICODE(50) (O3)
Ship Address, Notes: UNICODE(4096) (O4)
c. Compulsory data: Order ID (O5)
d. Uniqueness: Order ID (trivial: there may not be two
orders having same autogenerated id numbers) (O6)
16. Purchase Orders (The set of all purchase orders placed
by the Nortwhind Traders to their suppliers since the
beginning of the twentieth century.)
a. C
 ardinality:max(card(Purchase Orders)) = 109
(Purchase Order ID) (PO0)
b. Data ranges:
Submitted Date, Creation Date, Expected Date,
Payment Date, Approved Date: [1/1/1900, 12/31/9999] (PO1)
 Shipping Fee, Taxes, Payment Amount:
CURRENCY(14) (PO2)
Notes: UNICODE(4096) (PO3)
c. Compulsory data: Purchase Order ID (PO4)
d. Uniqueness: Purchase Order ID (trivial: there may
not be two purchase orders having same
autogenerated id numbers) (PO5)
17. Products (The set of products that the Nortwhind
Traders ever bought from their suppliers, stored in their
stocks, and/or sold to their customers.)
a. Cardinality:max(card(Products)) = 109 (ID) (P0)
b. Data ranges:
Product Code: UNICODE(25) (P1)
Product Name, Quantity Per Unit: UNICODE(50) (P2)
Standard Cost, List Price: CURRENCY(14) (P3)
 Minimum Reorder Quantity, Reorder Level (Inventory
quantity that triggers reordering),
Target Level (Desired Inventory level after a purchase
reorder): INT(9) (P4)
Description: UNICODE(4096) (P5)
562 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

Discontinued: BOOLEAN (P6)

Attachments: ATTACHMENT (P7)
c. Compulsory data: ID, List Price (P8)
d. Uniqueness: ID (trivial: there may not be two
products having same autogenerated id numbers) (P9)
18. Inventory Transactions (The set of all Nortwhind
Traders inventory transactions since the beginning of
the twentieth century.)
a. C ardinality:max(card(Inventory Transactions)) = 109
(Transaction ID) (IT0)
b. Data ranges:
Transaction Created Date, Transaction Modified Date:
[1/1/1900, 12/31/9999] (IT1)
Quantity: INT(9) (IT2)
Comments: UNICODE(255) (IT3)
c. Compulsory data: Transaction ID, Transaction Type,
Product ID, Quantity (IT4)
d. Uniqueness: Transaction ID (trivial: there may not be
two inventory transactions having same autogenerated
id numbers) (IT5)
19. Order Details (The set of all details of the orders placed
to the Nortwhind Traders by their customers since the
beginning of the twentieth century.)
a. Cardinality:max(card(Order Details)) = 109 (ID) (OD0)
b. Data ranges:
Date Allocated: [1/1/1900, 12/31/9999] (OD1)
Unit Price: CURRENCY(14) (OD2)
Quantity: RAT(18, 4) (OD3)
Discount: [0, 1] , RAT(0, 308) (OD4)
c. Compulsory data: ID, Order ID, Quantity, Discount (OD5)
d. Uniqueness: ID (trivial: there may not be two order
details having same autogenerated id numbers) (OD6)
20. Purchase Order Details (The set of all details of the
purchase orders placed by the Nortwhind Traders to their
suppliers since the beginning of the twentieth century.)
Relational
 Schemas Implementation and Reverse Engineering 563

a. C ardinality:max(card(Purchase Order Details)) =

109 (ID) (POD0)
b. Data ranges:
Date Received: [1/1/1900, 12/31/9999] (POD1)
Unit Cost: CURRENCY(14) (POD2)
Quantity: RAT(18, 4) (POD3)
Posted To Inventory: BOOLEAN (POD4)
c. Compulsory data: ID, Purchase Order ID, Quantity,
Unit Cost (POD5)
d. Uniqueness: ID (trivial: there may not be two purchase
order details having same autogenerated id numbers) (POD6)
21. Sales Reports (The set of all sales reports provided by
the Nortwhind Traders db application.)
a. C ardinality:max(card(Sales Reports)) = 1015
(Group By) (SR0)
b. Data ranges:
Group By, Display, Title: UNICODE(50) (SR1)
Filter Row Source: UNICODE(4096) (SR2)
Default: BOOLEAN (SR3)
c. Compulsory data: Group By (SR4)
d. Uniqueness: Group By (there may not be two sales
reports having same group by clauses) (SR5)
22. Strings (The set of all strings used by the Nortwhind Traders
demo db application for its titles, messages, etc.)
a. Cardinality:max(card(Strings)) = 109 (String ID) (S0)
b. Data ranges:
String Data: UNICODE(255) (S1)
c. Compulsory data: String ID (S2)
d. Uniqueness: String ID (trivial: there may not be two
strings having same autogenerated id numbers) (S3)
Figure 4.118 contains the corresponding description of the MS Access
Nortwhind Traders demo db.
b. Figure 4.119 shows the scheme of the MS Access Nortwhind Trad-
ers demo db Purchase Order Details table and Fig. 4.120 its indexes.
Besides its surrogate primary key ID, the table has the following 7
columns: Purchase Order ID, Product ID, Quantity, Unit Cost, Date
564 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

Received, Posted To Inventory, and Inventory ID. Besides the PrimaryKey

index, the table has no other unique index, so no other key.
i = 1:
ü Purchase Order ID is prime, but it is not a key, as any purchase
order may have several detail lines (products);
ü Product ID is prime, but it is not a key, as any product may be
ordered in several purchase orders and even partially received for a
same such order;
ü Inventory ID is a key, as no inventory transaction may correspond to
several purchase order details (products): K = { ID, Inventory ID };
ü Quantity is not a key, as there may be, even in a same purchase
order, several products ordered in a same quantity and it is not
prime either: from the uniqueness point of view, it does not matter
in what quantity a product is ordered and/or received, so Quantity
cannot participate in any uniqueness of purchase order details;
ü Unit Cost is not a key, as there may be, even in a same purchase
order, several products having a same unit cost, and it is not prime
either: from the uniqueness point of view, it does not matter which
is the unit cost of a product, so Unit Cost cannot participate in any
uniqueness of purchase order details;
ü Date Received is prime, but not a key, as there may be, even in a
same purchase order, several products having a same received date;
ü Posted To Inventory is not a key, as there may be, even in a same
purchase order, several products having been posted to inventory
and several other ones having not yet been posted to inventory, and
it is not prime either: from the uniqueness point of view, it does not
matter whether an ordered product received quantity was posted
to inventory or not (the ones not yet received cannot of course be
posted), so Posted To Inventory cannot participate in any unique-
ness of purchase order details.
i = 2:
In this step we are left with only three prime attributes: Purchase Order
ID, Product ID, and Date Received.
ü Purchase Order ID • Product ID is not a key, as even in a same
purchase order, some amounts of an ordered product may have
been received (at different dates) and the rest of the ordered
amount not yet.
Relational
 Schemas Implementation and Reverse Engineering 565

The MS Access Northwind Traders sample db consists of 2 unre-

lated and 20 related object sets. The unrelated sets are:
Ø Sales Reports, which stores all sales reports provided by the
Northwind Traders db application (grouped by column names
–at most 50 characters, compulsory and unique–, title, and dis-
played name –both at most 50 characters–, corresponding SQL
statements for computing the row source –at most 4096 charac-
ters–, and whether it is the default sales report or not);
Ø Strings, which stores the set of all strings used by the Northwind
Traders sample db application for its titles, messages, etc. (an
autonumber and the corresponding string data of at most 255
characters).
The related sets store data since 1/1/1900 on the following items
(all currency-type ones may have at most 14 mantissa digits; all
sets may have at most 109 elements and have an autonumber id,
which is their only uniqueness):
Ø Customers, Suppliers, Shippers, and Employees (company, last
and first name, e-mail address, job title, city, state/province, and
country/region –at most 50 characters each of them–, business,
home, mobile phone, and fax number –at most 25 characters
each of them–, address and notes –at most 4096 characters each
of them–, zip/postal code –at most 15 characters–, a web page
hyperlink, and file attachments);
Ø Products traded (code –at most 25 characters–, suppliers, cat-
egory –one of 18 predefined ones–, name and quantity per unit
–at most 50 characters each of them–, standard cost and compul-
sory list price –currency values of at most 14 digits–, minimum
reorder quantity, inventory quantity that triggers reordering, and
desired inventory level after a purchase reorder –integers of at
most 9 digits–, description –at most 4096 characters–, whether
discontinued or not, and file attachments);
Ø Invoices (customer order, issue and due dates, tax, shipping, and
amount due currency values);
Ø Inventory transactions (type, product, purchase/customer order,
create and modified dates, quantity –an integer of at most 9 digits–,

FIGURE 4.118  Access 2013 Northwind db description.

566 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

and comments –at most 255 characters–; compulsory are only the
type, product, and quantity);
Ø Purchase orders (supplier, status, employees that submitted,
created, and approved it, payment type –one of 3 predefined
ones–, submitted, creation, expected, payment, approved, and
(per product) received dates, shipping fee, taxes, payment
amount, and (per product) unit cost currencies, products and
corresponding inventory transactions, per product quantities –
rationals of at most 18 mantissa and 4 scale digits–, per product
notes –at most 4096 characters–, and whether received products
were posted to inventory or not; compulsory are only the id,
purchase order id, quantity, and unit cost);
Ø Customer orders (customer, status, tax status, and per product
detail status, employee in charge, shipper, payment type –one of
3 predefined ones–, order, shipped, paid, and (per product) allo-
cated dates, shipping fee, taxes, and (per product) unit cost cur-
rencies, products and corresponding inventory transactions and
purchase orders, per product quantities –rationals of at most 18
mantissa and 4 scale digits–, ship name, city, state/province, zip/
postal code, and country/region –at most 50 characters–, ship
address and notes –at most 4096 characters–, and per product
discount percentage; compulsory are only the order id, quantity,
and discount);
Ø Purchase and Customer Order and Details Status, including
tax ones, and inventory transaction types (compulsory name, at
most 50 characters);
Ø Employee Privilege Types (name, at most 50 characters);
Ø Employee Privileges (unique compulsory pairs of employees
and privilege types).
FIGURE 4.118  (Continued)

ü Purchase Order ID • Date Received is not a key, as even in a same

purchase order, some amounts of different ordered products may
have been received at a same date.
Relational
 Schemas Implementation and Reverse Engineering 567

ü Product ID • Date Received is not a key, as there may happen that

same products were received at same dates, but for different pur-
chase orders.
i = 3:
ü Purchase Order ID • Product ID • Date Received is a key, as no
product should appear more than once in any purchase order as
being received at a same date.
Consequently, the algorithm stops (as no other combinations of prime
attributes are possible) and the Purchase Order Details table should have

FIGURE 4.119  The Purchase Order Details table scheme.

568 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

FIGURE 4.120  The Purchase Order Details table initial indexes.

the following three keys: K = { ID, Inventory ID, Purchase Order ID •

Product ID • Date Received }.
The corresponding SQL DDL statements for enforcing these keys are
the following:

ALTER TABLE [Purchase Order Details]

ADD CONSTRAINT keyPODInventoryID UNIQUE ([Inventory
ID]);

ALTER TABLE [Purchase Order Details]

ADD CONSTRAINT keyPODProductID UNIQUE ([Purchase
Order ID],
[Date Received], [Product ID]);

Figure 4.121 shows the Index window of this table after declaring the
keyPODInventoryID above key, while Fig. 4.122 shows it after declaring
the keyPODProductID too.
Relational
 Schemas Implementation and Reverse Engineering 569

FIGURE 4.121  The Purchase Order Details keyPODInventoryID key.

FIGURE 4.122  The Purchase Order Details keyPODProductID key.

570 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

4.11.2  PROPOSED EXERCISES

Many difficulties which nature throws in our way may be smoothed

away by the exercise of intelligence.
—Titus Livius

4.3. Based on Section 4.2, for automatizing translations of ANSI SQL

DDL scripts into corresponding scripts of various RDBMS versions:
a. design the algorithms of the family AF8′ for DB2 10.5, Oracle 12c,
MySQL 5.7, SQL Server 2014, and Access 2013;
b. design such algorithms for at least one current version of another
RDBMS vendor or open source consortium (e.g., PostgreSQL,
MongoDB, etc.);
c. based on Tables 4.3–4.12, establish a consolidated table for pin-
pointing both equivalences and differences between the data types
provided by the five RDBMS versions considered in this book.
*4.4. Extend algorithm A8 with the following features:
a. structures and abstract data types;
b. XML;
c. geospatial data types.
4.5. For automatizing translations of various RDBMS versions’ SQL
DDL scripts into corresponding ANSI standard ones:
a. design the algorithms of the family REAF0′ for DB2 10.5, Oracle
12c, MySQL 5.7, and SQL Server 2014;
b. design such algorithms for at least one current version of another
RDBMS vendor or open source consortium (e.g., PostgreSQL,
MongoDB, etc.).
4.6. Design the reverse engineering algorithm REA0, for translating
SQL ANSI DDL scripts into relational db schemas.
4.7. Modify algorithm A8 in order for it to also treat the totally not
recommended concatenated foreign keys (hint: take as an example its loop
for generating concatenated (unique) keys).
4.8. Modify algorithm REA2013A0 such that it also copes with tables
having no primary keys, and/or having not autonumber or integer primary
keys referencing other autonumber ones, and/or having primary keys
made up of several columns, and/or having primary keys with other data
types than the integer ones.
Relational
 Schemas Implementation and Reverse Engineering 571

4.9. Consider the case of rdbs in which there is at least one cycle in
their corresponding E-RD graph in which no two arrows meet sharp point
to sharp point:
a. should such cycles be allowed by RDBMSs? Why?
b. if yes, design needed (if any) modifications to algorithm
REA2013A0 such that it also copes with such cases.
4.10. Apply algorithm A2014SQLS1–8 to:
a. the E-R data model of the public library case study from Subsec-
tions 2.8.1 and 2.8.2; compare its result with the DDL script from
Subsection 4.3.4
b. the E-R data metamodel of the E-RDM presented in Section 2.9,
in order to obtain a MS SQL Server DDL script for generating new
metacatalog tables for it (that might be used when designing and
developing an E-RDM GUI)
c. the RDM metamodel of itself presented in subsection 3.7.2
*d. compare the results of c) above with the SQL Server 2014′s cor-
responding metacatalog tables.
4.11.
a. Design algorithms similar to A2014SQLS1–8 for DB2 10.5, Oracle
12c, MySQL 5.7, Access 2013.
*b. Same thing as above for the latest versions of PostgreSQL and
MongoDB.
4.12.
a. Design algorithms similar to A2014SQLS1–8 for SQL Server 2014,
DB2 10.5, Oracle 12c, MySQL 5.7, Access 2013, but for corre-
sponding instructions on how to implement rdbs through their
GUIs (not through SQL DDL scripts).
*b. Same thing as above for the latest versions of PostgreSQL and
MongoDB.
4.13
a. Design extensions to both REA2013A0 and REA2013A0–2 so that
they also generate DML scripts to populate dbs with the source rdb
instances.
b. Apply these extensions to the public library case study db instance
from Figure 3.13 and compare their output with the DML script
from Subsection 4.3.5.
572 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

4.14 Design extensions to A2014SQLS1–8 so that it also accepts:

a. the totally not recommended tables without primary keys, and/or
having not autonumber or integer primary keys referencing other
autonumber ones, and/or having primary keys made up of several
columns, and/or having primary keys with other data types than the
integer ones;
b. small static lookup sets whose elements might contain single quotes.
4.15. For automatizing translations of various RDBMS managed rdbs
into corresponding E-R data models:
a. design the algorithms of the family REAF0–2 for DB2 10.5, Oracle
12c, MySQL 5.7, and SQL Server 2014;
b. design such algorithms for at least one current version of another
RDBMS vendor or open source consortium (e.g., PostgreSQL,
MongoDB, etc.).
4.16. Based on the intuition gained from Section 4.7:
a. extend the algorithm REA2013A0 so that it also includes in the
resulting ANSI standard SQL DDL scripts the default column val-
ues and the table indexes (hint: the corresponding syntaxes are
colName data-type … DEFAULT expression … and
INDEX idxName
(colName1, …, colNamen));
b. what would be the corresponding SQL DDL script from Section
4.7 if it were obtained with this extended version of the algorithm?
4.17. Slightly modify algorithm REA2013A0–2 such that for more than
8 tables it draws separate structural E-RDs, each of which having at most
16 object sets (just like in Fig. 4.86).
4.18. Fig. 4.123 shows the instance of the EntryTypes table from the db
Stocks (see Section 4.7):
a. based on it, improve the description of this db;
b. generalize the above and correspondingly improve the algo-
rithms of the REAF0–2 family, by also taking into account rdb
instances;
c. given the instance from Fig. 4.123, would you implement the
EntryTypes object set with a table? If not, explain why and cor-
respondingly correct both the E-R data model and the DDL script
from Subsection 4.7.2.
Relational
 Schemas Implementation and Reverse Engineering 573

FIGURE 4.123. Instance of the EntryTypes table from the db Stocks.

4.19.
a. Would you add to the Customers table of the Stocks db (see Section
4.7) two Boolean columns Customer? and Supplier? Why?
b. Make all necessary changes, both in the ANSI standard SQL DDL
script from Subsection 4.7.2 and in the E-R data model from Sub-
section 4.7.3 in order to add these two columns to the Stocks db.
4.20. Based on the examples provided by Proposition 3.1 and Exercise
4.1, characterize and prove the corresponding properties of the algorithms
REA2013A0, REA2013A0–2, and A2014SQLS1–8, as well as of all of the
other algorithms that you designed from the scratch or extended as asked
by the exercises 4.3.a and b, 4.4 to 4.8, 4.9.b, 4.11, 4.12, 4.13a, 4.14, 4.15,
4.16a, 4.17, and 4.18b.
4.21. Apply the algorithm A8 to the relational schemas from Figs. 3.12
and 3.20, Tables 3.7–3.13, as well as to all those obtained by solving Exer-
cises 3.9, 3.28.a, and 3.76–3.81.
4.22 Apply the algorithm A2014SQLS1–8 to the E-R data models from
subsections 2.9.1 and 2.9.2, 3.7.3.1 and 3.7.3.2, Figs. 3.28–3.30, as well as
to all those obtained by solving the exercises from Chapter 2.
574 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

4.23. Apply the algorithm REA2013A0 to the MS Access Northwind db

(see Exercise 4.2).
4.24. Consider the E-R data model obtained by the Exercise 4.23:
a. is it a good one? Why? (hints: for example, only referring to its
Customers object set: what’s the difference between Custom-
ers and Employees and Shippers and Suppliers? Are all of these
four distinct object sets needed, or could we consolidate them in
only two: PEOPLE and COMPANIES? Are Job Title, City, State/
Province, and Country/Region only strings? Is it ok that all of its
properties, except for the primary key, are not compulsory? Is it
enough to uniquely identify customers only through the surro-
gate key ID? Is this table at least in 3NF? Why? After inspecting
their instances too, do you think that the sets Inventory Transac-
tion Types, Order Detail Status, Orders Status, Orders Tax Status,
and Purchase Order Status should be thought of as object sets?
Why?);
b. correct this E-R data model;
c. apply to this corrected E-R data model the algorithms
A2014SQLS1–8 and A2013A1–8 (obtained by solving Exercise
4.11.a) and run the resulted SQL DDL scripts in MS SQL Server
and Access in order to obtain the corresponding correctly designed
Northwind rdbs.
4.25. Consider the MS SQL Server Northwind demo db (that can be
downloaded from http://www.microsoft.com/en-us/download/details.
aspx?id=23654):
a. obtain its corresponding E-R data model (through reverse engi-
neering);
b. analyze it: is it a good one? Why?
c. correct this E-R data model;
d. apply to this corrected E-R data model the algorithms
A2014SQLS1–8 and A7/8–3, and then run the resulted SQL DDL
script in MS SQL Server in order to obtain the corresponding cor-
rectly designed Northwind rdb.
4.26. Consider the solved problem 4.0 above:
a. add tables PRODUCTS and INVOICE_DETAILS (see Exercise 2.8);
Relational
 Schemas Implementation and Reverse Engineering 575

b. add triggers in order to automatically compute total amounts for

both invoice detail lines and invoices;
*c. enforce the nonrelational constraints through SQL Server 2014
triggers;
**d. generalize the experience you got when solving a. above and
extend the algorithm A2014SQLS1–8 such that it also enforces at
least the types of nonrelational constraints that you encountered in
this exercise.
4.27. Extend the algorithm REA2013A0–2 such that it also takes into
consideration Value List type combo-boxes (hint: for example, in the MS
Access 2013 Northwind demo db, column Category of table Products is
referencing such a value list; consequently, this value list was considered
in Fig. 4.116 as a (value object) set Categories (static and small) and
Category is considered as an arrow in Fig. 4.116, instead of an ellipse
from Fig. 4.111; same thing goes for the Payment Method of Purchase
Orders).
4.28. Extend the algorithm REA2013A0–2 such that for tables having
only one double key (primary or not) whose both columns are foreign
keys to generate corresponding diamonds instead of rectangles (hint: for
example, in the MS Access 2013 Northwind demo db, table Employee
Privileges only has such a double primary key—see Fig. 4.117).
4.29. Extend the algorithm REA2013A0–2 such that for tables having
no primary key the maximum cardinality of the corresponding object set
is 2,147,483,648 (i.e., 2 GB, the maximum size of an Access db) divided
by the average row length in bytes and for tables having a compound pri-
mary key made out of two foreign keys (see Exercise 4.28) it is this size
minus the sum of half of the average maximum sizes of the two referenced
table rows, everything divided by the average size of the referencing table
rows (see, for example, the maximum cardinal of the Employee Privileges
object set in the solution of problem 4.2 above).
4.30. Extend the algorithm REA2013A0–2 such that for tables having
non-numeric primary keys the maximum cardinality of the correspond-
ing object set is computed from the corresponding maximum cardinal
of the primary key domain, limited to 2 GB (the maximum size of an
Access db) divided by the average row length (hint: for example, table
Sales Reports from Exercise 4.2 above has a UNICODE(50) primary
576 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

key; card(UNICODE) = 65536; max(card(UNICODE(50))) = 6553650,

which is immensely larger than 2 GB; max(length(Sales Reports rows))
= 4096 + 154 = 4250 bytes; avg(length(Sales Reports rows)) = 2125
bytes = 0.000001.979060471057891845703125 GB; consequently,
max(card(Sales Reports)) = 2 GB/0.0000019790604710578918457031
25 = 1085102592571150.0950588235294118, which is approximately
1,085,102,592,571,150—that is a little bit more than 1015).
4.31. Apply Algorithm A7/8–3 to all of the tables:
a. of the dbs Stocks (see subsection 4.7.1) and MS Access Northwind
(see Exercise 4.0);
b. obtained when solving Exercises 4.22 and 4.25.d.
4.32. Apply algorithms A7/8–3 and RESQLS2014A0–2 (designed in
Exercise 4.15) to the MS SQL Server Pubs sample db.
4.33. Apply algorithms A7/8–3 and REO12cA0–2 (designed in Exer-
cise 4.15) to the Oracle Express HR sample db.
4.34. Apply algorithms A7/8–3 and REDB2105A0–2 (designed in
Exercise 4.15) to the DB2 Express-C Sample db.
4.35. Apply algorithms A7/8–3 and REMySQL57A0–2 (designed in
Exercise 4.15) to the MySQL Sakila and World sample dbs.
4.36. Apply at least one algorithm of the AF1–8 family to each E-R
data model obtained by solving Exercise 2.10.
*4.37. Extend the RDM metamodel of itself from Subsection 3.7.2
with all tables and columns that it does not have, but you can find in the
visible parts of the metacatalogs of:
a. MS Access 2013 (hint: its system “MSys” prefixed tables);
b. MS SQL Server 2014 Express (hint: its master db tables);
c. Oracle MySQL 5.7 (hint: its information_schema tables).
4.38. Apply algorithm REA1–2 to all extensions added in Exercise 4.37
so that to obtain corresponding extended E-R data models of the RDM
(see subsection 3.7.3).
*4.39. Apply algorithms of the family REAF0–2 to the views provided
by the metacatalogs of:
a. MS SQL Server 2014 Express (hint: its master db system views);
b. Oracle 12c Express (hint: only the most frequently ones indicated by
http://www.techonthenet.com/oracle/sys_tables/ would be enough
for this exercise);
Relational
 Schemas Implementation and Reverse Engineering 577

c. IBM DB2 10.5 Express-C (hint: only the most frequently SYS-
CAT ones indicated by the following web page and its next (3)
one http://www.devx.com/dbzone/Article/29585/0/page/2 would
be enough for this exercise).
**4.40.
a. Reapply algorithm A0 to the descriptions from the E-R data models
obtained in Exercise 4.39, in order to correct these E-R data mod-
els (i.e., conceptually reverse engineer the corresponding views,
in order to discover the metacatalog tables from which they were
computed).
b. Apply corresponding algorithms of the family AF1–8 to the cor-
rected E-R data models obtained at a. above, in order to obtain SQL
DDL scripts for creating the corresponding metacatalog tables, run
them in the corresponding RDBMS versions, and then design for
them the views similar to those in Exercise 4.39 (your starting
point to this journey; hint: trivially, you have to do it in user dbs of
your own, not in the corresponding system ones!).

4.12  PAST AND PRESENT

The past is the shadow of the future.

—Lucian Blaga
Look closely at the present you are constructing. It should look like
the future you are dreaming.
—Alice Walker

All of the best practice rules and algorithms presented in this section, as
well as their characterization propositions, application examples and exer-
cises come from (Mancas, 2001). All these algorithms are also included in
MatBase (see, e.g., Mancas and Mancas (2005)).
Algorithms similar to A8 are included in all RDBMSes that provide
translation of the rdbs they manage into SQL DDL scripts.
Algorithms similar to REA0 are included in all RDBMSes for creating
or modifying rdbs when executing SQL DDL scripts.
Different variants of algorithms from the AF1–8 family are embedded
in the data modeling tools provided by DB2, Oracle, SQL Server, as well
578 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

as by third party solutions (see, for example, the universal Toad one, now
owned by the Dell Corp.: http://www.quest.com/toad-data-modeler/).
There exist other third party tools as well for generating corresponding
SQL DDL scripts from Access dbs: for example, see the DBWScript from
DBWeigher.com (http://dbweigher.com/dbwscript.php).
For obtaining E-RD variants from rdbs there are also tools both RDBMS
embedded (like, for example, the MS Access Relationships and SQL Server
Database Diagrams, the Oracle SQL Developer Data Modeler, and the
IBM DB2 Data Studio or InfoSphere Data Architect) and third party (see,
for example, the CA ERwin Data Modeler, http://erwin.com/products/data-
modeler, the Visual Paradigm, http://www.visual-paradigm.com/features/
database-design/ or the SmartDraw, http://www.smartdraw.com/resources/
tutorials/entity-relationship-diagrams/). Unfortunately, since its 2013 ver-
sion, the MS Visio does not have this facility anymore.
Excellent DB2 10.5, Oracle 12.c, MySQL 5.7, SQL Server 2014, and
Access 2013 documentation is provided by IBM, Oracle, and Microsoft,
respectively (see references in the next subsection).
Even if only in the fourth chapter of the next volume will we tackle
rdb optimizations, some initial aspects of it were also introduced in this
chapter, both from own experience, RDBMS documentations, and from
Bradford (2011), Feuerstein (2008), MacDonald (2013), and Vaughn and
Blackburn (2006).
All URLs mentioned in this section were last accessed on Sep-
tember 30th, 2014.

KEYWORDS

•• (z)Linux
•• @@DBTS
•• @@IDENTITY
•• A1-7
•• A2014SQLS1-8
•• A7/8-3
•• A8
Relational
 Schemas Implementation and Reverse Engineering 579

•• Access
•• addSemanticsAndClose
•• addSQLSColumn
•• ADO
•• ADOX
•• AF1-8
•• AF8’
•• AFTER UPDATE
•• AIX
•• ALL_OBJECTS
•• Allow Zero Length
•• ALTER SESSION
•• ALTER TABLE
•• ANCHOR
•• ANSI
•• ANYDATA
•• ANYDATASET
•• ANYTYPE
•• Array
•• ASCII
•• AS ROW BEGIN
•• AS ROW CHANGE
•• AS ROW END
•• associative array
•• ATTACHMENT
•• AUTO_INCREMENT
•• Azure
•• BasicFile LOB
•• BEFORE INSERT
•• BEFORE UPDATE
•• BFILE
•• BIGINT
•• BINARY
580 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

•• BINARY_DOUBLE
•• BINARY_FLOAT
•• BIT
•• BITEMPORAL
•• BLOB
•• BLU
•• BOOL
•• BOOLEAN
•• built-in data type
•• BUSINESS_TIME
•• BYTE
•• cardinality
•• CASCADE DELETE
•• CASCADE UPDATE
•• CAST
•• CCSID
•• CHAR
•• check constraint
•• chooseSQLSDT
•• chooseSQLSnumericDT
•• chooseSQLStextDT
•• CLOB
•• CODEUNITS16
•• CODEUNITS32
•• COMMENT
•• completeSQLScheme
•• compulsory data
•• computed column
•• CONCAT
•• COUNT
•• COUNTER
•• CREATE DATABASE
•• CREATE FUNCTION
Relational
 Schemas Implementation and Reverse Engineering 581

•• CREATE SEQUENCE
•• createSQLSForeignKey
•• createSQLSTable
•• CREATE TABLE
•• CREATE TRIGGER
•• CREATE TYPE
•• CREATE USER
•• CurDate
•• CURRENCY
•• CURRENT_DATE
•• CURRENT_TIME
•• CURRENT_TIMESTAMP
•• CURRENT_TIMEZONE
•• CURRENT DATE
•• CURRENT LOCALE LC_TIME
•• CURRENT TIME
•• CURRENT TIMESTAMP
•• CURRENT TIMEZONE
•• CurTime
•• DANGLING
•• DAO
•• Data Architect
•• database
•• Database Partitioning Feature (DPF)
•• data event
•• DATALINK
•• Data Links Manager
•• data range
•• Data Studio
•• data type
•• DATE
•• DATE_ADD
•• DATE_SUB
582 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

•• DATETIME2
•• DATETIMEOFFSET
•• DB2
•• DB2SECURITYLABEL
•• DB2SQLSTATE
•• DBA_OBJECTS
•• DBCLOB
•• DBCS
•• DBMS_XMLSCHEMA
•• DBTimeZone
•• DBURIType
•• DBWScript
•• DDL
•• DECFLOAT
•• DECIMAL
•• DELETE RESTRICT
•• delimited identifier
•• DEREF
•• DESCRIBE
•• Design View
•• DETERMINED BY
•• DETERMINISTIC
•• Developer Data Modeler
•• directly usage
•• direct recursive type
•• DISTINCT
•• DML
•• domain constraint
•• DO statement
•• DOUBLE
•• DOUBLEPRECISION
•• DROP
•• dummy date
Relational
 Schemas Implementation and Reverse Engineering 583

•• E-R data model

•• EBCDIC
•• Enforce Referential Integrity
•• ENUM
•• ERWin Data Modeler
•• Excel
•• EXPLAIN
•• Express-C
•• extended data type
•• Extract
•• FIPS Flagger
•• FLOAT
•• FLUSH
•• FOR BIT DATA
•• FOR EACH ROW
•• foreign key
•• functional dependency
•• GENERATE_UNIQUE
•• GEOGRAPHY
•• GEOMETRY
•• geospatial data type
•• GetDate
•• GetUTCDate
•• global temporary table
•• global transaction identifier
•• GRAPHIC
•• grbit
•• GROUP BY
•• GTID
•• GUID
•• Hadoop
•• HEX
•• HIERARCHYID
584 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

•• HP-UX
•• HTTPURIType
•• IBM
•• identifier
•• IDENTITY
•• indexed
•• index file
•• indirectly usage
•• indirect recursive type
•• infinity
•• information_schema
•• InfoSphere Data Replication
•• InfoSphere Warehouse
•• INSERT INTO
•• INT
•• INTEGER
•• INTERVAL DAY
•• INTERVAL YEAR
•• iSeries
•• key
•• LAST_INSERT_ID
•• LIMIT
•• Limit to List
•• Linux
•• LOAD DATA INFILE
•• LOB technology
•• Local_Timestamp
•• local temporary table
•• LOGICAL
•• LONG
•• LONGBLOB
•• LONG RAW
•• LONGTEXT
Relational
 Schemas Implementation and Reverse Engineering 585

•• Lookup
•• Lookup Wizard
•• Management Studio
•• MatBase
•• materialized view
•• MAX_STRING_SIZE
•• MAXDB
•• MBCS
•• MEDIUMBLOB
•• MEDIUMINT
•• MEDIUMTEXT
•• Microsoft
•• minimal uniqueness
•• Mixed Based Replication (MBR)
•• MIXED DATA
•• MIXED DECP
•• MONEY
•• MongoDB
•• MSysRelationships
•• MULTILINESTRING
•• MULTIPLE BYTE
•• MULTIPOINT
•• MULTIPOLYGON
•• MyISAM
•• MySQL
•• NaN
•• NCHAR
•• NCLOB
•• NDBCLUSTER
•• NESTED TABLE
•• NEWID
•• Northwind
•• NOT NULL
586 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

•• not unique index

•• Now
•• number
•• NUMBER
•• number sequence
•• NVARCHAR
•• NVARCHAR2
•• OBJECT
•• ODBC
•• Office
•• OID
•• OLEDB
•• ON COMMIT PRESERVE ROWS
•• ON DELETE
•• ON UPDATE
•• OpenSolaris
•• OpenVMS
•• Oracle
•• Oracle Express
•• Oracle Golden Gate
•• ORDER BY
•• ordinary array
•• ordinary identifier
•• OS X
•• PARAMETERS
•• PIVOT
•• PL/SQL
•• POINT
•• POLYGON
•• PostgreSQL
•• primaryKey
•• primary key
•• pureScale
Relational
 Schemas Implementation and Reverse Engineering 587

•• QBE
•• QMF
•• Query Design
•• Query Management Facility
•• quiet NaN
•• QUOTED_IDENTIFIER
•• R-I-C-0
•• R-I-C-1
•• R-I-C-2
•• R-I-C-3
•• R-I-C-4
•• R-I-C-5
•• R-I-D-0
•• R-I-D-1
•• R-I-D-2
•• R-I-P-0
•• R-I-P-1
•• R-I-P-2
•• R-I-P-3
•• R-I-P-4
•• R-I-P-5
•• R-I-T-1
•• R-I-T-2
•• R-I-T-3
•• R-I-T-4
•• RAW
•• REA0
•• REA1
•• REA2
•• REA2013A0
•• REA2013A0-2
•• REAF0-1
•• REAF0-2
588 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

•• REAF0’
•• REAL
•• recursive type
•• REF
•• REFERENCE
•• REFERENCES
•• referential integrity
•• REGEXP
•• relational scheme
•• REPLACE
•• Required
•• RESET
•• reverse engineering
•• root table
•• root type
•• Row Based Replication (RBR)
•• ROW BEGIN
•• ROW CHANGE TIMESTAMP
•• ROW END
•• ROWGUIDCOL
•• ROWID
•• ROWVERSION
•• Runtime
•• Sakila
•• SBCS
•• SDO_GEOMETRY
•• SDO_GEORASTER
•• SDO_TOPO_GEOMETRY
•• SecureFile LOB
•• SELECT
•• Sequence
•• SERIAL
•• Server Pubs
Relational
 Schemas Implementation and Reverse Engineering 589

•• SessionTimeZone
•• SET
•• SharePoint
•• SHORT
•• SHORT TEXT
•• SHOW
•• Show System Objects
•• SI_AverageColor
•• SI_Color
•• SI_ColorHistogram
•• SI_FeatureList
•• SI_PositionalColor
•• SI_StillImage
•• SI_Texture
•• signaling NaN
•• SINGLE
•• SMALLDATETIME
•• SMALLINT
•• SMALLMONEY
•• Smart Draw
•• Solaris
•• Spatial and Graph
•• Spatial Extender
•• SQL
•• SQL/MM StillImage
•• SQL_VARIANT
•• SQL Server
•• Statement Based Replication (SBR)
•• STDDT
•• STRAIGHT_JOIN
•• strongly typed data type
•• STRUCTURED
•• subnormal number
590 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

•• subtable
•• SUM
•• Supertable
•• supertype
•• SYSCAT
•• SysDate
•• SysDateTime
•• SysDateTimeOffset
•• SYSDUMMY
•• SYSFUN
•• SYSIBM
•• SYSPROC
•• SYSTEM_TIME
•• system views
•• SysTimestamp
•• SysUTCDateTime
•• szColumn
•• szObject
•• szReferencedColumn
•• szReferencedObject
•• szRelationship
•• table
•• tablespace
•• target type
•• temporary table
•• TEXT
•• Time
•• TIMESTAMP
•• TIMESTAMP WITH LOCAL TIME ZONE
•• TIMESTAMP WITH TIME ZONE
•• TINYBLOB
•• TINYINT
•• TINYTEXT
Relational
 Schemas Implementation and Reverse Engineering 591

•• TOP
•• TRANSACTION START ID
•• TRANSFORM
•• TRANSLATE
•• Trigger
•• TRIM
•• tuple constraint
•• typed table
•• typed view
•• type hierarchy
•• UDT
•• UNICODE
•• UNION ALL
•• UNIQUEIDENTIFIER
•• unique index
•• uniqueness
•• Universal Naming Convention (UNC)
•• Unix
•• UNPIVOT
•• UNSIGNED
•• UPDATE OF
•• UPDATE RESTRICT
•• URIFactory
•• URIType
•• UROWID
•• USER_OBJECTS
•• user defined data type
•• UTF-16
•• UTF-8
•• VALIDATED
•• VALIDATING
•• Validation Rule
•• VARBINARY
592 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

•• VARCHAR
•• VARCHAR2
•• VARGRAPHIC
•• VARRAYS
•• VBA
•• virtual column
•• Visio
•• Visual Paradigm
•• VM/VSE
•• weakly typed data type
•• Windows
•• WTDDT
•• XML
•• XMLEXISTS
•• XMLQUERY
•• XMLType
•• XPath
•• XQuery
•• Year
•• Yes (No Duplicates)
•• YESNO
•• zOS

REFERENCES

So all of the music has reference, or is inspired by something of the dharma117

that I’ve come in contact with.
—Joseph Jarman
Bradford, R. (2011). Effective MySQL: Optimizing SQL Statements. McGraw-Hill Osborne
Media: New York, NY.
Feuerstein, S. (2008). Oracle PL/SQL Best Practices. 2nd edition. O’Reilly Media: Sebas-
topol, CA.
117
Dharma, in Sanskrit, means the “Law that upholds, supports, or maintains the regulatory order of
the universe”.
Relational
 Schemas Implementation and Reverse Engineering 593

IBM Corp. (2014). DB2 database product documentation. (http://www01.ibm.com/sup-

port/docview.wss?uid=swg2700.9474).
IBM Corp. (2014). DB2 for Linux, UNIX, and Windows Best Practices. (https://www.
ibm.com/developerworks/community/wikis/home?lang=en#!/wiki/Wc9a068d-
7f6a6_4434_aece_0d297ea80ab1/index?sort=updated&tag=best_practices).
MacDonald, M. (2013). Access 2013: The Missing Manual. O’Reilly Media: Sebastopol, CA.
Mancas, C. (2001). Databases for Undergraduates Lecture Notes: Computer Science
Dept., Ovidius State University, Constanta, Romania.
Mancas, C., Mancas, S. (2005). MatBase E-R Diagrams Subsystem Metacatalog Concep-
tual Design. In Proc. IASTED DBA 2005. Conf. on DB and App., 83–89, Acta Press,
Innsbruck, Austria.
Microsoft Corp. (2013). Access 2013. (http://msdn.microsoft.com/en-us/library/office/
fp179695(v=office.15). aspx).
Microsoft Corp. (2014). SQL Server Database Engine (http://msdn.microsoft.com/en-us/
library/ms1878.75.aspx).
Oracle Corp. (2014). MySQL 5.7 Reference Manual. (http://dev.mysql.com/doc/refman/5.7/en/).
Oracle Corp. (2014). Oracle Database Documentation Library 12c Release 1. (http://docs.
oracle.com/database/121/index.htm).
Vaughn, W. R., Blackburn, P. (2006). Hitchhiker’s Guide to Visual Studio and SQL Server:
Best Practice Architectures and Examples, 7th edition. Addison-Wesley: Boston, MA.
This page intentionally left blank
CHAPTER 5

CONCLUSION

The possession of knowledge does not kill the sense of wonder and mystery.
There is always more mystery.
—Anaïs Nin

CONTENTS

5.1  Database Axioms.......................................................................... 595

5.2 Why Do We Need Another Conceptual Level for Expert DB
Design?......................................................................................... 604
5.3 What Are the Most Important Things That We Should Be
Aware of in DBs?.......................................................................... 609
Keywords................................................................................................611

5.1  DATABASE AXIOMS

God does not care about our mathematical difficulties;

He integrates empirically.
—Albert Einstein

Generally, in order to be able to compare several solutions to a same prob-

lem, but also to judge a solution for correctness, elegance, optimality, etc.,
the existence of a widely accepted set of guiding axioms is a must.
Unfortunately, for database design, implementation, manipulation and
optimization there is no such established set of axioms. This section pres-
ents my corresponding set proposal.
596 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

5.1.1  DESIGN AXIOMS

We cannot solve our problems with the same thinking we used when we
created them.
—Albert Einstein

The relational db scheme design problem is, essentially, two folded: what
table schemes should be included in a db scheme? what columns and con-
straints should contain every of its table schemes?
The following db design axioms provide criteria for comparing and
hierarchizing alternative design solutions:
A-DA0. Non-relationally rdb design axiom: rdb schemes should not
be designed relationally; instead, they should first be designed by using
E-RDs and associated restriction sets; then, conceptual design should be
refined by using a more powerful data model, able to heavily assist us in
discovering and designing all existing constraints, as well as in correct-
ing and refining all errors in E-RDs; finally, resulted conceptual schemes
should be (preferably, automatically) translated into corresponding rela-
tional ones, plus associated non-relational constraint sets. As an emer-
gency temporary shortcut, you could skip the intermediate data model
level, but never the E-RDM one.
A-DA01. Data plausibility axiom: any db instance should always store
only plausible data; implausible (“garbage”) data might be stored only
temporarily, during updating transactions (i.e., any time before start and
after end of such a transaction all data should be plausible).
A-DA02. No semantic overloading axiom: no fundamental object set
or property should be semantically overloaded (i.e., any such object set
or property should have only one simple and clear associated semantics).
A-DA03. Non-redundancy axiom: in any db, any fundamental data
should be stored only once, such that inserting new data, as well as updat-
ing or deleting existing one should be done in/from only one table row.
A-DA04. Unique objects axiom: just like, generally, for set elements,
object sets do not allow for duplicates (i.e., each object for which data is
stored in a db should always be uniquely identifiable through its corre-
sponding data).118
118
Consequently, no db fundamental table should ever contain duplicate rows (the so-called table syn-
tactic horizontal minimality). Moreover, for such tables, except for some very few exceptions (namely,
tables corresponding to subsets or to object sets of the type poultry/rabbit/etc. cages), their rows should
Conclusion
 597

A-DA05. Best possible performance axiom: db design, implementation

and optimization should guarantee obtaining the maximum possible per-
formance (i.e., the overall best possible execution speed for critical que-
ries and updates, as well as the best possible average execution speed for
the noncritical ones).
A-DA06. Controlled-redundancy axiom: only for satisfying the best
possible performance axiom above, dbs might store information too (i.e.,
computed, thus redundant data), provided that redundancy is strictly con-
trolled: users should at most read information (i.e., never write or delete
it) and information should be automatically recomputed immediately after
corresponding fundamental data has changed.
A-DA07. No trivial constraints axiom: no trivial constraint (e.g., any
set includes both the empty set and itself, any set is equal to itself, etc.)
should ever be added to a db scheme.
A-DA08. No constraints, but keys on redundant data axiom: no con-
straint should be enforced on redundant data, except for keys: as redundant
data is automatically computed, it should always be valid, so adding con-
straints on it would only slow down updates speed. The exception for keys
is justified as, sometimes, redundant data is needed in order to correctly
model all existing keys.119
A-DA09. Constraints discovery axiom: for any nontrivial and noncon-
tradictory and not implied (i.e., computable) restriction existing in the
modeled subuniverse, any db should enforce a corresponding constraint.
A-DA10. No implausible constraints axiom: no db should enforce a
constraint that does not correspond to a restriction existing in the modeled
subuniverse.
A-DA11. Constraint sets optimality axiom: for any db scheme, no
implied constraint should be enforced and the constraint set should be
enforceable in the minimum possible time (i.e., there should not exist
another equivalent constraint set whose enforcement takes less time in the
given context).
A-DA12. Constraint sets coherence axiom: for any db scheme, the set
of its constraints should be coherent (i.e., it should not be contradictory,

not differ between them only in their surrogate key column, but in other columns as well (the so-called
table semantic horizontal minimality).
119
see, for example, keys *Country • Zip from Fig. 3.5 (Subsection 3.2.3) and ConstrName • *DB from
Table 3.11 (Subsection 3.7.2)
598 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

thus only allowing for some object sets or properties the empty set as their
instances).
A-DA13. Minimum inclusions axiom: in any db model, for any inclu-
sion between object sets T # S, S should be the smallest set including T.
A-DA14. Inclusions acyclicity axiom: no fundamental object set inclu-
sions chain should be a cycle (dually, for any db scheme, its fundamental
inclusions graph should be acyclic).
A-DA15. Sets equality axiom: no equality of two fundamental object
sets should ever be added to a conceptual db scheme (i.e., such equali-
ties, if any, are always to be interpreted simply as declaring object set
aliases).120
A-DA16. Mappings equality axiom: no equality of two fundamental
mappings should ever be added to a db scheme (i.e., such equalities, if
any, are always to be interpreted simply as declaring mappings aliases).121
A-DA17. ER-Ds axiom: except for an overview one, no ER-D should
have more than one letter/A4 page and its nodes and edges should not
intersect between them, except for tangent points of edges to their con-
nected nodes.
A-DA18. ER-D cycles axiom: all ER-D cycles should be thoroughly
analyzed in order to discover whether or not they should be broken or they
need declaring additional constraints in the corresponding db scheme.
A-DA19. Naming axiom: all dbs and their components (object sets,
mappings, constraints, tables, columns, queries, stored procedures,
indexes, etc.) should be consistently named (i.e., strictly adhering to some
naming conventions), always embedding as much unambiguous semantics
as possible in each name. Names given during data analysis should be kept
identical on all design levels and down to corresponding db implementa-
tions. Names for elements of any such sets should be unique within their
corresponding sets.
A-DA20. Documentation axiom: Document all of your software (con-
sequently, data analysis, db design, implementation, manipulation and
120
Dually, rdb implementations may also contain, for optimality reasons, several tables corresponding
to a same object set obtained by vertical and/or horizontal splitting, provided that all of them are cor-
rectly linked and processed: instances of those obtained horizontally should always remain pairwise
disjoint, whereas any of the n–1 schemas (n > 1, natural) obtained vertically should be implemented
as corresponding to subsets of the remaining (master) one.
121
That is no table should ever contain two columns whose values should always be equal (the so-
called table vertical minimality).
Conclusion
 599

optimization too) as if tomorrow you will be struck by total amnesia, but

do not overdocument.
Obviously, axioms A-DA02, A-DA05, A-DA09, A-DA10, A-DA17, and
A-DA19 cannot be guaranteed by any data model, but only by us, as their
users.

5.1.2  IMPLEMENTATION AXIOMS

Beware of the problem of testing too many hypotheses; the more you torture
the data, the more likely they are to confess, but confessions obtained
under duress may not be admissible in the court of scientific opinion.
—Stephen M. Stigler

A-DI0. Instances consistence axiom: any possible rdb instance should be

consistent (not only with respect to relational-type constraints, but to the
non-relational ones too).122
A-DI01. Tables’ semantics axiom: in any rdb, each fundamental table
should correspond to an object set (be it actual or virtual) in the modeled
subuniverse of discourse.
A-DI02. Non-static fundamental objects representation axiom: for each
nonstatic fundamental object set of the modeled subuniverse there should
correspond a fundamental table in the modeling rdb (dually, only static
fundamental object sets, like, for example, rainbow colors, sex types, etc.,
could be represented not by tables, but, for example, by enumerated sets).
A-DI03. Tables indivisibility axiom: any fundamental object set should
always be modeled with only one table; these tables may be horizontally
and/or vertically splitted in actual implementations only when needed, but
under strict db applications control (see footnote of axiom A-DA15 above).
A-DI04. Columns’ semantics axiom: in any table, each fundamental
column should correspond to a property of the objects in the object set
corresponding to that table.
A-DI05. Empty instances axiom: any fundamental table should accept
both empty and nonempty instances.

122
That is all existing constraints (in the corresponding subuniverse of discourse), be them relational
or not, should be enforced in the db.
600 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

A-DI06. Constraints enforcement axiom: for each fundamental table,

any of its constraints whose type is provided by the target RDBMS has
to be enforced through that RDBMS; all other db constraints should be
enforced by all software applications built on top that rdb.123
A-DI07. Domain constraints axiom: any fundamental column should
have an associated domain constraint restricting its range to a plausible
finite subset of the corresponding data type.
A-DI08. Rows uniqueness axiom: any fundamental table row should
be always uniquely identifiable according to all existing uniqueness con-
straints in the corresponding subuniverse and it should correspond to a
unique object from the corresponding objet set.
A-DI09. No void rows axiom: for any fundamental table, at least one
semantic fundamental column should not accept null values.
A-DI10. Referential integrity axiom: any foreign key value should
always reference an existing value (dually, no rdb should ever contain
dangling pointers).
A-DI11. Primary keys axiom: any fundamental table should have a sur-
rogate integer primary key whose range cardinality should be equal to the
maximum possible cardinality of the corresponding modeled object set;
except for tables corresponding to subsets, primary keys should store auto
generated values.
A-DI12. Foreign keys axiom: any foreign key should be simple (i.e.,
not concatenated), reference a surrogate integer primary key and have as
range exactly the range of the referenced primary key.
A-DI13. Key Propagation Principle axiom: any mapping (i.e., many-
to-one or one-to-many relationship) between two fundamental object sets
should be implemented according to the Key Propagation Principle: a for-
eign key referencing its codomain (i.e., the “one” side) is added to the
table corresponding to its domain (i.e., the “many” side).
A-DI14. Set inclusion relational implementation axiom: any inclusion
between two object sets T # S should be represented by a dedicated for-
eign key of T referencing the primary key of S; one of them may be the
primary key of T.
123
That is never leave relational constraints enforcement to db applications: otherwise, first of all,
developers might forget to enforce them in some particular cases (if not in all of them…) and, sec-
ondly, db instances might violate even those enforced by the applications, whenever they are updated
directly, not through application control.
Conclusion
 601

A-DI15. No superkeys axiom: no superkey should ever be enforced;

generally, no other constraints than those declared in the corresponding
conceptual db scheme should ever be enforced.124
A-DI16. No modification anomalies axiom: tables of any rdb should be
completely anomaly free.125
A-DI17. Minimal data storing axiom: no data should be stored more
than once in any db.
Obviously, axiom A-DI06 cannot be guaranteed by any data model or
DBMS, but only by us, as their users.

5.1.3  INSTANCES MANIPULATION AXIOMS

The competent programmer is fully aware of the limited size of his own skull.
He therefore approaches his task with full humility,
and avoids clever tricks like the plague.
—Edsger W. Dijkstra

A-DM0. Positive data axiom: all data stored in conventional dbs is positive.
A-DM01. Closed world axiom: facts that are not known to be true (i.e.,
for which there is no stored data) are implicitly considered as being false.
A-DM02. Closed domains axiom: there are no other objects than those
for which there is stored data.
A-DM03. Null values set axiom: there is a distinguished countable set
of null values.
A-DM04. Null values axiom: Null values may model either (tempo-
rary) unknown or inapplicable or no information data.
A-DM05. Available columns axiom: only the columns of the tables,
views and subqueries of the FROM clause are available for all other clauses
of the corresponding SQL statements.
A-DM06. GROUP BY (golden rule) axiom: in the presence of a GROUP
BY clause, corresponding SELECT clause can freely contain only column
expressions also present in the GROUP BY one; all other available columns
may be used in SELECT only as arguments of aggregation functions.
124
Obviously, this is a particularization of axiom A-DA11 above.
125
That is not only anomaly-free with respect to the domain and key constraints, but with respect to all
other constraints, regardless of their type.
602 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

A-DM07. Relevant data and processing axiom: any query should con-
sider and minimally process, in each of its steps, only relevant data; for
example,
ü the SELECT and FROM clauses should contain only needed col-
umns and tables, respectively;
ü use the GROUP BY clause only when needed (e.g., computing
groups always containing only one row is senseless);
ü in the presence of a GROUP BY clause, all logical conditions that
may be evaluated before grouping (i.e., not containing aggregate
functions) should be placed in the corresponding WHERE clause
and only those that contain aggregated columns should be placed
in the corresponding HAVING one;
ü ordering should be done only on needed columns (expressions);
ü joins, groupings, and as much as possible of the needed logical
conditions should be performed on numeric columns (surrogate
keys and foreign keys referencing them): only in the last step of
a query should surrogate keys and pointers (i.e., foreign keys) be
replaced by corresponding desired semantic columns.
A-DM08. ORDER BY axiom: except for well-founded exceptions, as
part of a minimal courtesy towards our customers, query results should
be presented ordered by the most interesting possible order (even if not
explicitly asked for); dually, ordering should be done only once per query,
as its last step.
A-DM09. Data importance axiom: as data is one of the most important
assets of our customers, preserving its desired values, as well as consis-
tence of the corresponding db instances is a must (e.g., do not ever insert,
update or delete data that should not be inserted/updated/deleted).
A-DM10. Query semantics correctness axiom: any syntactically cor-
rect query computes a table, but neither its instance, nor even its header is
guaranteed to be the result expected by our customers (e.g., some of our
queries are, in fact, equivalent possible definitions of the empty set).126
A-DM11. Fastest possible manipulation axiom: at least in production
environments, data manipulations should be done with best possible
126
In order to guarantee their semantic correctness (the only one relevant to our customers), if you
cannot formally prove it, at least informally read each query meaning immediately after you consider
it finalized and compare reading with corresponding initial request: if they do not match exactly, then
reconsider query design and/or development.
Conclusion
 603

algorithms and technologies, so that processing speed be the fastest

possible.
A-DM12. Regularly backup fundamental data intelligently axiom: fun-
damental data has to be regularly backed-up, by intelligently using all cor-
responding RDBMS’s facilities (incremental, hot, etc.).
A-DM13. Always process sets (not elements) axiom: in data manipula-
tion, do not think like in imperative (procedural, algebraic) programming,
in terms of elements of sets, but in terms of sets of elements.127
Obviously, axioms A-DM07, A-DM08, A-DM09, A-DM10, and
A-DM13, cannot be guaranteed by any data model or DBMS, but only by
us, as their users.

5.1.4  OPTIMIZATION AXIOMS

We must develop knowledge optimization initiatives

to leverage our key learnings.
—Scott Adams

Even if no intuitive support may be given for the time being for opti-
mization, which heavily depends on current corresponding rdb instances,
workload, and data access paths (see second volume of this book), for hav-
ing all axioms grouped together, optimization ones are already presented
in this section too.128
Optimization also heavily depends on the used DBMS versions; as
such, first are presented here only the very general ones.
Finally, specific optimization axioms are presented in the second vol-
ume for the five RDBMS versions considered as examples in this book.
A-DO0. Reduce data contention axiom: intelligently use hard disks,
big files, multiple tablespaces, partitions, and segments with adequate
block sizes, separate user and system data, avoid constant updates of a
same row, etc. in order to always reduce data contention to the minimum
possible.

127
Data is not like trees (that you have to cut one by one), but rather like grains (that you may crop one
by one, but then all of us would eventually starve).
128
Consequently, beginner readers might skip it at a first sequential read of this volume and return to
it after reading and understanding the second volume.
604 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

A-DO1. Minimize db size axiom: regularly shrink tablespaces, tables,

and indexes for maintaining high processing speeds and minimal db and
backup sizes.
A-DO2. Maximize use of RDBMS statistics axiom: regularly gather and
intelligently use statistics provided by RDBMSs for db fine-tuning.
A-DO3. Follow RDBMS advisors recommendations axiom: regularly
monitor and apply all recommendations of RDBMSs’ advisors.
A-DO4. Commit as infrequently as is practical axiom: for performance
reasons, RDBMSs are generally writing asynchronously to the db data
files, but committing requires synchronous writes, because they must be
guaranteed at their occurring moments of time.129
Obviously, axioms A-DO0, A-DO3 and A-DO4 cannot be guaranteed
by any data model or DBMS, but only by us, as their users.

5.2  WHY DO WE NEED ANOTHER CONCEPTUAL LEVEL FOR

EXPERT DB DESIGN?

Whenever you set out to do something, something else must be done first.
—(Corollary 6 to the first Murphy’s Law:
If anything can go wrong, it will.)

Essentially, this first volume advocates using at least three levels from Fig.
1.1 (namely the E-RDM, RDM, and RDBMS ones) and five correspond-
ing FE algorithms (A0, A1–7, A7/8–3, A8, and one of the AF8′), whenever
having to create a new rdb or to extend an existing one. Dually, when hav-
ing to understand a not enough documented rdb, it advocates using same
three levels and three corresponding RE algorithms (namely, one of the
REAF0’, REA0, and REA1–2).
Obviously, there is an even shorter possible path presented in this book
too, namely using one of the composed algorithms from the AF1–8 fam-
ily, instead of the composition of one of the AF8 and A1–7. This should
be considered only by exception, when under time pressure: as we have
already seen, skipping the RDM level means losing portability (sometimes
even between different versions of a same RDBMS).
129
The need for synchronous write is one of the reasons why the online redo log can become a bottle-
neck for applications.
Conclusion
 605

Anyhow, even in such cases, do not forget to also apply A7/8–3 in

the end, for all new and/or modified fundamental tables. Please note that,
although it is presented in the RDM framework, you can straightforwardly
apply it to any RDBMS version instead.
Dually, it is generally preferable not only under time pressure to
directly use the needed member of the REAF0–2 shortcut family, instead
of composing REA1–2 and the needed one of the REAF0 family: when
working with a legacy not documented rdb, it is very rarely that you would
also come across another equivalent rdb implemented in another RDBMS
version.

5.2.1  IS E-RDM NEEDED? IS IT ENOUGH?

Science does not know its debt to imagination.

—Ralph Waldo Emerson

Of course that, at “co-volume”, one might reduce this minimum advisable

forward path to only two algorithms: the needed one from a family, let’s
say, AF0–8, obtained by composing the corresponding one from AF1–8
and A0, and A7/8–3—thus eliminating the E-RDM. Personally, I wouldn’t
ever recommend such an extreme shortcut, at least from the following
three fundamental reasons:
• customers are very rarely experts in mastering the considered
RDBMS version and even too rarely in correctly and fully bridg-
ing between their subuniverse of interest and that RDBMS version
(think that, generally, among your decision maker customers are
also CFOs, CEOs, stakeholders, etc., not only CTOs): skipping the
E-RDM level results, from this point of view, in replacing its busi-
ness-oriented jargon with, let’s say, an Oracle 12c one, whenever
you communicate with your customers, which is obviously not an
advantage, but rather a major disadvantage;
• very few RDBMS versions have satisfactory graphical capabilities
for showing (foreign key) links between the tables of a rdb (e.g.,
MS SQL Server is showing only one line between any two tables,
regardless of how many links there are between them, and even
when there is only one such link, you cannot determine which are
606 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

the corresponding connected columns; not even MS Access, which

is far more capable from this point of view, showing any link apart,
exactly between the corresponding columns, is not satisfactory: for
one-to-one relationships you can only guess which is the foreign
key and which the corresponding referenced column, for linked
tables belonging to other dbs than the current one the type of the
relationship is not showed, and for each link between two tables
or same table instances they use a different table instance for the
referenced table, which is generally confusing very many begin-
ners, at least): skipping the E-RDM level results, from this point of
view, in losing most of the very powerful graphical advantages of
the E-RDs;
• even when very experienced, only using a technological level inev-
itably distorts your thinking habits, which tends to favor technol-
ogy over conceptual data analysis and db architecturing; sooner
or later, you will start neglecting, for example, also considering
the business rules that are formalizable only as non-relational con-
straints, as well as the informal descriptions of the corresponding
subuniverses of discourse: skipping the E-RDM level results, from
this point of view, in losing both the accuracy of your data models
and their informal descriptions.
Of course that, regarding this latter above reason, one might argue that
informal descriptions may be also obtained directly from the correspond-
ing rdbs and that, anyhow, at this level, we do not care at all about non-
relational constraint enforcement (so what’s the use of discovering and
including such constraints in our models?) and not even about making sure
that there are no other ones (apart those already discovered) that govern
the corresponding subuniverses as well.
To the first objection, regarding the informal descriptions, I agree, pro-
vided it is actually done every time and also including all of the non-rela-
tional constraints that govern the corresponding subuniverse.
To the second one, regarding the usefulness of the non-relational con-
straints, I would stress again that any business rule governing a subuniverse
of interest to our data models and corresponding dbs should be discov-
ered, included in the model, and enforced: otherwise, data plausibility
is compromised. Do, for example, civil engineering architects or car
Conclusion
 607

designers ignore any constraint in their fields, hoping that the workers who
implement their models will discover and enforce them during manufac-
turing or, even worse, that not the workers, but the users of those products
will do it?
Moreover, generally, not only me would always stress that, for its sake,
as well as for our’s (workers in this field or/and users of its artifacts), the
IT industry should rapidly mature at least at the level of, let’s say, the car
manufacturing’s one (as building has already more than 5,000 years and is
very, very hard to narrow such gap rapidly…): would you buy or even only
drive or use a car (driven by somebody else) that was artisanally manufac-
tured, without proper design, without taking into consideration all physics
and safety rules that govern this subuniverse?
To conclude the above considerations, I am only one of the very many
who are convinced that we should not skip the E-RDM, but, on the con-
trary, we should always make full and intelligent use of it in the first step
of the conceptual data analysis and modeling process.
Dually, unfortunately, even extended with restriction sets, E-RDM is
not enough for data modeling, at least from the following four points of
views:
• E-RDM is totally ignoring the following db design axioms (see
Subsection 5.1.1): A-DA03, A-DA05, A-DA07, A-DA08, A-DA11
to A-DA16, A-DA18, and A-DA20.
• E-RDM is only partially taking into account the following db
design axioms (see Subsection 5.1.1): A-DA01 (as discovery of all
existing business rules in any subuniverse of discourse is not even
an aim of restriction sets), A-DA04 (as for restriction sets declar-
ing only one uniqueness per object set is enough), and A-DA20 (as
only informal descriptions of the subuniverses of discourse to be
modeled are compulsory).
• E-RDM has no means to check at least the syntactical correctness
of E-R data models (e.g., are all nonfunctional relationship-type
object sets and all functional relationships and all attributes cor-
rectly defined mathematically?).
• Moreover, there are very few commercial (and even prototype)
DBMSs that provide E-R user interfaces which make completely
transparent to users the RDM (or any other) level of db scheme
608 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

implementation, and even much more fewer the ones that also pro-
vide E-R query languages at least as powerful as SQL.
All these four considerations imply that E-RDM alone is not enough
for accurate, expert conceptual data analysis, db design, implementation,
and usage.
To conclude this subsection, the answers to the questions in its title are:
“definitely yes” and “definitely no”, respectively.

5.2.2  IS RDM NEEDED? IS IT ENOUGH?

What is now proved was only once imagined.

—William Blake

Due to the fact that RDBMSs are still the norm in this field and that they
are so close to RDM (and, hence, to one another), there is a great tempta-
tion to skip RDM completely, in the processes of both db design and usage.
However, as already discussed in details in Section 3.13, the answers to
the questions in the title of this subsection are: “yes” (at least for portabil-
ity and RLs reasons) and “definitely no”, respectively.
I would only add here to the pros that it is always much better not to
try to solve any problem from the beginning in technological terms, but
in higher, more abstract ones. In particular, in the RDM case, despite
the great number of fundamental compatibilities between nearly all
RDBMS versions, there are, however, enough technological differences
too between them, sometimes even only between different versions of a
same RDBMS.
Dually, for the cons, I would also add here the following two other
reasons:
• RDM is totally ignoring the following db design axioms (see Sec-
tion 5.1): A-DA07, A-DA08, A-DA11 to A-DA16, A-DA18, A-DI01
to A-DI05, A-DI08, A-DI09, and A-DI12 to A-DI17.
• RDM is only partially taking into account the following db design
axioms (see Section 5.1): A-DA01, A-DI0 (as it does not take into
account non-relational constraints), and A-DA04 (as users are not
coerced to declare at least one key for any fundamental table).
Conclusion
 609

5.2.3  WHAT SHOULD AN INTERMEDIATE MODEL BETWEEN

E-RDM AND RDM BRING?

The important thing in science is not so much to obtain new facts

as to discover new ways of thinking about them.
—Sir William Bragg

Not only from my point of view, first of all, such a data model should fully
take into account all db axioms from Section 5.1: even for those that it
cannot automatically guarantee, it should at least provide a framework and
tools for its users to help them comply with.
Next, it should also incorporate:
• an algorithm of type A3, for the (at least syntactical) validation of
the initial E-R data models;
• one of type A4, for assisting E-RD cycles analysis (because, as
shown in the second chapter of the second volume of this book, they
are very frequently embedding lot of non-relational constraints);
• one of type A5, for assisting the not always obvious task of guaran-
teeing the coherence of the constraint sets;
• one of type A6, for assisting the much more difficult task of guar-
anteeing the minimality of the constraint sets.
Finally, above all, in order to guarantee data plausibility, such a model
should also incorporate all needed constraint types, not only the relational
ones: ideally, it should provide at least the subclass of closed Horn clauses130.
There are hundreds of higher (than RDM and E-RDM) level concep-
tual data models, but, to my knowledge, none of them incorporates all the
above as well, except the (E)MDM that is presented and used in the second
volume of this book.

5.3  WHAT ARE THE MOST IMPORTANT THINGS THAT WE

SHOULD BE AWARE OF IN DBS?

The most important of my discoveries have been suggested to me by my

failures.
—Humphrey Davy

130
Recall that the Horn clauses are the biggest class of first order predicate formulas for which the
implication problem is decidable.
610 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

As we have already seen, no data models and/or DBMS might take into
account all above db axioms and best practice rules. Consequently, I
would like to take this final (in the realm of this volume) opportunity to
summarize once more the main ideas behind the crucial ones in the fol-
lowing dodecalog:
1. Data is increasingly becoming the most important asset of custom-
ers: we always must commit to keep it plausible, safe, and avail-
able at the maximum possible speed.
2. Only humans can understand semantics and break it down into
atomic ones: algorithms and/or machinery can only assist us with it.
3. It is not only in our customers’, but also in our own interest to
never semantically overload either object sets or mappings.
4. Discovering all existing business rules that govern any subuniverse
of interest, adding them to our data models, and enforcing them in
any corresponding db is crucial in guaranteeing data plausibility:
missing anyone of them allows for storing implausible data.
5. Adding any implausible constraint to a data model should be
always banned, as it would not allow for storing plausible data in
the corresponding dbs.
6. For any db, its associated constraint set should be coherent and
minimal.
7. Without proper data analysis and conceptual db design, resulting
dbs have very few chances to satisfy customers.
8. Just like, for example, in the case of aircrafts, following prescrip-
tions of algorithms and best practice rules is the safest thing to do
in the db field too.
9. Redundant data should always be strictly controlled: correspond-
ing values should always be read-only for users and should be
immediately calculated automatically when needed.
10. Object elements represented in dbs only by null values or/and
which are not uniquely identifiable semantically (i.e., not only by
surrogate keys) within their object sets are useless, confusing, tak-
ing up storage space and slowing down query execution for noth-
ing: consequently, they should be banned from any db instance.
11. When executed, a query always returns the correct answer to the cor-
responding customer question if and only if it was also semantically
Conclusion
 611

(not only syntactically) correctly designed and coded (dually: only

syntactically correct queries are always returning correct answers to
a question, but not necessarily to the desired one).
12. When manipulating data, thinking in terms of sets of elements is
far better than thinking in terms of elements of sets, as it results in
much greater query execution speeds (just like in the case of paral-
lel versus sequential programming) and much higher SQL (be it
pure, extended, or embedded) developing productivity.

KEYWORDS

•• A-DA0
•• A-DA1
•• A-DA2
•• A-DA3
•• A-DA4
•• A-DA5
•• A-DA6
•• A-DA7
•• A-DA8
•• A-DA9
•• A-DA10
•• A-DA11
•• A-DA12
•• A-DA13
•• A-DA14
•• A-DA15
•• A-DA16
•• A-DA17
•• A-DA18
•• A-DA19
•• A-DA20
•• A-DI0
•• A-DI1
•• A-DI2
•• A-DI3
•• A-DI4
•• A-DI5
•• A-DI6
612 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

•• A-DI7
•• A-DI8
•• A-DI9
•• A-DI10
•• A-DI11
•• A-DI12
•• A-DI13
•• A-DI14
•• A-DI15
•• A-DI16
•• A-DI17
•• A-DM0
•• A-DM1
•• A-DM2
•• A-DM3
•• A-DM4
•• A-DM5
•• A-DM6
•• A-DM7
•• A-DM8
•• A-DM9
•• A-DM10
•• A-DM11
•• A-DM12
•• A-DM13
•• closed world
•• constraint enforcement
•• constraint set coherence
•• constraint set minimality
•• controlled redundancy
•• data analysis
•• data contention
•• data plausibility
•• database design
•• database design axiom
•• database implementation axiom
•• database instance manipulation axiom
•• database optimization axiom
•• (E)MDM
•• E-RDM
 613

•• GROUP BY golden rule

•• HAVING clause
•• key propagation principle
•• non-relational constraints
•• null value
•• ORDER BY clause
•• positive data
•• RDM
•• referential integrity
•• relational constraints
•• SQL
•• WHERE clause
This page intentionally left blank
APPENDIX

MATHEMATIC PREREQUISITES FOR

THE MATH BEHIND

Do not worry about your difficulties in Mathematics.

I can assure you mine are still greater.
—Albert Einstein

A set is a collection/ensemble/etc. of unique elements (i.e., sets do not

allow for duplicates) whose order is irrelevant. Examples: the set of web
sites, the set of your friends, the set of e-messages you received/sent this
year, etc.
The number of elements of a set A is called its cardinal and is denoted
by card(A). For example, the cardinal of the set of rainbow colors ({red,
orange, yellow, green, blue, indigo}) is 6, the one of the world’s countries
affiliated to the United Nations is 193, etc.
The empty set (denoted Ø) is the set with no elements; trivially, card(Ø)
= 0. For example, the set of (natural) satellites of the Mercury planet is
empty (since it has no satellites).
The union (ø) of two sets is the set made out of all elements from all
corresponding sets (and as sets do not allow for duplicates, the school
union definition adds “taken only once”): A ø B = {x | x [ A ~ x [ B}.
For example, if A = {a, b, c} and B = {b, c, d}, then A ø B = {a, b, c,
d}; CHILDREN ø PARENTS = PEOPLE.
The intersection (∩) of two sets is the set containing all and only the
elements belonging to both of them: A ∩ B = {x | x [ A ` x [ B}.
For example, if A = {a, b, c} and B = {b, c, d}, then A ∩ B = {b, c}; SIN-
GLES ∩ MARRIED_AT_LEAST_ONCE = WIDOWED ø DIVORCED.
Two sets are disjoint if they don’t share any element (i.e., their inter-
section is the empty set; for example, A and B are disjoint if A ∩ B = Ø).
616 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

For example, if A = {a, b, c} and B = {1,2,3}, then A ∩ B = Ø; NEVER_

MARRIED ∩ MARRIED_AT_LEAST_ONCE = Ø.
A set is distinguished if it is disjoint of any other set, i.e. none of its
elements belongs to any other set.
A set A is included in a set B (denoted A # B or B $ A) if all elements
of A (called the subset) also belong to B (called the superset). Trivially,
any set includes itself and the empty set (both of them called improper
subsets), and equal sets are each included in the other (i.e., A # B ` B # A
 A = B). For subsets that may not be equal to their supersets, symbols ,
and . are used instead. A ÷ B denotes noninclusion. For example,
• EMPLOYEES , PEOPLE,
• CHILDREN # PEOPLE,
• TAKEOFF_AIRPORTS # LANDING_AIRPORTS,
• LANDING_AIRPORTS # TAKEOFF_AIRPORTS
• (so, TAKEOFF_AIRPORTS = LANDING_AIRPORTS = AIR-
PORTS),
• OCEANS ÷ SEAS, etc.
The complement (C) of a set is the set of all elements not belonging to
that set: AC = {x | x Ó A}. For example, if A = {1}, AC contains the whole
Universe, except for the digit 1.
The relative complement of a set A with respect to a set B (or the dif-
ference between B and A) is the set of all elements of B that do not also
belong to A: B \ A = {x | x [ B ` x Ó A}.
If A # B, then B \ A is called the absolute complement of A with respect
to B.
For example, if A = {1,2,3}, B = {1}, then B \ A = Ø and A \ B = {2,3};
PEOPLE \ CHILDREN = ADULTS (so ADULTS is the absolute comple-
ment of CHILDREN with respect to PEOPLE); etc.
An axiom (postulate, assumption) is any premise mathematical state-
ment, so evident as to be accepted as true without controversy, which
serves as a starting point from which other statements are logically
derived. Within the system they define, axioms (unless redundant) can-
not be derived by principles of deduction, nor are they demonstrable by
mathematical proofs, simply because they are starting points; there is
nothing else from which they logically follow: otherwise, they would
be classified as theorems. For example, in Euclidian geometry, an axiom
Appendix
 617

states that all right angles are equal to one another, while Thales’ theo-
rem states that, in any circle, any triangle formed between three distinct
points of the circle is a right one if and only if two of the points are on a
circle’s diameter.
Especially in metalogic and computability (recursion) theory (but not
only), methods (procedures, algorithms131) taking some class of problems
and reducing the solution to a finite set of steps that, for all class problems
instances, always computes the right answers are called effective.
Functions (see their definition below) with an effective method are
called  effectively calculable. Several independent efforts to provide a
formal characterization of effective calculability led to a variety of pro-
posed definitions132 (that were ultimately shown to be equivalent): the
notion captured by these definitions is called  (recursive) computabil-
ity. The Church–Turing thesis133 states that these two notions coincide:
any number-theoretic function that is effectively calculable is recursively
computable and vice-versa.
Generally, any function defined on and taking values from a set of for-
mulas is called an inference (derivation, deductive, transformation) rule.
In classical logic (as well as the semantics of many other nonclassical
logics) inference rules are also truth preserving134: if the premises are true
(under an interpretation) then so is the conclusion. Usually, only rules that
are recursive (i.e., for which there is an effective procedure for determin-
ing whether, according to the rule, any given formula is the conclusion of
a given set of formulae) are important.
For example, “if x is an ancestor of y and z is x’s father, then z is an
ancestor of y too” is a (transitive recursive) inference rule. Modus ponens
((p ` (p  q))  q), modus tollens ((¬q ` (p  q))  ¬p), resolution ((p
~ q) ` (¬p ~ r))  (q ~ r)), hypothetical ((p  q) ` (q  r))  (p  r))
and disjunctive ((p ~ q) ` ¬p)  q) syllogisms are well-known examples
of propositional logic (see below) inference rules.
In linguistics, computer science, and mathematics, a formal language
is a set of strings of symbols that may be constrained by rules that are
131
Some authors consider only effective methods for calculating the values of a function as being
algorithms.
132
Namely, general recursion, Turing machines, and λ-calculus.
133
Note that this is not a mathematical statement, so it cannot be proven mathematically.
134
In many-valued logics, they preserve a general designation.
618 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

specific for it. The alphabet of a formal language is the set of symbols,
letters, or tokens from which the strings of the language may be formed;
frequently it is required to be finite. The strings formed from this alphabet
are called words, and the words that belong to a particular formal language
are sometimes called well-formed words/formulas. For example, all com-
puter programming languages (and not only) are studied using the formal
languages theory.
A formal language is often defined by means of a formal grammar135:
a set of formation rules for rewriting strings (rewriting rules), along with
a start symbol from which rewriting must start. The rules describe how to
form strings from the language’s alphabet that are valid according to the
language’s syntax. A grammar is usually thought of as a language genera-
tor. However, it can also sometimes be used as the basis for a “recognizer”
— a tool that determines whether a given string belongs to the language or
is grammatically incorrect.
A formal system is a mathematical well-defined system of abstract
thought consisting of finite sets of symbols (its alphabet), axioms and
inference rules (the latter two constituting its deductive system), plus
a grammar for generating and/or accepting the well-formed formulas
forming its formal language. For example, Euclid’s Elements (or even
Spinoza’s Ethics, its apparently nonmathematical imitation), the pred-
icate, propositional, lambda, and domain relational calculi are formal
systems.
Deductive systems should be sound (i.e., all provable statements are
true) and complete (i.e., all true statements are provable). Similarly, algo-
rithms are said to be sound and complete (if they analyze/compute only
and all possible valid cases/values of the corresponding subuniverse).
Propositional (or sentential) logic (or calculus) is a formal system in
which formulas of a formal language may be interpreted as representing
propositions. A system of inference rules and axioms allows certain for-
mulas (called theorems) to be derived, which may be interpreted as true
propositions. The series of formulas which is constructed within such a
system is called a derivation and the last formula of the series is a theorem,
whose derivation may be interpreted as a proof of the truth of the proposi-
tion represented by the theorem.
135
For example, regular, context-free, deterministic context-free, etc.
Appendix
 619

Truth-functional propositional logic is a propositional logic whose

interpretation limits the truth values of its propositions to two, usually
denoted as (Boolean) true and false. Its logical expressions (yielding either
true or false) operands are logical ones: the constants true and false, (in)
equalities (=, ≠, <, ≤, >, ≥—also called the standard math operators/rela-
tions), logical functions, subexpressions enclosed in parenthesis, etc.; their
fundamental operators are not (¬), and (`), and or (~); most important
derived operators are implication (, ) and equivalence (). Truth-
functional propositional logic is considered to be zeroth-order logic.

Zeroth-order logic truth tables are the following:

¬ true = false; ¬ false = true.
false ` true = false; false ` false = false; true ` true = true;
false ~ false = false; false ~ true = true; true ~ true = true;
a  b = ¬ a ~ b = b  a; a  b = a  b ` b  a.

First-order logic (first-order predicate calculus, lower predicate calcu-

lus, quantification theory, predicate logic) (1OL) is a formal system used
in mathematics, philosophy, linguistics, computer science, etc., which
extends propositional logic with the existential (“there is/are”, denoted
‘'’) and universal (“for any”, denoted ‘;’) quantifiers that are applied to
predicates’ variables.
A theory about some topic is usually first-order logic together with a
specified domain of discourse (over which the quantified variables range),
finitely many functions (which map from that domain into it), finitely many
predicates (defined on that domain), and a recursive set of axioms (which
are believed to hold for those things). Note that, sometimes, “theory” is
understood in a more formal sense, which is just a set of sentences in first-
order logic, and that, in first-order theories, predicates are often associated
with atomic sets (i.e., not sets of sets).
In 1OL there are two classes of formulas: open ones, containing at
least one variable occurrence not bound to any quantifier, and closed ones
(propositions), in which all variable occurrences are bound to a quantifier.
In any fixed interpretation, propositions are either true or false, whereas
open formulas truth values depend on their actual variables values.136
136
Note that db constraints are closed while queries are open formulas.
620 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

An operator θ over some set A is said to be:

ü associative, if a θ (bθ c) = (a θ b) θ c, ∀a, b, c[A (case in which
parenthesis may be dropped, and simply write a θ b θ c instead and,
generally, a1θ…θ an);
ü commutative, if aθ b = bθ a, ∀a, b[A;
ü anticommutative, if aθ b ≠ bθ a, ∀a, b[A;
ü distributive over another operator τ, if a θ (b τ c) = a θ b τ a θ b,
;a, b, c[A.
For example, for all number sets, addition is associative, commuta-
tive, and distributive over multiplication, while matrix multiplication is
anticommutative.
Given any two sets A and B, their Cartesian product, denoted A × B,
is the set of all possible pairs <x, y>, with x from A and y from B: A × B
= {(x, y) | x [ A ` y [ B}. Obviously, card (A × B) = card (A) * card(B).
As × is associative, it may also be considered to be a n-ary operator, n > 1
natural. For example, if A = {1,2}, B = {a, b}, and C = {true, false}, then
A × B × C = {<1, a, true>, <1, a, false>, <1, b, true>, <1, b, false>, <2, a,
true>, <2, a, false>, <2, b, true>, <2, b, false>}.
× is anticommutative: for any sets A and B, A × B ≠ B × A; the only
commutative Cartesian products are the homogenous ones that is those
defined on a same set: A × A = A × A, for any set A. For example, if A =
{a, b} and B = {1, 2, 3}, there are two (distinct!) Cartesian product sets on
them: A × B = {(a,1), (a,2), (a,3), (b,1), (b,2), (b,3)} and B × A = {(1, a), (2,
a), (3, a), (1, b), (2, b), (3, b)}.
Given any two sets A and B, a binary or dyadic (mathematical) rela-
tion R on them (“from A to B”) is any nonvoid subset of their Cartesian
product. For example, let A = {a, b}, B = {1, 2, 3}, and R = {(a,1), (a,3),
(b,2)} , A × B. A scheme and instance type notation is also used: R = (A,
B; GR), where A and B are called R’s underlying sets (forming its scheme)
and GR # A × B is called R’s graph (its instance). If A = B, R is said to be
homogenous. Relations may have any natural number of underlying sets
(i.e., they may be ternary, quaternary, etc., generally n-ary).
A mapping (function) is a binary relation satisfying the following two
properties (constraints): total definition and functionality. A mapping f
# A × B is denoted f : A → B, with A being called its domain and B its
codomain.
Appendix
 621

A binary relation f # A × B is totally defined if its graph contains at

least one pair for each element of its domain (A); it is functional if its graph
contains at most one pair for each element of its domain (A). Obviously,
these two properties may be consolidated into only one: f is a mapping
(function) if its graph contains exactly one pair (denoted (x, f(x))) for each
element of its domain (A). For example:
• R above is totally defined, but not functional (as there are two pairs
for a[A)
• f = {(a,1), (b,2)} , A × B is a mapping f : A → B, with f(a) = 1 and
f(b) = 2, where A = {a, b}, B = {1,2}.
• BirthPlace : PEOPLE → CITIES is a mapping, as anybody is/was
born in one and only one city.
Two mappings are equal if they have same domain, codomain, and
graph. For example, sin : (–∞, ∞) → (–∞, ∞) is not equal to sin : (–∞, ∞) →
[–1, 1], as they have different codomains.
For any Cartesian product A × B there are two unique distinguished map-
pings prA : A × B → A and prB : A × B → B, called its canonical (Cartesian)
projections, which are mapping each pair to its corresponding pair element:
prA(x, y) = x, prB(x, y) = y, ;(x, y)[A × B. For example, if A = {a, b} and B
= {1, 2, 3}, prA(a,1) = prA(a,2) = prA(a,3) = a, prA(b,1) = prA(b,2) = prA(b,3)
= b, prB(a,1) = prB(b,1) = 1, prB(a,2) = prB(b,2) = 2, prB(a,3) = prB(b,3) = 3.
For any mapping f : A → B, its image, denoted Im(f) = {y[B | x[A,
y=f(x)} # B, is the set of all of its values. For example, if f : {1,2,3} → {a,
b, c}, with f(1) = b, f(2) = b, f(3) = c, then Im(f) = {b, c}.
The mapping composition h = g ° f of mappings f : A → B and g : B →
C is the unique mapping h : A → C computed as h(x) = g(f(x)), ∀x[A. For
example, if f : {1,2,3} → {a, b, c}, with f(1) = b, f(2) = b, f(3) = c, g : {a,
b, c} → {true, false}, with g(a) = true, g(b) = false, g(c) = true, then h :
{1,2,3}→ {true, false}, with h(1) = g(f(1)) = g(b) = false, h(2) = g(f(2)) =
g(b) = false, h(3) = g(f(3)) = g(c) = true.
A mapping f : A → A is called an automapping (autofunction). For
example, ReportsTo : EMPLOYEE → EMPLOYEE is an automapping.
An automapping is said to be idempotent if f ° f = f 2 = f. For example, f :
{1,2,3} → {1,2,3}, with f(1) = 3, f(2) = 2, f(3) = 3, is idempotent, as f 2(1) =
f(f(1)) = f(3) = 3 = f(1), f 2(2) = f(f(2)) = f(2) = 2 = f(2), f 2(3) = f(f(3)) = f(3)
= 3 = f(3).
622 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

A mapping f : A → B is one-to-one (injective) if its graph contains at

most one pair for each element of its codomain (B). For example, g : A →
B, with g(a) = 2 and g(b) = 1 is one-to-one, whereas g : A → B, with g(a)
= 1 and g(b) = 1 is not; MAC : NETWORK_ADAPTERS → GUIDS is one-
to-one (as any network adapter has a unique GUID), whereas BirthPlace :
PEOPLE → CITIES is not (as several people are/were born in any city).
Obviously, one-to-oneness is a dual of functionality.
A mapping f : A → B is onto (surjective) if Im(f) = B.
For example, f : {1,2,3} → {a, b, c}, with f(1) = b, f(2) = b, f(3) = c is
not onto, whereas f : {1,2,3} → {a, b, c}, with f(1) = b, f(2) = a, f(3) = c
it is; sin : (–∞, ∞) → (–∞, ∞) is not onto, but sin : (–∞, ∞) → [–1, 1] it is;
BirthPlace : PEOPLE → CITIES is onto (as, in any city, at least one per-
son is/were born), whereas Capital : COUNTRIES → CITIES it is not (as
not all cities are capitals).
A mapping f : A → B is bijective if it is both one-to-one and onto; some-
times, bijective mappings are denoted f : A ↔ B; such A and B are called
equipotent (as card(A) = card(B)) or isomorphic.
For example, f : {1,2,3} → {a, b, c}, with f(1) = b, f(2) = a, f(3) = c is
bijective, whereas f : {1,2,3} → {a, b, c}, with f(1) = b, f(2) = b, f(3) = c it
is not; sin : (–∞, ∞) → [–1, 1] is not bijective, as it is not one-to-one (e.g.,
sin(0) = sin(π) = sin(2π) = … = sin(kπ) = 0, k integer), whereas sin : [–π/2,
π/2] → [–1, 1] it is; the real linear function (ax + b) is bijective, whereas
the squaring one (ax2 + bx + c, a ≠0) it is not.
For any set A there is a unique distinguished bijective auto-mapping 1A:
A → A called A’s unity function, which is mapping each of its elements to
itself: 1A(x) = x, x[A.
A mapping (Cartesian) product h = f • g of mappings f : A → B and g :
A → C is the unique mapping h : A → B × C computed as h(x) = (f(x), g(x)),
∀x[A.
For example, if f = FirstName : PEOPLE → ASCII(255) and g = Last-
Name : PEOPLE → ASCII(128), h = FirstName • LastName : PEOPLE →
ASCII(255) × ASCII(128); if f(x) = ‘Dinu’ and g(x) = ‘Lipatti’, then h(x)
= (‘Dinu’, ‘Lipatti’).
If f : A → B is one-to-one then, ;g : A → C, f • g is one-to-one too.
Appendix
 623

For example, as SSN : PEOPLE → NAT(9) is one-to-one (as there

may not be two persons having same SSN), SSN • BirthDate : PEOPLE →
NAT(9) × [1/1/1900, SysDate()] is one-to-one too.
A mapping (Cartesian) product p = f1 • … • fn, n > 1 natural, is mini-
mally one-to-one if it is one-to-one but none of its proper subproducts is
one-to-one; in particular (for n = 2), f • g is minimally one-to-one if it is
one-to-one and neither f, nor g are.
For example, SSN • BirthDate : PEOPLE → NAT(9) × [1/1/1900, Sys-
Date()] is not minimally one-to-one (as SSN is one-to-one), whereas Country
• State : STATES → COUNTRIES × ASCII(64) it is: not only it is one-to-one
(as there may not be two states with a same name in a same country), but
none of its component is (Country as there are several states in a country,
e.g. 50 + a federal district in the U.S., 41 in Romania, etc., and State as there
may be two states of different countries having same name, e.g. Cordoba, in
Spain and Argentina; La Rioja, in Spain, Peru, and Argentina, etc.).
[x], x real, denotes the integer part of x. For example, [π] = 3, [5.9] = 5.
n! = 1 * 2 * … * (n—1) * n is the factorial of n, n natural.
For example, 4! = 1 * 2 * 3 * 4 = 24.
xy denotes the y-th power of x, x, y being complex (real, integer, natural,
etc.) numbers. For example, 24 = 2 * 2 * 2 * 2 = 16.
Applying Newton’s binomial theorem for x = y = 1, 2n = C(n, 0) + … +
C(n, n), n natural, where C(n, j) denotes the (natural) number of “n choose
j”, j natural, n ≥ j (i.e., combinations of j elements taken from n ones: C(n,
j) = n!/(j! * (n—j)!)).
For example, ;n natural, C(n, 0) = C(n, n) = 1; C(4,3) = 4!/(3! * 1!) =
24/6 = 4; etc.
The number of operands of an operator or of arguments of a function is
called its arity. For example, constants have arity 0 (are nullar), the logical
not has arity 1 (is unary), just like the functions card and Im, while and
and or have arity 2 (are binary), just like the function (Cartesian) product.
As this product is associative, f1 • … • fn has arity n (or is n-ary), n natu-
ral. Notationally, arity is denoted just like the absolute value operator: for
example, | f1 • … • fn | = n.
A unary operator f is monotonic (monotone) if it preserves ordering that
is if ;x, y, (x ≤ y  (f(x) ≤ f(y)), where ≤ is an order (e.g., ≤ on numbers,
but also # on sets, etc.).
624 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

For example, the linear numeric function f(x) = ax + b, a ≠ 0 is mono-

tonic (and, in particular, increasing, when a > 0, and decreasing, when a <
0), whereas the squaring one (ax2 + bx + c, a ≠0) it is not,
A binary homogenous relation R over a set A is:
ü total (or complete), if aRb ~ bRa, a, b[A;
ü connected, if aRb ~ bRa, a, b[A, a ≠ b;
ü partial (not total), if a, b[A such that a¬Rb ` b¬Ra;
ü reflexive, if aRa, a[A;
ü irreflexive, if a¬Ra, a[A; (sometimes called antireflexive)
ü symmetric, if aRb  bRa, a, b[A;
ü antisymmetric, if aRb ` bRa  a=b, a, b[A;
ü asymmetric, if aRb  ¬(bRa), a, b[A;
ü transitive, if aRb ` bRc  aRc, a, b, c[A;
ü intransitive, if aRb ` bRc  ¬(aRc), a, b, c[A; (sometimes
called antitransitive)
ü Euclidean, if aRb ` aRc  bRc, a, b, c[A;
ü cyclic, if {a1, …, an} # A, n > 0, natural, such that a1Ra2 ` … `
an-1Ran ` anRa1; such a {a1, …, an} subset of A is called a (simple)
cycle (circuit, circle, polygon) (in R’s graph);
ü acyclic, if it has no cycles.
For example, if A = {a, b, c}, then:
ü R = {(a, b), (b, c), (c, a)} is total, irreflexive, asymmetric, intransi-
tive, and cyclic;
ü R = {(a, a)} is partial, reflexive, antisymmetric, and cyclic;
ü R = {(a, b), (b, a)} is partial, irreflexive, symmetric, and cyclic;
ü R = {(a, b), (b, c), (a, c)} is total, irreflexive, asymmetric, transi-
tive, Euclidean, and acyclic;
ü R = {(a, b), (a, c)} is total, irreflexive, asymmetric, intransitive, and
acyclic.
Trivially, if A = R (the set of real numbers), then, for example:
ü ≥ is total, reflexive, antisymmetric, transitive, and not asymmetric;
ü > is partial, irreflexive, asymmetric, antisymmetric, transitive, and
acyclic;
ü ≠ is irreflexive and symmetric.
Obviously, auto-mappings are particular types of homogenous binary
(dyadic) relations, so they might also be reflexive, irreflexive, etc.
Appendix
 625

A binary homogenous relation R is a partial order if it is reflexive,

antisymmetric and transitive.
A partially ordered (by a partial order po) set S is abbreviated poset and
denoted (S, po). Inclusions are the most well-known examples of partial orders.
For each inclusion A # B there is a unique one-to-one distinguished
mapping iA: A → B called the canonical inclusion function, which maps
each subset element to the corresponding superset one: iA(x) = x, x[A.
Trivially, when B = A, iA = 1A.
An order is called total if it is also total. <, ≤, >, ≥ over numeric sets are
the most well-known examples of total orders.
A binary homogenous relation R is a preorder (quasiorder) if it is
reflexive and transitive.
Functional database dependency is the best rdb example of a preorder.
A binary homogenous relation is an equivalence one if it is reflexive,
symmetric and transitive.
Two sets are equal (=) if they have exactly same elements; trivially, A
# B ` B # A  A = B, for any sets A and B. Equality (not only of sets) is
the most well-known example of an equivalent relation.
Other well-known examples of equivalence relations are geometric
equivalences (e.g., triangle, rectangle, etc.); in rdb, all keys of any relation
are equivalent (see Exercise 3.42(iv)), all attributes that mutually depend
on each other (see Exercise 3.42(i)) and all constraint sets having same
closure are equivalent too, etc.
Two or more subsets of a set are called its partitions if they are pair-
wise disjoint and their union is equal to the set. Equivalently, a set is par-
titioned into several partitions if each of its elements belongs to one and
only one partition.
Most often, partitions of a set are indicated with the derived direct sum
(!) operator: A = P1 ! … ! Pn, n>1, natural, where Pi are A’s partitions,
is equivalent with A = P1 ø … ø Pn ` Pi ∩ Pj = Ø, i ≠ j, 1 ≤ i, j ≤ n.
For example, PEOPLE = ADULTS ! CHILDREN, PEOPLE =
ELECTED ! NOT_ELECTED, ELECTED = PRESIDENT !
SENATORS ! CONGRESSMEN ! MAYORS ! COUNCILLORS, etc.
When a set is partitioned by an equivalence relation, then its partitions
are members of the corresponding quotient set. For any quotient set A/~
(generated by equivalence relation ~ over set A) there is a unique onto
626 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

mapping r~: A → A/~ called its canonical surjection, which maps any ele-
ment of A to the partition to which it belongs. Sometimes, partitions are
identified by a (unique) representative element from A, in which case A/~
# A is called a representative system (system of representatives).
For example, let A = CITIZENS and ~ # CITIZENS × CITIZENS,
defined by “;x, y[CITIZENS, x~y iff x and y have the same supreme
head of state”; then CITIZENS/~ includes, for example, the classes of all
subjects of HM Queen Elisabeth II of the U.K. (also including Canadians,
Australians, New-Zealanders, etc.), of all citizens of the U.S. and their
territories, of all German citizens, etc. The corresponding representa-
tive system would now include the Japanese Emperor Akihito, Queens
Elisabeth II of the U.K., Margrethe II of Denmark, presidents Barack
Obama, Xi Jinping, Vladimir Putin, François Hollande (of the U.S.,
China, Russia, and France, respectively), etc. The canonical surjection
SupremeStateHead : CITIZENS → CITIZENS/~ has, for example, the
following values: SupremeStateHead(Seiji Ozawa) = Emperor Akihito,
SupremeStateHead(Sir Elton Hercules John) = Queen Elisabeth II of the
U.K., SupremeStateHead(Céline Marie Claudette Dion) = Queen Elisa-
beth II of the U.K., SupremeStateHead(Eldrick Tont “Tiger” Woods) =
President Barack Obama, SupremeStateHead(Dalai Lama) = President Xi
Jinping, etc.
The kernel (nucleus) of a mapping f : A → B, denoted ker(f), is an
equivalence relation over its domain, partitioning it into classes grouping
together all of its elements for which f takes same values: ker(f) = { (x, x’)
[ A2 | f(x) = f(x’) }. Sometimes, instead of the functional notation ker(f),
an infix one is used: x =f x’  f(x) = f(x’). The corresponding induced par-
tition (and quotient set) is {{w [ A | f(x) = f(w)} | x [ A } = A/ker(f), which
is also called the function coimage (denoted coim(f)). Note that coimages
are naturally isomorphic to images, as there is a bijection from coim(f) to
Im(f). It is very easy to show that ker(f•g) = ker(f) ∩ ker(g), for any func-
tion product f•g.
For example, with Country : STATES → COUNTRIES and Region :
STATES → REGIONS, ker(Country) = {(Alabama, Alabama), (Ala-
bama, Alaska), …, (Wisconsin, Wyoming), (Wyoming, Wyoming),
(Alba, Alba), (Alba, Arad), …, (Vâlcea, Vrancea), (Vrancea, Vrancea),
…}, ker(Region) = {(Connecticut, Connecticut), (Connecticut, Maine),
Appendix
 627

…, (Oregon, Washington), (Washington, Washington), (Argeş, Argeş),

(Argeş, Brăila), …, (Covasna, Harghita), (Harghita, Harghita), …},
and ker(Country•Region) = ker(Country) ∩ ker(Region) = ker(Region)
= {(Connecticut, Connecticut), (Connecticut, Maine), …, (Oregon,
Washington), (Washington, Washington), (Argeş, Argeş), (Argeş,
Brăila), …, (Covasna, Harghita), (Harghita, Harghita), …}.
A set together with an associative binary operation on it forms a semi-
group. Semigroups are one of the simplest algebraic structures, that is
an arbitrary set (called underlying (carrier) set) with one or more finitary
operations (i.e., operations having a finite arity) defined on it. For example,
the set of integers forms semigroups both with addition and multiplication.
Given a set and a binary operation on it, the element of the set that
leaves all set elements (including itself) unchanged when combined with
them through the binary operation, if any, is called the identity (or neutral)
element (of that set, with respect to that binary operation) or, simply, the
identity (e.g., for any number sets, 0 for sum and subtraction, 1 for multi-
plication and division).
A semigroup with identity is called monoid. For example, obviously,
any number set (i.e., complex, real, rational, integers, naturals) forms
monoids both with addition and multiplication.
A set S of symbols (called the alphabet) equipped with the associative
binary operation called string concatenation (the operation of joining two
character strings end-to-end) and augmented with a unique distinguished
identity element (called the empty string, generally denoted by ε or λ, and
representing, by convention, the unique sequence of zero elements from
S) is called the freely137 (generated) monoid over S, denoted S*. Obviously,
S* elements are character strings of various lengths made up with the sym-
bols of S.
Examples: ASCII*, UNICODE*, etc. ASCII(n) , ASCII*, where n is a
natural, is the subset made out only of strings having maximum length n.
The closure of a homogeneous binary relation R is the relation R+ = {(x,
z) | (x, y)[R ` (y, z)[R}.
R’ is the transitive closure of a relation R if and only if:
137
Generally, an abstract monoid (or semigroup) S is described as free if it is isomorphic to the free
monoid (or semigroup) on some set, where, informally, a free algebraic object over a set S can be
thought as being a “generic” algebraic structure over S: the only equations that hold between elements
of a free object are those that follow from the defining axioms of the corresponding algebraic structure.
628 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

1. R’ is transitive,
2. R # R’, and
3. for any relation R”, if R # R” and R” is transitive, then R’ # R”
(i.e., R’ is the smallest relation that satisfies (1) and (2)).
Similarly, any other such closure (e.g., reflexive, symmetric, etc.) can
be defined.
Obviously, the transitive closure of a relation can be obtained by clos-
ing it, closing the result, and continuing to close the results of the previous
closures until no further elements are added. Note that the digraph of the
transitive closure of a relation is obtained from the digraph of the relation
by adding for each directed path the arc that shunts the path if one is not
already there.
Let R be a relation from X to Y and S one from Y to Z;
The composition (or join or concatenation) of R and S, written R.S, is
the relation R.S = {(x, z)[X × Z | xRy and ySz, y[Y}.
A function-style notation S°R is also sometimes used, although it is
quite inconvenient for relations. The notation R.S is easier to deal with, as
the relations are named in the order that leaves them adjacent to the ele-
ments that they apply to (thus, x(R.S)z, because xRy and ySz for some y).
Transitivity and composition may seem similar, but they are not: both
are defined using x, y, and z, for one thing, but transitivity is a property of
a single relation, while composition is an operator on two relations that
produces a third relation (which may or may not be transitive).
The product of two relations R and S is the relation R × S = {(w, x, y, z) |
wRx ` ySz}.
The converse (or transpose) of R, written R−1, is the relation R−1 = {(y,
x) | xRy}.
Symmetric and converse may also seem similar, but they are actually
unrelated: both are described by swapping the order of pairs, but sym-
metry is a property of a single relation, while converse is an operator that
takes a relation and produces another relation (which may or may not be
symmetric). It is trivially true, however, that the union of a relation with
its converse is a symmetric relation.
Examples:
Let X = {Airplane, Pool, Restaurant}, Y = {Goggles, Heels, Seatbelt,
Tuxedo}, Z = {Buckle, Pocket, Strap}, and R = {(Airplane, Seatbelt), (Air-
Appendix
 629

plane, Tuxedo), (Pool, Goggles), (Restaurant, Heels), (Restaurant, Tux-

edo)} be the relation “is where one wears”, and S = {(Goggles, Strap),
(Heels, Buckle), (Heels, Strap), (Seatbelt, Buckle), (Seatbelt, Strap), (Tux-
edo, Pocket)} the relation “can have a”; then:
• R.S = {(Airplane, Buckle), (Airplane, Pocket), (Airplane, Strap),
(Pool, Buckle), (Pool, Strap), (Restaurant, Buckle), (Restaurant,
Pocket), (Restaurant, Strap)}.
• R-1 = {(Goggles, Pool), (Heels, Restaurant), (Seatbelt, Airplane),
(Tuxedo, Airplane), (Tuxedo, Restaurant)}.
As relations are sets (even if of pairs), all the operations and relations
on sets also apply to relations as well; for example:
The intersection of any R, S, is the relation R ∩ S = {x(R∩S)y | xRy ` xSy}.
The union of R and S, is the relation R ø S = {x(RøS)y | xRy ~ xSy}.
The difference of R and S, is the relation R − S (or R \ S) = {x(R−S)y |
xRy ` ¬ xSy}.
Let U be a set and R and S be relations on U (their common underlying set);
R = S iff (∀x, y∈U)(xRy  xSy).
R # S iff (∀x, y∈U)(xRy  xSy).
Originally138, a sentence is said to be undecidable when it can neither
be proved, nor refuted. For example, the so-called axiom of choice (equiv-
alent to the statement “the Cartesian product of a collection of nonempty
sets is nonempty”) is undecidable in the standard axiomatization set theory
except for it.
A Turing machine (TM) is a theoretical device that manipulates sym-
bols on an infinite strip of tape according to a set (table) of rules, by reading
and/or writing one symbol at a time. A TM whose set of rules prescribes at
most one action to be performed for any given situation139 is called deter-
ministic (DTM) or ordinary, whereas when there are several actions for at
least one situation it is called nondeterministic (NTM).
A complexity class is a set of problems of related resource-based com-
plexity. A typical complexity class definition has the form “the set of prob-
lems that can be solved by an abstract machine M using O(f(n)) of resource
138
This first sense arouse from the Gödel’s theorems; in the computability theory, there is a second
sense for undecidability, which applies to decision problems (countably infinite sets of questions each
requiring a yes or no answer) rather than to sentences: there is no algorithm (computable or effectively
calculable function) that correctly answers every question in the problem set. In this sense, for ex-
ample, the problem of whether a Turing machine halts or not on any given program is undecidable.
139
That is a pair of current <state, symbol>.
630 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

R”, where n, natural, is the size of the input, f(n) is a function of n (e.g.,
n, n2, 2n, etc.), O(x) is the complexity operator (called big O), and, for
example, resources are the number of steps (e.g., needed to be performed
by an algorithm in the corresponding worst case), execution time, memory
space, etc.
For example, the NC class is a set of decision problems decidable
in polylogarithmic time on a parallel computer with a polynomial num-
ber of processors—a class of problems having highly efficient parallel
algorithms.
L (also known as LOGSPACE, LSPACE or DLOGSPACE) is the
complexity class containing decision problems that can be solved by a
DTM using a logarithmic amount of memory space. L contains precisely
those languages expressible in first-order logic with an added commuta-
tive transitive closure operator. This result has applications to database
query languages: the data complexity of a query is defined as the complex-
ity of answering a fixed query, considering the data size as the variable
input. For this measure, queries against relational databases with complete
information (i.e., having no notion of nulls) as expressed, for instance, in
relational algebra are in L.
The main idea of logspace is that you can store a polynomial-mag-
nitude number in logspace and use it to remember pointers to a position
of the input. The logspace class is therefore useful to model computation
where the input is too big to fit in the internal memory (RAM) of a com-
puter. Long DNA sequences and databases are good examples of problems
where only a constant part of the input will be in RAM at a given time and
where we have pointers to compute the next part of the input to inspect,
thus using only logarithmic memory.
L is a subclass of NL (NLOGSPACE), which is the class of languages
decidable in logarithmic space on a nondeterministic Turing machine.
It has been proved that NC # L # NL # NC2, that is: given a parallel
computer C with a polynomial number O(nk) of processors for some con-
stant k, any problem that can be solved on C in O(log n) time is in L, and
any problem in L can be solved in O(log2 n) time on C. Still open problems
include whether L = NL.
The polynomial complexity class (P, or PTIME, or DTIME(nO(1))) con-
tains all decision problems that can be solved by a deterministic Turing
Appendix
 631

machine using a polynomial amount of time. Cobham’s thesis holds that

P is the class of computational problems that are “efficiently solvable” or
“tractable”; however, in practice some problems not known to be in P have
practical solutions, and some that are in P do not. A decision problem is
P-complete if it is in P.
The most famous example of a P-complete problem is the Turing
machine halting one: given a Turing machine, an input for it, and a num-
ber T written in unary, does that machine halt on that input within the first
T steps? Other examples include: the Lexicographically First Depth First
Search Ordering (given a graph with fixed ordered adjacency lists and
nodes u and v of it, is vertex u visited before vertex v in a depth-first search
induced by the order of the adjacency lists?), the Context Free Grammar
Membership (given a context-free grammar and a string, can that string
be generated by that grammar?), and the Horn-satisfiability (given a set
of Horn clauses, is there a variable assignment which satisfies them?).
Dually, there are problems in P that are not known to be either P-complete
or NC, but are thought to be difficult to parallelize. Examples include the
decision problem of finding the greatest common divisor of two numbers
and the one for determining what answer the extended Euclidean algo-
rithm would return when given two numbers.
It has been shown that NL # P. An open problem is whether L = P?
The nondeterministic polynomial complexity class (NP) is the set of
decision problems whose solutions can be determined by a NTM in poly-
nomial time (i.e., in O(nk), where k is a natural constant). For example, the
traveling salesman problem (given an input matrix of distances between n
cities, determine if there is a route visiting all cities with total distance less
than some k) is in NP.
Informally, the nondeterministic polynomial-time hard complexity
class (NP-hard) is the set of decision problems that are at least as hard
as the hardest problems in NP. Formally, a problem P is NP-hard when
every problem R in NP can be reduced in polynomial time to P. As a con-
sequence, finding a polynomial algorithm to solve any NP-hard problem
would give polynomial algorithms for all the problems in NP140, which
is unlikely, as many of them are considered as hard. For example, the
A common mistake is to think that the NP in NP-hard stands for non-polynomial: although it is
140

widely suspected that there are no polynomial-time algorithms for NP-hard problems, this has not
been proven yet. Moreover, NP also contains all problems that can be solved in polynomial time.
632 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

decision subset sum problem (given a set of integers, does any nonempty
subset of them add up to zero?) is NP-hard.
A decision problem is NP-complete when it is both in NP and NP-
hard. The set of NP-complete problems is denoted by NP-C (or NPC).
Although any given solution to an NP-complete problem can be verified
in polynomial time, there is no known efficient way to find such a solution
in the first place (the most notable characteristic of NP-complete problems
is that no fast solution to them is known): the time required to solve such
problems using any currently known algorithm increases very quickly as
the size of the problem grows. As a consequence, determining whether or
not it is possible to solve these problems quickly, called the P versus NP
problem, is one of the main still unsolved problems in computer science.
NP-complete problems are generally addressed by using heuristic meth-
ods and approximation algorithms.
For example, the above NP-hard subset sum problem is also NP-com-
plete. Another famous NP-hard example is the Boolean Satisfiability Prob-
lem (also called the Propositional Satisfiability Problem and abbreviated
as SATISFIABILITY or SAT): there exists an interpretation that satisfies a
given Boolean formula? In other words, can the variables of a given Bool-
ean formula be consistently replaced by true or false in such a way that the
formula evaluates to true?141 Despite the fact that no algorithms are known
to solve SAT efficiently, correctly, and for all possible input instances,
many instances of SAT-equivalent problems that occur in practice (e.g., in
AI, circuit design, automatic theorem proving, etc.) can actually be solved
rather efficiently using heuristical SAT-solvers. Although such algorithms
are not believed to be efficient on all SAT instances, they tend to work well
for many practical applications.
There are decision problems that are NP-hard but not NP-complete,
such as the programs halting problem: given a program and its input, will
it run forever?142 For example, the Boolean satisfiability problem can be
reduced to the programs halting problem by transforming it to the descrip-
tion of a Turing machine that tries all truth value assignments and, when
141
If this is the case, the formula is called satisfiable. On the other hand, if no such assignment exists,
the function expressed by the formula is identically false for all possible variable assignments and the
formula is unsatisfiable. For example, the formula a ` ¬b is satisfiable, because a = true and b = false
make it true; dually, a ` ¬a is unsatisfiable.
142
That’s a yes/no question, so this is a decision problem.
Appendix
 633

it finds one that satisfies the formula, it halts; otherwise, it goes into an
infinite loop. The halting problem is not in NP since all problems in NP
are decidable in a finite number of operations, while the halting problem
is generally undecidable.
There are also NP-hard problems that are neither NP-complete nor
undecidable. For instance, the language of True quantified Boolean formu-
las (TQBF)143 is decidable in polynomial space, but not nondeterministic
polynomial time (unless NP = PSPACE).
The polynomial space complexity class (PSPACE) is the set of decision
problems that can be solved by a DTM only using a polynomial amount
of (memory) space.
A decision problem is PSPACE-complete if it can be solved using an
amount of memory that is polynomial in the input length and if every
other problem that can be solved in polynomial space can be transformed
to it in polynomial time. The problems that are PSPACE-complete can be
thought of as the hardest problems in PSPACE, because a solution to any
such problem could easily be used to solve any other problem in PSPACE.
The PSPACE-complete problems are thought to be outside of the P and
NP complexity classes, but this has not been proved.144
For example, the canonical PSPACE-complete problem is the quanti-
fied Boolean formula problem (QBF) one, a generalization of the Boolean
satisfiability problem, in which both existential and universal quantifiers
can be applied to each variable: it asks whether a quantified sentential
form over a set of Boolean variables is true or false; for example, the fol-
lowing is an instance of QBF: ;x 'y 'z ((x ~ z) ` y).
The complexity class EXPSPACE is the set of all decision problems solv-
able by a DTM in O(2p(n)) space, where p(n) is a polynomial function of n.
NEXPTIME (or NEXP) is the class of problems solvable in exponential
time by a nondeterministic Turing machine.

143
A (fully) quantified Boolean formula is a formula in quantified propositional logic where every
variable is quantified (or bound), using either existential or universal quantifiers at the beginning of
the sentence. Such a formula is equivalent to either true or false (since there are no free variables). If
such a formula evaluates to true, then it belongs to TQBF (also known as QSAT or Quantified SAT).
144
It is, however, known that they lie outside of the NC class, because problems in NC can be solved
in an amount of space polynomial in the logarithm of the input size, and the class of problems solvable
in such a small amount of space is strictly contained in PSPACE (which is proved by a space hierarchy
theorem).
634 Conceptual Data Modeling and Database Design: A Fully Algorithmic Approach

The complexity class EXPTIME (also called EXP or DEXPTIME) is

the set of all decision problems solvable by a DTM in O(2p(n)) time, where
p(n) is a polynomial function of n.
A decision problem is EXPTIME-complete if it is in EXPTIME and
every problem in EXPTIME has a polynomial-time many-one reduction
to it (i.e., there is a polynomial-time algorithm that transforms instances
of one to instances of the other with the same answer). Problems that are
EXPTIME-complete might be thought of as the hardest problems in EXP-
TIME. Notice that although we don’t know if NP is equal to P or not,
we do know that EXPTIME-complete problems are not in P; it has been
proven that these problems cannot be solved in polynomial time (by the
time hierarchy theorem).
One of the basic undecidable problems is that of deciding whether a
DTM halts. One of the most fundamental EXPTIME-complete problems
is a simpler version of this, which asks if a DTM halts in at most k steps:
it is in EXPTIME because a trivial simulation requires O(k) time, and the
input k is encoded using O(log k) bits; it is EXPTIME-complete because,
roughly speaking, we can use it to determine if a machine solving an EXP-
TIME problem accepts in an exponential number of steps; it will not use
more.145
We know that P # NP # PSPACE # EXPTIME # NEXPTIME #
EXPSPACE, with146 P , EXPTIME ` P ≠ EXPTIME, NP , NEXPTIME
` NP ≠ NEXPTIME, and PSPACE , EXPSPACE ` PSPACE ≠ EXP-
SPACE, which means that at least one of the first three inclusions and at
least one of the last three inclusions must be proper, but it is not known
which ones are. Most experts believe all the inclusions are proper.
It is also known that if P = NP, then EXPTIME = NEXPTIME.

145
Recall that the same problem with the number of steps written in unary is P-complete.
146
By the time hierarchy theorem and the space hierarchy theorem.