Advanced C Programming

Declarations, External Names, Memory Layout

Sebastian Hack
[email protected]
Christoph Weidenbach
[email protected]

04.11.2008

saarland
university
computer science

1
Overview
Declarations
Properties of Declarations
Storage Classes
Type Qualifiers
Declarators

External Names
Linkage Models

C++ Compatibility
Memory Layout (Linux)

Literature
I Harbison & Steele, C — A Reference Manual, Chapter 4
I ISO/IEC 9899:1999, Chapter 6
I C Design Rationale, Chapter 6.2.2
See course website for links

2
Declarations

I Associate identifier with C object

I C objects:
I variable
I function
I type
I type tag
I structure and union components
I enum constants
I labels (for goto)
I preprocessor macros

3
Structure of Declarations
A declaration in C consists of
I Storage class specifier: extern, static, auto, register

I For syntactical reasons, typedef is also a storage class specifier

I Type qualifiers: const, volatile, restrict (C99)
I redundant occurrences are error in C89 but not in C99!
I Type specifiers: unsigned, signed, char, int, . . .
I C89: missing type specifier equals to int
I Declarator
I Can be left out in certain cases
I considered bad style, so we don’t elaborate on it
I Initializer (One or none)
Example
unsigned volatile long extern int const j ;
extern const volatile unsigned long int i = 3;

Convention
Use following order: storage class, qualifier, specifier
4
Attributes of Declarations

Each declaration defines several attributes of the declared object:

Scope Range in the program text where the object’s identifier is
declared
Visibility Range in the program text where the declared object can
be accessed with its identifier
Name Space Which kinds of objects must have distinct names if they
shall be referenced at the same time in the same scope
Extent The lifetime of the object during program runtime
Linkage Is the object visible from other translation units?

5
Visibility
I One declaration can hide another
int foo = 10;
int main ( void ) {
float foo ; /* this foo hides outer foo */
...
}

I Where does the hiding start?

{
int i = 0;
{
int j = i ;
int i = 10;
}
}

j == 0 or j == 10?

6
Visibility
I One declaration can hide another
int foo = 10;
int main ( void ) {
float foo ; /* this foo hides outer foo */
...
}

I Where does the hiding start?

{
int i = 0;
{
int j = i ;
int i = 10;
}
}

j == 0 or j == 10?

Rule
Scope starts at declaration point, not at start of enclosing block

6
Name Spaces

I The same identifier can declare different kinds of objects at the

same time (aka overloading)
I These object have to be in different name spaces
(overloading classes)
I C defines the following
I Proprocessor macro names
I goto labels
I struct, union, and enum tags
I Names of components of structs and unions
I The rest: variables, functions, typedef names

7
Name Spaces

I The same identifier can declare different kinds of objects at the

same time (aka overloading)
I These object have to be in different name spaces
(overloading classes)
I C defines the following
I Proprocessor macro names
I goto labels
I struct, union, and enum tags
I Names of components of structs and unions
I The rest: variables, functions, typedef names
I Example
extern int howmany ; /* rest name space */
extern char str [10]; /* rest */
typedef double howmany (); /* rest : conflict ! */
extern struct str { int a , b ; } x ; /* tag : no conflict */

7
Extent

I Lifetime of object at runtime

I Static extent
I Storage allocated before program start
I Storage remains allocated until program ends
I All functions, top-level declared variables, and local variables
declared static or extern have static extent

I Local extent
I Created on entry to a block or function
I Destroyed at block’s (function’s) exit
I Re-created each time block/function is entered

8
Storage Classes
auto and register

auto (local variables)

I Cannot be used for global variables
I Seldom used explicitly
I Will have a revival in the new C++0x standard

register (local variables and function parameters)

I Equivalent to auto but:
I Hint for the compiler that variable is used frequently
I Nowadays, rarely used
I Modern register allocation is powerful enough

9
Storage Classes
extern and static
extern
I Static extent
I External Linkage
I Variables: non-defining declaration:
no memory will be allocated for the variable
I Functions: Default for top-level defined functions

static
I Static extent
I Internal linkage
I Variables: tentative declaration:
If no initializer is given, then variable will be
initialized to 0

Attention!
Note that top-level defined variables without storage class are not
extern. They have external linkage, but that is not identical to extern
10
Type Qualifiers
const
I Helps you:
avoid unintentional write to data that should not be written to
I Helps the compiler:
Can optimize memory access because it knows that const variables
cannot be modified
I Pay attention to pointer rules:
int * const const_pointer ;
const int * p o i n t e r _ t o _ c o n s t ;

I Never cast const variables to non-const ones

+ write access leads to undefined behavior
I Example
int *p , i ;
const int * pc , ic ;
pc = p = & i ; /* ok */ pc = & ic ; /* ok */
* p = 5; /* ok */ * pc = 5; /* invalid */
p = & ic ; /* invalid */
p = pc ; /* invalid */
p = ( int *) & ic ; /* works , but dangerous */

11
Type Qualifiers
const — Usage Example
I Use const for getters
struct coord {
int x , y ;
}

int coord_set_x ( struct coord *c , int x ) {

c->x = x;
}

int coord_get_x ( const struct coord * c ) {

return c - > x ;
}

12
Type Qualifiers
const — Usage Example
I Use const for getters
struct coord {
int x , y ;
}

int coord_set_x ( struct coord *c , int x ) {

c->x = x;
}

int coord_get_x ( const struct coord * c ) {

return c - > x ;
}

Rules
I Understand usage of const
I Use const where ever possible
I Never de-const-ify code
I Never cast const pointer to non-const pointer
12
Type Qualifiers
volatile
I Important for concurrency (software and hardware!)
I Usually, the compiler has some freedom where to store the contents
of variables
I Dependent on this storage location, concurrent updates might be
seen or not
I Example

int flag ; I Flag modified by another

void foo ( void ) { thread between two accesses
if ( flag )
do_something ; I Assume compiler keeps flag
in register in foo
/* flag modified
by another thread */
I Reasonable optimization to
save memory accesses
if ( flag )
d o _ a n o t h e r _ th i n g ;
I Concurrent update invisible!
} I When should contents of a
variable be visible to other
threads?
13
Type Qualifiers
volatile — Sequence Points

I When do the effects on volatile variables need to be visible?

I C standard defines so-called sequence points
I Between those sequence points volatile variables are not
synchronized with memory
I Basically, after each statement
I But not within (non-short-circuit) expressions
I Another argument to not have side effects in expressions
I See Annex C of C standard

14
Type Qualifiers
restrict (C99)

I Can only be used with pointers

I Annotation to help the compiler
I Helps memory disambiguation (later in the course)
I Example:
void add ( int n , int * restrict a , int * restrict b ) {
int i ;

for ( i = 0; i < n ; i ++)

a [ i ] += b [ i ];
}

I Inside add, the compiler can assume that arrays a and b do not
overlap
I If they do, the behavior may be undefined

15
Declarators
Overview

I C declarators can be hard to read:

int *(*(*(* x )())[10])();

I Rationale: Look like the use of the declared variable

I 2 golden rules:
1. Go from inner to outer
2. Arrays and functions have higher priority that pointers
I Example:
int (* x )[5]; /* Pointer to an array of 5 ints */
int * x [5]; /* Array of five pointers to ints */

I More examples:
int *(*(* fp1 )( int ))[10];
float (*(* b ())[])();
void *(* c )( char , int (*)());
...

16
Declarators

I Do not use complicated declarators

I Use typedefs to break them into pieces
const char *(*(* x )[10])( void *);

I Is a pointer to an array of pointers to functions, which take a void

pointer and return a string
I Write:
typedef const char *(* printer_t )( void *);
typedef printer_t printers_t [10];
printers_t * x ;

I You must be able to read array and function pointer typedefs

17
Initializations
Guidelines
I Separate declaration and initialization
I Multiple, comma-separated initializations are hard to read
I Avoid visibility problems (see earlier slides)

I Do not initialize eagerly

Not good Good

int x = 0; int x ;
/* x not used here */ /* x not used here */
x = y + 1; x = y + 1;

I Compiler (with -Wall) will tell you if variable is potentially undefined

I Limit scope as much as possible:
Not good Good
int x = 0; if (...) {
if (...) { int x ;
x = f (); x = f ();
... ...
printf ( " % d " , x ); printf ( " % d " , x );
} }

18
Implicit Declarations
I Usually, all identifiers have to be declared before they are used
I In C89 there is one exception that can lead to hard-to-find bugs
void f ( void ) {
g (2.718);
}

void g ( int x ) {
printf ( " % d \ n " , x );
}

I Will print garbage: depending on endianess, the lower or higher

32 bits of the double 2.718
I If function prototype not given before call
int func ();

is assumed
I Prototype does not describe the function but how it is called!
I Thus: Always provide correct prototype

19
External Names
I How to make objects visible/hidden to other translation units?
I Easy for functions:
I Give static for local linkage
I Give or omit extern for external linkage
I Whole program needs exactly one definition for a (used!) function in
one of the translation units

I More complicated for variables:

I static imposes local linkage
I Else we have external linkage
I Giving extern or not makes a difference!
I Remember: External linkage does not require extern

Major Question
Which declaration of a global-linkage variable creates storage?
+ There are four models (!) and the standard

20
External Names
Linkage Models
Common
I All declarations with external linkage (no matter if extern or not)
create storage.
I The linker puts all definitions of the same name to the same address
I Named after FORTRAN common zones

Relaxed Ref/Def
I Declarations with extern are pure references
+ no storage allocated
I Definitions are declarations without storage class
I In all translation units, at least one definition must exist
I Referencing declarations of unused vars may be ignored

Strict Ref/Def
I Like relaxed Ref/Def, but exactly one definition must exist

Initialization
I Only declarations that initialize the variable create storage
21
Linkage Models !"#$%"$&'
Overview (from C99 Design Rationale, Chapter 6.2.2)
'
*+$%,&'-./0'1234",+52#'26'+7&#8+6+&,'9+#:"$&'327&95'
;27&9' *+9&'/' *+9&'<'
12332#' !"#!$%&'%#&'(& !"#!$%&'%#&'(&
'%#&)*'%+&,& 52'3&0!12%3+&,&
-& -&
&&&&'&.&/(& &&&&#6'$3+&'&,(&
&&&&0!12%3+&,(& 4&
4&
=&9">&7'=&6?@&6' '%#&'(& '%#&'(&
'%#&)*'%+&,& 52'3&0!12%3+&,&
-& -&
&&&&'&.&/(& &&&&#6'$3+&'&,(&
&&&&0!12%3+&,(& 4&
4&
A8,+B8'=&6?@&6' '%#&'(& !"#!$%&'%#&'(&
'%#&)*'%+&,& 52'3&0!12%3+&,&
-& -&
&&&&'&.&/(& &&&&#6'$3+&'&,(&
&&&&0!12%3+&,(& 4&
4& &
C#+8+"9+D&,' '%#&'&.&7(& '%#&'(&
'%#&)*'%+&,& 52'3&0!12%3+&,&
-& -&
&&&&'&.&/(& &&&&#6'$3+&'&,(&
&&&&0!12%3+&,(& 4&
4&

!"#"$% &'()%*+',)*%-.%/0)12/./)3*% 22
Linkage Models
The Standard
I Combination of strict Ref/Def and Initialization
I Only one file has definition
I Definition is declaration without storage class specifier or extern
with initializer
I Having multiple definitions causes undefined behavior!
(does not mean that you get an error message!)

23
Linkage Models
The Standard
I Combination of strict Ref/Def and Initialization
I Only one file has definition
I Definition is declaration without storage class specifier or extern
with initializer
I Having multiple definitions causes undefined behavior!
(does not mean that you get an error message!)

Conclusion
I False assumption on linkage model can be source of bugs!
I gcc under Linux does not use the standard model, but the UNIX one
I Do not rely on that when you want to write portable code!

23
Linkage Models
The Standard
I Combination of strict Ref/Def and Initialization
I Only one file has definition
I Definition is declaration without storage class specifier or extern
with initializer
I Having multiple definitions causes undefined behavior!
(does not mean that you get an error message!)

Conclusion
I False assumption on linkage model can be source of bugs!
I gcc under Linux does not use the standard model, but the UNIX one
I Do not rely on that when you want to write portable code!

Guideline
I Use strict Ref/Def
I Exactly one file with definition (declaration without storage class)
I All other declarations use extern
23
C++ Compatibility
I Your C code might be used by a C++ project. Be prepared for that.
I Biggest issue: C++ mangles function names
extern int my_func ( double , const char *);

int main () {
return my_func (2.345 , " Hallo " );
}

I Mangled symbol name: __Z7my_funcdPKc (GCC 4.0.1)

I Just including a C header will apply mangling to C declarations
I Linker will not be able to find symbols
I Surround declarations in .h files with
# ifdef __cplusplus
extern " C " {
# endif
/* Here go the C declarations */
# ifdef __cplusplus
}
# endif

I Then, the C++ compiler knows that those are C declarations

24
Memory Layout (Unix/Linux)

I When executed, the memory of a C program is composed into

several segments
I Text
I Executable code
I Might be read-only to forbid accidental self-modification

I Initialized Data
I global/local linkage data that has been initialized
I can be set read-only for const variables

I Uninitialized Data (BSS)

I global/local linkage data that has not been initialized
I Is initialized with 0 by the kernel at load time
I No space in the binary needs to be wasted

25
Memory Layout (Unix/Linux)
. . . cont’d
I Heap
I Dynamically allocated data (malloc)
I Usually grows upwards
I Stack
I auto variables
I stack frames
I spilled registers
I usually grows downwards

I Know where the segments start

I Gives you an idea where your pointers point to
I Example:
I You debug and some pointer 0xe502f segfaults
I This address is strange (below data, heap, and stack)
I You must have overwritten the pointer’s contents

26
Memory Layout (Unix/Linux)
. . . cont’d
I Heap
I Dynamically allocated data (malloc)
I Usually grows upwards
I Stack
I auto variables
I stack frames
I spilled registers
I usually grows downwards

I Know where the segments start

I Gives you an idea where your pointers point to
I Example:
I You debug and some pointer 0xe502f segfaults
I This address is strange (below data, heap, and stack)
I You must have overwritten the pointer’s contents

Homework
Write a small program that prints the addresses of the segments
26