Fundamentals of

C++
Programming
A F T
DR
Richard L. Halterman
School of Computing
Southern.Adventist University

January 27, 2013

Copyright © 2013 Richard L. Halterman. All rights reserved.
 i

Contents

1 The Context of Software Development 1

1.1 Software . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.2 Development Tools . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.3 Learning Programming with C++ . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.4 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.5 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6

2 Writing a C++ Program 7

2.1 General Structure of a Simple C++ Program . . . . . . . . . . . . . . . . . . . . . . . . . 7
2.2 Compiling the source code . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
2.3 Variations of our simple program . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
2.4 Template for simple C++ programs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
2.5 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
2.6 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19

3 Values and Variables 21

3.1 Integer Values . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
3.2 Variables and Assignment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
3.3 Identifiers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
3.4 Floating-point Types . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
3.5 Constants . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
3.6 Other Numeric Types . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
3.7 Characters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
3.8 Enumerated Types . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
3.9 Type Inference with auto . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
3.10 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36

©2013 Richard L. Halterman Draft date: January 27, 2013

CONTENTS ii

3.11 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38

4 Expressions and Arithmetic 41

4.1 Expressions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
4.2 Mixed Type Expressions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
4.3 Operator Precedence and Associativity . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
4.4 Comments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
4.5 Formatting . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
4.6 Errors and Warnings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
4.6.1 Compile-time Errors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
4.6.2 Run-time Errors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
4.6.3 Logic Errors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
4.6.4 Compiler Warnings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
4.7 Arithmetic Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
4.8 More Arithmetic Operators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
4.9 Algorithms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
4.10 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68
4.11 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70

5 Conditional Execution 75
5.1 Type bool . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
5.2 Boolean Expressions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76
5.3 The Simple if Statement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78
5.4 Compound Statements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
5.5 The if/else Statement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82
5.6 Compound Boolean Expressions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86
5.7 Nested Conditionals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89
5.8 Multi-way if/else Statements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97
5.9 Errors in Conditional Statements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101
5.10 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102
5.11 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103

6 Iteration 107
6.1 The while Statement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107
6.2 Nested Loops . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 117

©2013 Richard L. Halterman Draft date: January 27, 2013

CONTENTS iii

6.3 Abnormal Loop Termination . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 126

6.3.1 The break statement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 126
6.3.2 The goto Statement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127
6.3.3 The continue Statement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 129
6.4 Infinite Loops . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 130
6.5 Iteration Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 134
6.5.1 Drawing a Tree . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 134
6.5.2 Printing Prime Numbers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 137
6.6 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 140
6.7 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 141

7 Other Conditional and Iterative Statements 147

7.1 The switch Statement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 147
7.2 The Conditional Operator . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 151
7.3 The do/while Statement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 152
7.4 The for Statement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 155
7.5 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 161
7.6 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 162

8 Using Functions 167

8.1 Introduction to Using Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 168
8.2 Standard Math Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 173
8.3 Maximum and Minimum . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 176
8.4 clock Function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 177
8.5 Character Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 179
8.6 Random Numbers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 180
8.7 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 184
8.8 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 185

9 Writing Functions 187

9.1 Function Basics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 188
9.2 Using Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 196
9.3 Call by Value . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 201
9.4 Function Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 202
9.4.1 Better Organized Prime Generator . . . . . . . . . . . . . . . . . . . . . . . . . . 202

©2013 Richard L. Halterman Draft date: January 27, 2013

CONTENTS iv

9.4.2 Command Interpreter . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 204

9.4.3 Restricted Input . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 206
9.4.4 Better Die Rolling Simulator . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 207
9.4.5 Tree Drawing Function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 209
9.4.6 Floating-point Equality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 210
9.4.7 Multiplication Table with Functions . . . . . . . . . . . . . . . . . . . . . . . . . 213
9.5 Commenting Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 216
9.6 Custom Functions vs. Standard Functions . . . . . . . . . . . . . . . . . . . . . . . . . . 218
9.7 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 220
9.8 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 221

10 More on Functions 227

10.1 Global Variables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 227
10.2 Static Variables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 235
10.3 Overloaded Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 238
10.4 Recursion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 239
10.5 Making Functions Reusable . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 244
10.6 Pointers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 249
10.7 Call by Reference . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 254
10.7.1 Call by Reference via Pointers . . . . . . . . . . . . . . . . . . . . . . . . . . . . 255
10.7.2 Call by Reference via References . . . . . . . . . . . . . . . . . . . . . . . . . . 257
10.8 Function Pointers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 258
10.9 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 260
10.10Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 261

11 Arrays 269
11.1 Declaring and Using Arrays . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 271
11.2 Arrays and Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 279
11.3 Prime Generation with an Array . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 282
11.4 Multidimensional Arrays . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 284
11.5 Pointers and Arrays . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 287
11.6 Array Ranges . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 291
11.7 C Strings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 292
11.8 Dynamic Arrays . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 295
11.9 The sizeof Operator . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 300

©2013 Richard L. Halterman Draft date: January 27, 2013

CONTENTS v

11.10Copying an Array . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 302

11.11Memory Management . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 303
11.12Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 307
11.13Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 308

12 Array Algorithms 313

12.1 Sorting . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 313
12.2 Flexible Sorting . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 316
12.3 Search . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 319
12.3.1 Linear Search . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 319
12.3.2 Binary Search . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 322
12.4 Array Permutations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 330
12.5 Randomly Permuting an Array . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 337
12.6 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 344
12.7 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 344

13 Standard C++ Classes 347

13.1 String Objects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 348
13.2 Vectors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 351
13.2.1 Declaring and Using Vectors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 352
13.2.2 Copying Vectors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 354
13.2.3 Vectors and Arrays . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 360
13.3 Input/Output Streams . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 362
13.4 File Streams . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 366
13.5 Complex Numbers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 370
13.6 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 371
13.7 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 373

14 Custom Objects 375

14.1 Object Basics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 375
14.2 Fields . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 377
14.3 Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 383
14.4 Constructors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 390
14.5 Destructors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 392
14.6 Defining a New Numeric Type . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 394

©2013 Richard L. Halterman Draft date: January 27, 2013

CONTENTS vi

14.7 Encapsulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 396

14.8 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 399
14.9 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 400

15 More on Objects 407

15.1 Pointers to Objects and Object Arrays . . . . . . . . . . . . . . . . . . . . . . . . . . . . 407
15.2 The this Pointer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 410
15.3 const Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 412
15.4 Separating Method Declarations and Definitions . . . . . . . . . . . . . . . . . . . . . . . 414
15.5 Preventing Multiple Inclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 421
15.6 Overloaded Operators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 424
15.6.1 Operator Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 424
15.6.2 Operator Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 428
15.7 static Members . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 429
15.8 Classes vs. structs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 433
15.9 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 434
15.10Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 435

16 Building some Useful Classes 437

16.1 A Better Rational Number Class . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 437
16.2 Stopwatch . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 440
16.3 Sorting with Logging . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 446
16.4 Linked Lists . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 450
16.5 Automating Testing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 461
16.6 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 465
16.7 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 465

17 Inheritance and Polymorphism 467

17.1 I/O Stream Inheritance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 467
17.2 Inheritance Mechanics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 469
17.3 Uses of Inheritance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 472
17.4 Polymorphism . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 479
17.5 Protected Members . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 485
17.6 Fine Tuning Inheritance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 494
17.7 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 505

©2013 Richard L. Halterman Draft date: January 27, 2013

CONTENTS vii

17.8 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 506

18 Generic Programming 509

18.1 Function Templates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 509
18.2 Class Templates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 516
18.3 The Standard Template Library . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 526
18.4 Iterators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 526
18.5 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 528
18.6 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 528

19 Exception Handling 529

19.1 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 529
19.2 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 529

Appendices 533

A Visual C++ Command Line Development 533

B Developing C++ Programs under Unix/Linux with the GNU Tools 537

C Introduction to Binary Numbers and Computer Arithmetic 539

C.1 The Integer Types . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 539
C.2 The Floating-Point Types . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 544

D Bitwise Operators 549

Bibliography 553

Index 554

©2013 Richard L. Halterman Draft date: January 27, 2013

CONTENTS viii

©2013 Richard L. Halterman Draft date: January 27, 2013

 ix

Preface

Legal Notices and Information

Permission is hereby granted to make hardcopies and freely distribute the material herein under the
following conditions:

• The copyright and this legal notice must appear in any copies of this document made in whole or in
part.
• None of material herein can be sold or otherwise distributed for commercial purposes without written
permission of the copyright holder.

• Instructors at any educational institution may freely use this document in their classes as a primary
or optional textbook under the conditions specified above.

A local electronic copy of this document may be made under the terms specified for hardcopies:

• The copyright and these terms of use must appear in any electronic representation of this document
made in whole or in part.
• None of material herein can be sold or otherwise distributed in an electronic form for commercial
purposes without written permission of the copyright holder.

• Instructors at any educational institution may freely store this document in electronic form on a local
server as a primary or optional textbook under the conditions specified above.

Additionally, a hardcopy or a local electronic copy must contain the uniform resource locator (URL)
providing a link to the original content so the reader can check for updated and corrected content. The
current standard URL is http://python.cs.southern.edu/cppbook/html.

©2013 Richard L. Halterman Draft date: January 27, 2013

 1

Chapter 1

The Context of Software Development

A computer program, from one perspective, is a sequence of instructions that dictate the flow of electrical
impulses within a computer system. These impulses affect the computer’s memory and interact with the
display screen, keyboard, and mouse in such a way as to produce the “magic” that permits humans to
perform useful tasks, solve high-level problems, and play games. One program allows a computer to assume
the role of a financial calculator, while another transforms the machine into a worthy chess opponent. Note
the two extremes here:

• at the lower, more concrete level electrical impulses alter the internal state of the computer, while

• at the higher, more abstract level computer users accomplish real-world work or derive actual plea-
sure.

So well is the higher-level illusion achieved that most computer users are oblivious to the lower-level
activity (the machinery under the hood, so to speak). Surprisingly, perhaps, most programmers today write
software at this higher, more abstract level also. An accomplished computer programmer can develop
sophisticated software with little or no interest or knowledge of the actual computer system upon which it
runs. Powerful software construction tools hide the lower-level details from programmers, allowing them
to solve problems in higher-level terms.
The concepts of computer programming are logical and mathematical in nature. In theory, computer
programs can be developed without the use of a computer. Programmers can discuss the viability of a
program and reason about its correctness and efficiency by examining abstract symbols that correspond
to the features of real-world programming languages but appear in no real-world programming language.
While such exercises can be very valuable, in practice computer programmers are not isolated from their
machines. Software is written to be used on real computer systems. Computing professionals known
as software engineers develop software to drive particular systems. These systems are defined by their
underlying hardware and operating system. Developers use concrete tools like compilers, debuggers, and
profilers. This chapter examines the context of software development, including computer systems and
tools.

©2013 Richard L. Halterman Draft date: January 27, 2013

1.1. SOFTWARE 2

1.1 Software

A computer program is an example of computer software. Software makes a computer a truly universal
machine transforming it into the proper tool for the task at hand. One can refer to a program as a piece of
software as if it were a tangible object, but software is actually quite intangible. It is stored on a medium. A
hard drive, a CD, a DVD, and a USB pen drive are all examples of media upon which software can reside.
The CD is not the software; the software is a pattern on the CD. In order to be used, software must be stored
in the computer’s memory. Typically computer programs are loaded into memory from a medium like the
computer’s hard disk. An electromagnetic pattern representing the program is stored on the computer’s hard
drive. This pattern of electronic symbols must be transferred to the computer’s memory before the program
can be executed. The program may have been installed on the hard disk from a CD or from the Internet. In
any case, the essence that was transferred from medium to medium was a pattern of electronic symbols that
direct the work of the computer system.
These patterns of electronic symbols are best represented as a sequence of zeroes and ones, digits from
the binary (base 2) number system. An example of a binary program sequence is

10001011011000010001000001001110

To the underlying computer hardware, specifically the processor, a zero here and three ones there might
mean that certain electrical signals should be sent to the graphics device so that it makes a certain part of
the display screen red. Unfortunately, only a minuscule number of people in the world would be able to
produce, by hand, the complete sequence of zeroes and ones that represent the program Microsoft Word
for an Intel-based computer running the Windows 8 operating system. Further, almost none of those who
could produce the binary sequence would claim to enjoy the task.
The Word program for older Mac OS X computers using a PowerPC processor works similarly to the
Windows version and indeed is produced by the same company, but the program is expressed in a com-
pletely different sequence of zeroes and ones! The Intel Core i7 processor in the Windows machine accepts
a completely different binary language than the PowerPC processor in the Mac. We say the processors have
their own machine language.

1.2 Development Tools

If very few humans can (or want) to speak the machine language of the computers’ processors and software
is expressed in this language, how has so much software been developed over the years?
Software can be represented by printed words and symbols that are easier for humans to manage than
binary sequences. Tools exist that automatically convert a higher-level description of what is to be done
into the required lower-level code. Higher-level programming languages like C++ allow programmers to
express solutions to programming problems in terms that are much closer to a natural language like English.
Some examples of the more popular of the hundreds of higher-level programming languages that have been
devised over the past 60 years include FORTRAN, COBOL, Lisp, Haskell, C, Perl, Python, Java, and C#.
Most programmers today, especially those concerned with high-level applications, usually do not worry
about the details of underlying hardware platform and its machine language.
One might think that ideally such a conversion tool would accept a description in a natural language,
such as English, and produce the desired executable code. This is not possible today because natural
languages are quite complex compared to computer programming languages. Programs called compilers
that translate one computer language into another have been around for over 60 years, but natural language

©2013 Richard L. Halterman Draft date: January 27, 2013

1.2. DEVELOPMENT TOOLS 3

processing is still an active area of artificial intelligence research. Natural languages, as they are used
by most humans, are inherently ambiguous. To understand properly all but a very limited subset of a
natural language, a human (or artificially intelligent computer system) requires a vast amount of background
knowledge that is beyond the capabilities of today’s software. Fortunately, programming languages provide
a relatively simple structure with very strict rules for forming statements that can express a solution to any
program that can be solved by a computer.
Consider the following program fragment written in the C++ programming language:
subtotal = 25;
tax = 3;
total = subtotal + tax;

These three lines do not make up a complete C++ program; they are merely a piece of a program. The
statements in this program fragment look similar to expressions in algebra. We see no sequence of bi-
nary digits. Three words, subtotal, tax, and total, called variables, are used to hold information.
Mathematicians have used variables for hundreds of years before the first digital computer was built. In
programming, a variable represents a value stored in the computer’s memory. Familiar operators (= and +)
are used instead of some cryptic binary digit sequence that instructs the processor to perform the operation.
Since this program is expressed in the C++ language, not machine language, it cannot be executed directly
on any processor. A C++ compiler is used to translate the C++ code into machine code.
The higher-level language code is called source code. The compiled machine language code is called
the target code. The compiler translates the source code into the target machine language.
The beauty of higher-level languages is this: the same C++ source code can be compiled to different
target platforms. The target platform must have a C++ compiler available. Minor changes in the source code
may be required because of architectural differences in the platforms, but the work to move the program
from one platform to another is far less than would be necessary if the program for the new platform had
to be rewritten by hand in the new machine language. Just as importantly, when writing the program the
human programmer is free to think about writing the solution to the problem in C++, not in a specific
machine language.
Programmers have a variety of tools available to enhance the software development process. Some
common tools include:

• Editors. An editor allows the user to enter the program source code and save it to files. Most pro-
gramming editors increase programmer productivity by using colors to highlight language features.
The syntax of a language refers to the way pieces of the language are arranged to make well-formed
sentences. To illustrate, the sentence

The tall boy runs quickly to the door.

uses proper English syntax. By comparison, the sentence

Boy the tall runs door to quickly the.

is not correct syntactically. It uses the same words as the original sentence, but their arrangement
does not follow the rules of English.
Similarly, programming languages have strict syntax rules that must be followed to create well-
formed programs. Only well-formed programs are acceptable and can be compiled and executed.
Some syntax-aware editors can use colors or other special annotations to alert programmers of syntax
errors before the program is compiled.

©2013 Richard L. Halterman Draft date: January 27, 2013

1.2. DEVELOPMENT TOOLS 4

Concept of
(Design
problem
program logic)
solution

Editor (Edit) Programmer’s

responsibility

#include <io

Source code
using namespace
std;
int main()
{
srand(23);
int n;
n = rand();

Library proc(n);

declarations
(source code)

Preprocessor
istream cin;
ostream cout;
int rand();
void sand();
(Preprocess)
typedef unsigned U
#define NULL (0)

#include <io

Enhanced using namespace

std;
int main()
{

source code srand(23);

int n;
n = rand();
proc(n);

Automated
Compiler (Compile)
by tools

Object code 101100010101

000010001100
1100001111010100
0011101101110011
1000000010000110
0111000000111111
1100111011001001

Pre-compiled 0000100001111000
0001110111101101
1101111011111010

libraries
(object code)
11000011110
00111011011
1000000010000110
0111000000111111
1100111011001001
Linker (Link)
0000100001111000
0001110111101101
1101111011111010

101100010101
000010001100

Executable 1100001111010100
0011101101110011
1000000010000110
0111000000111111

program 1100111011001001
0000100001111000
0001110111101101
1101111011111010

Figure 1.1: Source code to target code sequence

• Compilers. A compiler translates the source code to target code. The target code may be the machine
language for a particular platform or embedded device. The target code could be another source
language; for example, the earliest C++ compiler translated C++ into C, another higher-level language.
The resulting C code was then processed by a C compiler to produce an executable program. C++
compilers today translate C++ directly into machine language.
The complete set of build tools for C++ includes a preprocessor, compiler, and linker:

– Preprocessor—adds to or modifies the contents of the source file before the compiler begins
processing the code. We use the services of the preprocessor mainly to #include information
about library routines our programs use.
– Compiler—translates C++ source code to machine code.
– Linker—combines the compiler-generated machine code with precompiled library code or
compiled code from other sources to make a complete executable program. Most compiled
C++ code is incapable of running by itself and needs some additional machine code to make a
complete executable program. The missing machine code has been precompiled and stored in
a repository of code called a library. A program called a linker combines the programmer’s
compiled code and the library code to make a complete program.

©2013 Richard L. Halterman Draft date: January 27, 2013

1.3. LEARNING PROGRAMMING WITH C++ 5

We generally do not think about the preprocessor, compiler, and linker working as three separate
programs (although they do), because the tools we use make it appear as only one process is taking
place: translating our source code to an executable program.

• Debuggers. A debugger allows a programmer to more easily trace a program’s execution in order
to locate and correct errors in the program’s implementation. With a debugger, a developer can
simultaneously run a program and see which line in the source code is responsible for the program’s
current actions. The programmer can watch the values of variables and other program elements to see
if their values change as expected. Debuggers are valuable for locating errors (also called bugs) and
repairing programs that contain errors. (See Section 4.6 for more information about programming
errors.)

• Profilers. A profiler collects statistics about a program’s execution allowing developers to tune ap-
propriate parts of the program to improve its overall performance. A profiler indicates how many
times a portion of a program is executed during a particular run, and how long that portion takes to
execute. Profilers also can be used for testing purposes to ensure all the code in a program is actually
being used somewhere during testing. This is known as coverage. It is common for software to fail
after its release because users exercise some part of the program that was not executed anytime during
testing. The main purpose of profiling is to find the parts of a program that can be improved to make
the program run faster.

The programming components of the development process are illustrated in Figure 1.1.
Many developers use integrated development environments (IDEs). An IDE includes editors, debug-
gers, and other programming aids in one comprehensive program. Examples of IDEs for C++ include
Microsoft’s Visual Studio 2012, the Eclipse Foundation’s Eclipse CDT, and Apple’s XCode.
Despite the plethora of tools (and tool vendors’ claims), the programming process for all but trivial
programs is not automatic. Good tools are valuable and certainly increase the productivity of developers,
but they cannot write software. There are no substitutes for sound logical thinking, creativity, common
sense, and, of course, programming experience.

1.3 Learning Programming with C++

Bjarne Stroustrup of AT&T Bell Labs created C++ in the mid 1980s. C++ is an extension of the programming
language C, a product of AT&T Bell Labs from the early 1970s. C was developed to write the Unix
operating system, and C is widely used for systems-level software and embedded systems development.
C++ initially provided object-oriented programming features (see Chapter 13 and Chapter 14) and later
added generic programming capabilities. C++’s close relationship to C allows C++ programs to utilize a
large collection of code developed in C.
C++ is widely used in industry for commercial software development. It is an industrial strength pro-
gramming language used for developing complex systems in business, science, and engineering. Examples
of software written in C++ include Microsoft Windows 8, Microsoft Office, Mac OS X, and Adobe Creative
Suite.
In order to meet the needs of commercial software development and accomplish all that it does, C++
itself is complex. While experienced programmers can accomplish great things with C++, beginners some-
times have a difficult time with it. Professional software developers enjoy the flexible design options that
C++ permits, but beginners need more structure and fewer options so they can master simpler concepts
before moving on to more complex ones.

©2013 Richard L. Halterman Draft date: January 27, 2013

1.4. SUMMARY 6

This book does not attempt to cover all the facets of the C++ programming language. Experienced
programmers should look elsewhere for books that cover C++ in much more detail. The focus here is on
introducing programming techniques and developing good habits. To that end, our approach avoids some of
the more esoteric features of C++ and concentrates on the programming basics that transfer directly to other
imperative programming languages such as Java, C#, and Python. We stick with the basics and explore
more advanced features of C++ only when necessary to handle the problem at hand.

1.4 Summary

• Computers require both hardware and software to operate. Software consists of instructions that
control the hardware.
• At the lowest level, the instructions for a computer program can be represented as a sequence of zeros
and ones. The pattern of zeros and ones determine the instructions performed by the processor.
• Two different kinds of processors can have different machine languages.
• Application software can be written largely without regard to the underlying hardware. A tool called
a compiler translates the higher-level, abstract language into the machine language required by the
hardware.
• Programmers develop software using tools such as editors, compilers, debuggers, and profilers.
• C++ is a higher-level programming language.
• An IDE is an integrated development environment—one program that provides all the tools that
developers need to write software.

1.5 Exercises

1. What is a compiler?
2. How is compiled code different from source code?
3. What tool does a programmer use to produce C++ source code?
4. What tool(s) does a programmer use to convert C++ source code into executable machine code?
5. What does the linker do?
6. Does the linker deal with files containing source code or or machine language code?
7. What does the preprocessor do to source code?
8. List several advantages developing software in a higher-level language has over developing software
in machine language.
9. How can an IDE improve a programmer’s productivity?
10. Name a popular C++ IDE is used by programmers developing for Microsoft Windows.
11. Name a popular C++ IDE is used by programmers developing for Apple Mac OS X.

©2013 Richard L. Halterman Draft date: January 27, 2013

 7

Chapter 2

Writing a C++ Program

Properly written C++ programs have a particular structure. The syntax must be correct, or the compiler
will generate error messages and not produce executable machine language. This chapter introduces C++
by providing some simple example programs and associated fundamental concepts. Most of the concepts
presented in this chapter are valid in many other programming languages as well. While other languages
may implement the concepts using slightly different syntax, the ideas are directly transferable to other
languages like C, Java, C#, and Ada.

2.1 General Structure of a Simple C++ Program

Listing 2.1 (simple.cpp) is one of the simplest C++ programs that does something:

Listing 2.1: simple.cpp

1 #include <iostream>
2
3 using namespace std;
4
5 int main()
6 {
7 cout << "This is a simple C++ program!" << endl;
8 }

The text as shown in Listing 2.1 (simple.cpp) may be typed in an editor and can saved to a file named
simple.cpp. The actual name of the file is irrelevant, but the name “simple” accurately describes the nature
of this program. The extension .cpp is a common extension used for C++ source code.
After creating this file with a text editor and compiling it, it prints the message

This is a simple C++ program!

Listing 2.1 (simple.cpp) contains six non-blank lines of code:

©2013 Richard L. Halterman Draft date: January 27, 2013

2.1. GENERAL STRUCTURE OF A SIMPLE C++ PROGRAM 8

• #include <iostream>
This line is a preprocessing directive. All preprocessing directives within C++ source code begin with
a # symbol. This one directs the preprocessor to add some predefined source code to our existing
source code before the compiler begins to process it. This process is done automatically and is
invisible to us.
Here we want to use some parts of the iostream library, a collection precompiled C++ code that
can be used by other C++ programs (like ours). The iostream library contains routines that handle
input and output (I/O) that include functions such as printing to the display, getting user input from
the keyboard, and dealing with files.
Two items used in Listing 2.1 (simple.cpp), cout and endl, are not part of the C++ language it-
self. These items, along with many other things related to input and output, were developed in
C++compiled, and stored in the iostream library. The compiler needs to be made aware of these
iostream items so it can compile our program. The #include directive specifies a file, called a
header, that contains the specifications for the library code. The compiler checks our code against
the specifications in the header to ensure that we are using the library code correctly.
Most of the programs we write use this #include <iostream> directive, and some programs
we will write in the future will #include other information as well.

• using namespace std;

The two items our program needs to display a message on the screen, cout and endl, have longer
names: std::cout and std::endl. This using namespace std directive allows us to
omit the std:: prefixes and use their shorter names. This directive is optional, but if it is omitted,
the longer names must be used. Listing 2.2 (simple2.cpp) shows how the longer names are used. The
name std stands for “standard,” and the using namespace std line indicates that some of the
names we use in our program are part of the so-called “standard namespace.”

• int main()
This specifies the real beginning of our program. Here we are declaring a function named main. All
C++ programs must contain this function to be executable. Details about the meaning of int and
the parentheses will appear in later chapters. More general information about functions appear in
Chapter 8 and Chapter 9.

• {
The open curly brace marks the beginning of the body of a function. The body of a function contains
the statements to be carried out when the function executes.

• cout << "This is a simple C++ program!"<< endl;

The body of our main function contains only one statement. This statement causes the message
This is a simple C++ program! to be printed on the screen. A statement is the fundamental unit of
execution in a C++ program. Functions contain statements that are compiled into executable code.
C++ has a variety of different kinds of statements, and the chapters that follow explore these various
kinds of statements. All statements in C++ end with a semicolon (;). A more detailed explanation of
this statement appears below.

• }
The close curly brace marks the end of the body of a function. Both the open curly brace and close
curly brace are required for every function definition.

©2013 Richard L. Halterman Draft date: January 27, 2013

2.2. COMPILING THE SOURCE CODE 9

Note which lines in the program end with a semicolon (;) and which do not. Do
not put a semicolon after the #include preprocessor directive. Do not put a
semicolon on the line containing main, and do not put semicolons after the curly
braces.

2.2 Compiling the source code

C++ compilers come in a variety of forms. Some, such as Microsoft’s Visual Studio 2012 and Apple’s
XCode tools provide IDEs (see Section 1.2) with many features. In this section we focus on building
C++ programs with Microsoft’s Visual Studio 2012. The Visual C++ Express Edition, which is more than
adequate for building and running the programs in this text, is available as a free download from Microsoft’s
Visual C++ Express website at http://www.microsoft.com/express/vc. More information about Visual
Studio 2012 can be found at Microsoft’s Visual Studio website, http://msdn.microsoft.com/en-us/
vstudio/default.aspx.
While we will be concentrating on Visual C++, the code in this text is not Visual C++ specific. The code
herein aims to be C++ standards compliant and platform independent.
The following describes the task of C++ software development under Visual Studio:

• To begin creating a C++ program, you must first launch Visual Studio 2012 from the Start menu or
other relevant shortcut. Figure 2.1 shows the start menu of a typical Windows 7 system and the Visual
Studio application tile in Windows 8. You quickly should see a splash screen similar to the one shown

Figure 2.1: Launching Visual Studio from the Windows 7 start menu or Windows 8 application tile and the
ensuing .

in Figure 2.1.
If you never before have run the Visual Studio application, you must wait a few moments while it

©2013 Richard L. Halterman Draft date: January 27, 2013

2.2. COMPILING THE SOURCE CODE 10

configures the development environment for you. At this point you will indicate that Visual C++ is
your preferred development language (not necessary if you are using the Visual C++ Express Edition).
Figure 2.2 shows what Visual Studio looks like when it is fully loaded and ready to use.

Figure 2.2: Visual Studio Ready for Use

• After Visual Studio has started, you begin the process of developing a C++ program by creating a
new project. As Figure 2.3 shows, you can create a new project by following the menu sequence:
File→New→Project

Figure 2.3: Creating a New Project

• In the dialog that results, shown on the right of Figure 2.3, you should choose the project type to be
Visual C++ in the left pane, and use the Win32 Console Application option in the center pane. In the
name field near the bottom of the dialog, enter a name for the project; we will use the name simple.
You may change the location to a different folder if you like, or even a different drive (such as a USB
pen drive). In this example, we chose to not change the default location provided by Visual Studio.
• When you select OK on the project creation dialog, a Win32 Application Wizard as shown on the left
of Figure 2.4 appears. At this point, the instructions in the dialog say “Click Finish from any window
to accept the current settings.” Do not select Finish; instead, select Next to continue. We have one
more key step to complete so our project is set up correctly.

©2013 Richard L. Halterman Draft date: January 27, 2013

2.2. COMPILING THE SOURCE CODE 11

Figure 2.4: Configuring the New Project

• In the subsequent Applications Settings dialog (see the right image in Figure 2.4), select the em-
phEmpty project checkbox. Also, uncheck the Security Development Lifecycle box. The dialog
should look like the right image in Figure 2.4 before you proceed. Choose Finish when you are ready
to continue.

• At this point, the Solution Explorer panel shows the structure of the newly created, albeit empty,
project. The left image in Figure 2.5 shows the newly populated Solution Explorer pane.

Figure 2.5: Adding a C++ Source File

• Right click on the simple element in the Solution Explorer. As shown on the right of in Figure 2.5,
select Add and New Item from the resulting pop-up menu.

• In the Add New Item dialog box, shown on the left in Figure 2.6, select C++F̃ile (.cpp) and enter a
name for the C++ file in the text field at the bottom. You need not add .cpp to the end of the name,
as Visual Studio will do that for you. The file here will be named simple.cpp. Press Add when done.

• As shown on the right in Figure 2.6, the Solution Explorer pane now shows the file simple.cpp,
and the large editor pane is ready for you to type in the source code for the program. The new source
file is initially empty.

• In the editor pane with the simple.cpp tab at the top, type in the source code for our simple C++
program. Figure 2.7 shows the completed code.

©2013 Richard L. Halterman Draft date: January 27, 2013

2.2. COMPILING THE SOURCE CODE 12

Figure 2.6: Ready to Edit the C++ Source File

Figure 2.7: Editor Pane

• You may save the source file at this point by selecting Save from the File menu or by pressing pressing
at the same time the Ctrl S keys. If you do not save your program, Visual Studio will prompt
you to do so before it builds and runs your program.
To run the program, select Debug→Start Without Debugging, as shown on the left of Figure 2.8.
Visual Studio will attempt to build the executable program. It will prompt you to save your file if you

Figure 2.8: Building and Running the Program

have not saved it or have made changes since the last time you saved it. The progress of the build
process is displayed is in the Output panel in the bottom window. One of the lines has

1>Compiling. . .

and another later has

©2013 Richard L. Halterman Draft date: January 27, 2013

2.2. COMPILING THE SOURCE CODE 13

1>Linking. . .

The last few lines are very important:

1>Build succeeded.
1>
1>Time Elapsed 00:00:02.98
========= Build: 1 succeeded, 0 failed, 0 up-to-date, 0 skipped ==========

These lines indicate the build was successful.

• A console window appears with the output of your program. The right image in Figure 2.8 shows
this text window. You can press any key on the keyboard to close the window. If the console window
does not appear, you have typographical errors in your program; return to the editor, fix the errors,
and try to run the program again.
• When you are finished with Visual Studio, select the File→Exit menu items as shown in Figure 2.9.

Figure 2.9: Exiting Visual Studio

These are the steps for writing a basic C++ program in Visual Studio 2012. While the steps initially may
seem complex and tedious, they will become natural after you have written a few C++ programs.
When the program was run, the Visual Studio environment created a console window in which to run
the program.

This is a simple C++ program!

Press any key to continue . . .

The first line of the output was printed by the program. The second line prompting the user to press any
key to continue is added by the Visual C++ run-time environment so the console window stays visible long
enough for the user to see the output. If you run the program from a the standard Windows command shell
(CMD.EXE, usually found in the Start menu under Accessories and named Command Prompt), only the
program’s output will appear and the “Press any key to continue . . .” message will not appear.
The following summarizes the steps you should follow when writing a C++ program using the Visual
Studio IDE:

©2013 Richard L. Halterman Draft date: January 27, 2013

 2.3. VARIATIONS OF OUR SIMPLE PROGRAM 14

1. In the main menu:

File→New→Project (or Ctrl Shift N )

2. Select “Win32 Console Project” and click “Next” to set the “Empty Project” option
3. In the Solution Explorer pane right click on “Source Files” and select
Add→New Item...
4. Select C++ File (.cpp) and enter the name of your file
5. Type in the source code for your program in the editor panel
6. Save your file:
File→Save filename (or Ctrl S )

7. In the main menu:

Debug→Start Without Debugging (or Ctrl F5 )

It is possible to develop C++ programs without using Visual Studio’s integrated development environ-
ment. Visual Studio comes with some additional tools for command-line development. Appendix A de-
scribes how you can edit the C++ source code with a standalone text editor and compile and run the program
from the Windows command prompt (CMD.EXE). Command-line development under Linux and Mac OS
X is covered in B. Some programmers prefer the freedom of using their favorite editor, and the standalone
tools can be used in scripts to automate the build process.

2.3 Variations of our simple program

The two items Listing 2.1 (simple.cpp) needed to display a message on the screen, cout and endl, have
longer names: std::cout and std::endl. The using namespace std directive allows us to
omit the std:: prefixes and use their shorter names. This directive is optional, but if it is omitted, the
longer names must be used. For example, Listing 2.2 (simple2.cpp) shows an alternative way of writing
Listing 2.1 (simple.cpp).

Listing 2.2: simple2.cpp

1 #include <iostream>
2
3 int main()
4 {
5 std::cout << "This is a simple C++ program!" << std::endl;
6 }

Listing 2.3 (simple3.cpp) shows another way to use the shorter names for cout and endl within a C++
program.

Listing 2.3: simple3.cpp

1 #include <iostream>
2
3 using std::cout;
4 using std::endl;
5

©2013 Richard L. Halterman Draft date: January 27, 2013

 2.3. VARIATIONS OF OUR SIMPLE PROGRAM 15

6 int main()
7 {
8 cout << "This is a simple C++ program!" << endl;
9 }

We generally will not use the third approach, although you will encounter it in published C++ code. The
compiler will generate the same machine language code for all three versions. We generally will write
programs that use the using namespace directive and, therefore, use the shorter names.
The statement in the main function in any of the three versions of our program uses the services of an
object called std::cout. The std::cout object prints text on the computer’s screen. The text of the
message as it appears in the C++ source code is called a string, for string of characters. Strings are enclosed
within quotation marks (”). The symbols << make up the insertion operator. You can think of the message
to be printed as being “inserted” into the cout object. The cout object represents the output stream; that
is, text that the program prints to the console window. The endl word means “the end of line of printed
text,” and it causes the cursor to move down to the next line so any subsequent output will appear on the
next line. If you read the statement from left to right, the cout object which is responsible for displaying
text on the screen first receives the text to print and then receives the end-of-line directive to move to the
next line.
For simplicity, we’ll refer to this type of statement as a print statement, even though the word print does
not appear anywhere in the statement.
With minor exceptions, any statement in C++ must appear within a function definition. Our single print
statement appears within the function named main.
Any function, including main, may contain multiple statements. In Listing 2.4 (arrow.cpp), six print
statements draw an arrow on the screen:
Listing 2.4: arrow.cpp
1 #include <iostream>
2
3 using namespace std;
4
5 int main()
6 {
7 cout << " * " << endl;
8 cout << " *** " << endl;
9 cout << " ***** " << endl;
10 cout << " * " << endl;
11 cout << " * " << endl;
12 cout << " * " << endl;
13 }

The output of Listing 2.4 (arrow.cpp) is

*
***
*****
*
*
*

©2013 Richard L. Halterman Draft date: January 27, 2013

 2.3. VARIATIONS OF OUR SIMPLE PROGRAM 16

Each print statement “draws” a horizontal slice of the arrow. The six statements
cout << " * " << endl;
cout << " *** " << endl;
cout << " ***** " << endl;
cout << " * " << endl;
cout << " * " << endl;
cout << " * " << endl;

constitute the body of the main function. The body consists of all the statements between the open curly
brace () and close curly brace (). We say that the curly braces delimit the body of the function.
Listing 2.4 (arrow.cpp) can be rewritten to achieve the same effect with only one long print statement as
shown in Listing 2.5 (arrow2.cpp).

Listing 2.5: arrow2.cpp

1
2 #include <iostream>
3
4 using namespace std;
5
6 int main()
7 {
8 cout << " * " << endl
9 << " *** " << endl
10 << " ***** " << endl
11 << " * " << endl
12 << " * " << endl
13 << " * " << endl;
14 }

At first, Listing 2.4 (arrow.cpp) and Listing 2.5 (arrow2.cpp) may appear to be identical, but upon closer
inspection we see that cout appears only once within main, and only one semicolon (;) appears within
main. Since semicolons in C++ terminate statements, there really is only one statement. Notice that a
single statement can be spread out over several lines. The statement within main appearing as
cout << " * " << endl
<< " *** " << endl
<< " ***** " << endl
<< " * " << endl
<< " * " << endl
<< " * " << endl;

could have just as easily been written as

cout << " * " << endl << " *** "
<< endl << " ***** " << endl
<< " * " << endl
<< " * " << endl << " * " << endl;

but the first way of expressing it better portrays how the output will appear. Read this second version
carefully to convince yourself that the printed pieces will indeed flow to the cout printing object in the
proper sequence to produce the same picture of the arrow.

©2013 Richard L. Halterman Draft date: January 27, 2013

 2.4. TEMPLATE FOR SIMPLE C++ PROGRAMS 17

Consider the mistake of putting semicolons at the end of each of the lines in the
“one statement” version:
cout << " * " << endl;
<< " *** " << endl;
<< " ***** " << endl;
<< " * " << endl;
<< " * " << endl;
<< " * " << endl;
If this code fragment is put in main, the program will not compile. The reason is
simple—the semicolon at the end of the first line terminates the statement on that
line. The compiler expects a new statement on the next line, but
<< " *** " << endl;
is not a complete legal C++ statement since the << operator is missing the cout
object. The string " *** " and the end-of-line marker has nothing to “flow
into.”

Listing 2.6 (empty.cpp) is even simpler than Listing 2.1 (simple.cpp).

Listing 2.6: empty.cpp

1
2 int main()
3 {
4 }

Since Listing 2.6 (empty.cpp) uses neither the cout object nor endl, the #include and using direc-
tives are not needed. While it is legal and sometimes even useful in C++ to write functions with empty
bodies, such functions will do nothing when they execute. Listing 2.6 (empty.cpp) with its empty main
function is, therefore, truly the simplest executable C++ program that can be written, but it does nothing
when it runs!
In general, a C++ program may contain multiple functions, but such generality is deferred until Chap-
ter 9. For now, we will restrict our attention to programs with only a main function.

2.4 Template for simple C++ programs

For our immediate purposes all the programs we write will have the form shown in Figure 2.10.
Our programs generally will print something, so we need the #include directive. Since we will use
the short names for the printing objects, we also include the using namespace directive. The main
function definition is required for an executable program, and we will fill its body with statements that make
our program do as we wish. Later, our programs will become more sophisticated, and we will augment this
simple template.

©2013 Richard L. Halterman Draft date: January 27, 2013

2.5. SUMMARY 18

#include <iostream>

using namespace std;

int main()
{

program statements

Figure 2.10: The general structure of a C++ program.

2.5 Summary

• Any programs that print to the screen must use the #include directive to so that the program can
make use of the precompiled iostream library code to do I/O (input and output).

• All preprocessor directives begin with the # symbol.

• Messages can be printed on the screen by sending them to the std::cout object using the <<
operator.

• The << operator used in conjunction with the cout stream object is known as the insertion operator.

• The end of a printed line is signified by sending std::endl to the std::cout object.

• The shorter names cout and endl can be used if the using namespace std appears at the top
of the program’s source code.

• An executable C++ program must contain a function named main.

• Functions contain statements that are executed by the computer when the program runs. (The program
must be compiled for the computer to be able to execute the statements.)

• A function’s body consists of all the statements inside the delimiting curly braces ({}).

• A function’s body may be empty; however, such a function performs no useful activity when it
executes.

• Statements are terminated with semicolons (;).

• Within the Visual Studio IDE a program is developed within a project.

• Visual Studio’s command prompt environment can be used to develop simple C++ programs without
the overhead of creating a project.

• In a printing statement the characters within the quotation marks (") are printed literally on the screen.

©2013 Richard L. Halterman Draft date: January 27, 2013

2.6. EXERCISES 19

2.6 Exercises

1. What preprocessor directive must be used to allow statements to use the std::cout printing
stream?
2. What statement allows the short names cout and endl to be used instead of std::cout and
std::endl?
3. What does the name std stand for?

4. All C++ programs must have a function named what?

5. The body of main is enclosed within what symbols?
6. What operator directs information to the std::cout output stream?
7. Write a C++ program that prints your name in the console window.

8. Write a C++ program that prints your first and last name in the console window. Your first name
should appear on one line, and your last name appear on the next line.
9. What other files must be distributed with your executable file so that your program will run on a
Windows PC without Visual Studio installed? window.

10. Can a single statement in C++ span multiple lines in the source code?

©2013 Richard L. Halterman Draft date: January 27, 2013

2.6. EXERCISES 20

©2013 Richard L. Halterman Draft date: January 27, 2013

 21

Chapter 3

Values and Variables

In this chapter we explore some building blocks that are used to develop C++ programs. We experiment
with the following concepts:

• numeric values
• variables
• declarations
• assignment
• identifiers
• reserved words

In the next chapter we will revisit some of these concepts in the context of other data types.

3.1 Integer Values

The number four (4) is an example of a numeric value. In mathematics, 4 is an integer value. Integers are
whole numbers, which means they have no fractional parts, and they can be positive and negative. Examples
of integers include 4, −19, 0, and −1005. In contrast, 4.5 is not an integer, since it is not a whole number.
C++ supports a number of numeric and non-numeric values. In particular, C++ programs can use integer
values. It is easy to write a C++ program that prints the number four, as Listing 3.1 (number4.cpp) shows.

Listing 3.1: number4.cpp

1 #include <iostream>
2
3 using namespace std;
4
5 int main()
6 {
7 cout << 4 << endl;
8 }

©2013 Richard L. Halterman Draft date: January 27, 2013

 3.1. INTEGER VALUES 22

Notice that unlike the programs we saw earlier, Listing 3.1 (number4.cpp) does not use quotation marks
("). Compare Listing 3.1 (number4.cpp) to Listing 3.2 (number4-alt.cpp).

Listing 3.2: number4-alt.cpp

1 #include <iostream>
2
3 using namespace std;
4
5 int main()
6 {
7 cout << "4" << endl;
8 }

Both programs behave identically, but Listing 3.1 (number4.cpp) prints the value of the number four, while
Listing 3.2 (number4-alt.cpp) prints a message containing the digit four. The distinction here seems unim-
portant, but we will see in Section 3.2 that the presence or absence of the quotes can make a big difference
in the output.
In C++ source code, integers are not written with commas. The number two thousand, four hundred
sixty-eight would be written 2468, not 2,468. In mathematics, integers are unbounded; said another way,
the set of mathematical integers is infinite. In C++ the range of integers is limited, because all computers
have a finite amount of memory. The exact range of integers supported depends on the computer system
and particular C++ compiler. C++ on most 32-bit computer systems can represent integers in the range
−2,147,483,648 to +2,147,483,647.
What happens if you exceed the range of C++ integers? Try Listing 3.3 (exceed.cpp) on your system.

Listing 3.3: exceed.cpp

1 #include <iostream>
2
3 using namespace std;
4
5 int main()
6 {
7 cout << -3000000000 << endl;
8 }

Negative three billion is too large for 32-bit integers, however, and the program’s output is obviously wrong:

1294967296

The number printed was not even negative! (The Visual C++ compiler does issue a warning, but it still
produces the executable program.)
This limited range of values is common among programming languages since each number is stored in
a fixed amount of memory. Larger numbers require more storage in memory. In order to model the infinite
set of mathematical integers an infinite amount of memory would be needed! As we will see later, C++
supports an integer type with a greater range.
Appendix C.1 provides some details about the implementation of C++ integers.

©2013 Richard L. Halterman Draft date: January 27, 2013

 3.2. VARIABLES AND ASSIGNMENT 23

3.2 Variables and Assignment

In algebra, variables are used to represent numbers. The same is true in C++, except C++ variables also can
represent values other than numbers. Listing 3.4 (variable.cpp) uses a variable to store an integer value and
then prints the value of the variable.

Listing 3.4: variable.cpp

1 #include <iostream>
2
3 using namespace std;
4
5 int main()
6 {
7 int x;
8 x = 10;
9 cout << x << endl;
10 }

The main function in Listing 3.4 (variable.cpp) contains three statements:

• int x;
This is a declaration statement. All variables in a C++ program must be declared. A declaration
specifies the type of a variable. The word int indicates that the variable is an integer. The name
of the integer variable is x. We say that variable x has type int. C++ supports types other than
integers, and some types require more or less space in the computer’s memory. The compiler uses
the declaration to reserve the proper amount a memory to store the variable’s value. Due to the
declaration the compiler also can check to see if the variable is being used properly; for example,
we will see that integers can be added together just like in mathematics. For some other data types,
however, addition is not possible and so is not allowed. The compiler can ensure that a variable
involved in an addition is compatible with addition. It can report an error if it is not.
The compiler will issue an error if a programmer attempts to use an undeclared variable. The com-
piler cannot deduce the storage requirements and cannot verify the variable’s proper usage if it not
declared. Once declared, a particular variable cannot be redeclared in the same context. A variable
may not change its type during its lifetime.

• x = 10;
This is an assignment statement. An assignment statement associates a value with a variable. The
key to an assignment statement is the symbol = which is known as the assignment operator. Here the
value 10 is being assigned to the variable x. This means the value 10 will be stored in the memory
location the compiler has reserved for the variable named x. We need not be concerned about where
the variable is stored in memory; the compiler takes care of that detail.
After it has been declared, a variable may be assigned and reassigned as often as necessary.

• cout << x << endl;

This statement prints the variable x’s current value.

©2013 Richard L. Halterman Draft date: January 27, 2013

 3.2. VARIABLES AND ASSIGNMENT 24

Note that the lack of quotation marks here is very important. If x has the value
10, the statement
cout << x << endl;
prints 10, the value of the variable x, but the statement
cout << "x" << endl;
prints x, the message containing the single letter x.

The meaning of the assignment operator (=) is different from equality in mathematics. In mathematics,
= asserts that the expression on its left is equal to the expression on its right. In C++, = makes the variable
on its left take on the value of the expression on its right. It is best to read x = 5 as “x is assigned the
value 5,” or “x gets the value 5.” This distinction is important since in mathematics equality is symmetric:
if x = 5, we know 5 = x. In C++, this symmetry does not exist; the statement
5 = x;

attempts to reassign the value of the literal integer value 5, but this cannot be done because 5 is always 5
and cannot be changed. Such a statement will produce a compiler error:

error C2106: ’=’ : left operand must be l-value

Variables can be reassigned different values as needed, as Listing 3.5 (multipleassignment.cpp) shows.

Listing 3.5: multipleassignment.cpp

1 #include <iostream>
2
3 using namespace std;
4
5 int main()
6 {
7 int x;
8 x = 10;
9 cout << x << endl;
10 x = 20;
11 cout << x << endl;
12 x = 30;
13 cout << x << endl;
14 }

Observe the each print statement in Listing 3.5 (multipleassignment.cpp) is identical, but when the program
runs the print statements produce different results.
A variable may be given a value at the time of its declaration; for example, Listing 3.6 (variable-init.cpp)
is a variation of Listing 3.4 (variable.cpp).

Listing 3.6: variable-init.cpp

1 #include <iostream>
2
3 using namespace std;

©2013 Richard L. Halterman Draft date: January 27, 2013

 3.2. VARIABLES AND ASSIGNMENT 25

4
5 int main()
6 {
7 int x = 10;
8 cout << x << endl;
9 }

Notice that in Listing 3.6 (variable-init.cpp) the declaration and assignment of the variable x is performed in
one statement instead of two. This combined declaration and immediate assignment is called initialization.
C++ supports another syntax for initializing variables as shown in Listing 3.7 (alt-variable-init.cpp).

Listing 3.7: alt-variable-init.cpp

1 #include <iostream>
2
3 using namespace std;
4
5 int main()
6 {
7 int x(10);
8 cout << x << endl;
9 }

This alternate form is not commonly used for simple variables, but it necessary for initializing more com-
plicated kinds of variables called objects. Objects are introduced in Chapter 13 and Chapter 14.
Multiple variables of the same type can be declared and, if desired, initialized in a single statement. The
following statements declare three variables in one declaration statement:
int x, y, z;

The following statement declares three integer variables and initializes two of them:
int x = 0, y, z = 5;

Here y’s value is undefined.

The compiler maps a variable to a location in the computer’s memory. We can visualize a variable and
its corresponding memory location as a box as shown in Figure 3.1.

a
5
Figure 3.1: Representing a variable and its memory location as a box

We name the box with the variable’s name. Figure 3.2 shows how the following sequence of C++ code
affects memory.

©2013 Richard L. Halterman Draft date: January 27, 2013

3.3. IDENTIFIERS 26

int a, b;
a = 2;
b = 5;
a = b;
b = 4;

a b
int a, b; ? ?
a b
a = 2; 2 ?
a b
b = 5; 2 5

a b
a = b; 5 5
a b
b = 4; 5 4

Figure 3.2: How memory changes during variable assignment

Importantly, the statement

a = b;

does not mean a and b refer to the same box (memory location). After this statement a and b still refer
to separate boxes (memory locations). It simply means the value stored in b’s box (memory location) has
been copied to a’s box (memory location). a and b remain distinct boxes (memory locations). The original
value found in a’s box is overwritten when the contents of b’s box are copied into a. After the assignment
of b to a, the reassignment of b to 4 does not affect a.

3.3 Identifiers

While mathematicians are content with giving their variables one-letter names like x, programmers should
use longer, more descriptive variable names. Names such as sum, height, and subTotal are much
better than the equally permissible s, h, and st. A variable’s name should be related to its purpose within
the program. Good variable names make programs more readable by humans. Since programs often contain

©2013 Richard L. Halterman Draft date: January 27, 2013

3.3. IDENTIFIERS 27

many variables, well-chosen variable names can render an otherwise obscure collection of symbols more
understandable.
C++ has strict rules for variable names. A variable name is one example of an identifier. An identifier
is a word used to name things. One of the things an identifier can name is a variable. We will see in later
chapters that identifiers name other things such as functions and classes. Identifiers have the following
form:

• Identifiers must contain at least one character.

• The first character must be an alphabetic letter (upper or lower case) or the underscore
ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz_
• The remaining characters (if any) may be alphabetic characters (upper or lower case), the underscore,
or a digit
ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz_0123456789
• No other characters (including spaces) are permitted in identifiers.
• A reserved word cannot be used as an identifier (see Table 3.1).

Here are some examples of valid and invalid identifiers:

• All of the following words are valid identifiers and so can be used as variable names: x, x2, total,
port_22, and FLAG.
• None of the following words are valid identifiers: sub-total (dash is not a legal symbol in an
identifier), first entry (space is not a legal symbol in an identifier), 4all (begins with a digit),
#2 (pound sign is not a legal symbol in an identifier), and class (class is a reserved word).

C++ reserves a number of words for special use that could otherwise be used as identifiers. Called
reserved words or keywords, these words are special and are used to define the structure of C++ programs
and statements. Table 3.1 lists all the C++ reserved words.
The purposes of many of these reserved words are revealed throughout this book.
None of the reserved words in Table 3.1 can be used as identifiers. Fortunately, if you accidentally
attempt to use one of the reserved words in a program as a variable name, the compiler will issue an error
(see Section 4.6 for more on compiler errors).
In Listing 2.1 (simple.cpp) we used several reserved words: using, namespace, and int. Notice
that include, cout, endl, and main are not reserved words.
While mathematicians are content with giving their variables one-letter names like x, programmers
should use longer, more descriptive variable names. Names such as sum, height, and sub_total are
much better than the equally permissible s, h, and st. A variable’s name should be related to its purpose
within the program. Good variable names make programs more readable by humans. Since programs often
contain many variables, well-chosen variable names can render an otherwise obscure collection of symbols
more understandable.
Some programming languages do not require programmers to declare variables before they are used;
the type of a variable is determined by how the variable is used. Some languages allow the same variable
to assume different types as its use differs in different parts of a program. Such languages are known as
dynamically-typed languages. C++ is a statically-typed language. In a statically-typed language, the type of

©2013 Richard L. Halterman Draft date: January 27, 2013

3.3. IDENTIFIERS 28

alignas alignof and and_eq

asm auto bitand bitor
bool break case catch
char char16_t char32_t class
compl const constexpr const_cast
continue decltype default delete
do double dynamic_cast else
enum explicit export extern
false float for friend
goto if inline int
long mutable namespace new
noexcept not not_eq nullptr
operator or or_eq private
protected public register reinterpret_cast
return short signed sizeof
static static_assert static_cast struct
switch template this thread_local
throw true try typedef
typeid typename union unsigned
using virtual void volatile
wchar_t while xor xor_eq

Table 3.1: C++ keywords

a variable must be explicitly specified before it is used by statements in a program. While the requirement
to declare all variables may initially seem like a minor annoyance, it offers several advantages:

• When variables must be declared, the compiler can catch typographical errors that dynamically-typed
languages cannot detect. For example, consider the following section of code:
int ZERO;
ZER0 = 1;

The identifier in the first line ends with a capital “Oh.” In the second line, the identifier ends with the
digit zero. The distinction may be difficult or impossible to see in a particular editor or printout of
the code. A C++ compiler would immediately detect the typo in the second statement, since ZER0
(last letter a zero) has not been declared. A dynamically-typed language would create two variables:
ZERO and ZER0.

• When variables must be declared, the compiler can catch invalid operations. For example, a variable
may be declared to be of type int, but the programmer may accidentally assign a non-numeric
value to the variable. In a dynamically-typed language, the variable would silently change its type
introducing an error into the program. In C++, the compiler would report the improper assignment as
error, since once declared a C++ variable cannot change its type.

• Ideally, requiring the programmer to declare variables forces the programmer to plan ahead and think
more carefully about the variables a program might require. The purpose of a variable is tied to its
type, so the programmer must have a clear notion of the variable’s purpose before declaring it. When
variable declarations are not required, a programmer can “make up” variables as needed as the code is
written. The programmer need not do the simple double check of the variable’s purpose that writing

©2013 Richard L. Halterman Draft date: January 27, 2013

 3.4. FLOATING-POINT TYPES 29

Type Storage Smallest Magnitude Largest Magnitude Minimum Precision

float 4 bytes 1.17549 × 10−38 3.40282 × 10+38 6 digits
double 8 bytes 2.22507 × 10−308 1.79769 × 10+308 15 digits

Table 3.2: Characteristics of Floating-point Numbers on 32-bit Computer Systems

the variable’s declaration requires. While declaring the type of a variable specifies its purpose in only
a very limited way, any opportunity to catch such errors is beneficial.
• Statically-typed languages are generally more efficient than dynamically-typed languages. The com-
piler knows how much storage a variable requires based on its type. The space for that variable’s
value will not change over the life of the variable, since its type cannot change. In a dynamically
typed language that allows a variable to change its type, if a variable’s type changes during program
execution, the storage it requires may change also, so memory for that variable must be allocated
elsewhere to hold the different type. This memory reallocation at run time slows down the program’s
execution.

C++ is a case-sensitive language. This means that capitalization matters. if is a reserved word, but
none of If, IF, or iF are reserved words. Identifiers are case sensitive also; the variable called Name is
different from the variable called name.
Variable names should not be distinguished merely by differences in capitalization because it can be
confusing to human readers. For the same reason, it is considered poor practice to give a variable the same
name as a reserved word with one or more of its letters capitalized.

3.4 Floating-point Types

Many computational tasks require numbers that have fractional parts. For example, to compute the area of
a circle given the circle’s radius, the value π, or approximately 3.14159 is used. C++ supports such non-
integer numbers, and they are called floating-point numbers. The name implies that during mathematical
calculations the decimal point can move or “float” to various positions within the number to maintain the
proper number of significant digits. The types float and double represent different types of floating-
point numbers. The type double is used more often, since it stands for “double-precision floating-point,”
and it can represent a wider range of values with more digits of precision. The float type represents
single-precision floating-point values that are less precise. Table 3.2 provides some information about
floating-point values as commonly implemented on 32-bit computer systems. Floating point numbers can
be both positive and negative.
As you can see from Table 3.2, doubles provide more precision at the cost of using more memory.
Listing 3.8 (pi-print.cpp) prints an approximation of the mathematical value π.

Listing 3.8: pi-print.cpp

1 #include <iostream>
2
3 using namespace std;
4
5 int main()
6 {
7 double pi = 3.14159;

©2013 Richard L. Halterman Draft date: January 27, 2013

 3.4. FLOATING-POINT TYPES 30

8 cout << "Pi = " << pi << endl;

9 cout << "or " << 3.14 << " for short" << endl;
10 }

The first line in Listing 3.8 (pi-print.cpp) declares a variable named pi and assigns it a value. The second
line in Listing 3.8 (pi-print.cpp) prints the value of the variable pi, and the third line prints a literal value.
Any literal numeric value with a decimal point in a C++ program automatically has the type double, so
3.14

has type double. To make a literal floating-point value a float, an f or F must be appended to the
number, as in
3.14f

All floating-point literals are double values unless an f or F is appended to the

end of the number; for example, 2.0 is a double literal, while 2.0f is a float
literal.

Floating-point numbers are an approximation of mathematical real numbers. As in the case of the int
data type, the range of floating-point numbers is limited, since each value must be stored in a fixed amount
of memory. In some ways, though, ints are very different from doubles. Any integer within the range of
the int data type can be represented exactly. This is not true for the floating-point types. Consider the real
number π. Since π contains an infinite number of digits, its value only can be approximated. Because of the
limited number of digits available, some numbers with a finite number of digits can be only approximated;
for example, the number 23.3123400654033989 contains too many digits for the double type and must be
approximated as 23.3023498654034. An example of the problems that can arise due to the inexact nature
of floating-point numbers is demonstrated later in Listing 4.2 (imprecise.cpp).
Floating-point numbers can be expressed in scientific notation. Since most programming editors do not
provide superscripting and special symbols like ×, the normal scientific notation is altered slightly. The
number 6.022 × 1023 is written 6.022e23. The number to the left of the e (capital E can be used as well)
is the mantissa, and the number to the right of the e is the exponent of 10. As another example, −5.1 × 104
is expressed in C++ as -5.1e-4. Listing 3.9 (scientificnotation.cpp) prints some scientific constants using
scientific notation.
Listing 3.9: scientificnotation.cpp
1 #include <iostream>
2
3 using namespace std;
4
5 int main()
6 {
7 double avogadros_number = 6.022e23, c = 2.998e8;
8 cout << "Avogadro's number = " << avogadros_number << endl;
9 cout << "Speed of light = " << c << endl;
10 }

Appendix C.2 provides some insight into the implementation of C++ floating-point values and explains how
internally all floating-point numbers are stored in exponential notation with a mantissa and exponent.

©2013 Richard L. Halterman Draft date: January 27, 2013

 3.5. CONSTANTS 31

3.5 Constants

In Listing 3.9 (scientificnotation.cpp), Avogadro’s number and the speed of light are scientific constants;
that is, to the degree of precision to which they have been measured and/or calculated, they do not vary. C++
supports named constants. Constants are declared like variables with the addition of the const keyword:
const double PI = 3.14159;

Once declared and initialized, a constant can be used like a variable in all but one way—a constant may not
be reassigned. It is illegal for a constant to appear on the left side of the assignment operator (=) outside its
declaration statement. A subsequent statement like
PI = 2.5;

would cause the compiler to issue an error message:

error C3892: ’PI’ : you cannot assign to a variable that is const

and fail to compile the program. Since the scientific constants do not change, Listing 3.10 (const.cpp) is a
better version of Listing 3.9 (scientificnotation.cpp).

Listing 3.10: const.cpp

1 #include <iostream>
2
3 using namespace std;
4
5 int main()
6 {
7 const double avogadros_number = 6.022e23, c = 2.998e8;
8 cout << "Avogadro's number = " << avogadros_number << endl;
9 cout << "Speed of light = " << c << endl;
10 }

Because constants cannot be assigned outside their declaration statement, a constant must be given an
initial value when it is declared.
By convention, C++ programmers generally express constant names in all capital letters; in this way,
within the source code a human reader can distinguish a constant quickly from a variable.

3.6 Other Numeric Types

C++ supports several other numeric data types:

• long int—typically provides integers with a greater range than the int type; its abbreviated
name is long. It is guaranteed to provide a range of integer values at least as large as the int type.
• short int—typically provides integers with a smaller range than the int type; its abbreviated
name is short. It is guaranteed that the range of ints is at least as big as the range of shorts.
• unsigned int—is restricted to non-negative integers; its abbreviated name is unsigned. While
the unsigned type is limited in non-negative values, it can represent twice as many positive values

©2013 Richard L. Halterman Draft date: January 27, 2013

3.7. CHARACTERS 32

0 '\0' 16 —- 32 '' 48 '0' 64 @@ 80 'P' 96 '`' 112 'p'

1 —- 17 —- 33 '!' 49 '1' 65 'A' 81 'Q' 97 'a' 113 'q'
2 —- 18 —- 34 '"' 50 '2' 66 'B' 82 'R' 98 'b' 114 'r'
3 —- 19 —- 35 '#' 51 '3' 67 'C' 83 'S' 99 'c' 115 's'
4 —- 20 —- 36 '$' 52 '4' 68 'D' 84 'T' 100 'd' 116 't'
5 —- 21 —- 37 '%' 53 '5' 69 'E' 85 'U' 101 'e' 117 'u'
6 —- 22 —- 38 & 54 '6' 70 'F' 86 'V' 102 'f' 118 'v'
7 '\a' 23 —- 39 '\'' 55 '7' 71 'G' 87 'W' 103 'g' 119 'w'
8 '\b' 24 —- 40 '(' 56 '8' 72 'H' 88 'X' 104 'h' 120 'x'
9 '\t' 25 —- 41 ')' 57 '9' 73 'I' 89 'Y' 105 'i' 121 'y'
10 '\n' 26 —- 42 '*' 58 ':' 74 'J' 90 'Z' 106 'j' 122 'z'
11 —- 27 —- 43 '+' 59 ';' 75 'K' 91 '[' 107 'k' 123 '{'
12 '\f' 28 —- 44 ',' 60 '<' 76 'L' 92 '\\' 108 'l' 124 '|'
13 '\r' 29 —- 45 '-' 61 '=' 77 'M' 93 ']' 109 'm' 125 '}'
14 —- 30 —- 46 '.' 62 '>' 78 'N' 94 'ˆ' 110 'n' 126 '∼'
15 —- 31 —- 47 '/' 63 '?' 79 'O' 95 '_' 111 'o' 127 —-

Table 3.3: ASCII codes for characters

as the int type. (The name int is actually the short name for signed int and int can be
written as signed.)

• long double—can extend the range and precision of the double type.

While the C++ language standard specifies minimum ranges and precision for all the numeric data types,
a particular C++ compiler may exceed the specified minimums.
C++ provides such a variety of numeric types for specialized purposes usually related to building highly
efficient programs. We will have little need to use many of these types. Our examples will use mainly
the numeric types int for integers, double for an approximation of real numbers, and, less frequently,
unsigned when non-negative integral values are needed.

3.7 Characters

The char data type is used to represent single characters: letters of the alphabet (both upper and lower
case), digits, punctuation, and control characters (like newline and tab characters). Most systems support the
American Standard Code for Information Interchange (ASCII) character set. Standard ASCII can represent
128 different characters. Table 3.3 lists the ASCII codes for various characters.
In C++ source code, characters are enclosed by single quotes ('), as in
char ch = 'A';

Standard (double) quotes (") are reserved for strings, which are composed of characters, but strings and
chars are very different. C++ strings are covered in Section 11.7. The following statement would produce
a compiler error message:
ch = "A";

since a string cannot be assigned to a character variable.

Internally, chars are stored as integer values, and C++ permits assigning numeric values to char
variables and assigning characters to numeric variables. The statement

©2013 Richard L. Halterman Draft date: January 27, 2013

 3.7. CHARACTERS 33

ch = 65;

assigns a number to a char variable to show that this perfectly legal. The value 65 is the ASCII code for
the character A. If ch is printed, as in
ch = 65;
cout << ch;

the corresponding character, A, would be printed because ch’s declared type is char, not int or some
other numeric type.
Listing 3.11 (charexample.cpp) shows how characters can be used within a program.

Listing 3.11: charexample.cpp

1 #include <iostream>
2
3 using namespace std;
4
5 int main()
6 {
7 char ch1, ch2;
8 ch1 = 65;
9 ch2 = 'A';
10 cout << ch1 << ", " << ch2 << ", " << 'A' << endl;
11 }

The program displays

A, A, A

The first A is printed because the statement

ch1 = 65;

assigns the ASCII code for A to ch1. The second A is printed because the statement
ch2 = 'A';

assigns the literal character A to ch2. The third A is printed because the literal character 'A' is sent directly
to the output stream.
Integers and characters can be freely assigned to each other, but the range of chars is much smaller
than the range of ints, so care must be taken when assigning an int value to a char variable.
Some characters are non-printable characters. The ASCII chart lists several common non-printable
characters:

• '\n'—the newline character

• '\r'—the carriage return character
• '\b'—the backspace character
• '\a'—the “alert” character (causes a “beep” sound on many systems)
• '\t'—the tab character

©2013 Richard L. Halterman Draft date: January 27, 2013

3.8. ENUMERATED TYPES 34

• '\f'—the formfeed character

• '\0'—the null character (used in C strings, see Section 11.7)

These special non-printable characters begin with a backslash (\) symbol. The backslash is called
an escape symbol, and it signifies that the symbol that follows has a special meaning and should not be
interpreted literally. This means the literal backslash character must be represented as two backslashes:
'\\'.
These special non-printable character codes can be embedded within strings. To embed a backslash
within a string, you must escape it; for example, the statement
cout << "C:\\Dev\\cppcode" << endl;

would print

C:\Dev\cppcode

See what this statement prints:

cout << "AB\bCD\aEF" << endl;

The following two statements behave identically on most computer systems:

cout << "End of line" << endl;
cout << "End of line\n";

On the Microsoft Windows platform, the character sequence "\r\n" (carriage return, line feed) appears
at the end of lines in text files. Under Unix and Linux, lines in text files end with '\n' (line feed). On
Apple Macintosh systems, text file lines end with the '\r' (carriage return) character. The compilers that
adhere to the C++ standard will ensure that the '\n' character in a C++ program when sent to the output
stream will produce the same results as the endl manipulator.

3.8 Enumerated Types

C++ allows a programmer to create a new, very simple type and list all the possible values of that type. Such
a type is called an enumerated type, or an enumeration type. The enum keyword introduces an enumerated
type:
enum Color { Red, Orange, Yellow, Green, Blue, Violet };

Here, the new type is named Color, and a variable of type Color may assume one of the values that
appears in the list of values within the curly braces. The semicolon following the close curly brace is
required. Sometimes the enumerated type definition is formatted as
enum Color
{
Red,
Orange,
Yellow,

©2013 Richard L. Halterman Draft date: January 27, 2013

3.9. TYPE INFERENCE WITH AUTO 35

Green,
Blue,
Violet
};

but the formatting makes no different to the compiler.

The values listed with the curly braces constitute all the values that a variable of the enumerated type
can attain. The name of each value of an enumerated type must be a valid C++ identifier (see Section 3.3).
Given the Color type defined as above, we can declare and use variables of the enum type as shown
by the following code fragment:
Color myColor;
myColor = Orange;

Here the variable myColor has our custom type Color, and its value is Orange.
It is illegal to reuse an enumerated value name within another enumerated type within the same program.
In the following code, the enumerated value Light is used in both Shade and Weight:
enum Shade { Dark, Dim, Light, Bright };
enum Weight { Light, Medium, Heavy };

These two enumerated types are incompatible because they share the value Light, and so the compiler
will issue an error.
The value names within an enum type must be unique. The convention in C++ is to captialize the first
letter of an enum type and its associated values, although the language does not enforce this convention.
An enum type is handy for representing a small number of discrete, non-numeric options. For example,
consider a program that controls the movements made by a small robot. The allowed orientations are
forward, backward, left, right, up, and down. The program could encode these movements as integers,
where 0 means left, 1 means backward, etc. While that implementation will work, it is not ideal. Integers
may assume many more values than just the six values expected. The compiler cannot ensure that an integer
variable representing a robot move will stay in the range 0...5. What if the programmer makes a mistake and
under certain rare circumstances assigns a value outside of the range 0...5? The program then will contain
an error that may result in erratic behavior by the robot. With enum types, if the programmer uses only the
named values of the enum type, the compiler will ensure that such a mistake cannot happen.
A particular enumerated type necessarily has far fewer values than a type such as int. Imagine making
an integer enum type and having to list all of its values! (The standard 32-bit int type represents over four
billion values.) Enumerated types, therefore, are practical only for types that have a relatively small range
of values.

3.9 Type Inference with auto

C++ requires that a variable be declared before it is used. Ordinarily this means specifying the variable’s
type, as in
int count;
char ch;
double limit;

©2013 Richard L. Halterman Draft date: January 27, 2013

3.10. SUMMARY 36

A variable may be initialized when it is declared:

int count = 0;
char ch = 'Z';
double limit = 100.0;

Each of the values has a type: 0 is an int, 'Z' is a char, and 0.0 is a double. The auto keyword
allows the compiler to automatically deduce the type of a variable if it is initialized when it is declared:
auto count = 0;
auto ch = 'Z';
auto limit = 100.0;

The auto keyword may not be used without an accompanying initialization; for example, the following
declaration is illegal:
auto x;

because the compiler cannot deduce x’s type.

Automatic type inference is supported only by compilers that comply with the lat-
est C++11 standard. Programmers using older compilers must specify a variable’s
exact type during the variable’s declaration.

Automatic type deduction with auto is not useful to beginning C++ programmers. It is just as easy to
specify the variable’s type. The value of auto will become clearer when we consider some of the more
advanced features of C++ (see Section ??).

3.10 Summary

• C++ supports several kinds of numeric values and variables.

• In C++ commas cannot be used when expressing numeric literals.

• Numbers represented on a computer have limitations based on the finite nature of computer systems.

• All numeric types in C++ have limited ranges. This is because they must be stored in the computer’s
memory and be manipulated by the processor. The memory and processor deal with fixed-size quan-
tities, and in order to handle an unlimited range of values memory would have to be infinite.

• Variables are used to store values.

• All variables in C++ must be declared before they are used within the program.

• How can you declare a variable named total to be of type integer?

• The = operator means assignment, not mathematical equality.

• A variable can be reassigned at any time.

• A variable can be given an initial value at the time of its declaration.

©2013 Richard L. Halterman Draft date: January 27, 2013

3.10. SUMMARY 37

• Multiple variables can be declared in one declaration statement.

• A variable represents a location in memory capable of storing a value.

• The statement a = b; copies the value stored in variable b into variable a. a and b still refer to
there own memory locations.

• A variable name is an example of an identifier.

• The name of a variable must follow the identifier naming rules.

• All identifiers must consist of at least one character. The first symbol must be an alphabetic letter or
the underscore. Remaining symbols (if any) must be alphabetic letters, the underscore, or digits.

• Reserved words have special meaning within a C++ program and cannot be used as identifiers.

• Descriptive variable names are preferred over one-letter names.

• C++ is case sensitive; the name X is not the same as the name x.

• Floating numbers approximate mathematical real numbers.

• The float type represents single-precision floating-point numbers, and the double type represents
double-precision floating-point numbers.

• There are many values that floating-point numbers cannot represent exactly.

• The symbol f or F can be appended to a floating-point literal to signify that it is a float value
instead of a double value.

• Scientific notation literals of the form 1.0 × 101 can be expressed in C++ as 1.0e1.0.

• The qualifier const renders a “variable” unchangeable; it signifies a constant.

• On some systems longs may have extended range compared to normal ints.

• On some systems shortss may have reduced range compared to normal ints.

• The unsigned type represents non-negative integers.

• On some systems long doubles may have extended precision compared to doubles.

• The char type represents characters.

• Characters are stored internally as 8-bit integers. Their values correspond to the ASCII code.

• A char value can be assigned to an int variable, and an int value can be assigned to a char
variable. chara and ints are readily interchangeable in C++.

• Character literals appear within single quote marks ('), unlike string literals that appear within double
quote marks (").

• Special non-printable control codes like newline and tab are prefixed with the backslash escape char-
acter (\).

• The '\n' character is an alternative to using endl to end a printed line.

©2013 Richard L. Halterman Draft date: January 27, 2013

3.11. EXERCISES 38

3.11 Exercises

1. Will the following lines of code print the same thing? Explain why or why not.
cout << 6 << endl;
cout << "6" << endl;

2. Will the following lines of code print the same thing? Explain why or why not.
cout << x << endl;
cout << "x" << endl;

3. What is the largest int available on your system?

4. What is the smallest int available on your system?

5. What is the largest double available on your system?

6. What is the smallest double available on your system?

7. What C++ data type represents non-negative integers?

8. What happens if you attempt to use a variable within a program, and that variable is not declared?

9. What is wrong with the following statement that attempts to assign the value ten to variable x?
10 = x;

10. Once a variable has been properly declared and initialized can its value be changed?

11. What is another way to write the following declaration and initialization?
int x = 10;

12. In C++ can you declare more than variable in the same declaration statement? If so, how?

13. In the declaration

int a;
int b;

do a and b represent the same memory location?

14. Classify each of the following as either a legal or illegal C++ identifier:

(a) fred
(b) if
(c) 2x
(d) -4
(e) sum_total
(f) sumTotal
(g) sum-total
(h) sum total

©2013 Richard L. Halterman Draft date: January 27, 2013

3.11. EXERCISES 39

(i) sumtotal
(j) While
(k) x2
(l) Private
(m) public
(n) $16
(o) xTwo
(p) _static
(q) _4
(r) ___
(s) 10%
(t) a27834
(u) wilma's

15. What can you do if a variable name you would like to use is the same as a reserved word?
16. Why does C++ require programmers to declare a variable before using it? What are the advantages of
declaring variables?
17. What is the difference between float and double?
18. How can a programmer force a floating-point literal to be a float instead of a double?
19. How is the value 2.45 × 10−5 expressed as a C++ literal?
20. How can you ensure that a variable’s value can never be changed after its initialization?
21. How can you extend the range of int on some systems?
22. How can you extend the range and precision of double on some systems?
23. Write a program that prints the ASCII chart for all the values from 0 to 127.
24. Is "i" a string literal or character literal?
25. Is 'i' a string literal or character literal?
26. Is it legal to assign a char value to an int variable?
27. Is it legal to assign an int value to a char variable?
28. What is printed by the following code fragment?
int x;
x = 'A';
cout << x << endl;

29. What is the difference between the character 'n' and the character '\n'?
30. Write a C++ program that simply emits a beep sound when run.
31. Rewrite the following code fragment so that the code behaves exactly the same but the endl is not
used.
cout << endl << "Aye!" << endl << "Bye!" << endl;

©2013 Richard L. Halterman Draft date: January 27, 2013

3.11. EXERCISES 40

©2013 Richard L. Halterman Draft date: January 27, 2013

 41

Chapter 4

Expressions and Arithmetic

This chapter uses the C++ numeric types introduced in Chapter 3 to build expressions and perform arith-
metic. Some other important concepts are covered—user input, source formatting, comments, and dealing
with errors.

4.1 Expressions

A literal value like 34 and a properly declared variable like x are examples of a simple expressions. Values
and variables can be combined with operators to form more complex expressions. Listing 4.1 (adder.cpp)
shows how the addition operator (+) is used to add two integers.

Listing 4.1: adder.cpp

1 #include <iostream>
2
3 using namespace std;
4
5 int main()
6 {
7 int value1, value2, sum;
8 cout << "Please enter two integer values: ";
9 cin >> value1 >> value2;
10 sum = value1 + value2;
11 cout << value1 << " + " << value2 << " = " << sum << endl;
12 }

In Listing 4.1 (adder.cpp):

• int value1, value2, sum;

This statement declares three integer variables, but it does initialize them. The values of the variables
cannot be determined until the program is actually running, so it does no good to initialize them.
• cout << "Please enter two integer values: ";
This statement prompts the user to enter some information. This statement is our usual print state-
ment, but it is not terminated with the end-of-line marker endl. This is because we want the cursor

©2013 Richard L. Halterman Draft date: January 27, 2013

4.1. EXPRESSIONS 42

to remain at the end of the printed line so when the user types in values they appear on the same line
as the message prompting for the values. When the user presses the enter key to complete the input,
the cursor will automatically move down to the next line.
• cin >> value1 >> value2;
This statement causes the program’s execution to stop until the user types two numbers on the key-
board and then presses enter. The first number entered will be assigned to value1, and the second
number entered will be assigned to value2. Once the user presses the enter key, the value entered is
assigned to the variable. The user may choose to type one number, press enter, type the second num-
ber, and press enter again. Instead, the user may enter both numbers separated by one of more spaces
and then press enter only once. The program will not proceed until the user enters two numbers.

The cin input stream object can assign values to multiple variables in one state-
ment, as shown here:
int num1, num2, num3;
cin >> num1 >> num2 >> num3;
A common beginner’s mistake is use commas to separate the variables, as in
int num1, num2, num3;
cin >> num1, num2, num3;
The compiler will not generate an error message, because it is legal C++ code. The
statement, however, will not assign the three variables from user input as desired.
The comma operator in C++ has different meanings in different contexts, and here
it is treated like a statement separator; thus, the variables num2 and num3 are
not involved with the cin input stream object. We will have no need to use the
comma operator in this way, but you should be aware of this potential pitfall.

cin is a object that can be used to read input from the user. The >> operator is known as the
extraction operator. Notice that it is “backwards” from the << operator used with the cout object.
The cin object represents the input stream—information flowing into the program from user input
from the keyboard. The >> operator extracts the data from the input stream cin and assigns the
pieces of the data, in order, to the various variables on its right.
• sum = value1 + value2;
This is an assignment statement because it contains the assignment operator (=). The variable sum
appears to the left of the assignment operator, so sum will receive a value when this statement exe-
cutes. To the right of the assignment operator is an arithmetic expression involving two variables and
the addition operator. The expression is evaluated by adding together the values of the two variables.
Once the expression’s value has been determined, that value can be assigned to the sum variable.

All expressions have a value. The process of determining the expression’s value is called evaluation.
Evaluating simple expressions is easy. The literal value 54 evaluates to 54. The value of a variable named
x is the value stored in the memory location reserved for x. The value of a more complex expression is
found by evaluating the smaller expressions that make it up and combining them with operators to form
potentially new values.
The main C++ arithmetic operators are found in Table 4.1. The common arithmetic operations, addition,
subtraction, and multiplication, behave in the expected way. All these operators are classified as binary
operators because they operate on two operands. In the statement

©2013 Richard L. Halterman Draft date: January 27, 2013

4.1. EXPRESSIONS 43

Operator Meaning
+ addition
- subtraction
* multiplication
/ division
% modulus

Table 4.1: The simple C++ arithmetic operators

x = y + z;

the right side is an addition expression y + z. The two operands of the + operator are y and z.
Two of the operators above, + and -, can be used as unary operators. A unary operator has only one
operand. The - unary operator expects a single numeric expression (literal number, variable, or complex
numeric expression within parentheses) immediately to its right; it computes the additive inverse of its
operand. If the operand is positive (greater than zero), the result is a negative value of the same magnitude;
if the operand is negative (less than zero), the result is a positive value of the same magnitude. Zero is
unaffected. For example, the following code sequence
int x = 3;
int y = -4;
int z = 0;
cout << -x << " " << -y << " " -z << endl;

within a program would print

-3 4 0

The following statement

cout << -(4 - 5) << endl;

within a program would print

The unary + operator is present only for completeness; when applied to a numeric value, variable, or
expression, the resulting value is no different from the original value of its operand. Omitting the unary +
operator from the following statement
x = +y;

does not change its behavior.

All the arithmetic operators are subject to the limitations of the data types on which they operate; for
example, on a system in which the largest int is 2,147,483,647, the expression
2147483647 + 1

©2013 Richard L. Halterman Draft date: January 27, 2013

4.1. EXPRESSIONS 44

will not evaluate to the correct answer since the correct answer falls outside the range of ints.
When two ints are added, subtracted, multiplied, or divided, the result is an integer. As long as the
range of ints is not exceeded, the arithmetic works as expected. Division, however, is another matter. The
statement
cout << 10/3 << " " << 3/10 << endl;

prints

3 0

because in the first case 10 divided by 3 is 3 with a remainder of 1, and in the second case 3 divided by
10 is 0 with a remainder of 3. Since integers are whole numbers, any fractional part of the answer must
be discarded. The process of discarding the fractional part leaving only the whole number part is called
truncation. 10 divided by 3 should be 3.3333..., but that value is truncated to 3. Truncation is not rounding;
for example, 11 divided by 3 is 3.6666..., but it also truncates to 3.

Truncation simply removes any fractional part of the value. It does not round.
Both 10.01 and 10.999 truncate to 10.

The modulus operator (%) computes the remainder of integer division; thus,
cout << 10%3 << " " << 3%10 << endl;

prints

1 3

since 10 divided by 3 is 3 with a remainder of 1, and 3 divided by 10 is 0 with a remainder of 3.

The modulus operator is more useful than it may first appear. Listing 4.11 (timeconv.cpp) shows how it
can be used to convert a given number of seconds to hours, minutes, and seconds.
In contrast to integer arithmetic, floating-point arithmetic with doubles behaves as expected:
cout << 10.0/3.0 << " " << 3.0/10.0 << endl;

prints

3.33333 0.3

Recall from Section 3.4 that integers can be represented exactly, but floating-point numbers are imprecise
approximations of real numbers. Listing 4.2 (imprecise.cpp) clearly demonstrates their weaknesses.

©2013 Richard L. Halterman Draft date: January 27, 2013

 4.1. EXPRESSIONS 45

Listing 4.2: imprecise.cpp

1 #include <iostream>
2
3 using namespace std;
4
5 int main()
6 {
7 double one = 1.0,
8 one_third = 1.0/3.0,
9 zero = one - one_third - one_third - one_third;
10
11 cout << "one = " << one << ", one_third = " << one_third
12 << ", zero = " << zero << endl;
13
14 }

one = 1, one_third = 0.333333, zero = 1.11022e-016

The reported result is 1.11022 × 10−16 , or 0.000000000000000111022. While this number is very small,
with real numbers we get
1 1 1
1− − − = 0
3 3 3

Floating-point numbers are not real numbers, so the result of 1.0/3.0 cannot be represented exactly
without infinite precision. In the decimal (base 10) number system, one-third is a repeating fraction, so it
has an infinite number of digits. Even simple non-repeating decimal numbers can be a problem. One-tenth
(0.1) is obviously non-repeating, so it can be expressed exactly with a finite number of digits. As it turns out,
since numbers within computers are stored in binary (base 2) form, even one-tenth cannot be represented
exactly with double-precision floating-point numbers, as Listing 4.3 (imprecise10.cpp) illustrates.

Listing 4.3: imprecise10.cpp

1 #include <iostream>
2
3 using namespace std;
4
5 int main()
6 {
7 double one = 1.0,
8 one_tenth = 1.0/10.0,
9 zero = one - one_tenth - one_tenth - one_tenth
10 - one_tenth - one_tenth - one_tenth
11 - one_tenth - one_tenth - one_tenth
12 - one_tenth;
13
14 cout << "one = " << one << ", one_tenth = " << one_tenth
15 << ", zero = " << zero << endl;
16
17 }

©2013 Richard L. Halterman Draft date: January 27, 2013

4.2. MIXED TYPE EXPRESSIONS 46

The program’s output is

one = 1, one_tenth = 0.1, zero = 1.38778e-016

Surely the reported answer (1.38778 × 10−16 is close to the correct answer (zero). If you round it to the one-
hundred trillionth place (15 places behind the decimal point), it is correct. Listing 5.5 (samedifferent.cpp) in
Chapter 5 further emphasizes the point that programmers must use care when dealing with the imprecision
of floating-point quantities.
Despite their inexactness, double-precision floating-point numbers are used every day throughout the
world to solve sophisticated scientific and engineering problems. The limitations of floating-point numbers
are unavoidable since values with infinite characteristics must be represented in a finite way. Floating-point
numbers provide a good trade-off of precision for practicality.
Since a char is stored as a number (see Section 3.7), arithmetic can be performed on characters. We
will have little need to apply mathematics to characters, but sometimes it is useful. As an example, the
lower-case letters of the alphabet a–z occupy ASCII values 97–123, with a = 97, b = 98, etc. The upper-
case letters A–Z are coded as 65–91, with A = 65, B = 66, etc. To capitalize any lower-case letter, you need
only subtract 32, as in
char lower = 'd', upper = lower - 32;
cout << upper << endl;

This section of code would print D. If you do not remember the offset of 32 between upper- and lower-case
letter, you can compute it with the letters themselves:
upper = lower - ('a' - 'A');

In this case, if lower has been assigned any value in the range 'a' to 'z', upper will be assigned the
capitalized version of lower. On the other hand, if lower’s value is outside of that range, upper will
not be assigned a meaningful value.

4.2 Mixed Type Expressions

Expressions may contain mixed elements; for example, in the following program fragment
int x = 4;
double y = 10.2, sum;
sum = x + y;

an int is being added to a double, and the result is being assigned to a double. How is the arithmetic
performed?
As shown in Figure 4.1, the range of ints falls completely within the range of doubles; thus, any
int value can represented by a double. The int 4 also can be expressed as the double 4.0. In fact,
since the largest int on most systems is 2,147,483,647, the minimum 15 digits of double precision are
more than adequate to represent all integers exactly. This means that any int value can be represented by
a double. The converse is not true, however. 2,200,000,000 can be represented by a double but it is too
big for the int type. We say that the double type is wider than the int type and that the int type is
narrower than the double type.

©2013 Richard L. Halterman Draft date: January 27, 2013

4.2. MIXED TYPE EXPRESSIONS 47

Range of double
Range of int

+1.8 × 10308 -2.1 × 109 0 +2.1 × 109 +1.8 × 10308

Figure 4.1: Range of ints vs. range of doubles

It would be reasonable, then, to be able to assign int values to double variables. The process is
called widening, and it is always safe to widen an int to a double. The following code fragment
double d1;
int i1 = 500;
d1 = i1;
cout << "d1 = " << d1 << endl;

is legal C++ code, and when part of a complete program it would display

d1 = 500

Assigning a double to an int variable is not always possible, however, since the double value may
not be in the range of ints. Furthermore, if the double falls within the range of ints but is not a whole
number, the fractional part cannot be represented by the int. Consider the following code fragment:
double d = 1.6;
int i = d;

The second line assigns 1 to i. The fractional part (0.6) is truncated (see Section 4.1). Note that proper
rounding is not done. The Visual C++ compiler will warn us of a potential problem:

warning C4244: ’=’ : conversion from ’double’ to ’int’, possible loss of data

This warning reminds us that some information may be lost in the assignment. While the compiler and
linker will generate a executable program when warnings are present, all warnings should be carefully
scrutinized. This warning is particularly useful, since it is easy for errors to creep into calculations due to
the truncation of floating-point numbers.
Converting from a wider type to a narrower type (like double to int) is called narrowing. It often is
necessary to assign a floating-point value to an integer variable. If we know the value to assign is within
the range of ints, and the value has no fractional parts or its truncation would do no harm, the assignment
would be safe. To perform the assignment without a warning from the compiler, we use a procedure called
a cast, also called a type cast. The cast forces the compiler to accept the assignment without a issuing a
warning. The static_cast operator is used during the assignment:

©2013 Richard L. Halterman Draft date: January 27, 2013

 4.2. MIXED TYPE EXPRESSIONS 48

i = static_cast<int>(d);

The reserved word static_cast is used to perform a narrowing conversion. The item to convert is
placed in the parentheses (in this case the variable d), and the desired type is placed in the angle brackets
(in this case the type int). In the statement
i = static_cast<int>(d);

the type of the variable d is not changed; its value is simply copied to a temporary location in memory.
During the copying process the value is converted to an int.
Literal values and expressions can also be cast:
i = static_cast<int>(1.6);
i = static_cast<int>(x + 2.1);

When doubles are narrowed to ints, any fractional part is discarded. The as-
signment truncates; it does not round. The value 1.7 would be converted to the
int value 1.

Converting from a wider type to a narrower type (like double to int) is called narrowing. The
opposite conversion, from a narrower type to a wider type, is called widening. The widening conversion
is always safe, so a type cast is not required. Narrowing is a potentially dangerous operation, and using an
explicit cast does not remove the danger—it simply silences the compiler.
Narrowing is a potentially dangerous operation, and using an explicit cast does not remove the danger—
it simply silences the compiler. For example, consider Listing 4.4 (badnarrow.cpp).

Listing 4.4: badnarrow.cpp

1 #include <iostream>
2
3 using namespace std;
4
5 int main()
6 {
7 double d = 2200000000.0;
8 int i = d;
9 cout << "d = " << d << ", i = " << i << endl;
10 }

The Visual C++ compiler issues a warning about the possible loss of precision when assigning d to i.
Silencing the warning with a type cast is a bad idea; the program’s output indicates that the warning should
be heeded:

d = 2.2e+009, i = -2147483648

©2013 Richard L. Halterman Draft date: January 27, 2013

4.2. MIXED TYPE EXPRESSIONS 49

The printed values of i and d are not even close, nor can they be because it is impossible to represent the
value 2,200,000,000 as an int on a system that uses 32-bit integers. When assigning a value of a wider
type to a variable of a narrower type, the programmer must assume the responsibility to ensure that the
actual value to be narrowed is indeed within the range of the narrower type. The compiler cannot ensure
the safety of the assignment.
Casts should be used with great care because a cast creates a spot in the program that is immune to the
compiler’s type checking. A careless assignment can produce a garbage result introducing an error into the
program.
When mixed arithmetic must be performed—such as adding an int to a double—the compiler auto-
matically produces machine language code that copies the int value to a temporary memory location and
transforms it into its double equivalent. Double-precision floating-point arithmetic is then performed. In-
teger arithmetic is performed only when both operands and ints. 1/3 thus evaluates to 0, but 1.0/3.0,
1/3.0, and 1.0/3 all evaluate to 0.33333.
Since double is wider than int, we say that double dominates int. In a mixed type arithmetic
expression, the less dominant type is coerced into the more dominant type in order to perform the arithmetic
operation.
Section 3.8 introduced enumerated types. Behind the scenes, the compiler translates enumerated values
into integers. The first value in the enumeration is 0, the second value is 1, etc. The following code
demonstrates the relationship between enumerated types and integers:
enum Color { Red, Orange, Yellow, Green, Blue, Violet };
cout << Orange << " " << Green << endl;

The output of this code fragment is

1 3

C++ allows a programmer to assign an enum type to an integer variable; for example, the following state-
ment is legal:
int num = Orange;

Here, the variable num assumes the value 1. A programmer, however, may not directly assign an integer
value to a variable declared to be of an enum type; for example, the following statement:
Color col = 1;

is illegal. Sometimes it is convenient to convert an integer to an enumerated type, and a type cast enables
the assignment:
enum Color { Red, Orange, Yellow, Green, Blue, Violet };
Color myColor = static_cast<Color>(2);

Since the compiler encodes enumerated values with the integers 0, 1, 2, . . . in the order the values are spec-
ified in the enumeration list, the assignment statement above assigns the value Yellow to the variable
myColor. Casting, of course, can be dangerous; consider:
Color badColor = static_cast<Color>(45);

©2013 Richard L. Halterman Draft date: January 27, 2013

4.3. OPERATOR PRECEDENCE AND ASSOCIATIVITY 50

This statement assigns a meaningless value to the variable badColor, and the cast renders the compiler
powerless to help.
Even though enumerated types are encoded as integers internally, programmers may not perform arith-
metic on enumerated types without involving casts. Such opportunities should be very rare; if you need to
perform arithmetic on a variable, it most likely should be a numerical type, not an enumerated type.

4.3 Operator Precedence and Associativity

When different operators are used in the same expression, the normal rules of arithmetic apply. All C++
operators have a precedence and associativity:

• Precedence—when an expression contains two different kinds of operators, which should be applied
first?
• Associativity—when an expression contains two operators with the same precedence, which should
be applied first?

To see how precedence works, consider the expression

2 + 3 * 4

Should it be interpreted as
(2 + 3) * 4

(that is, 20), or rather is

2 + (3 * 4)

(that is, 14) the correct interpretation? As in normal arithmetic, in C++ multiplication and division have
equal importance and are performed before addition and subtraction. We say multiplication and division
have precedence over addition and subtraction. In the expression
2 + 3 * 4

the multiplication is performed before addition, since multiplication has precedence over addition. The
result is 14. The multiplicative operators (*, /, and %) have equal precedence with each other, and the
additive operators (binary + and -) have equal precedence with each other. The multiplicative operators
have precedence over the additive operators.
As in standard arithmetic, in C++ if the addition is to be performed first, parentheses can override the
precedence rules. The expression
(2 + 3) * 4

evaluates to 20. Multiple sets of parentheses can be arranged and nested in any ways that are acceptable in
standard arithmetic.
To see how associativity works, consider the expression
2 - 3 - 4

The two operators are the same, so they have equal precedence. Should the first subtraction operator be
applied before the second, as in

©2013 Richard L. Halterman Draft date: January 27, 2013

4.3. OPERATOR PRECEDENCE AND ASSOCIATIVITY 51

(2 - 3) - 4

(that is, −5), or rather is

2 - (3 - 4)

(that is, 3) the correct interpretation? The former (−5) is the correct interpretation. We say that the subtrac-
tion operator is left associative, and the evaluation is left to right. This interpretation agrees with standard
arithmetic rules. All binary operators except assignment are left associative. Assignment is an exception; it
is right associative. To see why associativity is an issue with assignment, consider the statement
w = x = y = z;

This is legal C++ and is called chained assignment. Assignment can be used as both a statement and an
expression. The statement
x = 2;

assigns the value 2 to the variable x. The expression

x = 2

assigns the value 2 to the variable x and evaluates to the value that was assigned; that is, 2. Since assignment
is right associative, the chained assignment example should be interpreted as
w = (x = (y = z));

which behaves as follows:

• The expression y = z is evaluated first. z’s value is assigned to y, and the value of the expression
y = z is z’s value.
• The expression x = (y = z) is evaluated. The value of y = z, that is z, is assigned to x. The
overall value of the expression x = y = z is thus the value of z. Now the values of x, y, and z are
all equal (to z).
• The expression w = (x = y = z) is evaluated. The value of the expression x = y = z is
equal to z’s value, so z’s value is assigned to w. The overall value of the expression w = x = y = z
is equal to z, and the variables w, x, y, and z are all equal (to z).

As in the case of precedence, parentheses can be used to override the natural associativity within an expres-
sion.
The unary operators have a higher precedence than the binary operators, and the unary operators are
right associative. This means the statements
cout << -3 + 2 << endl;
cout << -(3 + 2) << endl;

which display

-1
-5

©2013 Richard L. Halterman Draft date: January 27, 2013

4.4. COMMENTS 52

Arity Operators Associativity

Unary +, -
Binary *, /, % Left
Binary +, - Left
Binary = Right

Table 4.2: Operator precedence and associativity. The operators in each row have a higher precedence than
the operators below it. Operators within a row have the same precedence.

behave as expected.
Table 4.2 shows the precedence and associativity rules for some C++ operators. The * operator has a
unary form that has nothing to do with mathematics; it is covered in Section 10.6.

4.4 Comments

Good programmers annotate their code by inserting remarks that explain the purpose of a section of code or
why they chose to write a section of code the way they did. These notes are meant for human readers, not
the compiler. It is common in industry for programs to be reviewed for correctness by other programmers
or technical managers. Well-chosen identifiers (see Section 3.3) and comments can aid this assessment
process. Also, in practice, teams of programmers develop software. A different programmer may be
required to finish or fix a part of the program written by someone else. Well-written comments can help
others understand new code quicker and increase their productivity modifying old or unfinished code. While
it may seem difficult to believe, even the same programmer working on her own code months later can have
a difficult time remembering what various parts do. Comments can help greatly.
Any text contained within comments is ignored by the compiler. C++ supports two types of comments:
single line comments and block comments:

• Single line comment—the first type of comment is useful for writing a single line remark:
// Compute the average of the values
avg = sum / number;

The first line here is a comment that comment explains what the statement that follows it is supposed
to do. The comment begins with the double forward slash symbols (//) and continues until the end
of that line. The compiler will ignore the // symbols and the contents of the rest of the line. This
type of comment is also useful for appending a short comment to the end of a statement:
avg = sum / number; // Compute the average of the values

Here, an executable statement and the comment appear on the same line. The compiler will read the
assignment statement here, but it will ignore the comment. The compiler generates the same machine
code for this example as it does for the preceding example, but this example uses one line of source
code instead of two.
• Block comment—the second type of comment begins with the symbols /* and is in effect until
the */ symbols are encountered. The /* . . . */ symbols delimit the comment like parentheses
delimit a parenthetical expression. Unlike parentheses, however, these block comments cannot be
nested within other block comments.
The block comment is handy for multi-line comments:

©2013 Richard L. Halterman Draft date: January 27, 2013

 4.5. FORMATTING 53

/* After the computation is completed

the result is displayed. */
cout << result << endl;

What should be commented? Avoid making a remark about the obvious; for example:
result = 0; // Assign the value zero to the variable named result

The effect of this statement is clear to anyone with even minimal C++ programming experience. Thus, the
audience of the comments should be taken into account; generally, “routine” activities require no remarks.
Even though the effect of the above statement is clear, its purpose may need a comment. For example:
result = 0; // Ensures 'result' has a well-defined minimum value

This remark may be crucial for readers to completely understand how a particular part of a program works.
In general, programmers are not prone to providing too many comments. When in doubt, add a remark.
The extra time it takes to write good comments is well worth the effort.

4.5 Formatting

Program comments are helpful to human readers but ignored by the compiler. The way the source code is
formatted is also important to human readers but is of no consequence to the compiler. Consider Listing 4.5
(reformattedvariable.cpp) which is a reformatted version of Listing 3.4 (variable.cpp).

Listing 4.5: reformattedvariable.cpp

1 #include <iostream>
2 using
3 namespace
4 std
5 ;
6 int
7 main
8 (
9 )
10 {
11 int
12 x
13 ;
14 x
15 =
16 10
17 ;
18 cout
19 <<
20 x
21 <<
22 endl
23 ;
24 }

Listing 4.6 (reformattedvariable2.cpp) is another reformatted version of Listing 3.4 (variable.cpp).

©2013 Richard L. Halterman Draft date: January 27, 2013

 4.5. FORMATTING 54

Listing 4.6: reformattedvariable2.cpp

1 #include <iostream>
2 using namespace std;int main(){int x;x=10;cout<<x<<endl;}

Both reformatted programs are valid C++ and compile to the same machine language code as the original
version. Most would argue that the original version is easier to read and understand more quickly than either
of the reformatted versions. The elements in Listing 3.4 (variable.cpp) are organized better. What are some
distinguishing characteristics of Listing 3.4 (variable.cpp)?

• Each statement appears on its own line. A statement is not unnecessarily split between two lines of
text. Visually, one line of text implies one action (statement) to perform.

• The close curly brace aligns vertically with the corresponding open curly brace from a line above it.
This makes it easier to determine if the curly braces match and nest properly. It also better portrays
the logical structure of the program. The ability to accurately communicate the logical structure of
a program becomes very important as write more complex programs. Programs with complex logic
frequently use multiple nested curly braces (for example, see Listing 5.9 (troubleshoot.cpp)). Without
a consistent, organized arrangement of curly braces it can difficult to determine which opening brace
goes with a particular closing brace.

• The statements that constitute the body of main are indented several spaces. This visually em-
phasizes the fact that the elements are indeed logically enclosed. As with curly brace alignment,
indentation to emphasize logical enclosure becomes more important as more complex programs are
considered.

• Spaces are used to spread out statements and group pieces of the statement. Space around the op-
erators (=) makes it easier to visually separate the operands from the operators and comprehend the
details of the expression. Most people find the statement

total_sale = subtotal + tax;

much easier to read than

total_sale=subtotal+tax;

since the lack of space in the second version makes it more difficult to pick out the individual pieces
of the statement. In the first version with extra space, it is clearer where operators and variable names
begin and end.
In a natural language like English, a book is divided into distinct chapters, and chapters are composed
of paragraphs. One paragraph can be distinguished from another because the first line is indented or
an extra space appears between two paragraphs. Space is used to separate words in each sentence.
Consider how hard it would be to read a book if all the sentences were printed like this one:

Theboyranquicklytothetreetoseethestrandedcat.

Judiciously placed open space in a C++ program can greatly enhance its readability.

©2013 Richard L. Halterman Draft date: January 27, 2013

4.6. ERRORS AND WARNINGS 55

C++ gives the programmer a large amount of freedom in formatting source code.
The compiler reads the characters that make up the source code one symbol at a
time left to right within a line before moving to the next line. While extra space
helps readability, spaces are not allowed in some places:
– Variable names and reserved words must appear as unbroken units.
– Multi-symbol operators like << cannot be separated (< < is illegal).

One common coding convention that is universal in C++ programming is demonstrated in Listing 3.10
(const.cpp). While programmers usually use lower-case letters in variable names, they usually express
constant names with all capital letters; for example, PI is used for the mathematical constant π instead of
pi. C++ does not require constants to be capitalized, but capitalizing them aids humans reading the source
code so they can quickly distinguish between variables and constants.
Most software development organizations adopt a set of style guidelines, sometimes called code conven-
tions. These guidelines dictate where to indent and by how many spaces, where to place curly braces, how
to assign names to identifiers, etc. Programmers working for the organization are required to follow these
style guidelines for the code they produce. This better enables any member of the development team to read
and understand more quickly code written by someone else. This is necessary when code is reviewed for
correctness or when code must be repaired or extended, and the original programmer is no longer with the
development team.

4.6 Errors and Warnings

Beginning programmers make mistakes writing programs because of inexperience in programming in gen-
eral or because of unfamiliarity with a programming language. Seasoned programmers make mistakes due
to carelessness or because the proposed solution to a problem is faulty and the correct implementation of an
incorrect solution will not produce a correct program. Regardless of the reason, a programming error falls
under one of three categories:

• compile-time error
• run-time error
• logic error

4.6.1 Compile-time Errors

A compile-time error results from the programmer’s misuse of the language. A syntax error is a common
compile-time error. For example, in English one can say

The boy walks quickly.

This sentence uses correct syntax. However, the sentence

The boy walk quickly.

©2013 Richard L. Halterman Draft date: January 27, 2013

4.6. ERRORS AND WARNINGS 56

is not correct syntactically: the number of the subject (singular form) disagrees with the number of the verb
(plural form). It contains a syntax error. It violates a grammatical rule of the English language. Similarly,
the C++ statement
x = y + 2;

is syntactically correct because it obeys the rules for the structure of an assignment statement described in
Section 3.2. However, consider replacing this assignment statement with a slightly modified version:
y + 2 = x;

If a statement like this one appears in a program and the variables x and y have been properly declared, the
compiler will issue an error message; for example, the Visual C++ compiler reports (among other things):

error C2106: ’=’ : left operand must be l-value

The syntax of C++ does not allow an expression like y + 2 to appear on the left side of the assignment
operator.
(The term l-value in the error message refers to the left side of the assignment operator; the l is an
“elle,” not a “one.”.)
The compiler may generate an error for a syntactically correct statement like
x = y + 2;

if either of the variables x or y has not been declared; for example, if y has not been declared, Visual C++
reports:

error C2065: ’y’ : undeclared identifier

Other common compile-time errors include missing semicolons at the end of statements, mismatched curly
braces and parentheses, and simple typographical errors.
Compile-time errors usually are the easiest to repair. The compiler pinpoints the exact location of the
problem, and the error does not depend on the circumstances under which the program executes. The exact
error can be reproduced by simply recompiling the same source code.
Compilers have the reputation for generating cryptic error messages. They seem to provide little help
as far as novice programmers are concerned. Sometimes a combination of errors can lead to messages that
indicate errors on lines that follow the line that contains the actual error. Once you encounter the same
error several times and the compiler messages become more familiar, you become better able to deduce the
actual problem from the reported message. Unfortunately C++ is such a complex language that sometimes a
simple compile-time error can result in a message that is incomprehensible to beginning C++ programmers.

4.6.2 Run-time Errors

The compiler ensures that the structural rules of the C++ language are not violated. It can detect, for exam-
ple, the malformed assignment statement and the use of a variable before its declaration. Some violations of
the language cannot be detected at compile time, however. A program may not run to completion but instead
terminate with an error. We commonly say the program “crashed.” Consider Listing 4.7 (dividedanger.cpp)
which under certain circumstances will crash.

©2013 Richard L. Halterman Draft date: January 27, 2013

 4.6. ERRORS AND WARNINGS 57

Listing 4.7: dividedanger.cpp

1 // File dividedanger.cpp
2
3 #include <iostream>
4
5 using namespace std;
6
7 int main()
8 {
9 int dividend, divisor;
10
11 // Get two integers from the user
12 cout << "Please enter two integers to divide:";
13 cin >> dividend >> divisor;
14 // Divide them and report the result
15 cout << dividend << "/" << divisor << " = "
16 << dividend/divisor << endl;
17 }

The expression
dividend/divisor

is potentially dangerous. If the user enters, for example, 32 and 4, the program works nicely

Please enter two integers to divide: 32 4

32/4 = 8

and displays the answer of 8. If the user instead types the numbers 32 and 0, the program reports an error
and terminates. Division by zero is undefined in mathematics, and integer division by zero in C++ is illegal.
When the program attempts the division at run time, the system detects the attempt and terminates the
program.
This particular program can fail in other ways as well; for example, outside of the C++ world, 32.0 looks
like a respectable integer. If the user types in 32.0 and 8, however, the program crashes because 32.0 is not
a valid way to represent an integer in C++. When the compiler compiles the source line
cin >> dividend >> divisor;

given that dividend has been declared to be an int, it generates slightly different machine language
code than it would if dividend has been declared to be a double instead. The compiled code expects
the text entered by the user to be digits with no extra decoration. Any deviation from this expectation results
in a run-time error. Similar results occur if the user enters text that does not represent an integer, like fred.
Observe that in either case—entry of a valid but inappropriate integer (zero) or entry of a non-integer
(32.0 or fred)—it is impossible for the compiler to check for these problems at compile time. The compiler
cannot predict what the user will enter when the program is run. This means it is up to the programmer
to write code that can handle bad input that the user may provide. As we continue our exploration of
programming in C++, we will discover ways to make our programs more robust against user input (see
Listing 5.2 (betterdivision.cpp) in Chapter 5, for example). The solution involves changing the way the
program runs depending on the actual input provided by the user.

©2013 Richard L. Halterman Draft date: January 27, 2013

 4.6. ERRORS AND WARNINGS 58

4.6.3 Logic Errors

Consider the effects of replacing the expression

dividend/divisor;

in Listing 4.7 (dividedanger.cpp) with the expression:

divisor/dividend;

The program compiles with no errors. It runs, and unless a value of zero is entered for the dividend, no
run-time errors arise. However, the answer it computes is not correct in general. The only time the correct
answer is printed is when dividend = divisor. The program contains an error, but neither the compiler
nor the run-time system is able detect the problem. An error of this type is known as a logic error.
Listing 4.14 (faultytempconv.cpp) is an example of a program that contains a logic error. Listing 4.14
(faultytempconv.cpp) compiles and does not generate any run-time errors, but it produces incorrect results.
Beginning programmers tend to struggle early on with compile-time errors due to their unfamiliarity
with the language. The compiler and its error messages are actually the programmer’s best friend. As
the programmer gains experience with the language and the programs written become more complicated,
the number of compile-time errors decrease or are trivially fixed and the number of logic errors increase.
Unfortunately, both the compiler and run-time environment are powerless to provide any insight into the
nature and sometimes location of logic errors. Logic errors, therefore, tend to be the most difficult to find
and repair. Tools such as debuggers are frequently used to help locate and fix logic errors, but these tools
are far from automatic in their operation.
Errors that escape compiler detection (run-time errors and logic errors) are commonly called bugs. Be-
cause of the inability of the compiler to detect these problems, such bugs are the major source of frustration
for developers. The frustration often arises because in complex programs the bugs sometimes only reveal
themselves in certain situations that are difficult to reproduce exactly during testing. You will discover this
frustration as your programs become more complicated. The good news is that programming experience
and the disciplined application of good programming techniques can help reduce the number logic errors.
The bad news is that since software development in an inherently human intellectual pursuit, logic errors
are inevitable. Accidentally introducing and later finding and eliminating logic errors is an integral part of
the programming process.

4.6.4 Compiler Warnings

A warning issued by the compiler does mark a violation of the rules in the C++ language, but it is an notifi-
cation to the programmer that the program contains a construct that is a potential problem. In Listing 4.10
(tempconv.cpp) the programmer is attempting to print the value of a variable before it has been given a
known value.

Listing 4.8: uninitialized.cpp

1 // uninitialized.cpp
2
3 #include <iostream>
4
5 using namespace std;
6
7 int main()
8 {

©2013 Richard L. Halterman Draft date: January 27, 2013

 4.6. ERRORS AND WARNINGS 59

9 int n;
10 cout << n << endl;
11 }

An attempt to build Listing 4.8 (uninitialized.cpp) yields the following message from the Visual C++
compiler:

warning C4700: uninitialized local variable ’n’ used

The compiler issued a warning but still generated the executable file. When run, the program produces a
random result because it prints the value in memory associated with the variable, but the program does not
initialize that memory location.
Listing 4.9 (narrow.cpp) assigns a double value to an int variable, which we know from Section 4.1
truncates the result.

Listing 4.9: narrow.cpp

1 #include <iostream>
2
3 using namespace std;
4
5 int main()
6 {
7 int n;
8 double d = 1.6;
9 n = d;
10 cout << n << endl;
11 }

When compiled we see

warning C4244: ’=’ : conversion from ’double’ to ’int’, possible loss of data

Since it is a warning and not an error, the compiler generates the executable, but the warning should prompt
us to stop and reflect about the correctness of the code. The enhanced warning level prevents the program-
mer from being oblivious to the situation.
The default Visual C++ warning level is 3 when compiling in the IDE and level 1 on the command line
(that is why we use the /W3 option on the command line); the highest warning level is 4. You can reduce
the level to 1 or 2 or disable warnings altogether, but that is not recommended. The only reason you might
want to reduce the warning level is to compile older existing C++ source code that does meet newer C++
standards. When developing new code, higher warning levels are preferred since they provide more help to
the programmer. Unless otherwise noted, all the complete program examples in this book compile cleanly
under Visual C++ set at warning level 3. Level 3 is helpful for detecting many common logic errors.
Most warnings can be avoided by a simple addition to the code. Section 4.2 showed how static_cast
can be used to coerce a wider type to a narrower type. At Visual C++ warning Level 3, the compiler issues a
warning if the cast is not used. The little code that must be added should cause the programmer to stop and
reflect about the correctness of the construct. The enhanced warning level prevents the programmer from
being oblivious to the situation.

©2013 Richard L. Halterman Draft date: January 27, 2013

 4.7. ARITHMETIC EXAMPLES 60

Use the strongest level of warnings available to your compiler. Treat all warnings
as problems that must be corrected. Do not accept as completed a program that
compiles with warnings.

4.7 Arithmetic Examples

The kind of arithmetic to perform in a complex expression is determined on an operator by operator basis.
For example, consider Listing 4.10 (tempconv.cpp) that attempts to convert a temperature from degrees
Fahrenheit to degrees Celsius using the formula

◦ 5
C= × (◦ F − 32)
9

Listing 4.10: tempconv.cpp

1 // File tempconv.cpp
2
3 #include <iostream>
4
5 using namespace std;
6
7 int main()
8 {
9 double degreesF, degreesC;
10 // Prompt user for temperature to convert
11 cout << "Enter the temperature in degrees F: ";
12 // Read in the user's input
13 cin >> degreesF;
14 // Perform the conversion
15 degreesC = 5/9*(degreesF - 32);
16 // Report the result
17 cout << degreesC << endl;
18 }

Listing 4.10 (tempconv.cpp) contains comments that document each step explaining the code’s purpose.
An initial test is promising:

Enter the temperature in degrees F: 32

Degrees C = 0

Water freezes at 32 degrees Fahrenheit and 0 degrees Celsius, so the program’s behavior is correct for this
test. Several other attempts are less favorable—consider

©2013 Richard L. Halterman Draft date: January 27, 2013

 4.7. ARITHMETIC EXAMPLES 61

Enter the temperature in degrees F: 212

Degrees C = 0

Water boils at 212 degrees Fahrenheit which is 100 degrees Celsius, so this answer is not correct.

Enter the temperature in degrees F: -40

Degrees C = 0

The value −40 is the point where the Fahrenheit and Celsius curves cross, so the result should be −40, not
zero. The first test was only coincidentally correct.
Unfortunately, the printed result is always zero regardless of the input. The problem is the division 5/9
in the statement
degreesC = 5/9*(degreesF - 32);

Division and multiplication have equal precedence, and both are left associative; therefore, the division is
performed first. Since both operands are integers, integer division is performed and the quotient is zero (5
divided by 9 is 0, remainder 5). Of course zero times any number is zero, thus the result. The fact that a
floating-point value is involved in the expression (degreesF) and the overall result is being assigned to a
floating-point variable, is irrelevant. The decision about the exact type of operation to perform is made on
an operator-by-operator basis, not globally over the entire expression. Since the division is performed first
and it involves two integer values, integer division is used before the other floating-point pieces become
involved.
One solution simply uses a floating-point literal for either the five or the nine, as in
degreesC = 5.0/9*(degreesF - 32);

This forces a double-precision floating-point division (recall that the literal 5.0 is a double). The correct
result, subject to rounding instead of truncation, is finally computed.
Listing 4.11 (timeconv.cpp) uses integer division and modulus to split up a given number of seconds to
hours, minutes, and seconds.

Listing 4.11: timeconv.cpp

1 // File timeconv.cpp
2
3 #include <iostream>
4
5 using namespace std;
6
7 int main()
8 {
9 int hours, minutes, seconds;
10 cout << "Please enter the number of seconds:";
11 cin >> seconds;
12 // First, compute the number of hours in the given number
13 // of seconds

©2013 Richard L. Halterman Draft date: January 27, 2013

 4.7. ARITHMETIC EXAMPLES 62

14 hours = seconds / 3600; // 3600 seconds = 1 hours

15 // Compute the remaining seconds after the hours are
16 // accounted for
17 seconds = seconds % 3600;
18 // Next, compute the number of minutes in the remaining
19 // number of seconds
20 minutes = seconds / 60; // 60 seconds = 1 minute
21 // Compute the remaining seconds after the minutes are
22 // accounted for
23 seconds = seconds % 60;
24 // Report the results
25 cout << hours << " hr, " << minutes << " min, "
26 << seconds << " sec" << endl;
27 }

If the user enters 10000, the program prints 2 hr, 46 min, 40 sec. Notice the assignments to
the seconds variable, such as
seconds = seconds % 3600

The right side of the assignment operator (=) is first evaluated. The remainder of seconds divided by
3,600 is assigned back to seconds. This statement can alter the value of seconds if the current value of
seconds is greater than 3,600. A similar statement that occurs frequently in programs is one like
x = x + 1;

This statement increments the variable x to make it one bigger. A statement like this one provides further
evidence that the C++ assignment operator does not mean mathematical equality. The following statement
from mathematics
x = x+1
is surely never true; a number cannot be equal to one more than itself. If that were the case, I would deposit
one dollar in the bank and then insist that I really had two dollars in the bank, since a number is equal to
one more than itself. That two dollars would become 3.00, then 4.00, etc., and soon I would be rich. In
C++, however, this statement simply means “add one to x and assign the result back to x.”
A variation on Listing 4.11 (timeconv.cpp), Listing 4.12 (enhancedtimeconv.cpp) performs the same
logic to compute the time pieces (hours, minutes, and seconds), but it uses more simple arithmetic to
produce a slightly different output—instead of printing 11,045 seconds as 3 hr, 4 min, 5 sec, List-
ing 4.12 (enhancedtimeconv.cpp) displays it as 3:04:05. It is trivial to modify Listing 4.11 (timeconv.cpp)
so that it would print 3:4:5, but Listing 4.12 (enhancedtimeconv.cpp) includes some extra arithmetic to
put leading zeroes in front of single-digit values for minutes and seconds as is done on digital clock displays.

Listing 4.12: enhancedtimeconv.cpp

1 // File enhancedtimeconv.cpp
2
3 #include <iostream>
4
5 using namespace std;
6
7 int main()
8 {
9 int hours, minutes, seconds;
10 cout << "Please enter the number of seconds:";

©2013 Richard L. Halterman Draft date: January 27, 2013

 4.8. MORE ARITHMETIC OPERATORS 63

11 cin >> seconds;

12 // First, compute the number of hours in the given number
13 // of seconds
14 hours = seconds / 3600; // 3600 seconds = 1 hours
15 // Compute the remaining seconds after the hours are
16 // accounted for
17 seconds = seconds % 3600;
18 // Next, compute the number of minutes in the remaining
19 // number of seconds
20 minutes = seconds / 60; // 60 seconds = 1 minute
21 // Compute the remaining seconds after the minutes are
22 // accounted for
23 seconds = seconds % 60;
24 // Report the results
25 cout << hours << ":";
26 // Compute tens digit of minutes
27 int tens = minutes / 10;
28 cout << tens;
29 // Compute ones digit of minutes
30 int ones = minutes % 10;
31 cout << ones << ":";
32 // Compute tens digit of seconds
33 tens = seconds / 10;
34 cout << tens;
35 // Compute ones digit of seconds
36 ones = seconds % 10;
37 cout << ones << endl;
38 }

Listing 4.12 (enhancedtimeconv.cpp) uses the fact that if x is a one- or two-digit number, x % 10 is the
tens digit of x. If x % 10 is zero, x is necessarily a one-digit number.

4.8 More Arithmetic Operators

We will see shortly that variables are often modified in a regular way as programs execute. A variable may
increase by one or decrease by five. The statement
x = x + 1;

increments x by one, making it one bigger than it was before this statement was executed. C++ has a shorter
statement that accomplishes the same effect:
x++;

This is the increment statement. A similar decrement statement is available:

x--; // Same as x = x - 1;

These statements are more precisely post-increment and post-decrement operators. There are also pre-
increment and pre-decrement forms, as in
--x; // Same as x = x - 1;
++y; // Same as y = y + 1;

©2013 Richard L. Halterman Draft date: January 27, 2013

 4.8. MORE ARITHMETIC OPERATORS 64

When they appear alone in a statement, the pre- and post- versions of the increment and decrement
operators work identically. Their behavior is different when they are embedded within a more complex
statement. Listing 4.13 (prevspost.cpp) demonstrates how the pre- and post- increment operators work
slightly differently.

Listing 4.13: prevspost.cpp

1 #include <iostream>
2
3 using namespace std;
4
5 int main()
6 {
7 int x1 = 1, y1 = 10, x2 = 100, y2 = 1000;
8 cout << "x1=" << x1 << ", y1=" << y1
9 << ", x2=" << x2 << ", y2=" << y2 << endl;
10 y1 = x1++;
11 cout << "x1=" << x1 << ", y1=" << y1
12 << ", x2=" << x2 << ", y2=" << y2 << endl;
13 y2 = ++x2;
14 cout << "x1=" << x1 << ", y1=" << y1
15 << ", x2=" << x2 << ", y2=" << y2 << endl;
16 }

Listing 4.13 (prevspost.cpp) prints

x1=1, y1=10, x2=100, y2=1000

x1=2, y1=1, x2=100, y2=1000
x1=2, y1=1, x2=101, y2=101

If x1 has the value 1 just before the statement

y1 = x1++;

then immediately after the statement executes x1 is 2 and y1 is 1.

If x1 has the value 1 just before the statement
y1 = ++x1;

then immediately after the statement executes x1 is 2 and y1 is also 2.

As you can see, the pre-increment operator uses the new value of the incremented variable when eval-
uating the overall expression. In contrast, the post-increment operator uses the original value of the in-
cremented variable when evaluating the overall expression. The pre- and post-decrement operator behaves
similarly.
For beginning programmers it is best to avoid using the increment and decrement operators within more
complex expressions. We will use them frequently as standalone statements since there is no danger of
misinterpreting their behavior when they are not part of a more complex expression.
C++ provides a more general way of simplifying a statement that modifies a variable through simple
arithmetic. For example, the statement

©2013 Richard L. Halterman Draft date: January 27, 2013

4.8. MORE ARITHMETIC OPERATORS 65

x = x + 5;

can be shorted to

x += 5;

This statement means “increase x by five.” Any statement of the form

x op= exp;

where

• x is a variable.

• op= is an arithmetic operator combined with the assignment operator; for our purposes, the ones most
useful to us are +=, -=, *=, /=, and %=.

• exp is an expression compatible with the variable x.

Arithmetic reassignment statements of this form are equivalent to

x = x op exp;

This means the statement

x *= y + z;

is equivalent to

x = x * (y + z);

The version using the arithmetic assignment does not require parentheses. The arithmetic assignment is
especially handy if a variable with a long name is to be modified; consider

temporary_filename_length = temporary_filename_length / (y + z);

versus

temporary_filename_length /= y + z;

©2013 Richard L. Halterman Draft date: January 27, 2013

 4.9. ALGORITHMS 66

Do not accidentally reverse the order of the symbols for the arithmetic assignment
operators, like in the statement
x =+ 5;
Notice that the + and = symbols have been reversed. The compiler interprets this
statement as if it had been written
x = +5;
that is, assignment and the unary operator. This assigns x to exactly five instead
of increasing it by five.
Similarly,
x =- 3;
would assign −3 to x instead of decreasing x by three.

Appendix D examines some additional arithmetic operators available in C++.

4.9 Algorithms

An algorithm is a finite sequence of steps, each step taking a finite length of time, that solves a problem or
computes a result. A computer program is one example of an algorithm, as is a recipe to make lasagna. In
both of these examples, the order of the steps matter. In the case of lasagna, the noodles must be cooked
in boiling water before they are layered into the filling to be baked. It would be inappropriate to place the
raw noodles into the pan with all the other ingredients, bake it, and then later remove the already baked
noodles to cook them in boiling water separately. In the same way, the ordering of steps is very important
in a computer program. While this point may be obvious, consider the following sound argument:

1. The relationship between degrees Celsius and degrees Fahrenheit can be expressed as

◦ 5
C= × (◦ F − 32)
9

2. Given a temperature in degrees Fahrenheit, the corresponding temperature in degrees Celsius can be
computed.

Armed with this knowledge, Listing 4.14 (faultytempconv.cpp) follows directly.

Listing 4.14: faultytempconv.cpp

1 // File faultytempconv.cpp
2
3 #include <iostream>
4
5 using namespace std;
6
7 int main()
8 {
9 double degreesF = 0, degreesC = 0;

©2013 Richard L. Halterman Draft date: January 27, 2013

 4.9. ALGORITHMS 67

10 // Define the relationship between F and C

11 degreesC = 5.0/9*(degreesF - 32);
12 // Prompt user for degrees F
13 cout << "Enter the temperature in degrees F: ";
14 // Read in the user's input
15 cin >> degreesF;
16 // Report the result
17 cout << degreesC << endl;
18 }

Unfortunately, the executing program always displays

-17.7778

regardless of the input provided. The English description provided above is correct. No integer division
problems lurk, as in Listing 4.10 (tempconv.cpp). The problem lies simply in statement ordering. The
statement
degreesC = 5.0/9*(degreesF - 32);

is an assignment statement, not a definition of a relationship that exists throughout the program. At the
point of the assignment, degreesF has the value of zero. The variable degreesC is assigned before
degreesF’s value is received from the user.
As another example, suppose x and y are two variables in some program. How would we interchange
the values of the two variables? We want x to have y’s original value and y to have x’s original value. This
code may seem reasonable:
x = y;
y = x;

The problem with this section of code is that after the first statement is executed, x and y both have the same
value (y’s original value). The second assignment is superfluous and does nothing to change the values of
x or y. The solution requires a third variable to remember the original value of one the variables before it
is reassigned. The correct code to swap the values is
temp = x;
x = y;
y = temp;

This small example emphasizes the fact that algorithms must be specified precisely. Informal notions about
how to solve a problem can be valuable in the early stages of program design, but the coded program
requires a correct detailed description of the solution.
The algorithms we have seen so far have been simple. Statement 1, followed by Statement 2, etc. until
every statement in the program has been executed. Chapter 5 and Chapter 6 introduce some language
constructs that permit optional and repetitive execution of some statements. These constructs allow us to
build programs that do much more interesting things, but more complex algorithms are required to make it
happen. We must not lose sight of the fact that a complicated algorithm that is 99% correct is not correct.
An algorithm’s design and implementation can be derailed by inattention to the smallest of details.

©2013 Richard L. Halterman Draft date: January 27, 2013

4.10. SUMMARY 68

4.10 Summary

• The literal value 4 and integer sum are examples of simple C++ numeric expressions.

• 2*x + 4 is an example of a more complex C++ numeric expression.

• Expressions can be printed via the cout output stream object and assigned to variables.

• A binary operator performs an operation using two operands.

• With regard to binary operators: + represents arithmetic addition; - represents arithmetic subtrac-
tion; * represents arithmetic multiplication; / represents arithmetic division; % represents arithmetic
modulus, or integer remainder after division.

• The std::cin object can be used to assign user keyboard input to variables when the program is
executing.

• The >> operator is called the extraction operator. It assigns to variables data extracted from the input
stream object cin.

• A unary operator performs an operation using one operand.

• The - unary operator represents the additive inverse of its operand.

• The + unary operator has no effect on its operand.

• Arithmetic applied to integer operands yields integer results.

• With a binary operation, double-precision floating-point arithmetic is performed if at least one of its
operands is a floating-point number.

• Floating-point arithmetic is inexact and subject to rounding errors because all floating-point values
have finite precision.

• When a floating-point value is assigned to an integer variable, the value is truncated, not properly
rounded.

• A mixed expression is an expression that contains values and/or variables of differing types.

• Generally speaking, a type is narrower than another type if its range of values is smaller than the
other type; a type is wider than another type if its range of values is larger than the other type.

• A wider type dominates a narrower type.

• In mixed arithmetic with a binary operator, the type of arithmetic performed is determined by the
more dominant operand.

• Floating point values can be assigned to integer variables, but any fractional part will be truncated.
Programmers must ensure that the value assigned falls within the range of integers.

• In C++ a wider type may be assigned to a narrower type, but a floating-point to integer conversion
truncates, and a value outside the range of the narrower type results in a bogus value being assigned.
Programmers should use caution when assigning a wider type to a narrower type.

• In C++, operators have both a precedence and an associativity.

©2013 Richard L. Halterman Draft date: January 27, 2013

4.10. SUMMARY 69

• With regard to the arithmetic operators, C++ uses the same precedence rules as standard arithmetic:
multiplication and division are applied before addition and subtraction unless parentheses dictate
otherwise.

• The arithmetic operators associate left to right; assignment associates right to left.

• Chained assignment can be used to assign the same value to multiple variables within one statement.

• The unary operators + and - have precedence over the binary arithmetic operators *, /, and %,
which have precedence over the binary arithmetic operators + and -, which have precedence over the
assignment operator.

• Comments are notes within the source code. The compiler ignores comments when compiling the
source code.

• Comments inform human readers about the code.

• Comments should not state the obvious, but it is better to provide too many comments rather than too
few.

• A single line comment begins with the symbols // and continues until the end of the line.

• A block comment begins with the symbols /* and continues in the source code until the symbols */
terminate it.

• Source code should be formatted so that it is more easily read and understood by humans.

• C++ has some general formatting guidelines that should be followed:

– Each statement should appear on its own line.

– Curly braces should be aligned in order to be matched visually more quickly.
– The statements that comprise the body of a function should be indented; four spaces are gener-
ally accepted as ideal.
– Spaces should be used within statements separating the logical pieces to make the statements
easier to read.

• Compile-time errors are caused by the programmer’s misuse of the C++ language. Compile-time
errors are detected and reported by the compiler (and sometimes the linker).

• Runtime errors are errors that are detected when the program is executing. When a run-time error
arises, the program terminates with an error message.

• Logic errors elude detection by the compiler and run-time environment. A logic error is indicated
when the the program does not behave as expected.

• The compiler should be set at the highest warning level to check for as many problems as possible.
Warnings should be taken seriously, and a program should not be considered finished while warnings
remain.

• In complicated arithmetic expressions involving many operators and operands, the rules pertaining
to mixed arithmetic are applied on an operator-by-operator basis, following the precedence and asso-
ciativity laws, not globally over the entire expression.

• The ++ and -- operators can be used to increment and decrement variables.

©2013 Richard L. Halterman Draft date: January 27, 2013

4.11. EXERCISES 70

• The family of op= operators (+=, -=, *=, /=, and %=) allow variables to be changed by a given
amount using a particular arithmetic operator.
• C++ programs implement algorithms; as such, C++ statements do not declare statements of fact or
define relationships that hold throughout the program’s execution; rather they indicate how the values
of variables change as the execution of the program progresses.

4.11 Exercises

1. Is the literal 4 a valid C++ expression?

2. Is the variable x a valid C++ expression?
3. Is x + 4 a valid C++ expression?
4. What affect does the unary + operator have when applied to a numeric expression?
5. Sort the following binary operators in order of high to low precedence: +, -, *, /, %, =.
6. Given the following declaration:
int x = 2;

Indicate what each of the following C++ statements would print.

(a) cout << "x"<< endl;
(b) cout << 'x'<< endl;
(c) cout << x << endl;
(d) cout << "x + 1"<< endl;
(e) cout << 'x'+ 1 << endl;
(f) cout << x + 1 << endl;
7. Sort the following types in order from narrowest to widest: int, double, float, long, char.
8. Given the following declarations:
int i1 = 2, i2 = 5, i3 = -3;
double d1 = 2.0, d2 = 5.0, d3 = -0.5;

Evaluate each of the following C++ expressions.

(a) i1 + i2
(b) i1 / i2
(c) i2 / i1
(d) i1 * i3
(e) d1 + d2
(f) d1 / d2
(g) d2 / d1
(h) d3 * d1
(i) d1 + i2

©2013 Richard L. Halterman Draft date: January 27, 2013

4.11. EXERCISES 71

(j) i1 / d2
(k) d2 / i1
(l) i2 / d1
(m) i1/i2*d1
(n) d1*i1/i2
(o) d1/d2*i1
(p) i1*d1/d2
(q) i2/i1*d1
(r) d1*i2/i1
(s) d2/d1*i1
(t) i1*d2/d1

9. What is printed by the following statement:

cout << /* 5 */ 3 << endl;

10. Given the following declarations:

int i1 = 2, i2 = 5, i3 = -3;
double d1 = 2.0, d2 = 5.0, d3 = -0.5;

Evaluate each of the following C++ expressions.

(a) i1 + (i2 * i3)

(b) i1 * (i2 + i3)
(c) i1 / (i2 + i3)
(d) i1 / i2 + i3
(e) 3 + 4 + 5 / 3
(f) (3 + 4 + 5) / 3
(g) d1 + (d2 * d3)
(h) d1 + d2 * d3
(i) d1 / d2 - d3
(j) d1 / (d2 - d3)
(k) d1 + d2 + d3 / 3
(l) (d1 + d2 + d3) / 3
(m) d1 + d2 + (d3 / 3)
(n) 3 * (d1 + d2) * (d1 - d3)

11. How are single-line comments different from block comments?

12. Can block comments be nested?

13. Which is better, too many comments or too few comments?

14. What is the purpose of comments?

©2013 Richard L. Halterman Draft date: January 27, 2013

4.11. EXERCISES 72

15. The programs in Listing 3.4 (variable.cpp), Listing 4.5 (reformattedvariable.cpp), and Listing 4.6
(reformattedvariable2.cpp) compile to the same machine code and behave exactly the same. What
makes one of the programs clearly better than the others?

16. Why is human readability such an important consideration?

17. Consider the following program which contains some errors. You may assume that the comments
within the program accurately describe the program’s intended behavior.
#include <iostream>

using namespace std;

int main()
{
int n1, n2, d1; // 1
// Get two numbers from the user
cin << n1 << n2; // 2
// Compute sum of the two numbers
cout << n1 + n2 << endl; // 3
// Compute average of the two numbers
cout << n1+n2/2 << endl; // 4
// Assign some variables
d1 = d2 = 0; // 5
// Compute a quotient
cout << n1/d1 << endl; // 6
// Compute a product
n1*n2 = d1; // 7
// Print result
cout << d1 << endl; // 8
}

For each line listed in the comments, indicate whether or not a compile-time, run-time, or logic error
is present. Not all lines contain an error.

18. What distinguishes a compiler warning from a compiler error? Should you be concerned about
warnings? Why or why not?

19. What are the advantages to enhancing the warning reporting capabilities of the compiler?

20. Write the shortest way to express each of the following statements.

(a) x = x + 1;
(b) x = x / 2;
(c) x = x - 1;
(d) x = x + y;
(e) x = x - (y + 7);
(f) x = 2*x;
(g) number_of_closed_cases = number_of_closed_cases + 2*ncc;

©2013 Richard L. Halterman Draft date: January 27, 2013

4.11. EXERCISES 73

21. What is printed by the following code fragment?

int x1 = 2, y1, x2 = 2, y2;
y1 = ++x1;
y2 = x2++;
cout << x1 << " " << x2 << endl;
cout << y1 << " " << y2 << endl;

Why does the output appear as it does?

22. Consider the following program that attempts to compute the circumference of a circle given the
radius entered by the user. Given a circle’s radius, r, the circle’s circumference, C is given by the
formula:
C = 2πr

#include <iostream>

using namespace std;

int main()
{
double C, r;
const double PI = 3.14159;
// Formula for the area of a circle given its radius
C = 2*PI*r;
// Get the radius from the user
cout >> "Please enter the circle's radius: ";
cin << r;
// Print the circumference
cout << "Circumference is " << C << endl;
}

(a) The compiler issues a warning. What is the warning?

(b) The program does not produce the intended result. Why?
(c) How can it be repaired so that it not only eliminates the warning but also removes the logic
error?

23. Write a C++ program that ...

24. Write a C++ program that ...

©2013 Richard L. Halterman Draft date: January 27, 2013

4.11. EXERCISES 74

©2013 Richard L. Halterman Draft date: January 27, 2013

 75

Chapter 5

Conditional Execution

All the programs in the preceding chapters execute exactly the same statements regardless of the input, if
any, provided to them. They follow a linear sequence: Statement 1, Statement 2, etc. until the last statement
is executed and the program terminates. Linear programs like these are very limited in the problems they
can solve. This chapter introduces constructs that allow program statements to be optionally executed,
depending on the context (input) of the program’s execution.

5.1 Type bool

Arithmetic expressions evaluate to numeric values; a Boolean expression, sometimes called a predicate,
evaluates to true or false. While Boolean expressions may appear very limited on the surface, they are
essential for building more interesting and useful programs.
C++ supports the non-numeric data type bool, which stands for Boolean. The term Boolean comes
from the name of the British mathematician George Boole. A branch of discrete mathematics called
Boolean algebra is dedicated to the study of the properties and the manipulation of logical expressions.
Compared to the numeric types, the bool type is very simple in that it can represent only two values:
true or false. Listing 5.1 (boolvars.cpp) is a simple program that shows how Boolean variables can be
used.

Listing 5.1: boolvars.cpp

1 #include <iostream>
2
3 using namespace std;
4
5 int main()
6 {
7 // Declare some Boolean variables
8 bool a = true, b = false;
9 cout << "a = " << a << ", b = " << b << endl;
10 // Reassign a
11 a = false;
12 cout << "a = " << a << ", b = " << b << endl;
13 // Mix integers and Booleans
14 a = 1;

©2013 Richard L. Halterman Draft date: January 27, 2013

 5.2. BOOLEAN EXPRESSIONS 76

Operator Meaning
== Equal to
< Less than
> Greater than
<= Less than or equal to
>= Greater than or equal to
!= Not equal to

Table 5.1: C++ Relational operators

15 b = 1;
16 cout << "a = " << a << ", b = " << b << endl;
17 // Assign Boolean value to an integer
18 int x = a, y = true;
19 cout << "a = " << a << ", b = " << b
20 << ", x = " << x << ", y = " << y << endl;
21 // More mixing
22 a = 1725; // Warning issued
23 b = -19; // Warning issued
24 cout << "a = " << a << ", b = " << b << endl;
25 }

As you can see from running Listing 5.1 (boolvars.cpp), the Boolean values false and true are
represented as integer 0 and integer 1. More precisely, zero represents the bool value false, and any
non-zero integer (positive or negative) means true. The direct assignment to a bool variable of an integer
other than 0 or 1 may result in a warning (Visual C++ reports truncation of ’int’ to ’bool’), but the variable
is still interpreted as true. The data type bool is basically a convenience for programmers; any C++
program that uses bool variables can be rewritten using integers instead to achieve the same results. While
Boolean values and variables are freely compatible and interchangeable with integers, the bool type is
convenient and should be used when the context involves truth values instead of numbers.
It is important to note that the Visual C++ compiler issues warnings for the last two assignment state-
ments in Listing 5.1 (boolvars.cpp). Even though any non-zero value is considered true, 1 is the preferred
integer equivalent to true (as you can see when you attempt to print the literal value true). Since the
need to assign to a Boolean variable a value other than true or false or the equivalent 1 or 0 should be
extremely rare, the compiler’s message alerts the programmer to check to make sure the assignment is not
a mistake.

5.2 Boolean Expressions

The simplest Boolean expressions are false and true, the Boolean literals. A Boolean variable is also
a Boolean expression. An expression comparing numeric expressions for equality or inequality is also a
Boolean expression. These comparisons are done using relational operators. Table 5.1 lists the relational
operators available in C++.
Table 5.2 shows some simple Boolean expressions with their associated values. An expression like
10 < 20 is legal but of little use, since the expression true is equivalent, simpler, and less likely to
confuse human readers. Boolean expressions are extremely useful when their truth values depend on the
values of one or more variables.

©2013 Richard L. Halterman Draft date: January 27, 2013

5.2. BOOLEAN EXPRESSIONS 77

Expression Value
10 < 20 always true
10 >= 20 always false
x == 10 true only if x has the value 10
X != y true unless x and y have the same values

Table 5.2: Relational operator examples

The relational operators are binary operators and are all left associative. They all have a lower prece-
dence than any of the arithmetic operators; therefore, the expression
x + 2 < y / 10

is evaluated as if parentheses were placed as so:

(x + 2) < (y / 10)

C++ allows statements to be simple expressions; for example, the statement

x == 15;
may look like an attempt to assign the value 15 to the variable x, but it is not.
The = operator performs assignment, but the == operator checks for relational
equality. If you make a mistake and use == as shown here, Visual C++ will issue a
warning that includes the message
warning C4553: ’==’ : operator has no effect; did you intend ’=’?
Recall from Section 4.6.4 that a compiler warning does not indicate a violation of
the rules of the language; rather it alerts the programmer to a possible trouble spot
in the code.
Another example of an expression used as a statement is
x + 15;
This statement is a legal (but useless) C++ statement, and the compiler notifies us
accordingly:
warning C4552: ’+’ : operator has no effect; expected operator with
side-effect
Why are expressions allowed as statements? Some simple expressions have side
effects that do alter the behavior of the program. One example of such an ex-
pression is x++. Listing 4.13 (prevspost.cpp) showed how x++ can be used as
a standalone statement and behave as an expression within a larger statement. A
more common example is the use of a function call (which is an expression) as
standalone a statement. (We introduce functions in Chapter 8.) In order to keep
the structure of the language as uniform as possible, C++ tolerates useless ex-
pressions as statements to enable programmers to use the more useful expression-
statements. Fortunately, most compilers issue informative warnings about the use-
less expression-statements to keep developers on track.

©2013 Richard L. Halterman Draft date: January 27, 2013

 5.3. THE SIMPLE IF STATEMENT 78

5.3 The Simple if Statement

The Boolean expressions described in Section 5.2 at first may seem arcane and of little use in practical
programs. In reality, Boolean expressions are essential for a program to be able to adapt its behavior at run
time. Most truly useful and practical programs would be impossible without the availability of Boolean
expressions.
The run-time exceptions mentioned in Section 4.6 arise from logic errors. One way that Listing 4.7
(dividedanger.cpp) can fail is when the user enters a zero for the divisor. Fortunately, programmers can take
steps to ensure that division by zero does not occur. Listing 5.2 (betterdivision.cpp) shows how it might be
done.

Listing 5.2: betterdivision.cpp

1 #include <iostream>
2
3 using namespace std;
4
5 int main()
6 {
7 int dividend, divisor;
8
9 // Get two integers from the user
10 cout << "Please enter two integers to divide:";
11 cin >> dividend >> divisor;
12 // If possible, divide them and report the result
13 if ( divisor != 0 )
14 cout << dividend << "/" << divisor << " = "
15 << dividend/divisor << endl;
16 }

The second cout statement may not always be executed. In the following run

Please enter two integers to divide: 32 8

32/8 = 4

it is executed, but if the user enters a zero as the second number:

Please enter two integers to divide: 32 0

the program prints nothing after the user enters the values.
The last statement in Listing 5.2 (betterdivision.cpp) begins with the reserved word if. The if state-
ment allows code to be optionally executed. In this case, the printing statement is executed only if the
variable divisor’s value is not zero.
The Boolean expression
divisor != 0

©2013 Richard L. Halterman Draft date: January 27, 2013

5.3. THE SIMPLE IF STATEMENT 79

determines if the single statement that follows the right parenthesis is executed. If divisor is not zero,
the message is printed; otherwise, nothing is displayed.
Figure 5.1 shows how program execution flows through the if statement. of Listing 5.2 (betterdivision.cpp).

Is no
divisor ≠ 0?

yes

do the division
and print result

Figure 5.1: if flowchart

The general form of the if statement is:

if ( condition )
statement;

• The reserved word if begins the if statement.

• The Boolean expression condition determines whether or not the body will be executed. The Boolean
expression must be enclosed within parentheses as shown.

• The statement is the statement to be executed if the Boolean expression is true. The statement makes
up the body of the if statement. Section 5.4 shows how the body can be composed of multiple
statements.

Good coding style dictates the body should be indented to emphasize the optional execution and improve
the program’s readability. The indenting is not required by the compiler. Sometimes programmers will place
a short body on the same time as the if; for example, the following if statement optionally assigns y:

©2013 Richard L. Halterman Draft date: January 27, 2013

5.3. THE SIMPLE IF STATEMENT 80

if ( x < 10 )
y = x;

and could be written as

if ( x < 10 ) y = x;

but should not be written as

if ( x < 10 )
y = x;

because the lack of indentation hides the fact that the assignment statement is optionally executed. The
compiler will accept it, but it is misleading to human readers accustomed to the indentation convention.
The compiler, of course, will accept the code written as
if(x<10)y=x;

but the lack of spaces makes it difficult for humans to read.

When the if statement is written the preferred way using two lines of source
code, it is important not to put a semicolon at the end of the first line:
if ( x < 10 ); // No! Don't do this!
y = x;
Here, the semicolon terminates the if statement, but the indentation implies that
the second line is intended to be the body of the if statement. The compiler,
however, interprets the badly formatted if statement as if it were written as
if ( x < 10 )
; // This is what is really going on.
y = x;
This is legal in C++; it means the if statement has an empty body. In which
case the assignment is not part of the body. The assignment statement is after
the body and always will be executed regardless of the truth value of the Boolean
expression.

When checking for equality, as in

if ( x == 10 )
cout << "ten";
be sure to use the relational equality operator (==), not the assignment operator
(=). Since an assignment statement has a value (the value that is assigned, see Sec-
tion 4.3), C++ allows = within the conditional expression. It is, however, almost
always a mistake when beginning programmers use = in this context. Visual C++ at
warning Level 4 checks for the use of assignment within a conditional expression;
the default Level 3 does not.

©2013 Richard L. Halterman Draft date: January 27, 2013

 5.4. COMPOUND STATEMENTS 81

5.4 Compound Statements

Sometimes more than one statement must be optionally executed. Listing 5.3 (alternatedivision.cpp) shows
how curly braces are used to group multiple statements together into one compound statement.

Listing 5.3: alternatedivision.cpp

1 #include <iostream>
2
3 using namespace std;
4
5 int main()
6 {
7 int dividend, divisor, quotient;
8
9 // Get two integers from the user
10 cout << "Please enter two integers to divide:";
11 cin >> dividend >> divisor;
12 // If possible, divide them and report the result
13 if ( divisor != 0 )
14 {
15 quotient = dividend / divisor;
16 cout << dividend << " divided by " << divisor << " is "
17 << quotient << endl;
18 }
19 }

The assignment statement and printing statement are both a part of the body of the if statement. Given
the truth value of the Boolean expression divisor != 0 during a particular program run, either both
statements will be executed or neither statement will be executed.
As a matter of style many programmers always use curly braces to delimit the body of an if statement
even if the body contains only one statement:

if ( x < 10 )
{
y = x;
}

They do this because it is easy to introduce a logic error if additional statements are added to the body later
and the programmer forgets to add then required curly braces.

©2013 Richard L. Halterman Draft date: January 27, 2013

 5.5. THE IF/ELSE STATEMENT 82

The format of the following code

if ( x < 10 )
y = x;
z = x + 5;
implies that both assignments are part of the body of the if statement. Since
multiple statements making up the body must be in a compound statement within
curly braces, the compiler interprets the code fragment as if it had been written
if ( x < 10 )
y = x;
z = x + 5;
Such code will optionally execute the first assignment statement and always exe-
cute the second assignment statement.
The programmer probably meant to write it as
if ( x < 10 )
{
y = x;
z = x + 5;
}

The curly braces are optional if the body consists of a single statement. If the body consists of only one
statement and curly braces are not used, then the semicolon that terminates the statement in the body also
terminates the if statement. If curly braces are used to delimit the body, a semicolon is not required after
the body’s close curly brace.

5.5 The if/else Statement

One undesirable aspect of Listing 5.2 (betterdivision.cpp) is if the user enters a zero divisor, nothing is
printed. It may be better to provide some feedback to the user to indicate that the divisor provided cannot
be used. The if statement has an optional else clause that is executed only if the Boolean expression is
false. Listing 5.4 (betterfeedback.cpp) uses the if/else statement to provide the desired effect.

Listing 5.4: betterfeedback.cpp

1 #include <iostream>
2
3 using namespace std;
4
5 int main()
6 {
7 int dividend, divisor;
8
9 // Get two integers from the user
10 cout << "Please enter two integers to divide:";
11 cin >> dividend >> divisor;
12 // If possible, divide them and report the result

©2013 Richard L. Halterman Draft date: January 27, 2013

 5.5. THE IF/ELSE STATEMENT 83

13 if ( divisor != 0 )
14 cout << dividend << "/" << divisor << " = "
15 << dividend/divisor << endl;
16 else
17 cout << "Division by zero is not allowed" << endl;
18 }

A given program run will execute exactly one of either the if body or the else body. Unlike in
Listing 5.2 (betterdivision.cpp), a message is always displayed.

Please enter two integers to divide: 32 0

Division by zero is not allowed

The else clause contains an alternate body that is executed if the condition is false. The program’s flow
of execution is shown in Figure 5.2.

yes Is no
divisor ≠ 0?

do the division
castigate user
and print result

Figure 5.2: if/else flowchart

Listing 5.4 (betterfeedback.cpp) avoids the division by zero run-time error that causes the program to
terminate prematurely, but it still alerts the user that there is a problem. Another application may handle
the situation in a different way; for example, it may substitute some default value for divisor instead of
zero.
The general form of an if/else statement is

©2013 Richard L. Halterman Draft date: January 27, 2013

5.5. THE IF/ELSE STATEMENT 84

if ( condition )
if statement ;
else
else statement ;
• The reserved word if begins the if/else statement.
• The condition is a Boolean expression that determines whether or not the if statement or the else
statement will be executed. The condition must be in parentheses.
• The if statement is the statement to be executed if the condition is true. To make the if/else
statement more readable, the if statement is indented more spaces than the if line. This part of the
if statement is sometimes called the body of the if.
• The reserved word else begins the second part of the if/else statement.
• The else statement is the statement to be executed if the condition is false. To make the if/else
statement more readable, the else statement is indented more spaces than the else line. This part of
the if/else statement is sometimes called the body of the else.

The body of the else clause of an if/else statement may be a compound statement:
if ( x == y )
cout << x;
else
{
x = 0;
cout << y;
}

or the if body alone may be a compound statement:

if ( x == y )
{
cout << x;
x = 0;
}
else
cout << y;

or both parts may be compound:

if ( x == y )
{
cout << x;
x = 0;
}
else
{
cout << y;
y = 0;
}

©2013 Richard L. Halterman Draft date: January 27, 2013

 5.5. THE IF/ELSE STATEMENT 85

or, as in Listing 5.4 (betterfeedback.cpp), both the if body and the else body can be simple statements.

Remember, if you wish to associate more than one statement with the body of
the if or else, you must use a compound statement. Compound statements are
enclosed within curly braces ({}).

Due to the imprecise representation of floating-point numbers (see Listing 4.3 (imprecise10.cpp) in Sec-
tion 4.1), the equality operator (==) by itself should not be used when comparing floating-point expressions.
Listing 5.5 (samedifferent.cpp) uses an if/else statement to demonstrate the perils of using the equality
operator with floating-point quantities.

Listing 5.5: samedifferent.cpp

1 #include <iostream>
2 #include <iomanip>
3
4 using namespace std;
5
6 int main()
7 {
8 double d1 = 1.11 - 1.10,
9 d2 = 2.11 - 2.10;
10 cout << "d1 = " << d1 << endl;
11 cout << "d2 = " << d2 << endl;
12 if ( d1 == d2 )
13 cout << "Same" << endl;
14 else
15 cout << "Different" << endl;
16 cout << "d1 = " << setprecision(20) << d1 << endl;
17 cout << "d2 = " << setprecision(20) << d2 << endl;
18 }

In Listing 5.5 (samedifferent.cpp) the displayed values of d1 and d2 are rounded so they appear equivalent,
but internally the exact representations are slightly different. By including the header iomanip we can
use the setprecision stream manipulator to force cout to display more decimal places in the floating-
point number it prints. Observe from the output of Listing 5.5 (samedifferent.cpp) that the two quantities
that should be identically 0.01 are actually slightly different.

d1 = 0.01
d2 = 0.01
Different
d1 = 0.010000000000000009
d2 = 0.0099999999999997868

This result should not discourage you from using floating-point numbers where they truly are needed.
In Section 9.4.6 we will see how to handle floating-point comparisons properly.

©2013 Richard L. Halterman Draft date: January 27, 2013

5.6. COMPOUND BOOLEAN EXPRESSIONS 86

e1 e2 e1 && e2 e1 || e2 !e1
false false false false true
false true false true true
true false false true false
true true true true false

Table 5.3: Logical operators—e1 and e2 are Boolean expressions

5.6 Compound Boolean Expressions

Simple Boolean expressions, each involving one relational operator, can be combined into more complex
Boolean expressions using the logical operators && (and), || (or), and ! (not). A combination of two or
more Boolean expressions using logical operators is called a compound Boolean expression.
To introduce compound Boolean expressions, consider a computer science degree that requires, among
other computing courses, Operating Systems and Programming Languages. If we isolate those two courses,
we can say a student must successfully complete both Operating Systems and Programming Languages to
qualify for the degree. A student that passes Operating Systems but not Programming Languages will not
have met the requirements. Similarly, Programming Languages without Operating Systems is insufficient,
and a student completing neither Operating Systems nor Programming Languages surely does not qualify.
Logical AND works in exactly the same way. If e1 and e2 are two Boolean expressions, e1 && e2 is true
only if e1 and e2 are both true; if either one is false or both are false, the compound expression is false.
To illustrate logical OR, consider two mathematics courses, Differential Equations and Linear Algebra.
A computer science degree requires one of those two courses. A student who successfully completes
Differential Equations but does not take Linear Algebra meets the requirement. Similarly, a student may
take Linear Algebra but not Differential Equations. It is important to note the a student may elect to take
both Differential Equations and Linear Algebra (perhaps on the way to a mathematics minor), but the
requirement is no less fulfilled.
Logical OR works in a similar fashion. Given our Boolean expressions e1 and e2 , the compound ex-
pression e1 || e2 is false only if e1 and e2 are both false; if either one is true or both are true, the compound
expression is true. Note that logical OR is an inclusive or, not an exclusive or. In informal conversion we
often imply exclusive or in a statement like “Would you like cake or ice cream for dessert?” The implica-
tion is one or the other, not both. In computer programming the or is inclusive; if both subexpressions in
an or expression are true, the or expression is true.
Logical NOT simply reverses the truth value of the expression to which it is applied. If e is a true
Boolean expression, !e is false; if e is false, !e is true.
Table 5.3 is called a truth table. It shows all the combinations of truth values for two simple expres-
sions and the values of compound Boolean expressions built from applying the &&, ||, and ! C++ logical
operators.
Both && and || are binary operators; that is, they require two operands, both of which must be Boolean
expressions. Logical not (!) is a unary operator (see Section 4.1); it requires a single Boolean operand
immediately to its right.
Operator ! has higher precedence than both && and ||. && has higher precedence than ||. && and ||
are left associative; ! is right associative. && and || have lower precedence than any other binary operator
except assignment. This means the expression
x <= y && x <= z

©2013 Richard L. Halterman Draft date: January 27, 2013

5.6. COMPOUND BOOLEAN EXPRESSIONS 87

is evaluated
(x <= y) && (x <= z)

Some programmers prefer to use the parentheses as shown here even though they are not required. The
parentheses improve the readability of complex expressions, and the compiled code is no less efficient.

The relational operators such as < compare two operands. The result of the com-
parison is a Boolean value, which is freely convertible to an integer. The misappli-
cation of relational operators can lead to surprising results; consider, for example,
the expression
1 <= x <= 10
This expression is always true, regardless of the value of x! If the programmer’s
intent is to represent the mathematical notion of x falling within the range 1...10
inclusive, as in 1 ≤ x ≤ 10, the above C++ expression is not equivalent.
The expression
1 <= x <= 10
is evaluated as
(1 <= x) <= 10
If x is greater than or equal to one, the subexpression 1 ∼ <= ∼ x evaluates to
true, or integer 1. Integer 1, however, is always less than 10, so the overall expres-
sion is true. If instead x is less than one, the subexpression 1 ∼ <= ∼ x evaluates
to false, or integer 0. Integer 0 is always less than 10, so the overall expression is
true. The problem is due to the fact that C++ does not strictly distinguish between
Boolean and integer values.
The correct way to represent the mathematical notion of 1 ≤ x ≤ 10 is
1 <= x && x <= 10
In this case x must simultaneously be greater than or equal to 1 and less than
or equal to 10. The revised Boolean expression is a little more verbose than the
mathematical representation, but it is the correct formulation for C++.

The following section of code assigns the indicated values to a bool:

bool b;
int x = 10;
int y = 20;
b = (x == 10); // assigns true to b
b = (x != 10); // assigns false to b
b = (x == 10 && y == 20); // assigns true to b
b = (x != 10 && y == 20); // assigns false to b
b = (x == 10 && y != 20); // assigns false to b
b = (x != 10 && y != 20); // assigns false to b
b = (x == 10 || y == 20); // assigns true to b
b = (x != 10 || y == 20); // assigns true to b
b = (x == 10 || y != 20); // assigns true to b

©2013 Richard L. Halterman Draft date: January 27, 2013

5.6. COMPOUND BOOLEAN EXPRESSIONS 88

b = (x != 10 || y != 20); // assigns false to b

Convince yourself that the following expressions are equivalent:

(x != y)
!(x == y)
(x < y || x > y)

In the expression e1 && e2 both subexpressions e1 and e2 must be true for the overall expression to be
true. Since the && operator evaluates left to right, this means that if e1 is false, there is no need to evaluate
e2 . If e1 is false, no value of e2 can make the expression e1 && e2 true. The logical and operator first tests the
expression to its left. If it finds the expression to be false, it does not bother to check the right expression.
This approach is called short-circuit evaluation. In a similar fashion, in the expression e1 || e2 , if e1 is
true, then it does not matter what value e2 has—a logical or expression is true unless both subexpressions
are false. The || operator uses short-circuit evaluation also.
Why is short-circuit evaluation important? Two situations show why it is important to consider:

• The order of the subexpressions can affect performance. When a program is running, complex ex-
pressions require more time for the computer to evaluate than simpler expressions. We classify an
expression that takes a relatively long time to evaluate as an expensive expression. If a compound
Boolean expression is made up of an expensive Boolean subexpression and an less expensive Boolean
subexpression, and the order of evaluation of the two expressions does not affect the behavior of the
program, then place the more expensive Boolean expression second. If the first subexpression is false
and && is being used, then the expensive second subexpression is not evaluated; if the first subex-
pression is true and || is being used, then, again, the expensive second subexpression is avoided.

• Subexpressions can be ordered to prevent run-time errors. This is especially true when one of the
subexpressions depends on the other in some way. Consider the following expression:

(x != 0) && (z/x > 1)

Here, if x is zero, the division by zero is avoided. If the subexpressions were switched, a run-time
error would result if x is zero.

©2013 Richard L. Halterman Draft date: January 27, 2013

 5.7. NESTED CONDITIONALS 89

Arity Operators Associativity

unary (post) ++, (post) --, static_cast right
unary (pre) ++, (pre) --, !, +, - right
binary *, /, % left
binary +, - left
binary <<, >> left
binary >, <, >=, <= left
binary ==, != left
binary && left
binary || left
binary =, +=, /=, *=, /=, %= left

Table 5.4: Precedence of C++ Operators (High to Low)

Suppose you wish to print the word “OK” if a variable x is 1, 2, or 3. An informal

translation from English might yield:
if ( x == 1 || 2 || 3 )
cout << "OK" << endl;
Unfortunately, x’s value is irrelevant; the code always prints the word “OK.” Since
the == operator has lower precedence than ||, the expression
x == 1 || 2 || 3
is interpreted as
(x == 1) || 2 || 3
The expression x == 1 is either true or false, but integer 2 is always interpreted
as true, and integer 3 is interpreted as true is as well.
The correct statement would be
if ( x == 1 || x == 2 || x == 3 )
cout << "OK" << endl;
The revised Boolean expression is more verbose and less similar to the English
rendition, but it is the correct formulation for C++.

Our current list of C++ operators is shown in Table 5.4.

5.7 Nested Conditionals

The statements in the body of the if or the else may be any C++ statements, including other if/else
statements. These nested if statements can be used to develop arbitrarily complex control flow logic.
Consider Listing 5.6 (checkrange.cpp) that determines if a number is between 0 and 10, inclusive.

Listing 5.6: checkrange.cpp

1 #include <iostream>
2
3 using namespace std;

©2013 Richard L. Halterman Draft date: January 27, 2013

 5.7. NESTED CONDITIONALS 90

4
5 int main()
6 {
7 int value;
8 cout << "Please enter an integer value in the range 0...10: ";
9 cin >> value;
10 if ( value >= 0 ) // First check
11 if ( value <= 10 ) // Second check
12 cout << "In range";
13 cout << endl;
14 }

Listing 5.6 (checkrange.cpp) behaves as follows:

• The first condition is checked. If value is less than zero, the second condition is not evaluated and
the statement following the outer if is executed. The statement after the outer if simply prints a
newline.

• If the first condition finds value to be greater than or equal to zero, the second condition is then
checked. If the second condition is met, the In range message is displayed; otherwise, it is not.

Both conditions of this nested if must be met for the message to be printed. Said another way, the
first condition and the second condition must be met for the In range message to be printed. From this
perspective, the program can be rewritten to behave the same way with only one if statement, as Listing 5.7
(newcheckrange.cpp) shows.

Listing 5.7: newcheckrange.cpp

1 #include <iostream>
2
3 using namespace std;
4
5 int main()
6 {
7 int value;
8 cout << "Please enter an integer value in the range 0...10: ";
9 cin >> value;
10 if ( value >= 0 && value <= 10 )
11 cout << "In range";
12 cout << endl;
13 }

Listing 5.7 (newcheckrange.cpp) uses a logical && to check both conditions at the same time. Its logic
is simpler, using only one if statement, at the expense of a slightly more complex Boolean expression in
its condition. The second version is preferable here, because simpler logic is usually a desirable goal.
Sometimes a program’s logic cannot be simplified as in Listing 5.7 (newcheckrange.cpp). In Listing 5.8
(enhancedcheckrange.cpp) only one if statement is insufficient to implement the necessary behavior.

Listing 5.8: enhancedcheckrange.cpp

1 #include <iostream>
2
3 using namespace std;

©2013 Richard L. Halterman Draft date: January 27, 2013

 5.7. NESTED CONDITIONALS 91

4
5 int main()
6 {
7 int value;
8 cout << "Please enter an integer value in the range 0...10: ";
9 cin >> value;
10 if ( value >= 0 ) // First check
11 if ( value <= 10 ) // Second check
12 cout << value << " is acceptable";
13 else
14 cout << value << " is too large";
15 else
16 cout << value << " is too small";
17 cout << endl;
18 }

Listing 5.8 (enhancedcheckrange.cpp) provides a more specific message instead of a simple notification
of acceptance. Exactly one of three messages is printed based on the value of the variable. A single if or
if/else statement cannot choose from among more than two different execution paths.
Listing 5.9 (troubleshoot.cpp) implements a very simple troubleshooting program that (an equally sim-
ple) computer technician might use to diagnose an ailing computer.

Listing 5.9: troubleshoot.cpp

1 #include <iostream>
2
3 using namespace std;
4
5 int main()
6 {
7 cout << "Help! My computer doesn't work!" << endl;
8 char choice;
9 cout << "Does the computer make any sounds "
10 << "(fans, etc.) or show any lights? (y/n):";
11 cin >> choice;
12 // The troubleshooting control logic
13 if ( choice == 'n' ) // The computer does not have power
14 {
15 cout << "Is it plugged in? (y/n):";
16 cin >> choice;
17 if ( choice == 'n' ) // It is not plugged in, plug it in
18 {
19 cout << "Plug it in. If the problem persists, "
20 << "please run this program again." << endl;
21 }
22 else // It is plugged in
23 {
24 cout << "Is the switch in the \"on\" position? (y/n):";
25 cin >> choice;
26 if ( choice == 'n' ) // The switch is off, turn it on!
27 {
28 cout << "Turn it on. If the problem persists, "
29 << "please run this program again." << endl;
30 }

©2013 Richard L. Halterman Draft date: January 27, 2013

 5.7. NESTED CONDITIONALS 92

31 else // The switch is on

32 {
33 cout << "Does the computer have a fuse? (y/n):";
34 cin >> choice;
35 if ( choice == 'n' ) // No fuse
36 {
37 cout << "Is the outlet OK? (y/n):";
38 if ( choice == 'n' ) // Fix outlet
39 {
40 cout << "Check the outlet's circuit "
41 << "breaker or fuse. Move to a "
42 << "new outlet, if necessary. "
43 << "If the problem persists, "
44 << "please run this program "
45 << "again." << endl;
46 }
47 else // Beats me!
48 {
49 cout << "Please consult a service "
50 << "technician." << endl;
51 }
52 }
53 else // Check fuse
54 {
55 cout << "Check the fuse. Replace if "
56 << "necessary. If the problem "
57 << "persists, then "
58 << "please run this program again."
59 << endl;
60 }
61 }
62 }
63 }
64 else // The computer has power
65 {
66 cout << "Please consult a service technician." << endl;
67 }
68 }

This very simple troubleshooting program attempts to diagnose why a computer does not work. The
potential for enhancement is unlimited, but this version only deals with power issues that have simple fixes.
Notice that if the computer has power (fan or disk drive makes sounds or lights are visible), the program
indicates that help should be sought elsewhere! The decision tree capturing the basic logic of the program
is shown in Figure 5.3. The steps performed are:

1. Is it plugged in? This simple fix is sometimes overlooked.

2. Is the switch in the on position? This is another simple fix.

3. If applicable, is the fuse blown? Some computer systems have a user-serviceable fuse that can blow
out during a power surge. (Most newer computers have power supplies that can handle power surges
and have no user-serviceable fuses.)

4. Is there power at the receptacle? Perhaps the outlet’s circuit breaker or fuse has a problem.

©2013 Richard L. Halterman Draft date: January 27, 2013

5.7. NESTED CONDITIONALS 93

Figure 5.3: Decision tree for troubleshooting a computer system

The easiest checks are made first. Progressively more difficult checks are introduced as the program runs.
Based on your experience with troubleshooting computers that do not run properly, you may be able to
think of many enhancements to this simple program.
Note that in Listing 5.9 (troubleshoot.cpp) curly braces are used in many places where they strictly are
not necessary. Their inclusion in Listing 5.9 (troubleshoot.cpp) improves the readability of the program and
makes the logic easier to understand. Even if you do not subscribe to the philosophy of using curly braces
for every if/else body, it is a good idea to use them in situations that improve the code’s readability.
Recall the time conversion program in Listing 4.11 (timeconv.cpp). If the user enters 10000, the
program runs as follows:

Please enter the number of seconds:10000

2 hours 46 minutes 40 seconds

and if the user enters 9961, the program prints:

Please enter the number of seconds:9961

2 hours 46 minutes 1 second

©2013 Richard L. Halterman Draft date: January 27, 2013

 5.7. NESTED CONDITIONALS 94

Suppose we wish to improve the English presentation by not using abbreviations. If we spell out hours,
minutes, and seconds, we must be careful to use the singular form hour, minute, or second when the
corresponding value is one. Listing 5.10 (timeconvcond1.cpp) uses if/else statements to express to
time units with the correct number.

Listing 5.10: timeconvcond1.cpp

1 // File timeconvcond1.cpp
2
3 #include <iostream>
4
5 using namespace std;
6
7 int main()
8 {
9 // Some useful conversion constants
10 const int SECONDS_PER_MINUTE = 60,
11 SECONDS_PER_HOUR = 60*SECONDS_PER_MINUTE; // 3600
12 int hours, minutes, seconds;
13 cout << "Please enter the number of seconds:";
14 cin >> seconds;
15 // First, compute the number of hours in the given number
16 // of seconds
17 hours = seconds / SECONDS_PER_HOUR; // 3600 seconds = 1 hours
18 // Compute the remaining seconds after the hours are
19 // accounted for
20 seconds = seconds % SECONDS_PER_HOUR;
21 // Next, compute the number of minutes in the remaining
22 // number of seconds
23 minutes = seconds / SECONDS_PER_MINUTE; // 60 seconds = 1 minute
24 // Compute the remaining seconds after the minutes are
25 // accounted for
26 seconds = seconds % SECONDS_PER_MINUTE;
27 // Report the results
28 cout << hours;
29 // Decide between singular and plural form of hours
30 if ( hours == 1 )
31 cout << " hour ";
32 else
33 cout << " hours ";
34 cout << minutes;
35 // Decide between singular and plural form of minutes
36 if ( minutes == 1 )
37 cout << " minute ";
38 else
39 cout << " minutes ";
40 cout << seconds;
41 // Decide between singular and plural form of seconds
42 if ( seconds == 1 )
43 cout << " second";
44 else
45 cout << " seconds";
46 cout << endl;
47 }

©2013 Richard L. Halterman Draft date: January 27, 2013

 5.7. NESTED CONDITIONALS 95

The if/else statements within Listing 5.10 (timeconvcond1.cpp) are responsible for printing the correct
version—singular or plural—for each time unit. One run of Listing 5.10 (timeconvcond1.cpp) produces

Please enter the number of seconds:10000

2 hours 46 minutes 40 seconds

All the words are plural since all the value are greater than one. Another run produces

Please enter the number of seconds:9961

2 hours 46 minutes 1 second

Note the word second is singular as it should be.

Please enter the number of seconds:3601

1 hour 0 minutes 1 second

Here again the printed words agree with the number of the value they represent.
An improvement to Listing 5.10 (timeconvcond1.cpp) would not print a value and its associated time
unit if the value is zero. Listing 5.11 (timeconvcond2.cpp) adds this feature.

Listing 5.11: timeconvcond2.cpp

1 // File timeconvcond1.cpp
2
3 #include <iostream>
4
5 using namespace std;
6
7 int main()
8 {
9 // Some useful conversion constants
10 const int SECONDS_PER_MINUTE = 60,
11 SECONDS_PER_HOUR = 60*SECONDS_PER_MINUTE; // 3600
12 int hours, minutes, seconds;
13 cout << "Please enter the number of seconds:";
14 cin >> seconds;
15 // First, compute the number of hours in the given number
16 // of seconds
17 hours = seconds / SECONDS_PER_HOUR; // 3600 seconds = 1 hours
18 // Compute the remaining seconds after the hours are
19 // accounted for
20 seconds = seconds % SECONDS_PER_HOUR;
21 // Next, compute the number of minutes in the remaining
22 // number of seconds
23 minutes = seconds / SECONDS_PER_MINUTE; // 60 seconds = 1 minute

©2013 Richard L. Halterman Draft date: January 27, 2013

 5.7. NESTED CONDITIONALS 96

24 // Compute the remaining seconds after the minutes are

25 // accounted for
26 seconds = seconds % SECONDS_PER_MINUTE;
27 // Report the results
28 if ( hours > 0 ) // Print hours at all?
29 {
30 cout << hours;
31 // Decide between singular and plural form of hours
32 if ( hours == 1 )
33 cout << " hour ";
34 else
35 cout << " hours ";
36 }
37 if ( minutes > 0 ) // Print minutes at all?
38 {
39 cout << minutes;
40 // Decide between singular and plural form of minutes
41 if ( minutes == 1 )
42 cout << " minute ";
43 else
44 cout << " minutes ";
45 }
46 // Print seconds at all?
47 if ( seconds > 0 || (hours == 0 && minutes == 0 && seconds == 0) )
48 {
49 cout << seconds;
50 // Decide between singular and plural form of seconds
51 if ( seconds == 1 )
52 cout << " second";
53 else
54 cout << " seconds";
55 }
56 cout << endl;
57 }

In Listing 5.11 (timeconvcond2.cpp) each code segment responsible for printing a time value and its English
word unit is protected by an if statement that only allows the code to execute if the time value is greater
than zero. The exception is in the processing of seconds: if all time values are zero, the program should
print 0 seconds. Note that each of the if/else statements responsible for determining the singular or
plural form is nested within the if statement that determines whether or not the value will be printed at all.
One run of Listing 5.11 (timeconvcond2.cpp) produces

Please enter the number of seconds:10000

2 hours 46 minutes 40 seconds

All the words are plural since all the value are greater than one. Another run produces

Please enter the number of seconds:9961

2 hours 46 minutes 1 second

©2013 Richard L. Halterman Draft date: January 27, 2013

 5.8. MULTI-WAY IF/ELSE STATEMENTS 97

Note the word second is singular as it should be.

Please enter the number of seconds:3601

1 hour 1 second

Here again the printed words agree with the number of the value they represent.

Please enter the number of seconds:7200

2 hours

Another run produces:

Please enter the number of seconds:60

1 minute

Finally, the following run shows that the program handles zero seconds properly:

Please enter the number of seconds:0

0 seconds

5.8 Multi-way if/else Statements

A simple if/else statement can select from between two execution paths. Suppose we wish to choose
one execution path from among several possible paths, as shown in Figure 5.4? Listing 5.8 (enhancedcheckrange.cpp)
showed how to select from among three options. What if exactly one of many actions should be taken?
Nested if/else statements are required, and the form of these nested if/else statements is shown in
Listing 5.12 (digittoword.cpp).

Listing 5.12: digittoword.cpp

1 #include <iostream>
2
3 using namespace std;
4
5 int main()
6 {
7 int value;
8 cout << "Please enter an integer in the range 0...5: ";
9 cin >> value;
10 if ( value < 0 )
11 cout << "Too small";

©2013 Richard L. Halterman Draft date: January 27, 2013

 5.8. MULTI-WAY IF/ELSE STATEMENTS 98

Figure 5.4: Flowchart with multiple optional execution pathways

12 else
13 if ( value == 0 )
14 cout << "zero";
15 else
16 if ( value == 1 )
17 cout << "one";
18 else
19 if ( value == 2 )
20 cout << "two";
21 else
22 if ( value == 3 )
23 cout << "three";
24 else
25 if ( value == 4 )
26 cout << "four";
27 else
28 if ( value == 5 )
29 cout << "five";
30 else
31 cout << "Too large";
32 cout << endl;
33 }

Observe the following about Listing 5.12 (digittoword.cpp):

• It prints exactly one of eight messages depending on the user’s input.

• Notice that each if body contains a single printing statement and each else body, except the last
one, contains an if statement. The control logic forces the program execution to check each condi-
tion in turn. The first condition that matches wins, and its corresponding if body will be executed.
If none of the conditions are true, the last else’s Too large message will be printed.

©2013 Richard L. Halterman Draft date: January 27, 2013

 5.8. MULTI-WAY IF/ELSE STATEMENTS 99

• No curly braces are necessary to delimit the if or else bodies since each body contains only a
single statement (although a single deeply nested if/else statement is a mighty big statement).

Listing 5.12 (digittoword.cpp) is formatted according to the conventions used in earlier examples. As
a consequence, the mass of text drifts to the right as more conditions are checked. A commonly used
alternative style, shown in Listing 5.13 (restyleddigittoword.cpp), avoids this rightward drift.

Listing 5.13: restyleddigittoword.cpp

1 #include <iostream>
2
3 using namespace std;
4
5 int main()
6 {
7 int value;
8 cout << "Please enter an integer in the range 0...5: ";
9 cin >> value;
10 if ( value < 0 )
11 cout << "Too small";
12 else if ( value == 0 )
13 cout << "zero";
14 else if ( value == 1 )
15 cout << "one";
16 else if ( value == 2 )
17 cout << "two";
18 else if ( value == 3 )
19 cout << "three";
20 else if ( value == 4 )
21 cout << "four";
22 else if ( value == 5 )
23 cout << "five";
24 else
25 cout << "Too large";
26 cout << endl;
27 }

Based on our experience so far, the formatting of Listing 5.13 (restyleddigittoword.cpp) somewhat hides
the true structure of the program’s logic, but this style of formatting multi-way if/else statements is
so common that it is regarded as acceptable by most programmers. The sequence of else if lines all
indented to the same level identifies this construct as a multi-way if/else statement.
Listing 5.14 (datetransformer.cpp) uses a multi-way if/else to transform a numeric date in month/-
day format to an expanded US English form and an international Spanish form; for example, 2/14 would
be converted to February ∼ 14 and 14 ∼ febrero.

Listing 5.14: datetransformer.cpp

1 #include <iostream>
2
3 using namespace std;
4
5 int main()
6 {
7 cout << "Please enter the month and day as numbers: ";

©2013 Richard L. Halterman Draft date: January 27, 2013

 5.8. MULTI-WAY IF/ELSE STATEMENTS 100

8 int month, day;

9 cin >> month >> day;
10 // Translate month into English
11 if ( month == 1 )
12 cout << "January";
13 else if ( month == 2 )
14 cout << "February";
15 else if ( month == 3 )
16 cout << "March";
17 else if ( month == 4 )
18 cout << "April";
19 else if ( month == 5 )
20 cout << "May";
21 else if ( month == 6 )
22 cout << "June";
23 else if ( month == 7 )
24 cout << "July";
25 else if ( month == 8 )
26 cout << "August";
27 else if ( month == 9 )
28 cout << "September";
29 else if ( month == 10 )
30 cout << "October";
31 else if ( month == 11 )
32 cout << "November";
33 else
34 cout << "December";
35 // Add the day
36 cout << " " << day << " or " << day << " ";
37 // Translate month into Spanish
38 if ( month == 1 )
39 cout << "enero";
40 else if ( month == 2 )
41 cout << "febrero";
42 else if ( month == 3 )
43 cout << "marzo";
44 else if ( month == 4 )
45 cout << "abril";
46 else if ( month == 5 )
47 cout << "mayo";
48 else if ( month == 6 )
49 cout << "junio";
50 else if ( month == 7 )
51 cout << "julio";
52 else if ( month == 8 )
53 cout << "agosto";
54 else if ( month == 9 )
55 cout << "septiembre";
56 else if ( month == 10 )
57 cout << "octubre";
58 else if ( month == 11 )
59 cout << "noviembre";
60 else
61 cout << "diciembre";
62 cout << endl;

©2013 Richard L. Halterman Draft date: January 27, 2013

 5.9. ERRORS IN CONDITIONAL STATEMENTS 101

63 }

A sample run of Listing 5.14 (datetransformer.cpp) is shown here:

Please enter the month and day as numbers: 5 20

May 20 or 20 mayo

5.9 Errors in Conditional Statements

Consider Listing 5.15 (badequality.cpp).

Listing 5.15: badequality.cpp

1 #include <iostream>
2
3 using namespace std;
4
5 int main()
6 {
7 int input;
8 cout << "Please enter an integer:";
9 cin >> input;
10 if ( input = 2 )
11 cout << "two" << endl;
12 cout << "You entered " << input << endl;
13 }

Listing 5.15 (badequality.cpp) demonstrates a common mistake—using the assignment operator where
the equality operator is intended. This program, when run, always prints the message “two” and insists the
user entered 2 regardless of the actual input. Recall from Section 4.3 that the assignment expression has a
value. The value of an assignment expression is same as the value that is assigned; thus, the expression
input = 2

has the value 2. When you consider also that every integer can be treated as a Boolean value (see Sec-
tion 5.1) and any non-zero value is interpreted as true, you can see that the condition of if statement
if ( input = 2 )
cout << "two" << endl;

is always true. Additionally, the variable input is always assigned the value 2.
Because it is such a common error, most C++ compilers can check for such misuse of assignment. At
warning Level 4, for example, Visual C++ will issue a warning when assignment appears where a conditional
expression is expected:

warning C4706: assignment within conditional expression

This check is not performed by default because sometimes the use of assignment within a conditional
expression is warranted. We have no reason to do so, however, so it is a good idea to allow the Visual C++
compiler to perform the extra check.

©2013 Richard L. Halterman Draft date: January 27, 2013

5.10. SUMMARY 102

Carefully consider each compound conditional used, such as

value > 0 && value <= 10

found in Listing 5.7 (newcheckrange.cpp). Confusing logical and and logical or is a common programming
error. The Boolean expression
x > 0 || x <= 10

is always true, no matter what value is assigned to the variable x. A Boolean expression that is always true
is known as a tautology. Think about it. If x is an int, what value could the variable x assume that would
make this Boolean expression false? Regardless of its value, one or both of the subexpressions will be true,
so the compound logical or expression is always true. This particular or expression is just a complicated
way of expressing the value true.
Another common error is contriving compound Boolean expressions that are always false, known as
contradictions. Suppose you wish to exclude values from a given range; for example, reject values in the
range 0...10 and accept all other numbers. Is the Boolean expression in the following code fragment up to
the task?
// All but 0, 1, 2, ..., 10
if ( value < 0 && value > 10 )
/* Code to execute goes here . . . */

A closer look at the condition reveals it can never be true. What number can be both less than zero and
greater than ten at the same time? None can, of course, so the expression is a contradiction and a compli-
cated way of expressing false. To correct this code fragment, replace the && operator with ||.

5.10 Summary

• The bool data type represents the values true and false.

• The name bool comes from Boolean algebra, the mathematical study of operations on truth values.

• In C++ the value true is represented by the integer one, and false is represented by zero.

• Any integer value except zero is treated as true.

• Integers and bools are interchangeable and can be assigned to each other.

• Expressions involving the relational operators (==, !=, <, >, <=, and >=) evaluate to Boolean values.

• ! is the unary not operator.

• Boolean expressions can be combined via && (logical AND) and || (logical OR).

• The if statement can be used to optionally execute statements.

• A compound statement is a sequence of statements within a pair of curly braces.

• The if statement has an optional else clause to require the selection between two alternate paths
of execution.

• The if/else statements can be nested to achieve arbitrary complexity.

©2013 Richard L. Halterman Draft date: January 27, 2013

5.11. EXERCISES 103

• The bodies of if/else statements should be indented to aid human readers. Indentation doe not
affect the logic of the program; when multiple statements are to be part of the body of an if or else,
the statements must be part of a compound statement.
• Beware placing a semicolon immediately after the close parenthesis of an if statement’s condition.
• Complex Boolean expressions require special attention, as they are easy to get wrong.
•
•
•

5.11 Exercises

1. What values can a variable of type bool assume?

2. Where does the term bool originate?
3. What is the integer equivalent to true in C++?
4. What is the integer equivalent to false in C++?
5. Is the value -16 interpreted as true or false?
6. May an integer value be assigned to a bool variable?
7. Can true be assigned to an int variable?
8. Given the following declarations:
int x = 3, y = 5, z = 7;
bool b1 = true, b2 = false, b3 = x == 3, b4 = y < 3;

evaluate the following Boolean expressions:

(a) x == 3
(b) x < y
(c) x >= y
(d) x <= y
(e) x != y - 2
(f) x < 10
(g) x >= 0 && x < 10
(h) x < 0 && x < 10
(i) x >= 0 && x < 2
(j) x < 0 || x < 10
(k) x > 0 || x < 10
(l) x < 0 || x > 10
(m) b1

©2013 Richard L. Halterman Draft date: January 27, 2013

5.11. EXERCISES 104

(n) !b1
(o) !b2
(p) b1 && b2

9. Express the following Boolean expressions in simpler form; that is, use fewer operators. x is an int.
(a) !(x == 2)
(b) x < 2 || x == 2
(c) !(x < y)
(d) !(x <= y)
(e) x < 10 && x > 20
(f) x > 10 || x < 20
(g) x != 0
(h) x == 0
10. What is the simplest tautology?
11. What is the simplest contradiction?
12. Write a C++ program that requests an integer value from the user. If the value is between 1 and 100
inclusive, print “OK;” otherwise, do not print anything.
13. Write a C++ program that requests an integer value from the user. If the value is between 1 and 100
inclusive, print “OK;” otherwise, print “Out of range.”
14. Write a C++ program that requests an integer value from the user. If the value is between 1 and 100
inclusive,
15. The following program attempts to print a message containing the the English word correspond-
ing to a given integer input. For example, if the user enters the value 3, the program should print
"You entered a three". In its current state, the program contains logic errors. Locate the
problems and repair them so the program will work as expected.
#include <iostream>

using namespace std;

int main()
{
cout << "Please in value in the range 1...5: ";
int value;
cin >> value;
// Translate number into its English word
if ( month == 1 )
cout << "You entered a";
cout << "one";
cout << endl;
else if ( month == 2 )
cout << "You entered a";
cout << "two";
cout << endl;

©2013 Richard L. Halterman Draft date: January 27, 2013

5.11. EXERCISES 105

else if ( month == 3 )
cout << "You entered a";
cout << "three";
cout << endl;
else if ( month == 4 )
cout << "You entered a";
cout << "four";
cout << endl;
else if ( month == 5 )
cout << "You entered a";
cout << "five";
cout << endl;
else // Value out of range
cout << "You entered a";
cout << "value out of range";
cout << endl;
}

16. Consider the following section of C++ code:

// i, j, and k are ints
if ( i < j )
{
if ( j < k )
i = j;
else
j = k;
}
else
{
if ( j > k )
j = i;
else
i = k;
}
cout << "i = " << i << " j = " << j << " k = " << k << endl;

What will the code print if the variables i, j, and k have the following values?
(a) i is 3, j is 5, and k is 7
(b) i is 3, j is 7, and k is 5
(c) i is 5, j is 3, and k is 7
(d) i is 5, j is 7, and k is 3
(e) i is 7, j is 3, and k is 5
(f) i is 7, j is 5, and k is 3
17. Consider the following C++ program that prints one line of text:
#include <iostream>
using namespace std;

©2013 Richard L. Halterman Draft date: January 27, 2013

5.11. EXERCISES 106

int main()
{
int input;
cin >> input;
if ( input < 10 )
{
if ( input != 5 )
cout << "wow ";
else
input++;
}
else
{
if ( input == 17 )
input += 10;
else
cout << "whoa ";
}
cout << input << endl;
}

What will the program print if the user provides the following input?
(a) 3
(b) 21
(c) 5
(d) 17
(e) -5
18. Why does the following section of code always print "ByeHi"?
int x;
cin >> x;
if ( x < 0 );
cout << "Bye";
cout << "Hi" << endl;

19. Write a C++ program that requests five integer values from the user. It then prints the maximum and
minimum values entered. If the user enters the values 3, 2, 5, 0, and 1, the program would indicate
that 5 is the maximum and 0 is the minimum. Your program should handle ties properly; for example,
if the user enters 2, 4 2, 3 and 3, the program should report 2 as the minimum and 4 as maximum.
20. Write a C++ program that requests five integer values from the user. It then prints one of two
things: if any of the values entered are duplicates, it prints "DUPLICATES"; otherwise, it prints
"ALL UNIQUE".
21. Write a C++ program that ...

©2013 Richard L. Halterman Draft date: January 27, 2013

 107

Chapter 6

Iteration

Iteration repeats the execution of a sequence of code. Iteration is useful for solving many programming
problems. Iteration and conditional execution form the basis for algorithm construction.

6.1 The while Statement

Listing 6.1 (counttofive.cpp) counts to five by printing a number on each output line.

Listing 6.1: counttofive.cpp

1 #include <iostream>
2
3 using namespace std;
4
5 int main()
6 {
7 cout << 1 << endl;
8 cout << 2 << endl;
9 cout << 3 << endl;
10 cout << 4 << endl;
11 cout << 5 << endl;
12 }

When compiled and run, this program displays

1
2
3
4
5

How would you write the code to count to 10,000? Would you copy, paste, and modify 10,000 printing
statements? You could, but that would be impractical! Counting is such a common activity, and computers

©2013 Richard L. Halterman Draft date: January 27, 2013

 6.1. THE WHILE STATEMENT 108

routinely count up to very large values, so there must be a better way. What we really would like to do is
print the value of a variable (call it count), then increment the variable (count++), and repeat this process
until the variable is large enough (count == 5 or count == 10000). This process of executing the
same section of code over and over is known as iteration, or looping, and C++ can accomplish it in several
different ways.
Listing 6.2 (iterativecounttofive.cpp) uses a while statement to count to five:

Listing 6.2: iterativecounttofive.cpp

1 #include <iostream>
2
3 using namespace std;
4
5 int main()
6 {
7 int count = 1; // Initialize counter
8 while ( count <= 5 )
9 {
10 cout << count << endl; // Display counter, then
11 count++; // Increment counter
12 }
13 }

Listing 6.2 (iterativecounttofive.cpp) uses a while statement to display a variable that is counting
up to five. Unlike the approach taken in Listing 6.1 (counttofive.cpp), it is trivial to modify Listing 6.2
(iterativecounttofive.cpp) to count up to 10,000—just change the literal value 5 to 10000.
The line
while ( count <= 5 )

begins the while statement. The expression within the parentheses must be a Boolean expression. If the
Boolean expression is true when the while statement is executed, the body of the while statement is
executed, and the body is executed repeatedly as long as the Boolean expression remains true.
The statements
cout << count << endl;
count++;

constitute the body of the while statement. The curly braces are necessary since more than one statement
makes up the body.
The while statement has the general form:

while ( condition )
statement ;

• The reserved word while begins the while statement.

• The Boolean expression condition determines whether the body will be (or will continue to be) exe-
cuted. The expression must be enclosed within parentheses as shown.

©2013 Richard L. Halterman Draft date: January 27, 2013

6.1. THE WHILE STATEMENT 109

• The statement is the statement to be executed while the Boolean expression is true. The statement
makes up the body of the while statement. The statement may be a compound statement (multiple
statements enclosed within curly braces, see Section 5.4).

Except for the reserved word while instead of if, while statements look identical to if statements.
Sometimes beginning programmers confuse the two or accidentally type if when they mean while or
vice-versa. Usually the very different behavior of the two statements reveals the problem immediately;
however, sometimes, especially in nested complex logic, this mistake can be hard to detect.
Figure 6.1 shows how program execution flows through Listing 6.2 (iterativecounttofive.cpp).

Figure 6.1: while flowchart for Listing 6.2 (iterativecounttofive.cpp)

The condition is checked before the body is executed, and then each time after the body has been
executed. If the condition is initially false the body is not executed. If the condition is initially true, the
body is executed repeatedly until the condition becomes false, at which point the loop terminates. Observe
that the body may never be executed if the Boolean expression in the condition is initially false.
Listing 6.3 (countup.cpp) counts up from zero as long as the user wishes to do so.

©2013 Richard L. Halterman Draft date: January 27, 2013

 6.1. THE WHILE STATEMENT 110

Listing 6.3: countup.cpp

1 /*
2 * Counts up from zero. The user continues the count by entering
3 * 'Y'. The user discontinues the count by entering 'N'.
4 */
5
6 #include <iostream>
7
8 using namespace std;
9
10 int main()
11 {
12 char input; // The users choice
13 int count = 0; // The current count
14 bool done = false; // We are not done
15
16 while ( !done )
17 {
18
19 // Print the current value of count
20 cout << count << endl;
21 cout << "Please enter \"Y\" to continue or \"N\" to quit: ";
22 cin >> input;
23 // Check for "bad" input
24 if ( input != 'Y' && input != 'y' && input != 'N' && input != 'n' )
25 cout << "\"" << input << "\""
26 << " is not a valid choice" << endl;
27 else if ( input == 'Y' || input == 'y' )
28 count++; // Keep counting
29 else if ( input == 'N' || input == 'n' )
30 done = true; // Quit the loop
31 }
32 }

A sample run of Listing 6.3 (countup.cpp) produces

©2013 Richard L. Halterman Draft date: January 27, 2013

 6.1. THE WHILE STATEMENT 111

0
Please enter "Y" to continue or "N" to quit: y
1
Please enter "Y" to continue or "N" to quit: y
2
Please enter "Y" to continue or "N" to quit: y
3
Please enter "Y" to continue or "N" to quit: q
"q" is not a valid choice
3
Please enter "Y" to continue or "N" to quit: r
"r" is not a valid choice
3
Please enter "Y" to continue or "N" to quit: W
"W" is not a valid choice
3
Please enter "Y" to continue or "N" to quit: Y
4
Please enter "Y" to continue or "N" to quit: y
5
Please enter "Y" to continue or "N" to quit: n

In Listing 6.3 (countup.cpp) the expression

input != 'Y' && input != 'y' && input != 'N' && input != 'n'

is true if the character variable input is not equal to one of the listed character literals. The Boolean
variable done controls the loop’s execution. It is important to note that the expression
!done

inside the while’s condition evaluates to the opposite truth value of the variable done; the expression
does not affect the value of done. In other words, the ! operator applied to a variable does not modify the
variable’s value. In order to actually change the variable done, you would need to reassign it, as in
done = !done; // Invert the truth value

For Listing 6.3 (countup.cpp) we have no need to invert its value. We ensure that its value is false initially
and then make it true when the user enters a capital or lower-case N.
Listing 6.4 (addnonnegatives.cpp) is a program that allows a user to enter any number of non-negative
integers. When the user enters a negative value, the program no longer accepts input, and it displays the
sum of all the non-negative values. If a negative number is the first entry, the sum is zero.

Listing 6.4: addnonnegatives.cpp

1 /*
2 * Allow the user to enter a sequence of non-negative
3 * integers. The user ends the list with a negative
4 * integer. At the end the sum of the non-negative
5 * integers entered is displayed. The program prints
6 * zero if the users no non-negative integers.

©2013 Richard L. Halterman Draft date: January 27, 2013

 6.1. THE WHILE STATEMENT 112

7 */
8
9 #include <iostream>
10
11 using namespace std;
12
13 int main()
14 {
15 int input = 0, // Ensure the loop is entered
16 sum = 0; // Initialize sum
17
18 // Request input from the user
19 cout << "Enter numbers to sum, negative number ends list:";
20
21 while ( input >= 0 ) // A negative number exits the loop
22 {
23 cin >> input; // Get the value
24 if ( input >= 0 )
25 sum += input; // Only add it if it is non-negative
26 }
27 cout << "Sum = " << sum << endl; // Display the sum
28 }

The initialization of input to zero coupled with the condition input >= 0 of the while guarantees
that the body of the while loop will be executed at least once. The if statement ensures that a negative
entry will not be added to sum. (Could the condition have used > instead of >= and achieved the same
results?) Upon entry of a negative value, sum will not be updated and the condition of the while will no
longer be true. The loop then terminates and the print statement is finally executed.
Listing 6.4 (addnonnegatives.cpp) shows that a while loop can be used for more than simple counting.
The program does not keep track of the number of values entered. The program simply accumulates the
entered values in the variable named sum.
It is a little awkward in Listing 6.4 (addnonnegatives.cpp) that the same condition appears twice, once
in the while and again in the if. Furthermore, what if the user wishes to enter negative values along
with non-negative values? The code can be simplified with a common C++ idiom that uses cin and the
extraction operator as a condition within an while statement.
If x is an integer, the expression
cin >> x

evaluates to false if the user does not enter a valid integer literal. Armed with this knowledge we can
simplify and enhance Listing 6.4 (addnonnegatives.cpp) as shown in Listing 6.5 (addnumbers.cpp).

Listing 6.5: addnumbers.cpp

1 #include <iostream>
2
3 using namespace std;
4
5 int main()
6 {
7 int input, sum = 0;
8 cout << "Enter numbers to sum, type 'q' to end the list:";
9 while ( cin >> input )

©2013 Richard L. Halterman Draft date: January 27, 2013

 6.1. THE WHILE STATEMENT 113

10 sum += input;
11 cout << "Sum = " << sum << endl;
12 }

The condition reads a value from the input stream and, if it is successful, it is interpreted as true.
When the user enters 'q', the loop is terminated. If the user types 'q' at the beginning, the loop is not
entered. The if statement is no longer necessary, since the statement
sum += input;

can be executed only if input has been legitimately assigned. Also, the variable input no longer needs
to initialized with a value simply so the loop is entered the first time; now it is assigned and then checked
within the condition of the while.
In Listing 6.5 (addnumbers.cpp), the program’s execution will terminate with any letter the user types;
an entry of 'x' or Ctrl-Z will terminate the sequence just as well as 'q'.

If you use the

while ( cin >> x )
{
// Do something . . .
}
idiom, be aware that when the loop is exited due to the input stream being unable
to read a value of the type of variable x, the cin input stream object is left in an
error state and cannot be used for the rest of the program’s execution. It you wish
to use this technique and reuse cin later, you must reset cin and extract and
discard keystrokes entered since the last valid use of the extractor operator. This
recovery process is covered in Section 13.3, but, for now, use this idiom to control
a loop only if the program does not require additional user input later during its
execution.

Listing 6.6 (powersof10.cpp) prints the powers of 10 from 1 to 1,000,000,000 (the next power of ten,
10,000,000,000, is outside the range of the int type).

Listing 6.6: powersof10.cpp

1 #include <iostream>
2
3 using namespace std;
4
5 int main()
6 {
7 int power = 1;
8 while ( power <= 1000000000 )
9 {
10 cout << power << endl;
11 power *= 10;
12 }
13 }

Listing 6.6 (powersof10.cpp) produces

©2013 Richard L. Halterman Draft date: January 27, 2013

 6.1. THE WHILE STATEMENT 114

1
10
100
1000
10000
100000
1000000
10000000
100000000
1000000000

It is customary to right justify a column of numbers, but Listing 6.6 (powersof10.cpp) prints the powers
of ten with their most-significant digit left aligned. We can right align the numbers using a stream object
called a steam manipulator. The specific stream manipulator we need is named setw. setw means “set
width.” It can be used as

cout << setw(3) << x << endl;

This statement prints the value of x right justified within a three character horizontal space on the screen.
Listing 6.7 (powersof10justified.cpp) shows the affects of setw.

Listing 6.7: powersof10justified.cpp

1 #include <iostream>
2 #include <iomanip>
3
4 using namespace std;
5
6 // Print the powers of 10 from 1 to 1,000,000,000
7 int main()
8 {
9 int power = 1;
10 while ( power <= 1000000000 )
11 {
12 // Right justify each number in a field 10 wide
13 cout << setw(10) << power << endl;
14 power *= 10;
15 }
16 }

Listing 6.7 (powersof10justified.cpp) prints

©2013 Richard L. Halterman Draft date: January 27, 2013

6.1. THE WHILE STATEMENT 115

1
10
100
1000
10000
100000
1000000
10000000
100000000
1000000000

Observe that in order to use setw the compiler needs to be made aware of it. The needed information about
setw is not found in the iomanip header file, so an additional preprocessor include directive is required:
#include <iomanip>

The setw manipulator “conditions” the output stream for the next item to be printed. The values passed
to the “conditioned” stream are all right justified within the number of spaces specified by setw.
A while statement can be used to make Listing 5.9 (troubleshoot.cpp) more convenient for the user.
Recall that the computer troubleshooting program forces the user to rerun the program once a potential
program has been detected (for example, turn on the power switch, then run the program again to see what
else might be wrong). A more desirable decision logic is shown in Figure 6.2.

Figure 6.2: Decision tree for troubleshooting a computer system Listing 6.8 (troubleshootloop.cpp)

Listing 6.8 (troubleshootloop.cpp) incorporates a while statement so that the program’s execution
continues until the problem is resolved or its resolution is beyond the capabilities of the program.

©2013 Richard L. Halterman Draft date: January 27, 2013

 6.1. THE WHILE STATEMENT 116

Listing 6.8: troubleshootloop.cpp

1 #include <iostream>
2
3 using namespace std;
4
5 int main()
6 {
7 cout << "Help! My computer doesn't work!" << endl;
8 char choice;
9 bool done = false; // Initially, we are not done
10
11 while ( !done ) // Continue until we are done
12 {
13 cout << "Does the computer make any sounds "
14 << "(fans, etc.) or show any lights? (y/n):";
15 cin >> choice;
16 // The troubleshooting control logic
17 if ( choice == 'n' ) // The computer does not have power
18 {
19 cout << "Is it plugged in? (y/n):";
20 cin >> choice;
21 if ( choice == 'n' ) // It is not plugged in, plug it in
22 {
23 cout << "Plug it in." << endl;
24 }
25 else // It is plugged in
26 {
27 cout << "Is the switch in the \"on\" position? (y/n):";
28 cin >> choice;
29 if ( choice == 'n' ) // The switch is off, turn it on!
30 {
31 cout << "Turn it on." << endl;
32 }
33 else // The switch is on
34 {
35 cout << "Does the computer have a fuse? (y/n):";
36 cin >> choice;
37 if ( choice == 'n' ) // No fuse
38 {
39 cout << "Is the outlet OK? (y/n):";
40 if ( choice == 'n' ) // Fix outlet
41 {
42 cout << "Check the outlet's circuit "
43 << "breaker or fuse. Move to a "
44 << "new outlet, if necessary. "
45 << endl;
46 }
47 else // Beats me!
48 {
49 cout << "Please consult a service "
50 << "technician." << endl;
51 done = true; // Exhausted simple fixes
52 }
53 }

©2013 Richard L. Halterman Draft date: January 27, 2013

 6.2. NESTED LOOPS 117

54 else // Check fuse

55 {
56 cout << "Check the fuse. Replace if "
57 << "necessary." << endl;
58 }
59 }
60 }
61 }
62 else // The computer has power
63 {
64 cout << "Please consult a service technician." << endl;
65 done = true; // Only troubleshoots power issues
66 }
67 }
68 }

The bulk of the body of the Listing 6.8 (troubleshootloop.cpp)is wrapped by a while statement. The
Boolean variable done is often called a flag. You can think of the flag being down when the value is
false and raised when it is true. In this case, when the flag is raised, it is a signal that the program should
terminate.
Notice the last 11 lines of Listing 6.8 (troubleshootloop.cpp):
}
}
}
}
else // The computer has power
{
cout << "Please consult a service technician." << endl;
done = true; // Only troubleshoots power issues
}
}
}

In the way this code is organized, the matching opening curly brace of a particular closing curly brace can be
found by scanning upward in the source code until the closest opening curly brace at the same indentation
level is found. Our programming logic is now getting complex enough that the proper placement of curly
braces is crucial for human readers to more quickly decipher how the program should work. See Section 4.5
for guidelines on indentation and curly brace placement to improve code readability.

6.2 Nested Loops

Just like in if statements, while bodies can contain arbitrary C++ statements, including other while
statements. A loop can therefore be nested within another loop. To see how nested loops work, consider a
program that prints out a multiplication table. Elementary school students use multiplication tables, or times
tables, as they learn the products of integers up to 10 or even 12. Figure 6.3 shows a 10 × 10 multiplication
table. We want our multiplication table program to be flexible and allow the use to specify the table’s size.
We will begin our development work with a simple program and add features as we go. First, we will not
worry about printing the table’s row and column titles, nor will we print the lines separating the titles from
the contents of the table. Initially we will print only the contents of the table. We will see we need a nested

©2013 Richard L. Halterman Draft date: January 27, 2013

 6.2. NESTED LOOPS 118

Figure 6.3: A 10 × 10 multiplication table

loop to print the table’s contents, but that still is too much to manage in our first attempt. In our first attempt
we will print the rows of the table in a very rudimentary manner. Once we are satisfied that our simple
program works we can add more features. Listing 6.9 (timestable-1st-try.cpp) shows our first attempt at a
muliplication table.

Listing 6.9: timestable-1st-try.cpp

1 #include <iostream>
2
3 using namespace std;
4
5 int main()
6 {
7 int size; // The number of rows and columns in the table
8 cout << "Please enter the table size: ";
9 cin >> size;
10 // Print a size x size multiplication table
11 int row = 1;
12 while ( row <= size ) // Table has 10 rows.
13 {
14 cout << "Row #" << row << endl;
15 row++; // Next row
16 }
17 }

The output of Listing 6.9 (timestable-1st-try.cpp) is somewhat underwhelming:

©2013 Richard L. Halterman Draft date: January 27, 2013

 6.2. NESTED LOOPS 119

Please enter the table size: 10

Row #1
Row #2
Row #3
Row #4
Row #5
Row #6
Row #7
Row #8
Row #9
Row #10

Listing 6.9 (timestable-1st-try.cpp) does indeed print each row in its proper place—it just does not supply
the needed detail for each row. Our next step is to refine the way the program prints each row. Each row
should contain size numbers. Each number within each row represents the product of the current row and
current column; for example, the number in row 2, column 5 should be 2 × 5 = 10. In each row, therefore,
we must vary the column number from from 1 to size. Listing 6.10 (timestable-2nd-try.cpp) contains the
needed refinement.

Listing 6.10: timestable-2nd-try.cpp

1 #include <iostream>
2
3 using namespace std;
4
5 int main()
6 {
7 int size; // The number of rows and columns in the table
8 cout << "Please enter the table size: ";
9 cin >> size;
10 // Print a size x size multiplication table
11 int row = 1;
12 while ( row <= size ) // Table has size rows.
13 {
14 int column = 1; // Reset column for each row.
15 while ( column <= size ) // Table has size columns.
16 {
17 int product = row*column; // Compute product
18 cout << product << " "; // Display product
19 column++; // Next element
20 }
21 cout << endl; // Move cursor to next row
22 row++; // Next row
23 }
24 }

We use a loop to print the contents of each row. The outer loop controls how many total rows the program
prints, and the inner loop, executed in its entirity each time the program prints a row, prints the individual
elements that make up a row.
The result of Listing 6.10 (timestable-2nd-try.cpp) is

©2013 Richard L. Halterman Draft date: January 27, 2013

 6.2. NESTED LOOPS 120

Please enter the table size: 10

1 2 3 4 5 6 7 8 9 10
2 4 6 8 10 12 14 16 18 20
3 6 9 12 15 18 21 24 27 30
4 8 12 16 20 24 28 32 36 40
5 10 15 20 25 30 35 40 45 50
6 12 18 24 30 36 42 48 54 60
7 14 21 28 35 42 49 56 63 70
8 16 24 32 40 48 56 64 72 80
9 18 27 36 45 54 63 72 81 90
10 20 30 40 50 60 70 80 90 100

The numbers within each column are not lined up nicely, but the numbers are in their correct positions rela-
tive to each other. We can use the setw stream manipulator introduced in Listing 6.7 (powersof10justified.cpp)
to right justify the numbers within a four-digit area. Listing 6.11 (timestable-3rd-try.cpp) contains this align-
ment adjustment.

Listing 6.11: timestable-3rd-try.cpp

1 #include <iostream>
2 #include <iomanip>
3
4 using namespace std;
5
6 int main()
7 {
8 int size; // The number of rows and columns in the table
9 cout << "Please enter the table size: ";
10 cin >> size;
11 // Print a size x size multiplication table
12 int row = 1;
13 while ( row <= size ) // Table has size rows.
14 {
15 int column = 1; // Reset column for each row.
16 while ( column <= size ) // Table has size columns.
17 {
18 int product = row*column; // Compute product
19 cout << setw(4) << product; // Display product
20 column++; // Next element
21 }
22 cout << endl; // Move cursor to next row
23 row++; // Next row
24 }
25 }

Listing 6.11 (timestable-3rd-try.cpp) produces the table’s contents in an attractive form:

©2013 Richard L. Halterman Draft date: January 27, 2013

6.2. NESTED LOOPS 121

Please enter the table size: 10

1 2 3 4 5 6 7 8 9 10
2 4 6 8 10 12 14 16 18 20
3 6 9 12 15 18 21 24 27 30
4 8 12 16 20 24 28 32 36 40
5 10 15 20 25 30 35 40 45 50
6 12 18 24 30 36 42 48 54 60
7 14 21 28 35 42 49 56 63 70
8 16 24 32 40 48 56 64 72 80
9 18 27 36 45 54 63 72 81 90
10 20 30 40 50 60 70 80 90 100

Input values of 5:

Please enter the table size: 5

1 2 3 4 5
2 4 6 8 10
3 6 9 12 15
4 8 12 16 20
5 10 15 20 25

and 15:

Please enter the table size: 15

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
2 4 6 8 10 12 14 16 18 20 22 24 26 28 30
3 6 9 12 15 18 21 24 27 30 33 36 39 42 45
4 8 12 16 20 24 28 32 36 40 44 48 52 56 60
5 10 15 20 25 30 35 40 45 50 55 60 65 70 75
6 12 18 24 30 36 42 48 54 60 66 72 78 84 90
7 14 21 28 35 42 49 56 63 70 77 84 91 98 105
8 16 24 32 40 48 56 64 72 80 88 96 104 112 120
9 18 27 36 45 54 63 72 81 90 99 108 117 126 135
10 20 30 40 50 60 70 80 90 100 110 120 130 140 150
11 22 33 44 55 66 77 88 99 110 121 132 143 154 165
12 24 36 48 60 72 84 96 108 120 132 144 156 168 180
13 26 39 52 65 78 91 104 117 130 143 156 169 182 195
14 28 42 56 70 84 98 112 126 140 154 168 182 196 210
15 30 45 60 75 90 105 120 135 150 165 180 195 210 225

also give good results.

All that is left is to add the row and column titles and the lines that bound the edges of the table.
Listing 6.12 (timestable.cpp) adds the necessary code.

©2013 Richard L. Halterman Draft date: January 27, 2013

 6.2. NESTED LOOPS 122

Listing 6.12: timestable.cpp

1 #include <iostream>
2 #include <iomanip>
3
4 using namespace std;
5
6 int main()
7 {
8 int size; // The number of rows and columns in the table
9 cout << "Please enter the table size: ";
10 cin >> size;
11 // Print a size x size multiplication table
12
13 // First, print heading: 1 2 3 4 5 etc.
14 cout << " ";
15 // Print column heading
16 int column = 1;
17 while ( column <= size )
18 {
19 cout << setw(4) << column; // Print heading for this column.
20 column++;
21 }
22 cout << endl;
23
24 // Print line separator: +------------------
25 cout << " +";
26 column = 1;
27 while ( column <= size )
28 {
29 cout << "----"; // Print line for this column.
30 column++;
31 }
32 cout << endl;
33
34 // Print table contents
35 int row = 1;
36 while ( row <= size ) // Table has size rows.
37 {
38 cout << setw(2) << row << " |"; // Print heading for this row.
39 int column = 1; // Reset column for each row.
40 while ( column <= size ) // Table has size columns.
41 {
42 int product = row*column; // Compute product
43 cout << setw(4) << product; // Display product
44 column++; // Next element
45 }
46 row++; // Next row
47 cout << endl; // Move cursor to next row
48 }
49 }

When the user supplies the value 10, Listing 6.12 (timestable.cpp) produces

©2013 Richard L. Halterman Draft date: January 27, 2013

6.2. NESTED LOOPS 123

Please enter the table size: 10

1 2 3 4 5 6 7 8 9 10
+----------------------------------------
1 | 1 2 3 4 5 6 7 8 9 10
2 | 2 4 6 8 10 12 14 16 18 20
3 | 3 6 9 12 15 18 21 24 27 30
4 | 4 8 12 16 20 24 28 32 36 40
5 | 5 10 15 20 25 30 35 40 45 50
6 | 6 12 18 24 30 36 42 48 54 60
7 | 7 14 21 28 35 42 49 56 63 70
8 | 8 16 24 32 40 48 56 64 72 80
9 | 9 18 27 36 45 54 63 72 81 90
10 | 10 20 30 40 50 60 70 80 90 100

An input of 15 yields

Please enter the table size: 15

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
+------------------------------------------------------------
1 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
2 | 2 4 6 8 10 12 14 16 18 20 22 24 26 28 30
3 | 3 6 9 12 15 18 21 24 27 30 33 36 39 42 45
4 | 4 8 12 16 20 24 28 32 36 40 44 48 52 56 60
5 | 5 10 15 20 25 30 35 40 45 50 55 60 65 70 75
6 | 6 12 18 24 30 36 42 48 54 60 66 72 78 84 90
7 | 7 14 21 28 35 42 49 56 63 70 77 84 91 98 105
8 | 8 16 24 32 40 48 56 64 72 80 88 96 104 112 120
9 | 9 18 27 36 45 54 63 72 81 90 99 108 117 126 135
10 | 10 20 30 40 50 60 70 80 90 100 110 120 130 140 150
11 | 11 22 33 44 55 66 77 88 99 110 121 132 143 154 165
12 | 12 24 36 48 60 72 84 96 108 120 132 144 156 168 180
13 | 13 26 39 52 65 78 91 104 117 130 143 156 169 182 195
14 | 14 28 42 56 70 84 98 112 126 140 154 168 182 196 210
15 | 15 30 45 60 75 90 105 120 135 150 165 180 195 210 225

If the user enters 7, the program prints

©2013 Richard L. Halterman Draft date: January 27, 2013

6.2. NESTED LOOPS 124

Please enter the table size: 7

1 2 3 4 5 6 7
+----------------------------
1 | 1 2 3 4 5 6 7
2 | 2 4 6 8 10 12 14
3 | 3 6 9 12 15 18 21
4 | 4 8 12 16 20 24 28
5 | 5 10 15 20 25 30 35
6 | 6 12 18 24 30 36 42
7 | 7 14 21 28 35 42 49

The user even can enter a 1:

Please enter the table size: 1

1
+----
1 | 1

As we can see, the table automatically adjusts to the size and spacing required by the user’s input.
This is how Listing 6.12 (timestable.cpp) works:

• It is important to distinguish what is done only once (outside all loops) from that which is done
repeatedly. The column heading across the top of the table is outside of all the loops; therefore, it is
printed all at once.

• The work to print the heading for the rows is distributed throughout the execution of the outer loop.
This is because the heading for a given row cannot be printed until all the results for the previous row
have been printed.

• A code fragment like

if ( x < 10 )
cout << " ";
cout << x;

prints x in one of two ways: if x is a one-digit number, it prints a space before it; otherwise, it does
not print the extra space. The net effect is to right justify one and two digit numbers within a two
character space printing area. This technique allows the columns within the times table to be properly
right aligned.

• In the nested loop, row is the control variable for the outer loop; column controls the inner loop.

• The inner loop executes size times on every single iteration of the outer loop. How many times is
the statement
cout << product << " "; // Display product

executed? size × size times, one time for every product in the table.

©2013 Richard L. Halterman Draft date: January 27, 2013

 6.2. NESTED LOOPS 125

• A newline is printed after the contents of each row is displayed; thus, all the values printed in the
inner (column) loop appear on the same line.

Nested loops are used when an iterative process itself must be repeated. In our times table example, a
while loop is used to print the contents of each row, but multiple rows must be printed. The inner loop
prints the contents of each row, while the outer is responsible for printing all the rows.
Listing 6.13 (permuteabc.cpp) uses a triply-nested loop to print all the different arrangements of the
letters A, B, and C. Each string printed is a permutation of ABC.

Listing 6.13: permuteabc.cpp

1 // File permuteabc.cpp
2
3 #include <iostream>
4
5 using namespace std;
6
7 int main()
8 {
9 char first = 'A'; // The first letter varies from A to C
10 while ( first <= 'C' )
11 {
12 char second = 'A';
13 while ( second <= 'C' ) // The second varies from A to C
14 {
15 if ( second != first ) // No duplicate letters
16 {
17 char third = 'A';
18 while ( third <= 'C' ) // The third varies from A to C
19 { // Don't duplicate first or second letter
20 if ( third != first && third != second )
21 cout << first << second << third << endl;
22 third++;
23 }
24 }
25 second++;
26 }
27 first++;
28 }
29 }

Notice how the if statements are used to prevent duplicate letters within a given string. The output of
Listing 6.13 (permuteabc.cpp) is all six permutations of ABC:

ABC
ACB
BAC
BCA
CAB
CBA

©2013 Richard L. Halterman Draft date: January 27, 2013

 6.3. ABNORMAL LOOP TERMINATION 126

6.3 Abnormal Loop Termination

Normally, a while statement executes until its condition becomes false. This condition is checked only at
the “top” of the loop, so the loop is not immediately exited if the condition becomes false due to activity
in the middle of the body. Ordinarily this behavior is not a problem because the intention is to execute all
the statements within the body as an indivisible unit. Sometimes, however, it is desirable to immediately
exit the body or recheck the condition from the middle of the loop instead. C++ provides the break and
continue statements to to give programmers more flexibility designing the control logic of loops.

6.3.1 The break statement

C++ provides the break statement to implement middle-exiting control logic. The break statement causes
the immediate exit from the body of the loop. Listing 6.14 (addmiddleexit.cpp) is a variation of Listing 6.4
(addnonnegatives.cpp) that illustrates the use of break.

Listing 6.14: addmiddleexit.cpp

1 #include <iostream>
2
3 using namespace std;
4
5 int main()
6 {
7 int input, sum = 0;
8 cout << "Enter numbers to sum, negative number ends list:";
9 while ( true )
10 {
11 cin >> input;
12 if ( input < 0 )
13 break; // Exit loop immediately
14 sum += input;
15 }
16 cout << "Sum = " << sum << endl;
17 }

The condition of the while is a tautology, so the body of the loop will be entered. (At warning Level 4,
Visual C++ will issue a warning about the tautology.) Since the condition of the while can never be false,
the break statement is the only way to get out of the loop. The break statement is executed only when the
user enters a negative number. When the break statement is encountered during the program’s execution,
the loop is immediately exited. Any statements following the break within the body are skipped. It is not
possible, therefore, to add a negative number to the sum variable.
The break statement should be used sparingly because it introduces an exception into the normal
control logic of the loop. Ideally, every loop should have a single entry point and single exit point. While
Listing 6.14 (addmiddleexit.cpp) has a single exit point (the break statement), programmers commonly
use break statements within while statements with conditions that are not always true. In such a
while loop, adding a break statement adds an extra exit point (the top of the loop where the condition
is checked is one point, and the break statement is another). Using multiple break statements within a
single loop is particularly dubious and should be avoided. Why have the break statement at all if its use is
questionable and it is dispensable? The logic in Listing 6.4 (addnonnegatives.cpp) is fairly simple, so the
restructuring of Listing 6.14 (addmiddleexit.cpp) is straightforward; in general, the effort may complicate
the logic a bit and require the introduction of an additional Boolean variable. As shown in Figure 6.4, any

©2013 Richard L. Halterman Draft date: January 27, 2013

 6.3. ABNORMAL LOOP TERMINATION 127

program that uses a break statement can be rewritten so that the break statement is not used. The no-

bool looping = true;

while ( Condition 1 ) while ( looping && Condition 1 )
{ {
Part A Part A

if ( Condition 2 ) if ( Condition 2 )
{ {
Part B Eliminate Part B
the
break; break looping = false;
} }
statement
else
Part C
{
}
Part C
}
}

Figure 6.4: The code on the left generically represents any loop that uses a break statement. It is possible
to transform the code on the left to eliminate the break statement, as the code on the right shows.

break version introduces a Boolean variable, and the loop control logic is a little more complicated. The
no-break version uses more space (an extra variable) and more time (requires an extra check in the loop
condition on every iteration), and its logic is more complex. The more complicated the control logic for a
given section of code, the more difficult the code is to write correctly. In some situations, even though it
violates the “single entry point, single exit point” principle, a simple break statement can be an acceptable
loop control option.

6.3.2 The goto Statement

The break statement exits the single loop in which it is located. A break statement is insufficient to
jump completely out of the middle of a nested loop. The goto statement allows the program’s execution
flow to jump to a specified location within the function. Listing 6.15 (exitnested.cpp) shows how a goto
statement can be used to jump out from the middle of a nested loop.

Listing 6.15: exitnested.cpp

1 #include <iostream>
2
3 using namespace std;
4
5 int main()
6 {
7 // Compute some products

©2013 Richard L. Halterman Draft date: January 27, 2013

 6.3. ABNORMAL LOOP TERMINATION 128

8 int op1 = 2;
9 while ( op1 < 100 )
10 {
11 int op2 = 2;
12 while ( op2 < 100 )
13 {
14 if ( op1 * op2 == 3731 )
15 goto end;
16 cout << "Product is " << (op1 * op2) << endl;
17 op2++;
18 }
19 op1++;
20 }
21 end:
22 cout << "The end" << endl;
23 }

When op1 * op2 is 3731, program flow will jump to the specified label within the program. In this
example, the label is named end, but this name is arbitrary. Like variable names, label names should be
chosen to indicate their intended purpose. The label here named end comes after and outside the nested
while loops.
A label’s name is an identifier (see Section 3.3), and a label is distinguished by the colon that immedi-
ately follows its name. A label represents a target to which a goto can jump. A goto label must appear
before a statement within a function.
With the goto statement, the while is superfluous; for example, Listing 6.2 (iterativecounttofive.cpp)
could be rewritten without the while statement as shown in Listing 6.16 (gotoloop.cpp).

Listing 6.16: gotoloop.cpp

1 #include <iostream>
2
3 using namespace std;
4
5 int main()
6 {
7 int count = 1; // Initialize counter
8 top:
9 if ( count > 5 )
10 goto end;
11 cout << count << endl; // Display counter, then
12 count++; // Increment counter
13 goto top;
14 end:
15 ; // Target is an empty statement
16 }

Early programming languages like FORTRAN and early versions of BASIC did not have structured
statements like while, so programmers were forced to use goto statements to write loops. The problem
with using goto statements is that it is easy to develop program logic that is very difficult to under-
stand, even for the original author of the code. See the Wikipedia article about spaghetti code (http:
//en.wikipedia.org/wiki/Spaghetti_code). The structured programming revolution of the 1960s in-
troduced constructs such as the while statement and resulted in the disappearance of the use of goto in
most situations. All modern programming languages have a form of the while statement, so the goto

©2013 Richard L. Halterman Draft date: January 27, 2013

 6.3. ABNORMAL LOOP TERMINATION 129

statement in C++ is largely ignored except for the case of breaking out of a nested loop. You similarly
should restrict your use of the goto statement to the abnormal exit of nested loops.

6.3.3 The continue Statement

The continue statement is similar to the break statement. During a program’s execution, when the
break statement is encountered within the body of a loop, the remaining statements within the body of the
loop are skipped, and the loop is exited. When a continue statement is encountered within a loop, the
remaining statements within the body are skipped, but the loop condition is checked to see if the loop should
continue or be exited. If the loop’s condition is still true, the loop is not exited, but the loop’s execution
continues at the top of the loop. Listing 6.17 (continueexample.cpp) shows how the continue statement
can be used.

Listing 6.17: continueexample.cpp

1 #include <iostream>
2
3 using namespace std;
4
5 int main()
6 {
7 int input, sum = 0;
8 bool done = false;
9 while ( !done )
10 {
11 cout << "Enter positive integer (999 quits): ";
12 cin >> input;
13 if ( input < 0 )
14 {
15 cout << "Negative value " << input << " ignored"
16 << endl;
17 continue; // Skip rest of body for this iteration
18 }
19 if ( input != 999 )
20 {
21 cout << "Tallying " << input << endl;
22 sum += input;
23 }
24 else
25 done = (input == 999); // 999 entry exits loop
26 }
27 cout << "sum = " << sum << endl;
28 }

The continue statement is not used as frequently as the break statement since it is often easy to
transform the code into an equivalent form that does not use continue. Listing 6.18 (nocontinueexample.cpp)
works exactly like Listing 6.17 (continueexample.cpp), but the continue has been eliminated.

Listing 6.18: nocontinueexample.cpp

1 #include <iostream>
2
3 using namespace std;
4

©2013 Richard L. Halterman Draft date: January 27, 2013

 6.4. INFINITE LOOPS 130

5 int main()
6 {
7 int input, sum = 0;
8 bool done = false;
9 while ( !done )
10 {
11 cout << "Enter positive integer (999 quits): ";
12 cin >> input;
13 if ( input < 0 )
14 {
15 cout << "Negative value " << input << " ignored"
16 << endl;
17 }
18 else
19 if ( input != 999 )
20 {
21 cout << "Tallying " << input << endl;
22 sum += input;
23 }
24 else
25 done = (input == 999); // 999 entry exits loop
26 }
27 cout << "sum = " << sum << endl;
28 }

As shown in Figure 6.5, any program that uses a continue statement can be rewritten so that the
continue statement is not used. The transformation is simpler than for break elimination (see Fig-
ure 6.4), since the loop’s condition remains the same, and no additional variable is needed. The version that
uses continue is no more efficient than version that uses else; in fact, the Visual C++ and GNU C++ com-
pilers generate the same machine language code for Listing 6.17 (continueexample.cpp) and Listing 6.18
(nocontinueexample.cpp). Also, the logic of the else version is no more complex than the continue
version. Therefore, unlike the break statement above, there is no compelling reason to use the continue
statement. Sometimes a continue statement is added at the last minute to an existing loop body to handle
an exceptional condition (like ignoring negative numbers in the example above) that initially went unno-
ticed. If the body of the loop is lengthy, a conditional statement with a continue can be added easily
near the top of the loop body without touching the logic of the rest of the loop. Therefore, the continue
statement merely provides a convenient alternative to the programmer. The else version is preferred.

6.4 Infinite Loops

An infinite loop is a loop that is never exited. Once the program flow enters the loop’s body it cannot
escape. Infinite loops are sometimes designed. For example, a long-running server application, like a Web
server, may need to continuously check for incoming, connections. This checking can be performed within
a loop that runs indefinitely. All too often for beginning programmers, however, infinite loops are created
by accident and represent logical errors in their programs.
Intentional infinite loops should be made obvious. For example,
while ( true )
{
/* Do something forever . . . */
}

©2013 Richard L. Halterman Draft date: January 27, 2013

 6.4. INFINITE LOOPS 131

while ( Condition 1 ) while ( Condition 1 )

{ {
Part A Part A

if ( Condition 2 ) if ( Condition 2 )
{ {
Part B Eliminate Part B
the
continue; }
continue
} else
statement
{
Part C
Part C
}
}
}

Figure 6.5: The code on the left generically represents any loop that uses a continue statement. It is
possible to transform the code on the left to eliminate the continue statement, as the code on the right
shows.

The Boolean literal true is always true, so it is impossible for the loop’s condition to be false. The only
ways to exit the loop is via a break statement, return statement (see Chapter 9), or an exit call (see
Section 8.1) embedded somewhere within its body.
Intentional infinite loops are easy to write correctly. Accidental infinite loops are quite common, but
can be puzzling for beginning programmers to diagnose and repair. Consider Listing 6.19 (findfactors.cpp)
that attempts to print all the integers with their associated factors from 1 to 20.

Listing 6.19: findfactors.cpp

1 #include <iostream>
2
3 using namespace std;
4
5 int main()
6 {
7 // List of the factors of the numbers up to 20
8 int n = 1;
9 const int MAX = 20;
10 while ( n <= MAX )
11 {
12 int factor = 1;
13 cout << n << ": ";
14 while ( factor <= n )
15 if ( n % factor == 0 )
16 {

©2013 Richard L. Halterman Draft date: January 27, 2013

 6.4. INFINITE LOOPS 132

17 cout << factor << " ";

18 factor++;
19 }
20 cout << endl; // Go to next line for next n
21 n++;
22 }
23 }

It displays

1: 1
2: 1 2
3: 1

and then “freezes up” or “hangs,” ignoring any user input (except the key sequence Ctrl C on most
systems which interrupts and terminates the running program). This type of behavior is a frequent symptom
of an unintentional infinite loop. The factors of 1 display properly, as do the factors of 2. The first factor
of 3 is properly displayed and then the program hangs. Since the program is short, the problem may be
easy to locate. In some programs, though, the error may be challenging to find. Even in Listing 6.19
(findfactors.cpp) the debugging task is nontrivial since nested loops are involved. (Can you find and fix the
problem in Listing 6.19 (findfactors.cpp) before reading further?)
In order to avoid infinite loops, we must ensure that the loop exhibits certain properties:

• The loop’s condition must not be a tautology (a Boolean expression that can never be false). For
example,
while ( i >= 1 || i <= 10 )
{
/* Body omitted */
}

is an infinite loop since any value chosen for i will satisfy one or both of the two subconditions.
Perhaps the programmer intended to use a && instead of || to stay in the loop as long as i remains
in the range 1...10.
In Listing 6.19 (findfactors.cpp) the outer loop condition is
n <= MAX

If n is 21 and MAX is 20, then the condition is false, so this is not a tautology. Checking the inner
loop condition:
factor <= n

we see that if factor is 3 and n is 2, then the expression is false; therefore, it also is not a tautology.
• The condition of a while must be true initially to gain access to its body. The code within the body
must modify the state of the program in some way so as to influence the outcome of the condition
that is checked at each iteration. This usually means one of the variables used in the condition is
modified in the body. Eventually the variable assumes a value that makes the condition false, and the
loop terminates.

©2013 Richard L. Halterman Draft date: January 27, 2013

6.4. INFINITE LOOPS 133

In Listing 6.19 (findfactors.cpp) the outer loop’s condition involves the variable n and constant MAX.
MAX cannot change, so to avoid an infinite loop it is essential that n be modified within the loop.
Fortunately, the last statement in the body of the outer loop increments n. n is initially 1 and MAX is
20, so unless the circumstances arise to make the inner loop infinite, the outer loop should eventually
terminate.

The inner loop’s condition involves the variables n and factor. No statement in the inner loop
modifies n, so it is imperative that factor be modified in the loop. The good news is factor is
incremented in the body of the inner loop, but the bad news is the increment operation is protected
within the body of the if statement. The inner loop contains one statement, the if statement. That
if statement in turn has two statements in its body:

while ( factor <= n )

if ( n % factor == 0 )
{
cout << factor << " ";
factor++;
}

If the condition of the if is ever false, the state of the program will not change when the body of the
inner loop is executed. This effectively creates an infinite loop. The statement that modifies factor
must be moved outside of the if statement’s body:

while ( factor <= n )

if ( n % factor == 0 )
cout << factor << " ";
factor++;

This new version runs correctly.

A debugger can be used to step through a program to see where and why an infinite loop arises. Another
common technique is to put print statements in strategic places to examine the values of the variables
involved in the loop’s control. The original inner loop can be so augmented:

while ( factor <= n )

{
cout << "factor = " << factor
<< " n = " << n << endl;
if ( n % factor == 0 )
{
cout << factor << " ";
factor++;
}
}

It produces the following output:

©2013 Richard L. Halterman Draft date: January 27, 2013

6.5. ITERATION EXAMPLES 134

1: factor = 1 n = 1
1
2: factor = 1 n = 2
1 factor = 2 n = 2
2
3: factor = 1 n = 3
1 factor = 2 n = 3
factor = 2 n = 3
factor = 2 n = 3
factor = 2 n = 3
factor = 2 n = 3
factor = 2 n = 3
.
.
.

The program continues to print the same line until the user interrupts its execution. The output demonstrates
that once factor becomes equal to 2 and n becomes equal to 3 the program’s execution becomes trapped
in the inner loop. Under these conditions:

1. 2 < 3 is true, so the loop continues and

2. 3 % 2 is equal to 1, so the if statement will not increment factor.

It is imperative that factor be incremented each time through the inner loop; therefore, the statement
incrementing factor must be moved outside of the if’s guarded body.

6.5 Iteration Examples

We can implement some sophisticated algorithms in C++ now that we are armed with if and while
statements. This section provides several examples that show off the power of conditional execution and
iteration.

6.5.1 Drawing a Tree

Suppose a triangular tree must be drawn, but its height is provided by the user. A tree that is five levels tall
would look like

*
***
*****
*******
*********

©2013 Richard L. Halterman Draft date: January 27, 2013

 6.5. ITERATION EXAMPLES 135

whereas a three-level tree would look like

*
***
*****

If the height of the tree is fixed, we can write the program as a simple variation of Listing 2.4 (arrow.cpp)
which uses only printing statements and no loops. Our program, however, must vary its height and width
based on input from the user.
Listing 6.20 (startree.cpp) provides the necessary functionality.

Listing 6.20: startree.cpp

1 #include <iostream>
2
3 using namespace std;
4
5 int main()
6 {
7 int height; // Height of tree
8 cout << "Enter height of tree: ";
9 cin >> height; // Get height from user
10 int row = 0; // First row, from the top, to draw
11 while ( row < height ) // Draw one row for every unit of height
12 {
13 // Print leading spaces
14 int count = 0;
15 while ( count < height - row )
16 {
17 cout << " ";
18 count++;
19 }
20 // Print out stars, twice the current row plus one:
21 // 1. number of stars on left side of tree
22 // = current row value
23 // 2. exactly one star in the center of tree
24 // 3. number of stars on right side of tree
25 // = current row value
26 count = 0;
27 while ( count < 2*row + 1 )
28 {
29 cout << "*";
30 count++;
31 }
32 // Move cursor down to next line
33 cout << endl;
34 // Change to the next row
35 row++;
36 }
37 }

When Listing 6.20 (startree.cpp) is run and the user enters, for example, 7, the output is:

©2013 Richard L. Halterman Draft date: January 27, 2013

6.5. ITERATION EXAMPLES 136

Enter height of tree: 7

*
***
*****
*******
*********
***********
*************

Listing 6.20 (startree.cpp) uses two sequential while loops nested within a while loop. The outer
while loop is responsible for drawing one row of the tree each time its body is executed:

• As long as the user enters a value greater than zero, the body of the outer while loop will be
executed; if the user enters zero or less, the program terminates and does nothing. This is the expected
behavior.

• The last statement in the body of the outer while:

row++;

ensures that the variable row increases by one each time through the loop; therefore, it eventually
will equal height (since it initially had to be less than height to enter the loop), and the loop will
terminate. There is no possibility of an infinite loop here.

• The body of the outer loop consists of more than one statement; therefore, the body must be enclosed
within curly braces. Whenever a group of statements is enclosed within curly braces a block is
formed. Any variable declared within a block is local to that block. A variable’s scope (the section
of the source code in which the variable exists and can be used) is from its point of declaration to
the end of the block. For example, the variables height and row are declared in the block that is
main’s body; thus, they are local to main. The variable count is declared within the block that is
the body of the outer while statement; therefore, count is local to the outer while statement. An
attempt to use count outside the body of the outer while statement would be an error.

What does it mean for a variable x to be local to a particular section of code? It means x does not exist
outside its scope. There may be other variables in the program named x, but they are different variables. If
it seems odd that you can have two different variables in the same program with the same name, consider
the fact that there can be two people in the same room with the same name. They are different people, but
they have the same name. Similarly, the meaning of a variable depends on its context, and its name is not
necessarily unique.
The two inner loops play distinct roles:

• The first inner loop prints spaces. The number of spaces printed is equal to the height of the tree the
first time through the outer loop and decreases each iteration. This is the correct behavior since each
succeeding row moving down contains fewer leading spaces but more asterisks.

• The second inner loop prints the row of asterisks that make up the tree. The first time through the
outer loop, row is zero, so no left side asterisks are printed, one central asterisk is printed (the top of
the tree), and no right side asterisks are printed. Each time through the loop the number of left-hand
and right-hand stars to print both increase by one and the same central asterisk is printed; therefore,

©2013 Richard L. Halterman Draft date: January 27, 2013

 6.5. ITERATION EXAMPLES 137

the tree grows one wider on each side each line moving down. Observe how the 2*row + 1 value
expresses the needed number of asterisks perfectly.

• While it seems asymmetrical, note that no third inner loop is required to print trailing spaces on the
line after the asterisks are printed. The spaces would be invisible, so there is no reason to print them!

6.5.2 Printing Prime Numbers

A prime number is an integer greater than one whose only factors (also called divisors) are one and itself.
For example, 29 is a prime number (only 1 and 29 divide into it with no remainder), but 28 is not (2, 4, 7,
and 14 are factors of 28). Prime numbers were once merely an intellectual curiosity of mathematicians, but
now they play an important role in cryptography and computer security.
The task is to write a program that displays all the prime numbers up to a value entered by the user.
Listing 6.21 (printprimes.cpp) provides one solution.

Listing 6.21: printprimes.cpp

1 #include <iostream>
2
3 using namespace std;
4
5 int main()
6 {
7
8 int max_value;
9 cout << "Display primes up to what value? ";
10 cin >> max_value;
11 int value = 2; // Smallest prime number
12 while ( value <= max_value )
13 {
14 // See if value is prime
15 bool is_prime = true; // Provisionally, value is prime
16 // Try all possible factors from 2 to value - 1
17 int trial_factor = 2;
18 while ( trial_factor < value )
19 {
20 if ( value % trial_factor == 0 )
21 {
22 is_prime = false; // Found a factor
23 break; // No need to continue; it is NOT prime
24 }
25 trial_factor++;
26 }
27 if ( is_prime )
28 cout << value << " "; // Display the prime number
29 value++; // Try the next potential prime number
30 }
31 cout << endl; // Move cursor down to next line
32 }

Listing 6.21 (printprimes.cpp), with an input of 90, produces:

©2013 Richard L. Halterman Draft date: January 27, 2013

6.5. ITERATION EXAMPLES 138

Display primes up to what value? 90

2 3 5 7 11 13 17 19 23 29 31 37 41 43 47 53 59 61 67 71 73 79 83 89

The logic of Listing 6.21 (printprimes.cpp) is a little more complex than that of Listing 6.20 (startree.cpp).
The user provides a value for max_value. The main loop (outer while iterates over all the values from
two to max_value:

• Two new variables, local to the body of the outer loop, are introduced: trial_factor and
is_prime. is_prime is initialized to true, meaning value is assumed to be prime unless our
tests prove otherwise. trial_factor takes on all the values from two to value - 1 in the inner
loop:
int trial_factor = 2;
while ( trial_factor < value )
{
if ( value % trial_factor == 0 )
{
is_prime = false; // Found a factor
break; // No need to continue; it is NOT prime
}
trial_factor++;
}

The expression value % trial_factor is zero when trial_factor divides into value
with no remainder—exactly when trial_factor is a factor of value. If any of the values of
trial_factor is determined to actually be a factor of value, then is_prime is set to false,
and the loop is exited via the break. If the loop continues to completion, is_prime will never be
set to false, which means no factors were found and value is indeed prime.
• The if statement after the inner loop:
if ( is_prime )
cout << value << " "; // Display the prime number

simply checks the status of is_prime. If is_prime is true, then value must be prime, so value
is printed along with an extra space to separate from other factors that may be printed during the next
iterations.

Some important questions can be asked.

1. If the user enters a 2, will it be printed?

In this case max_value = value = 2, so the condition of the outer loop
value <= max_value

is true, since 2 <= 2. is_prime is set to true, but the condition of the inner loop
trial_factor < value

is not true (2 is not less than 2). Thus, the inner loop is skipped, is_prime is not changed from
true, and 2 is printed. This behavior is correct, because 2 is the smallest prime number (and the only
even prime).

©2013 Richard L. Halterman Draft date: January 27, 2013

6.5. ITERATION EXAMPLES 139

2. if the user enters a number less than 2, is anything printed?

The while condition ensures that values less than two are not considered. The body of the while
will never be entered. Only the newline is printed, and no numbers are displayed. This behavior is
correct.
3. Is the inner loop guaranteed to always terminate?
In order to enter the body of the inner loop, trial_factor must be less than value. value
does not change anywhere in the loop. trial_factor is not modified anywhere in the if
statement within the loop, and it is incremented within the loop immediately after the if state-
ment. trial_factor is, therefore, incremented during each iteration of the loop. Eventually,
trial_factor will equal value, and the loop will terminate.
4. Is the outer loop guaranteed to always terminate?
In order to enter the body of the outer loop, value must be less than or equal to max_value.
max_value does not change anywhere in the loop. value is increased in the last statement within
the body of the outer loop, and value is not modified anywhere else. Since the inner loop is guaran-
teed to terminate as shown in the previous answer, eventually value will exceed max_value and
the loop will end.

The logic of the inner while can be rearranged slightly to avoid the break statement. The current
version is:
while ( trial_factor < value )
{
if ( value % trial_factor == 0 )
{
is_prime = false; // Found a factor
break; // No need to continue; it is NOT prime
}
trial_factor++;
}

It can be rewritten as:

while ( is_prime && trial_factor < value )
{
is_prime = (value % trial_factor != 0);
trial_factor++; // Try next factor
}

This version without the break introduces a slightly more complicated condition for the while but re-
moves the if statement within its body. is_prime is initialized to true before the loop. Each time through
the loop it is reassigned. trial_factor will become false if at any time value % trial_factor is
zero. This is exactly when trial_factor is a factor of value. If is_prime becomes false, the loop
cannot continue, and if is_prime never becomes false, the loop ends when trial_factor becomes
equal to value. Because of operator precedence, the parentheses are not necessary. The parentheses do
improve readability, since an expression including both == and != is awkward for humans to parse. When
parentheses are placed where they are not needed, as in
x = (y + 2);

the compiler simply ignores them, so there is no efficiency penalty in the compiled code.

©2013 Richard L. Halterman Draft date: January 27, 2013

6.6. SUMMARY 140

This loop can be further shortened:

while ( is_prime && trial_factor < value )
is_prime = (value % trial_factor++ != 0);

This version uses the post-increment operator within the test expression (see Section 4.8). Recall that with
the post-increment operator the value of the variable is used in the surrounding expression (if any), and then
the variable is incremented. Since the while’s body now contains only one statement, the curly braces are
not needed.

6.6 Summary

• The while statement allows the execution of code sections to be repeated multiple times.

• The condition of the while controls the execution of statements within the while’s body.

• The statements within the body of a while are executed over and over until the condition of the
while is false.

• If the while’s condition is initially false, the body is not executed at all.

• In an infinite loop, the while’s condition never becomes false.

• The statements within the while’s body must eventually lead to the condition being false; otherwise,
the loop will be infinite.

• Do not confuse while statements with if statements; their structure is very similar (while re-
served word instead of the if word), but they behave differently.

• Infinite loops are rarely intentional and are usually accidental.

• An infinite loop can be diagnosed by putting a printing statement inside its body.

• An assignment expression has a value; the expression’s value is the same as the expression on the
right side of the assignment operator. This fact can be used to streamline the the control of a loop that
repeats based on user input.

• A loop contained within another loop is called a nested loop.

• Iteration is a powerful mechanism and can be used to solve many interesting problems.

• A block is any section of source code enclosed within curly braces. A compound statement is one
example of a block.

• A variable declared within a block is local to that block.

• Complex iteration using nested loops mixed with conditional statements can be difficult to do cor-
rectly.

• Sometimes simple optimizations can speed up considerably the execution of loops.

• The break statement immediately exits a loop, skipping the rest of the loop’s body, without checking
to see if the condition is true or false. Execution continues with the statement immediately following
the body of the loop.

©2013 Richard L. Halterman Draft date: January 27, 2013

6.7. EXERCISES 141

• In a nested loop, the break statement exits only the loop in which the break is found.

• The goto statement directs the program’s execution to a labeled statement within the function. The
goto statement is legitimately used only to exit completely from the depths of a nested loop.

• The continue statement immediately checks the loop’s condition, skipping the rest of the loop’s
body. If the condition is true, the execution continues at the top of the loop as usual; otherwise, the
loop is terminated and execution continues with the statement immediately following the loop’s body.
false.

• In a nested loop, the continue statement affects only the loop in which the continue is found.

6.7 Exercises

1. In Listing 6.4 (addnonnegatives.cpp) could the condition of the if statement have used > instead of
>= and achieved the same results? Why?

2. In Listing 6.4 (addnonnegatives.cpp) could the condition of the while statement have used > instead
of >= and achieved the same results? Why?

3. Use a loop to rewrite the following code fragment so that it uses just one cout and one endl.
cout << 2 << endl;
cout << 4 << endl;
cout << 6 << endl;
cout << 8 << endl;
cout << 10 << endl;
cout << 12 << endl;
cout << 14 << endl;
cout << 16 << endl;

4. In Listing 6.4 (addnonnegatives.cpp) what would happen if the statement containing cin is moved
out of the loop? Is moving the assignment out of the loop a good or bad thing to do? Why?

5. How many asterisks does the following code fragment print?

int a = 0;
while ( a < 100 )
{
cout << "*";
a++;
}
cout << endl;

6. How many asterisks does the following code fragment print?

int a = 0;
while ( a < 100 )
cout << "*";
cout << endl;

7. How many asterisks does the following code fragment print?

©2013 Richard L. Halterman Draft date: January 27, 2013

6.7. EXERCISES 142

int a = 0;
while ( a > 100 )
{
cout << "*";
a++;
}
cout << endl;

8. How many asterisks does the following code fragment print?

int a = 0;
while ( a < 100 )
{
int b = 0;
while ( b < 55 )
{
cout << "*";
b++;
}
cout << endl;
}

9. How many asterisks does the following code fragment print?

int a = 0;
while ( a < 100 )
{
if ( a % 5 == 0 )
cout << "*";
a++;
}
cout << endl;

10. How many asterisks does the following code fragment print?
int a = 0;
while ( a < 100 )
{
int b = 0;
while ( b < 40 )
{
if ( (a + b) % 2 == 0 )
cout << "*";
b++;
}
cout << endl;
a++;
}

11. How many asterisks does the following code fragment print?

©2013 Richard L. Halterman Draft date: January 27, 2013

6.7. EXERCISES 143

int a = 0;
while ( a < 100 )
{
int b = 0;
while ( b < 100 )
{
int c = 0;
while ( c < 100 )
{
cout << "*";
c++;
}
b++;
}
a++;
}
cout << endl;

12. What is printed by the following code fragment?

int a = 0;
while ( a < 100 )
cout << a++;
cout << endl;

13. What is printed by the following code fragment?

int a = 0;
while ( a > 100 )
cout << a++;
cout << endl;

14. Rewrite the following code fragment using a break statement and eliminating the done variable.
Your code should behave identically to this code fragment.
bool done = false;
int n = 0, m = 100;
while ( !done && n != m )
{
cin >> n;
if ( n < 0 )
done = true;
cout << "n = " << endl;
}

15. Rewrite the following code fragment so it does not use a break statement. Your code should behave
identically to this code fragment.
// Code with break ...

16. Rewrite the following code fragment so it eliminates the continue statement. Your new code’s
logic should be simpler than the logic of this fragment.

©2013 Richard L. Halterman Draft date: January 27, 2013

6.7. EXERCISES 144

int x = 100, y;
while ( x > 0 )
{
cin >> y;
if ( y == 25 )
{
x--;
continue;
}
cin >> x;
cout << "x = " << x << endl;
}

17. Suppose you were given some code from the 1960s in a language that did not support structured
statements like while. Your task is to modernize it and adapt it to C++. The following code fragment
has been adapted to C++ already, but you must now structure it with a while statement to replace
the gotos. Your code should be goto free and still behave identically to this code fragment.
int i = 0;
top: if ( i >= 10 )
goto end;
cout << i << endl;
i++;
goto top;
end:

18. Simplify with goto...

19. What is printed by the following code fragment?

int a = 0;
while ( a < 100 );
cout << a++;
cout << endl;

20. Write a C++ program that accepts a single integer value entered by the user. If the value entered is
less than one, the program prints nothing. If the user enters a positive integer, n, the program prints
an n × n box drawn with * characters. If the users enters 1, for example, the program prints

If the user enters a 2, it prints

**
**

An entry of three yields

©2013 Richard L. Halterman Draft date: January 27, 2013

6.7. EXERCISES 145

***
***
***

and so forth. If the user enters 7, it prints

*******
*******
*******
*******
*******
*******
*******

that is, a 7 × 7 box of * symbols.

21. Write a C++ program that allows the user to enter exactly twenty double-precision floating-point
values. The program then prints the sum, average (arithmetic mean), maximum, and minimum of the
values entered.

22. Write a C++ program that allows the user to enter any number of non-negative double-precision
floating-point values. The user terminates the input list with any negative value. The program then
prints the sum, average (arithmetic mean), maximum, and minimum of the values entered. The
terminating negative value is not used in the computations.

23. Redesign Listing 6.20 (startree.cpp) so that it draws a sideways tree pointing right; for example, if
the user enters 7, the program would print

*
**
***
****
*****
******
*******
******
*****
****
***
**
*

24. Redesign Listing 6.20 (startree.cpp) so that it draws a sideways tree pointing left; for example, if the
user enters 7, the program would print

©2013 Richard L. Halterman Draft date: January 27, 2013

6.7. EXERCISES 146

*
**
***
****
*****
******
*******
******
*****
****
***
**
*

©2013 Richard L. Halterman Draft date: January 27, 2013

 147

Chapter 7

Other Conditional and Iterative

Statements

The if/else and while statements are sufficient to implement any algorithms that involve conditional
execution and looping. The break and continue statements are convenient but are not necessary. C++
provides some additional conditional and iterative statements that are more convenient to use in some
circumstances. These additional statements include

• switch: an alternative to some multi-way if/else statements

• the conditional operator: an expression that exhibits the behavior of an if/else statement

• do/while: a loop that checks its condition after its body is executed

• for: a loop convenient for counting

These alternate constructs allow certain parts of algorithms to expressed more clearly and succinctly.
This chapter explores these other forms of expressing conditional execution and iteration.

7.1 The switch Statement

The switch statement provides a convenient alternative for some multi-way if/else statements like the
one in Listing 5.13 (restyleddigittoword.cpp). Listing 7.1 (switchdigittoword.cpp) is a new implementation
of Listing 5.13 (restyleddigittoword.cpp) that uses a switch statement instead of a multi-way if/else
statement.

Listing 7.1: switchdigittoword.cpp

1 #include <iostream>
2
3 using namespace std;
4
5 int main()
6 {
7 int value;

©2013 Richard L. Halterman Draft date: January 27, 2013

 7.1. THE SWITCH STATEMENT 148

8 cout << "Please enter an integer in the range 0...5: ";

9 cin >> value;
10 switch ( value )
11 {
12 case 0:
13 cout << "zero";
14 break;
15 case 1:
16 cout << "one";
17 break;
18 case 2:
19 cout << "two";
20 break;
21 case 3:
22 cout << "three";
23 break;
24 case 4:
25 cout << "four";
26 break;
27 case 5:
28 cout << "five";
29 break;
30 default:
31 if ( value < 0 )
32 cout << "Too small";
33 else
34 cout << "Too large";
35 break;
36 }
37 cout << endl;
38 }

The general form of a switch is:

©2013 Richard L. Halterman Draft date: January 27, 2013

7.1. THE SWITCH STATEMENT 149

switch ( integral expression ) {

case integral constant 1 :
statement(s)
break;
case integral constant 2 :
statement(s)
break;
case integral constant 3 :
statement(s)
break;
.
.
.
case integral constant n :
statement(s)
break;
default:
statement(s)
break;
}
In a switch statement

• The reserved word switch identifies a switch statement.

• The required parenthesized expression that follows the word switch must evaluate to an integral
value. Any integer type, characters, and Boolean expressions are acceptable. Floating point expres-
sions and other non-integer types are forbidden.

• The body of the switch is enclosed by required curly braces.

• Each occurrence of the word case is followed by an integral constant and a colon (:). We call the
integral constant a case label. This label can be either a literal value or a const symbolic value
(see Section 3.5). In particular, non-const variables and other expressions are expressly forbidden.
The case label defines a position within the code; it is not an executable statement. A case label
represents a target to which the program’s execution flow can jump.
If the case label matches the switch’s expression, then the statements that follow that label are
executed up until the break statement is encountered. The statements and break statement that
follow each case label are optional. One way to execute one set of statements for more than one
case label is to provide empty statements for one or more of the labels, as in:

©2013 Richard L. Halterman Draft date: January 27, 2013

7.1. THE SWITCH STATEMENT 150

cin >> key; // get key from user

switch ( key )
{
case 'p':
case 'P':
cout << "You choose \"P\"" << endl;
break;
case 'q':
case 'Q':
done = true;
break;
}

Here either an upper- or lowercase P result in the same action— You chose P is printed. If the user
enters either an upper- or lowercase Q, the done Boolean variable is set to true. If the user enters
neither P nor Q, none of the statements in the switch is executed.
The break statement is optional. When a case label is matched, the statements that follow are
executed until a break statement is encountered. The control flow then transfers out of the body of
the switch. In this way, the break within a switch works just like a break within a loop: the
rest of the body of the statement is skipped and program execution resumes at the next statement fol-
lowing the body. A missing break statement, a common error, when its omission is not intentional,
causes the statements of the succeeding case label to be executed. The process continues until a
break is encountered or the end of the switch body is reached.

• The default label is matched if none of the case labels match. It serves as a catch all option like
the final else in a multi-way if/else statement. The default label is optional. If it is missing
and none of the case labels match the expression, then no statement within the switch’s body is
executed.

The switch statement has two restrictions that make it less general than the multi-way if/else:

• The switch argument must be an integral expression.

• Case labels must be constant integral values. Integral literals and constants are acceptable. Variables
or expressions are not allowed.

To illustrate these restrictions, consider the following if/else statement that translates easily to an equiv-
alent switch statement:
if ( x == 1 )
{
// Do 1 stuff here . . .
}
else if ( x == 2 )
{
// Do 2 stuff here . . .
}
else if ( x == 3 )
{
// Do 3 stuff here . . .
}

©2013 Richard L. Halterman Draft date: January 27, 2013

7.2. THE CONDITIONAL OPERATOR 151

The corresponding switch statement is:

switch ( x )
{
case 1:
// Do 1 stuff here . . .
break;
case 2:
// Do 2 stuff here . . .
break;
case 3:
// Do 3 stuff here . . .
break;
}

Now consider the following if/else:

if ( x == y )
{
// Do "y" stuff here . . .
}
else if ( x > 2 )
{
// Do "> 2" stuff here . . .
}
else if ( z == 3 )
{
// Do 3 stuff here . . .
}

This code cannot be easily translated into a switch statement. The variable y cannot be used as a case
label. The second choice checks for an inequality instead of an exact match, so direct translation to a case
label is impossible. In the last condition, a different variable is checked, z instead of x. The control flow
of a switch statement is determined by a single value (for example, the value of x), but a multi-way
if/else statement is not so constrained.
Where applicable, a switch statement allows programmers to compactly express multi-way selection
logic. Most programmers find a switch statement easier to read than an equivalent multi-way if/else
construct.
A positive consequence of the switch statement’s restrictions is that it allows the compiler to produce
more efficient code for a switch than for an equivalent if/else. If a choice must be made from one of
several or more options, and the switch statement can be used, then the switch statement will likely be
faster than the corresponding multi-way if/else.

7.2 The Conditional Operator

As purely a syntactical convenience, C++ provides an alternative to the if/else construct called the
conditional operator. It has limited application but is convenient nonetheless. The following code fragment
assigns one of two things to x:

• the result of y/z, if z is nonzero, or

©2013 Richard L. Halterman Draft date: January 27, 2013

7.3. THE DO/WHILE STATEMENT 152

• zero, if z is zero; we wish to avoid the run-time error of division by zero.

// Assign a value to x:
if ( z != 0 )
x = y/z; // Division is possible
else
x = 0; // Assign a default value instead

This code has two assignment statements, but only one is executed at any given time. The conditional
operator makes for a more compact statement:
// Assign a value to x:
x = ( z != 0 ) ? y/z : 0;

The general form of a conditional expression is:

( condition ) ? exp1 : exp2

• condition is a normal Boolean expression that might appear in an if statement. Parentheses around
the condition are not required but should be used to improve the readability.
• exp1 is the overall value of the conditional expression if condition is true.
• exp2 is the overall value of the conditional expression if condition is false.

The conditional operator uses two symbols (? and :) and three operands. Since it has three operands it
is classified as a ternary operator (C++’s only one). The overall type of a conditional expression is the more
dominant of exp1 and exp2 The conditional expression can be used anywhere an expression can be used. It
is not a statement itself; it is used within a statement.
As another example, the absolute value of a number is defined in mathematics by the following formula:

n, when n ≥ 0
|n| =
−n, when n < 0

In other words, the absolute value of a positive number or zero is the same as that number; the abso-
lute value of a negative number is the additive inverse (negative of) of that number. The following C++
expression represents the absolute value of the variable n:
(n < 0) ? -n : n

Some argue that the conditional operator is cryptic, and thus its use reduces a program’s readability. To
seasoned C++ programmers it is quite understandable, but it is used sparingly because of its very specific
nature.

7.3 The do/while Statement

The while statement (Section 6.1) checks its condition before its body is executed; thus, it is a top-
checking loop. Its body is not executed if its condition is initially false. At times, this structure is inconve-
nient; for example, consider Listing 7.2 (goodinputonly.cpp).

©2013 Richard L. Halterman Draft date: January 27, 2013

 7.3. THE DO/WHILE STATEMENT 153

Listing 7.2: goodinputonly.cpp

1 #include <iostream>
2
3 using namespace std;
4
5 int main()
6 {
7 int in_value = -1;
8 cout << "Please enter an integer in the range 0-10: ";
9 // Insist on values in the range 0...10
10 while ( in_value < 0 || in_value > 10 )
11 cin >> in_value;
12 // in_value at this point is guaranteed to be within range
13 cout << "Legal value entered was " << in_value << endl;
14 }

The loop in Listing 7.2 (goodinputonly.cpp) traps the user in the while until a number in the desired
range is provided. Here’s how it works:

• The initialization of in_value to −1 ensures the condition of the while will be true initially, and,
thus, the body of the loop will be entered.

• The condition of the while specifies a set that includes all values that are not in the desired range.
in_value is initially in this set, so the loop is entered.

• The user does not get a chance to enter a value until program execution is inside the loop.

• The only way the loop can be exited is if the user enters a value that violates the condition—precisely
a value in the desired range.

The initialization of in_value before the loop check is somewhat artificial. It is there only to ensure
entry into the loop. It seems unnatural to check for a valid value before the user gets a chance to enter
it. A loop that checks its condition after its body is executed at least once would be more appropriate.
The do/while statement is a bottom-checking loop that behaves exactly in this manner. Listing 7.3
(betterinputonly.cpp) uses a do/while statement to check for valid input.

Listing 7.3: betterinputonly.cpp

1 #include <iostream>
2
3 using namespace std;
4
5 int main()
6 {
7 int in_value;
8 cout << "Please enter an integer in the range 0-10: ";
9 // Insist on values in the range 0...10
10 do
11 cin >> in_value;
12 while ( in_value < 0 || in_value > 10 );
13 // in_value at this point is guaranteed to be within range
14 cout << "Legal value entered was " << in_value << endl;
15 }

©2013 Richard L. Halterman Draft date: January 27, 2013

7.3. THE DO/WHILE STATEMENT 154

Notice that there is no need to initialize in_value since its value is not used until after it is assigned
through the input stream cin. Figure 7.1 compares the flowcharts of a while and do/while loop.

Figure 7.1: The flowcharts for while and do/while loops

The do/while statement has the general form:

do
statement;
while (condition );

• The reserved words do and while identify a do/while statement. The do and while keywords
delimit the loop’s body, but curly braces are still required if the body consists of more than one
statement.

• The condition is associated with the while at the end of the loop. The condition is a Boolean
expression and must be enclosed within parentheses.

• The statement is like the body of the while loop. It can be a compound statement enclosed within
curly braces.

The body of a do/while statement, unlike the while statement, is guaranteed to execute at least
once.

©2013 Richard L. Halterman Draft date: January 27, 2013

7.4. THE FOR STATEMENT 155

7.4 The for Statement

Recall Listing 6.2 (iterativecounttofive.cpp). It simply counts from one to five. Counting is a frequent
activity performed by computer programs. Certain program elements are required in order for any program
to count:

• A variable must be used to keep track of the count; in Listing 6.2 (iterativecounttofive.cpp), count
is the aptly named counter variable.
• The counter variable must be given an initial value. In the case of Listing 6.2 (iterativecounttofive.cpp),
the initial value is 1.
• The variable must be modified (usually incremented) as the program counts. The statement
count++;

increments count in Listing 6.2 (iterativecounttofive.cpp).

• A way must be provided to determine if the counting has completed. In Listing 6.2 (iterativecounttofive.cpp),
the condition of the while statement determines if the counting is complete or must contine.

C++ provides a specialized loop that packages these four programming elements into one convenient
statement. Called the for statement, its general form is

for ( initialization ; condition ; modification )

statement;
• The reserved word for identifies a for statement.
• The loop is controlled by a special variable called the loop variable.
• The header, contained in parentheses, contains three parts, each separated by semicolons:
– Initialization. The initialization part assigns an initial value to the loop variable. The loop
variable may be declared here as well; if it is declared here, then its scope is limited to the for
statement. This means you may use that loop variable only within the loop. It also means you
are free to reuse that variable’s name outside the loop to declare a different variable with the
same name as the loop variable.
The initialization part is performed one time.
– Condition. The condition part is a Boolean expression, just like the condition of a while
statement. The condition is checked each time before the body is executed.
– Modification. The modification part generally changes the loop variable. The change should be
such that the condition will eventually become false so the loop will terminate. The modification
is performed during each iteration after the body is executed.
Notice that the last part (modification) is not following by a semicolon; semicolons are used
strictly to separate the three parts.
• The statement is like the body of any other loop. It may be a compound statement within curly braces.

Any for loop can be rewritten as a while loop. The general form of the for loop given above can be
written equivalently as

©2013 Richard L. Halterman Draft date: January 27, 2013

 7.4. THE FOR STATEMENT 156

initialization;
while ( condition )
{
statement;
modification;
}
Listing 7.4 (forcounttofive.cpp) uses a for statement to count to five.

Listing 7.4: forcounttofive.cpp

1 #include <iostream>
2
3 using namespace std;
4
5 int main()
6 {
7 for ( int count = 1; count <= 5; count++ )
8 cout << count << endl; // Display counter
9 }

With a while loop, the four counting components (variable declaration, initialization, condition, and
modification can be scattered throughout the code. With a for loop, a programmer should be able to
determine all the important information about the loop’s control by looking at one statement.
Recall Listing 6.12 (timestable.cpp) that prints a multiplication table on the screen. We can organize
its code better by converting all the while statements to for statements. The result uses far less code, as
shown in Listing 7.5 (bettertimestable.cpp).

Listing 7.5: bettertimestable.cpp

1 #include <iostream>
2 #include <iomanip>
3
4 using namespace std;
5
6 int main()
7 {
8 int size; // The number of rows and columns in the table
9 cout << "Please enter the table size: ";
10 cin >> size;
11 // Print a size x size multiplication table
12
13 // First, print heading
14 cout << " ";
15 for ( int column = 1; column <= size; column++ )
16 cout << setw(4) << column; // Print heading for this column.
17 cout << endl;
18 // Print line separator
19 cout << " +";
20 for ( int column = 1; column <= size; column++ )
21 cout << "----"; // Print separator for this column.

©2013 Richard L. Halterman Draft date: January 27, 2013

 7.4. THE FOR STATEMENT 157

22 cout << endl;

23 // Print table contents
24 for ( int row = 1; row <= size; row++ )
25 {
26 cout << setw(4) << row << " |"; // Print row label.
27 for ( int column = 1; column <= size; column++ )
28 cout << setw(4) << row*column; // Display product
29 cout << endl; // Move cursor to next row
30 }
31 }

A for loop is ideal for stepping through the rows and columns. The information about the control of
both loops is now packaged in the respective for statements instead of being spread out in various places
in main. In the while version, it is easy for the programmer to forget to update one or both of the counter
variables (row and/or column). The for makes it harder for the programmer to forget the loop variable
update, since it is done right up front in the for statement header.
It is considered bad programming practice to do either of the following in a for statement:

• Modify the loop control variable within the body of the loop—if the loop variable is modified
within the body, then the logic of the loop’s control is no longer completely isolated to the for state-
ment’s header. The programmer must look elsewhere within the statement to understand completely
how the loop works.

• Prematurely exit the loop with a break—this action also violates the concept of keeping all the
loop control logic in one place (the for’s header).

The language allows both of these practices, but experience shows that it is best to avoid them. If it seems
necessary to violate this advice, consider using a different kind of loop. The while and do/while loops
do not imply the same degree of control regularity expected in a for loop.
Listing 7.6 (permuteabcd.cpp) is a rewrite of Listing 6.13 (permuteabc.cpp) that replaces its while
loops with for loops and adds an additional character.

Listing 7.6: permuteabcd.cpp

1 // File permuteabcd.cpp
2
3 #include <iostream>
4
5 using namespace std;
6
7 int main()
8 {
9 for ( char first = 'A'; first <= 'D'; first ++ )
10 for ( char second = 'A'; second <= 'D'; second++ )
11 if ( second != first ) // No duplicate letters
12 for ( char third = 'A'; third <= 'D'; third++ )
13 if ( third != first && third != second )
14 for ( char fourth = 'A'; fourth <= 'D'; fourth++ )
15 if ( fourth != first && fourth != second && fourth != third )
16 cout << first << second << third << fourth << endl;
17 }

©2013 Richard L. Halterman Draft date: January 27, 2013

 7.4. THE FOR STATEMENT 158

Notice that since all the variable initialization and incrementing is taken care of in the for statement
headers, we no longer need compound statements in the loop bodies, so the curly braces are unnecessary.
Listing 7.6 (permuteabcd.cpp) prints all 24 permutations of ABCD:

ABCD
ABDC
ACBD
ACDB
ADBC
ADCB
BACD
BADC
BCAD
BCDA
BDAC
BDCA
CABD
CADB
CBAD
CBDA
CDAB
CDBA
DABC
DACB
DBAC
DBCA
DCAB
DCBA

Listing 7.7 (forprintprimes.cpp) is a rewrite of Listing 6.21 (printprimes.cpp) that replaces its while
loops with for loops.

Listing 7.7: forprintprimes.cpp

1 #include <iostream>
2
3 using namespace std;
4
5 int main()
6 {
7
8 int max_value;
9 cout << "Display primes up to what value? ";
10 cin >> max_value;
11 for ( int value = 2; value <= max_value; value++ )
12 {
13 // See if value is prime
14 bool is_prime = true; // Provisionally, value is prime
15 // Try all possible factors from 2 to value - 1
16 for ( int trial_factor = 2;

©2013 Richard L. Halterman Draft date: January 27, 2013

 7.4. THE FOR STATEMENT 159

17 is_prime && trial_factor < value;

18 trial_factor++ )
19 is_prime = (value % trial_factor != 0 );
20 if ( is_prime )
21 cout << value << " "; // Display the prime number
22 }
23 cout << endl; // Move cursor down to next line
24 }

As shown in Listing 7.7 (forprintprimes.cpp), the conditional expression in the for loop is not limited
to a simple test of the loop control variable; it can be any legal Boolean expression. The logical and
(&&), or (||), and not (!) operators can be used to create complex Boolean expressions, if necessary.
The modification part of the for loop is not limited to simple arithmetic and can be quite elaborate. For
example:
for ( double d = 1000; d >= 1; cin >> d )
{
/* Body goes here */
}

Here d is reassigned from the input stream. If necessary, multiple variables can be initialized in the initial-
ization part:
for ( int i = 0, j = 100; i < j; i++ )
{
/* Body goes here */
}

While the for statement supports such complex headers, simpler is usually better. Ordinarily the for
loop should manage just one control variable, and the initialization, condition, and modification parts should
be straightforward. If a particular programming situation warrants an overly complicated for construction,
consider using another kind of loop.
Any or all of the parts of the for statement (initialization, condition, modification, and body) may be
omitted:

• Initialization. If the initialization is missing, as in

for (; i < 10; i++ )
/* Body goes here */

then no initialization is performed by the for loop, and it must be done elsewhere.
• Condition. If the condition is missing, as in
for ( int i = 0; ; i++ )
/* Body goes here */

then the condition is true by default. A break or goto must appear in the body unless an infinite
loop is intended.
• Modification. If the modification is missing, as in
for ( int i = 0; i < 10; )
/* Body goes here */

©2013 Richard L. Halterman Draft date: January 27, 2013

7.4. THE FOR STATEMENT 160

then the for performs no automatic modification; the modification must be done by a statement in
the body to avoid an infinite loop.

• Body. If the body is missing, as in

for ( int i = 0; i < 10; i++ ) {}

or
for ( int i = 0; i < 10; i++ );

then an empty loop results. This can be used for a non-portable delay (slower computers will delay
longer than faster computers), but some compilers may detect that such code has no functional effect
and optimize away such an empty loop. This means the compiler will ignore the for altogether.

As mentioned in Section 5.3, be careful about accidentally putting a semicolon at

the end of the for header, as in
for ( int i = 0; i < 10; i++ );
/* Intended body goes here */
The semicolon terminates the for statement, and the intended body that follows
is not the body, even though it may be properly indented.

One common C/C++ idiom to make an intentional infinite loop is to use a for statement with all control
information missing:
for ( ;; )
/* Body goes here */

Omitting all the parts of the for header is a statement from the programmer that says “I know what I am
doing—I really want an infinite loop here.” Of course, it actually may not be infinite because the body may
contain a break or goto statement.
While the for statement supports the omission of parts of its header, such constructs should be avoided.
The intention of the for loop is to allow the programmer to see all the aspects of the loop’s control in one
place. If some of these control responsibilities are to be handled elsewhere (not in the for’s header) then
consider using another kind of loop.
Programmers usually select a simple name for the control variable of a for statement. Recall that
variable names should be well chosen to reflect the meaning of their use within the program. It may come
as a surprise that i is probably the most common name used for an integer control variable in a for loop.
This practice has its roots in mathematics where variables such as i, j, and k are commonly used to index
vectors and matrices. Such mathematical structures have programming analogs in arrays (chrefch:arrays)
and vectors (Section 13.2). Computer programmers make considerable use of for loops in array and
vector processing, so programmers have universally adopted this convention of short control variable names.
Thus, it generally is acceptable to use simple identifiers like i as loop control variables.
C++ allows the break, continue, and goto statements to be used in the body of a for statement.
Like with the while and do/while statements, break causes immediate loop termination, continue
causes the condition to be immediately checked to determine if the iteration should continue, and goto
jumps to a label somewhere in the function. As previously mentioned, however, for loop control should be
restricted to its header, and the use of break, continue, and goto within for loops should be avoided.

©2013 Richard L. Halterman Draft date: January 27, 2013

7.5. SUMMARY 161

Any for loop can be rewritten with a while loop and behave identically. For example, consider the
for loop
for ( int i = 1; i <= 10; i++ )
cout << i << endl;

and next consider the while loop that behaves exactly the same way:
int i = 1;
while ( i <= 10 )
{
cout << i << endl;
i++;
}

Which is better? The for loop conveniently packages the loop control information in its header, but in
the while loop this information is distributed throughout the small section of code. The for loop thus
provides a better organization of the loop control code. Does one loop outperform the other? No, most
compilers produce essentially the same code for both constructs. Thus, the for loop is preferred in this
example.

7.5 Summary

• The switch statement is an alternative to some multi-way if/else statements.

• Not all multi-way if/else statements can be converted directly to a switch statement.

• The expression to evaluate within the switch statement must evaluate to an integral value (int,
short, long, char, or bool).

• The case labels within a switch statement must be integral literalsliterals or constants (int,
short, long, char, or bool). Specifically, case labels may not be variables or other expres-
sions.

• Program execution jumps to the case label with the value that matches the switch expression.

• Once a case label is matched, program execution continues within the switch statement until a
break statement is encountered.

• If no case labels match the switch expression, the program execution jumps to the default
label, if it is present. If no case labels match and no default label is present, no part of the
switch body is executed.

• The conditional operator is an expression that evaluates to one of two values depending on a given
condition.

• The do/while is a bottom-checking loop, unlike the while loop, which is a top-checking loop.

• The body of a do/while loop is always executed at least once, regardless of the condition. The
body of a while loop is not executed if the condition is initially false.

• The for loop is a top-checking loop that, when used properly, concentrates all the information about
its control in one convenient location.

©2013 Richard L. Halterman Draft date: January 27, 2013

7.6. EXERCISES 162

• The three parts of the for loop control are initialization, condition, and modification.
• The for statement is best used for a loop that can be controlled by a single variable with a definite
starting value, a definite ending value, and a regular way to update the variable’s value.
• A for loop is ideal for counting.
• Any or all of the three parts in the for loop header can be omitted; however, if you feel the need to
omit one or more of the parts, the while statement may be a better choice.
• Best practice avoids modifying the control variable of a for statement within the loop’s body; the
modification should be limited to the third part of the for header.
• The <iomanip> library provides the setw object that provides special formatting for output sent
to cout.

7.6 Exercises

1. Consider the following code fragment.

int x;
cin >> x;
switch ( x + 3 )
{
case 5:
cout << x << endl;
break;
case 10:
cout << x - 3 << endl;
break;
case 20:
cout << x + 3 << endl;
break;
}

(a) What is printed when the user enters 2?

(b) What is printed when the user enters 5?
(c) What is printed when the user enters 7?
(d) What is printed when the user enters 17?
(e) What is printed when the user enters 20?
(f) What is printed when the user enters 5?

2. Consider the following code fragment.

char ch;
cin >> ch;
switch ( ch )
{
case 'a':
cout << "*" << endl;

©2013 Richard L. Halterman Draft date: January 27, 2013

7.6. EXERCISES 163

break;
case 'A':
cout << "**" << endl;
break;
case 'B':
case 'b':
cout << "***" << endl;
case 'C':
case 'c':
cout << "****" << endl;
break;
default:
cout << "*****" << endl;
}

(a) What is printed when the user enters a?

(b) What is printed when the user enters A?
(c) What is printed when the user enters b?
(d) What is printed when the user enters B?
(e) What is printed when the user enters C?
(f) What is printed when the user enters c?
(g) What is printed when the user enters t?

3. What is printed by the following code fragment?

int x = 0;
do
{
cout << x << " ";
x++;
} while ( x < 10 );
cout << endl;

4. What is printed by the following code fragment?

int x = 20;
do
{
cout << x << " ";
x++;
} while ( x < 10 );
cout << endl;

5. What is printed by the following code fragment?

for ( int x = 0; x < 10; x++ )
cout << "*";
cout << endl;

6. Rewrite the following code fragment so that a switch is used instead of the if/else statements.

©2013 Richard L. Halterman Draft date: January 27, 2013

7.6. EXERCISES 164

int value;
char ch;
cin >> ch;
if ( ch == 'A' )
value = 10;
else if ( ch == 'P' )
value = 20;
else if ( ch == 'T' )
value = 30;
else if ( ch == 'V' )
value = 40;
else
value = 50;
cout << value << endl;

7. Rewrite the following code fragment so that a multi-way if/else is used instead of the switch
statement.
int value;
char ch;
cin >> ch;
switch( ch )
{
case 'A':
value = 10;
break;
case 'P':
cin >> value;
break;
case 'T':
value = ch;
break;
case 'V':
value = ch + 1000;
break;
default:
value = 50;
}
cout << value << endl;

8. Rewrite the following code fragment so that a multi-way if/else is used instead of the switch
statement.
int value;
char ch;
cin >> ch;
switch( ch )
{
case 'A':
cout << ch << endl;
value = 10;

©2013 Richard L. Halterman Draft date: January 27, 2013

7.6. EXERCISES 165

break;
case 'P':
case 'E':
cin >> value;
break;
case 'T':
cin >> ch;
value = ch;
case 'C':
value = ch;
cout << "value=" << value << ", ch=" << ch << endl;
break;
case 'V':
value = ch + 1000;
break;
}
cout << value << endl;

9. Rewrite the following code fragment so a while loop is used instead of the for statement.
for ( int i = 100; i > 0; i-- )
cout << i << endl;

10. Rewrite the following code fragment so that it uses the conditional operator instead of an if state-
ment:
if ( value % 2 != 0 ) // Is value even?
value = value + 1; // If not, make it even.

11. Rewrite the following code fragment so that it uses the conditional operator instead of an if/else
statement:
if ( value % 2 == 0 ) // Is value even?
value = 0; // If so, make it zero.
else
value = value + 1; // Otherwise, make it even.

12. Would the following multi-way if/else be a good candidate to rewrite as a switch statement?
If so, rewrite the code using a switch; otherwise, explain why it is impractical to do so.
int x, y;
cin >> x >> y;
if ( x < 10 )
y = 10;
else if ( x == 5 )
y = 5;
else if ( x == y )
y = 0;
else if ( y > 10 )
x = 10;
else
x = y;

©2013 Richard L. Halterman Draft date: January 27, 2013

7.6. EXERCISES 166

©2013 Richard L. Halterman Draft date: January 27, 2013

 167

Chapter 8

Using Functions

Suppose you must write a C++ program that computes the square root of a number supplied by the user.
Listing 8.1 (computesquareroot.cpp) provides a simple implementation.

Listing 8.1: computesquareroot.cpp

1 // File squareroot.cpp
2
3 #include <iostream>
4
5 using namespace std;
6
7 int main()
8 {
9 double input;
10
11 // Get value from the user
12 cout << "Enter number: ";
13 cin >> input;
14 double diff;
15 // Compute a provisional square root
16 double root = 1.0;
17
18 do // Loop until the provisional root
19 { // is close enough to the actual root
20 root = (root + input/root) / 2.0;
21 cout << "root is " << root << endl;
22 // How bad is the approximation?
23 diff = root * root - input;
24 }
25 while (diff > 0.0001 || diff < -0.0001);
26
27 // Report approximate square root
28 cout << "Square root of " << input << " = " << root << endl;
29 }

The program is based on a simple algorithm that uses successive approximations to zero in on an answer
that is within 0.00000001 of the true answer.

©2013 Richard L. Halterman Draft date: January 27, 2013

8.1. INTRODUCTION TO USING FUNCTIONS 168

One sample run is

Enter number: 2
root is 1.5
root is 1.41667
root is 1.41422
Square root of 2 = 1.41422

The actual square root is approximately 1.4142135623730951 and so the result is within our accepted
tolerance (0.0001). Another run is

Enter number: 100

root is 50.5
root is 26.2401
root is 15.0255
root is 10.8404
root is 10.0326
root is 10.0001
root is 10
Square root of 100 = 10

which is, of course, the exact answer.

While this code may be acceptable for many applications, better algorithms exist that work faster and
produce more precise answers. Another problem with this code is, what if you are working on a significant
scientific or engineering application and must compute square roots in various places throughout the source
code? Must we copy and paste the relevant portions of this square root code to each site in our source code
that requires a square root computation? Also, what if we develop other programs that use square root?
Do we copy this code into every program that needs to compute square roots, or is there a better way to
package the square root code and reuse it?
Code is made reusable by packaging it in functions. A function is a unit of reusable code. In Chapter 9
we will write our own reusable functions, but in this chapter we examine some of the functions available
in the C/C++ standard library. C++ provides a collection of standard precompiled C and C++ code stored
in libraries. Programmers can use parts of this library code within their own code to build sophisticated
programs.

8.1 Introduction to Using Functions

In mathematics, a function computes a result from a given value; for example, from the function definition
f (x) = 2x + 3, we can compute f (5) = 13 and f (0) = 3. A function in C++ works like a mathematical function.
To introduce the function concept, we will look at the standard C++ function that implements mathematical
square root.
In C++, a function is a named sequence of code that performs a specific task. A program uses a function
when specific processing is required. One example of a function is the mathematical square root function.

©2013 Richard L. Halterman Draft date: January 27, 2013

 8.1. INTRODUCTION TO USING FUNCTIONS 169

Such a function, named sqrt, is available to C and C++ programs (see Section 8.2). The square root
function accepts one double value and produces another double value; for example, the square root of
16 is 4, so when presented with 16.0, sqrt responds with 4.0. The square root function is visualized in
Figure 8.1. To the user of the square root function, the function is a black box; the user is concerned more

Figure 8.1: Conceptual view of the square root function

about what it does, not how it does it.

This sqrt function is exactly what we need for our square root program, Listing 8.1 (computesquareroot.cpp).
The new version, Listing 8.2 (standardsquareroot.cpp), uses the library function sqrt and eliminates the
complex logic of the original code.

Listing 8.2: standardsquareroot.cpp

1 #include <iostream>
2 #include <cmath>
3
4 using namespace std;
5
6 int main()
7 {
8 double input;
9
10 // Get value from the user
11 cout << "Enter number: ";
12 cin >> input;
13
14 // Compute the square root
15 double root = sqrt(input);
16
17 // Report result
18 cout << "Square root of " << input << " = " << root << endl;
19 }

The expression
sqrt(input)

is a function invocation, also known as a function call. A function provides a service to the code that uses
it. Here, our main function is the caller1 that uses the service provided by the sqrt function. We say
main calls, or invokes, sqrt passing it the value of input. The expression sqrt(input) evaluates to
the square root of the value of the variable input.
1 The term client can be used as well, although we reserve the term client for code that interacts with objects (see Chapter 13).

©2013 Richard L. Halterman Draft date: January 27, 2013

 8.1. INTRODUCTION TO USING FUNCTIONS 170

When calling a function, the function’s name is followed by parentheses that contain the information to
pass to the function so it can perform its task. In the expression
sqrt(input)

input is the information the function needs to do its work. We say input is the argument, or parameter,
passed to the function. The function cannot change the value of main’s input variable, it simply uses the
variable’s value to perform the computation. The sqrt function can be called in other ways, as illustrated
in Listing 8.3 (usingsqrt.cpp):

Listing 8.3: usingsqrt.cpp

1 /*
2 * This program shows the various ways the
3 * sqrt function can be used.
4 */
5
6 #include <iostream>
7 #include <cmath>
8
9 using namespace std;
10
11 int main()
12 {
13 double x = 16.0;
14 // Pass a literal value and display the result
15 cout << sqrt(16.0) << endl;
16 // Pass a variable and display the result
17 cout << sqrt(x) << endl;
18 // Pass an expression
19 cout << sqrt(2 * x - 5) << endl;
20 // Assign result to variable
21 double y = sqrt(x);
22 // Use result in an expression
23 y = 2 * sqrt(x + 16) - 4;
24 // Use result as argument to a function call
25 y = sqrt(sqrt(256.0));
26 cout << y << endl;
27 }

The sqrt function accepts a single numeric argument that must be a float, double, or long double.
The parameter that a caller can pass to sqrt can be a literal number, a numeric variable, an arithmetic
expression, or even a function invocation that produces an acceptable numeric result.
Some functions, like sqrt, compute a value that is returned to the caller. The caller can use this result
in various ways, as shown in Listing 8.3 (usingsqrt.cpp). The next to the last statement computes the square
root of the square root of 256 (which is 4).
If the caller code attempts to pass a parameter to the function that is incompatible with type expected
by the function, the compiler will issue an error.
cout << sqrt("16") << endl; // Illegal, a string is not a number

Listing 8.3 (usingsqrt.cpp) shows that a program can call the sqrt function as many times and in as
many places as needed. As noted in Figure 8.1, to the caller of the square root function, the function is a

©2013 Richard L. Halterman Draft date: January 27, 2013

8.1. INTRODUCTION TO USING FUNCTIONS 171

black box; the caller is concerned strictly about what the function does, not how the function accomplishes
its task.
We safely can treat all functions like black boxes. We can use the service that a function provides
without being concerned about its internal details. We are guaranteed that we can influence the function’s
behavior only via the parameters that we pass, and that nothing else we do can affect what the function does
or how it does it. Furthermore, the function cannot affect any of our code, apart from what we do with the
value it computes.
Some functions take more than one parameter; for example, the C++ max function requires two argu-
ments in order to produce a result. The max function selects and returns the larger of the two parameters.
The max function is visualized in Figure 8.2. The max function could be used as

Figure 8.2: Conceptual view of the maximum function

cout << "The larger of " << 4 << " and " << 7 << " is " << max(4, 7) << endl;

Notice that the parameters are contained in parentheses following the function’s name, and the parameters
are separated by commas.
From the caller’s perspective a function has three important parts:

• Name. Every function has a name that identifies the location of the code to be executed. Function
names follow the same rules as variable names; a function name is another example of an identifier
(see Section 3.3).
• Parameter type(s). A function must be called with the exact number and types of parameters that it
expects. If a caller attempts to call a function with too many or too few parameters, the compiler issues
an error message and does not compile the code. Similarly, if the parameters the caller passes during
a call are not compatible with the types specified for the function, the compiler reports appropriate
error messages.
• Result type. A function can compute a result and return this value to the caller. The caller’s use
of this result must be compatible with the function’s specified result type. The result type and the
parameter types can be completely unrelated.

These three crucial pieces of information are formally described for each function in a specification
known as a function prototype. The prototype for the sqrt function is
double sqrt(double)

and the max function’s prototype can be expressed as 2

2 The prototype for the actual library function max uses generic types; generic types are beyond the scope of this introductory

book, so the prototype provided here is strictly for illustrating a point.

©2013 Richard L. Halterman Draft date: January 27, 2013

8.1. INTRODUCTION TO USING FUNCTIONS 172

int max(int, int)

In a function prototype, the return type is first listed, followed by the function’s name, and then the param-
eter types appear in parentheses. Sometimes it is useful to list parameter names in the function’s prototype,
as in
double sqrt(double n)

or
int max(int a, int b)

The specific parameter names are irrelevant. The names make it easier to describe what the function does;
for example, sqrtcomputes the square root of n and max determines the larger of a and b.
When using a library function the programmer must include the appropriate #include directive in
the source code. The file specified in an #include directive contains prototypes for library functions. In
order to use the sqrt function, a program must include the
#include <cmath>

preprocessor directive. For the max function, include

#include <algorithm>

although under Visual C++, #includeing <iostream> is sufficient.

Armed with the function prototypes, the compiler can check to see if the calling code is using the library
functions correctly. An attempt to use the sqrt function as
cout << sqrt(4.0, 7.0) << endl; // Error

will result in an error, because the prototype for sqrt specifies only one numeric parameter, not two.
Some functions do not accept parameters; for example, the C++ function to generate a pseudorandom
number, rand, is called with no arguments:
cout << rand() << endl;

The rand function returns an int value, but the caller does not pass the function any information to do its
task. The rand prototype is
int rand()

Notice the empty parentheses that indicate this function does not accept any parameters.
Unlike mathematical functions that must produce a result, a C++ function is not required to return a
value to the caller. The C++ function exit expects an integer value from the caller, but it does not return a
result back to the caller. A prototype for a function that returns nothing uses void as the return type, as in:
void exit(int);

The exit function immediately terminates the program’s execution. The integer argument passed to exit
is returned to the operating system which can use the value to determine if the program terminated normally
or due to an error. C++ programs automatically return zero when main finishes executing—no exit call
is necessary.
Note that since exit does not return a value to the caller, code such as

©2013 Richard L. Halterman Draft date: January 27, 2013

8.2. STANDARD MATH FUNCTIONS 173

cout << exit(8) << endl; // Illegal!

will not compile since the expression exit(8) evaluates to nothing, and the cout stream object requires
an actual value of some kind to print. A void function is called for the side effects it produces instead a
value it computes.
We will see in Chapter 9 that a function may accept no parameters and return nothing, but there are no
standard library functions of this form.

8.2 Standard Math Functions

The cmath library provides much of the functionality of a scientific calculator. Table 8.1 lists only a few
of the available functions.

mathfunctions Module
double sqrt(double x) √
Computes the square root of a number: sqrt(x) = x
double exp(double x)
Computes e raised a power: exp(x) = ex
double log(double x)
Computes the natural logarithm of a number: log(x) = loge x = ln x
double log10(double x)
Computes the common logarithm of a number: log(x) = log10 x
double cos(double)
Computes the cosine of a value specified in radians: cos(x) = cos x; other trigonometric
functions include sine, tangent, arc cosine, arc sine, arc tangent, hyperbolic cosine, hyper-
bolic sine, and hyperbolic tangent
double pow(double x, double y)
Raises one number to a power of another: pow(x, y) = xy
double fabs(double x)
Computes the absolute value of a number: fabs(x) = |x|

Table 8.1: A few of the functions from the cmath library

The cmath library also defines a constant named HUGE_VAL. This constant can be used to represent
infinity or an undefined value such the slope of a vertical line or a fraction with a zero denominator. A
complete list of the numeric functions available to C++ can be found at http://www.cplusplus.com/
reference/clibrary/cmath/.

Be careful to put the function’s arguments in the proper order when calling a
function; for example, the call pow(10,2) computes 102 = 100, but the call
pow(2,10) computes 210 = 1, 024.

A C++ program that uses any of the functions from the cmath library must use the following prepro-
cessor #include directive:

©2013 Richard L. Halterman Draft date: January 27, 2013

8.2. STANDARD MATH FUNCTIONS 174

#include <cmath>

Figure 8.3 shows a problem that can be solved using functions found in the cmath library. Suppose a
spacecraft is at a fixed location in space some distance from a planet. A satellite is orbiting the planet in a
circular orbit. We wish to compute how much farther away the satellite will be from the spacecraft when it
has progressed 10 degrees along its orbital path.

(x2 ,y2 )

d2
θ
(x1 ,y1 ) d1
(0,0) (px ,py )

Figure 8.3: Orbital distance problem

We will let the origin of our coordinate system (0,0) be located at the center of the planet which cor-
responds also to the center of the circular orbital path. The satellite is initially at point (x1 , y1 ) and the
spacecraft is stationary at point (px , py ). The spacecraft is located in the same plane as the satellite’s orbit.
We need to compute the difference in the distances between the moving point (satellite) and the fixed point
(spacecraft) at two different times during the satellite’s orbit.
Two problems must be solved, and facts from mathematics provide the answers:

1. Problem: The location of the moving point must be recomputed as it moves along the circle.
Solution: Given an initial position (x1 , y1 ) of the moving point, a rotation of θ degrees around the
origin will yield a new point at (x2 , y2 ), where

x2 = x1 cos θ − y1 sin θ
y2 = x1 sin θ + y1 cos θ

2. Problem: The distance between the moving point and the fixed point must be recalculated as the
moving point moves to a new position.
Solution: The distance d1 in Figure 8.3 between two points (px , py ) and (x1 , y1 ) is given by the
formula q
d1 = (x1 − px )2 + (y1 − py )2

Similarly, the distance d2 in Figure 8.3 is

q
d2 = (x2 − px )2 + (y2 − py )2

Listing 8.4 (orbitdist.cpp) uses these mathematical results to compute the difference in the distances.

©2013 Richard L. Halterman Draft date: January 27, 2013

 8.2. STANDARD MATH FUNCTIONS 175

Listing 8.4: orbitdist.cpp

1 #include <iostream>
2 #include <cmath>
3
4 using namespace std;
5
6 int main()
7 {
8 // Location of orbiting point is (x,y)
9 double x; // These values change as the
10 double y; // satellite moves
11 const double PI = 3.14159;
12
13 // Location of fixed point is always (100, 0),
14 // AKA (p_x, p_y). Change these as necessary.
15 const double p_x = 100;
16 const double p_y = 0;
17
18 // Radians in 10 degrees
19 const double radians = 10 * PI/180;
20
21 // Precompute the cosine and sine of 10 degrees
22 const double COS10 = cos(radians);
23 const double SIN10 = sin(radians);
24
25 // Get starting point from user
26 cout << "Enter initial satellite coordinates (x,y):";
27 cin >> x >> y;
28
29 // Compute the initial distance
30 double d1 = sqrt((p_x - x)*(p_x - x) + (p_y - y)*(p_y - y));
31
32 // Let the satellite orbit 10 degrees
33 double x_old = x; // Remember x's original value
34 x = x*COS10 - y*SIN10; // Compute new x value
35 // x's value has changed, but y's calculate depends on
36 // x's original value, so use x_old instead of x.
37 y = x_old*SIN10 + y*COS10;
38
39 // Compute the new distance
40 double d2 = sqrt((p_x - x)*(p_x - x) + (p_y - y)*(p_y - y));
41
42 // Print the difference in the distances
43 cout << "Difference in distances: " << d2 - d1 << endl;
44 }

We can use the square root function to improve the efficiency of our primes program. Instead of trying
all the factors of n up to n − 1, we need only try potential factors up to the square root of n. Listing 8.5
(moreefficientprimes.cpp) uses the sqrt function to reduce the number of factors that need be considered.

Listing 8.5: moreefficientprimes.cpp

1 #include <iostream>
2 #include <cmath>
3

©2013 Richard L. Halterman Draft date: January 27, 2013

 8.3. MAXIMUM AND MINIMUM 176

4 using namespace std;

5
6 int main()
7 {
8
9 int max_value;
10 cout << "Display primes up to what value? ";
11 cin >> max_value;
12 for ( int value = 2; value <= max_value; value++ )
13 {
14 // See if value is prime
15 bool is_prime = true; // Provisionally, value is prime
16 double r = value, root = sqrt(r);
17 // Try all possible factors from 2 to the square
18 // root of value
19 for ( int trial_factor = 2;
20 is_prime && trial_factor <= root; trial_factor++ )
21 is_prime = (value % trial_factor != 0 );
22 if ( is_prime )
23 cout << value << " "; // Display the prime number
24 }
25 cout << endl; // Move cursor down to next line
26 }

The sqrt function comes in three forms:

double sqrt(double)
float sqrt(float)
long double sqrt(long double)

The function names are the same, but the parameter types differ. We say that the sqrt function is over-
loaded. (Overloaded functions are covered in more detail in Section 10.3.) When a caller invokes the sqrt
function, the compiler matches the call to the closest matching prototype. If the caller passes a double
parameter, the compiler generates code to call the double version. If the caller instead passes a float
variable, the compiler selects the float version of sqrt. When an int is passed to sqrt, the compiler
cannot decide which version to use, because an int can be converted automatically to either a float,
double, or long double. The compiler thus needs some help to resolve the ambiguity, so we intro-
duced an additional variable of type double so the compiler will use the double version of the sqrt
function. Another option is to use a type cast to convert the integer value into one of the types acceptable
to the sqrt function.

8.3 Maximum and Minimum

C++ provides standard functions for determining the maximum and minimum of two numbers. Listing 8.6
(maxmin.cpp) shows how they can be used.

Listing 8.6: maxmin.cpp

1 #include <iostream>
2
3 using namespace std;
4

©2013 Richard L. Halterman Draft date: January 27, 2013

 8.4. CLOCK FUNCTION 177

5 int main()
6 {
7 int value1, value2;
8 cout << "Please enter two integer values: ";
9 cin >> value1 >> value2;
10 cout << "max = " << max(value1, value2)
11 << ", min = " << min(value1, value2) << endl;
12 }

8.4 clock Function

The clock function from the <ctime> library requests from the operating system the amount of time an
executing program has been running. The units returned by the call clock() is system dependent, but it
can be converted into seconds with the constant CLOCKS_PER_SEC, also defined in the ctime library.
Under Visual C++, the CLOCKS_PER_SEC constant is 1,000, which means the call clock() returns the
number of milliseconds that the program has been running.
Using two calls to the clock function you can measure elapsed time. Listing 8.7 (timeit.cpp) measures
how long it takes a user to enter a character from the keyboard.

Listing 8.7: timeit.cpp

1 #include <iostream>
2 #include <ctime>
3
4 using namespace std;
5
6 int main()
7 {
8 char letter;
9 cout << "Enter a character: ";
10 clock_t seconds = clock(); // Record starting time
11 cin >> letter;
12 clock_t other = clock(); // Record ending time
13 cout << static_cast<double>(other - seconds)/CLOCKS_PER_SEC
14 << " seconds" << endl;
15 }

The type clock_t is a type defined in the <ctime> header file. clock_t is equivalent to an
unsigned long, and you can perform arithmetic on clock_t values and variables just as if they are
unsigned longs. In the expression
static_cast<double>(other - seconds)/CLOCKS_PER_SEC

the cast is required to force floating-point division; otherwise, the result is truncated to an integer value.
Listing 8.8 (measureprimespeed.cpp) measures how long it takes a program to display all the prime
numbers up to half a million using the algorithm from Listing 7.7 (forprintprimes.cpp).

Listing 8.8: measureprimespeed.cpp

1 #include <iostream>
2 #include <ctime>

©2013 Richard L. Halterman Draft date: January 27, 2013

 8.4. CLOCK FUNCTION 178

3 #include <cmath>
4
5 using namespace std;
6
7 // Display the prime numbers between 2 and 500,000 and
8 // time how long it takes
9
10 int main()
11 {
12 clock_t start_time = clock(), // Record start time
13 end_time;
14 for ( int value = 2; value <= 500000; value++ )
15 {
16 // See if value is prime
17 bool is_prime = true; // Provisionally, value is prime
18 // Try all possible factors from 2 to n - 1
19 for ( int trial_factor = 2;
20 is_prime && trial_factor < value;
21 trial_factor++ )
22 is_prime = (value % trial_factor != 0 );
23 if ( is_prime )
24 cout << value << " "; // Display the prime number
25 }
26 cout << endl; // Move cursor down to next line
27 end_time = clock();
28 // Print the elapsed time
29 cout << "Elapsed time: "
30 << static_cast<double>(end_time - start_time)/CLOCKS_PER_SEC
31 << " sec." << endl;
32 }

On one system, the program took 93 seconds, on average, to print all the prime numbers up to 500,000.
By comparison, the newer, more efficient version, Listing 8.5 (moreefficientprimes.cpp), which uses the
square root optimization takes only 15 seconds to display all the primes up to 500,000. Exact times will
vary depending on the speed of the computer.
As it turns out, much of the program’s execution time is taken up printing the output, not computing the
prime numbers to print. We can compare the algorithms better by redirecting the program’s output to a file.
If the executable program is named primes.exe, you can redirect its output at the command line by issuing
the command

primes > run1.out

This creates a text file named run1.out that can be viewed with any text editor. Its contents are exactly what
would have been printed to the screen if the redirection is not used.
When run using redirection, the time difference is even more dramatic: The unoptimized version gen-
erates the prime numbers up to 500,000 in 77 seconds, while the optimized square root version requires
only 2 seconds to generate the same number of primes! An even faster prime generator can be found in
Listing 11.7 (fasterprimes.cpp); it uses a completely different algorithm to generate prime numbers.
The ctime header must be #included to use the standard time function in a program.

©2013 Richard L. Halterman Draft date: January 27, 2013

 8.5. CHARACTER FUNCTIONS 179

8.5 Character Functions

The C library provides a number of character functions that are useful to C++ programmers. Listing 8.9
(touppercase.cpp) converts lowercase letters to uppercase letters.

Listing 8.9: touppercase.cpp

1 #include <iostream>
2 #include <cctype>
3
4 using namespace std;
5
6 int main()
7 {
8 for ( char lower = 'a'; lower <= 'z'; lower++ )
9 {
10 char upper = toupper(lower);
11 cout << lower << " => " << upper << endl;
12 }
13 }

The first lines printed by Listing 8.9 (touppercase.cpp) are

a => A
b => B
c => C
d => D

Interestingly, the toupper function returns an int, not a char. At the enhanced warning level 4 for
Visual C++ a cast is required to assign the result to the variable upper:

char upper = static_cast<char>(toupper(lower));

Some of the more useful character functions are described in Table 8.2.

©2013 Richard L. Halterman Draft date: January 27, 2013

8.6. RANDOM NUMBERS 180

charfunctions Module
int toupper(int ch)
Returns the uppercase version of the given character; returns the original character if no
uppercase version exists (such as for punctuation or digits)
int tolower(int ch)
Returns the lowercase version of the given character; returns the original character if no
lowercase version exists (such as for punctuation or digits)
int isupper(int ch)
Returns a nonzero value (true) if ch is an uppercase letter (’A’–’Z’); otherwise, it returns
0 (false)
int islower(int ch)
Returns a nonzero value (true) if ch is an lowercase letter (’a’–’z’); otherwise, it returns
0 (false)
int isalpha(int ch)
Returns a nonzero value (true) if ch is a letter from the alphabet (’A’–’Z’ or ’a’–’z’);
otherwise, it returns 0 (false)
int isdigit(int ch)
Returns a nonzero value (true) if ch is a digit (’0’–’9’); otherwise, it returns 0 (false)

Table 8.2: A few of the functions from the cctype library

Other functions exist to determine if a character is a punctuation character like a comma or semicolon
(ispunct), a space, tab, or newline character (isspace).
To use the standard C character functions in your C++ program, you must include the <cctype> header
file.

8.6 Random Numbers

Some applications require behavior that appears random. Random numbers are useful particularly in games
and simulations. For example, many board games use a die (one of a pair of dice) to determine how many
places a player is to advance. (See Figure 8.4.) A die or pair of dice are used in other games of chance. A
die is a cube containing spots on each of its six faces. The number of spots range from one to six. A player
rolls a die or sometimes a pair of dice, and the side(s) that face up have meaning in the game being played.
The value of a face after a roll is determined at random by the complex tumbling of the die. A software
adaptation of a game that involves dice would need a way to simulate the random roll of a die.
All algorithmic random number generators actually produce pseudorandom numbers, not true random
numbers. A pseudorandom number generator has a particular period, based on the nature of the algorithm
used. If the generator is used long enough, the pattern of numbers produced repeats itself exactly. A
sequence of true random numbers would not contain such a repeating subsequence. The good news is
that all practical algorithmic pseudorandom number generators have periods that are large enough for most
applications.
C++ programmers can use two standard C functions for generating pseudorandom numbers: srand and
rand:
void srand(unsigned)
int rand()

©2013 Richard L. Halterman Draft date: January 27, 2013

 8.6. RANDOM NUMBERS 181

Figure 8.4: A pair of dice

srand establishes the first value in the sequence of pseudorandom integer values. Each call to rand
returns the next value in the sequence of pseudorandom values. Listing 8.10 (simplerandom.cpp) shows
how a sequence of 100 pseudorandom numbers can be printed.

Listing 8.10: simplerandom.cpp

1 #include <iostream>
2 #include <cstdlib>
3
4 using namespace std;
5
6 int main()
7 {
8 srand(23);
9 for ( int i = 0; i < 100; i++ )
10 {
11 int r = rand();
12 cout << r << " ";
13
14 }
15 cout << endl;
16 }

The numbers printed by the program appear to be random. The algorithm is given a seed value to
begin, and a formula is used to produce the next value. The seed value determines the sequence of numbers
generated; identical seed values generate identical sequences. If you run the program again, the same
sequence is displayed, because the same seed value, 23, is used. In order to allow each program run
to display different sequences, the seed value must be different for each run. How can we establish a
different seed value for each run? The best way to make up a “random” seed at run time is to use the time
function which is found in the ctime library. The call time(0) returns the number of seconds since
midnight January 1, 1970. This value obviously differs between program runs, so each execution will use
a different seed value, and the generated pseudorandom number sequences will be different. Listing 8.11
(betterrandom.cpp) incorporates the time function to improve its randomness over multiple executions.

Listing 8.11: betterrandom.cpp

1 #include <iostream>
2 #include <cstdlib>

©2013 Richard L. Halterman Draft date: January 27, 2013

 8.6. RANDOM NUMBERS 182

3 #include <ctime>
4
5 using namespace std;
6
7 int main()
8 {
9 srand(static_cast<unsigned>(time(0)));
10 for ( int i = 0; i < 100; i++ )
11 {
12 int r = rand();
13 cout << r << " ";
14
15 }
16 cout << endl;
17 }

Each execution of Listing 8.11 (betterrandom.cpp) produces a different pseudorandom number se-
quence. The actual type of value that time returns is time_t, so the result from a call to time must
be cast to unsigned int before being used with srand.
Notice that the numbers returned by rand can be rather large. The pseudorandom values range from 0
to a maximum value that is implementation dependent. The maximum value for Visual C++ś rand function
is 32,767, which corresponds to the largest 16-bit signed int value. The cstdlib header defines the
constant RAND_MAX that represents the largest value in the range. The following statement
cout << RAND_MAX << endl;

would print the value of RAND_MAX for a particular system.

Ordinarily we need values in a more limited range, like 1...100. Simple arithmetic with the modulus
operator can produce the result we need. If n is any non-negative integer and m is any positive integer, the
expression

n%m

produces a value in the range 0 . . . − 1.

This means the statement
int r = rand() % 100;

can assign only values in the range 0...99 to r. If we really want values in the range 1...100, what can we
do? We simply need only add one to the result:
int r = rand() % 100 + 1;

This statement produces pseudorandom numbers in the range 1...100.

We now have all we need to write a program that simulates the rolling of a die.
Listing 8.12 (die.cpp) simulates rolling die.

Listing 8.12: die.cpp

1 #include <iostream>
2 #include <cstdlib>

©2013 Richard L. Halterman Draft date: January 27, 2013

 8.6. RANDOM NUMBERS 183

3 #include <ctime>
4
5 using namespace std;
6
7 int main()
8 {
9 // Set the random seed value
10 srand(static_cast<unsigned>(time(0)));
11
12 // Roll the die three times
13 for ( int i = 0; i < 3; i++ )
14 {
15 // Generate random number in the range 1...6
16 int value = rand() % 6 + 1;
17
18
19 // Show the die
20 cout << "+-------+" << endl;
21 switch ( value )
22 {
23 case 1:
24 cout << "| |" << endl;
25 cout << "| * |" << endl;
26 cout << "| |" << endl;
27 break;
28 case 2:
29 cout << "| * |" << endl;
30 cout << "| |" << endl;
31 cout << "| * |" << endl;
32 break;
33 case 3:
34 cout << "| * |" << endl;
35 cout << "| * |" << endl;
36 cout << "| * |" << endl;
37 break;
38 case 4:
39 cout << "| * * |" << endl;
40 cout << "| |" << endl;
41 cout << "| * * |" << endl;
42 break;
43 case 5:
44 cout << "| * * |" << endl;
45 cout << "| * |" << endl;
46 cout << "| * * |" << endl;
47 break;
48 case 6:
49 cout << "| * * * |" << endl;
50 cout << "| |" << endl;
51 cout << "| * * * |" << endl;
52 break;
53 default:
54 cout << " *** Error: illegal die value ***" << endl;
55 break;
56 }
57 cout << "+-------+" << endl;

©2013 Richard L. Halterman Draft date: January 27, 2013

 8.7. SUMMARY 184

58 }
59 }

The output of one run of Listing 8.12 (die.cpp) is

+-------+
| * * |
| |
| * * |
+-------+
+-------+
| * * * |
| |
| * * * |
+-------+
+-------+
| |
| * |
| |
+-------+

Since the values are pseudorandomly generated, actual output will vary from one run to the next.

8.7 Summary

• The C and C++ standard library provides a collection of routines that can be incorporated into code
that you write.

• When faced with the choice of using a standard library function or writing your own code to solve
the same problem, choose the library function. The standard function will be tested thoroughly, well
documented, and likely more efficient than the code you would write.

• The function is a standard unit of reuse in C++.

• Code that uses a function is known as calling code.

• A function has a name, a list of parameters (which may be empty), and a result type (which may be
void). A function performs some computation or action that is useful to callers. Typically a function
produces a result based on the parameters passed to it.

• Callers communicate information to a function via its parameters (also known as arguments).

• In order to use a standard function, callers must #include the proper header file.

• The arguments passed to a function by a caller consist of a comma-separated list enclosed by paren-
theses.

• Calling code (code outside the function’s definition that invoke the function) must pass the correct
number and types of parameters that the function expects.

©2013 Richard L. Halterman Draft date: January 27, 2013

8.8. EXERCISES 185

• The C standard library (<cmath>) includes a variety of mathematical functions.

• The time function from the <ctime> library returns the number of seconds since midnight, January
1, 1970.

• The C standard library (<cctype>) includes a variety of character manipulation functions.

• The rand function (from <cstdlib>) returns a pseudorandom integer. The first pseudorandom
number can be set with the srand function. The constant RAND_MAX is largest possible value
returned by rand.

8.8 Exercises

1. Suppose you need to compute the square root of a number in a C++ program. Would it be a good idea
to write the code to perform the square root calculation? Why or why not?
2. Which one of the following values could be computed by the rand function?
4.5 34 -1 MAX_RAND + 1

3. Classify each of the following expressions as legal or illegal. Each expression represents a call to a
standard C/C++ library function.
(a) sqrt(4.5)
(b) sqrt(4.5, 3.1)
(c) rand(4)
(d) srand()
(e) tolower('A')
4. From geometry: Write a computer program that given the lengths of the two sides of a right triangle
adjacent to the right angle computes the length of the hypotenuse of the triangle. (See Figure 8.5.)
If you are unsure how to solve the problem mathematically, do a web search for the Pythagorean
theorem.
5. From trigonometry: Write a computer program that given the lengths of the two sides of a right
triangle adjacent to the right angle computes the length of the hypotenuse of the triangle. If you are
unsure how to solve the problem mathematically, do a web search for the Pythagorean theorem.
6.

©2013 Richard L. Halterman Draft date: January 27, 2013

8.8. EXERCISES 186

Hy
po
ten
Side 2

us
e

Side 1

Figure 8.5: Right triangle

©2013 Richard L. Halterman Draft date: January 27, 2013

 187

Chapter 9

Writing Functions

As programs become more complex, programmers must structure their programs in such a way as to ef-
fectively manage their complexity. Most humans have a difficult time keeping track of too many pieces of
information at one time. It is easy to become bogged down in the details of a complex problem. The trick
to managing complexity is to break down the problem into more manageable pieces. Each piece has its
own details that must be addressed, but these details are hidden as much as possible within that piece. The
problem is ultimately solved by putting these pieces together to form the complete solution.
So far all of our programs have been written within one function—main. As the number of statements
within a function increases, the function can become unwieldy. The code within such a function that does
all the work by itself is called monolithic code. Monolithic code that is long and complex is undesirable for
several reasons:

• It is difficult to write correctly. All the details in the entire piece of code must be considered when
writing any statement within that code.

• It is difficult to debug. If the sequence of code does not work correctly, it is often difficult to find
the source of the error. The effects of an erroneous statement that appears earlier in the code may not
become apparent until a correct statement later uses the erroneous statement’s incorrect result.

• It is difficult to extend. All the details in the entire sequence of code must be well understood before
it can be modified. If the code is complex, this may be a formidable task.

Using a divide and conquer strategy, a programmer can decompose a complicated function (like main)
into several simpler functions. The original function can then do its job by delegating the work to these
other functions. In this way the original function can be thought of as a “work coordinator.”
Besides their code organization aspects, functions allow us to bundle functionality into reusable parts. In
Chapter 8 we saw how library functions can dramatically increase the capabilities of our programs. While
we should capitalize on library functions as much as possible, sometimes we need a function exhibiting
custom behavior that is not provided by any standard function. Fortunately we can create our own functions,
and the same function may be used (called) in numerous places within a program. If the function’s purpose
is general enough and we write the function properly, we may be able to reuse the function in other programs
as well.

©2013 Richard L. Halterman Draft date: January 27, 2013

 9.1. FUNCTION BASICS 188

9.1 Function Basics

Recall the “handwritten” square root code we saw in Listing 8.1 (computesquareroot.cpp). We know that
the better option is the standard library function sqrt; however, we will illustrate custom function devel-
opment by writing our own square root function based on the code in Listing 8.1 (computesquareroot.cpp).
In Listing 9.1 (customsquareroot.cpp) we see the definition for the square_root function.

Listing 9.1: customsquareroot.cpp

1 #include <iostream>
2 #include <iomanip>
3 #include <cmath>
4
5 using namespace std;
6
7 // Compute an approximation of
8 // the square root of x
9 double square_root(double x)
10 {
11 double diff;
12 // Compute a provisional square root
13 double root = 1.0;
14
15 do // Loop until the provisional root
16 { // is close enough to the actual root
17 root = (root + x/root) / 2.0;
18 //cout << "root is " << root << endl;
19 // How bad is the approximation?
20 diff = root * root - x;
21 }
22 while (diff > 0.0001 || diff < -0.0001);
23 return root;
24 }
25
26 int main()
27 {
28 // Compare the two ways of computing the square root
29 for (double d = 1.0; d <= 10.0; d += 0.5)
30 cout << setw(7) << square_root(d) << " : " << sqrt(d) << endl;
31 }

The main function in Listing 9.1 (customsquareroot.cpp) compares the behavior of our custom square_root
function to the sqrt library function. Its output:

©2013 Richard L. Halterman Draft date: January 27, 2013

9.1. FUNCTION BASICS 189

1 : 1
1.22474 : 1.22474
1.41422 : 1.41421
1.58116 : 1.58114
1.73205 : 1.73205
1.87083 : 1.87083
2 : 2
2.12132 : 2.12132
2.23607 : 2.23607
2.34521 : 2.34521
2.44949 : 2.44949
2.54952 : 2.54951
2.64577 : 2.64575
2.73861 : 2.73861
2.82843 : 2.82843
2.91548 : 2.91548
3 : 3
3.08221 : 3.08221
3.16228 : 3.16228

shows a few small differences in the results. Clearly we should use the standard sqrt function instead of
ours.
There are two aspects to every C++ function:

• Function definition. The definition of a function specifies the function’s return type and parameter
types, and it provides the code that determines the function’s behavior. In Listing 9.1 (customsquareroot.cpp)
the definition of the square_root function appears above the main function.
• Function invocation. A programmer uses a function via a function invocation. The main function
invokes both our square_root function and the sqrt function. Every function has exactly one
definition but may have many invocations.

A function definition consists of four parts:

• Name—every function in C++ has a name. The name is an identifier (see Section 3.3). As with
variable names, the name chosen for a function should accurately portray its intended purpose or
describe its functionality.
• Type—every function has a return type. If the function returns a value to its caller, its type corre-
sponds to the type of the value it returns. The special type void signifies that the function does not
return a value.
• Parameters—every function must specify the types of parameters that it accepts from callers. The
parameters appear in a parenthesized comma-separated list like in a function prototype (see Sec-
tion 8.1). Unlike function prototypes, however, parameters usually have names associated with each
type.
• Body—every function definition has a body enclosed by curly braces. The body contains the code to
be executed when the function is invoked.

©2013 Richard L. Halterman Draft date: January 27, 2013

 9.1. FUNCTION BASICS 190

Type of value the Name of Type of value the

function computes function function requires the
caller to provide

double square_root(double x)
{
double diff;
The name
// Compute a provisional square root the function
double root = 1.0; uses for the value
provided by the caller
do // Loop until the provisional root
Body of { // is close enough to the actual root
function root = (root + x/root) / 2.0;
// How bad is the approximation?
diff = root * root - x;
}
while (diff > 0.0001 || diff < -0.0001);
return root;
}

Figure 9.1: Function definition dissection

Figure 9.1 dissects a our square_root function definition.

The simplest function accepts no parameters and returns no value to the caller. Listing 9.2 (simplefunction.cpp)
is a variation of Listing 4.1 (adder.cpp) that contains such a simple function definition.

Listing 9.2: simplefunction.cpp

1 #include <iostream>
2
3 using namespace std;
4
5 // Definition of the prompt function
6 void prompt()
7 {
8 cout << "Please enter an integer value: ";
9 }
10
11 int main()
12 {
13 int value1, value2, sum;
14 cout << "This program adds together two integers." << endl;

©2013 Richard L. Halterman Draft date: January 27, 2013

 9.1. FUNCTION BASICS 191

15 prompt(); // Call the function

16 cin >> value1;
17 prompt(); // Call the function again
18 cin >> value2;
19 sum = value1 + value2;
20 cout << value1 << " + " << value2 << " = " << sum << endl;
21 }

The prompt function simply prints a message. The program runs as follows:

1. The program’s execution, like in all C++ programs, begins with the first executable statement in the
function named main. The first line in the main function simply declares some variables needed for
compiler housekeeping, so the next line actually begins the executable code.

2. The first executable statement prints the message of the program’s intent.

3. The next statement is a call of the prompt function. At this point the program’s execution transfers
to the body of the prompt function. The code within prompt is executed until the end of its body
or until a return statement is encountered. Since prompt contains no return statement, all of
prompt’s body (the one print statement) will be executed.

4. When prompt is finished, control is passed back to the point in main immediately after the call of
prompt.

5. The next action after prompt call reads the value of value1 from the keyboard.

6. A second call to prompt transfers control back to the code within the prompt function. It again
prints its message.

7. When the second call to prompt is finished, control passes back to main at the point of the second
input statement that assigns value2 from the keyboard.

8. The remaining two statements in main are executed, and then the program’s execution terminates.

As another simple example, consider Listing 9.3 (countto10.cpp).

Listing 9.3: countto10.cpp

1 #include <iostream>
2
3 using namespace std;
4
5 int main()
6 {
7 for ( int i = 1; i <= 10; i++ )
8 cout << i << endl;
9 }

which simply counts to ten:

©2013 Richard L. Halterman Draft date: January 27, 2013

 9.1. FUNCTION BASICS 192

1
2
3
4
5
6
7
8
9
10

If counting to ten in this way is something we want to do frequently within a program, we can write a
function as shown in Listing 9.4 (countto10func.cpp) and call it as many times as necessary.

Listing 9.4: countto10func.cpp

1 #include <iostream>
2
3 using namespace std;
4
5 // Count to ten and print each number on its own line
6 void count_to_10()
7 {
8 for ( int i = 1; i <= 10; i++ )
9 cout << i << endl;
10 }
11
12 int main()
13 {
14 cout << "Going to count to ten . . .";
15 count_to_10();
16 cout << "Going to count to ten again. . .";
17 count_to_10();
18 }

Our prompt and countto10 functions are a bit underwhelming. The prompt function could be
eliminated, and each call to prompt could be replaced with the statement in its body. The same could
be said for the countto10 function, although it is convenient to have the simple one-line statement
that hides the complexity of the loop. Using the prompt function does have one advantage, though. If
prompt is removed and the two calls to prompt are replaced with the print statement within prompt, we
have to make sure that the two messages printed are identical. If we simply call prompt, we know the two
messages printed will be identical, because only one possible message can be printed (the one in the body
of prompt).
We can alter the behavior of a function through a mechanism called parameter passing. If a function
is written to accept information from the caller, the caller must supply the information in order to use the
function. The caller communicates the information via one or more parameters as required by the function.
The countto10 function does us little good if we sometimes want to count up to a different number.
Listing 9.5 (countton.cpp) generalizes Listing 9.4 (countto10func.cpp) to count as high as the caller needs.

Listing 9.5: countton.cpp

©2013 Richard L. Halterman Draft date: January 27, 2013

 9.1. FUNCTION BASICS 193

1 #include <iostream>
2
3 using namespace std;
4
5 // Count to n and print each number on its own line
6 void count_to_n(int n)
7 {
8 for ( int i = 1; i <= n; i++ )
9 cout << i << endl;
10 }
11
12 int main()
13 {
14 cout << "Going to count to ten . . .";
15 count_to_n(10);
16 cout << "Going to count to five . . .";
17 count_to_n(5);
18 }

When the caller, in this case main, issues the call

count_to_n(10);

the argument 10 is assigned to n before the function’s statements begin executing.

A caller must pass exactly one integer parameter (or other type that is assignment-compatible with
integers) to countton during a call. An attempt to do otherwise one will result in a compiler error or
warning:
count_to_n(); // Error, missing parameter during the call
count_to_n(3, 5); // Error, too many parameters during the call
count_to_n(3.2); // Warning, possible loss of data (double to int)

We can enhance the prompt function’s capabilities as shown in Listing 9.6 (betterprompt.cpp)

Listing 9.6: betterprompt.cpp

1 #include <iostream>
2
3 using namespace std;
4
5 // Definition of the prompt function
6 int prompt()
7 {
8 int value;
9 cout << "Please enter an integer value: ";
10 cin >> value;
11 return value;
12 }
13
14 int main()
15 {
16 int value1, value2, sum;
17 cout << "This program adds together two integers." << endl;
18 value1 = prompt(); // Call the function

©2013 Richard L. Halterman Draft date: January 27, 2013

 9.1. FUNCTION BASICS 194

19 value2 = prompt(); // Call the function again

20 sum = value1 + value2;
21 cout << value1 << " + " << value2 << " = " << sum << endl;
22 }

In this version, prompt takes care of the input, so main does not have to use any input statements. The
assignment statement within main:
value1 = prompt();

implies prompt is no longer a void function; it must return a value that can be assigned to the variable
value1. Furthermore, the value that prompt returns must be assignment compatible with an int, be-
cause value1’s declared type is int. A quick look at the first line of prompt’s definition confirms our
assumption:
int prompt()

This indicates that prompt returns an int value.

Because prompt is declared to return an int value, it must contain a return statement. A return
statement specifies the exact value to return to the caller. When a return is encountered during a func-
tion’s execution, control immediately passes back to the caller. The value of the function call is the value
specified by the return statement, so the statement
value1 = prompt();

assigns to the variable value1 the value indicated when the return statement executed during the call.
Note that in Listing 9.6 (betterprompt.cpp), we declared a variable named value inside the prompt
function. This variable is local to the function, meaning we cannot use this particular variable outside of
prompt. It also means we are free to use that same name outside of the prompt function in a different
context, and that use will not interfere with the value variable within prompt. We say that value is a
local variable.
We can further enhance our prompt function. Currently prompt always prints the same message.
Using parameters, we can customize the message that prompt prints. Listing 9.7 (evenbetterprompt.cpp)
shows how parameters are used to provide a customized message within prompt.

Listing 9.7: evenbetterprompt.cpp

1 #include <iostream>
2
3 using namespace std;
4
5 // Definition of the prompt function
6 int prompt(int n)
7 {
8 int value;
9 cout << "Please enter integer #" << n << ": ";
10 cin >> value;
11 return value;
12 }
13
14 int main()
15 {
16 int value1, value2, sum;

©2013 Richard L. Halterman Draft date: January 27, 2013

 9.1. FUNCTION BASICS 195

17 cout << "This program adds together two integers." << endl;
18 value1 = prompt(1); // Call the function
19 value2 = prompt(2); // Call the function again
20 sum = value1 + value2;
21 cout << value1 << " + " << value2 << " = " << sum << endl;
22 }

In Listing 9.7 (evenbetterprompt.cpp), the parameter influences the message that it printed. The user is
now prompted to enter value #1 or value #2. The call
value1 = prompt(1);

passes the value 1 to the prompt function. Since prompt’s parameter is named n, the process works as
if the assignment statement
n = 1;

were executed as the first action within prompt.

In the first line of the function definition:
int prompt(int n)

n is called the formal parameter. A formal parameter is used like a variable within the function’s body, but
it is declared in the function’s parameter list; it is not declared in the function’s body. A formal parameter
is a parameter as used in the formal definition of the function.
At the point of the function call:
value1 = prompt(1);

the parameter (or argument) passed into the function, 1, is called the actual parameter. An actual parameter
is the parameter actually used during a call of the function. When a function is called, any actual parameters
are assigned to their corresponding formal parameters, and the function begin executing. Another way to say
it is that during a function call, the actual parameters are bound to their corresponding formal parameters.

The parameters used within a function definition are called formal parameters.
Formal parameters behave as local variables within the function’s body; as such,
the name of a formal parameter will not conflict with any local variable or for-
mal parameter names from other functions. This means as a function developer
you may choose a parameter name that best represents the parameter’s role in the
function.
If you are writing a function, you cannot predict the caller’s actual parameters.
You must be able to handle any value the caller sends. The compiler will ensure
that the types of the caller’s parameters are compatible with the declared types of
your formal parameters.
To remember the difference between formal and actual parameters, remember this:
• A formal parameter is a parameter declared and used in a function’s formal
definition.
• An actual parameter is a parameter supplied by the caller when the caller
actually uses (invokes or calls) the function.

©2013 Richard L. Halterman Draft date: January 27, 2013

9.2. USING FUNCTIONS 196

When the call

value1 = prompt(1);

is executed in main, and the statement

cout << "Please enter integer #" << n << ": ";

within the body of prompt is executed, n will have the value 1. Similarly, when the call
value2 = prompt(2);

is executed in main, and the statement

cout << "Please enter integer #" << n << ": ";

within the body of prompt is executed, n will have the value 2. In the case of
value1 = prompt(1);

n within prompt is bound to 1, and in the case of

value2 = prompt(2);

n within prompt is bound to 2.

A function’s definition requires that all formal parameters be declared in the paren-
theses following the function’s name. A caller does not provide actual parameter
type declarations when calling the function. Given the square_root function
defined in Listing 9.1 (customsquareroot.cpp), the following caller code fragment
is illegal:
double number = 25.0;
// Legal, pass the variable's value to the function
cout << square_root(number) << endl;
// Illegal, do not declare the parameter during the call
cout << square_root(double number) << endl;
The function definition is responsible for declaring the types of its parameters, not
the caller.

9.2 Using Functions

The general form of a function definition is

type name( parameterlist )

{
body
}
©2013 Richard L. Halterman Draft date: January 27, 2013
9.2. USING FUNCTIONS 197

• The type of the function indicates the type of value the function returns. Often a function will perform
a calculation and the result of the calculation must be communicated back to the place where the
function was invoked. The special type void indicates that the function does not return a value.

• The name of the function is an identifier (see Section 3.3). The function’s name should indicate the
purpose of the function.

• The parameterlist is a comma separated list of pairs of the form

type name

where type is a C++ type and name is an identifier representing a parameter. The caller of the func-
tion communicates information into the function via parameters. The parameters specified in the
parameter list of a function definition are called formal parameters. A parameter is also known as an
argument. The parameter list may be empty; an empty parameter list indicates that no information
may be passed into the function by the caller.

• The body is the sequence of statements, enclosed within curly braces, that define the actions that the
function is to perform. The statements may include variable declarations, and any variables declared
within the body are local to that function.
The body may contain only one statement, many statements, or no statements at all; regardless, the
curly braces always are required.

Observe that multiple pieces of information can be passed into a function via multiple parameters, but
only one piece of information can be passed out of the function via the return value. Recall the greatest
common divisor (also called greatest common factor) from elementary methematics. To determine the GCD
of 24 and 18 we list all of their common factors and select the largest one:
24: 1, 2, 3, 4, 6 , 8, 12, 24
18: 1, 2, 3, 6 , 9, 18
The following code defines a function that that computes the greatest common divisor of two integers.
It determines largest factor (divisor) common to its parameters:
int gcd(int num1, int num2)
{
// Determine the smaller of num1 and num2
int min = (num1 < num2) ? num1 : num2;
// 1 is definitely a common factor to all ints
int largestFactor = 1;
for ( int i = 1; i <= min; i++ )
if ( num1 % i == 0 && num2 % i == 0 )
largestFactor = i; // Found larger factor
return largestFactor;
}

This function is named gcd and expects two integer arguments. Its formal parameters are named num1 and
num2. It returns an integer result. Its body declares three local variables: min, largestFactor, and i
(i is local to the for statement). The last line in its body is a return statement. A return statement is
required for functions that return a value. A void function is not required to have a return statement. If a
void function does have a return statement, it must simply consist of return followed by a semicolon
(in other words, it cannot return a value, like gcd’s return statement does). A void function that does
not contain a return statement simply returns at the end of its body.

©2013 Richard L. Halterman Draft date: January 27, 2013

9.2. USING FUNCTIONS 198

The greatest common divisor function can be used to reduce fractions to lowest terms; for example,
18
consider the fraction . The greatest common divisor of 18 and 24 is 6, and so we divide the numerator
24
18 ÷ 6 3
and the denominator of the fraction by 6: = .
24 ÷ 6 4
The GCD function has applications in other areas besides reducing fractions to lowest terms. Consider
the problem of dividing a piece of plywood 24 inches long by 18 inches wide into square pieces of max-
imum size without wasting any material. Since the GCF(24, 18) = 6, we can cut the plywood into twelve
6 inch × 6 inch square pieces as shown in Figure 9.2.

18 inches

24 inches

Figure 9.2: Cutting plywood

If we cut the plywood into squares of any other size without wasting the any of the material, the squares
would have to be smaller than 6 inches × 6 inches; for example, we could make fourty-eight 3 inch × 3 inch
squares as shown in pieces as shown in Figure 9.3. If we cut squares larger than 6 inches × 6 inches, not

3
3

18 inches

24 inches

Figure 9.3: Squares too small

all the plywood can be used to make the squares. Figure 9.4. shows how some larger squares would fare.
The GCD function also is useful in cryptography.

©2013 Richard L. Halterman Draft date: January 27, 2013

9.2. USING FUNCTIONS 199

9 in.

9 in.

Waste
18 inches

24 inches

8 in.

8 in.

18 inches

Waste

24 inches

Figure 9.4: Squares too large

Recall from Section 6.5 that local variables have meaning only within their scope. This means that
when you write a function you can name a local variable without fear that its name may be used already
in another part of the program. Two different functions can use local variables named x, and these are two
different variables that have no influence on each other. Anything local to a function definition is hidden to
all code outside that function definition.
Since a formal parameter is a local variable, you can reuse the names of formal parameters in different
functions without a problem.
Another advantage of local variables is that they occupy space in the computer’s memory only when
the function is executing. Space is allocated for local variables and parameters when the function begins
executing. When the function is finished and control returns to the caller, the variables and parameters go
out of scope, and the memory they held is freed up for other purposes within the running program. This
process of local variable allocation and deallocation happens each time a caller invokes the function. More
information about how C++ handles memory management during a program’s execution can be found in
Section 11.11.
Once a function has been defined it can be used. A programmer-defined function is invoked in exactly
the same way as a standard library function like sqrt (Section 8.2) or rand (Section 8.6). If the function

©2013 Richard L. Halterman Draft date: January 27, 2013

9.2. USING FUNCTIONS 200

returns a type (that is, it is not declared void), then its invocation can be used anywhere an expression
of that type can be used. The parameters used for the function call are known as actual parameters. The
function gcd can be called as part of an assignment statement:
int factor = gcd(val, 24);

This call uses the variable val as its first actual parameter and the literal value 24 as its second actual
parameter. Variables, expressions, and literals can be freely used as actual parameters. The function then
computes and returns its result. This result is assigned to the variable factor.
How does the function call and parameter mechanism work? It’s actually quite simple. The actual
parameters, in order, are assigned (bound) to each of the formal parameters in the function definition, then
control is passed to the body of the function. When the function’s body is finished executing, control passes
back to the point in the program where the function was called. The value returned by the function, if any,
replaces the function call expression. In the statement
int factor = gcd(val, 24);

an integer value is assigned to factor. The expression on the right is a function call, so the function is
invoked to determine what to assign. The value of the variable val is assigned to the formal parameter
num1, and the literal value 24 is assigned to to the formal parameter num2. The body of the gcd function
is then executed. When the return statement in the body is encountered, program execution returns back
to where the function was called. The argument of the return statement becomes the value that is assigned
to factor. This process of copying actual parameters to formal parameters works exactly like assignment,
so widening and narrowing (see Section 4.2 is performed automatically as needed, For example, if val is
declared to a char, its value would automatically be copied to a temporary location and converted to an
int. This temporary value would then be bound to the formal parameter num1. Note that gcd could be
called from many different places within the same program, and, since different parameter values could be
passed at each of these different invocations, gcd could compute a different result at each invocation.
Other invocation examples include:

• cout << gcd(36, 24);

This example simply prints the result of the invocation. The value 36 is bound to num1
and 24 is bound to num2 for the purpose of the function call. The value 12 will be printed,
since 12 is the greatest common divisor of 36 and 24..

• x = gcd(x - 2, 24);

The execution of this statement would evaluate x - 2 and bind its value to num1. num2
would be assigned 24. The result of the call is then assigned to x. Since the right side of
the assignment statement is evaluated before being assigned to the left side, the original
value of x is used when calculating x ∼ - ∼ 2, and the function return value then updates
x.

• x = gcd(x - 2, gcd(10, 8));

This example shows two invocations in one statement. Since the function returns an inte-
ger value its result can itself be used as an actual parameter in a function call. Passing the
result of one function call as an actual parameter to another function call is called function
composition.

©2013 Richard L. Halterman Draft date: January 27, 2013

 9.3. CALL BY VALUE 201

It is illegal (compiler syntax error) if a function call does not agree with the function’s definition. Pos-
sible problems include:

• Number of actual parameters do not agree with the number of formal parameters. The number
of parameters must agree exactly. For example, the statement
int factor = gcd(24); // Error: too few parameters

is illegal given the above definition of gcd, since only one actual parameter is provided when two
are required.
• Passing an actual parameter that is not assignment compatible with the formal parameter. For
example, passing the cout object when an int has been defined, as in
int factor = gcd(36, cout); // Error: second parameter is wrong type

The compiler will detect that cout is not a valid int and report an error.
• Using the result in a context where an expression of that type is not allowed. For example, a
function that returns void cannot be used where an int is expected:
cout << srand(2); // Error: srand does not return anything

The compiler will disallow this code.

9.3 Call by Value

The default parameter passing mechanism in C++ is classified as call by value, also known as pass by value.
This means the value of the actual parameter is copied to the formal parameter for the purpose of executing
the function’s code. When the function executes it cannot affect the value of the actual parameter because it
is working on a copy of actual parameter. Listing 9.8 (callbyvalue.cpp) illustrates the consequences of call
by value.

Listing 9.8: callbyvalue.cpp

1 #include <iostream>
2
3 using namespace std;
4
5 /*
6 * increment(x)
7 * Illustrates call by value protocol.
8 */
9 void increment(int x)
10 {
11 cout << "Beginning execution of increment, x = "
12 << x << endl;
13 x++; // Increment x
14 cout << "Ending execution of increment, x = "
15 << x << endl;
16 }
17
18 int main()
19 {

©2013 Richard L. Halterman Draft date: January 27, 2013

 9.4. FUNCTION EXAMPLES 202

20 int x = 5;
21 cout << "Before increment, x = " << x << endl;
22 increment(x);
23 cout << "After increment, x = " << x << endl;
24 }

For additional drama we chose to name the actual parameter the same as the formal parameter, but, of
course, the names do not matter because they represent completely different memory locations.
Listing 9.8 (callbyvalue.cpp) produces

Before increment, x = 5
Beginning execution of increment, x = 5
Ending execution of increment, x = 6
After increment, x = 5

The memory for the variable x in main is unaffected since increment works on a copy of the actual
parameter.
C++ supports another way of passing parameters called call by reference. Call by reference is introduced
in Section 10.7.

9.4 Function Examples

This section contains a number of examples of how functions can be used to organize a program’s code.

9.4.1 Better Organized Prime Generator

Listing 9.9 (primefunc.cpp) is a simple enhancement of Listing 8.5 (moreefficientprimes.cpp). It uses the
square root optimization and adds a separate is_prime function.

Listing 9.9: primefunc.cpp

1 #include <iostream>
2 #include <cmath>
3
4 using namespace std;
5
6 /*
7 * is_prime(n)
8 * Determines the primality of a given value
9 * n an integer to test for primality
10 * Returns true if n is prime; otherwise, returns false
11 */
12 bool is_prime(int n)
13 {
14 bool result = true; // Provisionally, n is prime
15 double r = n, root = sqrt(r);
16 // Try all possible factors from 2 to the square

©2013 Richard L. Halterman Draft date: January 27, 2013

 9.4. FUNCTION EXAMPLES 203

17 // root of n
18 for ( int trial_factor = 2;
19 result && trial_factor <= root; trial_factor++ )
20 result = (n % trial_factor != 0 );
21 return result;
22 }
23
24 /*
25 * main
26 * Tests for primality each integer from 2
27 * up to a value provided by the user.
28 * If an integer is prime, it prints it;
29 * otherwise, the number is not printed.
30 */
31 int main()
32 {
33 int max_value;
34 cout << "Display primes up to what value? ";
35 cin >> max_value;
36 for ( int value = 2; value <= max_value; value++ )
37 if ( is_prime(value) ) // See if value is prime
38 cout << value << " "; // Display the prime number
39 cout << endl; // Move cursor down to next line
40 }

Listing 9.9 (primefunc.cpp) illustrates several important points about well-organized programs:

• The complete work of the program is no longer limited to the main function. The effort to test for
primality is delegated to a separate function. main is focused on a simpler task: generating all the
numbers to be considered and using another function (is_prime) to do the hard work of determin-
ing if a given number is prime. main is now simpler and more logically coherent. A function is
coherent when it is focused on a single task. Coherence is a desirable property of functions. If a
function becomes too complex by trying to do too many different things, it can be more difficult to
write correctly and debug when problems are detected. A complex function should be decomposed
into several, smaller, more coherent functions. The original function would then call these new sim-
pler functions to accomplish its task. Here, main is not concerned about how to determine if a given
number is prime; main simply delegates the work to is_prime and makes use of the is_prime
function’s findings.

• Each function is preceded by a thorough comment that describes the nature of the function. It explains
the meaning of each parameter, and it indicates what the function should return. The comment for
main is usually not as thorough as for other functions, because it usually has no parameters, and it
always returns a code to the operating system upon the program’s termination.

• While the exterior comment indicates what the function is to do, comments within each function
explain in more detail how the function accomplishes its task.

The call to is_prime return true or false depending on the value passed to it. The means a condition
like
if ( is_prime(value) == true ) . . .

can be expressed more compactly as

©2013 Richard L. Halterman Draft date: January 27, 2013

 9.4. FUNCTION EXAMPLES 204

if ( is_prime(value) ) . . .

because if is_prime(value) is true, true ∼ == ∼ true is true, and if is_prime(value) is false,

false ∼ == ∼ true is false. The expression is_prime(value) suffices.
Just as it is better for a loop to have exactly one entry point and exactly one exit point, preferably a
function will have a single return statement. Simple functions with a small number of returns are
generally tolerable, however. Consider the following version of is_prime:
bool is_prime(int n)
{
for ( int trialFactor = 2;
trialFactor <= sqrt(static_cast<double>(n));
trialFactor++ )
if ( n % trialFactor == 0 ) // Is trialFactor a factor?
return false; // Yes, return right away
return true; // Tried them all, must be prime
}

This version uses two return statements, but eliminates the need for a local variable (result). No
break statement is necessary, because a return statement exits the function immediately. The two
return statements are close enough textually in source code that the logic is fairly transparent.

9.4.2 Command Interpreter

Some functions are useful even if they accept no information from the caller and return no result. List-
ing 9.10 (calculator.cpp) uses such a function.

Listing 9.10: calculator.cpp

1 #include <iostream>
2 #include <cmath>
3
4 using namespace std;
5
6 /*
7 * help_screen
8 * Displays information about how the program works
9 * Accepts no parameters
10 * Returns nothing
11 */
12 void help_screen()
13 {
14 cout << "Add: Adds two numbers" << endl;
15 cout << " Example: a 2.5 8.0" << endl;
16 cout << "Subtract: Subtracts two numbers" << endl;
17 cout << " Example: s 10.5 8.0" << endl;
18 cout << "Print: Displays the result of the latest operation"
19 << endl;
20 cout << " Example: p" << endl;
21 cout << "Help: Displays this help screen" << endl;
22 cout << " Example: h" << endl;
23 cout << "Quit: Exits the program" << endl;

©2013 Richard L. Halterman Draft date: January 27, 2013

 9.4. FUNCTION EXAMPLES 205

24 cout << " Example: q" << endl;

25 }
26
27 /*
28 * menu
29 * Display a menu
30 * Accepts no parameters
31 * Returns the character entered by the user.
32 */
33 char menu()
34 {
35 // Display a menu
36 cout << "=== A)dd S)ubtract P)rint H)elp Q)uit ===" << endl;
37 // Return the char entered by user
38 char ch;
39 cin >> ch;
40 return ch;
41 }
42
43 /*
44 * main
45 * Runs a command loop that allows users to
46 * perform simple arithmetic.
47 */
48 int main()
49 {
50 double result = 0.0, arg1, arg2;
51 bool done = false; // Initially not done
52 do
53 {
54 switch ( menu() )
55 {
56 case 'A': // Addition
57 case 'a':
58 cin >> arg1 >> arg2;
59 result = arg1 + arg2;
60 cout << result << endl;
61 break;
62 case 'S': // Subtraction
63 case 's':
64 cin >> arg1 >> arg2;
65 result = arg1 - arg2;
66 // Fall through, so it prints the result
67 case 'P': // Print result
68 case 'p':
69 cout << result << endl;
70 break;
71 case 'H': // Display help screen
72 case 'h':
73 help_screen();
74 break;
75 case 'Q': // Quit the program
76 case 'q':
77 done = true;
78 break;

©2013 Richard L. Halterman Draft date: January 27, 2013

 9.4. FUNCTION EXAMPLES 206

79 }
80 }
81 while ( !done );
82 }

The help_screen function needs no information from main, nor does it return a result. It behaves
exactly the same way each time it is called. The menu function returns the character entered by the user.

9.4.3 Restricted Input

Listing 7.3 (betterinputonly.cpp) forces the user to enter a value within a specified range. We now can easily
adapt that concept to a function. Listing 9.11 (betterinputfunc.cpp) uses a function named get_int_range
that does not return until the user supplies a proper value.

Listing 9.11: betterinputfunc.cpp

1 #include <iostream>
2
3 using namespace std;
4
5 /*
6 * get_int_range(first, last)
7 * Forces the user to enter an integer within a
8 * specified range
9 * first is either a minimum or maximum acceptable value
10 * last is the corresponding other end of the range,
11 * either a maximum or minimum * value
12 * Returns an acceptable value from the user
13 */
14 int get_int_range(int first, int last)
15 {
16 // If the larger number is provided first,
17 // switch the parameters
18 if ( first > last )
19 {
20 int temp = first;
21 first = last;
22 last = temp;
23 }
24 // Insist on values in the range first...last
25 cout << "Please enter a value in the range "
26 << first << "..." << last << ": ";
27 int in_value; // User input value
28 bool bad_entry;
29 do
30 {
31 cin >> in_value;
32 bad_entry = (in_value < first || in_value > last);
33 if ( bad_entry )
34 {
35 cout << in_value << " is not in the range "
36 << first << "..." << last << endl;
37 cout << "Please try again: ";
38 }

©2013 Richard L. Halterman Draft date: January 27, 2013

 9.4. FUNCTION EXAMPLES 207

39 }
40 while ( bad_entry );
41 // in_value at this point is guaranteed to be within range
42 return in_value;
43 }
44
45 /*
46 * main
47 * Tests the get_int_range function
48 */
49 int main()
50 {
51 cout << get_int_range(10, 20) << endl;
52 cout << get_int_range(20, 10) << endl;
53 cout << get_int_range(5, 5) << endl;
54 cout << get_int_range(-100, 100) << endl;
55 }

Listing 9.11 (betterinputfunc.cpp) forces the user to enter a value within a specified range. This func-
tionality could be useful in many programs.
In Listing 9.11 (betterinputfunc.cpp)

• The high and low values are specified by parameters. This makes the function more flexible since
it could be used elsewhere in the program with a completely different range specified and still work
correctly.
• The function is supposed to be called with the lower number passed as the first parameter and the
higher number passed as the second parameter. The function will also accept the parameters out of
order and automatically swap them to work as expected; thus,
num = get_int_range(20, 50);

will work exactly like

num = get_int_range(50, 20);

• The Boolean variable bad_entry is used to avoid evaluating the Boolean expression twice (once
to see if the bad entry message should be printed and again to see if the loop should continue).

9.4.4 Better Die Rolling Simulator

Listing 9.12: betterdie.cpp

1 #include <iostream>
2 #include <cstdlib>
3 #include <ctime>
4
5 using namespace std;
6
7 /*
8 * initialize_die
9 * Initializes the randomness of the die
10 * /

©2013 Richard L. Halterman Draft date: January 27, 2013

 9.4. FUNCTION EXAMPLES 208

11 void initialize_die()
12 {
13 // Set the random seed value
14 srand(static_cast<unsigned>(time(0)));
15 }
16
17 /*
18 * show_die(spots)
19 * Draws a picture of a die with number of spots
20 * indicated
21 * spots is the number of spots on the top face
22 */
23 void show_die(int spots)
24 {
25 cout << "+-------+" << endl;
26 switch ( spots )
27 {
28 case 1:
29 cout << "| |" << endl;
30 cout << "| * |" << endl;
31 cout << "| |" << endl;
32 break;
33 case 2:
34 cout << "| * |" << endl;
35 cout << "| |" << endl;
36 cout << "| * |" << endl;
37 break;
38 case 3:
39 cout << "| * |" << endl;
40 cout << "| * |" << endl;
41 cout << "| * |" << endl;
42 break;
43 case 4:
44 cout << "| * * |" << endl;
45 cout << "| |" << endl;
46 cout << "| * * |" << endl;
47 break;
48 case 5:
49 cout << "| * * |" << endl;
50 cout << "| * |" << endl;
51 cout << "| * * |" << endl;
52 break;
53 case 6:
54 cout << "| * * * |" << endl;
55 cout << "| |" << endl;
56 cout << "| * * * |" << endl;
57 break;
58 default:
59 cout << " *** Error: illegal die value ***" << endl;
60 break;
61 }
62 cout << "+-------+" << endl;
63 }
64
65 /*

©2013 Richard L. Halterman Draft date: January 27, 2013

 9.4. FUNCTION EXAMPLES 209

66 * roll
67 * Returns a pseudorandom number in the range 1...6
68 */
69 int roll()
70 {
71 return rand() % 6 + 1;
72 }
73
74 /*
75 * main
76 * Simulates the roll of a die three times
77 */
78 int main()
79 {
80
81 // Initialize the die
82 initialize_die();
83
84 // Roll the die three times
85 for ( int i = 0; i < 3; i++ )
86 show_die(roll());
87 }

In Listing 9.12 (betterdie.cpp), main is no longer concerned with the details of pseudorandom number
generation, nor is it responsible for drawing the die. These important components of the program are now
in functions, so their details can be perfected independently from main.
Note how the result of the call to roll is passed directly as an argument to show_die.

9.4.5 Tree Drawing Function

Listing 9.13: treefunc.cpp

1 #include <iostream>
2
3 using namespace std;
4
5 /*
6 * tree(height)
7 * Draws a tree of a given height
8 * height is the height of the displayed tree
9 */
10 void tree(int height)
11 {
12 int row = 0; // First row, from the top, to draw
13 while ( row < height ) // Draw one row for every unit of height
14 {
15 // Print leading spaces
16 int count = 0;
17 while ( count < height - row )
18 {
19 cout << " ";
20 count++;
21 }

©2013 Richard L. Halterman Draft date: January 27, 2013

 9.4. FUNCTION EXAMPLES 210

22 // Print out stars, twice the current row plus one:

23 // 1. number of stars on left side of tree
24 // = current row value
25 // 2. exactly one star in the center of tree
26 // 3. number of stars on right side of tree
27 // = current row value
28 count = 0;
29 while ( count < 2*row + 1 )
30 {
31 cout << "*";
32 count++;
33 }
34 // Move cursor down to next line
35 cout << endl;
36 // Change to the next row
37 row++;
38 }
39
40 }
41
42 /*
43 * main
44 * Allows users to draw trees of various heights
45 */
46 int main()
47 {
48 int height; // Height of tree
49 cout << "Enter height of tree: ";
50 cin >> height; // Get height from user
51 tree(height);
52 }

Observe that the name height is being used as a local variable in main and as a formal parameter
name in tree. There is no conflict here, and the two heights represent two different locations in memory.
Furthermore, the fact that the statement
tree(height);

uses main’s height as an actual parameter and height happens to be the name as the formal parameter
is simply a coincidence. During the call, the value of main’s height variable is copied into to the formal
parameter in tree also named height. The compiler can keep track of which height is which based
on where each is declared.

9.4.6 Floating-point Equality

Recall from Listing 4.2 (imprecise.cpp) that floating-point numbers are not mathematical real numbers;
a floating-point number is finite, and is represented internally as a quantity with a binary mantissa and
exponent. Just as 1/3 cannot be represented finitely in the decimal (base 10) number system, 1/10 cannot be
represented exactly in the binary (base 2) number system with a fixed number of digits. Often, no problems
arise from this imprecision, and in fact many software applications have been written using floating-point
numbers that must perform precise calculations, such as directing a spacecraft to a distant planet. In such
cases even small errors can result in complete failures. Floating-point numbers can and are used safely and
effectively, but not without appropriate care.

©2013 Richard L. Halterman Draft date: January 27, 2013

 9.4. FUNCTION EXAMPLES 211

To build our confidence with floating-point numbers, consider Listing 9.14 (simplefloataddition.cpp),
which adds two double-precision floating-point numbers are checks for a given value.

Listing 9.14: simplefloataddition.cpp

1 #include <iostream>
2
3 using namespace std;
4
5 int main()
6 {
7 double x = 0.9;
8 x += 0.1;
9 if ( x == 1.0 )
10 cout << "OK" << endl;
11 else
12 cout << "NOT OK" << endl;
13 }

All seems well judging by Listing 9.14 (simplefloataddition.cpp). Next, consider Listing 9.15 (badfloatcheck.cpp)
which attempts to control a loop with a double-precision floating-point number.

Listing 9.15: badfloatcheck.cpp

1 #include <iostream>
2
3 using namespace std;
4
5 int main()
6 {
7 // Count to ten by tenths
8 for ( double i = 0.0; i != 1.0; i += 0.1 )
9 cout << "i = " << i << endl;
10 }

When compiled and executed, Listing 9.15 (badfloatcheck.cpp) begins as expected, but it does not end
as expected:

©2013 Richard L. Halterman Draft date: January 27, 2013

 9.4. FUNCTION EXAMPLES 212

i = 0
i = 0.1
i = 0.2
i = 0.3
i = 0.4
i = 0.5
i = 0.6
i = 0.7
i = 0.8
i = 0.9
i = 1
i = 1.1
i = 1.2
i = 1.3
i = 1.4
i = 1.5
i = 1.6
i = 1.7
i = 1.8
i = 1.9
i = 2
i = 2.1

We expect it stop when the loop variable i equals 1, but the program continues executing until the user
types Ctrl-C. We are adding 0.1, just as in Listing 9.14 (simplefloataddition.cpp), but now there is a prob-
lem. Since 0.1 cannot be represented exactly within the constraints of the double-precision floating-point
representation, the repeated addition of 0.1 leads to round off errors that accumulate over time. Whereas
0.1 + 0.9 rounded off may equal 1, 0.1 added to itself 10 times may be 1.000001 or 0.999999, neither of
which is exactly 1.
Listing 9.15 (badfloatcheck.cpp) demonstrates that the == and != operators are of questionable worth
when comparing floating-point values. The better approach is to check to see if two floating-point values
are close enough, which means they differ by only a very small amount. When comparing two floating-
point numbers x and y, we essentially must determine if the absolute value of their difference is small; for
example, |x − y| < 0.00001. The C abs function was introduced in Section 8.2, and we can incorporate it
into an equals function, as shown in Listing 9.16 (floatequals.cpp).

Listing 9.16: floatequals.cpp

1 #include <iostream>
2 #include <cmath>
3
4 using namespace std;
5
6 /*
7 * equals(a, b, tolerance)
8 * Returns true if a = b or |a - b| < tolerance.
9 * If a and b differ by only a small amount
10 * (specified by tolerance), a and b are considered
11 * "equal." Useful to account for floating-point
12 * round-off error.

©2013 Richard L. Halterman Draft date: January 27, 2013

 9.4. FUNCTION EXAMPLES 213

13 * The == operator is checked first since some special

14 * floating-point values such as HUGE_VAL require an
15 * exact equality check.
16 */
17 bool equals(double a, double b, double tolerance)
18 {
19 return a == b || fabs(a - b) < tolerance;
20 }
21
22 int main()
23 {
24 for ( double i = 0.0; !equals(i, 1.0, 0.0001); i += 0.1 )
25 cout << "i = " << i << endl;
26 }

The third parameter, named tolerance, specifies how close the first two parameters must be in order
to be considered equal. The == operator must be used for some special floating-point values such as
HUGE_VAL, so the function checks for == equality as well. Since C++ uses short-circuit evaluation for
Boolean expressions involving logical OR (see Section 5.2), if the == operator indicates equality, the more
elaborate check is not performed.
You should use a function like equals when comparing two floating-point values for equality.

9.4.7 Multiplication Table with Functions

When we last visited our multiplication table program in Listing 7.5 (bettertimestable.cpp), we used nested
for loops to display a user-specified size. We can decompose the code into functions as shown in List-
ing 9.17 (timestablefunction.cpp). Our goal is to have a collection of functions that each are very simple.
We also want the program’s overall structure to be logically organized.

Listing 9.17: timestablefunction.cpp

1 #include <iostream>
2 #include <iomanip>
3
4 using namespace std;
5
6 // Print the column labels for an n x n multiplication table.
7 void col_numbers(int n)
8 {
9 cout << " ";
10 for ( int column = 1; column <= n; column++ )
11 cout << setw(4) << column; // Print heading for this column.
12 cout << endl;
13 }
14
15 // Print the table's horizontal line at the top of the table
16 // beneath the column labels.
17 void col_line(int n)
18 {
19 cout << " +";
20 for ( int column = 1; column <= n; column++ )
21 cout << "----"; // Print separator for this row.
22 cout << endl;

©2013 Richard L. Halterman Draft date: January 27, 2013

 9.4. FUNCTION EXAMPLES 214

23 }
24
25 // Print the title of each column across the top of the table
26 // including the line separator.
27 void col_header(int n)
28 {
29 // Print column titles
30 col_numbers(n);
31
32 // Print line separator
33 col_line(n);
34 }
35
36 // Print the title that appears before each row of the table's
37 // body.
38 void row_header(int n)
39 {
40 cout << setw(4) << n << " |"; // Print row label.
41 }
42
43 // Print the line of text for row n
44 // This includes the row number and the
45 // contents of each row.
46 void print_row(int row, int columns)
47 {
48 row_header(row);
49 for ( int col = 1; col <= columns; col++ )
50 cout << setw(4) << row*col; // Display product
51 cout << endl; // Move cursor to next row
52 }
53
54 // Print the body of the n x n multiplication table
55 void print_contents(int n)
56 {
57 for ( int current_row = 1; current_row <= n; current_row++ )
58 print_row(current_row, n);
59 }
60
61 // Print a multiplication table of size n x n.
62 void timestable(int n)
63 {
64 // First, print column heading
65 col_header(n);
66
67 // Print table contents
68 print_contents(n);
69 }
70
71 // Forces the user to enter an integer within a
72 // specified range first is either a minimum or maximum
73 // acceptable value last is the corresponding other end
74 // of the range, either a maximum or minimum value
75 // Returns an acceptable value from the user
76 int get_int_range(int first, int last)
77 {

©2013 Richard L. Halterman Draft date: January 27, 2013

 9.4. FUNCTION EXAMPLES 215

78 // If the larger number is provided first,

79 // switch the parameters
80 if ( first > last )
81 {
82 int temp = first;
83 first = last;
84 last = temp;
85 }
86 // Insist on values in the range first...last
87 cout << "Please enter a value in the range "
88 << first << "..." << last << ": ";
89 int in_value; // User input value
90 bool bad_entry;
91 do
92 {
93 cin >> in_value;
94 bad_entry = (in_value < first || in_value > last);
95 if ( bad_entry )
96 {
97 cout << in_value << " is not in the range "
98 << first << "..." << last << endl;
99 cout << "Please try again: ";
100 }
101 }
102 while ( bad_entry );
103 // in_value at this point is guaranteed to be within range
104 return in_value;
105 }
106
107 int main()
108 {
109 // Get table size from user; allow values in the
110 // range 1...18.
111 int size = get_int_range(1, 18);
112
113 // Print a size x size multiplication table
114 timestable(size);
115 }

In Listing 9.17 (timestablefunction.cpp), each function plays a very specific role; for example, row_header
prints the label for a particular row and then prints the vertical bar that separates the row label from the body
of the table. We reuse the get_int_range function from Listing 9.11 (betterinputfunc.cpp). Notice that
there no longer appear to be any nested loops in the program. The nested loop is not really gone, it now is
hidden by the program’s functional composition. Observe that the print_contents function contains a
loop. Each time through the loop it calls print_row, but print_row itself contains a loop. The nested
iteration, therefore, is still present.
Realistically, the functional decomposition within Listing 9.17 (timestablefunction.cpp) is extreme. The
relative simplicity of the program does not not really justify eight separate functions each with such a
narrow focus; however, more complex software systems are decomposed in this very manner. Not only
does Listing 9.17 (timestablefunction.cpp) give us insight into how we can take a complicated problem and
break it down into simpler, more manageable pieces, we can use the program to better understand how the
function invocation process works. To see how, consider the situation where a user wishes to print a 3 × 3
multiplication table using Listing 9.17 (timestablefunction.cpp). Figure
In Figure
9.5. 9.5 an arrow pointing down

©2013 Richard L. Halterman Draft date: January 27, 2013

9.5. COMMENTING FUNCTIONS 216

timestablefunc.cpp Function Call Sequence

Program Execution Timeline
main
1,18 3 3

get_int_range timestable

3 3

col_header print_contents
3 3 1,3 2,3 3,3

col_numbers print_line print_row print_row print_row

1 2 3

row_header row_header row_header

Figure 9.5: Traces the activation of the various functions in Listing 9.17 (timestablefunction.cpp) when the
user enters 3 for a 3 × 3 multiplication table. Time flows from left to right. The horizontal bars show the
lifetimes of the various functions involved.

corresponds to a function call. The labels on the down arrows represent parameters passed during the call.
The up arrows indicate the return from a function. The label on an up arrow is the function’s return value.
Functions of type void have no labels on their return arrows.
Each horizontal bar in Figure 9.5 represents the duration of the program’s execution that the function
is active. The main function always is the first function executed in a C++ program, and in Listing 9.17
(timestablefunction.cpp) main immediately calls the get_int_range function. During the lifetime of
get_int_range’s execution notice that main is still active. This means any local variables it may
maintain are still stored in memory and have not lost their values. This means that main’s value variable
exists for the effective life of the program. This is not true for the variables used in get_int_range:
temp, in_value, bad_entry, and the parameters first and last. These variables maintained by
get_int_range appear automatically when get_int_range begins executing (the left end of its
bar) and their space is released when get_int_range is finished (the right end of its bar)1 . During
the printing of a 3 × 3 table the program calls print_row three times, and for each call the function’s
parameters row and column and local variable col come to life and then disappear when the function
returns.

9.5 Commenting Functions

It is good practice to comment a function’s definition with information that aids programmers who may
need to use or extend the function. The essential information includes:
1 Technically, the run-time environment does not allocate the space for a local variable until after its point of declaration. For

variables declared within blocks, like temp with the if body, the variable is discarded at the end of the block’s execution.

©2013 Richard L. Halterman Draft date: January 27, 2013

9.5. COMMENTING FUNCTIONS 217

• The purpose of the function. The function’s purpose is not always evident merely from its name.
This is especially true for functions that perform complex tasks. A few sentences explaining what the
function does can be helpful.
• The role of each parameter. The parameter names and types are obvious from the definition, but the
purpose of a parameter may not be apparent merely from its name. It is helpful indicate the purpose
of each parameter, if any.
• The nature of the return value. While the function may do a number of interesting things as
indicated in the function’s purpose, what exactly does it return to the caller? It is helpful to clarify
exactly what value the function produces, if any.

Other information is often required in a commercial environment:

• Author of the function. Specify exactly who wrote the function. An email address can be included.
If questions about the function arise, this contact information can be invaluable.
• Date that the function’s implementation was last modified. An additional comment can be added
each time the function is updated. Each update should specify the exact changes that were made and
the person responsible for the update.
• References. If the code was adapted from another source, list the source. The reference may consist
of a Web URL.

The following fragment shows the beginning of a well-commented function definition:

/*
* distance(x1, y1, x2, y2)
* Computes the distance between two geometric points
* x1 is the x coordinate of the first point
* y1 is the y coordinate of the first point
* x2 is the x coordinate of the second point
* y2 is the y coordinate of the second point
* Returns the distance between (x1,y1) and (x2,y2)
* Author: Joe Algori ([email protected])
* Last modified: 2010-01-06
* Adapted from a formula published at
* http://en.wikipedia.org/wiki/Distance
*/
double distance(double x1, double y1, double x2, double y2)
{
...

From the information provided

• callers know what the function can do for them,

• callers know how to use the function,
• subsequent programmers can contact the original author if questions arise about its use or implemen-
tation,
• subsequent programmers can check the Wikipedia reference if questions arise about its implementa-
tion, and
• the quality of the algorithm may evaluated based upon the quality of its source of inspiration (Wikipedia).

©2013 Richard L. Halterman Draft date: January 27, 2013

 9.6. CUSTOM FUNCTIONS VS. STANDARD FUNCTIONS 218

9.6 Custom Functions vs. Standard Functions

Armed with our knowledge of function definitions, we can ewrite Listing 8.1 (computesquareroot.cpp) so
the program uses a custom square root function. Listing 9.18 (squarerootfunction.cpp) shows one possibil-
ity.

Listing 9.18: squarerootfunction.cpp

1 #include <iostream>
2
3 using namespace std;
4
5 double square_root(double x)
6 {
7 double diff;
8 // Compute a provisional square root
9 double root = 1.0;
10
11 do // Loop until the provisional root
12 { // is close enough to the actual root
13 root = (root + x/root) / 2.0;
14 cout << "root is " << root << endl;
15 // How bad is the approximation?
16 diff = root * root - x;
17 }
18 while (diff > 0.0001 || diff < -0.0001);
19 return root;
20 }
21
22 int main()
23 {
24 double input;
25
26 // Get value from the user
27 cout << "Enter number: ";
28 cin >> input;
29 // Report square root
30 cout << "Square root of " << input << " = "
31 << square_root(input) << endl;
32 }

Is Listing 9.18 (squarerootfunction.cpp) better than Listing 8.2 (standardsquareroot.cpp) which uses the
standard sqrt function from the cmath library? Generally speaking, if you have the choice of using a
standard library function or writing your own custom function that provides the same functionality, choose
to use the standard library routine. The advantages of using the standard library routine include:

• Your effort to produce the custom code is eliminated entirely; you can devote more effort to other
parts of the application’s development.
• If you write your own custom code, you must thoroughly test it to ensure its correctness; standard
library code, while not immune to bugs, generally has been subjected to a complete test suite. Library
code is used by many developers, and thus any lurking errors are usually exposed early; your code is
exercised only by the programs you write, and errors may not become apparent immediately. If your
programs are not used by a wide audience, bugs may lie dormant for a long time. Standard library

©2013 Richard L. Halterman Draft date: January 27, 2013

 9.6. CUSTOM FUNCTIONS VS. STANDARD FUNCTIONS 219

routines are well known and trusted; custom code, due to its limited exposure, is suspect until it gains
wider exposure and adoption.
• Standard routines are typically tuned to be very efficient; it takes a great deal of effort to make custom
code efficient.
• Standard routines are well-documented; extra work is required to document custom code, and writing
good documentation is hard work.

Listing 9.19 (squarerootcomparison.cpp) tests our custom square root function over a range of 1,000,000,000
floating-point values.

Listing 9.19: squarerootcomparison.cpp

1 #include <iostream>
2 #include <cmath>
3
4 using namespace std;
5
6 // Consider two floating-point numbers equal when
7 // the difference between them is very small.
8 bool equals(double a, double b, double tolerance)
9 {
10 return a == b || fabs(a - b) < tolerance;
11 }
12
13 double square_root(double x)
14 {
15 // Compute a provisional square root
16 double root = 1.0;
17
18 do // Loop until the provisional root
19 { // is close enough to the actual root
20 root = (root + x/root) / 2.0;
21 }
22 while ( !equals(root*root, x, 0.0001) );
23 return root;
24 }
25
26 int main()
27 {
28 for ( double d = 0.0; d < 100000.0; d += 0.0001 )
29 {
30 if ( !equals(square_root(d), sqrt(d), 0.001) )
31 cout << d << ": Expected " << sqrt(d) << ", but computed "
32 << square_root(d) << endl;
33 }
34 }

Listing 9.19 (squarerootcomparison.cpp) uses our equals function from Listing 9.16 (floatequals.cpp).
The third parameter specifies a tolerance; if the difference between the first two parameters is less than
the specified tolerance, the first two parameters are considered equal. Our new custom square_root
function uses the equals function. The main function uses the equals function as well. Observe,
however, that the tolerance used within the square root computation is smaller than the tolerance main
uses to check the result. The main function, therefore, uses a less strict notion of equality.

©2013 Richard L. Halterman Draft date: January 27, 2013

9.7. SUMMARY 220

The output of Listing 9.19 (squarerootcomparison.cpp):

0: Expected 0, but computed 0.0078125

0.0001: Expected 0.01, but computed 0.0116755
0.0008: Expected 0.0282843, but computed 0.0298389
0.0009: Expected 0.03, but computed 0.0313164
0.001: Expected 0.0316228, but computed 0.0327453

shows that our custom square root function produces results outside of main’s acceptable tolerance for five
values. Five wrong answers out of one billion tests represents a 0.0000005% error rate. While this error
rate is very small, indicates our square_root function is not perfect. Our function is not trustworthy
because one of values that causes the function to fail may be very important to a particular application.

9.7 Summary

• The development of larger, more complex programs is more manageable when the program consists
of multiple programmer-defined functions.

• Every function has one definition but can have many invocations.

• A function definition includes the function’s name, parameters, return type, and body.

• A function name, like a variable name, is an identifier.

• Formal parameters are the parameters as they appear in a function’s definition; actual parameters are
the arguments supplied by the caller.

• Formal parameters are essentially variables local to the function; actual parameters may be variables,
expressions, or literal values.

• The values of the actual parameters are copied into the formal parameters when a function is invoked.

• Callers must pass to functions the number of parameters specified in the function definition. The
types of the actual parameters must be compatible with the types of the formal parameters.

• With C++’s default call-by-value parameter passing mechanism, modifying a formal parameter within
a function will not affect the value of the caller’s actual parameter. The formal parameter is simply a
copy of the caller’s actual parameter.

• Variables declared within a function are local to that function definition. Local variables cannot be
seen by code outside the function definition.

• During a program’s execution, local variables live only when the function is executing. When a
particular function call is finished, the space allocated for its local variables is freed up.

• Programmers should supply comments for each function indicating its purpose and the role(s) of
its parameter(s) and return value. Additional information about the function’s author, date of last
modification, and other information may be required in some situations.

©2013 Richard L. Halterman Draft date: January 27, 2013

9.8. EXERCISES 221

9.8 Exercises

1. Is the following a legal C++ program?

int proc(int x)
{
return x + 2;
}

int proc(int n)
{
return 2*n + 1;
}

int main()
{
int x = proc(5);
}

2. Is the following a legal C++ program?

int proc(int x)
{
return x + 2;
}

int main()
{
int x = proc(5),
y = proc(4);
}

3. Is the following a legal C++ program?

#include <iostream>

void proc(int x)
{
std::cout << x + 2 << std::endl;
}

int main()
{
int x = proc(5);
}

4. Is the following a legal C++ program?

#include <iostream>

void proc(int x)
{

©2013 Richard L. Halterman Draft date: January 27, 2013

9.8. EXERCISES 222

std::cout << x + 2 << std::endl;

}

int main()
{
proc(5);
}

5. Is the following a legal C++ program?

#include <iostream>

int proc(int x, int y)

{
return 2*x + y*y;
}

int main()
{
std::cout << proc(5, 4) << std::endl;
}

6. Is the following a legal C++ program?

#include <iostream>

int proc(int x, int y)

{
return 2*x + y*y;
}

int main()
{
std::cout << proc(5) << std::endl;
}

7. Is the following a legal C++ program?

#include <iostream>

int proc(int x)
{
return 2*x*x;
}

int main()
{
std::cout << proc(5, 4) << std::endl;
}

8. Is the following a legal C++ program?

©2013 Richard L. Halterman Draft date: January 27, 2013

9.8. EXERCISES 223

#include <iostream>

proc(int x)
{
std::cout << 2*x*x << std::endl;
}

int main()
{
proc(5);
}

9. The programmer was expecting the following program to print 200. What does it print instead? Why
does it print what it does?
#include <iostream>

void proc(int x)
{
x = 2*x*x;
}

int main()
{
int num = 10;
proc(num);
std::cout << num << std::endl;
}

10. Is the following program legal since the variable x is used in two different places (proc and main)?
Why or why not?
#include <iostream>

int proc(int x)
{
return 2*x*x;
}

int main()
{
int x = 10;
std::cout << proc(x) << std::endl;
}

11. Is the following program legal since the actual parameter has a different name from the formal pa-
rameter (y vs. x)? Why or why not?
#include <iostream>

int proc(int x)
{

©2013 Richard L. Halterman Draft date: January 27, 2013

9.8. EXERCISES 224

return 2*x*x;
}

int main()
{
int y = 10;
std::cout << proc(y) << std::endl;
}

12. Complete the distance function started in Section 9.5. Test it with several points to convince
yourself that is correct.
What happens if a caller passes too many parameters to a function?
13. What happens if a caller passes too few parameters to a function?
14. What are the rules for naming a function in C++?
15. Consider the following function definitions:
#include <iostream>

int fun1(int n)
{
int result = 0;
while ( n )
{
result += n;
n--;
}
return result;
}

void fun2(int stars)

{
for ( int i = 0; i < stars; i++ )
std::cout << "*";
std::cout << std::endl;
}

double fun3(double x, double y)

{
return 2*x*x + 3*y;
}

bool fun4(char ch)

{
return ch >='A' && ch <= 'Z';
}

bool fun5(int a, int b, int c)

{
return (a <= b) ? (b <= c) : false;

©2013 Richard L. Halterman Draft date: January 27, 2013

9.8. EXERCISES 225

int fun6()
{
return rand() % 2;
}

Examine each of the following print statements. If the statement is illegal, explain why it is illegal;
otherwise, indicate what the statement will print.
(a) cout << fun1(5) << endl;
(b) cout << fun1() << endl;
(c) cout << fun1(5, 2) << endl;
(d) cout << fun2(5) << endl;
(e) fun2(5);
(f) cout << fun3(5, 2) << endl;
(g) cout << fun3(5.0, 2.0) << endl;
(h) cout << fun3('A', 'B') << endl;
(i) cout << fun3(5.0) << endl;
(j) cout << fun3(5.0, 0.5, 1.2) << endl;
(k) cout << fun4('T') << endl;
(l) cout << fun4('t') << endl;
(m) cout << fun4(5000) << endl;
(n) fun4(5000);
(o) cout << fun5(2, 4, 6) << endl;
(p) cout << fun5(4, 2, 6) << endl;
(q) cout << fun5(2, 2, 6) << endl;
(r) cout << fun5(2, 6) << endl;
(s) if ( fun5(2, 2, 6) )
cout << "Yes"<< endl;
else
cout << "No"<< endl;
(t) cout << fun6() << endl;
(u) cout << fun6(4) << endl;
(v) cout << fun3(fun1(3), 3) << endl;
(w) cout << fun3(3, fun1(3)) << endl;
(x) cout << fun1(fun1(fun1(3))) << endl;
(y) cout << fun6(fun6()) << endl;

©2013 Richard L. Halterman Draft date: January 27, 2013

9.8. EXERCISES 226

©2013 Richard L. Halterman Draft date: January 27, 2013

 227

Chapter 10

More on Functions

This chapter covers some additional aspects of functions in C++. Recursion, a key concept in computer
science is introduced.

10.1 Global Variables

All variables to this point have been local to functions or local to blocks within the bodies of conditional or
iterative statements. Local variables have some very desirable properties:

• The memory required to store a local variable is used only when the variable is in scope. When the
program execution leaves the scope of a local variable, the memory for that variable is freed up and
can be used for a local variable in another function when that function is invoked.

• The same variable name can be used in different functions without any conflict. The compiler derives
all of its information about a local variable used within a function from the declaration of that variable
in that function. The compiler will not look for the declaration of a local variable in the definition
of another function. Thus, there is no way a local variable in one function can interfere with a local
variable declared in another function.

A local variable is transitory, so its value is lost in between function invocations. Sometimes it is desirable
to have a variable that lives as long as the program is running; that is, until the main function completes.
In contrast to a local variable, a global variable is declared outside of all functions and is not local to any
particular function. In fact, any function that appears in the text of the source code after the point of the
global variable’s declaration may legally access and/or modify that global variable.
Listing 10.1 (globalcalculator.cpp) is a modification of Listing 9.10 (calculator.cpp) that uses a global
variable named result that is shared by several functions in the program.

Listing 10.1: globalcalculator.cpp

1 #include <iostream>
2 #include <cmath>
3
4 using namespace std;
5

©2013 Richard L. Halterman Draft date: January 27, 2013

 10.1. GLOBAL VARIABLES 228

6 /*
7 * help_screen
8 * Displays information about how the program works
9 * Accepts no parameters
10 * Returns nothing
11 * /
12 void help_screen()
13 {
14 cout << "Add: Adds two numbers" << endl;
15 cout << " Example: a 2.5 8.0" << endl;
16 cout << "Subtract: Subtracts two numbers" << endl;
17 cout << " Example: s 10.5 8.0" << endl;
18 cout << "Print: Displays the result of the latest operation"
19 << endl;
20 cout << " Example: p" << endl;
21 cout << "Help: Displays this help screen" << endl;
22 cout << " Example: h" << endl;
23 cout << "Quit: Exits the program" << endl;
24 cout << " Example: q" << endl;
25 }
26
27 /*
28 * menu
29 * Display a menu
30 * Accepts no parameters
31 * Returns the character entered by the user.
32 */
33 char menu()
34 {
35 // Display a menu
36 cout << "=== A)dd S)ubtract P)rint H)elp Q)uit ===" << endl;
37 // Return the char entered by user
38 char ch;
39 cin >> ch;
40 return ch;
41 }
42
43 /*
44 * Global variables used by several functions
45 */
46 double result = 0.0, arg1, arg2;
47
48 /*
49 * get_input
50 * Assigns the globals arg1 and arg2 from user keyboard
51 * input
52 */
53 void get_input()
54 {
55 cin >> arg1 >> arg2;
56 }
57
58 /*
59 * report
60 * Reports the value of the global result

©2013 Richard L. Halterman Draft date: January 27, 2013

 10.1. GLOBAL VARIABLES 229

61 */
62 void report()
63 {
64 cout << result << endl;
65 }
66
67 /*
68 * add
69 * Assigns the sum of the globals arg1 and arg2
70 to the global variable result
71 */
72
73 void add()
74 {
75 result = arg1 + arg1;
76 }
77
78 /*
79 * subtract
80 * Assigns the difference of the globals arg1 and arg2
81 * to the global variable result
82 */
83
84 void subtract()
85 {
86 result = arg1 - arg2;
87 }
88
89 /*
90 * main
91 * Runs a command loop that allows users to
92 * perform simple arithmetic.
93 */
94 int main()
95 {
96 bool done = false; // Initially not done
97 do
98 {
99 switch ( menu() )
100 {
101 case 'A': // Addition
102 case 'a':
103 get_input();
104 add();
105 report();
106 break;
107 case 'S': // Subtraction
108 case 's':
109 get_input();
110 subtract();
111 report();
112 case 'P': // Print result
113 case 'p':
114 report();
115 break;

©2013 Richard L. Halterman Draft date: January 27, 2013

 10.1. GLOBAL VARIABLES 230

116 case 'H': // Display help screen

117 case 'h':
118 help_screen();
119 break;
120 case 'Q': // Quit the program
121 case 'q':
122 done = true;
123 break;
124 }
125 }
126 while ( !done );
127 }

Listing 10.1 (globalcalculator.cpp) uses global variables result, arg1, and arg2. These names no
longer appear in the main function. These global variables are accessed and/or modified in four different
functions: get_input, report, add, and subtract.
When in the course of translating the statements within a function to machine language, the compiler
resolves a variable it encounters as follows:

• If the variable has a local declaration (that is, it is a local variable or parameter), the compiler will
use the local variable or parameter, even if a global variable of the same name exists. Local vari-
ables, therefore, take precedence over global variables. We say the local declaration hides the global
declaration in the scope of the local variable.
• If the variable has no local declaration but is declared as a global variable, the compiler will use the
global variable.
• If the variable has neither a local declaration nor a global declaration, then the variable is undefined,
and its use is an error.

In the situation where a local variable hides a global variable of the same name, there is a way to access
both the local variable and like-named global variable within the local variable’s scope. Suppose a program
has a global variable named x and a function with a local variable named x. The statement
x = 10;

within the scope of the local variable will assign the local x. The following statement will assign the global
variable x in the scope of the local variable of the same name:
::x = 10;

The :: operator is called the scope resolution operator. This special syntax may be used whenever a global
variable is accessed within a function, but usually it only used when necessary to access a hidden global
variable.
If the value of a local variable is used by a statement before that variable has been given a value,
either through initialization or assignment, the compiler will issue a warning. For example, the Visual C++
compiler will issue a warning about code in the following function:
void uninitialized()
{
int x; // Declare the variable
cout << x; // Then use it
}

©2013 Richard L. Halterman Draft date: January 27, 2013

10.1. GLOBAL VARIABLES 231

The warning is

warning C4700: uninitialized local variable ’x’ used

(The only way to avoid this warning in Visual C++ is to turn off all warnings.) A local variable has an
undefined value after it is declared without being initialized. Its value should not be used until it has been
properly assigned. Global variables, however, do not need to be initialized before they are used. Numeric
global variables are automatically assigned the value zero. This means the initialization of result in List-
ing 10.1 (globalcalculator.cpp) is superfluous, since result will be assigned zero automatically. Boolean
global variables are automatically assigned zero as well, as zero represents false (see Section 5.1).
When it is acceptable to use global variables, and when is it better to use local variables? In general,
local variables are preferred to global variables for several reasons:

• When a function uses local variables exclusively and performs no other input operations (like using
the cin object), its behavior is influenced only by the parameters passed to it. If a non-local variable
appears, the function’s behavior is affected by every other function that can modify that non-local
variable. As a simple example, consider the following trivial function that appears in a program:
int increment(int n)
{
return n + 1;
}

Can you predict what the following statement within that program will print?
cout << increment(12) << endl;

If your guess is 13, you are correct. The increment function simply returns the result of adding
one to its argument. The increment function behaves the same way each time it is called with the
same argument.
Next, consider the following three functions that appear in some program:
int process(int n)
{
return n + m; // m is a global integer variable
}

void assign_m()
{
m = 5;
}

void inc_m()
{
m++;
}

Can you predict the what the following statement within the program will print?
cout << process(12) << endl;

We cannot predict what this statement in isolation will print. The following scenarios all produce
different results:

©2013 Richard L. Halterman Draft date: January 27, 2013

10.1. GLOBAL VARIABLES 232

assign_m();
cout << process(12) << endl;

prints 17,
m = 10;
cout << process(12) << endl;

prints 22,
m = 0;
inc_m();
inc_m();
cout << process(12) << endl;

prints 14, and

assign_m();
inc_m();
inc_m();
cout << process(12) << endl;

prints 19. The identical printing statements print different values depending on the cumulative effects
of the program’s execution up to that point.
It may be difficult to locate an error if that function fails because it may be the fault of another func-
tion that assigned an incorrect value to the global variable. The situation may be more complicated
than the simple examples above; consider:
assign_m();
.
. /* 30 statements in between, some of which may change a,
. b, and m */
.
if ( a < 2 && b <= 10 )
m = a + b - 100;
.
. /* 20 statements in between, some of which may change m
.
cout << process(12) << endl;

• A nontrivial program that uses non-local variables will be more difficult for a human reader to un-
derstand than one that does not. When examining the contents of a function, a non-local variable
requires the reader to look elsewhere (outside the function) for its meaning:
// Linear function
double f(double x)
{
return m*x + b;
}

What are m and b? How, where, and when are they assigned or re-assigned?
• A function that uses only local variables can be tested for correctness in isolation from other func-
tions, since other functions do not affect the behavior of this function. This function’s behavior is
only influenced only by its parameters, if it has any.

©2013 Richard L. Halterman Draft date: January 27, 2013

 10.1. GLOBAL VARIABLES 233

The exclusion of global variables from a function leads to functional independence. A function that
depends on information outside of its scope to correctly perform its task is a dependent function. When
a function operates on a global variable it depends on that global variable being in the correct state for
the function to complete its task correctly. Nontrivial programs that contain many dependent functions
are more difficult debug and extend. A truly independent function that use no global variables and uses
no programmer-defined functions to help it out can be tested for correctness in isolation. Additionally, an
independent function can be copied from one program, pasted into another program, and work without
modification. Functional independence is a desirable quality.
Unlike global variables, global constants are generally safe to use. Code within functions that use
global constants are dependent on those constants, but since constants cannot be changed, developers need
not worry that other functions that have access to the global constants might disturb their values.
The use of global constants within functions has drawbacks in terms of program maintenance. As a
program evolves, code is added and removed. If a global constant is removed or its meaning changes during
the course of the program’s development, the change will affect any function using the global constant.
Listing 10.2 (digitaltimer.cpp) uses global constants to assist the display of a digital timer.

Listing 10.2: digitaltimer.cpp

1 #include <iostream>
2 #include <iomanip>
3 #include <ctime>
4
5 using namespace std;
6
7 // Some conversions from seconds
8 const clock_t SEC_PER_MIN = 60, // 60 sec = 1 min
9 SEC_PER_HOUR = 60 * SEC_PER_MIN, // 60 min = 1 hr
10 SEC_PER_DAY = 24 * SEC_PER_HOUR; // 24 hr = 24 hr
11
12 /*
13 * print_time
14 * Displays the time in hr:min:sec format
15 * seconds is the number of seconds to display
16 */
17
18 void print_time(clock_t seconds)
19 {
20 clock_t hours = 0, minutes = 0;
21
22 // Prepare to display time =============================
23 cout << endl;
24 cout << " ";
25
26 // Compute and display hours ===========================
27 hours = seconds/SEC_PER_HOUR;
28 cout << setw(2) << setfill('0') << hours << ":";
29
30 // Remove the hours from seconds
31 seconds %= SEC_PER_HOUR;
32
33 // Compute and display minutes =========================
34 minutes = seconds/SEC_PER_MIN;
35 cout << setw(2) << setfill('0') << minutes << ":";

©2013 Richard L. Halterman Draft date: January 27, 2013

 10.1. GLOBAL VARIABLES 234

36 // Remove the minutes from seconds

37 seconds %= SEC_PER_MIN;
38
39 // Compute and display seconds =========================
40 cout << setw(2) << setfill('0') << seconds << endl;
41 }
42
43 int main()
44 {
45 clock_t start = clock(); // Record starting time
46 clock_t elapsed = (clock() - start)/CLOCKS_PER_SEC, // Elapsed time in sec.
47 previousElapsed = elapsed;
48 // Counts up to 24 hours (1 day), then stops
49 while ( elapsed < SEC_PER_DAY )
50 {
51 // Update the display only every second
52 if ( elapsed - previousElapsed >= 1 )
53 {
54 // Remember when we last updated the display
55 previousElapsed = elapsed;
56 print_time(elapsed);
57 }
58 // Update elapsed time
59 elapsed = (clock() - start)/CLOCKS_PER_SEC;
60 }
61 }

In Listing 10.2 (digitaltimer.cpp):

• The main function controls the time initialization and update and deals strictly in seconds. The logic
in main is kept relatively simple.
• The code that extracts the hours, minutes, and seconds from a given number of seconds is isolated in
print_time. The print_time function can now be used anytime a value in seconds needs to
be expressed in the hours : minutes : seconds format.
• The second conversion constants (SEC_PER_HOUR, SEC_PER_MIN, and SEC_PER_DAY) are
global constants so that both functions can access them if necessary. In this case the functions use
different constants , but it makes sense to place all the conversion factors in one place.

Since the two functions divide the responsibilities in a way that each can be developed independently,
the design is cleaner and the program is easier to develop and debug. The use of constants ensures that the
shared values cannot be corrupted by either function.
The exclusion from a function’s definition of global variables and global constants does not guarantee
that it will always produce the same results given the same parameter values; consider
int compute(int n)
{
int favorite;
cout << "Please enter your favorite number: ";
cin >> favorite;
return n + favorite;
}

©2013 Richard L. Halterman Draft date: January 27, 2013

10.2. STATIC VARIABLES 235

The compute function avoid globals, yet we cannot predict the value of the expression compute(12).
Recall the increment function from above:
int increment(int n)
{
return n + 1;
}

Its behavior is totally predictable. Furthermore, increment does not modify any global variables, mean-
ing it cannot in any way influence the overall program’s behavior. We say that increment is a pure
function. A pure function cannot perform any input or output (for example, use the cout and cin ob-
jects), nor may it use global variables. While increment is pure, the compute function is impure. The
following function is impure also, since it performs output:
int increment_and_report(int n)
{
cout << "Incrementing " << n << endl;
return n + 1;
}

A pure function simply computes its return value and has no other observable side effects.
A function that uses only pure functions and otherwise would be considered pure is itself a pure func-
tion; for example:
int double_increment(int n)
{
return increment(n) + 1;
}

double_increment is a pure function since increment is pure; however, double_increment_with_report:

int double_increment_with_report(int n)
{
return increment_and_report(n) + 1;
}

is not a pure function since it calls increment_and_report which is impure.

10.2 Static Variables

Space in the computer’s memory for local variables and function parameters is allocated at run time when
the function begins executing. When the function is finished and returns, the memory used for the function’s
local variables and parameters is freed up for other purposes. If a function is never called, the variable’s
local variables and parameters will never occupy the computer’s memory.
Because a function’s locals are transitory, a function cannot ordinarily retain any information between
calls. C++ provides a way in which a variable local to a function can be retained in between calls. List-
ing 10.3 (counter.cpp) shows how declaring a local variable static allows it to remain in the computer’s
memory for the duration of the program’s execution.

Listing 10.3: counter.cpp

©2013 Richard L. Halterman Draft date: January 27, 2013

 10.2. STATIC VARIABLES 236

1 #include <iostream>
2 #include <iomanip>
3
4 using namespace std;
5
6 /*
7 * count
8 * Keeps track of a count.
9 * Returns the current count.
10 */
11
12 int count()
13 {
14 // cnt's value is retained between calls because it
15 // is declared static
16 static int cnt = 0;
17 return ++cnt; // Increment and return current count
18 }
19
20 int main()
21 {
22 // Count to ten
23 for ( int i = 0; i < 10; i++ )
24 cout << count() << ' ';
25 cout << endl;
26 }

In Listing 10.3 (counter.cpp), the count function is called 10 times. Each time it returns a tally of the
number of times it has been called:

1 2 3 4 5 6 7 8 9 10

By contrast, if you remove the word static from Listing 10.3 (counter.cpp), recompile it, and rerun it, it
prints:

1 1 1 1 1 1 1 1 1 1

because new memory is allocated for the cnt variable each time count is called.
The local declaration
static int cnt = 0;

allocates space for cnt and assigns zero to it once—at the beginning of the program’s execution. The space
set aside for cnt is not released until the program finishes executing.
Recall Listing 9.7 (evenbetterprompt.cpp) that included the following function:
int prompt(int n)

©2013 Richard L. Halterman Draft date: January 27, 2013

 10.2. STATIC VARIABLES 237

{
int value;
cout << "Please enter integer #" << n << ": ";
cin >> value;
return value;
}

The caller, main, used prompt function as follows:

int value1, value2, sum;
cout << "This program adds together two integers." << endl;
value1 = prompt(1); // Call the function
value2 = prompt(2); // Call the function again
sum = value1 + value2;

The first call to prompt prints the message

Please enter integer #1:

and awaits the user’s input. The second call prints the message

Please enter integer #2:

Another caller might use prompt within a loop like this:

int sum = 0;
for ( int i = 1; i < 10; i++ )
sum += prompt(i);

Notice that it is the caller’s responsibility to keep track of the proper number to pass to prompt. The caller
may make a mistake and pass the wrong number or may not want to manage such details. It would be better
to move the responsibility of tracking input count to prompt. static variables make that possible, as
Listing 10.4 (promptwithstatic.cpp)

Listing 10.4: promptwithstatic.cpp

1 #include <iostream>
2
3 using namespace std;
4
5 /*
6 * prompt requests an integer from the user and
7 * keeps track of the cumulative number of entries.
8 * Returns the value entered by the user.
9 */
10 int prompt()
11 {
12 static int count = 0;
13 int value;

©2013 Richard L. Halterman Draft date: January 27, 2013

 10.3. OVERLOADED FUNCTIONS 238

14 cout << "Please enter integer #" << ++count << ": ";
15 cin >> value;
16 return value;
17 }
18
19 int main()
20 {
21 int value1, value2, sum;
22 cout << "This program adds together two integers." << endl;
23 value1 = prompt(); // Call the function
24 value2 = prompt(); // Call the function again
25 sum = value1 + value2;
26 cout << value1 << " + " << value2 << " = " << sum << endl;
27 }

Listing 10.4 (promptwithstatic.cpp) behaves just like Listing 9.7 (evenbetterprompt.cpp) but in the new
version main does not have to keep track of the number of user entries.
Local static variables were inherited from the C programming language, but their need has dimin-
ished with the introduction of objects in C++ (see Chapter 14. Functions with static variables provide a
way to implement executable code with persistent state. Objects provide a more natural and more flexible
way to achieve the same effect.

10.3 Overloaded Functions

In C++, a program can have multiple functions with the same name. When two or more functions within
a program have the same name, the function is said to be overloaded. The functions must be different
somehow, or else the compiler would not know how to associate a call with a particular function definition.
The compiler identifies a function by more than its name; a function is uniquely identified by its signature.
A function signature consists of the function’s name and its parameter list. In the parameter list, only the
types of the formal parameters are important, not their names. If the parameter types do not match exactly,
both in number and position, then the function signatures are different. Consider the following overloaded
functions:

1. void f() { /* ... */ }

This version has no parameters, so its signature differs from all the others which each have at least
one parameter.

2. void f(int x) { /* ... */ }

This version differs from Version 3, since its single parameter is an int, not a double.

3. void f(double x) { /* ... */ }

This version differs from Version 2, since its single parameter is a double, not an int.

4. void f(int x, double y) { /* ... */ }

This version differs from Version 5 because, even though Versions 4 and 5 have the same number of
parameters with the same types, the order of the types is different.

©2013 Richard L. Halterman Draft date: January 27, 2013

 10.4. RECURSION 239

5. void f(double x, int y) { /* ... */ }

See the comments for Version 4.

Overloaded functions are a convenience for programmers. If overloaded functions were not allowed
(many programming languages do not support function overloading), new function names must be created
for different functions that perform basically the same task but accept different parameter types. It is better
for the programmer to choose the same name for the similar functions and let the compiler properly resolve
the differences. Overloading becomes a more important issue for constructors, special functions called
during object creation (Chapter 14).

10.4 Recursion

The factorial function is widely used in combinatorial analysis (counting theory in mathematics), probabil-
ity theory, and statistics. The factorial of n is often expressed as n!. Factorial is defined for non-negative
integers as
n! = n · (n − 1) · (n − 2) · (n − 3) · · · 2 · 1
and 0! is defined to be 1. Thus 6! = 6 · 5 · 4 · 3 · 2 · 1 = 720. Mathematicians precisely define factorial in this
way: 
 1 if n = 0
n! =
n · (n − 1)! otherwise.


This definition is recursive since the ! function is being defined, but ! is also used in the definition. A C++
function can be defined recursively as well. Listing 10.5 (factorialtest.cpp) includes a factorial function that
exactly models the mathematical definition.

Listing 10.5: factorialtest.cpp

1 #include <iostream>
2
3 using namespace std;
4
5 /*
6 * factorial(n)
7 * Computes n!
8 * Returns the factorial of n.
9 */
10 int factorial(int n)
11 {
12 if ( n == 0 )
13 return 1;
14 else
15 return n * factorial(n - 1);
16 }
17
18 int main()
19 {
20 // Try out the factorial function
21 cout << " 0! = " << factorial(0) << endl;
22 cout << " 1! = " << factorial(1) << endl;
23 cout << " 6! = " << factorial(6) << endl;

©2013 Richard L. Halterman Draft date: January 27, 2013

 10.4. RECURSION 240

24 cout << "10! = " << factorial(10) << endl;

25 }

Observe that the factorial function in Listing 10.5 (factorialtest.cpp) uses no loop to compute its
result. The factorial function simply calls itself. The call factorial(6) is computed as follows:
factorial(6) = 6 * factorial(5)
= 6 * 5 * factorial(4)
= 6 * 5 * 4 * factorial(3)
= 6 * 5 * 4 * 3 * factorial(2)
= 6 * 5 * 4 * 3 * 2 * factorial(1)
= 6 * 5 * 4 * 3 * 2 * 1 * factorial(0)
= 6 * 5 * 4 * 3 * 2 * 1 * 1
= 6 * 5 * 4 * 3 * 2 * 1
= 6 * 5 * 4 * 3 * 2
= 6 * 5 * 4 * 6
= 6 * 5 * 24
= 6 * 120
= 720

Note that the factorial function can be slightly optimized by changing the if’s condition from
( n == 0 ) to ( n < 2 ). This change results in a function execution trace that eliminates two
function calls at the end:
factorial(6) = 6 * factorial(5)
= 6 * 5 * factorial(4)
= 6 * 5 * 4 * factorial(3)
= 6 * 5 * 4 * 3 * factorial(2)
= 6 * 5 * 4 * 3 * 2 * 1
= 6 * 5 * 4 * 3 * 2
= 6 * 5 * 4 * 6
= 6 * 5 * 24
= 6 * 120
= 720

Figure 10.1 shows the call sequence for factorial(6) invoked from with a main function.
A correct simple recursive function definition is based on four key concepts:

1. The function must optionally call itself within its definition; this is the recursive case.
2. The function must optionally not call itself within its definition; this is the base case.
3. Some sort of conditional execution (such as an if/else statement) selects between the recursive
case and the base case based on one or more parameters passed to the function.
4. Each invocation that does correspond to the base case must call itself with parameter(s) that move the
execution closer to the base case. The function’s recursive execution must converge to the base case.

Each recursive invocation must bring the function’s execution closer to it base case. The factorial
function calls itself in the else clause of the if/else statement. Its base case is executed if the condition
of the if statement is true. Since the factorial is defined only for non-negative integers, the initial invocation
of factorial must be passed a value of zero or greater. A zero parameter (the base case) results in no

©2013 Richard L. Halterman Draft date: January 27, 2013

 10.4. RECURSION 241

factorial(6) Function Call Sequence

(called from main)

Program Execution Timeline

main
6 720

factorial
5 120

factorial
4 24

factorial
3 6

factorial
2 2

factorial
1 1

factorial

Figure 10.1: Traces the function activations of the recursive function factorial when called from main
with an argument of 6. The arrows into an activation bar indicates the argument passed by the caller; the
arrows out show the value passed back to the caller.

recursive call. Any other positive parameter results in a recursive call with a parameter that is closer to zero
than the one before. The nature of the recursive process progresses towards the base case, upon which the
recursion terminates.
We can easily write a non-recursive factorial function, as Listing 10.6 (nonrecursfact.cpp) shows.

Listing 10.6: nonrecursfact.cpp

1 #include <iostream>
2
3 using namespace std;
4
5 /*
6 * factorial(n)
7 * Computes n!
8 * Returns the factorial of n.
9 */
10 int factorial(int n)
11 {

©2013 Richard L. Halterman Draft date: January 27, 2013

 10.4. RECURSION 242

12 int product = 1;
13 for ( int i = n; i > 0; i-- )
14 product *= i;
15 return product;
16 }
17
18 int main()
19 {
20 // Try out the factorial function
21 cout << " 0! = " << factorial(0) << endl;
22 cout << " 1! = " << factorial(1) << endl;
23 cout << " 6! = " << factorial(6) << endl;
24 cout << "10! = " << factorial(10) << endl;
25 }

Which factorial function is better, the recursive or non-recursive version? Generally, if the same
basic algorithm is being used by both the recursive and non-recursive functions, then the non-recursive
function will be more efficient. A function call is a relatively expensive operation compared to a variable
assignment or comparison. The body of the non-recursive factorial function invokes no functions,
but the recursive version calls a function—it calls itself—during all but the last recursive invocation. The
iterative version of factorial is therefore more efficient than the recursive version.
Even though the iterative version of the factorial function is technically more efficient than the recursive
version, on most systems you could not tell the difference. The reason is the factorial function “grows” fast,
meaning it returns fairly large results for relatively small arguments. For example, with 32-bit integers,
the 13! is the largest integer value that can be computed. The execution time difference between the two
versions is negligible on most systems.
Recall the gcd functions from Section 9.2. It computed he greatest common divisor (also known as
greatest common factor) of two integer values. It works, but it is not very efficient. A better algorithm is
used in Listing 10.7 (gcd.cpp). It is based on one of the oldest algorithms known, developed by Euclid
around 300 B.C.

Listing 10.7: gcd.cpp

1 #include <iostream>
2
3 using namespace std;
4
5 /*
6 * gcd(m, n)
7 * Uses Euclid's method to compute
8 * the greatest common divisor
9 * (also called greatest common
10 * factor) of m and n.
11 * Returns the GCD of m and n.
12 */
13 int gcd(int m, int n)
14 {
15 if ( n == 0 )
16 return m;
17 else
18 return gcd(n, m % n);
19 }
20

©2013 Richard L. Halterman Draft date: January 27, 2013

 10.4. RECURSION 243

21 int iterative_gcd(int num1, int num2)

22 {
23 // Determine the smaller of num1 and num2
24 int min = (num1 < num2) ? num1 : num2;
25 // 1 is definitely a common factor to all ints
26 int largestFactor = 1;
27 for ( int i = 1; i <= min; i++ )
28 if ( num1 % i == 0 && num2 % i == 0 )
29 largestFactor = i; // Found larger factor
30 return largestFactor;
31 }
32
33 int main()
34 {
35 // Try out the gcd function
36 for ( int num1 = 1; num1 <= 100; num1++ )
37 for ( int num2 = 1; num2 <= 100; num2++ )
38 cout << "gcd(" << num1 << "," << num2
39 << ") = " << gcd(num1, num2) << endl;
40 }

Note that this gcd function is recursive. The algorithm it uses is much different from our original iterative
version. Because of the difference in the algorithms, this recursive version is actually much more efficient
than our original iterative version. A recursive function, therefore, cannot be dismissed as inefficient just
because it is recursive.
Listing 10.8 (histobar.cpp) provides another example of a recursive function. The segments1 function
uses iteration to draw segments that make up a bar that could be part of a histogram. The segments2
function does that same thing, except it uses recursion.

Listing 10.8: histobar.cpp

1 #include <iostream>
2
3 using namespace std;
4
5 // Draws a bar n segments long
6 // using iteration.
7 void segments1(int n)
8 {
9 while ( n > 0 )
10 {
11 cout << "*";
12 n--;
13 }
14 cout << endl;
15 }
16
17 // Draws a bar n segments long
18 // using recursion.
19 void segments2(int n)
20 {
21 if ( n > 0 )
22 {
23 cout << "*";

©2013 Richard L. Halterman Draft date: January 27, 2013

 10.5. MAKING FUNCTIONS REUSABLE 244

24 segments2(n - 1);
25 }
26 else
27 cout << endl;
28 }
29
30 int main()
31 {
32 segments1(3);
33 segments1(10);
34 segments1(0);
35 segments1(5);
36 cout << "-------------" << endl;
37 segments2(3);
38 segments2(10);
39 segments2(0);
40 segments2(5);
41 }

The outout of Listing 10.8 (histobar.cpp) shows that the two functions produce the same results:

***
**********

*****
-------------
***
**********

*****

10.5 Making Functions Reusable

In a function definition we can package functionality that can be used in many different places within a
program. Thus far, however, we have not seen how function definitions can be reused easily in other pro-
grams. For example, our is_prime function in Listing 9.9 (primefunc.cpp) works well within Listing 9.9
(primefunc.cpp), and it could be put to good use in other programs that need to test primality (encryption
software, for example, makes heavy use of prime numbers). We could use the copy-and-paste feature of
our favorite text editor to copy the is_prime function definition from Listing 9.9 (primefunc.cpp) into the
new encryption program we are developing.
It is possible to reuse a function in this way only if the function definition does not use any programmer-
defined global variables, programmer-defined global constants, nor other programmer-defined functions. If
a function does use any of these programmer-defined external entities, they must be included for the function
to compile. Said another way, the code in the function definition ideally will use only local variables and
parameters. Such a function is a truly independent function can be reused easily in multiple programs.
The notion of copying source code from one program to another is not ideal, however. It is too easy for
the copy to be incomplete or for some other error to be introduced during the copy. Furthermore, such code

©2013 Richard L. Halterman Draft date: January 27, 2013

 10.5. MAKING FUNCTIONS REUSABLE 245

duplication is wasteful. If 100 programs on a particular system all need to use the is_prime function,
under this scheme they must all include the is_prime code. This redundancy wastes space. Finally, in
perhaps the most compelling demonstration of the weakness of this copy-and-paste approach, what if a
bug is discovered in the is_prime function that all 100 programs are built around? When the error is
discovered and fixed in one program, the other 99 programs will still contain the bug. Their source code
must be updated, and they each then must be recompiled. The situation would be the same if a correct
is_prime function were updated to be made more efficient. The problem is this: all the programs using
is_prime define their own is_prime function; while the function definitions are meant to be identical,
there is no mechanism tying all these common definitions together. We would really like to reuse the
function as is without copying it.
Fortunately, C++ provides a way to develop functions in separate files, and these independently devel-
oped functions can be used by multiple programs. A way must be provided for the compiler to check if the
calling code is using the functions correctly, and the linker must combine the separately compiled files into
a complete program.
10.9 provides the first step in making a reusable is_prime function.

Listing 10.9: prime.h

1 bool is_prime(int);

The simple one-line code in 10.9 can be stored in a file called prime.h. In Visual Studio, you simply add a
new item to your project specified as a header file, name it prime.h, and you are ready to type the one line of
code into the newly created file. This file contains the prototype for our is_prime function (see our earlier
discussion of function prototypes in Section 8.1). Caller code that intends to use our is_prime function
must #include this file so that compiler can check to see if the caller is using our is_prime function
properly. An attempt, for example, to pass two arguments to is_prime would result in a compiler error
since the prototype specifies a single integer argument.
A second file, which could be named prime.cpp, appears in Listing 10.10 (prime.cpp).

Listing 10.10: prime.cpp

1 // prime.cpp
2 #include <cmath> // Needed for sqrt
3 #include "prime.h" // is_prime prototype
4
5 /*
6 * is_prime(n)
7 * Determines the primality of a given value
8 * n an integer to test for primality
9 * Returns true if n is prime; otherwise, returns false
10 */
11 bool is_prime(int n)
12 {
13 bool result = true; // Provisionally, n is prime
14 double r = n, root = sqrt(r);
15 // Try all possible factors from 2 to the square
16 // root of n
17 for ( int trial_factor = 2; result && trial_factor <= root; trial_factor++ )
18 result = (n % trial_factor != 0 );
19 return result;
20 }

©2013 Richard L. Halterman Draft date: January 27, 2013

 10.5. MAKING FUNCTIONS REUSABLE 246

The code in Listing 10.10 (prime.cpp) is placed in prime.cpp, a different file from prime.h. It provides an
implementation of the is_prime function. Notice the #include preprocessor directive in Listing 10.10
(prime.cpp) that references the file prime.h. While not required, this serves as a good check to see if the
implementation code in this file is faithful to the prototype specified in prime.h. This prime.cpp file is
compiled separately, and the compiler will report an error if the implementation of is_prime disagrees
with the information in the header file.
Note that the file prime.cpp does not contain a main function; main will appear in another file. Also
observe that we do not need to #include the iostream header, since the cout and cin objects are not
used anywhere in this file. The cmath header is #included since is_prime uses the sqrt function.
The final piece of the program is the calling code. Suppose Listing 10.11 (primetester.cpp) is added to
the project in a file named primetester.cpp.

Listing 10.11: primetester.cpp

1 #include <iostream>
2 #include "prime.h"
3
4 using namespace std;
5
6 /*
7 * main
8 * Tests for primality each integer from 2
9 * up to a value provided by the user.
10 * If an integer is prime, it prints it;
11 * otherwise, the number is not printed.
12 */
13 int main()
14 {
15 int max_value;
16 cout << "Display primes up to what value? ";
17 cin >> max_value;
18 for ( int value = 2; value <= max_value; value++ )
19 if ( is_prime(value) ) // See if value is prime
20 cout << value << " "; // Display the prime number
21 cout << endl; // Move cursor down to next line
22 }

Note that the file primetester.cpp uses a function named is_prime, but its definition is missing. The
definition for is_prime is found, of course, in prime.cpp.
Visual Studio will automatically compile and link the .cpp files when it builds the project. Each .cpp
is compiled independently on its own merits.
If you are using the Visual Studio Command Line tool, in order to build the program you would type

cl /EHsc /Za /W3 primetester.cpp prime.cpp

The executable file’s name is determined by the name of the first source file listed, in this case primetester.
If you are using the GCC tools instead of Visual Studio, in order to make the executable program named
primetester (or primetester.exe under the Microsoft Windows version of the GCC tools), you would issue
the command

©2013 Richard L. Halterman Draft date: January 27, 2013

 10.5. MAKING FUNCTIONS REUSABLE 247

g++ -o primetester -Wall -O1 -std=c++11 prime.cpp primetester.cpp

The GNU C++ compiler will separately compile the two source files producing two machine language object
files. The linker will then use those object files to create the executable. When the linker has created the
executable, it automatically deletes the two object files.
The is_prime function is now more readily available to other programs. If it becomes an often used
function in many programs, it can be compiled and placed into a special file called a library. In this form
it need not be recompiled each time a new program is built that requires it. If our is_prime is placed in
a dynamic library, its code can be loaded and linked at run time and shared by many executing programs.
We do not cover library creation in this text.
In Listing 8.7 (timeit.cpp), Listing 8.8 (measureprimespeed.cpp), and Listing 10.2 (digitaltimer.cpp) we
used the clock function from the <ctime> library to measure the elapsed time of sections of various
executing programs. In each of these programs the programmer must be aware of the clock_t type and
CLOCKS_PER_SEC constant, both defined in the <ctime> header file. Furthermore, the programmer
must use the clock function properly and correctly perform some arithmetic and include a messy type
cast operation. Armed with our knowledge of global variables (Section 10.1) and separate compilation, we
can provide a better programmer interface to the lower-level timing functions provided to the C library.
Listing 10.12 (timermodule.h) specifies some convenient timing functions.

Listing 10.12: timermodule.h

1 // Header file timermodule.h
2
3 // Reset the timer so it reads 0 seconds
4 void reset_timer();
5
6 // Start the timer. The timer will
7 // begin measuring elapsed time.
8 void start_timer();
9
10 // Stop the timer. The timer will
11 // retain the current elapsed time, but
12 // it will not measure any time while
13 // it is stopped.
14 void stop_timer();
15
16 // Return the cummulative time (in seconds)
17 // kept by the timer since it last was reset
18 double elapsed_time();

Listing 10.13 (timermodule.cpp) implements the functions declared in Listing 10.12 (timermodule.h).

Listing 10.13: timermodule.cpp

1 // File timermodule.h
2 // Implements the program timer module
3
4 #include <ctime>
5
6 // Global variable that keeps track of the

©2013 Richard L. Halterman Draft date: January 27, 2013

 10.5. MAKING FUNCTIONS REUSABLE 248

7 // elapsed time.
8 double elapsed;
9
10 // Global variable that counts the number of
11 // clock ticks since the most recent start time.
12 clock_t start_time;
13
14 // Global flag that indicates whether or not the
15 // timer is running.
16 bool running;
17
18 // Reset the timer so it reads 0 seconds
19 void reset_timer()
20 {
21 elapsed = 0.0;
22 running = false; // Ensure timer is not running
23 }
24
25 // Start the timer. The timer will
26 // begin measuring elapsed time.
27 // Starting the timer if it already is running has no effect
28 void start_timer()
29 {
30 // Starting an already running timer has no effect
31 if ( !running )
32 {
33 running = true; // Note that the timer is running
34 start_time = clock(); // Record start time
35 }
36 }
37
38 // Stop the timer. The timer will
39 // retain the current elapsed time, but
40 // it will not measure any time while
41 // it is stopped.
42 // Stopping the timer if it is not currently running has no
43 // effect.
44 void stop_timer()
45 {
46 // Stopping a non-running timer has no effect
47 if ( running )
48 {
49 clock_t stop_time = clock(); // Record stop time
50 running = false; // Stop the clock
51 // Add to the elapsed time how long it has been since we last
52 // started the timer
53 elapsed += static_cast<double>((stop_time - start_time))
54 / CLOCKS_PER_SEC;
55 }
56 }
57
58 // Return the cummulative running time (in seconds)
59 // kept by the timer since it last was reset
60 double elapsed_time()
61 {

©2013 Richard L. Halterman Draft date: January 27, 2013

 10.6. POINTERS 249

62 if ( running ) // Compute time since last reset

63 {
64 clock_t current_time = clock(); // Record current time
65 return elapsed + static_cast<double>((current_time - start_time))
66 / CLOCKS_PER_SEC;
67 }
68 else // Timer stopped; elapsed already computed in stop_timer
69 return elapsed;
70 }

Observe that the code in Listing 10.13 (timermodule.cpp) allows client code to stop the timer and restart it
later without losing any previously accumulated time. The implementation uses three global variables—
elapsed, start_time, and running)—to maintain the state of the timer. One or more of these global
variables is influenced by three functions—start_timer, stop_timer, reset_time. The fourth
function returns the value of the elapsed variable.
Listing 10.14 (bettertimeit.cpp) simplifies Listing 8.7 (timeit.cpp) with our new timer module.

Listing 10.14: bettertimeit.cpp

1 #include <iostream>
2 #include "timermodule.h" // Out timer module
3
4 using namespace std;
5
6 int main()
7 {
8 char letter;
9 cout << "Enter a character: ";
10 start_timer(); // Start timing
11 cin >> letter;
12 stop_timer(); // Stop timing
13 cout << elapsed_time() << " seconds" << endl;
14 }

The code within Listing 10.14 (bettertimeit.cpp) is simpler and cleaner than the code in Listing 8.7 (timeit.cpp).
All traces of the clock_t type and the messy arithmetic and casting are gone. The timer module provides
a simple interface to callers that hides the details on how the timing actually happens.
Despite the ease of use of our timer module, it has a servere limitation. Suppose you wish to measure
how long it takes for a function to execute and also, during that function’s execution, separately time a
smaller section of code within that function. When the function is finished executing, you would like to
know how long it took the function to do its job and how long a portion of its code took to execute. We
essentially need two independent timers, but with our timer module it is not possible to conduct simul-
taneously more than one timing. We will see a far superior way to model a program execution timer in
Section 16.2. We will use objects to enable us to maintain as many simultaneous stopwatches as we need.

10.6 Pointers

Ordinarily we need not be concerned about where variables live in the computer’s memory during a pro-
gram’s execution. The compiler generates machine code that takes care of those details for us. Some
systems software like operating systems and device drivers need to access specific memory locations in

©2013 Richard L. Halterman Draft date: January 27, 2013

10.6. POINTERS 250

order to interoperate with hardware. Systems programmers, therefore, must be able to write code that
can access such lower-level detail. Developers of higher-level applications sometimes need to access the
address of variables to achieve specialized effects.
Each byte in a computer’s memory is numbered with a unique address. The first address is 0, and
the locations are numbered sequentially up to some maximum value allowed by the operating system and
hardware. A C++ variable is stored in memory, so each variable lives at a particular address.
If x is a variable of any type, the expression
&x

represents the address of x. The expression

&x

is really just a number—the numeric address of the variable’s memory location, but except for some situa-
tions in systems programming, programmers rarely need to treat this value as a number. The & operator is
called the address of operator.
While an address is really just a non-negative integer value, C++ uses a special notation when dealing
with addresses. In the following declaration
int *p;

the variable p is not an int itself; the * symbol in the declaration signifies that p is a pointer to an int.
This means that p can be assigned to the address of an int. The following sequence of code
int x;
x = 3;
int *p;
p = &x;

assigns the address of x to p. We can visualize the execution of these four statements in Figure 10.2. The *
symbol used as shown above during a variable declaration indicates that the variable is a pointer. It will be
used to refer to another variable or some other place in memory. In this case, the sequence of assignments
allows pointer p to refer to variable x.
In order to access memory via a pointer, we use the unary * operator. When not used in the context of a
declaration, the unary * operator is called the pointer dereferencing operator. Continuing the code sequence
above,
int x;
x = 3;
int *p;
p = &x;
*p = 5;

the statement
*p = 5;

copies the value 5 into the address referenced by the pointer p. This means that x’s value is changed.
Figure 10.3 illustrates the full sequence. Notice that the assignment to *p modifies variable x’s value. The
pointer p provides another way to access the memory allocated for x.
It is important to note that the statement

©2013 Richard L. Halterman Draft date: January 27, 2013

10.6. POINTERS 251

Figure 10.2: Pointer declaration and assignment

*p = 5;

is the first assignment statement we have seen that uses more than just a single variable name on the left of
the assignment operator. The statement is legal because the expression *p represents a memory location
that can store a value. Here, *p stands in the place of the variable x, and we easily can assign a value to x.

The unary * operator has two distinct meanings depending on the context:
1. At the pointer’s declaration, for example,
double *p;

the * indicates that p is a pointer to an double; it is not itself a double.

2. After the pointer’s declaration, for example,
*p = 0.08;

the expression *p represents the contents of memory at the address assigned

to p.

The * operator is the inverse of the & operator. & finds the address of a variable, and, when this address
is assigned to a pointer variable like p, the * operator can be used to access the memory referenced by the
pointer.

©2013 Richard L. Halterman Draft date: January 27, 2013

10.6. POINTERS 252

Figure 10.3: Assignment via a pointer

A pointer may be declared and not assigned, as in

int *p;

We say p is an uninitialized pointer, sometimes called a wild pointer. If p is a local variable, its contents
are undetermined bits. Because of p’s declared type, these bits are interpreted as an address, so the net
effect is that the uninitialized pointer p points to a random location in the computer’s memory. An attempt
to dereference p, as in
*p = 500;

is certainly asking for trouble. This statement attempts to write the value 500 at some unknown-to-the-
programmer memory location. Often the address is not part of the area of memory set aside for the executing
program, so the operating system steps in and issues a run-time error. This is the best possible result for
misusing a wild pointer. It is possible, however, that the spurious address in within the executing program’s
domain. In this case the value 500 may overwrite another variable or the compiled machine language
instructions of the program itself! Such errors are difficult to track down because the overwritten value of
the variable cannot be detected until the program attempts to use the variable. The statement that misuses
the uninitialized pointer may be far away in the source code (even in a different source file) from the code
that attempts to use the clobbered variable. When the program fails, the programmer naturally looks around
in the code where the failure occurred—the code in the vicinity where the clobbered variable appears.

©2013 Richard L. Halterman Draft date: January 27, 2013

10.6. POINTERS 253

Unfortunately, the misbehaving code is fine, and the error lies elsewhere. There often is no easy way to
locate the true source of the error.

Compare the assignment

*p = 5;
to the following (illegal C++) statement:
p = 5; // Illegal statement as is
This statement attempts to make p to refer to the memory address 5; it does not
assign the value 5 to the memory to which p refers. C++ is very strict about
disallowing the mixing of pointers and non-pointers across assignment.
Systems programmers sometimes need to assign a pointer to a particular address
in memory, and C++ permits the assignment with a special kind of type cast, the
reinterpret_cast:
p = reinterpret_cast<int *>(5); // Legal
The familiar static_cast (see Section 4.2) will not work. Why is C++ so strict
when it comes to assignments of pointers to non-pointers and vice versa? It is easy
to make a mistake such as omitting the * operator when it is needed, so the special
cast forces the programmer to pause and consider whether the mixed assignment
truly is necessary or whether it is a programming error.

C++ provides the reserved word nullptr to represent a pointer to “nothing.” It stands for “null
pointer.” This keyword is new to the current ISO C++11 standard and is supported by both Visual C++
11.0 in Visual Studio 2012 and GCC 4.6. The following statement
p = nullptr;

indicates that p is pointing nowhere. On most platforms, nullptr maps to address zero, which is out of
bounds for any running program. Dereferencing p thus would result in a run-time error, and the problem
can be located quickly in a debugger.
C++ allows the literal value 0 to be used in place of nullptr. The statement
p = 0;

achieves the same result as assigning nullptr to p. This is how C++ programmers assigned a pointer
to point to nothing before the nullptr keyword was available. Since newer compilers support existing
C++ source code, the literal zero assignment still works. You should use the nullptr keyword if your
compiler supports it, because it improves the source code readability. Since 0 can represent both an integral
value and a pointer to any type, the following code (p is declared to be a pointer to an integer):
p = 0;
*p = 0;

The first statement assigns null to p, while the second statement sets the data to which p points to zero. Said
another way, the first statement changes what p points to; the second statement changes where p points.
Superficially, the two statements look very similar and are easy to confuse. Next, consider the statements
p = nullptr; // OK
*p = nullptr; // Error, will not compile

©2013 Richard L. Halterman Draft date: January 27, 2013

 10.7. CALL BY REFERENCE 254

The second statement contains an error, because the nullptr literal may be assigned only to a pointer
type.

The nullptr reserved word is part of the C++11 standard. The name nullptr
is simply an identifier (for example, a variable or function name) for older com-
pilers. Before the nullptr constant the literal 0 (zero) was considered the null
pointer reference. For backwards compatibility with older code C++11 allows you
to use 0 in place of nullptr, but if possible you should avoid this practice when
writing new code. The nullptr literal allows the complier to perform better
type checking. To see why, suppose the programmer believes the variable p is a
pointer to an integer, but p is instead a simple integer:
// Programmer believes p is a pointer to an integer
p = 0;
In this case the compiler is powerless to detect a problem because p is an integer,
and zero is a valid integer value. If p really is an integer rather a pointer, the
compiler will flag the following code:
// Programmer believes p is a pointer to an integer
p = nullptr;
because the nullptr may not be assigned to a non-pointer type.

The nullptr constant and the notion of a pointer pointing nowhere becomes more useful when build-
ing dynamic data structures (see, for example, Section 16.4).

10.7 Call by Reference

The default technique for passing parameters to functions is call by value (see Section 9.3). C++ also
supports call by reference, also known as pass by reference, which allows functions to alter the values of
formal parameters passed by callers.
Consider Listing 10.15 (faultyswap.cpp), which uses a function that attempts to interchange the values
of its two integer parameters.

Listing 10.15: faultyswap.cpp

1 #include <iostream>
2
3 using namespace std;
4
5 /*
6 * swap(a, b)
7 * Attempts to interchange the values of
8 * its parameters a and b. That it does, but
9 * unfortunately it only affects the local
10 * copies.
11 */
12 void swap(int a, int b)
13 {

©2013 Richard L. Halterman Draft date: January 27, 2013

 10.7. CALL BY REFERENCE 255

14 int temp = a;
15 a = b;
16 b = temp;
17 }
18
19 /*
20 * main
21 * Attempts to interchange the values of
22 * two variables using a faulty swap function.
23 */
24 int main()
25 {
26 int var1 = 5, var2 = 19;
27 cout << "var1 = " << var1 << ", var2 = " << var2 << endl;
28 swap(var1, var2);
29 cout << "var1 = " << var1 << ", var2 = " << var2 << endl;
30 }

The output of Listing 10.15 (faultyswap.cpp) is

var1 = 5, var2 = 19
var1 = 5, var2 = 19

Unfortunately, the swap function simply interchanges copies of the actual parameters, not the actual pa-
rameters themselves. We really would like to write a function that interchanges the caller’s variables.
Call by reference can be used to achieve the desired effect. C++ can do call by reference in two ways:
pointer parameters and reference parameters.

10.7.1 Call by Reference via Pointers

Pointers (see Section 10.6) allow us to access the memory locations of variables. We can use this capa-
bility to allow a function to modify the values of variables that are local to the functions that called it.
Listing 10.16 (swapwithpointers.cpp) provides a correct version of our variable interchange program.

Listing 10.16: swapwithpointers.cpp

1 #include <iostream>
2
3 using namespace std;
4
5 /*
6 * swap(a, b)
7 * Interchanges the values of memory
8 * referenced by its parameters a and b.
9 * It effectively interchanges the values
10 * of variables in the caller's context.
11 */
12 void swap(int *a, int *b)
13 {
14 int temp = *a;

©2013 Richard L. Halterman Draft date: January 27, 2013

 10.7. CALL BY REFERENCE 256

15 *a = *b;
16 *b = temp;
17 }
18
19 /*
20 * main
21 * Interchanges the values of two variables
22 * using the swap function.
23 */
24 int main()
25 {
26 int var1 = 5, var2 = 19;
27 cout << "var1 = " << var1 << ", var2 = " << var2 << endl;
28 swap(&var1, &var2);
29 cout << "var1 = " << var1 << ", var2 = " << var2 << endl;
30 }

The output of Listing 10.16 (swapwithpointers.cpp) is

var1 = 5, var2 = 19
var1 = 19, var2 = 5

which is the result we were trying to achieve. The swap function can manipulate main’s variables directly
since we passed it pointers to those variables.
In Listing 10.16 (swapwithpointers.cpp):

• The formal parameters to swap, a and b, are pointers to integers; they are not integers themselves. In
order to access the integer to which the pointer named a refers, it must be dereferenced. That is why
any use of a in swap’s body is prefixed with the pointer dereferencing operator, *. The statement
int temp = *a;

assigns to the local variable temp the value of the variable to which a points. Since main passes
var1 as the first parameter in its call to swap, in this case a points to var1, so *a is effectively
another way to express var1. The local temp thus is assigned the value of main’s var1.

• In swap’s statement
*a = *b;

since *a is effectively main’s var1 and *b is effectively main’s var2, this statement assigns the
value of main’s var2 to its var1 variable.

• The swap function’s

*b = temp;

statement assigns the local temp value to main’s var2 variable, since swap’s b parameter points
to main’s var2.

• Within main, the call

©2013 Richard L. Halterman Draft date: January 27, 2013

 10.7. CALL BY REFERENCE 257

swap(&var1, &var2);

passes the addresses of its local variables to swap.

In reality, call by reference with pointers is still using call by value. Instead of passing copies of values,
we are passing copies of addresses. In Listing 10.16 (swapwithpointers.cpp), for example, the values of the
addresses of var1 and var2 are copied to the formal parameters a and b. The difference is we are not
attempting to reassign a or b; we are reassigning memory to which a and b point. Whether we use the
original address or a copy of the address, it is still the same address—the same numeric location in memory.

10.7.2 Call by Reference via References

Call by reference with pointers (see Section 10.7.1) can be used in both C and C++. Since C++ programs
often use C libraries, C++ programmers must be familiar with the pointer technique for call by reference.
C++, however, provides a simpler way of implementing call by reference using reference parameters. List-
ing 10.17 (swapwithreferences.cpp) shows how reference parameters can be used in our variable inter-
change program.

Listing 10.17: swapwithreferences.cpp

1 #include <iostream>
2
3 using namespace std;
4
5 /*
6 * swap(a, b)
7 * Interchanges the values of memory
8 * referenced by its parameters a and b.
9 * It effectively interchanges the values
10 * of variables in the caller's context.
11 */
12 void swap(int& a, int& b)
13 {
14 int temp = a;
15 a = b;
16 b = temp;
17 }
18
19 /*
20 * main
21 * Interchanges the values of two variables
22 * using the swap function.
23 */
24 int main()
25 {
26 int var1 = 5, var2 = 19;
27 cout << "var1 = " << var1 << ", var2 = " << var2 << endl;
28 swap(var1, var2);
29 cout << "var1 = " << var1 << ", var2 = " << var2 << endl;
30 }

The syntax of Listing 10.17 (swapwithreferences.cpp) is a bit cleaner than that of Listing 10.16 (swapwithpointers.cpp).
In Listing 10.17 (swapwithreferences.cpp):

©2013 Richard L. Halterman Draft date: January 27, 2013

 10.8. FUNCTION POINTERS 258

• The formal parameters to swap, a and b, are references to integers; this is signified by the &amp;
symbol following int in their declarations. a is used exactly like an integer; there is no need to
reference it with *. Because a is a reference, however, it is a alias to another variable or memory
location elsewhere. This means if you modify a, you also modify the variable it references, in this
case var1 in main. The statement
int temp = a;

assigns to the local variable temp the value of a, but since a is another way to get to var1, this
statement ultimately assigns var1’s value to temp.
• Within main, the call
swap(var1, var2);

does not require any special decoration. It looks like a normal call by value function invocation.
A programmer must be aware that swap uses reference parameters and that any function that uses
reference parameters can change the actual values passed to it.

Reference parameters were introduced into C++ so that some of the more advanced object-oriented
features could be implemented more cleanly. Some argue that for simpler situations like the swap function,
pointer call by reference is more desirable than reference parameter call by reference, because with pointer
call by reference, the caller is forced to pass addresses of the actual parameters that may be modified. In
this way there can be no doubt that call by reference is going on. With reference parameters, call by value
and call by reference cannot be distinguished at the call site, since a call with reference parameters looks
exactly like a call by value invocation.
In general, pure call by value functions are preferred to call by reference functions. Functions using
call by reference cause side effects. This means they can change the state of the program in ways that
can be determined only by looking inside of the function and seeing how it works. Functions that access
global variables (see Section 10.1) can also cause side effects. Program development is much easier when
functions can be treated as black boxes that perform computations in isolation without the possibility of
affecting anything outside of their local context. The result of a function’s work can be assigned to a
variable thus changing the state of the program, but that change is the responsibility of the caller, not the
responsibility of the function itself. With call by value, the function’s parameters and local variables come
into existence when the function executes, the parameters and local variables disappear when the function
is finished, and nothing else is affected by the function’s execution.
Side-effect-free functions can be developed and tested in isolation from the rest of the program. Once
programmers are satisfied with their correctness, they need not be touched again as the remainder of the
system is developed. Functions with side effects, however, have dependencies to other parts of the program,
and changes elsewhere in the system may require programmers to modify and re-evaluate existing functions.

10.8 Function Pointers

(Add some commentary.)

Listing 10.18: arithmeticeval.cpp

1 #include <iostream>
2
3 using namespace std;
4

©2013 Richard L. Halterman Draft date: January 27, 2013

 10.8. FUNCTION POINTERS 259

5 int add(int x, int y)

6 {
7 return x + y;
8 }
9
10 int multiply(int x, int y)
11 {
12 return x * y;
13 }
14
15 int evaluate(int (*f)(int, int), int x, int y)
16 {
17 return f(x, y);
18 }
19
20 int main()
21 {
22 cout << add(2, 3) << endl;
23 cout << multiply(2, 3) << endl;
24 cout << evaluate(&add, 2, 3) << endl;
25 cout << evaluate(&multiply, 2, 3) << endl;
26 }

The first parameter of the evaluate function,

int (*f)(int, int)

represents a function pointer. The parameter’s name is f, and f is a pointer to a function that accepts two
integer parameters and returns an integer result. Any function with a prototype that matches this one can be
passed as the first argument to evaluate. Both add and multiply qualify, and the expression
evaluate(&add, 2, 3)

passes the address of the add function to evaluate. The evaluate then invokes the function at the
address of its first parameter, passing parameters two and three as arguments to that function.
C++ has a somewhat relaxed syntax for function pointers that cannot be applied to pointers to data. The
asterisk can be omitted from the evaluate function’s definition, as in
int evaluate(int (f)(int, int), int x, int y)
{
return f(x, y);
}

The parentheses around f are still required, and due to these parentheses the compiler can deduce f is a
pointer to a function. Similarly, when evaluate is invoked, the ampersand can be omitted in front of the
function name, as is
cout << evaluate(add, 2, 3) << endl;

Since the compiler knows that add is a function and since add is not followed by parentheses, the compiler
can determine that this in not a call to add but rather the address of the add function.
Function pointers ... [finish this]

©2013 Richard L. Halterman Draft date: January 27, 2013

10.9. SUMMARY 260

10.9 Summary

• A global variable is available to any function defined after the global variable.

• Local variables are not automatically initialized, but global numeric variables are automatically ini-
tialized to zero.

• A global variable exists for the life of the program, but local variables are created during a function
call and are discarded when the function’s execution has completed.

• A local variable with the same name as a global variable hides the global from the code within the
function. The scope resolution operator (::) can be used to access the hidden global variable.

• Modifying a global variable directly affects the behavior of any function that uses that global vari-
able. A function that uses a global variable cannot be tested in isolation since its behavior will vary
depending on how code outside the function modifies the global variable.

• The behavior of an independent function is determined strictly by the parameters passed into it. An
independent function will not use global variables.

• Local variables are preferred to global variables, since the indiscriminate use of global variables leads
to functions that are less flexible, less reusable, and more difficult to understand.

• Unlike normal local variables, static local variables exist throughout the life of the program.

• Unlike normal local variables, a static local variable retains its value in between function invoca-
tions.

• All static local variables are initialized once, at the beginning of the program’s execution.

• A function’s signature consists of its name along with the types of parameters it accepts.

• A C++ program may have multiple functions with the same names as long as their signatures differ.

• A recursive function must optionally call itself or not as determined by a conditional statement. The
call of itself is the recursive case, and the base case does not make the recursive all. Each recursive
call should move the computation closer to the base case.

• Function prototypes can be placed in a .h header file which is #included by calling code. The
function definitions are then placed in a .cpp file which can be compiled separately from the calling
code.

• A pointer is a variable representing a memory address.

• The address of a variable can be determined with the address-of operator (&amp;). This address can
be assigned to a pointer variable.

• The * symbol is used during declaration to signify a pointer variable.

• When not used during a declaration, the unary * operator is the pointer dereferencing operator. The
pointer dereferencing operator in conjunction with a pointer variable is used to examine or modify
the contents of the memory referenced by the pointer.

• The pointer dereferencing operator (*) and address-of operator (&amp;) perform the inverse opera-
tions of each other. The &amp; operator yields the address of an object, while the * operator provides
the object at a given address.

©2013 Richard L. Halterman Draft date: January 27, 2013

10.10. EXERCISES 261

• In call-by-reference parameter passing, an actual parameter value is not copied into the formal pa-
rameter. Instead, the address of an actual parameter is passed as the formal parameter. The code
within the function can manipulate the value of the actual parameter through its address. The result
is that functions can modify actual parameters via call by reference.

• C++ implements call by reference two different ways: via pointer parameters and reference parame-
ters. The C programming language supports call by reference only with pointer parameters.

• You should use call by reference when a function would need to return more than one result. Since a
function can return only one value, multiple variables can be passed by reference to a function, and
the function can modify their values directly.

10.10 Exercises

1. Consider the following C++ code:

#include <iostream>

using namespace std;

int sum1(int n)
{
int s = 0;
while ( n > 0 )
{
s++;
n--;
}
return s;
}

int input;

int sum2()
{
int s = 0;
while ( input > 0 )
{
s++;
input--;
}
return s;
}

int sum3()
{
int s = 0;
for ( int i = input; i > 0; i-- )
s++;
return s;

©2013 Richard L. Halterman Draft date: January 27, 2013

10.10. EXERCISES 262

int main()
{
// See each question below for details
}

(a) What is printed if main is written as follows?

int main()
{
input = 5;
cout << sum1(input) << endl;
cout << sum2() << endl;
cout << sum3() << endl;
}

(b) What is printed if main is written as follows?

int main()
{
input = 5;
cout << sum1(input) << endl;
cout << sum3() << endl;
cout << sum2() << endl;
}

(c) What is printed if main is written as follows?

int main()
{
input = 5;
cout << sum2() << endl;
cout << sum1(input) << endl;
cout << sum3() << endl;
}

(d) Which of the functions sum1, sum2, and sum3 produce a side effect? What is the side effect?
(e) Which function may not use the input variable?
(f) What is the scope of the variable input? What is its lifetime?
(g) What is the scope of the variable i? What is its lifetime?
(h) Which of the functions sum1, sum2, and sum3 manifest good functional independence? Why?

2. Consider the following C++ code:

#include <iostream>

using namespace std;

int next_int1()
{
static int cnt = 0;

©2013 Richard L. Halterman Draft date: January 27, 2013

10.10. EXERCISES 263

cnt++;
return cnt;
}

int next_int2()
{
int cnt = 0;
cnt++;
return cnt;
}

int global_count = 0;

int next_int3()
{
global_count++;
return global_count;
}

int main()
{
for ( int i = 0; i < 5; i++ )
cout << next_int1() << " "
<< next_int2() << " "
<< next_int3() << endl;
}

(a) What does the program print?

(b) Which of the functions next_int1, next_int2, and next_int3 is the best function for
the intended purpose? Why?
(c) What is a better name for the function named next_int2?
(d) The next_int3 function works in this context, but why is it not a good implementation of
function that always returns the next largest integer?

3. Consider the following C++ code:

#include <iostream>

using namespace std;

int max(int n)
{
return n;
}

int max(int m, int n)

{
return (m >= n)? m : n;
}

©2013 Richard L. Halterman Draft date: January 27, 2013

10.10. EXERCISES 264

int max(int m, int n, int r)

{
int x = m;
if ( n > x )
x = n;
if ( r > x )
x = r;
return x;
}

int main()
{
cout << max(4) << endl;
cout << max(4, 5) << endl;
cout << max(5, 4) << endl;
cout << max(1, 2, 3) << endl;
cout << max(2, 1, 3) << endl;
cout << max(2, 1, 2) << endl;
}

(a) Is the program legal since there are three different functions named max?
(b) What does the program print?

4. Consider the following function:

int proc(int n)
{
if ( n < 1 )
return 1;
else
return proc(n/2) + proc(n - 1);
}

Evaluate each of the following expressions:

(a) proc(0)
(b) proc(1)
(c) proc(2)
(d) proc(3)
(e) proc(5)
(f) proc(10)
(g) proc(0)

5. Rewrite the gcd function so that it implements Euclid’s method but uses iteration instead of recur-
sion.

6. If x is a variable, how would you determine its address in the computer’s memory?

7. What is printed by the following code fragment?

©2013 Richard L. Halterman Draft date: January 27, 2013

10.10. EXERCISES 265

int x = 5, y = 3, *p = &x, *q = &y;

cout << "x = " << x << ", y = " << y << endl;
x = y;
cout << "x = " << x << ", y = " << y << endl;
x = 7;
cout << "x = " << x << ", y = " << y << endl;
*p = 10;
cout << "x = " << x << ", y = " << y << endl;
p = q;
*p = 20;
cout << "x = " << x << ", y = " << y << endl;

8. Given the declarations:

int x, y, *p, *q;

indicate what each of the following code fragments will print.

(a) p = &x;
x = 5;
cout << *p << endl;

(b) x = 5;
p = &x;
cout << *p << endl;

(c) p = &x;
*p = 8;
cout << *p << endl;

(d) p = &x;
q = &y;
x = 100;
y = 200;
*q = *p;
cout << x << ' ' << y << endl;
cout << *p << ' ' << *q << endl;

(e) p = &x;
q = &y;
x = 100;
y = 200;
q = p;
cout << x << ' ' << y << endl;
cout << *p << ' ' << *q << endl;

(f) x = 5;
y = 10;
p = q = &y;
cout << *p << ' ' << *q << endl;

©2013 Richard L. Halterman Draft date: January 27, 2013

10.10. EXERCISES 266

*p = 100;
*q = 1;
cout << x << ' ' << y << endl;

(g) x = 5;
y = 10;
p = q = &x;
*p = *q = y;
cout << x << ' ' << y << endl;

9. The following function does not behave as expected:

/*
* (Faulty function)
*
* get_range
* Establishes a range of integers. The lower value must
* be greater than or equal to min, and the upper value
* must be less than or equal to max.
* min is the lowest acceptable lower value.
* max is the highest acceptable upper value.
* lower is assigned the lower limit of the range
* upper is assigned the upper limit of the range
* /
void get_range(int min, int max, int lower, int upper)
{
cout << "Please enter a data range within the bounds "
<< min << "..." << max << ": ";
do // Loop until acceptable values are provided
{
cin >> lower >> upper;
if ( lower < min )
cout << lower << " is too low, please try again." << endl;
if ( upper > max )
cout << upper << " is too high, please try again." << endl;
}
while ( lower < min || upper > max );
}

(a) Modify the function so that it works using call by reference with pointers.
(b) Modify the function so that it works using call by reference with reference parameters.

10. Classify the following functions as pure or impure. x is a global variable and LIMIT is a global
constant.

(a) int f1(int m, int n)

{
return 2*m + 3*n;
}

©2013 Richard L. Halterman Draft date: January 27, 2013

10.10. EXERCISES 267

(b) int f2(int n)

{
return n - LIMIT;
}

(c) int f3(int n)

{
return n - x;
}

(d) void f4(int n)

{
cout << 2*n << endl;
}

(e) int f5(int n)

{
int m;
cin >> m;
return m * n;
}

(f) int f6(int n)

{
int m = 2*n, p;
p = 2*m - 5;
return p - n;
}

11.

©2013 Richard L. Halterman Draft date: January 27, 2013

10.10. EXERCISES 268

©2013 Richard L. Halterman Draft date: January 27, 2013

 269

Chapter 11

Arrays

The variables we have used to this point can assume only one value at a time. As we have seen, individual
variables can be used to create some interesting and useful programs; however, variables that can represent
only one value at a time do have their limitations. Consider Listing 11.1 (averagenumbers.cpp) which
averages five numbers entered by the user.

Listing 11.1: averagenumbers.cpp

1 #include <iostream>
2
3 using namespace std;
4
5 int main()
6 {
7 double n1, n2, n3, n4, n5;
8 cout << "Please enter five numbers: ";
9 // Allow the user to enter in the five values.
10 cin >> n1 >> n2 >> n3 >> n4 >> n5;
11 cout << "The average of " << n1 << ", " << n2 << ", "
12 << n3 << ", " << n4 << ", " << n5 << " is "
13 << (n1 + n2 + n3 + n4 + n5)/5 << endl;;
14 }

A sample run of Listing 11.1 (averagenumbers.cpp) looks like:

Please enter five numbers: 9 3.5 0.2 100 15.3

The average of 9.0, 3.5, 0.2, 100.0, 15.3 is 25.6

The program conveniently displays the values the user entered and then computes and displays their aver-
age.
Suppose the number of values to average must increase from five to 25. If we use Listing 11.1
(averagenumbers.cpp) as a guide, twenty additional variables must be introduced, and the overall length of
the program will necessarily grow. Averaging 1,000 numbers using this approach is impractical.
Listing 11.2 (averagenumbers2.cpp) provides an alternative approach for averaging numbers.

©2013 Richard L. Halterman Draft date: January 27, 2013

 270

Listing 11.2: averagenumbers2.cpp

1 #include <iostream>
2
3 using namespace std;
4
5 int main()
6 {
7 double sum = 0.0, num;
8 const int NUMBER_OF_ENTRIES = 5;
9 cout << "Please enter " << NUMBER_OF_ENTRIES
10 << " numbers: ";
11 for ( int i = 0; i < NUMBER_OF_ENTRIES; i++ )
12 {
13 cin >> num;
14 sum += num;
15 }
16 cout << "The average of " << NUMBER_OF_ENTRIES
17 << " values is " << sum/NUMBER_OF_ENTRIES << endl;
18 }

Listing 11.2 (averagenumbers2.cpp) behaves slightly differently from Listing 11.1 (averagenumbers.cpp),
as the following sample run using the same data shows:

Please enter five numbers: 9 3.5 0.2 100 15.3

The average of the five values is 25.6

Listing 11.2 (averagenumbers2.cpp) can be modified to average 25 values much more easily than List-
ing 11.1 (averagenumbers.cpp) that must use 25 separate variables—just change the constant NUMBER_OF_ENTRIES.
In fact, the coding change to average 1,000 numbers is no more difficult. However, unlike the original aver-
age program, this new version does not display the numbers entered. This is a significant difference; it may
be necessary to retain all the values entered for various reasons:

• All the values can be redisplayed after entry so the user can visually verify their correctness.
• The values may need to be displayed in some creative way; for example, they may be placed in a
graphical user interface component, like a visual grid (spreadsheet).
• The values entered may need to be processed in a different way after they are all entered; for example,
we may wish to display just the values entered above a certain value (like greater than zero), but the
limit is not determined until after all the numbers are entered.

In all of these situations we must retain the values of all the variables for future recall.
We need to combine the advantages of both of the above programs; specifically we want

• the ability to retain individual values, and

• the ability to dispense with creating individual variables to store all the individual values

These may seem like contradictory requirements, but C++ provides a standard data structure that simultane-
ously provides both of these advantages—the array.

©2013 Richard L. Halterman Draft date: January 27, 2013

11.1. DECLARING AND USING ARRAYS 271

11.1 Declaring and Using Arrays

An array in C++ is a variable that refers to a block of memory that can hold multiple values simultaneously;
an array, therefore, represents a collection of values. An array has a name, and the values it contains are
accessed via their position within the block of memory designated for the array. An array stores a sequence
of values, and the values must all be of the same type. A collection of values all of the same type is said to
be homogeneous.
Like any other variable, an array can be local or global, and it must be declared before it is used. The
following code fragment declares an array variable named list that is assigned to a block of memory that
can hold 25 integer values:
int list[25];

At the point of an array’s declaration the square brackets indicate the amount of space reserved for the
array’s elements. The compiler generates machine language code that that ensures space for 25 integers
will be reserved at run time. On many systems integers are four bytes each, in which case an array of 25
integers would occupy 100 bytes of memory when the program executes. The exact amount of memory
required depends on the size of an integer for a particular system.
If list is declared within a function, the array’s contents are undetermined. This means that even
though space is set aside for 25 elements, the values of those individual elements are unspecified. If instead
the array is declared as a global variable, list will contain initially 25 zeros.
Once we declare an array, we can access its elements using the square bracket operator:
int list[3]; // Declare list to be an array of three ints
list[0] = 5 ; // Make the first element 5
list[1] = -3 ; // Make the second element -3
list[2] = 12 ; // Make the last element 12

This code fragment shows that the square brackets used after the point of declaration allow us to access an
individual element based on that element’s position within the block of memory assigned to the array. The
number within the square brackets indicates the distance from the beginning of the array. The expression
list[0] therefore indicates the element at the very beginning (a distance of zero from the beginning),
and list[1] is the second element (a distance of one away from the beginning).
After executing these assignment statements, the list array conceptually looks like Figure 11.1.
Once an array has been properly declared it can be used within a program. The square brackets, [], are
classified within C++ as a binary operator, since it requires two operands: an array name and an index. An
array’s elements can be accessed with the [] operator, as shown in the following code fragment:
int list[3]; // Declare list to be an array of three ints
list[0] = 5 ; // Make the first element 5
list[1] = -3 ; // Make the second element -3
list[2] = 12 ; // Make the last element 12
cout << list[1]; // Print element in position 1
list[1] = 7; // Assign 7 to element in position 1
cout << list[1]; // Print element in position 1

The [] operator is used to specify a position within an array. The array’s name in conjunction with []
operator represents an element in the array.
In Section 10.6 we saw an assignment statement in which the left operand of the assignment operator
involved more than just a single, simple variable. Observe that a statement such as

©2013 Richard L. Halterman Draft date: January 27, 2013

11.1. DECLARING AND USING ARRAYS 272

int list[3];
list[0] = 5;
list[1] = -3;
list[2] = 12;

list
5 -3 12
0 1 2

Figure 11.1: A simple array with three elements

list[1] = -3 ; // Make the second element -3

is a similar situation. The expression to the left of the assignment operator is not a simple, single variable
name. This is legal because the expression to the left of the assignment operator refers to a memory location
that can store a value. The memory location it identifies is a part of the memory reserved for the list array.

The array square brackets operator has two distinct meanings depending on the
context:
1. At the array’s declaration, for example,
double nums[10];

the number within the square brackets specifies the number of elements that
the array can hold. The compiler uses this number along with the type of
the array to determine how much memory to allocate for the array.

2. After the array’s declaration, for example,

nums[3] = 10.45;

the number within the square brackets represents an index locating a specific
element within the memory allocated to the array.

Declarations of arrays and non-array values can be interspersed, as in:

int list[5], x, s[3], *p, reg;

Here, list and s are integer arrays, but x and reg are simple integer variables, and p is a pointer to an
integer (see Section 10.6, for a discussion of pointers).
The contents of an array may be initialized when it is declared. The following code fragment declares
three arrays and specifies their initial contents:

©2013 Richard L. Halterman Draft date: January 27, 2013

11.1. DECLARING AND USING ARRAYS 273

int list[] = { 2, 4, 6, 8 };
double collection[10] = { 1.0, 3.5, 0.5, 7.2 };
char letters[] = { 'a', 'b', 'c' };

The elements of the array are provided by a comma-separated list within curly braces. This comma sepa-
rated list of initial values is called an array initialization list. When an array initialization list is provided,
the square brackets can be left empty as the compiler can calculate the array’s size based on the number of
elements in the list. If a number is provided within the square brackets, the array is allocated with the spec-
ified size, and the first positions within the array are filled with the elements specified in the initialization
list.
The code fragment above declares three arrays: list, collection, and letters. The list array
is created to hold four integer values consisting of 2, 4, 6, and 8. The collection array is created to hold
10 double-precision floating-point values, and the first four elements consist of 1.0, 3.5, 0.5, and 7.2. The
letters array holds the three characters shown. Figure 11.2 visualizes the contents of the three arrays.

int list[] = { 2, 4, 6, 8 };
list
2 4 6 8
0 1 2 3

double collection[10] = { 1.0, 3.5, 0.5, 7.2 };

collection
1.0 3.5 0.5 7.2 ? ? ? ? ? ?
0 1 2 3 4 5 6 7 8 9

char letters[] = { 'a', 'b', 'c' };

letters
‘a’ ‘b’ ‘c’
0 1 2

Figure 11.2: Array declarations and memory layout

We can observe three key points from Figure 11.2:

• An array name represents the location of the first element in an array. For an array named list, the
expressions list and &list[0] within C++ source code are equivalent.

• Elements in an array are stored in a contiguous block of memory. This means, for example, the
memory occupied by the element at index 3 follows immediately after the memory occupied by the

©2013 Richard L. Halterman Draft date: January 27, 2013

11.1. DECLARING AND USING ARRAYS 274

element at index 2 and immediately before the element at index 4. The size specified in the array’s
declaration determines the size this block of contiguous memory.
• Elements in an array are located by a numeric index. The first element is at index zero, not one.

The array initialization list may be used only when an array is declared. It cannot be used to assign or
reassign an array after the array has been declared; for example, the code
int nums[10];
nums = { 1, 2, 3, 4, 5, 6, 7, 8, 9, 10 }; // Illegal, will not compile

is illegal, and the compiler will report an error. The second statement is a simple assignment statement,
not a declaration and initialization statement. An array variable may not appear all by itself on the left
side of the assignment operator. An array, therefore, may be modified after its point of declaration only by
reassigning individual elements using the [] operator.
In an expression such as
list[3]

the expression within the square brackets is called an index or subscript. The subscript terminology is
borrowed from mathematicians who use subscripts to reference elements in a mathematical vector (for
example, V2 represents the second element in vector V). Unlike the convention often used in mathematics,
however, the first element in an array is at position zero, not one. The expression list[2] can be read
aloud as “list sub two.” The index indicates the distance from the beginning; thus, the very first element is
at a distance of zero from the beginning of the array. The first element of array a is a[0]. As a consequence
of a zero beginning index, if array a holds n elements, the last element in a is a[n − 1], not a[n].
To assign a specific element in an array you must specify the element’s position. The following code
fragment
double nums[10];
nums[0] = 0.5;
nums[3] = 5.3;
nums[9] = 0.2;

declares an array of doubles that initially holds 10 elements. The first element is set to 0.5, the item at
index 3 (that is, the fourth item) is assigned 5.3, and the last element in the array is 0.2. The values of the
other elements in the array nums are unspecified. Figure 11.3 illustrates this code sequence.
The elements of an array can be treated as any other variable; for example,
// Print the fourth element
cout << nums[3] << endl;
// The third element is the average of two other elements
nums[2] = (nums[0] + nums[9])/2;
// Get elements at indices 1 and 4 from the user
cin >> nums[1] >> nums[4];

In C++ source code an array name references a block of memory, not any individual element. The
following code fragment:
int nums[] = { 2, 4, 6, 8, 10 };
cout << nums << endl;

is unable to print the elements of the array. Instead, on most systems it prints a cryptic string like

©2013 Richard L. Halterman Draft date: January 27, 2013

11.1. DECLARING AND USING ARRAYS 275

double nums[10];
nums
? ? ? ? ? ? ? ? ? ?
0 1 2 3 4 5 6 7 8 9

nums[0] = 0.5;
nums
0.5 ? ? ? ? ? ? ? ? ?
0 1 2 3 4 5 6 7 8 9

nums[3] = 5.3;
nums
0.5 ? ? 5.3 ? ? ? ? ? ?
0 1 2 3 4 5 6 7 8 9

nums[9] = 0.2;
nums
0.5 ? ? 5.3 ? ? ? ? ? 0.2
0 1 2 3 4 5 6 7 8 9

Figure 11.3: Array element assignment

0036FA54

This is a hexadecimal (base 16) number representing a memory address—the actual beginning address of
the block of memory assigned to the array.
The expression within the [] operator must evaluate to an integer; some examples include

• an integer literal: a[34]

• an integer variable: a[x]

• an integer arithmetic expression: a[x + 3]

• an integer result of a function call that returns an int: a[max(x, y)]

©2013 Richard L. Halterman Draft date: January 27, 2013

 11.1. DECLARING AND USING ARRAYS 276

• an element of an integer array: a[b[3]]

The action of moving through an array visiting each element is known as traversal. for loops are ideal
for array traversals. If a is an integer array containing 10 elements, the following loop prints each element
in a:
for ( int i = 0; i < 10; i++ )
cout << a[i] << endl;

The loop control variable, i, steps through each valid index of array a. i starts at 0 and ends at 9, the last
valid position in array a.
The following loop prints contents of array a in reverse order:
for ( int i = 9; i >= 0; i-- )
cout << a[i] << endl;

The following code produces an array named set containing the integer sequence 0, 1, 2, 3, 4, 5, 6, 7, 8, 9:
int set[10];
for ( int i = 0; i < 10; i++ )
set[i] = i;

We now have all the tools we need to build a program that flexibly averages numbers while retaining
all the values entered. Listing 11.3 (arrayaverage.cpp) uses an array and a loop to achieve the generality of
Listing 11.2 (averagenumbers2.cpp) with the ability to retain all input for later redisplay.

Listing 11.3: arrayaverage.cpp

1 #include <iostream>
2
3 using namespace std;
4
5 int main()
6 {
7 double sum = 0.0;
8 const int NUMBER_OF_ENTRIES = 5;
9 double numbers[NUMBER_OF_ENTRIES];
10
11 cout << "Please enter " << NUMBER_OF_ENTRIES
12 << " numbers: ";
13 // Allow the user to enter in the values.
14 for ( int i = 0; i < NUMBER_OF_ENTRIES; i++ )
15 {
16 cin >> numbers[i];
17 sum += numbers[i];
18 }
19 cout << "The average of ";
20 for ( int i = 0; i < NUMBER_OF_ENTRIES - 1; i++ )
21 cout << numbers[i] << ", ";
22 // No comma following last element
23 cout << numbers[NUMBER_OF_ENTRIES - 1] << " is "
24 << sum/NUMBER_OF_ENTRIES << endl;
25 }

©2013 Richard L. Halterman Draft date: January 27, 2013

 11.1. DECLARING AND USING ARRAYS 277

The output of Listing 11.3 (arrayaverage.cpp) is similar to the original Listing 11.1 (averagenumbers.cpp)
program:

Please enter 5 numbers: 9 3.5 0.2 100 15.3

The average of 9.0, 3.5, 0.2, 100.0, 15.3 is 25.6

Unlike the original program, however, we now conveniently can extend this program to handle as many
values as we wish. We need only change the definition of the NUMBER_OF_ENTRIES constant to allow the
program to handle any number of values. This centralization of the definition of the array’s size eliminates
duplicating a hard-coded value and leads to a program that is more maintainable. Suppose every occurrence
of NUMBER_OF_ENTRIES were replaced with the literal value 5. The program would work exactly the
same way, but changing the size would require touching many places within the program. When duplicate
information is scattered throughout a program, it is a common error to update some but not all of the
information when a change is to be made. If all of the duplicate information is not updated to agree,
the inconsistencies result in errors within the program. By faithfully using the NUMBER_OF_ENTRIES
constant throughout the program instead of the literal numeric value, the problems with inconsistency can
be avoided.
The first loop in Listing 11.3 (arrayaverage.cpp) collects all five input values from the user. The second
loop only prints the first four because it also prints a trailing comma after each element. Since no comma
should be displayed after the last element, the last element is printed outside the loop.
The compiler will insist that the programmer use an integral value for an index, but the programmer
must ensure that the index used is within the bounds of the array. Since the index may consist of an
arbitrary integer expression whose value cannot be determined until run time, the compiler cannot check
for out-of-bound array accesses; for example, in the code
int x;
int v[10]; // Make an array with 10 spaces available
cin >> x; // User enters x at run time
v[x] = 1; // Is this OK? What is x?

the compiler cannot predict what number the user will enter. This means that misuse of an array index can
lead to run-time errors. To illustrate the problem, consider Listing 11.4 (arrayoutofbounds.cpp).

Listing 11.4: arrayoutofbounds.cpp

1 #include <iostream>
2 using namespace std;
3
4 int main()
5 {
6 const unsigned SIZE = 3;
7 int a[SIZE] = { 5, 5, 5 };
8 // Print the contents of the array
9 cout << "a contains ";
10 for ( int i = 0; i < SIZE; i++ )
11 cout << a[i] << " ";
12 cout << endl;
13 // Change all the 5s in array a to 8s
14 for ( int i = 0; i <= SIZE; i++ ) // Bug: <= should be <
15 a[i] = 8;

©2013 Richard L. Halterman Draft date: January 27, 2013

 11.1. DECLARING AND USING ARRAYS 278

16 // Reprint the contents of the array

17 cout << "a contains ";
18 for ( int i = 0; i < SIZE; i++ )
19 cout << a[i] << " ";
20 cout << endl;
21 }

Listing 11.4 (arrayoutofbounds.cpp) contains a logic error; the reassignment loop goes one past the
end of array a. Attempting to access elements outside the bounds of an array produces what is known
as undefined behavior. The C++ language standard uses this term to indicate a program’s behavior is not
specified, and complier writers are free to do whatever they want. Some possible behaviors include:

• The compiled code could check to ensure the index is within range and produce an intelligible run-
time error message,
• the program could simply crash mysteriously,
• the program could continue with no indication of a problem and appear to behave correctly most of
the time,

or any number of other possible outcomes. In other words, the actual program behavior is system dependent
and compiler dependent. The code within a program that exhibits undefined behavior is considered a logic
error, since its action is inconsistent across platforms and compilers. Simply said, the program’s behavior
is unpredictable. Unpredictable behavior is incorrect behavior.
In most cases, an out-of-bounds access simply accesses memory outside the array. If this includes
memory that does not belong to the executing program, modern operating systems will kill the program and
produce some error message. Under Visual C++ when the program is built as a debug version, the program
prints the contents of the array the first time but crashes before it can print it out a second time. Microsoft
Windows then displays the dialog box shown in Figure 11.4. The program running under Linux or OS X

Figure 11.4: Memory access error under Visual C++

may simply print in the console:

©2013 Richard L. Halterman Draft date: January 27, 2013

 11.2. ARRAYS AND FUNCTIONS 279

Segmentation fault

If your program is using an array, and it terminates with such a message, you should check its source code
carefully for places where out-of-bounds array accesses are possible.
The following code fragment shows some proper and improper array accesses:
int numbers[10]; // Declare the array
numbers[0] = 5; // Put value 5 first
numbers[9] = 2; // Put value 2 last
numbers[-1] = 5; // Out of bounds; first valid index is 0
numbers[10] = 5; // Out of bounds; last valid index is 9
numbers[1.3] = 5; // Compile error, 1.3 is not an int

Notice that the compiler does check that the type of the index is a valid integer expression.
In an array traversal with a for loop such as
for ( int i = 0; i < SIZE; i++ )
cout << a[i] << endl;

it is easy to ensure that i cannot fall outside the bounds of array a, but arbitrary index values should be
checked before their use:
int x;
int a[10]; // Make an array with 10 spaces available
cin >> x; // User enters x at run time
// Ensure x is within the bounds of the array
if ( 0 <= x && x < 10 )
a[x] = 1; // This is safe now
else
cout << "Index provided is out of range" << endl;

Here, the if statement ensures the array access is within the array’s bounds.

11.2 Arrays and Functions

An array can be passed to a function, as shown in Listing 11.5 (arraytofunc.cpp)

Listing 11.5: arraytofunc.cpp

1 #include <iostream>
2
3 using namespace std;
4
5 /*
6 * print(a, n)
7 * Prints the contents of an int array
8 * a is the array to print
9 * n is the size of the array
10 * /

©2013 Richard L. Halterman Draft date: January 27, 2013

 11.2. ARRAYS AND FUNCTIONS 280

11 void print(int a[], int n)

12 {
13 for ( int i = 0; i < n; i++ )
14 cout << a[i] << " ";
15 cout << endl;
16 }
17
18 /*
19 * sum(a, n)
20 * Adds up the contents of an int array
21 * a is the array to sum
22 * n is the size of the array
23 */
24 int sum(int a[], int n)
25 {
26 int result = 0;
27 for ( int i = 0; i < n; i++ )
28 result += a[i];
29 return result;
30 }
31
32 int main()
33 {
34 int list[] = { 2, 4, 6, 8 };
35 // Print the contents of the array
36 print(list, 4);
37 // Compute and display sum
38 cout << sum(list, 4) << endl;
39 // Zero out all the elements of list
40 for ( int i = 0; i < 4; i++ )
41 list[i] = 0;
42 // Reprint the contents of the array
43 print(list, 4);
44 // Compute and display sum
45 cout << sum(list, 4) << endl;
46 }

As shown in the print function’s definition:

void print(int a[], int n)

when an array formal parameter is declared, its square brackets are left empty. The programmer can supply
a number within the square brackets, but it is ignored by the compiler. When calling a function that accepts
an array parameter, as in
print(list, 4);

the programmer must pass the array’s size along with the array name. This is because an array simply
references a block of memory and has has no notion of its size.
An undecorated array name by itself in C++ source code behaves like a constant pointer to the beginning
element of the array. Consequently, when an array is passed as an actual parameter during a function call,
as in Listing 11.5 (arraytofunc.cpp):
print(list, 4);

©2013 Richard L. Halterman Draft date: January 27, 2013

 11.2. ARRAYS AND FUNCTIONS 281

the function is passed the address of the array—the array’s address is bound to the formal parameter. So,
while the function cannot affect the address of the array itself, the function has total access to the array’s
contents. In Listing 11.6 (cleararray.cpp), the function clear modifies the contents of any array sent to it,
making all the elements zero.

Listing 11.6: cleararray.cpp

1 #include <iostream>
2
3 using namespace std;
4
5 /*
6 * print(a, n)
7 * Prints the contents of an int array
8 * a is the array to print
9 * n is the size of the array
10 */
11 void print(int a[], int n)
12 {
13 for ( int i = 0; i < n; i++ )
14 cout << a[i] << " ";
15 cout << endl;
16 }
17
18 /*
19 * clear(a, n)
20 * Makes all the elements of array a zero
21 * a is the array to zero out
22 * n is the size of the array
23 */
24 void clear(int a[], int n)
25 {
26 for ( int i = 0; i < n; i++ )
27 a[i] = 0;
28 }
29
30 int main()
31 {
32 int list[] = { 2, 4, 6, 8 };
33 // Print the contents of the array
34 print(list, 4);
35 // Zero out the array
36 clear(list, 4);
37 // Reprint the contents of the array
38 print(list, 4);
39 }

The output of Listing 11.6 (cleararray.cpp) is

2 4 6 8
0 0 0 0

©2013 Richard L. Halterman Draft date: January 27, 2013

 11.3. PRIME GENERATION WITH AN ARRAY 282

The clear function actually modifies the contents of main’s list array. The function works on the
array’s actual elements, not copies of its elements.
Arrays, therefore, by default are passed by reference with respect to the elements they contain. By
default, an array’s contents are open to corruption by errant functions. In order to protect an array so that a
function may read its contents but not modify its contents, the parameter must be declared const, as in
int sum(const int a[], int n)

In Listing 11.5 (arraytofunc.cpp) there is no reason why the print and sum functions should be able to to
modify the contents of their array parameters, so the const specifier should be added to their definitions.
As a general rule, all functions that accept array parameters should declare the array contents as const
unless they need to modify the elements of the array.

As a general rule, all functions that accept array parameters should declare the
array contents as const unless they need to modify the elements of the array.
For example, use
int sum(const int a[], int n)
rather than
int sum(int a[], int n)
if the sum function must be able to examine the contents of the array but is not
intended to modify its contents.

C++ does not directly support empty arrays. A physical array must contain at least one element. Some-
times it is appropriate to consider an array that is conceptually empty; for example, what if we wish to
pass an “empty” array to the sum function? We can pass 0 as the second parameter since an empty array
contains no elements, but what should we pass as the first parameter? Any array will do, but there is no
need create an real array when none truly is needed. Instead, we may use nullptr:
quanitity = sum(nullptr, 0);

If you scrutinize the body of the sum function, you will see that this call will assign to quantity the
correct answer, namely zero.

11.3 Prime Generation with an Array

Listing 11.7 (fasterprimes.cpp) uses an algorithm developed by the Greek mathematician Eratosthenes who
lived from 274 B.C. to 195 B.C. The principle behind the algorithm is simple: Make a list of all the integers
two and larger. Two is a prime number, but any multiple of two cannot be a prime number (since a multiple
of two has two as a factor). Go through the rest of the list and mark out all multiples of two (4, 6, 8, ...).
Move to the next number in the list (in this case, three). If it is not marked out, it must be prime, so go
through the rest of the list and mark out all multiples of that number (6, 9, 12, ...). Continue this process
until you have listed all the primes you want.

Listing 11.7: fasterprimes.cpp

1 // File primesieve.cpp

©2013 Richard L. Halterman Draft date: January 27, 2013

 11.3. PRIME GENERATION WITH AN ARRAY 283

2
3 #include <iostream>
4 #include <ctime>
5
6 using namespace std;
7
8 // Display the prime numbers between 2 and 500,000 and
9 // time how long it takes
10
11 // Largest potential prime considered
12 //const int MAX = 500000;
13 const int MAX = 500;
14
15 // Each position in the Boolean array indicates
16 // if the number of that position is not prime:
17 // false means "prime," and true means "composite."
18 // Initially all numbers are prime until proven otherwise
19 bool nonprimes[MAX]; // Global array initialized to all zeros
20
21 int main()
22 {
23 clock_t start_time = clock(); // Record start time
24
25 // First prime number is 2; 0 and 1 are not prime
26 nonprimes[0] = nonprimes[1] = true;
27
28 // Start at the first prime number, 2.
29 for ( int i = 2; i < MAX; i++ )
30 {
31 // See if i is prime
32 if ( !nonprimes[i] )
33 {
34 cout << i << " ";
35 // It is prime, so eliminate all of its
36 // multiples that cannot be prime
37 for ( int j = 2*i; j < MAX; j += i )
38 nonprimes[j] = true;
39 }
40 }
41 cout << endl; // Move cursor down to next line
42 // Print the elapsed time
43 cout << "Elapsed time: "
44 << static_cast<double>(clock() - start_time)/CLOCKS_PER_SEC
45 << " seconds" << endl;
46 }

Recall Listing 8.8 (measureprimespeed.cpp), which also prints all the prime numbers up to 500,000.
Using redirection (see Section 8.4), Listing 8.8 (measureprimespeed.cpp) takes 77 seconds. In comparison,
Listing 11.7 (fasterprimes.cpp) takes about one second to perform the same task on the system. This is
comparable to the square root version, Listing 8.5 (moreefficientprimes.cpp), which takes two seconds to
run. If the goal is to print the prime numbers up to 1,000,000, the original version averages 271 seconds,
or about four and one-half minutes. The square root version averages 4.5 seconds. The new, array-based
version averages 2 seconds. For 5,000,000 values the unoptimized original version takes a little over an
hour and 33 minutes, the square root version takes a respectable 39 seconds, but array version averages

©2013 Richard L. Halterman Draft date: January 27, 2013

11.4. MULTIDIMENSIONAL ARRAYS 284

only 9 seconds to run.

11.4 Multidimensional Arrays

The arrays we have seen thus far have been one dimensional—simple sequences of values. C++ supports
higher-dimensional arrays. A two-dimensional array is a block of memory that is best visualized as a table
with rows and columns. The statement
int a[2][3];

declares a to be a two-dimensional (2D) array of integers. A 2D array is sometimes called a matrix. In

this case, the declaration specifies that array a contains two rows and three columns. Figure 11.5 shows the
logical structure of the array created by the following sequence of code:
int a[2][3]; // a is a 2D array
a[0][0] = 5;
a[0][1] = 19;
a[0][2] = 3;
a[1][0] = 22;
a[1][1] = -8;
a[1][2] = 10;

a
0 5 19 3

1 22 -8 10
0 1 2

Figure 11.5: A 2 × 3 two-dimensional array

The two-dimensional array a is said to be a 2 × 3 array, meaning it has two rows and three columns (as
shown in Figure 11.5). Rows are arranged horizontally, and the values in columns are arranged vertically.
In each of the assignment statements above, for example
a[1][0] = 22;

the first index (here 1) signifies the row and the second index (here 0) denotes the column of the element
within the array.
Using a syntax similar to the initialization lists of one-dimensional arrays, the 2D array a from above
could have been declared and created as:

©2013 Richard L. Halterman Draft date: January 27, 2013

11.4. MULTIDIMENSIONAL ARRAYS 285

int a[2][3] = { { 5, 19, 3 },

{ 22, -8, 10 } };

Note that each row appears within its own set of curly braces, and each row looks like 1D array initialization
list. You may omit the first index, as shown here:
int a[][3] = { { 5, 19, 3 },
{ 22, -8, 10 } };

For 2D arrays initialized in this manner the first subscript is optional, but the second subscript (that is, the
size of each column) is required.
To access an element of a 2D array, use two subscripts:
a[r][c] = 4; // Assign element at row r, column c
cout << a[m][n] << endl; // Display element at row m, column n

The following function prints the contents of a ROWS × COLUMNS 2D array of doubles:
void print(const double m[ROWS][COLUMNS])
{
for ( int row = 0; row < ROWS; row++ )
{
for ( int col = 0; col < COLUMNS; col++ )
cout << setw(5) << m[row][col];
cout << endl;
}
}

The declaration of the parameter m is quite complicated. We can simplify the syntax by using a C++
typedef statement. The statement
typedef double Matrix[ROWS][COLUMNS];

defines a new type named Matrix. If you omit the word typedef from the above statement, it becomes a
declaration of a 2D matrix named Matrix. The typedef keyword changes the meaning of the statement
so that instead of declaring a variable it defines a new type. typedef statements usually appear near the
top of a source file and most often have global scope. It is uncommon to see a local typedef statement
within a function body.
Given the typedef statement defining the new type name Matrix, the parameter for the print
function may be expressed more simply:
void print(const Matrix m)
{
for ( int row = 0; row < ROWS; row++ )
{
for ( int col = 0; col < COLUMNS; col++ )
cout << setw(5) << m[row][col];
cout << endl;
}
}

Listing 11.8 (twodimarray.cpp) experiments with 2D arrays and takes advantage of the typedef statement
to simplify the code.

©2013 Richard L. Halterman Draft date: January 27, 2013

 11.4. MULTIDIMENSIONAL ARRAYS 286

Listing 11.8: twodimarray.cpp

1 #include <iostream>
2 #include <iomanip>
3
4 using namespace std;
5
6 const int ROWS = 3,
7 COLUMNS = 5;
8
9 // The name Matrix represents a new type
10 // that means a ROWS x COLUMNS
11 // two-dimensional array of double-precision
12 // floating-point numbers.
13 typedef double Matrix[ROWS][COLUMNS];
14
15 // Allow the user to enter the elements of a matrix
16 void populate_matrix(Matrix m)
17 {
18 cout << "Enter the " << ROWS << " rows of the matrix. " << endl;
19 for ( int row = 0; row < ROWS; row++ )
20 {
21 cout << "Row #" << row << " (enter " << COLUMNS << " values):";
22 for ( int col = 0; col < COLUMNS; col++ )
23 cin >> m[row][col];
24 }
25 }
26
27 void print_matrix(const Matrix m)
28 {
29 for ( int row = 0; row < ROWS; row++ )
30 {
31 for ( int col = 0; col < COLUMNS; col++ )
32 cout << setw(5) << m[row][col];
33 cout << endl;
34 }
35 }
36
37 int main()
38 {
39 // Declare the 2D array
40 Matrix mat;
41 // Populate the array
42 populate_matrix(mat);
43 // Print the array
44 print_matrix(mat);
45 }

An expression that uses just one index with a 2D array represents a single row within the 2D array. This
row is itself a 1D array. Thus, if a is a 2D array and i is an integer, then the expression

a[i]

is a 1D array representing row i.

©2013 Richard L. Halterman Draft date: January 27, 2013

 11.5. POINTERS AND ARRAYS 287

C++ allows arrays with dimensions higher than two. Each “slice” of a 3D array is simply a 2D array, a
4D array is an array of 3D arrays, etc. For example, the statement
matrix[x][y][z][t] = 1.0034;

assigns 1.0034 to an element in a 4D array of doubles. In practice, arrays with more than two dimensions
are rare, but advanced scientific and engineering applications sometimes require higher-dimensional arrays.

11.5 Pointers and Arrays

An array name used in C++ source code references a location in memory, the address of the first element
(element at index 0) in the array. In this way an array name is similar to a constant pointer (see Section 10.6
for more information about C++ pointers). Because of this, an array identifier may be treated in some ways
like a pointer. Similarly and more commonly, a pointer can be directed to point to an array, and then the
pointer itself can be treated as if it were an array.
Listing 11.9 (pointerprint.cpp) uses a pointer to traverse an array.

Listing 11.9: pointerprint.cpp

1 #include <iostream>
2
3 using namespace std;
4
5 int main()
6 {
7 int a[] = { 2, 4, 6, 8, 10, 12, 14, 16, 18, 20 },
8 *p;
9
10 p = &a[0]; // p points to first element of array a
11
12 // Print out the contents of the array
13 for ( int i = 0; i < 10; i++ )
14 {
15 cout << *p << ' '; // Print the element p points to
16 p++; // Increment p so it points to the next element
17 }
18 cout << endl;
19 }

The statement
p = &a[0];

sets p to point to the first element of array a. A shorter way to accomplish the same thing is
p = a;

since a is itself a reference to the array’s location in memory. This assignment statement clearly demon-
strates the association between array variables and pointer variables. Note that the opposite assignment
(a = p) is impossible, because array a declared as above may not appear by itself on the left side of the
assignment operator.
Pointer variables can participate in addition and subtraction expressions. The statement

©2013 Richard L. Halterman Draft date: January 27, 2013

11.5. POINTERS AND ARRAYS 288

p++;

changes the address stored in p so subsequently the pointer will point to the next integer position in memory.
If p is assigned to array a, incrementing p redirects it to point to a’s next element. 1 In Figure 11.6, pointer
p is assigned to array a and then incremented to refer to various elements within a.

int a[] = { 2, 4, 6, 8 }, *p
a
2 4 6 8

p 0 1 2 3

p = a;
a
2 4 6 8

p 0 1 2 3

p++;
a
2 4 6 8

p 0 1 2 3

*p = 7;
a
2 7 6 8

p 0 1 2 3

Figure 11.6: Incrementing a pointer to an array

The expression
p[0]

is another way to write

1 Each byte in memory has a unique numeric address. Since most C++ types require more than one byte of storage, incrementing

a pointer by 1 does not simply add 1 to the address it holds. The amount added depends on the type of the pointer; for example, on
systems using 32-bit (4 byte) integers, adding 1 to an integer pointer variable increments its address by 4, not 1. The compiler knows
the type of the pointer variable because programmers must declare all variables. The compiler, therefore, can automatically adjust
the arithmetic to work properly. If a pointer points to memory within an array and is of the same type as the array, incrementing the
pointer correctly repositions the pointer to point to the next element in the array.

©2013 Richard L. Halterman Draft date: January 27, 2013

 11.5. POINTERS AND ARRAYS 289

*p

so the array indexing operator ([]) can be used with pointers. The expression
p[5]

represents the element at index 5 within the array pointed to by p.

Listing 11.10 (pointerprint2.cpp) uses pointers in a different way to traverse an array.

Listing 11.10: pointerprint2.cpp

1 #include <iostream>
2
3 using namespace std;
4
5 int main()
6 {
7 int a[] = { 2, 4, 6, 8, 10, 12, 14, 16, 18, 20 },
8 *begin, *end, *cursor;
9
10 begin = a; // begin points to the first element of array a
11 end = a + 10; // end points to just after the last element
12
13 // Print out the contents of the array
14 cursor = begin;
15 while ( cursor != end )
16 {
17 cout << *cursor << ' '; // Print the element
18 cursor++; // Increment cursor so it points to the next element
19 }
20 cout << endl;
21 }

In Listing 11.10 (pointerprint2.cpp), the code

cursor = begin;
while ( cursor != end )
{
cout << *cursor << ' '; // Print the element
cursor++; // Increment cursor so it points to the next element
}

can be expressed more succinctly as a for loop:

for ( cursor = begin; cursor != end; cursor++ )
cout << *cursor << ' '; // Print the element

If pointer p points to an array, the element at index i can be accessed as either p[i] or *(p + i).
The expression *(p + i) dereferences the address that is i positions away from the address referenced
by p.
Sometimes pointer notation is used to represent an array parameter to a function. The array print
function that begins
void print(const int a[], int n)

©2013 Richard L. Halterman Draft date: January 27, 2013

 11.5. POINTERS AND ARRAYS 290

could instead be written

void print(const int *a, int n)

where a is a pointer to an array. The compiler treats the two forms identically in the machine language it
produces.
Listing 11.11 (recursivearrayprint.cpp) uses two array printing functions, iterative_print and
recursive_print. Both use the pointer notation when declaring their array parameters.

Listing 11.11: recursivearrayprint.cpp

1 #include <iostream>
2
3 using namespace std;
4
5 /*
6 * iterative_print(a,n)
7 * Prints the contents of array a
8 * a is the array to print; a is not modified
9 * n is number of elements in a
10 */
11 void iterative_print(const int *a, int n)
12 {
13 for ( int i = 0; i < n; i++ )
14 cout << a[i] << ' ';
15 }
16
17 /*
18 * recursive_print(a,n)
19 * Prints the contents of array a
20 * a is the array to print; a is not modified
21 * n is number of elements in a
22 */
23 void recursive_print(const int *a, int n)
24 {
25 if ( n > 0 )
26 {
27 cout << *a << ' '; // Print the first element of the array
28 recursive_print(a + 1, n - 1); // Print rest of the array
29 }
30 }
31
32 int main()
33 {
34 int list[] = { 23, -3, 4, 215, 0, -3, 2, 23, 100, 88, -10 };
35 iterative_print(list, 11);
36 recursive_print(list, 11);
37 }

The function iterative_print uses a loop, while recursive_print uses recursion instead.
Inside the recursive_print function, the expression a + 1 points to the second element in array a.
a + 1 essentially represents the rest of the array—everything except for the first element. The statement
recursive_print(a + 1, n - 1);

©2013 Richard L. Halterman Draft date: January 27, 2013

11.6. ARRAY RANGES 291

calls the function recursively, passing the remainder of the array (a + 1) with a length that is one less
(n - 1) than before. The recursion terminates when the array’s length is zero.

11.6 Array Ranges

In Section 11.2 we saw one way that we can pass an array to a function. We noted that since an array
parameter is just the address of the array’s first element, callers must include additional information so that
the function knows the size of the array. Another approach passes the address of the array’s first element and
the address of the imaginary element that is just past the end of the array. This is known as the range-based
approach to passing array parameters.
Consider the array print function in Listing 11.5 (arraytofunc.cpp) in Section 11.2:
void print(int a[], int n)
{
for ( int i = 0; i < n; i++ )
cout << a[i] << " ";
cout << endl;
}

This print function requires the caller to provide the array and the array’s size, as in
int my_list[] = { 20, 30, 40, 50 };
print(my_list, 4);

A range-based version looks like

void print(int *begin, int *end)
{
for ( int *cursor = begin; cursor != end; cursor++ )
cout << *cursor << " ";
cout << endl;
}

In the range-based approach the caller passes the array (that is, the address of the element at index zero)
and the address of the imaginary element just past the end of the array:
int my_list[] = { 20, 30, 40, 50 };
print(my_list, my_list + 4);

Here we see the caller passes two pointers to this range-based print function:

• my_list points to the first element (at index 0) of the array, and
• the expression my_list + 4, because of pointer arithmetic (see Section 11.5), points to the address
just past the element at the back of the array.

The local variable in the range-based print function, cursor, is a pointer that the loop directs to each
element in the array. When cursor passes the last element in the array, the iterator ceases. This is because
the loop can continue only while the expression cursor != end is true.
The C++ standard library generalizes this range-based approach to algorithms that process more sophis-
ticated kinds of data structures.

©2013 Richard L. Halterman Draft date: January 27, 2013

11.7. C STRINGS 292

11.7 C Strings

A string is a sequence of characters. C and C++ implement strings as as arrays of char. The C++ language
additionally supports string objects (see Section 13.1). In the C language, the only option is a char array.
We use the term C string to refer to an array of characters as used in the C language. In this section, any
mention of the term string refers to a C string.
A string is an array of characters. A string literal is a sequence of characters enclosed within quotation
marks, as in
cout << "Howdy!" << endl;

All proper C strings are null terminated. This means the last character in the array is ASCII zero, which
in C++ is represented by the character literal '\0'. Figure 11.7 shows the physical layout of the string
"Howdy!" in memory.

Figure 11.7: Physical layout of a C string

Since strings are actually arrays, care must be taken when using string variables:

• Enough space must be reserved for number of characters in the string, including the null terminating
character.

• The array of characters must be properly null terminated.

The following code fragment is safe and acceptable:

char *word = "Howdy!";
cout << word << endl;

The variable word is declared to be a pointer to a character, and it is initialized to point to a string literal.
The following code fragment is less safe:
char word[256];
cin >> word;

word can hold 255 viable characters plus the null terminator. If the user types in relatively short words
(length less than 255 characters), there is no problem. If at any time the user types in more characters than
will fit in the word array, the executing program will have a problem. The problem is known as a buffer
overrun. In the best case, buffer overruns lead to buggy programs. In the worst case, clever users can exploit
buffer overruns to compromise software systems. Buffer overruns are always logic errors and you should
take great care to avoid them.
The following code provides a safe way to get user input:

©2013 Richard L. Halterman Draft date: January 27, 2013

 11.7. C STRINGS 293

char word[10];
fgets(word, 10, stdin);
cout << word << endl;

The fgets function is a standard C function. The second parameter specifies the maximum length of the
string, including the terminating null character, that will be placed in the string word. The last argument,
stdin is a C construct related to the C++ object cin. In order to use fgets within a program you must
include the <cstdio> header.
The following code begs for disaster:
char *word;
cin >> word;

In this case word points to a random location in memory, and no buffer has been allocated to receive the
input characters from cin. The program’s behavor executing this code is undefined, but it likley will lead to
the program crashing. Insidiously, depending on how the operating system manages memory, the program
may run fine much of the time and crash only rarely. Regardless, the program contains a serious bug.
When passing an array to a function a caller must provide the size of the array so that the function may
process the array properly. Since C strings are null terminated, such size information is not necessary. The
find_char function in Listing 11.12 (findchar.cpp) determines if a particular character is present in a
string:

Listing 11.12: findchar.cpp

1 #include <iostream>
2
3 using namespace std;
4
5 bool find_char(const char *s, char ch)
6 {
7 while ( *s != '\0' ) // Scan until we see the null character
8 {
9 if ( *s == ch )
10 return true; // Found the matching character
11 s++; // Advance to the next position within the string
12 }
13 return false; // Not found
14 }
15
16 int main()
17 {
18 char *phrase = "this is a phrase";
19 // Try all the characters a through z
20 for ( char ch = 'a'; ch <= 'z'; ch++ )
21 {
22 cout << '\'' << ch << '\''<< " is ";
23 if ( !find_char(phrase, ch) )
24 cout << "NOT ";
25 cout << "in " << '\"' << phrase << '\"' << endl;
26 }
27 }

The output of Listing 11.12 (findchar.cpp) is

©2013 Richard L. Halterman Draft date: January 27, 2013

11.7. C STRINGS 294

’a’ is in "this is a phrase"

’b’ is NOT in "this is a phrase"
’c’ is NOT in "this is a phrase"
’d’ is NOT in "this is a phrase"
’e’ is in "this is a phrase"
’f’ is NOT in "this is a phrase"
’g’ is NOT in "this is a phrase"
’h’ is in "this is a phrase"
’i’ is in "this is a phrase"
’j’ is NOT in "this is a phrase"
’k’ is NOT in "this is a phrase"
’l’ is NOT in "this is a phrase"
’m’ is NOT in "this is a phrase"
’n’ is NOT in "this is a phrase"
’o’ is NOT in "this is a phrase"
’p’ is in "this is a phrase"
’q’ is NOT in "this is a phrase"
’r’ is in "this is a phrase"
’s’ is in "this is a phrase"
’t’ is in "this is a phrase"
’u’ is NOT in "this is a phrase"
’v’ is NOT in "this is a phrase"
’w’ is NOT in "this is a phrase"
’x’ is NOT in "this is a phrase"
’y’ is NOT in "this is a phrase"
’z’ is NOT in "this is a phrase"

The find_char function in Listing 11.12 (findchar.cpp) uses pointer notation to traverse the string. It
does not need to know in advance the number of characters in the string because it starts at the beginning
and keeps scanning each character in turn until it finds the character it is looking for or encounters the null
terminating character.
Recall from Section 5.1 that for Boolean conditions C++ treats a zero value as false and any non-zero
value as true. Because of such a loose interpretation of Boolean expressions, the find_char function
above may be written more compactly as
char find_char(const char *s, char ch)
{
// Scan until we see the null character or the character
// we seek
while ( *s != '\0' && *s != ch )
s++; // Advance to the next position within the string
return *s; // Null character = false, any other is true
}

The only way out of the loop is to scan the null terminating character or the character sought. Here, if the
loop encounters the null terminating character, it exits and returns that null character. The null character
is simply ASCII zero—Boolean false. If the loop locates the sought character, it exits and returns that
character which will not be ASCII zero. Any character except the null character has an ASCII value greater
than zero; therefore, Boolean true.

©2013 Richard L. Halterman Draft date: January 27, 2013

11.8. DYNAMIC ARRAYS 295

Most routines that process strings depend on the strings to be null terminated in order to work properly.
Some standard C string functions include

• int strlen(const char *s) returns the number of characters in string s, not including the
null terminator.

• char *strcpy(char *s, const char *t) copies the contents of string t into string s up
to and including the null terminator; s must point to a buffer large enough to hold all the characters
of C string t.

• char *strncpy(char *s, const char *t, unsigned n) works like strcpy but copies
a maximum of n characters; s must point to a buffer that can hold at least n characters.

• int strcmp(const char *s, const char *t) compares two strings for lexicographic
(dictionary) ordering. The function returns an integer less than zero if s appears lexicographically
before t. The function returns an integer greater than zero if s appears lexicographically after t. The
function returns zero if the two strings are identical.
The following code fragment
cout << strcmp("ABC", "XYZ") << endl;
cout << strcmp("XYZ", "ABC") << endl;
cout << strcmp("ABC", "ABC") << endl;

prints

-1
1
0

• int strncmp(const char *s, const char *t, int n) compares the first n charac-
ters of two strings for lexicographic (dictionary) ordering. The function returns an integer less than
zero if the first n characters of s appear lexicographically before the first n characters of t; that is, s
would appear before t is a dictionary. The function returns an integer greater than zero if the first n
characters of s appear lexicographically after the first n characters of t. The function returns zero if
the first n characters of the two strings are identical.

You should be familiar with C strings because C++ code can use C libraries, and C libraries often use C
strings. Since C strings tend to be problematic, however, in most cases you should use the newer string
objects (see Section 13.1) whenever possible.

11.8 Dynamic Arrays

In all of the programs we have examined so far the size of the arrays were determined at compile time; for
example, in Listing 11.3 (arrayaverage.cpp), the program that allowed the user to enter a list of numbers to
average, the statements
const int NUMBER_OF_ENTRIES = 5;
double numbers[NUMBER_OF_ENTRIES];

©2013 Richard L. Halterman Draft date: January 27, 2013

 11.8. DYNAMIC ARRAYS 296

fix the program to accept exactly five numbers. The programmer may change the constant NUMBER_OF_ENTRIES,
but the program then must be recompiled. We would like a program that truly is flexible and is built to han-
dle any number of entries the user may need to process.
One approach to the flexibly-sized array is to define the array to hold as many items as conceivably will
be needed. Listing 11.13 (largearrayaverage.cpp) uses this approach.

Listing 11.13: largearrayaverage.cpp

1 #include <iostream>
2
3 using namespace std;
4
5 // Maximum number of expected values is 1,000,000
6 const int MAX_NUMBER_OF_ENTRIES = 1000000;
7 double numbers[MAX_NUMBER_OF_ENTRIES];
8
9 int main()
10 {
11 double sum = 0.0;
12 int size; // Actual number of entries
13
14 // Get effective size of the array
15 cout << "Please enter number of values to process: ";
16 cin >> size;
17
18 if ( size > 0 ) // Nothing to do with no entries
19 {
20 cout << "Please enter " << size << " numbers: ";
21 // Allow the user to enter in the values.
22 for ( int i = 0; i < size; i++ )
23 {
24 cin >> numbers[i];
25 sum += numbers[i];
26 }
27 cout << "The average of ";
28 for ( int i = 0; i < size - 1; i++ )
29 cout << numbers[i] << ", ";
30 // No comma following last element
31 cout << numbers[size - 1] << " is "
32 << sum/size << endl;
33 }
34 }

Listing 11.13 (largearrayaverage.cpp) creates an array that can hold one million entries. The variable
size keeps track of the actual size needed by the user. While the array’s maximum size is one million, the
part of the array actually used during a given execution may be much smaller.
Notice that the array numbers is a global variable and is not local to main as it was in Listing 11.3
(arrayaverage.cpp). This is because most systems limit the amount of storage available to local variables
within functions. Local variables are stored in an area of memory known as the stack. Local variables exist
in memory only when the function that uses them is invoked; therefore, the stack grows and shrinks as
functions execute and return. Global variables, on the other hand, exist in the computer’s memory for the
life of the program’s execution. Global variables are stored in what is known as static memory. There is
a limit to the amount of static memory available for global variables (the amount of memory in RAM or

©2013 Richard L. Halterman Draft date: January 27, 2013

 11.8. DYNAMIC ARRAYS 297

virtual memory on disk, if nothing else), but the global limit usually is much higher then the local limit.
One million double-precision floating-point numbers consumes 8,000,000 bytes (8 megabytes) in Visual
C++ and on many other systems. The default stack size for local variables under Visual C++ is only one
megabyte, although the stack size can be increased by adjusting the compiler and linker build options.
While the approach taken in Listing 11.13 (largearrayaverage.cpp) works, it wastes memory resources.
In modern computing, a user may have multiple programs open at the same time. If each program is tying
up the maximum amount of memory it may ever need, there may not be enough real memory (RAM) to
go around, and the computer will be forced to use more virtual memory (shuttling portions of the running
program’s memory to and from the disk drive). Virtual memory access greatly degrades the speed of a
program, and so the user’s experience suffers. It is important that each program uses its resources wisely.
Fortunately, there is a way for a programmer to create an array of the exact size needed, even if that size
is not known until run time. Listing 11.14 (flexiblearrayaverage.cpp) shows how.

Listing 11.14: flexiblearrayaverage.cpp

1 #include <iostream>
2
3 using namespace std;
4
5
6 int main()
7 {
8 double sum = 0.0;
9 double *numbers; // Note: numbers is a pointer, not an array
10 int size; // Actual number of entries
11
12 // Get effective size of the array
13 cout << "Please enter number of values to process: ";
14 cin >> size;
15
16 if ( size > 0 ) // Nothing to do with no entries
17 {
18 cout << "Please enter " << size << " numbers: ";
19 // Allocate the exact size needed
20 numbers = new double[size]; // Dynamically allocated array
21 // Allow the user to enter in the values.
22 for ( int i = 0; i < size; i++ )
23 {
24 cin >> numbers[i];
25 sum += numbers[i];
26 }
27 cout << "The average of ";
28 for ( int i = 0; i < size - 1; i++ )
29 cout << numbers[i] << ", ";
30 // No comma following last element
31 cout << numbers[size - 1] << " is "
32 << sum/size << endl;
33 delete [] numbers; // Free up the space held by numbers
34 }
35 }

Notice that numbers is not declared to be an array; it is a pointer:

double *numbers; // Note: numbers is a pointer, not an array

©2013 Richard L. Halterman Draft date: January 27, 2013

 11.8. DYNAMIC ARRAYS 298

This statement makes space for a single pointer, merely four bytes on a 32-bit system or eight bytes on a
64-bit system. The The expression
new double[size]

allocates at run time a block of memory for an array that can hold exactly size double-precision floating-
point elements. The value of this expression is the starting address of the newly allocated memory block.
The statement
numbers = new double[size]; // Dynamically allocated array

therefore assigns to numbers the address of the allocated memory.

The reserved word new allocates memory for the array’s elements. new technically is classified as an
operator. The area of memory used for dynamic allocation via new is different from both the area used for
local variables (the stack) and global variables (static memory). Dynamic memory comes from the heap.
Section 11.11 provides more details about the memory available to an executing C++ program.
The variable numbers is local to main, so it lives on the stack. It is a pointer, so it stores a memory
address. The new expression returns an address to memory in the heap. So, even though the numbers
variable itself is stored on the stack, the memory it references is located in the heap. We do not need to
worry about our array being too big to be local to main, since its contents do not consume stack space.
Local variables are automatically allocated on the stack when a function is invoked and automatically
deallocated when the function returns. The programmer does not need to explicitly manage local variables.
Dynamically allocated memory, however, requires more attention on the part of the programmer. The
statement
delete [] numbers; // Free up the space held by numbers

uses the delete operator to free up the memory held by the numbers array. The programmer is respon-
sible for deallocating memory that was allocated with new. Every use of the new operator should have a
corresponding use of delete somewhere later in the program’s execution. Notice that the square brackets
([]) are empty. The run-time environment keeps track of the amount of memory to free up.
delete is a C++ reserved word.
You should develop the habit of ensuring that every use of the new operator has an associated call of
delete. If the delete statement is omitted from Listing 11.14 (flexiblearrayaverage.cpp), the program
in this case likely will behave no differently since most operating systems reclaim all the memory held a
programs when the program finishes executing. Serious problems arise in longer running programs that
must allocate and deallocate dynamic memory over time. Programs that run for extended periods of time,
like web servers and operating system kernels, can crash due to failures to properly deallocate memory. The
condition is known a memory leak—the program overtime allocates more and more memory via new but
never releases the memory back with delete. Eventually the program uses up all of the available heap
space and crashes.
Listing 11.14 (flexiblearrayaverage.cpp) requires the user to enter up front the number of values to
average. This is inconvenient, and people are notoriously poor counters. One solution is to allocate a
minimal size array, and then resize it as necessary when it fills up. Listing 11.15 (resizearray.cpp) uses this
approach.

Listing 11.15: resizearray.cpp

1 #include <iostream>
2

©2013 Richard L. Halterman Draft date: January 27, 2013

 11.8. DYNAMIC ARRAYS 299

3 using namespace std;

4
5 int main()
6 {
7 double sum = 0.0, // Sum of the elements in the list
8 *numbers, // Dynamic array of numbers
9 input; // Current user entry
10
11 // Initial size of array and amount to expand full array
12 const int CHUNK = 3;
13 int size = 0, // Current number of entries
14 capacity = CHUNK; // Initial size of array
15
16
17 // Allocate a modest-sized array to begin with
18 numbers = new double[capacity];
19
20 cout << "Please enter any number of non-negative values "
21 << "(negative value ends the list): ";
22 cin >> input;
23
24 while ( input >= 0 ) // Continue until negative number entered
25 {
26 if ( size >= capacity ) // Room left to add an element?
27 {
28 capacity += CHUNK; // Expand array
29 double *temp = new double[capacity]; // Allocate space
30 for ( int i = 0; i < size; i++ )
31 temp[i] = numbers[i]; // Copy existing values
32 delete [] numbers; // Free up old space
33 numbers = temp; // Update numbers to new location
34 cout << "Expanding by " << CHUNK << endl;
35 }
36 numbers[size] = input; // Add number to array at last position
37 size++; // Update last position
38 sum += input; // Add to running sum
39 cin >> input; // Get next number
40 }
41 if ( size > 0 ) // Can't average less than one number
42 {
43 cout << "The average of ";
44 for ( int i = 0; i < size - 1; i++ )
45 cout << numbers[i] << ", ";
46 // No comma following last element
47 cout << numbers[size - 1] << " is "
48 << sum/size << endl;
49 }
50 else
51 cout << "No numbers to average" << endl;
52 delete [] numbers; // Free up the space held by numbers
53 }

©2013 Richard L. Halterman Draft date: January 27, 2013

11.9. THE SIZEOF OPERATOR 300

11.9 The sizeof Operator

The size of a static array is fixed at compile time. The C++ sizeof operator computes the amount of
memory in bytes that a static array consumes. The code
int list[100],
set[] = { 1, 2, 3, 4, 5 };
cout << "list: " << (sizeof list) << ", set: " << (sizeof set) << endl;

prints

list: 400, set: 20

on a system in which integers use four bytes of storage. This is because list holds 100 integers at four
bytes each, and set holds five integers at four bytes each. The sizeof operator computes the storage
required for any type; for example,
cout << "int: " << (sizeof int) << ", double: " << (sizeof double) << endl;

prints

int: 4, double: 8

on systems that store integers in fours bytes and double-precision floating-point numbers in eight bytes.
The number of elements in a statically allocated array can be obtained from a simple formula: divide
the memory occupied by the array by the number of bytes required for each element. The following code
fragment
int list[100];
double collection[] = { 2.8, 3.33, 0.4, 0.0, 10.4 };
cout << "list: " << (sizeof list)/(sizeof list[0])
<< ", collection: " << (sizeof collection)/(sizeof collection[0]) << endl;

prints

list: 100, collection: 5

on all systems regardless of how much memory integers and floating-point numbers consume.
The code in the main function of Listing 11.5 (arraytofunc.cpp)
int list[] = { 2, 4, 6, 8 };
// Print the contents of the array
print(list, 4);

©2013 Richard L. Halterman Draft date: January 27, 2013

11.9. THE SIZEOF OPERATOR 301

therefore could be written

int list[] = { 2, 4, 6, 8 };
// Print the contents of the array
print(list, (sizeof list)/(sizeof list[0]));

to eliminate the literal value 4 as the second parameter. The programmer then may add extra elements to,
or remove elements from, the initialization list without changing the call to to print.
The sizeof operator can be used only to compute sizes known at compile time and is not helpful for
dynamically allocated arrays. The code
int *list;
list = new int[100];
cout << "list: " << (sizeof list)/(sizeof list[0]) << endl;

on most systems prints

list: 1

because pointers and integers often require the same amount of memory. Since list’s declared type is
int *, the sizeof operator returns the size of a pointer, not the number of elements in the dynamically-
allocated array list points to.
Because of the way arrays are passed to functions in C++, we must always pass the size of the array
along with the array itself. The function declaration
void print(int a[])
{
cout << (sizeof a) << endl; // Prints 4
// Other details omitted . . .
}

is equivalent to
void print(int a[100])
{
cout << (sizeof a) << endl; // Prints 4
// Other details omitted . . .
// Details omitted . . .
}

because the compiler ignores any numbers within the square brackets of an array parameter’s declaration.
Arrays are always passed as pointers to functions. That is why a declaration of the form
void print(int *a)
{
cout << (sizeof a) << endl; // Prints 4
// Other details omitted . . .
// Details omitted . . .
}

is so common—it most accurately reflects how arrays are passed to functions.

©2013 Richard L. Halterman Draft date: January 27, 2013

11.10. COPYING AN ARRAY 302

11.10 Copying an Array

It is important to remember that within C++ source code an array variable behaves similarly to a constant
pointer. At first it may seem plausible to make a copy of an array as follows:
int a[10], b[10]; // Declare two arrays
for ( int i = 0; i < 10; i++ ) // Populate one of them
a[i] = i; // a is filled with increasing values
b = a; // Make a copy of array a?

Since b behaves like a constant pointer, it cannot be reassigned; that is, the name b cannot appear on the
left side of the assignment operator all by itself. Wherever b points, it must continue to point there during
its lifetime.
Perhaps the solution is to use a dynamic array (see Section 11.8 for information about dynamic arrays).
If b is not const, can the array be copied through a simple assignment? The following code is legal:
int a[10], *b; // Declare two arrays, one dynamic
for ( int i = 0; i < 10; i++ ) // Populate one of them
a[i] = i; // a is filled with increasing values
b = a; // Make a copy of array a?

but b is not a copy of a. b aliases a, so changing the contents of a will also change the contents of b
identically, since b refers to the exact memory to which a refers. Figure 11.8 illustrates the aliasing.

int a[] = { 2, 4, 6, 8 }, *b;

a
2 4 6 8
Declare

b 0 1 2 3

b = a;
a
2 4 6 8 Alias, no
copy

b 0 1 2 3 made

Figure 11.8: Faulty array copy

The following code shows the proper way to make a copy of a:

int a[10], *b; // Declare two arrays, one dynamic
for ( int i = 0; i < 10; i++ ) // Populate one of them
a[i] = i; // a is filled with increasing values

©2013 Richard L. Halterman Draft date: January 27, 2013

11.11. MEMORY MANAGEMENT 303

// Really make a copy of array a

b = new int[10]; // Allocate b
for ( int i = 0; i < 10; i++ )
b[i] = a[i];

Separate space for the dynamic array must be allocated, and then each element from the original array must
be copied into the new array. Figure 11.9 shows how this process works.

int a[] = { 2, 4, 6, 8 }, *b;

a
2 4 6 8 Declare

b 0 1 2 3

b = new int[4];
a
2 4 6 8 Allocate
space for

b 0 1 2 3 copy

? ? ? ?
0 1 2 3

for ( int i = 0; i < 4; i++ )

b[i] = a[i];
a
2 4 6 8 Copy the
elements

b 0 1 2 3

2 4 6 8
0 1 2 3

Figure 11.9: Correct array copy

It is important to note that since the code above allocates b dynamically with new, code elsewhere
within the program should eventually use delete to free up b’s space when it is no longer used. Failure
to properly deallocate b constitutes a memory leak.

11.11 Memory Management

Memory management issues with new and delete frequently are the source of difficult to find and fix
logic errors. Programmers must adhere to the following tenets:

• Every call to new should have an associated call to delete when the allocated memory is no

©2013 Richard L. Halterman Draft date: January 27, 2013

11.11. MEMORY MANAGEMENT 304

longer needed. It sounds simple enough, but it is not always clear when delete should be used.
The following function exhibits a memory leak:
void calc(int n)
{
// ...
// Do some stuff
// ...
int *p = new int[n];
// ...
// Do some stuff with p
// ...
// Exit function without deleting p's memory
}

If a program calls function calc enough times, the program will eventually run out of memory.
Figure 11.10 shows how memory is stranded when local pointers go out of scope at the end of a
function’s execution. In the calc function p is a local variable. As such, p lives on the stack

Figure 11.10: Memory leak

(see Section 11.8). When a particular call to calc completes, the function’s clean up code automat-
ically releases the space help by the variable p. This is because all functions automatically manage
the memory for their parameters and local variables. The problem is p is assigned via new to point
to memory allocated from the heap, not the stack. Function executions manage the stack memory
only. When the function’s execution completes, p is gone, and the memory to which p pointed is not
deallocated automatically. Worse yet, the heap memory formerly referenced by the local variable p
now is unreachable because no other variables that remain viable know anything about it.
• The delete operator never should be used to free up memory that was not allocated by a
previous call to new. This code fragment illustrates one such example:
int list[10], *p = list; // p points to list
// ...

©2013 Richard L. Halterman Draft date: January 27, 2013

11.11. MEMORY MANAGEMENT 305

// Do some stuff
// ...
delete [] p; // Logic error, attempt to deallocate p's memory

The space referenced by pointer p was not allocated by new, so delete should not be used to
attempt to free up its memory.
• delete must not be used to deallocate the same memory more than once. This can happen
when two pointers refer to the same memory. Such pointers are called aliases. The following code
fragment illustrates the situation:
int *p = new int[10], *q = p; // q aliases p
// ...
// Do some stuff with p and/or q
// ...
delete [] p; // Free up p's memory
// ...
// Do some other stuff
// ...
delete [] q; // Logic error, q's memory already freed!

Since pointer p and pointer q point to the same memory, they are aliases of each other. Deallocate
the memory referenced by one of them, and the other’s memory is also deallocated, since it is the
same memory.
• Memory previously deallocated via delete should never be accessed. It is a logic error to access
memory previously deallocated by delete. The code fragment
int *list = new int[10];
// ...
// Use list, then
// ...
delete [] list; // Deallocate list's memory
// ...
// Sometime later
// ...
int x = list[2]; // Logic error, but sometimes works!

illustrates how such a situation can arise. For efficiency reasons the delete operator generally
marks heap space as “available” without modifying the contents of that memory. Careless program-
mers can accidentially use that memory obliviously as if it were still live. The problem manifests
itself when the freed up memory eventually gets reallocated elsewhere via a call to new. The result
is that programs seem to “work” for a while and mysteriously fail at unpredictable times. Debugging
such situations can be very difficult. The problem often happens because of aliasing:
int *list = new int[10];
// ...
int *arr = list; // arr aliases list
// ...
delete [] list; // Deallocate list's memory
// ...
// Sometime later
// ...
int x = arr[2]; // Same problem!

©2013 Richard L. Halterman Draft date: January 27, 2013

11.11. MEMORY MANAGEMENT 306

We deleted list’s memory, not arr’s memory, didn’t we? No, since arr is an alias for list,
arr references memory previously deallocated with delete.

The space for non-static local variables is automatically allocated when a function begins executing,
and that space is automatically deallocated when the function returns. A programmer need not explicitly
use something like new and delete to manage memory for locals. The space for function parameters is
handled automatically also.
The space for global variables and static local variables is automatically allocated before the pro-
gram begins executing its main function.
Figure 11.11 illustrates the typical organization of the memory of an executing C++ program. Memory

High memory
Stack

Heap

Data

Code
Low memory

Figure 11.11: Typical memory organization in an executing program

for an executing program is divided into four sections:

• Code. The code section of memory holds the program’s compiled executable instructions. The code
section should never change while the program executes, and some operating systems enforce this
restriction. The size of the code segment does not change while the program executes.

©2013 Richard L. Halterman Draft date: January 27, 2013

11.12. SUMMARY 307

• Data. The data section of memory contains global variables (see Section 10.1) and persistent local
variables (static locals, see Section 10.2). The variables in the data section exist for the life of
the executing program, but, unless they are constants (see Section 3.5), the executing program may
freely change their values. The size of the data segment does not change while the program executes.

• Heap. The heap is where an executing program obtains dynamic memory. The new operator gets
its memory from the heap, and delete returns previously allocated memory back to the heap. The
size of the heap grows and shrinks during the program’s execution as the program allocates and
deallocates dynamic memory using new and delete.

• Stack. The stack is where local variables and function parameters live. Space for local variables and
parameters appears when a function is called and disappears when the function returns. The size of
the stack grows and shrinks during the program’s execution as various functions execute.

Notice that the stack and heap segments grow toward each other. Deep recursion can consume a consider-
able amount of stack space. An improperly written recursive function, for example, one that omits the base
case and thus exhibits “infinite” recursion, will cause the stack to collide with the heap. Modern operating
systems will terminate a process that consumes all of its stack space, but on some embedded systems this
stack overflow may go undetected. Similarly, a program that allocates a lot of dynamic memory, especially
a program with memory leaks, may cause the heap to collide with the stack. In either case the executing
program does not have enough memory available to satisfy its needs, so the operating system will terminate
the program’s execution.

11.12 Summary

• An array represents a collection of homogeneous data values.

• An array variable references the starting address of a chunk of contiguous memory that holds the
array’s elements.

• The type of values contained by an array and the number values the array can hold are specified when
the array is declared; for example,
double list[20];

• An element of an array is accessed via a non-negative integer index using the [] operator; for exam-
ple,
v[3] = 4.1;

The index of the desired element is placed within the square brackets.

• The first element in an array is at index zero. If the array contains n elements, the last element is at
index n − 1.

• For non-dynamic arrays, an array’s size must be specified when it is declared; for example,
int numbers[10]; // numbers holds 10 integers

• An array’s contents may be initialized when the array is declared, as in

int arr[] = { 2, 4, 6, 8 };

©2013 Richard L. Halterman Draft date: January 27, 2013

11.13. EXERCISES 308

• Any expression compatible with an integer can can placed within the [] operator. Literals, variables,
and expressions can be used. The compiler will issue an error if a floating-point expression is used
within the square brackets.
• A for loop is a convenient way to traverse the contents of an array.
• Like other variables, an array variable can be local, static local, or global.
• An array can be passed to a function. The actual parameter passed to the function is a pointer to
the first element of the array. Ordinarily, the caller must pass an additional parameter specifying the
number of elements within the array, since an array does not keep track of its own size.
• If a function is not supposed to modify the contents of an array, best programming practices insist the
array parameter should be declared const.
• It is the programmer’s responsibility to stay within the bounds of an array. Venturing outside the
bounds of an array results in undefined behavior that introduces a logic error and often leads to a
program crash.
• One array may not be assigned to another through simple assignment; a loop must be used to copy
each element individually.
• Two-dimensional arrays conceptually store their elements in rows and columns. C++ supports arrays
of arbitrary dimensions.
• In a 2D array expression like a[i][j] the first index (here i) represents a row, and the second index
(here j) specifies a column.
• The space for an array’s elements may be allocated dynamically at run time using new.
• During a program’s execution, the elements of statically allocated local arrays live on the stack, the
elements of statically allocated global arrays live in static memory, and the elements of dynamically
allocated arrays (local or global) live on the heap.
• The use of the new operator allocates resources that can be reclaimed only through the use of
delete. Failure to use a matching delete with each use of new results in a memory leak.

11.13 Exercises

1. Can you declare an array to hold a mixture of ints and doubles?

2. What happens if you attempt to access an element of an array using a negative index?
3. Given the declaration
int list[100];

(a) What expression represents the very first element of list?

(b) What expression represents the very last element of list?
(c) Write the code fragment that prints out the contents of list.
(d) Is the expression list[3.0] legal or illegal?

4. Is the following code fragment legal or illegal?

©2013 Richard L. Halterman Draft date: January 27, 2013

11.13. EXERCISES 309

int list1[5], list2[5] = { 3, 3, 3, 3, 3 };

list1 = list2;

5. Provide a single declaration statement that declares an integer array named list that contains the
values 45, −3, 16 and 8?
6. Does an array keep track of the number of elements it contains?
7. Complete the following function that adds up all the positive values in an array of integers. For
example, if array arr contains the elements 3, -3, 5, 2, −1, and 2, the call sum_positive(arr)
would evaluate to 12, since 3 + 5 + 2 + 2 = 12. The function returns zero if the array is empty (that
is, n ¡ 1).
// Array a with length n
int sum_positive(const int *a, int n)
{
// Add your code...
}

8. Complete the following function that counts the even numbers in an array of integers. For example,
if array arr contains the elements 3, 5, 2, −1, and 2, the call count_evens(arr) would evaluate
to 4, since 2 + 2 = 4. The function returns zero if the array is empty (that is, n ¡ 1). The function does
not affect the contents of the array.
// Array a with length n
int count_evens(const int *a, int n)
{
// Add your code...
}

9. Complete the following function that counts the even numbers in a 2D array of integers.
// 2D array v with r rows and c columns
int count_evens(const int **v, int r, int c)
{
// Add your code...
}

10. Suppose your task is to implement the function with the prototype
void proc(int a[]);

When you implement the body of proc, how can you determine the size of array a?
11. Consider the declaration
int collection[100][200];

What does the expression collection[15][29] represent?

12. Consider the declaration
int collection[100][200];

How many elements does collection hold?

©2013 Richard L. Halterman Draft date: January 27, 2013

11.13. EXERCISES 310

13. Consider the declaration

int collection[100][200];

Write the C++ code that prints all the elements in collection. All the elements in the same row
should appear on the same line, and but each successive row should appear on its own line.

14. Consider the declaration

int collection[100][200];

What does the expression collection[15] represent?

15. Consider the declaration

int mesh[100][200][100][50];

How many elements does mesh hold?

16. Rewrite the following expressions using pointer notation instead of array notation.

(a) a[4]
(b) a[1]
(c) a[0]

17. Rewrite the following expressions using array notation instead of pointer notation.

(a) *(a + 3)
(b) *a
(c) *(a + 0)

18. Rewrite the following code fragment using array notation instead of pointer notation:
void display(int *a, int n)
{
while ( n )
{
cout << *a << " ";
a++;
n--;
}
cout << endl;
}

19. Rewrite the following code fragment using pointer notation instead of array notation:
int sum(int *a, int n)
{
int s = 0;
for ( int i = 0; i < n; i++ )
s += a[i];
return s;
}

©2013 Richard L. Halterman Draft date: January 27, 2013

11.13. EXERCISES 311

20. Consider the following declaration:

char *word = "abcde";

Evaluate each of the following expressions.

(a) word[1]
(b) *word
(c) word[5]
21. Suppose your task is to implement the function with the prototype
void proc(char *s);

where s is a C string. When you implement the body of proc, how can you determine the length of
string s?
22. Given the following declarations which appear in a function body:
double nums[100], *grid = new double[100];

(a) Where will the nums array live, static memory, the stack, or the heap?
(b) Where will the grid array live, static memory, the stack, or the heap?

23. What operator should eventually be used when the new operator is used to allocate memory? What
is the consequence of its omission?
24. List some common errors programmers make when dealing with dynamic memory.
25. Complete the following function that counts the number of negative values in a 10 × 10 integer 2D
array.
int count_negatives(int a[10][10])
{
// Add your code...
}

©2013 Richard L. Halterman Draft date: January 27, 2013

11.13. EXERCISES 312

©2013 Richard L. Halterman Draft date: January 27, 2013

 313

Chapter 12

Array Algorithms

12.1 Sorting

Sorting—arranging the elements within an array into a particular order—is a common activity. For example,
an array of integers may be arranged in ascending order (that is, from smallest to largest). An array of words
(strings) may be arranged in lexicographical (commonly called alphabetic) order. Many sorting algorithms
exist, and some perform much better than others. We will consider one sorting algorithm that is relatively
easy to implement.
The selection sort algorithm is relatively easy to implement, and it performs acceptably for smaller
arrays. If A is the array, and i represents an array index, selection sort works as follows:

1. Set i = 0.

2. Examine all the elements A[ j], where j > i. If any of these elements is less than A[i], then exchange
A[i] with the smallest of these elements. (This ensures that all elements after position i are greater
than or equal to A[i].)

3. If i is less than the length of A, set i equal to i + 1, and goto Step 2.

If the condition in Step 3 is not met, the algorithm terminates with a sorted array. The command to “goto
Step 2” in Step 3 represents a loop. We can begin to translate the above description into C++ as follows:
// n is A's length
for ( int i = 0; i < n - 1; i++ )
{
// Examine all the elements
// A[j], where j > i.
// If any of these A[j] is less than A[i],
// then exchange A[i] with the smallest of these elements.
}

The directive at Step 2 beginning with “Examine all the elements A[ j], where j > i” must also be
implemented as a loop. We continue refining our implementation with:
// n is A's length

©2013 Richard L. Halterman Draft date: January 27, 2013

 12.1. SORTING 314

for ( int i = 0; i < n - 1; i++ )

{
for ( int j = i + 1; j < n; j++ )
{
// Examine all the elements
// A[j], where j > i.
}
// If any A[j] is less than A[i],
// then exchange A[i] with the smallest of these elements.
}

In order to determine if any of the elements is less than A[i], we introduce a new variable named
small. The purpose of small is to keep track of the position of the smallest element found so far. We
will set small equal to i initially, because we wish to locate any element less than the element found at
position i.
// n is A's length
for ( int i = 0; i < n - 1; i++ )
{
// small is the position of the smallest value we've seen
// so far; we use it to find the smallest value less than A[i]
int small = i;
for ( int j = i + 1; j < n; j++ )
{
if ( A[j] < A[small] )
small = j; // Found a smaller element, update small
}
// If small changed, we found an element smaller than A[i]
if ( small != i )
// exchange A[small] and A[i]
}

Listing 12.1 (sortintegers.cpp) provides the complete C++ implementation of the selection_sort
function within a program that tests it out.

Listing 12.1: sortintegers.cpp

1 #include <iostream>
2
3 using namespace std;
4
5 /*
6 * swap(a, b)
7 * Interchange the values of
8 * its parameters a and b which are
9 * passed by reference.
10 */
11 void swap(int& a, int& b)
12 {
13 int temp = a;
14 a = b;
15 b = temp;
16 }

©2013 Richard L. Halterman Draft date: January 27, 2013

 12.1. SORTING 315

17
18 /*
19 * selection_sort
20 * Arranges the elements of array a into ascending order.
21 * a is an array of ints.
22 * n is the array's length.
23 * The contents of a are physically rearranged.
24 */
25 void selection_sort(int a[], int n)
26 {
27 for ( int i = 0; i < n - 1; i++ )
28 {
29 // Note: i,small, and j represent positions within a
30 // a[i], a[small], and a[j] represents the elements at
31 // those positions.
32 // small is the position of the smallest value we've seen
33 // so far; we use it to find the smallest value less
34 // than a[i]
35 int small = i;
36 // See if a smaller value can be found later in the array
37 for ( int j = i + 1; j < n; j++ )
38 if ( a[j] < a[small] )
39 small = j; // Found a smaller value
40 // Swap a[i] and a[small], if a smaller value was found
41 if ( i != small )
42 swap(a[i], a[small]);
43 }
44 }
45
46 /*
47 * print
48 * Prints the contents of an int array
49 * a is the array to print.
50 * n is the array's length.
51 * a is not modified.
52 */
53 void print(const int a[], int n)
54 {
55 cout << '{';
56 if ( n > 0 )
57 {
58 cout << a[0]; // Print the first element
59 for ( int i = 1; i < n; i++ )
60 cout << ',' << a[i]; // Print the rest
61 }
62 cout << '}';
63 }
64
65
66 int main()
67 {
68 int list[] = { 23, -3, 4, 215, 0, -3, 2, 23, 100, 88, -10 };
69 int len = (sizeof list)/(sizeof list[0]);
70 cout << "Before: ";
71 print(list, len);

©2013 Richard L. Halterman Draft date: January 27, 2013

 12.2. FLEXIBLE SORTING 316

72 cout << endl;

73 selection_sort(list, len);
74 cout << "After: ";
75 print(list, len);
76 cout << endl;
77 }

Listing 12.1 (sortintegers.cpp) uses a fancier print routine, separating the array’s elements with com-
mas. The program’s output is

Before: {23,-3,4,215,0,-3,2,23,100,88,-10}
After: {-10,-3,-3,0,2,4,23,23,88,100,215}

12.2 Flexible Sorting

What if want to change the behavior of the sorting function in Listing ?? (selectionsort.cpp) so that it
arranges the elements in descending order instead of ascending order? It is actually an easy modification;
simply change the line
if ( a[j] < a[small] )

to be
if ( a[j] > a[small] )

What if instead we want to change the sort so that it sorts the elements in ascending order except that all
the even numbers in the array appear before all the odd numbers? This modification would be a little more
complicated, but it could accomplished (as we will see) in that if statement’s conditional expression.
The next question is more intriguing: How can we rewrite the selection_sort function so that, by
passing an additional parameter, it can sort the array in any way we want?
We can make our sort function more flexible by using a function pointer (see Section 10.8). Listing 12.2
(flexiblesort.cpp) arranges the elements in an array two different ways using the same selection_sort
function.

Listing 12.2: flexiblesort.cpp

1 #include <iostream>
2
3 using namespace std;
4
5 /*
6 * swap(a, b)
7 * Interchange the values of
8 * its parameters a and b which are
9 * passed by reference.
10 */
11 void swap(int& a, int& b)
12 {
13 int temp = a;

©2013 Richard L. Halterman Draft date: January 27, 2013

 12.2. FLEXIBLE SORTING 317

14 a = b;
15 b = temp;
16 }
17
18 /*
19 * less_than(a, b)
20 * Returns true if a < b; otherwise, returns
21 * false.
22 */
23 bool less_than(int a, int b)
24 {
25 return a < b;
26 }
27
28 /*
29 * greater_than(a, b)
30 * Returns true if a > b; otherwise, returns
31 * false.
32 */
33 bool greater_than(int a, int b)
34 {
35 return a > b;
36 }
37
38 /*
39 * selection_sort(a, n, compare)
40 * Arranges the elements of a in an order determined
41 * by the compare function.
42 * a is an array of ints.
43 * n is the array's length.
44 * compare is a function that compares the ordering of
45 * two integers.
46 * The contents of a are physically rearranged.
47 */
48 void selection_sort(int a[], int n, bool (compare)(int, int))
49 {
50 for ( int i = 0; i < n - 1; i++ )
51 {
52 // Note: i,small, and j represent positions within a
53 // a[i], a[small], and a[j] represents the elements at
54 // those positions.
55 // small is the position of the smallest value we've seen
56 // so far; we use it to find the smallest value less
57 // than a[i]
58 int small = i;
59 // See if a smaller value can be found later in the array
60 for ( int j = i + 1; j < n; j++ )
61 if ( compare(a[j], a[small]) )
62 small = j; // Found a smaller value
63 // Swap a[i] and a[small], if a smaller value was found
64 if ( i != small )
65 swap(a[i], a[small]);
66 }
67 }
68

©2013 Richard L. Halterman Draft date: January 27, 2013

 12.2. FLEXIBLE SORTING 318

69 /*
70 * print
71 * Prints the contents of an int array
72 * a is the array to print.
73 * n is the array's length.
74 * a is not modified.
75 */
76 void print(const int a[], int n)
77 {
78 cout << '{';
79 if ( n > 0 )
80 {
81 cout << a[0]; // Print the first element
82 for ( int i = 1; i < n; i++ )
83 cout << ',' << a[i]; // Print the rest
84 }
85 cout << '}';
86 }
87
88
89 int main()
90 {
91 int list[] = { 23, -3, 4, 215, 0, -3, 2, 23, 100, 88, -10 };
92 int len = (sizeof list)/(sizeof list[0]);
93 cout << "Original: ";
94 print(list, len);
95 cout << endl;
96 selection_sort(list, len, less_than);
97 cout << "Ascending: ";
98 print(list, len);
99 cout << endl;
100 selection_sort(list, len, greater_than);
101 cout << "Descending: ";
102 print(list, len);
103 cout << endl;
104 }

The output of Listing 12.2 (flexiblesort.cpp) is

Original: {23,-3,4,215,0,-3,2,23,100,88,-10}
Ascending: {-10,-3,-3,0,2,4,23,23,88,100,215}
Descending: {215,100,88,23,23,4,2,0,-3,-3,-10}

The comparison function passed to the sort routine customizes the sort’s behavior. The basic structure of
the sorting algorithm does not change, but its notion of ordering is adjustable. If the second parameter to
selection_sort is less_than, the elements are arranged in ascending order. If greater_than
is used instead, the array is sorted in descending order. More creative orderings are possible with more
elaborate comparison functions.
Selection sort is a relatively efficient simple sort, but more advanced sorts are, on average, much faster
than selection sort, especially for large data sets. One such general purpose sort is Quicksort, devised by

©2013 Richard L. Halterman Draft date: January 27, 2013

 12.3. SEARCH 319

C. A. R. Hoare in 1962. Quicksort is the fastest known general purpose sort. Since sorting is a common
data processing activity, the standard C library provides a function named qsort that implements Quick-
sort. More information about Quicksort and qsort is available at http://en.wikipedia.org/wiki/
Quicksort.

12.3 Search

Searching an array for a particular element is a common activity. We will consider two kinds of search
strategies: linear search and binary search.

12.3.1 Linear Search

Listing 12.3 (linearsearch.cpp) uses a function named locate that returns the position of the first occur-
rence of a given element in an array of integers; if the element is not present, the function returns −1.

Listing 12.3: linearsearch.cpp

1 #include <iostream>
2 #include <iomanip>
3
4 using namespace std;
5
6 /*
7 * locate(a, n, seek)
8 * Returns the index of element seek in array a.
9 * Returns -1 if seek is not an element of a.
10 * a is the array to search.
11 * n is the array's length.
12 * seek is the element to find.
13 */
14 int locate(const int a[], int n, int seek)
15 {
16 for ( int i = 0; i < n; i++ )
17 if ( a[i] == seek )
18 return i; // Return position immediately
19 return -1; // Element not found
20 }
21
22 /*
23 * format(i)
24 * Prints integer i right justified in a 4-space
25 * field. Prints "****" if i > 9,999.
26 */
27 void format(int i)
28 {
29 if ( i > 9999 )
30 cout << "****" << endl; // Too big!
31 else
32 cout << setw(4) << i;
33 }
34
35

©2013 Richard L. Halterman Draft date: January 27, 2013

 12.3. SEARCH 320

36 /*
37 * print(v, n)
38 * Prints the contents of an int array.
39 * a is the array to print.
40 * n is the array's length.
41 * /
42 void print(const int a[], int n)
43 {
44 for ( int i = 0; i < n; i++ )
45 format(a[i]);
46 }
47
48
49 /*
50 * display(a, n, value)
51 * Draws an ASCII art arrow showing where
52 * the given value is within the array.
53 * a is the array.
54 * n is the array's length.
55 * value is the element to locate.
56 */
57 void display(const int a[], int n, int value)
58 {
59 int position = locate(a, n, value);
60 if ( position >= 0 )
61 {
62 print(a, n); // Print contents of the array
63 cout << endl;
64 position = 4*position + 7; // Compute spacing for arrow
65 cout << setw(position);
66 cout << " ˆ " << endl;
67 cout << setw(position);
68 cout << " | " << endl;
69 cout << setw(position);
70 cout << " +-- " << value << endl;
71 }
72 else
73 {
74 cout << value << " not in ";
75 print(a, n);
76 cout << endl;
77 }
78 cout << "======" << endl;
79 }
80
81 int main()
82 {
83 int list[] = { 100, 44, 2, 80, 5, 13, 11, 2, 110 };
84 int len = (sizeof list)/(sizeof list[0]);
85 display(list, len, 13);
86 display(list, len, 2);
87 display(list, len, 7);
88 display(list, len, 100);
89 display(list, len, 110);
90 }

©2013 Richard L. Halterman Draft date: January 27, 2013

12.3. SEARCH 321

The output of Listing 12.3 (linearsearch.cpp) is

100 44 2 80 5 13 11 2 110
ˆ
|
+-- 13
======
100 44 2 80 5 13 11 2 110
ˆ
|
+-- 2
======
7 not in 100 44 2 80 5 13 11 2 110
======
100 44 2 80 5 13 11 2 110
ˆ
|
+-- 100
======
100 44 2 80 5 13 11 2 110
ˆ
|
+-- 110
======

The key function in Listing 12.3 (linearsearch.cpp) is locate; all the other functions simply lead to a
more interesting display of locate’s results. If locate finds a match, it immediately returns the position
of the matching element; otherwise, if locate considers all the elements of the array and finds no match,
it returns −1. −1 is a good indication of failure, since −1 is not a valid index in a C++ array.
The other functions are

• format prints an integer right justified within a four-space field. Extra spaces pad the beginning of
the number if necessary.

• print prints out the elements in any integer array using the format function to properly format
the values. This alignment simplifies the display function.

• display uses locate, print, and setw to provide a graphical display of the operation of the
locate function.

The kind of search performed by locate is known as linear search, since the process takes a straight
line path from the beginning to the end of the array, and it considers each element in order. Figure 12.1
illustrates linear search.

©2013 Richard L. Halterman Draft date: January 27, 2013

 12.3. SEARCH 322

int list[] = { 100, 44, 2, 80, 5, 13, 11, 2, 110 };

int x = locate(list, 9, 13);

list
100 44 2 80 5 13 11 2 110
0 1 2 3 4 5 6 7 8
13?
5

Figure 12.1: Linear search. The grayed out elements are not considered during the search process.

12.3.2 Binary Search

Linear search is acceptable for relatively small arrays, but the process of examining each element in a large
array is time consuming. An alternative to linear search is binary search. In order to perform binary search
an array’s elements must be in sorted order. Binary search exploits this sorted structure of the array using a
clever but simple strategy that quickly zeros in on the element to find:

1. If the array is empty, return −1.

2. Check the element in the middle of the array. If it is what you are seeking, return its position. If the
middle element is larger than the element you are seeking, perform a binary search on the first half of
the array. If the middle element is smaller than the element you are seeking, perform a binary search
on the second half of the array.

This approach is analogous to looking for a telephone number in the phone book in this manner:

1. Open the book at its center. If the name of the person is on one of the two visible pages, look at the
phone number.
2. If not, and the person’s last name is alphabetically less the names on the visible pages, apply the
search to the left half of the open book; otherwise, apply the search to the right half of the open book.
3. Discontinue the search with failure if the person’s name should be on one of the two visible pages
but is not present.

Listing 12.4 (binarysearch.cpp) contains a C++ function that implements the binary search algorithm.

Listing 12.4: binarysearch.cpp

1 #include <iostream>
2 #include <iomanip>
3
4 using namespace std;
5
6 /*
7 * binary_search(a, n, seek)

©2013 Richard L. Halterman Draft date: January 27, 2013

 12.3. SEARCH 323

8 * Returns the index of element seek in array a;

9 * returns -1 if seek is not an element of a
10 * a is the array to search; a's contents must be
11 * sorted in ascending order.
12 * n is the length of the array.
13 * seek is the element to find.
14 */
15 int binary_search(const int a[], int n, int seek)
16 {
17 int first = 0, // Initially the first element in array
18 last = n - 1, // Initially the last element in array
19 mid; // The middle of the array
20 while ( first <= last )
21 {
22 mid = first + (last - first + 1)/2;
23 if ( a[mid] == seek )
24 return mid; // Found it
25 else if ( a[mid] > seek )
26 last = mid - 1; // continue with 1st half
27 else // a[mid] < seek
28 first = mid + 1; // continue with 2nd half
29 }
30 return -1; // Not there
31 }
32
33 /*
34 * format(i)
35 * Prints integer i right justified in a 4-space
36 * field. Prints "****" if i > 9,999.
37 */
38 void format(int i)
39 {
40 if ( i > 9999 )
41 cout << "****" << endl; // Too big!
42 else
43 cout << setw(4) << i;
44 }
45
46 /*
47 * print(v, n)
48 * Prints the contents of an int array.
49 * a is the array to print.
50 * n is the array's length.
51 */
52 void print(const int a[], int n)
53 {
54 for ( int i = 0; i < n; i++ )
55 format(a[i]);
56 }
57
58
59 /*
60 * display(a, n, value)
61 * Draws an ASCII art arrow showing where
62 * the given value is within the array.

©2013 Richard L. Halterman Draft date: January 27, 2013

 12.3. SEARCH 324

63 * a is the array.
64 * n is the array's length.
65 * value is the element to locate.
66 */
67 void display(const int a[], int n, int value)
68 {
69 int position = binary_search(a, n, value);
70 if ( position >= 0 )
71 {
72 print(a, n); // Print contents of the array
73 cout << endl;
74 position = 4*position + 7; // Compute spacing for arrow
75 cout << setw(position);
76 cout << " ˆ " << endl;
77 cout << setw(position);
78 cout << " | " << endl;
79 cout << setw(position);
80 cout << " +-- " << value << endl;
81 }
82 else
83 {
84 cout << value << " not in ";
85 print(a, n);
86 cout << endl;
87 }
88 cout << "======" << endl;
89 }
90
91 int main()
92 {
93 // Check binary search on even- and odd-length arrays and
94 // an empty array
95 int even_list[] = { 1, 2, 3, 4, 5, 6, 7, 8 },
96 odd_list[] = { 1, 2, 3, 4, 5, 6, 7, 8, 9 },
97 *empty_list = nullptr;
98
99 for ( int i = -1; i <= 10; i++ )
100 display(even_list, 8, i);
101 for ( int i = -1; i <= 10; i++ )
102 display(odd_list, 9, i);
103 for ( int i = -1; i <= 10; i++ )
104 display(empty_list, 0, i);
105 }

In the binary_search function:

• The initializations of first and last:

int first = 0, // Initially the first element in array
last = n - 1, // Initially the last element in array

ensure that first is less than or equal to last for a nonempty array. If the array is empty, first
is zero, and last is equal to n − 1 which equals −1. So in the case of an empty array the loop body
will be skipped, and −1 will be returned. This is correct behavior because an empty array cannot
possibly contain any item we seek.

©2013 Richard L. Halterman Draft date: January 27, 2013

12.3. SEARCH 325

• The calculation of mid ensures that first ≤ mid ≤ last.

• If mid is the location of the sought element (checked in the first if statement), the loop terminates,
and returns the correct position.

• The second if statement ensures that either last decreases or first increases each time through
the loop. Thus, if the loop does not terminate for other reasons, eventually first will be larger than
last, and the loop will terminate. If the loop terminates for this reason, −1 is returned. This is the
correct behavior.

• The second if statement excludes the irrelevant elements from further search. The number of ele-
ments to consider is cut in half.

Figure 12.2 illustrates how binary search works.

int list[] = { 10, 14, 20, 28, 29, 33, 34, 45, 48 };
int x = binary_search(list, 9, 33);

33?
list
10 14 20 28 29 33 34 45 48
0 1 2 3 4 5 6 7 8

Figure 12.2: Binary search. The grayed out elements are not considered during the search.

The implementation of the binary search algorithm is more complicated than the simpler linear search
algorithm. Ordinarily simpler is better, but for algorithms that process data structures that potentially hold
large amounts of data, more complex algorithms employing clever tricks that exploit the structure of the
data (as binary search does) often dramatically outperform simpler, easier-to-code algorithms.
For a fair comparison of linear vs. binary search, suppose we want to locate an element in a sorted array.
That the array is ordered is essential for binary search, but it can be helpful for linear search as well. The
revised linear search algorithm is
// This version requires array a to be sorted in
// ascending order.
int linear_search(const int a[], int n, int seek)
{
for ( int i = 0; i < n && a[i] <= seek; i++ )
if ( a[i] == seek )
return i; // Return position immediately
return -1; // Element not found
}

©2013 Richard L. Halterman Draft date: January 27, 2013

 12.3. SEARCH 326

Notice that, as in the original version of linear search, the loop will terminate when all the elements have
been examined, but it also will terminate early when it encounters an element larger than the sought element.
Since the array is sorted, there is no need to continue the search once you have begun seeing elements larger
than your sought value; seek cannot appear after a larger element in a sorted array.
Suppose an array to search contains n elements. In the worst case—looking for an element larger than
any currently in the array—the loop in linear search takes n iterations. In the best case—looking for an
element smaller than any currently in the array—the function immediately returns without considering any
other elements. The number of loop iterations thus ranges from 1 to n, and so on average linear search
requires n/2 comparisons before the loop finishes and the function returns.
Now consider binary search. After each comparison the size of the array left to consider is one-half the
original size. If the sought item is not found on the first probe, the number of remaining elements to search
is n/2. The next time through the loop, the number of elements left to consider drops to n/4, and so forth.
The problem of determining how many times a set of things can be divided in half until only one element
remains can be solved with a base-2 logarithm. For binary search, the worst case scenario of not finding the
sought element requires the loop to make log2 n iterations.
How does this analysis help us determine which search is better? The quality of an algorithm is judged
by two key characteristics:

• How much time (processor cycles) does it take to run?

• How much space (memory) does it take to run?

In our situation, both search algorithms process the array with only a few extra local variables, so for large
arrays they both require essentially the same space. The big difference here is speed. Binary search performs
more elaborate computations each time through the loop, and each operation takes time, so perhaps binary
search is slower. Linear search is simpler (fewer operations through the loop), but perhaps its loop executes
many more times than the loop in binary search, so overall it is slower.
We can deduce the faster algorithm in two ways: empirically and analytically. An empirical test is an
experiment; we carefully implement both algorithms and then measure their execution times. The analyt-
ical approach analyzes the source code to determine how many operations the computer’s processor must
perform to run the program on a problem of a particular size.
Listing 12.5 (searchcompare.cpp) gives us some empirical results.

Listing 12.5: searchcompare.cpp

1 #include <iostream>
2 #include <iomanip>
3 #include <ctime>
4
5 using namespace std;
6
7 /*
8 * binary_search(a, n, seek)
9 * Returns the index of element seek in array a;
10 * returns -1 if seek is not an element of a
11 * a is the array to search; a's contents must be
12 * sorted in ascending order.
13 * n is the length of the array.
14 * seek is the element to find.
15 */

©2013 Richard L. Halterman Draft date: January 27, 2013

 12.3. SEARCH 327

16 int binary_search(const int a[], int n, int seek)

17 {
18 int first = 0, // Initially the first element in array
19 last = n - 1, // Initially the last element in array
20 mid; // The middle of the array
21 while ( first <= last )
22 {
23 mid = first + (last - first + 1)/2;
24 if ( a[mid] == seek )
25 return mid; // Found it
26 else if ( a[mid] > seek )
27 last = mid - 1; // continue with 1st half
28 else // a[mid] < seek
29 first = mid + 1; // continue with 2nd half
30 }
31 return -1; // Not there
32 }
33
34
35 /*
36 * linear_search(a, n, seek)
37 * Returns the index of element seek in array a.
38 * Returns -1 if seek is not an element of a.
39 * a is the array to search.
40 * n is the array's length.
41 * seek is the element to find.
42 * This version requires array a to be sorted in
43 * ascending order.
44 * /
45 int linear_search(const int a[], int n, int seek)
46 {
47 for ( int i = 0; i < n && a[i] <= seek; i++ )
48 if ( a[i] == seek )
49 return i; // Return position immediately
50 return -1; // Element not found
51 }
52
53 const unsigned SIZE = 100000;
54 int list[SIZE];
55
56 int main()
57 {
58 // Ensure the elements are ordered low to high
59 for ( int i = 0; i < SIZE; i++ )
60 list[i] = i;
61 // Search for all the elements in list using linear search
62 clock_t start_time = clock(), // Record start time
63 end_time;
64 for ( int i = 0; i < SIZE; i++ )
65 linear_search(list, SIZE, i);
66 end_time = clock();
67 // Print the elapsed time
68 cout << "Elapsed time, linear search: "
69 << static_cast<double>(end_time - start_time)/CLOCKS_PER_SEC
70 << " seconds" << endl;

©2013 Richard L. Halterman Draft date: January 27, 2013

 12.3. SEARCH 328

Array Size Linear Search Binary Search

100,000 33.282 0.023
200,000 132.882 0.050
300,000 298.936 0.077
400,000 531.244 0.104
500,000 829.599 0.133
Execution Times in Seconds

Table 12.1: Running Times (seconds) for Linear and Binary Search

71 // Search for all the elements in list using binary search

72 start_time = clock(); // Record start time
73 for ( int i = 0; i < SIZE; i++ )
74 binary_search(list, SIZE, i);
75 end_time = clock();
76 // Print the elapsed time
77 cout << "Elapsed time, binary search: "
78 << static_cast<double>(end_time - start_time)/CLOCKS_PER_SEC
79 << " seconds" << endl;
80 }

When the constant SIZE is 100,000, linear search takes on average 33 seconds on one system, while
binary search requires less than one second. If you increase SIZE to 500,000, linear search runs in 830 sec-
onds (13 minutes, 50 seconds), while binary search still takes less than one second!
Table 12.1 lists the results for various sized arrays. Empirically, binary search performs dramatically
better than linear search.
In addition to using the empirical approach, we can judge which algorithm is better by analyzing the
source code for each function. Each arithmetic operation, assignment, logical comparison, and array access
requires time to execute. We will assume each of these activities requires one unit of processor “time.” This
assumption is not strictly true, but it will give acceptable results for relative comparisons. Since we will
follow the same rules when analyzing both search algorithms, the relative results for comparison purposes
will be fairly accurate.
We first consider linear search:
int linear_search(const int a[], int n, int seek)
{
for ( int i = 0; i < n && a[i] <= seek; i++ )
if ( a[i] == seek )
return i; // Return position immediately
return -1; // Element not found
}

We determined that, on average, the loop makes n/2 iterations for an array of length n. The initialization of
i happens only one time during each call to linear_search. All other activity involved with the loop
except the return statements happens n/2 times. Either i or −1 will be returned, and only one return is
executed during each call. Table 12.2 shows the breakdown for linear search.
The running time of the linear_search function can thus be expressed as a simple linear function:
l(n) = 3n + 2.

©2013 Richard L. Halterman Draft date: January 27, 2013

12.3. SEARCH 329

Action Operations Operation Cost Iterations Cost

i = 0 = 1 1 1
i < size && a[i] <= seek <=,&&,[],<= 4 n/2 2n
a[i] == seek [],== 2 n/2 n
return i or return -1 return 1 1 1
Total Cost 3n + 2

Table 12.2: Analysis of Linear Search

Action Operations Operation Cost Iterations Cost

first = 0 = 1 1 1
last = n - 1 =,- 2 1 2
first <= last <= 1 log2 n log2 n
mid = first + (last - first + 1)/2 =,+,-,+,/ 5 log2 n 5 log2 n
v[mid] == seek [],== 2 log2 n 2 log2 n
v[mid] > seek [],> 2 log2 n 2 log2 n
last = mid - 1 or first = mid + 1 =,± 2 log2 n 2 log2 n
return mid or return -1 return 1 1 1
Total Cost 12 log2 n + 3

Table 12.3: Analysis of Binary Search

Next, we consider binary search:

int binary_search(const int a[], int n, int seek)

{
int first = 0, // Initially the first element in array
last = n - 1, // Initially the last element in array
mid; // The middle of the array
while ( first <= last )
{
mid = first + (last - first + 1)/2;
if ( a[mid] == seek )
return mid; // Found it
else if ( a[mid] > seek )
last = mid - 1; // continue with 1st half
else // a[mid] < seek
first = mid + 1; // continue with 2nd half
}
return -1; // Not there
}

We determined that in the worst case the loop in binary_search iterates log2 n times if the array contains
n elements. The two initializations before the loop are performed once per call. Most of the actions within
the loop occur log2 n times, except that only one return statement can be executed per call, and in the
if/else statement only one path can be chosen per call. Table 12.3 shows the complete analysis of binary
search.

©2013 Richard L. Halterman Draft date: January 27, 2013

 12.4. ARRAY PERMUTATIONS 330

We will call our binary search function b(n). Figure 12.3 shows the plot of the two functions l(n) =
3n + 2 and b(n) = 12 log2 n + 3. For n < 16, the linear function 3n + 2 is less than the binary function

250
Time in Seconds

200 Linear
Logarithmic
150

100

50

0
0 10 20 30 40 50 60 70 80 90

List Size

Figure 12.3: A graph of the functions derived from analyzing the linear and binary search routines

12 log2 n + 3. This means that linear search should perform better than binary search for array sizes less
than 16. This is because the code for linear search is less complicated, and it can complete its work on
smaller arrays before the binary search finishes its more sophisticated computations. At n = 16, however,
the two algorithms should perform about the same because

l(16) = 3(16) + 2 = 48 + 2 = 50 ≈ 51 = 48 + 3 = 12(4) + 3 = 12 log2 16 + 3 = b(16)

Figure 12.3 shows that for all n > 16 binary search outperforms linear search, and the performance gap
increases rapidly as n grows. This wide performance discrepancy agrees with our empirical observations
(Table 12.1). Binary search is fast even for very large arrays, while linear search is impractical for large
arrays.

12.4 Array Permutations

Sometimes it is useful to consider all the possible arrangements of the elements within an array. A sorting
algorithm, for example, must work correctly on any initial arrangement of elements in an array. To test a sort
function, a programmer could check to see to see if it produces the correct result for all arrangements of a
relatively small array. A rearrangement of a collection of ordered items is called a permutation. Listing 12.6
(arraypermutations.cpp) prints all the permutations of the contents of a given array.

Listing 12.6: arraypermutations.cpp

1 #include <iostream>
2
3 using namespace std;
4
5 /*
6 * print
7 * Prints the contents of an int array

©2013 Richard L. Halterman Draft date: January 27, 2013

 12.4. ARRAY PERMUTATIONS 331

8 * a is the array to print; a is not modified

9 * n is the number of elements in the array
10 */
11 void print(const int *a, int n)
12 {
13 cout << "{";
14 if ( n > 0 )
15 {
16 cout << a[0]; // Print the first element
17 for ( int i = 1; i < n; i++ )
18 cout << ',' << a[i]; // Print the rest
19 }
20 cout << "}";
21 }
22
23 /*
24 * Interchange the element at index m with the
25 * element at index n within array a. The function's
26 * behavior is undefined if m or n represent an
27 * index outside the bounds of array a.
28 */
29 void swap(int *a, int m, int n)
30 {
31 int temp = a[m];
32 a[m] = a[n];
33 a[n] = temp;
34 }
35
36 /*
37 * Prints all the permutations of array a in the
38 * index range begin...end, inclusive. The function's
39 * behavior is undefined if begin or end
40 * represents an index outside of the bounds of array a.
41 */
42 void permute(int *a, int begin, int end)
43 {
44 if ( begin == end )
45 {
46 print(a, end + 1);
47 cout << endl;
48 }
49 else
50 {
51 for ( int i = begin; i <= end; i++ )
52 {
53 // Interchange the element at the first position
54 // with the element at position i
55 swap(a, begin, i);
56 // Recursively permute the rest of the array
57 permute(a, begin + 1, end);
58 // Interchange the current element at the first position
59 // with the current element at position i
60 swap(a, begin, i);
61 }
62 }

©2013 Richard L. Halterman Draft date: January 27, 2013

 12.4. ARRAY PERMUTATIONS 332

63 }
64
65 /*
66 * Tests the permutation functions
67 */
68 int main()
69 {
70 // Get number of values from the user
71 cout << "Please enter number of values to permute: ";
72 int number;
73 cin >> number;
74 // Create the array to hold all the values
75 int *list = new int[number]; // Dynamically allocate space
76 // Initialize the array
77 for ( int i = 0; i < number; i++ )
78 list[i] = i;
79
80 // Print original list
81 print(list, number);
82 cout << endl << "----------" << endl;
83 // Print all the permutations of list
84 permute(list, 0, number - 1);
85 cout << endl << "----------" << endl;
86 // Print list after all the manipulations
87 print(list, number);
88 delete [] list; // Free up space held by list array
89 }

A sample run of Listing 12.6 (arraypermutations.cpp) when the user enters 4 prints

©2013 Richard L. Halterman Draft date: January 27, 2013

12.4. ARRAY PERMUTATIONS 333

Please enter number of values to permute: 4

{0,1,2,3}
----------
{0,1,2,3}
{0,1,3,2}
{0,2,1,3}
{0,2,3,1}
{0,3,2,1}
{0,3,1,2}
{1,0,2,3}
{1,0,3,2}
{1,2,0,3}
{1,2,3,0}
{1,3,2,0}
{1,3,0,2}
{2,1,0,3}
{2,1,3,0}
{2,0,1,3}
{2,0,3,1}
{2,3,0,1}
{2,3,1,0}
{3,1,2,0}
{3,1,0,2}
{3,2,1,0}
{3,2,0,1}
{3,0,2,1}
{3,0,1,2}

----------
{0,1,2,3}

The permute function in Listing 12.6 (arraypermutations.cpp) is a recursive function, as it calls itself
inside of its definition. We have seen how recursion can be an alternative to iteration (see for example List-
ing 11.11 (recursivearrayprint.cpp)); however, the permute function here uses both iteration and recursion
together to generate all the arrangements of an array. At first glance, the combination of these two algorithm
design techniques as used here may be difficult to follow, but we actually can understand the process better
if we ignore some of the details of the code.
First, notice that in the recursive call begin is one larger, and end remains the same. This means as
the recursion progresses the ending index never changes, and the beginning index keeps increasing until it
reaches the ending index. The recursion terminates when begin becomes equal to end.
In its simplest form the function looks like this:
void permute(int *a, int begin, int end)
{
if ( begin == end )
// Print the whole array
else
// Do the interesting part of the algorithm

©2013 Richard L. Halterman Draft date: January 27, 2013

12.4. ARRAY PERMUTATIONS 334

Let us zoom in on the interesting part of the algorithm:

for ( int i = begin; i <= end; i++ )
{
swap(a, begin, i);
permute(a, begin + 1, end);
swap(a, begin, i);
}

If the mixture of iteration and recursion is confusing, eliminate iteration!

If a loop iterates a fixed number of times, you may replace the loop with the statements in its body
duplicated that number times; for example, the code
for ( int i = 0; i < 5; i++ )
cout << i << endl;

may be rewritten as
cout << 0 << endl;
cout << 1 << endl;
cout << 2 << endl;
cout << 3 << endl;
cout << 4 << endl;

Notice that the loop is gone. This process of transforming a loop into the series of statements that the
loop would perform is known as loop unrolling. Compilers sometimes unroll loops to make the code’s
execution faster. After unrolling the loop the loop control variable (in this case i) is gone, so there is no
need to initialize it (done once) and, more importantly, no need to check its value and update it during each
iteration of the loop.
Our purpose for unrolling the loop in perform is not to optimize it. Instead we are trying to understand
better how the algorithm works. In order to unroll perform’s loop, we will consider the case for arrays
containing exactly three elements. In this case the for in the perform function would be hardcoded as
for ( int i = 0; i <= 2; i++ )
{
swap(a, begin, i);
permute(a, begin + 1, end);
swap(a, begin, i);
}

and we can transform this code into

swap(a, begin, 0);
permute(a, begin + 1, end);
swap(a, begin, 0);

swap(a, begin, 1);

permute(a, begin + 1, end);
swap(a, begin, 1);

©2013 Richard L. Halterman Draft date: January 27, 2013

12.4. ARRAY PERMUTATIONS 335

swap(a, begin, 2);

permute(a, begin + 1, end);
swap(a, begin, 2);

Once the loop is gone, we see we have simply a series of recursive calls of permute sandwiched by calls
to swap. The first call to swap interchanges an element in the array with the first element. The second
call to swap reverses the effects of the first swap. This series of swap-permute-swap operations allows
each element in the array to have its turn being the first element in the permuted array. The permute
recursive call generates all the permutations of the rest of the list. Figure 12.4 traces the recursive process
of generating all the permutations of the array {0,1,2}. The leftmost third of Figure 12.4 shows the

i =1
3 {0,1,2}
2 {0,1,2} 4
5
i =0 {0,1,2}
6
9 7
{0,2,1} 8 {0,2,1}
i =2
1
10

i =1
12 {1,0,2} 13 {1,0,2}
i =1 14
11 15
{0,1,2} 20 {1,0,2}
16
19
{1,2,0} 17 {1,2,0}
18
i =2
21 0
3

i =2
i =1
23
22 {2,1,0} 24 {2,1,0}
25
{2,1,0}
26
27
29 {2,0,1} 28 {2,0,1}
i =2

Initial call to permute Recursive call Base case

begin = 0 begin = 1 begin = 2
end = 2 end = 2 end = 2

Figure 12.4: A tree mapping out the recursive process of the permute function operating on the array {}.

original contents of the array and the initial call of permute. The three branches represent the three
iterations of the for loop: i varying from begin (0) to end (2). The arrays indicate the state of the array
after the first swap but before the recursive call to permute.
The middle third of Figure 12.4 shows the state of the array during the first recursive call to permute.
The two branches represent the two iterations of the for loop: i varying from begin (1) to end (2). The
arrays indicate the state of the array after the first swap but before the next recursive call to permute. At
this level of recursion the element at index zero is fixed, and the remainder of the processing during this

©2013 Richard L. Halterman Draft date: January 27, 2013

 12.4. ARRAY PERMUTATIONS 336

chain of recursion is restricted to indices greater than zero.

The rightmost third of Figure 12.4 shows the state of the array during the second recursive call to
permute. At this level of recursion the elements at indices zero and one are fixed, and the remainder
of the processing during this chain of recursion is restricted to indices greater than one. This leaves the
element at index two, but this represents the base case of the recursion because begin (2) equals end (2).
The function makes no more recursive calls to itself. The function merely prints the current contents of the
array.
The arrows in Figure 12.4 represent a call to, or a return from, permute. They illustrate the recursive
call chain. The arrows pointing left to right represent a call, and the arrows pointing from right to left
represent a return from the function. The numbers associated with arrow indicate the order in which the
calls and returns occur during the execution of permute.
The second column from the left shows the original contents of the array after the first swap call but
before the first recursive call to permute. The swapped elements appear in red. The third column shows
the contents of the array at the second level of recursion. In the third column the elements at index zero
are fixed, as this recursion level is using begin with a value of one instead of zero. The for loop within
this recursive call swaps the elements highlighted in red. The rightmost column is the point where begin
equals end, and so the permute function does not call itself effectively terminating the recursion.
While Listing 12.6 (arraypermutations.cpp) is a good exercise in array manipulation and recursion,
the C++ standard library provides a function named next_permutation that rearranges the elements
of an array. Listing 12.7 (stlpermutations.cpp) uses next_permutation within a loop to print all the
permutations of the array’s elements.

Listing 12.7: stlpermutations.cpp

1 #include <iostream>
2 #include <algorithm>
3
4 using namespace std;
5
6 /*
7 * print
8 * Prints the contents of an int array
9 * a is the array to print; a is not modified
10 * n is the number of elements in the array
11 */
12 void print(const int *a, int n)
13 {
14 cout << "{";
15 if ( n > 0 )
16 {
17 cout << a[0]; // Print the first element
18 for ( int i = 1; i < n; i++ )
19 cout << ',' << a[i]; // Print the rest
20 }
21 cout << "}";
22 }
23
24
25 int main()
26 {
27 int nums[] = { 0, 1, 2, 3 };
28 cout << "---------------" << endl;

©2013 Richard L. Halterman Draft date: January 27, 2013

 12.5. RANDOMLY PERMUTING AN ARRAY 337

29 do
30 {
31 print(nums, 4);
32 cout << endl;
33 } // Compute the next ordering of elements
34 while ( next_permutation(nums, nums + 4) );
35 }

12.5 Randomly Permuting an Array

Section 12.4 showed how we can generate all the permutations of an array in an orderly fashion. More
often, however, we need to produce one of those permutations chosen at random. For example, we may
need to randomly rearrange the contents of an ordered array so that we can test a sort function to see if
it will produce the original ordered sequence. We could generate all the permutations, put each one in
an array of arrays, and select a permutation at random from that array of arrays. This approach is very
inefficient, especially as the length of the array to permute grows larger. Fortunately, we can randomly
permute the contents of an array easily and quickly. Listing 12.8 (randompermute.cpp) contains a function
named permute that randomly permutes the elements of an array.

Listing 12.8: randompermute.cpp

1 #include <iostream>
2 #include <ctime>
3
4 using namespace std;
5
6 /*
7 * print
8 * Prints the contents of an int array
9 * a is the array to print; a is not modified
10 * n is the number of elements in the array
11 */
12 void print(const int *a, int n)
13 {
14 cout << "{";
15 if ( n > 0 )
16 {
17 cout << a[0]; // Print the first element
18 for ( int i = 1; i < n; i++ )
19 cout << ',' << a[i]; // Print the rest
20 }
21 cout << "}";
22 }
23
24 /*
25 * Returns a pseudorandom number in the range begin...end - 1,
26 * inclusive. Returns 0 if begin >= end.
27 */
28 int random(int begin, int end)
29 {
30 if ( begin >= end )
31 return 0;

©2013 Richard L. Halterman Draft date: January 27, 2013

 12.5. RANDOMLY PERMUTING AN ARRAY 338

32 else
33 {
34 int range = end - begin;
35 return begin + rand()%range;
36 }
37 }
38
39 /*
40 * Randomly permute an array of integers.
41 * a is the array to permute, and n is its length.
42 */
43 void permute(int *a, int n)
44 {
45 for ( int i = 0; i < n - 1; i++ )
46 { // Select a pseudorandom location from the current
47 // location to the end of the collection
48 swap(a[i], a[random(i, n)]);
49 }
50 }
51
52 // Tests the permute function that randomly permutes the
53 // contents of an array
54 int main()
55 {
56 // Initialize random generator seed
57 srand(static_cast<int>(time(0)));
58
59 // Make the array {1,2,3,4,5,6,7,8}
60 int arr[] = { 1, 2, 3, 4, 5, 6, 7, 8 },
61 size = (sizeof arr)/(sizeof arr[0]);
62
63 // Print array before
64 print(arr, size);
65 cout << endl;
66
67 permute(arr, size);
68
69 // Print array after
70 print(arr, size);
71 cout << endl;
72 }

One run of Listing 12.8 (randompermute.cpp) produces

1 2 3 4 5 6 7 8
2 7 1 6 4 8 3 5

Notice that the permute function in Listing 12.8 (randompermute.cpp) uses a simple un-nested loop and
no recursion. The permute function varies the i index variable from 0 to the index of the next to last
element in the array. An index greater than i is chosen pseudorandomly using rand (see Section 8.6), and
the elements at position i and the random position are exchanged. At this point all the elements at index i

©2013 Richard L. Halterman Draft date: January 27, 2013

 12.5. RANDOMLY PERMUTING AN ARRAY 339

and smaller are fixed and will not change as the function’s execution continues. The index i is incremented,
and the process continues until all the i values have been considered.
To be correct, our permute function must be able to generate any valid permutation of the array. It is
important that our permute function is able to produce all possible permutations with equal probability;
said another way, we do not want our permute function to generate some permutations more often than
others. The permute function in Listing 12.8 (randompermute.cpp) is fine, but consider a slight variation
of the algorithm:
// Randomly permute an array?
void faulty_permute(int *a, int n)
{
for ( int i = 0; i < n; i++ )
{ // Select a pseudorandom position somewhere in the array
swap(a[i], a[random(0, n)]);
}
}

Do you see the difference between faulty_permute and permute? In faulty_permute, the
random index is chosen from all valid array indices, whereas permute restricts the random index to
valid indices greater than or equal to i. This means that any element within array a can be exchanged
with the element at position i during any loop iteration. While this approach may superficially appear
to be just as good as permute, it in fact produces an uneven distribution of permutations. Listing 12.9
(comparepermutations.cpp) exercises each permutation function 1,000,000 times on the array {1, 2, 3}
and tallies each permutation. There are exactly six possible permutations of this three-element array.

Listing 12.9: comparepermutations.cpp

1 #include <iostream>
2 #include <ctime>
3
4 using namespace std;
5
6 /*
7 * print
8 * Prints the contents of an int array
9 * a is the array to print; a is not modified
10 * n is the number of elements in the array
11 */
12 void print(const int *a, int n)
13 {
14 cout << "{";
15 if ( n > 0 )
16 {
17 cout << a[0]; // Print the first element
18 for ( int i = 1; i < n; i++ )
19 cout << ',' << a[i]; // Print the rest
20 }
21 cout << "}";
22 }
23
24 /*
25 * Copy the contents of one array into another.
26 * src is the source array.
27 * dest is the destination array.

©2013 Richard L. Halterman Draft date: January 27, 2013

 12.5. RANDOMLY PERMUTING AN ARRAY 340

28 * n is the number of elements to copy.

29 */
30 void copy(const int *src, int *dest, int n)
31 {
32 for ( int i = 0; i < n; i++ )
33 dest[i] = src[i];
34 }
35
36 /*
37 * Returns a pseudorandom number in the range begin...end - 1,
38 * inclusive. Returns 0 if begin >= end.
39 */
40 int random(int begin, int end)
41 {
42 if ( begin >= end )
43 return 0;
44 else
45 {
46 int range = end - begin;
47 return begin + rand()%range;
48 }
49 }
50
51 /*
52 * Randomly permute an array of integers.
53 * a is the array to permute, and n is its length.
54 */
55 void permute(int *a, int n)
56 {
57 for ( int i = 0; i < n - 1; i++ )
58 { // Select a pseudorandom location from the current
59 // location to the end of the collection
60 swap(a[i], a[random(i, n)]);
61 }
62 }
63
64 /* Randomly permute an array? */
65 void faulty_permute(int *a, int n)
66 {
67 for ( int i = 0; i < n; i++ )
68 { // Select a pseudorandom position somewhere in the collection
69 swap(a[i], a[random(0, n)]);
70 }
71 }
72
73 /* Classify an array as one of the six permutations */
74 int classify(const int *a)
75 {
76 switch ( 100*a[0] + 10*a[1] + a[2] )
77 {
78 case 123: return 0;
79 case 132: return 1;
80 case 213: return 2;
81 case 231: return 3;
82 case 312: return 4;

©2013 Richard L. Halterman Draft date: January 27, 2013

 12.5. RANDOMLY PERMUTING AN ARRAY 341

83 case 321: return 5;

84 }
85 return -1;
86 }
87
88 /* Report the accumulated statistics */
89 void report(const int *a)
90 {
91 cout << "1,2,3: " << a[0] << endl;
92 cout << "1,3,2: " << a[1] << endl;
93 cout << "2,1,3: " << a[2] << endl;
94 cout << "2,3,1: " << a[3] << endl;
95 cout << "3,1,2: " << a[4] << endl;
96 cout << "3,2,1: " << a[5] << endl;
97 }
98
99 /*
100 * Fill the given array with zeros.
101 * a is the array, and n is its length.
102 */
103 void clear(int *a, int n)
104 {
105 for ( int i = 0; i < n; i++ )
106 a[i] = 0;
107 }
108
109 int main()
110 {
111 // Initialize random generator seed
112 srand(static_cast<int>(time(0)));
113
114 // permutation_tally array keeps track of each permutation pattern
115 // permutation_tally[0] counts {1,2,3}
116 // permutation_tally[1] counts {1,3,2}
117 // permutation_tally[2] counts {2,1,3}
118 // permutation_tally[3] counts {2,3,1}
119 // permutation_tally[4] counts {3,1,2}
120 // permutation_tally[5] counts {3,2,1}
121 int permutation_tally[] = { 0, 0, 0, 0, 0, 0 };
122
123 // original always holds the array {1,2,3}
124 const int original[] = { 1, 2, 3 };
125 const int size = (sizeof original)/(sizeof original[0]);
126
127 // working holds a copy of original is gets permuted and tallied
128 int working[size];
129
130 // Run each permutation one million times
131 const int RUNS = 1000000;
132
133 cout << "--- Random permute #1 -----" << endl;
134 clear(permutation_tally, 6);
135 for ( int i = 0; i < RUNS; i++ ) // Run 1,000,000 times
136 {
137 // Make a copy of the original array

©2013 Richard L. Halterman Draft date: January 27, 2013

 12.5. RANDOMLY PERMUTING AN ARRAY 342

138 copy(original, working, size);

139 // Permute the array with the first algorithm
140 permute(working, size);
141 // Count this permutation
142 permutation_tally[classify(working)]++;
143 }
144 report(permutation_tally); // Report results
145
146 cout << "--- Random permute #2 -----" << endl;
147 clear(permutation_tally, 6);
148 for ( int i = 0; i < RUNS; i++ ) // Run 1,000,000 times
149 {
150 // Make a copy of the original array
151 copy(original, working, size);
152 // Permute the array with the second algorithm
153 faulty_permute(working, size);
154 // Count this permutation
155 permutation_tally[classify(working)]++;
156 }
157 report(permutation_tally); // Report results
158 }

In Listing 12.9 (comparepermutations.cpp)’s output, permute #1 corresponds to our original permute

function, and permute #2 is the faulty_permute function. The output of Listing 12.9 (comparepermutations.cpp)
reveals that the faulty permutation function favors some permutations over others:

--- Random permute #1 -----

1,2,3: 166608
1,3,2: 166820
2,1,3: 166350
2,3,1: 166702
3,1,2: 166489
3,2,1: 167031
--- Random permute #2 -----
1,2,3: 148317
1,3,2: 184756
2,1,3: 185246
2,3,1: 185225
3,1,2: 148476
3,2,1: 147980

In one million runs, the permute function provides an even distribution of the six possible permutations of
{1, 2, 3}. The faulty_permute function generates the permutations {1, 3, 2}, {2, 1, 3},
and {2, 3, 1} more often then the permutations {1, 2, 3}, {3, 1, 2}, and {3, 2, 1}.
To see why faulty_permute misbehaves, we need to examine all the permutations it can produce
during one call. Figure 12.5 shows a hierarchical structure that maps out how faulty_permute trans-
forms the contents of its array parameter each time through the for loop. The top of the tree shows the
original array, {1, 2, 3}. The second row shows the three possible resulting configurations after the
first iteration of the for loop. The leftmost 3-tuple represents the element at index zero swapped with

©2013 Richard L. Halterman Draft date: January 27, 2013

12.5. RANDOMLY PERMUTING AN ARRAY 343

123

123 213 321

213 123 132 123 213 231 231 321 312

312 231 213 321 132 123 231 123 132 321 132 123 312 231 213 132 213 231 132 213 231 123 312 321 213 321 312

Figure 12.5: A tree mapping out the ways in which faulty permute can transform the array 1, 2, 3 at each
iteration of its for loop

the element at index zero (effectively no change). The second 3-tuple on the second row represents the
interchange of the elements at index 0 and index 1. The third 3-tuple on the second row results from the
interchange of the elements at positions 0 and 2. The underlined elements represent the elements most
recently swapped. If only one item in the 3-tuple is underlined, the function merely swapped the item with
itself. The bottom row of 3-tuples contains all the possible outcomes of the faulty_permute function
given the array {1, 2, 3}.
As Figure 12.5 shows, the array {1, 3, 2}, {2, 1, 3}, and {2, 3, 1} each appear five times
in the last row, while {1, 2, 3}, {3, 1, 2}, and {3, 2, 1} each appear only four times. There
4
are a total of 27 possible outcomes, so some permutations appear = 14.815% of the time, while the
27
5
others appear = 18.519% of the time. Notice that these percentages agree with our experimental results
27
from Listing 12.9 (comparepermutations.cpp).

123

123 213 321

123 132 213 231 321 312

Figure 12.6: A tree mapping out the ways in which permute can transform the array 1, 2, 3 at each iteration
of its for loop

Compare Figure 12.5 to Figure 12.6. The second row of the tree for permute is identical to the
second row of the tree for faulty_permute, but the third row is different. The second time through
its loop the permute function does not attempt to exchange the element at index zero with any other
elements. We see that none of the first elements in the 3-tuples in row three are underlined. The third row
contains exactly one instance of each of the possible permutations of {1, 2, 3}. This means that the
correct permute function is not biased towards any of the individual permutations, and so the function can

©2013 Richard L. Halterman Draft date: January 27, 2013

12.6. SUMMARY 344

1
generate all the permutations with equal probability. The permute function has a = 16.667% probability
6
of generating a particular permutation; this number agrees with our the experimental results of Listing 12.9
(comparepermutations.cpp).

12.6 Summary

• Various algorithms exist for sorting arrays. Selection sort is a simple algorithm for sorting an array.

• Linear search is useful for finding elements in an unordered array. Binary search can be used on
ordered arrays, and due to the nature of the algorithm, binary search is very fast, even on large arrays.

• Permutations

• More permutations

12.7 Exercises

1. Complete the following function that reorders the contents of an array so they are reversed from
their original order. For example, an array containing the elements 2, 6, 2, 5, 0, 1, 2, 3 would be
transformed into 3, 2, 1, 0, 5, 2, 6, 2. Note that your function must physically rearrange the elements
within the array, not just print the elements in reverse order.
// Array a with length n
void reverse(int a[], int n)
{
// Add your code...
}

2. Complete the following function that reorders the contents of an array so that all the even numbers
appear before any odd number. The even values are sorted in ascending order with respect to them-
selves, and the odd numbers that follow are also sorted in ascending order with respect to themselves.
For example, an array containing the elements 2, 1, 10, 4, 3, 6, 7, 9, 8, 5 would be transformed into
2, 4, 6, 8, 10, 1, 3, 5, 7, 9 Note that your function must physically rearrange the elements within the
array, not just print the elements in reverse order.
// Array a with length n
void special_sort(int *a, int n)
{
// Add your code...
}

3. Complete the following function that shifts all the elements of an array backward one place. The
last element that gets shifted off the back end of the array is copied into the first (0th) position. For
example, if an array containing the elements 2, 1, 10, 4, 3, 6, 7, 9, 8, 5 is passed to the function,
it would be transformed into 5, 2, 1, 10, 4, 3, 6, 7, 9, 8 Note that your function must physically
rearrange the elements within the array, not just print the elements in the shifted order.

©2013 Richard L. Halterman Draft date: January 27, 2013

12.7. EXERCISES 345

// Array a with length n

void rotate(int a[], int n)
{
// Add your code...
}

4. Complete the following function that determines if the number of even and odd values in an array is
the same. The function would return true if the array contains 5, 1, 0, 2 (two evens and two odds),
but it would return false for the array containing 5, 1, 0, 2, 11 (too many odds). The function should
return true if the array is empty, since an empty array contains the same number of evens and odds
(0). The function does not affect the contents of the array.
// Array a with length n
bool balanced(const int *a, int n)
{
// Add your code...
}

5. Complete the following function that returns true if array a contains duplicate elements; it returns
false if all the elements in a are unique. For example, the array 2, 3, 2, 1, 9 contains duplicates (2
appears more than once), but the array 2, 1, 0, 3, 8, 4 does not (none of the elements appear more
than once).
An empty array has no duplicates.
The function does not affect the contents of the array.
// Array a with length n
bool has_duplicates(const int *a, int n)
{
// Add your code...
}

6. Can binary search be used on an unsorted array? Why or why not?

7. Can linear search be used on an unsorted array? Why or why not?

©2013 Richard L. Halterman Draft date: January 27, 2013

12.7. EXERCISES 346

©2013 Richard L. Halterman Draft date: January 27, 2013

 347

Chapter 13

Standard C++ Classes

In the hardware arena, a desktop computer is built by assembling

• a motherboard (a circuit board containing sockets for a processor and assorted supporting cards),
• a processor,
• memory,
• a video card,
• an input/output card (USB ports, parallel port, and mouse port),
• a disk controller,
• a disk drive,
• a case,
• a keyboard,
• a mouse, and
• a monitor.

(Some of these components like the I/O, disk controller, and video may be integrated with the mother-
board.)
The video card is itself a sophisticated piece of hardware containing a video processor chip, memory,
and other electronic components. A technician does not need to assemble the card; the card is used as is
off the shelf. The video card provides a substantial amount of functionality in a standard package. One
video card can be replaced with another card from a different vendor or with another card with different
capabilities. The overall computer will work with either card (subject to availability of drivers for the
operating system), because standard interfaces allow the components to work together.
Software development today is increasingly component based. Software components are used like
hardware components. A software system can be built largely by assembling pre-existing software building
blocks. C++ supports various kinds of software building blocks. The simplest of these is the function that
we investigated in Chapter 8 and Chapter 9. A more powerful technique uses built-in and user designed
software objects.

©2013 Richard L. Halterman Draft date: January 27, 2013

13.1. STRING OBJECTS 348

C++ is object-oriented. It was not the first OO programming language, but it was the first OO language
that gained widespread use in a variety of application areas. An OO programming language allows the
programmer to define, create, and manipulate objects. Variables representing objects can have considerable
functionality compared to the primitive numeric variables like ints and doubles. Like a normal variable,
every C++ object has a type. We say an object is an instance of a particular class, and class means the same
thing as type. An object’s type is its class. We have been using the cout and cin objects for some time.
cout is an instance of the ostream class—which is to say cout is of type ostream. cin is an instance
of the istream class.
Code that uses an object is a client of that object; for example, the following code fragment
cout << "Hello" << endl;

uses the cout object and, therefore, is a client of cout. Many of the functions we have seen so far have
been clients of the cout and/or cin objects. Objects provide services to their clients.

13.1 String Objects

As mentioned in Section 11.7, a string is a sequence of characters. A C string in particular is a null-

terminated array of characters. As do arrays, C strings present a number of challenges. Programmers must
ensure proper memory management (allocation and deallocation, Section 11.8) and be vigilant to prevent
buffer overruns (Section 11.7). Simple assignment of char * variables produces aliases, which usually
is not the programmer’s intention. Proper assignment and comparison of C strings requires specialized
functions such as strcpy and strcmp. As with all primitive arrays, a C string does not keep track of its
own length; programmers must use the strlen function which iterates over all the characters in the string
until it finds the null character.
The C++ standard library provides the class string which specifies string objects. A string variable
is an example of a C++ object. Objects can be more sophisticated than simple primitive-type variables. If
variable x has been declared to be of type int, we may assign a value to it, read its value, and find its
address in memory, but that is all. On the other hand, a single object can represent multiple values of
differing types and can provide special operations that allow clients of the object to do much more than
simply assign it, look at it, and find its address. These special operations are called methods or member
functions, and they are invoked with a special syntax.
A string object is declared like any other variable:
string name;

A string object may be assigned literal character sequences via the familiar string quotation syntax:
string name = "joe";
cout << name << endl;
name = "jane";
cout << name << endl;

In order to use string objects, you must provide the preprocessor directive
#include <string>

The string class is part of the standard namespace, which means their full type name is std::string.
If you use the

©2013 Richard L. Halterman Draft date: January 27, 2013

13.1. STRING OBJECTS 349

using namespace std;

or
using std::string;

statements in your code, you can use the abbreviated name string.
One string object may be assigned to another using the simple assignment operator:
string name1 = "joe", name2;
name2 = name1;
cout << name1 << " " << name2 << endl;

In this case, the contents of name1 are copied into name2. After the assignment both name1 and name2
have their own copies of the characters that make up the string; they do not share their contents. After the
assignment, changing one string will not affect the other string. Code within the string class defines how
the assignment operator should work in the context of string objects.
The string class provide a number of methods. To invoke a method on behalf of an object, you
ordinarily write the object’s name, then a dot (.), then the name of the method with any required parameters
within parentheses. A method call looks like a regular function call, but it must be used with an object of
a type that supports that method. One method that the string class provides is called length. The
following code fragment
string word = "computer";
cout << "\"" << word << "\" contains " << word.length()
<< " letters." << endl;

prints

"computer" contains 8 letters.

The expression:
word.length()

invokes the length method on behalf of the word object. Like functions, some methods expect parame-
ters, but the length method does not accept any arguments.
C++ allows classes to define the way operators work in the context of their objects. For example, the
string class defines a method named operator[] that allows a programmer to access a character
within a string, as in
cout << "The letter at index 3 is " << word.operator[](3) << endl;

Here the operator[] method uses the same syntax as the length method, but the operator[]
method expects a single integer parameter. The word operator is a C++ reserved word, and it means that
the method can be invoked with a special syntax. Here we see that operator[] is a method that can be
used with string objects. The above code fragment is better written as
cout << "The letter at index 3 is " << word[3] << endl;

©2013 Richard L. Halterman Draft date: January 27, 2013

 13.1. STRING OBJECTS 350

The expression
word.operator[](3)

is equivalent to the expression

word[3]

Programmers can use the familiar array indexing notation with string objects resulting in an easier to read
syntax. We say the array indexing operator, [], is overloaded for the string class. The term syntactic
sugar identifies a language feature that allows programmers to express a more complicated, less attractive
expression of code in a cleaner, more attractive or more natural fashion. The string class provides a
number of other overloaded operator methods that sweeten the syntax of string operations.
The following code fragment uses some of the methods available to string objects:

Listing 13.1: stringoperations.cpp

1 #include <iostream>
2 #include <string>
3
4 using namespace std;
5
6 int main()
7 {
8 // Declare a string object and initialize it
9 string word = "fred";
10 // Prints 4, since word contains four characters
11 cout << word.length() << endl;
12 // Prints "not empty", since word is not empty
13 if ( word.empty() )
14 cout << "empty" << endl;
15 else
16 cout << "not empty" << endl;
17 // Makes word empty
18 word.clear();
19 // Prints "empty", since word now is empty
20 if ( word.empty() )
21 cout << "empty" << endl;
22 else
23 cout << "not empty" << endl;
24 // Assign a string using operator= method
25 word = "good";
26 // Prints "good"
27 cout << word << endl;
28 // Append another string using operator+= method
29 word += "-bye";
30 // Prints "good-bye"
31 cout << word << endl;
32 // Print first character using operator[] method
33 cout << word[0] << endl;
34 // Print last character
35 cout << word[word.length() - 1] << endl;
36 // Prints "od-by", the substring starting at index 2 of length 5
37 cout << word.substr(2, 5);
38 string first = "ABC", last = "XYZ";

©2013 Richard L. Halterman Draft date: January 27, 2013

 13.2. VECTORS 351

39 // Splice two strings with + operator

40 cout << first + last << endl;
41 cout << "Compare " << first << " and ABC: ";
42 if ( first == "ABC" )
43 cout << "equal" << endl;
44 else
45 cout << "not equal" << endl;
46 cout << "Compare " << first << " and XYZ: ";
47 if ( first == "XYZ" )
48 cout << "equal" << endl;
49 else
50 cout << "not equal" << endl;
51 }

The statement
word = "good";

is equivalent to
word.operator=("good");

Here we see the explicit use of the dot (.) operator to invoke the method. Similarly,
word += "-bye";

is the syntactally sweetened way to write

word.operator+=("-bye");

The + operator performs string concatenation, making a new string by appending one string to the back of
another.
In addition to string methods, the standard string library provides a number of global functions that
process strings. These functions use operator syntax and allow us to compare strings via <, ==, >=, etc.
A more complete list of string methods and functions can be found at http://www.cplusplus.com/
reference/string/.

13.2 Vectors

As we saw in Chapter 11, arrays are useful for holding a collection of values. Arrays have their limitations,
however:

• The space for the contents of an array must be allocated up front. The programmer needs to know
how many items the array is to hold before any items can be placed into the array.
• Once an array is allocated, its size is fixed. If you need to add more elements, you must allocate a
new array of a larger size, copy all the existing elements into the new array, and then add the new
elements to the new array. If the original array was dynamically allocated (see Section 11.8), you
must explicitly free up its space via the delete operator to avoid a memory leak.
• Out-of-bounds accesses are logic errors that may not manifest themselves immediately during testing
and may be difficult to locate and fix.

©2013 Richard L. Halterman Draft date: January 27, 2013

13.2. VECTORS 352

• Dynamically allocated arrays must be deallocated when no longer needed to prevent memory leaks.
Other memory management problems include accidentally using delete more than once for the
same memory and accidentally using delete to deallocate memory that was not allocated via new.
Such actions result in undefined behavior that can corrupt memory and cause the program to crash.

• Arrays cannot keep track of their own size. An array name represents merely the memory address of
the array’s first element. The programmer must use an additional variable to keep track of an array’s
size and must pass an explicit size parameter to any functions that process arrays.

• It is not easy to represent an empty array; all arrays must contain at least one element.

Arrays are low-level data structures that are error prone and often inconvenient for programmers to
use. When it comes to a built-in data structure that holds a linear collection of elements, the array is the
only option that the C programming language provides. C++ has a more convenient alternative to arrays
called vectors. Defined in the vector class, a vector object uses an array internally, but it insulates the
programmer from many of the troublesome aspects of arrays.

13.2.1 Declaring and Using Vectors

Like an array, a vector is a variable that can hold multiple values simultaneously; it represents a collection
of values. A vector has a name, and the values it contains are accessed via their position within the vector.
A vector stores a sequence of values, and the values must all be of the same type.
In order to use a vector within your program, you must add the preprocessor directive
#include <vector>

The vector type is part of the standard (std) namespace, so its full name is std::vector, just like
the full name of cout is std::cout (see Section 2.1). If you include the statement
using namespace std;

or
using std::vector;

at the top of your code, the shorter name vector can be used.
Like any other variable, a vector can be local or global, and it must be declared before it is used. The
following code fragment declares a vector named list that can hold integer values:
vector<int> list;

The angle brackets (<>) indicate the type of values that the vector can contain. list is a vector that holds
int elements. Any valid C++ type can appear within the angle brackets. A vector declared this way is
initially empty.
Vectors support a number of methods, but we will focus on five of them:

• push_back—inserts a new element onto the back of a vector

• pop_back—removes the last element from a vector

• operator[]—provides access to the value stored at a given index within the vector

©2013 Richard L. Halterman Draft date: January 27, 2013

13.2. VECTORS 353

• at—provides bounds-checking access to the value stored at a given position within the vector

• size—returns the number of values currently stored in the vector.

• clear—makes the vector empty.

In order to add an element to a vector, we use the push_back method as follows:

vector<int> list; // Declare list to be a vector
list.push_back(5); // Add 5 to the end of list
list.push_back(-3); // Add -3 to the end of the list
list.push_back(12); // Add 12 to the end of list

After executing the three push_back calls above, the list vector conceptually looks like Fig-
ure 13.1. The size of the vector adjusts automatically as new elements are inserted onto the back.

5 -3 12
0 1 2

Figure 13.1: A simple vector with three elements

The pop_back method removes the last element from the vector:
vector<int> list; // Declare list to be a vector
list.push_back(5); // Add 5 to the end of list
list.push_back(-3); // Add -3 to the end of the list
list.push_back(12); // Add 12 to the end of list
list.pop_back(); // Removes 12 from the list
list.pop_back(); // Removes -3 from the list

The list vector defined above is initially empty. It has no elements until the programmer uses
push_back to insert items into the vector. It is a logic error to attempt to use the [] operator on an
empty vector. It is possible to declare a vector with a beginning size:
vector<double> nums(10);

In this case nums initially holds 10 double-precision floating-point values. Each of the values by default is
equal to 0.0. A second argument within the parentheses can specify a different initial value for the elements:
vector<int> collection(20, 8);

Here collection is a vector of 20 integers, all equal to 8. Even though a vector can be declared with
an initial capacity, we still can add elements with push_back and increase the vector’s size beyond the
initial specification.

©2013 Richard L. Halterman Draft date: January 27, 2013

13.2. VECTORS 354

The C++11 standard allows vectors to be initialized with an initialization list in a

manner similar to array initialization:
vector<int> collection = { 2, 4, 6, 8 };
Not all C++ compilers currently support this initialization list syntax; in particular
Visual C++ 11.0 in Visual Studio 2012 does not support it.

These different ways of initializing vectors takes advantage of a feature that many classes provide—
constructors. A constructor provides custom code that executes when an object is created. The vector
class provides a number of different constructors that allow vector objects to be initialized in different ways.
See Section 14.4 for more information about constructors.
As with an array, the programmer must ensure that the index used is within the bounds of the vector.
Since the index may consist of an arbitrary integer expression whose value cannot be determined until run
time, the compiler cannot check for out-of-bound vector accesses; for example, in the code
int x;
vector<int> v(10); // Make a vector with 10 spaces available
cin >> x; // User enters x at run time
v[x] = 1; // Is this OK? What is x?

the compiler cannot predict what number the user will enter. This means that, just as with an array, the
misuse of a vector index with operator[] can lead to undefined behavior. In this way, vectors are no
safer than raw arrays. Vectors provide an additional method, at, that works exactly like operator[],
but at performs run-time bounds checking on the access. Instead of
v[x] = 1;

one can write

v.at(x) = 1;

If the index is out of bounds, the at method guarantees its detection.

Vectors model one-dimensional, a sequential lists of values. Since a vector can hold values of any type,
it is possible to declare a vector of vectors, as in
vector<vector<int>> table;

Here table is a vector holding vectors of integers. The table variable thus represents the equivalent of
a 2D array.

13.2.2 Copying Vectors

In Section 11.10 we saw that array assignment is more complicated than simple variable assignment. With
arrays, the programmer must ensure that enough space is available in the destination array and then copy
individually each element in the source array into the destination array. Vector assignment, on the other
hand, is as easy as with any of the simple types such as int. Like the string class, the vector class
overloads operator=. When one vector is assigned to another vector, the elements of one vector are
copied into the other; for example,

©2013 Richard L. Halterman Draft date: January 27, 2013

 13.2. VECTORS 355

vector<int> vec(3, 4), // vec contains 4,4,4

nums(4, 1); // nums contains 1,1,1,1
vec = nums; // vec now contains 1,1,1,1
nums[0] = 2; // nums is 2,1,1,1, but vec is still 1,1,1,1

Each vector maintains its own unique memory space, so after assigning one vector to another, the elements
are duplicated, not shared. Vector assignment automatically ensures the destination vector has adequate
space to hold the assigned elements.
A vector can be passed to a function, as shown in Listing 13.2 (vectortofunc.cpp)

Listing 13.2: vectortofunc.cpp

1 #include <iostream>
2 #include <vector>
3
4 using namespace std;
5
6 /*
7 * print(v)
8 * Prints the contents of an int vector
9 * v is the vector to print
10 */
11 void print(vector<int> v)
12 {
13 int n = v.size();
14 for ( int i = 0; i < n; i++ )
15 cout << v[i] << " ";
16 cout << endl;
17 }
18
19 /*
20 * sum(v)
21 * Adds up the contents of an int vector
22 * v is the vector to sum
23 * Returns the sum of all the elements
24 * or zero if the vector is empty.
25 */
26 int sum(vector<int> v)
27 {
28 int result = 0, n = v.size();
29 for ( int i = 0; i < n; i++ )
30 result += v[i];
31 return result;
32 }
33
34 int main()
35 {
36 vector<int> list;
37 list.push_back(2);
38 list.push_back(4);
39 list.push_back(6);
40 list.push_back(8);
41 // Print the contents of the vector
42 print(list);

©2013 Richard L. Halterman Draft date: January 27, 2013

 13.2. VECTORS 356

43 // Compute and display sum

44 cout << sum(list) << endl;
45 // Zero out all the elements of list
46 int n = list.size();
47 for ( int i = 0; i < n; i++ )
48 list[i] = 0;
49 // Reprint the contents of the vector
50 print(list);
51 // Compute and display sum
52 cout << sum(list) << endl;
53 }

As shown in the print function’s definition:

void print(vector<int> v)

a vector formal parameter is declared just like non-vector formal parameters. The actual parameter (list)
is copied to the formal parameter (v) when the function is called at run time.
Since a vector potentially can be quite large, it is inefficient to pass a vector by value as shown above.
In call by value, a new vector must be created for the formal parameter, and all the elements of the actual
parameter must be copied into the new vector local to the function. A better approach uses call by reference,
with a twist:
void print(const vector<int>& v)
{
int n = v.size();
for ( int i = 0; i < n; i++ )
cout << v[i] << " ";
cout << endl;
}

The & symbol indicates that v is being passed by reference (see Section 10.7) to print. This means the
address of vector v is copied to the formal parameter instead of copying the whole vector itself. Passing
the address is much more efficient, because on most systems an address is the same size as a single int,
whereas the vector could contain 1,000,000 ints. With call by value all those 1,000,000 integers would
have to be copied from the actual parameter into the formal parameter when the function is invoked.
Unfortunately, as indicated in Section 10.7, call-by-value parameter passing is preferred to call-by-
reference parameter passing because with call by value the function cannot cause the side effect of modify-
ing the actual parameter. In the case of our new print function, however, the formal parameter is declared
to be a const reference. This means the function cannot modify the actual parameter being passed in.
Passing a vector object as a constant reference allows us to achieve the efficiency of call by reference with
the safety of call by value.
As with the new print function, sum should be rewritten so it uses a constant vector reference as
its formal parameter. If a function’s purpose is to modify a vector, the reference should not be const.
Listing 13.3 (makerandomvector.cpp) uses a function named make_random that fills a vector with pseu-
dorandom integer values.

Listing 13.3: makerandomvector.cpp

1 #include <iostream>
2 #include <vector>
3 #include <cstdlib> // For rand

©2013 Richard L. Halterman Draft date: January 27, 2013

 13.2. VECTORS 357

4
5 using namespace std;
6
7 /*
8 * print(v)
9 * Prints the contents of an int vector
10 * v is the vector to print; v is not modified
11 */
12 void print(const vector<int>& v)
13 {
14 int n = v.size();
15 for ( int i = 0; i < n; i++ )
16 cout << v[i] << " ";
17 cout << endl;
18 }
19
20 /*
21 * make_random(v)
22 * Fills an int vector with pseudorandom numbers
23 * v is the vector to fill; v is modified
24 * size is the maximum size of the vector
25 */
26 void make_random(vector<int>& v, int size)
27 {
28 v.clear(); // Empties the contents of vector
29 int n = rand() % size + 1; // Random size for v
30 for ( int i = 0; i < n; i++ )
31 v.push_back(rand()); // Populate with random values
32 }
33
34 int main()
35 {
36 srand(2);
37 vector<int> list;
38 // Print the contents of the vector
39 cout << "Vector initially: ";
40 print(list);
41 make_random(list, 20);
42 cout << "1st random vector: ";
43 print(list);
44 make_random(list, 20);
45 cout << "2nd random vector: ";
46 print(list);
47 make_random(list, 20);
48 cout << "3rd random vector: ";
49 print(list);
50 }

The make_random function in Listing 13.3 (makerandomvector.cpp) calls the vector method clear
which makes a vector empty. It then proceeds to add a random number of random integers to the empty
vector.
Listing 13.4 (stringsort.cpp) uses the string and vector classes and allows the user to enter words
separated by spaces and newlines. The user terminates the input pressing by pressing the enter key and then
pressing Ctrl-Z. The program first prints the words in the same order as they were entered and then prints

©2013 Richard L. Halterman Draft date: January 27, 2013

 13.2. VECTORS 358

them out in alphabetical (more precisely, lexicographical) order.

Listing 13.4: stringsort.cpp

1 // File stringsort.cpp
2
3 #include <iostream>
4 #include <string>
5 #include <vector>
6 #include <algorithm>
7
8 using namespace std;
9
10 /*
11 * print_words(words)
12 * Prints the contents of words, a vector of strings.
13 * words is not affected by the function.
14 */
15 void print_words(const vector<string>& words)
16 {
17 int n = words.size();
18 for ( int i = 0; i < n; i++ )
19 cout << words[i] << " ";
20 cout << endl;
21 }
22
23 int main()
24 {
25 vector<string> word_list; // List of words
26 string word;
27
28 // Read in the words and put them into the list
29 while ( cin >> word )
30 word_list.push_back(word);
31
32 // Print the list of words
33 print_words(word_list);
34
35 // Sort the list of words
36 sort(word_list.begin(), word_list.end());
37
38 // Print the words again
39 print_words(word_list);
40 }

The begin and end methods of the vector class return iterator objects. Without going into detail, an
iterator object behaves like a pointer. The begin method returns an object that behaves like a pointer to
the first element in the vector. The end method returns an object that behaves like a pointer to the element
just past the last element in the vector. In Listing 13.4 (stringsort.cpp) we see how the begin and end
methods can be used with the standard sort function to sort our vector of strings.
A function may return a vector object. Listing 13.5 (primelist.cpp) is a practical example of a function
that returns a vector.

Listing 13.5: primelist.cpp

©2013 Richard L. Halterman Draft date: January 27, 2013

 13.2. VECTORS 359

1 #include <iostream>
2 #include <vector>
3 #include <cmath>
4
5 using namespace std;
6
7 /*
8 * print(v)
9 * Prints the contents of an int vector
10 * v is the vector to print; v is not modified
11 */
12 void print(const vector<int>& v)
13 {
14 int n = v.size();
15 for ( int i = 0; i < n; i++ )
16 cout << v[i] << " ";
17 cout << endl;
18 }
19
20 /*
21 * is_prime(n)
22 * Determines the primality of a given value
23 * n an integer to test for primality
24 * Returns true if n is prime; otherwise, returns false
25 */
26 bool is_prime(int n)
27 {
28 if ( n < 2 )
29 return false;
30 else
31 {
32 bool result = true; // Provisionally, n is prime
33 double r = n, root = sqrt(r);
34 // Try all possible factors from 2 to the square
35 // root of n
36 for ( int trial_factor = 2; result && trial_factor <= root;
37 trial_factor++ )
38 result = (n % trial_factor != 0 );
39 return result;
40 }
41 }
42
43 /*
44 * primes(begin, end)
45 * Returns a vector containing the prime
46 * numbers in the range begin...end.
47 * begin is the first number in the range
48 * end is the last number in the range
49 */
50 vector<int> primes(int begin, int end)
51 {
52 vector<int> result;
53 for ( int i = begin; i <= end; i++ )
54 if ( is_prime(i) )

©2013 Richard L. Halterman Draft date: January 27, 2013

 13.2. VECTORS 360

55 result.push_back(i);
56 return result;
57 }
58
59 int main()
60 {
61 int low, high;
62 cout << "Please enter lowest and highest values in "
63 << "the range: ";
64 cin >> low >> high;
65 vector<int> prime_list = primes(low, high);
66 print(prime_list);
67 }

Returning a vector from a function raises the same efficiency concerns as passing a vector by value to a
function; the return value of the primes function must be copied into the caller’s prime_list variable.
If the vector is large, as in the call
prime_list = primes(2, 1000000);

all the elements in the returned vector must be copied into the caller’s prime_list vector.
Due to the overhead involved in vector return values, a better implementation passes the caller’s vector
as a non-constant reference. The primes function then simply populates the caller’s vector directly:
void primes(vector<int>& v, int begin, int end)
{
if ( !v.empty() ) // Ensure that the list starts
v.clear(); // out empty
for ( int i = begin; i <= end; i++ )
if ( is_prime(i) )
v.push_back(i);
}

The caller’s code in main is then written as

vector<int> prime_list;
primes(prime_list, low, high);
print(prime_list);

This call-by-reference version is more efficient than the vector return version.

13.2.3 Vectors and Arrays

It is easy to create a vector from a raw array, and it is easy to get access to the internal array that a vector
object maintains. The ability to convert from one form to the other is handy if you have access to a C library
that can process arrays but knows nothing of vector objects. Listing 13.6 (vecarrayconv.cpp) shows how a
vector may be “passed” to a function expecting an array and how an array can be turned into a vector.

Listing 13.6: vecarrayconv.cpp

1 #include <iostream>
2 #include <vector>
3

©2013 Richard L. Halterman Draft date: January 27, 2013

 13.2. VECTORS 361

4 using namespace std;

5
6 // sum(a,n)
7 // Adds up the elements in an integer array
8 // a is an array of integers
9 // n is the number of elements in a
10 int sum(const int *a, int n)
11 {
12 int s = 0;
13 for ( int i = 0; i < n; i++ )
14 s += a[i];
15 return s;
16 }
17
18 // sum(v)
19 // Adds up the elements in an integer vector
20 // v is a vector of integers
21 int sum(const vector<int>& v)
22 {
23 int s = 0, n = v.size();
24 for ( int i = 0; i < n; i++ )
25 s += v[i];
26 return s;
27 }
28
29 int main()
30 {
31 // Make a big vector filled with 2s
32 vector<int> set(1000, 2);
33 // Pass the vector to a function expecting a vector
34 cout << sum(set) << endl;
35 // Pass the vector's interal array to a function
36 // expecting a raw array
37 cout << sum(&set[0], set.size()) << endl;
38
39 cout << "-------------------" << endl;
40
41 // Declare an array and fill it with 3s
42 int arr[1000];
43 for ( int i = 0; i < 1000; i++ )
44 arr[i] = 3;
45 // Pass the array to a function expecting an array
46 cout << sum(arr, 1000) << endl;
47 // Make a vector from the array and pass it to a
48 // function expecting a vector
49 vector<int> arr_vec(arr, arr + 1000);
50 cout << sum(arr_vec) << endl;
51 }

The first vector constructor call in main:

vector<int> set(1000, 2);

initializes set to hold 1,000 integer elements, all with the value of 2. The second vector constructor call:
vector<int> arr_vec(arr, arr + 1000);

©2013 Richard L. Halterman Draft date: January 27, 2013

13.3. INPUT/OUTPUT STREAMS 362

Creates a vector object named arr_vec with its internal array holding a copy of the elements in arr.
This form of the vector constructor expects two pointer parameters: a pointer to the beginning of the array
(arr) and a pointer to the imaginary element just past the end of the array (arr + 1000). The vector
constructor makes a copy of the array, so vector’s internal array is distinct from the array referenced by
the constructor. After the construction process, changing an element of arr_vec does not affect arr and
vice-versa.
The statement
cout << sum(&set[0], set.size()) << endl;

passes the vector’s internal array to the sum function that expects a primitive array. The expression
&set[0]

is the actual memory location of the first element of vector set’s internal array. This means a function
accepting a non-const array or pointer argument, such as
void zero_it(int *a, int n)
{
for ( int i = 0; i < n; i++ )
a[i] = 0;
}

can directly modify the contents of a vector. The call

zero_it(&set[0], set.size());

would change all the elements in the vector set to zero.

Note that the sum function is overloaded in Listing 13.6 (vecarrayconv.cpp). One version of sum
accepts a pointer to an array and the array’s size, while the other version expects a single integer vector as
an argument. For more information about overloaded functions in C++, see Section 10.3.
In Chapter 11 we examined C++ arrays. Vectors provide all the capabilities of arrays and add consider-
able functionality. Since vectors are more modern, fuller-featured, and more convenient to use, why should
you even consider using an array? The main reason is C language compatibility. The C programming
language does not support the vector type, since vectors are objects and, unlike C++, C is not an object-
oriented language. C++ programs often use C libraries. Many C functions require array parameters and are
unable to use C++ vectors. In order to use those C library functions from within a C++ program, you must
use an array. Fortunately, we have seen that it is easy to convert vectors to arrays and vice-versa.
The C++ standard guarantees that the elements in a vector object reside in a contiguous memory block,
just like an array. A vector object is a thin wrapper around a standard array. When used properly, under
most circumstances there is no performance penalty for using a vector instead of a raw array. For most
applications, since arrays require more programmer attention and care to avoid logic errors, if you have the
choice between using a vector or an array, you should use a vector. The enhanced programmer productivity
gained by using vectors instead of arrays outweighs any potential slight performance advantage that an
array might provide over a vector.

13.3 Input/Output Streams

We have used iostream objects from the very beginning. cout is the output stream object that prints
to the screen. cin is the input stream object that receives values from the keyboard. The precise type of

©2013 Richard L. Halterman Draft date: January 27, 2013

 13.3. INPUT/OUTPUT STREAMS 363

cout is ostream, and cin’s type is istream.

Like other objects, cout and cin have methods. The << and >> operators actually are the methods
operator<< and operator>>. (The operators << and >> normally are used on integers to perform
left and right bitwise shift operations; see Appendix D for more information.) The following code fragment
cin >> x;
cout << x;

can be written in the explicit method call form as:

cin.operator>>(x);
cout.operator<<(x);

The first statement calls the operator>> method on behalf of the cin object passing in variable x by
reference. The second statement calls the operator<< method on behalf of the cout object passing the
value of variable x. A statement such as
cout << x << endl;

is a more pleasant way of expressing

cout.operator<<(x).operator<<(endl);

Reading the statement left to right, the expression cout.operator<<(x) prints x’s value on the screen
and returns the cout object itself. The return value (simply cout) then is used to invoke the operator<<
method again with endl as its argument.
A statement such as
cin >> x >> y;

can be written
cin.operator>>(x).operator>>(y);

As is the case of operator<< with cout, reading left to right, the expression cin.operator>>(x)
calls the operator>> method passing variable x by reference. It reads a value from the keyboard and
assigns it to x. The method call returns cin itself, and the return value is used immediately to invoke
operator>> passing variable y by reference.
You probably have noticed that it is easy cause a program to fail by providing input that the program
was not expecting. For example, compile and run Listing 13.7 (naiveinput.cpp).

Listing 13.7: naiveinput.cpp

1 #include <iostream>
2
3 using namespace std;
4
5 int main()
6 {
7 int x;
8 // I hope the user does the right thing!
9 cout << "Please enter an integer: ";
10 cin >> x;
11 cout << "You entered " << x << endl;
12 }

©2013 Richard L. Halterman Draft date: January 27, 2013

 13.3. INPUT/OUTPUT STREAMS 364

Listing 13.7 (naiveinput.cpp) works fine as long as the user enters an integer value. What if the user
enters the word “five,” which arguably is an integer? The program produces incorrect results. We can
use some additional methods available to the cin object to build a more robust program. Listing 13.8
(betterintinput.cpp) detects illegal input and continues to receive input until the user provides an acceptable
value.

Listing 13.8: betterintinput.cpp

1 #include <iostream>
2 #include <limits>
3
4 using namespace std;
5
6 int main()
7 {
8 int x;
9 // I hope the user does the right thing!
10 cout << "Please enter an integer: ";
11 // Enter and remain in the loop as long as the user provides
12 // bad input
13 while ( !(cin >> x) )
14 {
15 // Report error and re-prompt
16 cout << "Bad entry, please try again: ";
17 // Clean up the input stream
18 cin.clear(); // Clear the error state of the stream
19 // Empty the keyboard buffer
20 cin.ignore(numeric_limits<streamsize>::max(),'\n');
21 }
22 cout << "You entered " << x << endl;
23 }

We learned in Section 6.1 that the expression

cin >> x

has a Boolean value that can be used within a conditional or iterative statement. If the type of value entered
by the user is compatible with the declared type of the variable, the expression evaluates to true; otherwise,
it is interpreted as false. The negation
!(cin >> x)

is true if the input is bad, so the only way to execute the body of the loop is provide illegal input. As long
as the user provides bad input, the program’s execution stays inside the loop.
While determining whether of not a user’s entry is correct seems sufficient for the programmer to make
corrective measures, it is not. Two additional steps are necessary:

• The bad input characters cause the cin to enter an error state. The input stream object remains in an
error state until the programmer manually resets it. The call
cin.clear();

resets the stream object so it can process more input.

©2013 Richard L. Halterman Draft date: January 27, 2013

 13.3. INPUT/OUTPUT STREAMS 365

• Whatever characters the user typed in that cannot be assigned to the given variable remain in the
keyboard buffer. Clearing the stream object does not remove the leftover keystrokes. Asking the user
to retry without clearing the bad characters entered from before results in the same problem—the
stream object re-enters the error state and the bad characters remain in the keyboard buffer. The
solution is to flush from the keyboard buffer all of the characters that the user entered since the last
valid data entry. The statement
cin.ignore(numeric_limits<streamsize>::max(),'\n');

removes from the buffer all the characters, up to and including the newline character ('\n'). The
function call
numeric_limits<streamsize>::max()

returns the maximum number of characters that the buffer can hold, so the ignore method reads
and discards characters until it reads the newline character ('\n') or reaches the end of the buffer,
whichever comes first.

Once the stream object has been reset from its error state and the keyboard buffer is empty, user input can
proceed as usual.
The ostream and istream classes have a number of other methods, but we will not consider them
here.
istream objects use whitespace (spaces and tabs) as delimiters when they get user input from the
user. This means you cannot use the operator>> to assign a complete line of text from the keyboard.
Listing 13.9 (faultyreadline.cpp) illustrates.

Listing 13.9: faultyreadline.cpp

1 #include <iostream>
2 #include <string>
3
4 using namespace std;
5
6 int main()
7 {
8 string line;
9 cout << "Please enter a line of text: ";
10 cin >> line;
11 cout << "You entered: \"" << line << "\"" << endl;
12 }

A sample run of Listing 13.9 (faultyreadline.cpp) reveals:

Please enter a line of text: Mary had a little lamb.
You entered: "Mary"

As you can see, Listing 13.9 (faultyreadline.cpp) does not assign the complete line of text to the sting
variable line. The text is truncated at the first space in the input.
To read in a complete line of text from the keyboard, including any embedded spaces that may be
present, use the global getline function. As Listing 13.10 (readline.cpp) shows, the getline function
accepts an istream object and a string object to assign.

©2013 Richard L. Halterman Draft date: January 27, 2013

 13.4. FILE STREAMS 366

Listing 13.10: readline.cpp

1 #include <iostream>
2 #include <string>
3
4 using namespace std;
5
6 int main()
7 {
8 string line;
9 cout << "Please enter a line of text: ";
10 getline(cin, line);
11 cout << "You entered: \"" << line << "\"" << endl;
12 }

A sample run of Listing 13.10 (readline.cpp) produces:

Please enter a line of text: Mary has a little lamb.
You entered: "Mary has a little lamb."

13.4 File Streams

Many applications allow users to create and manipulate data. Truly useful applications allow users to store
their data to files; for example, word processors can save and load documents.
Vectors would be more useful it they were persistent. Data is persistent when it exists between program
executions. During one execution of a particular program the user may create and populate a vector. The
user then saves the contents of the vector to disk and quits the program. Later, the user can run the program
again and reload the vector from the disk and resume work.
C++ fstream objects allow programmers to build persistence into their applications. Listing 13.11
(numberlist.cpp) is a simple example of a program that allows the user to save the contents of a vector to a
text file and load a vector from a text file.

Listing 13.11: numberlist.cpp

1 // File file_io.cpp
2
3 #include <iostream>
4 #include <fstream>
5 #include <string>
6 #include <vector>
7
8 using namespace std;
9
10 /*
11 * print_vector(v)
12 * Prints the contents of vector v.
13 * v is a vector holding integers.
14 * /
15 void print_vector(const vector<int>& vec)
16 {
17 cout << "{";
18 int len = vec.size();

©2013 Richard L. Halterman Draft date: January 27, 2013

 13.4. FILE STREAMS 367

19 if ( len > 0 )
20 {
21 for ( int i = 0; i < len - 1; i++ )
22 cout << vec[i] << ","; // Comma after elements
23 cout << vec[len - 1]; // No comma after last element
24 }
25 cout << "}" << endl;
26 }
27
28 /*
29 * save_vector(filename, v)
30 * Writes the contents of vector v.
31 * filename is name of text file created. Any file
32 * by that name is overwritten.
33 * v is a vector holding integers. v is unchanged by the
34 * function.
35 */
36 void save_vector(const string& filename, const vector<int>& vec)
37 {
38 ofstream out;
39 out.open(filename.c_str());
40 if ( out.good() ) // Make sure the file was opened properly
41 {
42 int n = vec.size();
43 for ( int i = 0; i < n; i++ )
44 out << vec[i] << " "; // Space delimited
45 out << endl;
46 out.close();
47 }
48 else
49 cout << "Unable to save the file" << endl;
50 }
51
52 /*
53 * load_vector(filename, v)
54 * Reads the contents of vector v from text file
55 * filename. v's contents are replaced by the file's
56 * contents. If the file cannot be found, the vector v
57 * is empty.
58 * v is a vector holding integers.
59 */
60 void load_vector(const string& filename, vector<int>& vec)
61 {
62 ifstream in;
63 in.open(filename.c_str());
64 if ( in.good() ) // Make sure the file was opened properly
65 {
66 vec.clear(); // Start with empty vector
67 int value;
68 while ( in >> value ) // Read until end of file
69 vec.push_back(value);
70 in.close();
71 }
72 else
73 cout << "Unable to load in the file" << endl;

©2013 Richard L. Halterman Draft date: January 27, 2013

 13.4. FILE STREAMS 368

74 }
75
76 int main()
77 {
78 vector<int> list;
79 bool done = false;
80 char command;
81 while ( !done )
82 {
83 cout << "I)nsert <item> P)rint "
84 << "S)ave <filename> L)oad <filename> "
85 << "E)rase Q)uit: ";
86 cin >> command;
87 int value;
88 string filename;
89 switch ( command )
90 {
91 case 'I':
92 case 'i':
93 cin >> value;
94 list.push_back(value);
95 break;
96 case 'P':
97 case 'p':
98 print_vector(list);
99 break;
100 case 'S':
101 case 's':
102 cin >> filename;
103 save_vector(filename, list);
104 break;
105 case 'L':
106 case 'l':
107 cin >> filename;
108 load_vector(filename, list);
109 break;
110 case 'E':
111 case 'e':
112 list.clear();
113 break;
114 case 'Q':
115 case 'q':
116 done = true;
117 break;
118 }
119 }
120 }

Listing 13.11 (numberlist.cpp) is command driven with a menu, and when the user types
I)nsert <item> P)rint S)ave <filename> L)oad <filename> E)rase Q)uit: S data1.text

the current contents of the vector are saved to a file named data1.text. The user can erase the contents of
the vector:
I)nsert <item> P)rint S)ave <filename> L)oad <filename> E)rase Q)uit: E

©2013 Richard L. Halterman Draft date: January 27, 2013

13.4. FILE STREAMS 369

and then restore the contents with a load command:

I)nsert <item> P)rint S)ave <filename> L)oad <filename> E)rase Q)uit: L data1.text

The user may also quit the program, and later re-run the program and load in the previously saved list of
numbers. Different number lists can be saved to different files.
An ofstream object writes data to files. The open method can be used to associate the object with
the file to be created. The c_str method used with the file name string object in the call to open
returns a primitive C string, not a C++ string object. C strings are covered in Section 11.7. A C string is
required when associating an ifstream or ofstream object with an actual file on the disk. A string
literal can be used as well, as such as

ofstream fout;
int x = 10;
fout.open("myfile.dat");
if ( fout.good() ) // Make sure the file was opened properly
{
fout << "x = " << x << endl;
fout.close();
}
else
cout << "Unable to write to the file \"myfile.dat\"" << endl;

After the call to open, programmers should verify that the file was properly opened by calling the file
stream object’s good method. An output file stream may fail for various reasons, including the disk being
full or insufficient permissions to create a file in a given folder.
Once the file has been opened, an ofstream object can be used like the cout output stream object,
except the data is recorded in a text file instead of being printed on the screen. Just like with cout, you can
send an ofstream object stream manipulators like endl and setw.
If the file was properly opened, the programmer must call the file stream object’s close method when
finished with the file. Failure to do so may mean that some or all of the data written to the file is not
completely saved to the disk.
An ifstream object reads data from files. The open method can be used to associate the object with
the file to be read. As with ofstream objects, you must use the c_str method because open expects a
C string, not a C++ string object.
After calling open, call good to ensure the file was successfully opened. An input stream object often
fails to open properly because the file does not exist; perhaps the file name is misspelled, or the path to the
file is incorrect. An input stream can also fail because of insufficient permissions or because of bad sectors
on the disk.
Once the file has been opened, an ifstream object can be used like the cin input stream object,
except the data is read from a file instead the keyboard.
As with an output stream object, the file associated with an input stream object should be closed via the
object’s close method when the program has finished reading the file.

©2013 Richard L. Halterman Draft date: January 27, 2013

 13.5. COMPLEX NUMBERS 370

13.5 Complex Numbers

C++ supports mathematical complex numbers via the complex class. Recall from mathematics that a
complex number has a real component and an imaginary component. Often written as a + bi, a is the real
part, an ordinary real number, and bi is the imaginary part where b is a real number and i2 = −1.
The complex class in C++ is a template class like vector. In the angle brackets you specify the
precision of the complex number’s components:

complex<float> fc;
complex<double> dc;
complex<long double> ldc;

Here, the real component and imaginary coefficient of fc are single-precision floating-point values. dc
and ldc have the indicated precisions. Listing 13.12 (complex.cpp) is small example that computes the
product of complex conjugates (which should be real numbers).

Listing 13.12: complex.cpp

1 // File complex.cpp
2 #include <iostream>
3 #include <complex>
4
5 using namespace std;
6
7 int main()
8 {
9 // c1 = 2 + 3i, c2 = 2 - 3i; c1 and c2 are complex conjugates
10 complex<double> c1(2.0, 3.0), c2(2.0, -3.0);
11
12 // Compute product "by hand"
13 double real1 = c1.real(),
14 imag1 = c1.imag(),
15 real2 = c2.real(),
16 imag2 = c2.imag();
17 cout << c1 << " * " << c2 << " = "
18 << real1*real2 + imag1*real2 + real1*imag2 - imag1*imag2 << endl;
19
20 // Use complex arithmetic
21 cout << c1 << " * " << c2 << " = " << c1*c2 << endl;
22 }

Observe that the complex number 2 − 3i is displayed as the ordered pair (2,-3). The first element of
the pair is the real part, and the second element is the imaginary coefficient. If the imaginary part is zero,
the number is a real number (or, in this case, a double).
Imaginary numbers have scientific and engineering applications that exceed the scope of this book, so
this concludes our brief into C++’s complex class. If you need to solve problems that involve complex
numbers, more information can be found at http://www.cplusplus.com/reference/std/complex/.

©2013 Richard L. Halterman Draft date: January 27, 2013

13.6. SUMMARY 371

13.6 Summary

• C++ allows programmers to define new data types that provide additional capabilities over the built-in
data types like int and double.

• A programmer-defined data type is specified by a class. The C++ standard library contains a number
predefined classes.

• An object is an instance of a class. Object is to class as variable is to type.

• Objects can hold data and provide methods to access and manipulate their data.

• A method is also known as a member function. A method is function associated with a class of
objects.

• To invoke a non-operator method named f on behalf of object obj, the syntax is obj.f().

• Operator methods allow objects to use C++ operators to invoke a method with more convenient syntax;
for example, for a method named operator+=, the call
obj.operator+=(2);

would be better written

obj += 2;

• string is a data type from the standard C++ library representing a sequence of characters. A
string object is a better alternative to a C string when compatibility with existing C code is not
required.

• The directive
#include <string>

must appear in programs that use C++ string objects.

• The string class methods include

– length—returns the number of characters that make up the string

– operator[]—provides access to a particular character in a string; the first character is at
index zero. The [] operator with string objects works exactly as it does with arrays.
– at—provides bounds-checking access to the value stored at a given position within the vector
– operator+=—appends a string to the end of a given string.
– clear—removes all the characters from the string.
– substr—returns a substring of the given string.
– c_str—returns the C string representation of the given string; useful for functions that require
C strings instead of C++ string objects.

• An vector is an object that encapsulates an array. Like an array, a vector represents a collection of
homogeneous data values.

• The directive
#include <vector>

©2013 Richard L. Halterman Draft date: January 27, 2013

13.6. SUMMARY 372

must appear in programs that use vector objects.

• The type of values contained by a vector is specified within <> when the vector is declared; for
example,
vector<double> list;

• A vector is a C++ object, and, as such, provides methods that allow clients to interact with vectors in
interesting ways:

– push_back adds elements to the end of the vector. The size of the vector increases automati-
cally as new items are added.
pop_back removes the last element from the end of the vector. The size of the vector decreases
automatically as items are removed.
– operator[] is used to access individual elements of a vector with familiar array syntax.
– size returns the number of elements in the vector.
– clear makes the vector empty.

• Like with an array, an element of a vector is accessed via a non-negative integer index using the []
operator; for example,
v[3] = 4.1;

The index of the desired element is placed within the square brackets.

• The first element in a vector is at index zero. If the vector contains n elements, the last element is at
index n − 1.

• Unlike an array, a vector make be empty.

• The initial size of a vector can be specified when the vector is declared; for example,
vector<int> numbers(10); // numbers holds 10 integers

The vector’s size is not fixed, and it will expand automatically as new elements are added via the
push_back method.

• A vector may be filled with a specified value when it is declared, as in

vector<int> v(10, 2); // v contains 10 twos

• Any expression compatible with an integer can be placed within the [] operator. Literals, variables,
and expressions can be used. The compiler will issue a warning if a floating-point expression is used
within the square brackets.

• A for loop is a convenient way to traverse the contents of a vector.

• Like other variables, a vector variable can be local, static local, or global.

• A vector may be passed to a function. For efficiency reasons, a vector should be passed by reference.
If call-by-value safety is desired, pass a constant reference to the vector so the function cannot modify
the contents of the vector.

• It is the programmer’s responsibility to stay within the bounds of a vector. Venturing outside the
bounds of a vector is a logic error and often leads to a program crash.

©2013 Richard L. Halterman Draft date: January 27, 2013

13.7. EXERCISES 373

• The at method can be used in place of operator[]. The at method performs bounds checking.
• One vector can be assigned to another vector; for example,
vector<int> list(3, 5), // list contains 5,5,5
vec(4, 1); // vec contains 1,1,1,1
vec = list; // Now vec will contain 5,5,5
list[0] = 7; // Changes list but not vec

In the vector assignment process all the elements from one vector are copied into the other.
• C++ supports multidimensional vectors. A 2-D vector is simply a vector of vectors.
• The rows in a 2-D vector may have different lengths.
• The << and >> operators used, respectively, with cout and cin are actually methods that have a
special syntax.
• Programs can save data to and load data from files using ofstream and ifstream objects.
• The directive
#include <fstream>

is required for programs that use file stream objects.

• With file stream objects, every successful open call should be followed by a call to the object’s
close method when the program is finished processing the file.
• The programmer should call the good method on a file stream object to ensure the file was opened
successfully.
• C++’s complex class allows programmers to solve problems that require complex numbers.

13.7 Exercises

1. What advantages does a string object have over a C string array or characters?
2. Suppose you wish to use a C library function defined as follows:
void process_string(const char *s) { /* Details omitted */ }

but you need to pass a string object named word. Provide the code that would allow the process_string
function to process your word object.
3. A palindrome is a string of text that reads the same forwards and backwards. Examples include:
• “TENET”
• “STEP ON NO PETS”
• “RATS LIVE ON NO EVIL STAR”
All these strings are examples of strict palindromes where spacing and punctuation must reverse
exactly. Complete the following function that determines if a given string is a strict palindrome. It
should return true if the string is a strict palindrome and false otherwise.
The empty string reads the same forward and backward, so it is a palindrome. Your function should
ignore capitalization (that is, the string “Tenet” is regarded as a palindrome).

©2013 Richard L. Halterman Draft date: January 27, 2013

13.7. EXERCISES 374

bool palindrome(const string& s)

{
// Your code goes here . . .
}

4. What advantages does a vector have over an array?

5. What advantages does an array have have over a vector?

6. Suppose you wish to use a C library function defined as follows:

void process_array(int *arr, int len) { /* Details omitted */ }

where arr is an array and len is the array’s length, but you need to pass a vector of integers named
collection. Provide the code that would allow the process_array function to process your
collection object.
7. Complete the following function that filters negative elements out of a vector containing integers.
The function modifies the vector passed to it, removing the negative values. For example, if a vector
containing the elements 2, -16, 2, -5, 0, 1, -2, -3 is passed to the function, the function would modify
the vector so it contains only 2, 2, 0, 1. Note the original ordering of the non-negative values is
unchanged in the result.
void filter(vector<int>& v)
{
// Add your code...
}

©2013 Richard L. Halterman Draft date: January 27, 2013

 375

Chapter 14

Custom Objects

The introduction of variables in association with higher-level programming languages was a great step
forward in the late 1950s. No longer did programmers need to be concerned with the lower-level details
of the processor and absolute memory addresses. Named variables and functions allow programmers to
abstract away such machine-level details and concentrate on concepts that transcend computers, such as
integers and characters. Objects provide a level of abstraction above that of variables. Objects allow
programmers to go beyond simple values—developers can focus on more complex things like geometric
shapes, bank accounts, and aircraft wings. Programming objects that represent these real-world things can
possess capabilities that go far beyond the simple variables we have studied to this point.
A C++ object typically consists of a collection of data and code. By bundling data and code together,
objects store information and provide services to other parts of the software system. An object forms a
computational unit that makes up part of the overall computer application. A programming object can
model a real-world object more naturally than can a collection of simple variables since it can encapsulate
considerable complexity. Objects make it easier for developers to build complex software systems.
C++ is classified as an object-oriented language. Most modern programming languages have some
degree of object orientation. This chapter shows how programmers can define, create, and use custom
objects.

14.1 Object Basics

Consider the task of dealing with geometric points. Mathematicians represent a single point as an ordered
pair of real numbers, usually expressed as (x, y). In C++, the double type serves to approximate a subset of
the mathematical real numbers. A point with coordinates within the range of double-precision floating-point
numbers, therefore, can be modeled by two double variables. Notice something that conceptually is one
thing requires two variables in C++. As a consequence, a function that computes the distance between two
points requires four parameters—x1 , y1 , x2 , and y2 —rather than two points—(x1 , y1 ) and (x2 , y2 ). Ideally,
we should be able to use one variable to represent a point.
One approach could use a two-element array, for example:
typedef double Point[2];

This type definition statement declares the Point identifier to mean the same as a double-precision

©2013 Richard L. Halterman Draft date: January 27, 2013

14.1. OBJECT BASICS 376

floating-point array of size two. With this we can use single variable to represent a point object as in
the following declaration:

Point pt;

The problem with this approach is we must use numeric indices to distinguish between the two components
of a point object instead of names. We may agree that pt[0] means the x coordinate of point pt and
pt[1] means the y coordinate of point pt, but the compiler is powerless to detect the error if a programmer
uses an expression like pt[19] or pt[-3]. An array representation of an object has other problems as
well. Consider a bank account object. A bank account object could include, among many other diverse
things, an account number (an integer), a customer name (a string), and an interest rate (a double-precision
floating-point number). An array implementation of an object would not work for a bank account object
because the elements in an array must all be of the same type.
In additional to storing data, we want our objects to be active agents that can do computational tasks.
We need to be able associate code with a class of objects. We need a fundamentally different programming
construct to represent objects.
Before examining how C++ specifically handles objects, we first will explore what capabilities are de-
sirable. Consider an automobile. An automobile user, the driver, uses the car for transportation. The user’s
interface to the car is fairly simple, considering an automobile’s overall complexity. A driver provides
input to the car via its steering wheel, accelerator and brake pedals, turn signal control, shift lever, etc.
The automobile produces output to the driver with its speedometer, tachometer, various instrument lights
and gauges, etc. These standardized driver-automobile interfaces enable an experienced driver to drive any
modern car without the need for any special training for a particular make or model.
The typical driver can use a car very effectively without understanding the details of how it works. To
drive from point A to point B a driver does not need to know the number of cylinders in the engine, the
engine’s horsepower, or whether the vehicle is front-wheel drive or rear-wheel drive. A driver may look
under the hood at the engine, but the driver cannot confirm any details about what is inside the engine itself
without considerable effort or expense. Many details are of interest only to auto enthusiasts or mechanics.
There may be only a select few automotive engineers capable of understanding and appreciating other more
esoteric details about the vehicle’s design and implementation.
In some ways programming objects as used in object-oriented programming languages are similar to
automobiles. An object may possess considerable capability, but a programmer using the object needs to
know only what the object can do without needing to know how it works. An object provides an interface
to any client code that wishes to use that object. A typical object selectively exposes some parts of itself to
clients and keeps other parts hidden from clients. The object’s designer, on the other hand, must know the
complete details of the object’s implementation and must be an expert on both the what the object does and
how it works.
C++ uses the class keyword to define the structure of an object. A class serves as a pattern or template
from which objects are produced. We will concentrate on five things facilitating object-oriented program-
ming with C++ classes:

1. A class can specify the data that constitute an object’s state.

2. A class can define code to be executed on an object’s behalf that provides services to clients that use
the object.

3. A class may define code that automatically initializes a newly-created object ensuring that it begins
its life in a well-defined state.

©2013 Richard L. Halterman Draft date: January 27, 2013

 14.2. FIELDS 377

4. A class may define code that automatically cleans up any resources an object may be using when its
life is finished.

5. The C++ language provides a way for the developer of a class to specify which parts of its objects are
visible and which parts are hidden from clients.

A class is a programmer-defined type. An object is an instance of a class. The terms object and instance
may be used interchangeably.

14.2 Fields

The simplest kind of object stores only data. We can define a point type as follows:
class Point
{
public:
double x;
double y;
};

Notice the semicolon that follows the close curly brace of the class definition. This semicolon is required,
but it is easy to overlook. By convention class names begin with a captial letter, but class names are just
identifiers like variable names and function names. Here, our class name is Point. The body of the class
appears within the curly braces.
The Point class specifies two double-precision floating-point data components named x and y. These
are called fields or data members. They both appear within the class body after the public label. We say
that x and y are public fields of the Point class; this means client code has full access to the x and y
fields. Any client may examine and modify the x and y components of a Point object.
Once this Point class definition is available, a client can create and use Point objects as shown in
Listing 14.1 (mathpoints.cpp).

Listing 14.1: mathpoints.cpp

1 #include <iostream>
2
3 using namespace std;
4
5 // The Point class defines the structure of software
6 // objects that model mathematical, geometric points
7 class Point
8 {
9 public:
10 double x; // The point's x coordinate
11 double y; // The point's y coordinate
12 };
13
14 int main()
15 {
16 // Declare some point objects
17 Point pt1, pt2;
18 // Assign their x and y fields

©2013 Richard L. Halterman Draft date: January 27, 2013

 14.2. FIELDS 378

19 pt1.x = 8.5; // Use the dot notation to get to a part of the object
20 pt1.y = 0.0;
21 pt2.x = -4;
22 pt2.y = 2.5;
23 // Print them
24 cout << "pt1 = (" << pt1.x << "," << pt1.y << ")" << endl;
25 cout << "pt2 = (" << pt2.x << "," << pt2.y << ")" << endl;
26 // Reassign one point from the other
27 pt1 = pt2;
28 cout << "pt1 = (" << pt1.x << "," << pt1.y << ")" << endl;
29 cout << "pt2 = (" << pt2.x << "," << pt2.y << ")" << endl;
30 // Are pt1 and pt2 aliases? Change pt1's x coordinate and see.
31 pt1.x = 0;
32 cout << "pt1 = (" << pt1.x << "," << pt1.y << ")" << endl;
33 // Note that pt2 is unchanged
34 cout << "pt2 = (" << pt2.x << "," << pt2.y << ")" << endl;
35
36 }

Listing 14.1 (mathpoints.cpp) prints

pt1 = (8.5,0)
pt2 = (-4,2.5)
pt1 = (-4,2.5)
pt2 = (-4,2.5)
pt1 = (0,2.5)
pt2 = (-4,2.5)

It is important to note that Point is not an object. It represents a class of objects. It is a type. The variables
pt1 and pt2 are the objects, or instances, of the class Point. Each of the objects pt1 and pt2 has its
own copies of fields named x and y. Figure 14.1 provides a conceptual view of point objects pt1 and
pt2. On most systems double-precision floating-point numbers require eight bytes of memory. Since each

pt1 pt2

x 8.5 x -4.0

y 0.0 y 2.5

Figure 14.1: Two Point objects with their individual data fields

©2013 Richard L. Halterman Draft date: January 27, 2013

 14.2. FIELDS 379

Point object stores two doubles, a Point object uses at least 16 bytes of memory. In practice, an object
may be slightly bigger than the sum its individual components because most computer architectures restrict
how data can be arranged in memory. This means some objects include a few extra bytes for “padding.”
We can use the sizeof operator to determine the exact number of bytes an object occupies. Under Visual
C++, the expression sizeof pt1 evaluates to 16.
A client may use the dot (.) operator with an object to access one of the object’s members. The
expression pt1.x represents the x coordinate of object pt1. The dot operator is a binary operator; its left
operand is a class instance (object), and its right operand is a member of the class.
The assignment statement in Listing 14.1 (mathpoints.cpp)
pt1 = pt2;

and the statements that follow demonstrate that one object may be assigned to another directly without
the need to copy each individual member of the object. Compare object assignment to array assignment:
Such a direct assignment is not possible with static arrays, and direct assignment of dynamic arrays leads
to aliasing (see Section 11.10). As with the primitive types like ints and doubles, assigning one object
to another does not create an alias.
As another example, suppose we wish to implement a simple bank account object. We determine that
the necessary information for each account consists of a name, ID number, and a balance (amount of money
in the account). We can define our bank account class as
class Account
{
public:
// String representing the name of the account's owner
string name;
// The account number
int id;
// The current balance;
double balance;
};

The name, id, and balance fields constitute the Account class, and the fields represent three different
types. Listing 14.2 (bankaccount.cpp) uses an array of Account objects.

Listing 14.2: bankaccount.cpp

1 #include <iostream>
2 #include <string>
3 #include <vector>
4
5 using namespace std;
6
7 class Account
8 {
9 public:
10 // String representing the name of the account's owner
11 string name;
12 // The account number
13 int id;
14 // The current account balance
15 double balance;

©2013 Richard L. Halterman Draft date: January 27, 2013

 14.2. FIELDS 380

16 };
17
18 // Allows the user to enter via the keyboard information
19 // about an account and adds that account to the database.
20 void add_account(vector<Account>& accts)
21 {
22 string name;
23 int number;
24 double amount;
25
26 cout << "Enter name, account number, and account balance: ";
27 cin >> name >> number >> amount;
28 Account acct;
29 acct.name = name;
30 acct.id = number;
31 acct.balance = amount;
32 accts.push_back(acct);
33 }
34
35 // Print all the accounts in the database
36 void print_accounts(const vector<Account>& accts)
37 {
38 int n = accts.size();
39 for ( int i = 0; i < n; i++ )
40 cout << accts[i].name << "," << accts[i].id
41 << "," << accts[i].balance << endl;
42 }
43
44 void swap(Account& er1, Account& er2)
45 {
46 Account temp = er1;
47 er1 = er2;
48 er2 = temp;
49 }
50
51 bool less_than_by_name(const Account& e1, const Account& e2)
52 {
53 return e1.name < e2.name;
54 }
55
56
57 bool less_than_by_id(const Account& e1, const Account& e2)
58 {
59 return e1.id < e2.id;
60 }
61
62 bool less_than_by_balance(const Account& e1, const Account& e2)
63 {
64 return e1.balance < e2.balance;
65 }
66
67 // Sorts a bank account database into ascending order
68 // The comp parameter determines the ordering
69 void sort(vector<Account>& db,
70 bool (comp)(const Account&, const Account&))

©2013 Richard L. Halterman Draft date: January 27, 2013

 14.2. FIELDS 381

71 {
72 int size = db.size();
73 for ( int i = 0; i < size - 1; i++ )
74 {
75 int smallest = i;
76 for ( int j = i + 1; j < size; j++ )
77 if ( comp(db[j], db[smallest]) )
78 smallest = j;
79 if ( smallest != i )
80 swap(db[i], db[smallest]);
81 }
82 }
83
84
85 // Allows a user interact with a bank account database.
86 int main()
87 {
88 // The simple database of bank accounts
89 vector<Account> customers;
90
91 // User command
92 char cmd;
93
94 // Are we done yet?
95 bool done = false;
96
97 do
98 {
99 cout << "[A]dd [N]ame [I]D [B]alance [Q]uit==> ";
100 cin >> cmd;
101 switch ( cmd)
102 {
103 case 'A':
104 case 'a':
105 // Add an account
106 add_account(customers);
107 break;
108 case 'P':
109 case 'p':
110 // Print customer database
111 print_accounts(customers);
112 break;
113 case 'N':
114 case 'n':
115 // Sort database by name
116 sort(customers, less_than_by_name);
117 print_accounts(customers);
118 break;
119 case 'I':
120 case 'i':
121 // Sort database by ID (account number)
122 sort(customers, less_than_by_id);
123 print_accounts(customers);
124 break;
125 case 'B':

©2013 Richard L. Halterman Draft date: January 27, 2013

 14.2. FIELDS 382

126 case 'b':

127 // Sort database by account balance
128 sort(customers, less_than_by_balance);
129 print_accounts(customers);
130 break;
131 case 'Q':
132 case 'q':
133 done = true;
134 break;
135 }
136 }
137 while ( !done );
138 }

Listing 14.2 (bankaccount.cpp) stores bank account objects in a vector (see Section 13.2). Also, a bank
account object contains a string object as a field. This shows that objects can contain other objects, and
our objects can have arbitrarily complex structures.
A sample run of Listing 14.2 (bankaccount.cpp) prints

[A]dd [N]ame [I]D [B]alance [Q]uit==> a

Enter name, account number, and account balance: Sheri 34 100.34
[A]dd [N]ame [I]D [B]alance [Q]uit==> a
Enter name, account number, and account balance: Mary 10 1323.00
[A]dd [N]ame [I]D [B]alance [Q]uit==> a
Enter name, account number, and account balance: Larry 88 55.05
[A]dd [N]ame [I]D [B]alance [Q]uit==> a
Enter name, account number, and account balance: Terry 33 423.50
[A]dd [N]ame [I]D [B]alance [Q]uit==> a
Enter name, account number, and account balance: Gary 11 7.27
[A]dd [N]ame [I]D [B]alance [Q]uit==> n
Gary,11,7.27
Larry,88,55.05
Mary,10,1323
Sheri,34,100.34
Terry,33,423.5
[A]dd [N]ame [I]D [B]alance [Q]uit==> i
Mary,10,1323
Gary,11,7.27
Terry,33,423.5
Sheri,34,100.34
Larry,88,55.05
[A]dd [N]ame [I]D [B]alance [Q]uit==> b
Gary,11,7.27
Larry,88,55.05
Sheri,34,100.34
Terry,33,423.5
Mary,10,1323
[A]dd [N]ame [I]D [B]alance [Q]uit==> q

©2013 Richard L. Halterman Draft date: January 27, 2013

14.3. METHODS 383

The program allows users to sort the bank account database in several different ways using different com-
parison functions. Notice that the less_than_by_name and similar comparison functions use const
reference parameters for efficiency (see Section 13.2).
Observe that the add_account function has a local variable named name. All Account objects
have a field named name. The add_account function uses the name identifier in both ways without a
problem. The compiler can distinguish between the two uses of the identifier because one is qualified with
an object variable and the dot (.) operator and the other is not.

14.3 Methods

The objects we have seen so far have been passive entities that have no built-in functionality. In addition
to defining the structure of the data for its objects, a class can define functions that operate on behalf of its
objects.
Recall the bacnk account class, Account, from Section 14.2:
class Account
{
public:
// String representing the name of the account's owner
string name;
// The account number
int id;
// The current account balance
double balance;
};

Suppose this design is given to programmer Sam who must write a withdraw function. Sam is a careful
programmer, so his withdraw function checks for overdrafts:
/*******************************************************
* withdraw(acct, amt)
* Deducts amount amt from Account acct, if possible.
* Returns true if successful; otherwise, it returns false.
* A call can fail if the withdraw would
* cause the balance to fall below zero
*
* acct: a bank account object
* amt: funds to withdraw
*
* Author: Sam Coder
* Date: September 3, 2012
*******************************************************/
bool withdraw(Account& acct, double amt)
{
bool result = false; // Unsuccessful by default
if ( acct.balance - amt >= 0 )
{
acct.balance -= amt;
result = true; // Success

©2013 Richard L. Halterman Draft date: January 27, 2013

14.3. METHODS 384

}
return result;
}

The following code fragment shows the expected code a client might write when withdrawing funds from
the bank:
// Good client code
// ----------------
// A simple database of bank customers
Account accountDB[DB_SIZE];

// Get transaction information

int number;
double debit;
cout << "Please enter account number and amount to withdraw:";
cin >> number >> debit;

// Locate account
Account acct;
bool found = false;
for ( int i = 0; i < DB_SIZE; i++ )
if ( accountDB[i].id == number )
{
acct = accountDB[i];
found = true;
}

// Perform the transaction, if possible

if ( found )
{
if ( withdraw(acct, debit) )
cout << "Withdrawal successful" << endl;
else
cout << "Cannot perform the withdraw" << endl;
}
else
cout << "Account does not exist" << endl;

Unfortunately, nothing prevents a client from being sloppy:

// Bad client code
// ----------------
// A simple database of bank customers
Account accountDB[DB_SIZE];

// Get transactive information

int number;
double debit;
cout << "Please enter account number and amount to withdraw:";
cin >> number >> debit;

©2013 Richard L. Halterman Draft date: January 27, 2013

14.3. METHODS 385

// Locate account
Account acct;
bool found = false;
for ( int i = 0; i < DB_SIZE; i++ )
if ( accountDB[i].id == number )
{
acct = accountDB[i];
found = true;
}

// Perform the transaction

if ( found )
acct.balance -= debit; // What if amount is too big?
else
cout << "Account does not exist" << endl;

Nothing in a Account object itself can prevent an overdraft or otherwise prevent clients from improperly
manipulating the balance of an account object.
We need to be able to protect the internal details of our bank account objects and yet permit clients to
interact with them in a well-defined, controlled manner.
Consider a non-programming example. If I deposit $1,000.00 dollars into a bank, the bank then has
custody of my money. It is still my money, so I theoretically can reclaim it at any time. The bank stores
money in its safe, and my money is in the safe as well. Suppose I wish to withdraw $100 dollars from my
account. Since I have $1,000 total in my account, the transaction should be no problem. What is wrong
with the following scenario:

1. Enter the bank.

2. Walk past the teller into a back room that provides access to the safe.
3. The door to the safe is open, so enter the safe and remove $100 from a stack of $20 bills.
4. Exit the safe and inform a teller that you got $100 out of your account.
5. Leave the bank.

This is not the process a normal bank uses to handle withdrawals. In a perfect world where everyone is
honest and makes no mistakes, all is well. In reality, many customers might be dishonest and intentionally
take more money than they report. Even though I faithfully counted out my funds, perhaps some of the bills
were stuck to each other and I made an honest mistake by picking up six $20 bills instead of five. If I place
the bills in my wallet with other money that already may be present, I may never detect the error. Clearly a
bank needs more controlled procedure for customer withdrawals.
When working with programming objects, in many situations it is better to restrict client access to the
internals of an object. Client code should not be able to change directly bank account objects for various
reasons, including:

• A withdrawal should not exceed the account balance.

• Federal laws dictate that deposits above a certain amount should be reported to the Internal Revenue
Service, so a bank would not want customers to be able to add funds to an account in a way to
circumvent this process.

©2013 Richard L. Halterman Draft date: January 27, 2013

14.3. METHODS 386

• An account number should never change for a given account for the life of that account.

How do we protect the internal details of our bank account objects and yet permit clients to interact with
them in a well-defined, controlled manner? The trick is to hide completely from clients the object’s fields
and provide special functions, called methods, that have access to the hidden fields. These methods provide
the only means available to clients of changing the object’s internal state.
In the following revised Account class:
class Account
{
// String representing the name of the account's owner
string name;
// The account number
int id;
// The current account balance
double balance;
public:
// Methods will be added here . . .
};

all the fields no longer reside in the public section of the class definition. This makes the following client
code impossible:
Account acct;
// Set the account's balance
acct.balance = 100; // Illegal
// Withdraw some funds
acct.balance -= 20; // Illegal

The balance field is in the private area of the class. You may use the literal private label, as in
class Account
{
private:
// String representing the name of the account's owner
string name;
// The account number
int id;
// The current account balance
double balance;
public:
// Methods will be added here . . .
};

Here the private label is not necessary because any parts of a class not explicitly labeled are implicitly
private. Said another way, all members of a class are automatically private unless otherwise labeled. Some
programmers like to put the public members before the private members within a class definition, as in
class Account
{
public:
// Methods will be added here . . .

©2013 Richard L. Halterman Draft date: January 27, 2013

14.3. METHODS 387

private:
// String representing the name of the account's owner
string name;
// The account number
int id;
// The current account balance
double balance;
};

In this case the private label is necessary.

In order to enforce the spirit of the withdraw function, we will make it a method, and add a deposit
method to get funds into an account:
class Account
{
// String representing the name of the account's owner
string name;
// The account number
int id;
// The current account balance
double balance;
public:
/*******************************************************
* deposit(amt)
* Adds amount amt to the account's balance.
*
* Author: Sam Coder
* Date: September 3, 2012
*******************************************************/
void deposit(double amt)
{
balance += amt;
}

/*******************************************************
* withdraw(amt)
* Deducts amount amt from the account's balance,
* if possible.
* Returns true if successful; otherwise, it returns false.
* A call can fail if the withdraw would
* cause the balance to fall below zero
*
* amt: funds to withdraw
*
* Author: Sam Coder
* Date: September 3, 2012
*******************************************************/
bool withdraw(double amt)
{
bool result = false; // Unsuccessful by default
if ( balance - amt >= 0 )

©2013 Richard L. Halterman Draft date: January 27, 2013

14.3. METHODS 388

{
balance -= amt;
result = true; // Success
}
return result;
}
};

In this new definition of Account, the members named deposit and withdraw are not fields; they
are method definitions. A method definition looks like a function definition, but it appears within a class
definition. A method is also known as a member function.
A client accesses a method with the dot (.) operator:
// Withdraw money from the Account object named acct
acct.withdraw(100.00);

The withdraw method definition uses three variables: amt, result, and balance. The variables
amt and result are local to withdraw—amt is the method’s parameter, and result is declared
within the body of withdraw. Where is balance declared? It is the field declared in the private section
of the class. The withdraw method affects the balance field of the object upon which it is called:
// Affects the balance field of acct1 object
acct1.withdraw(100.00);
// Affects the balance field of acct2 object
acct2.withdraw(25.00);

Methods may be overloaded just like global functions (see Section 10.3). This means multiple methods is
the same class may have the same names, but their signatures must be different. Recall that a function’s
signature consists of its name and types of parameters; a method’s signature too consists of its name and
types of parameters.
We saw in Section 14.2 that each object provides storage space for its own data fields. An object does not
require any space for its methods. This means the only things about an individual object that an executing
program must maintain are the object’s fields. While the exact organization of memory organization varies
among operating systems, Figure 11.11 shows a typical layout. All the data processed by a program appears
in one of three sections: stack, heap, or static memory. As with simple data types like ints, the fields of
an object declared local to a function or method reside in the segment of memory known as the stack. Also
like simple data types, the fields of an object allocated with new appear on the heap. The fields in global
and static local objects reside in the static section of the executing program’s memory.
Consider the following simple class:
class Counter
{
int count;
public:
// Allow clients to reset the counter to zero
void clear()
{
count = 0;
}

// Allow clients to increment the counter

©2013 Richard L. Halterman Draft date: January 27, 2013

14.3. METHODS 389

void inc()
{
count++;
}

// Return a copy of the counter's current value

int get()
{
return count;
}
};

For this code we see that each Counter object will store a single integer value (its count field). Under
Visual C++ sizeof Counter is the same as sizeof int: four. A local Counter object consumes
four bytes of stack space, a global Counter object uses four bytes of static memory, and a dynamically-
allocated Counter object uses four bytes of heap space.
The segment of memory in Figure 11.11 labeled code stores the machine language for all the program’s
functions and methods. The compiler translates methods into machine language as it does regular func-
tions. Internally, the method inc in the Counter class is identified by a longer name, Counter::inc.
Although Counter::inc is a method, in the compiled code it works exactly like a normal function
unrelated to any class. In the client code
Counter ctr1, ctr2; // Declare a couple of Counter objects
ctr1.clear(); // Reset the counters to zero
ctr2.clear();
ctr1.inc(); // Increment the first counter

the statement
ctr1.clear();

sets the private count field of ctr1 to zero, while the statement
ctr2.clear();

sets ctr2’s count field to zero. Since all Counter objects share the same reset method, how does
each call to Counter::clear reset the field of the proper Counter object? The trick is this: While it
appears that the reset method of the Counter class accepts no parameters, it actually receives a secret
parameter that corresponds to the address of the object on left side of the dot. The C++ source code statement
ctr1.clear(); // Actual C++ code

internally is treated like

Counter::clear(&ctr1); // <-- Not real C++ code

in the compiled code. The code within the method can influence the field of the correct object via the
pointer it receives from the caller. The clear method contains the single statement
count = 0;

Since the count variable is not declared locally within the clear method and it is not a global variable,
it must be the field named count of the object pointed to by the secret parameter passed to clear.
Section 15.2 shows how programmers can access this secret pointer passed to methods.

©2013 Richard L. Halterman Draft date: January 27, 2013

 14.4. CONSTRUCTORS 390

14.4 Constructors

One crucial piece still is missing. How can we make sure the fields of an object have reasonable initial
values before a client begins using the object? A class may define a constructor that looks similar to a
method definition. The code within a constructor executes on behalf of an object when a client creates
the object. For some classes, the client can provide information for the constructor to use when initial-
izing the object. As with functions and methods, class constructors may be overloaded. Listing 14.3
(bankaccountmethods.cpp) exercises an enhanced Account class that offers deposit and withdraw
methods, as well as a constructor.

Listing 14.3: bankaccountmethods.cpp

1 #include <iostream>
2 #include <iomanip>
3 #include <string>
4
5 using namespace std;
6
7 class Account
8 {
9 // String representing the name of the account's owner
10 string name;
11 // The account number
12 int id;
13 // The current account balance
14 double balance;
15 public:
16 // Initializes a bank account object
17 Account(const string& customer_name, int account_number,
18 double amount):
19 name(customer_name), id(account_number), balance(amount)
20 {
21 if ( amount < 0 )
22 {
23 cout << "Warning: negative account balance" << endl;
24 balance = 0.0;
25 }
26 }
27
28 // Adds amount amt to the account's balance.
29 void deposit(double amt)
30 {
31 balance += amt;
32 }
33
34 // Deducts amount amt from the account's balance,
35 // if possible.
36 // Returns true if successful; otherwise, it returns false.
37 // A call can fail if the withdraw would
38 // cause the balance to fall below zero
39 bool withdraw(double amt)
40 {
41 bool result = false; // Unsuccessful by default
42 if ( balance - amt >= 0 )

©2013 Richard L. Halterman Draft date: January 27, 2013

 14.4. CONSTRUCTORS 391

43 {
44 balance -= amt;
45 result = true; // Success
46 }
47 return result;
48 }
49
50 // Displays information about the account object
51 void display()
52 {
53 cout << "Name: " << name << ", ID: " << id
54 << ", Balance: " << balance << endl;
55 }
56 };
57
58
59 int main()
60 {
61 Account acct1("Joe", 2312, 1000.00);
62 Account acct2("Moe", 2313, 500.29);
63 acct1.display();
64 acct2.display();
65 cout << "---------------------" << endl;
66 acct1.withdraw(800.00);
67 acct2.deposit(22.00);
68 acct1.display();
69 acct2.display();
70 }

Listing 14.3 (bankaccountmethods.cpp) produces

Name: Joe, ID: 2312, Balance: 1000

Name: Moe, ID: 2313, Balance: 500.29
---------------------
Name: Joe, ID: 2312, Balance: 200
Name: Moe, ID: 2313, Balance: 522.29

The following characteristics differentiate a constructor definition from a regular method definition:

• A constructor has the same name as the class.

• A constructor has no return type, not even void.

The constructor in Listing 14.3 (bankaccountmethods.cpp) initializes all the fields with values supplied by
the client. The comma-separated list between the colon and the curly brace that begins the body of the
constructor is called the constructor initialization list. An initialization list contains the name of each field
with its initial value in parentheses. All of the fields that make up an object must be initialized before the
body of the constructor executes. In this case the code within the constructor adjusts the balance to zero and
issues a warning if a client attempts to create an account with an initial negative balance. The constructor
thus ensures an account object’s balance can never begin with a negative value. The withdraw method

©2013 Richard L. Halterman Draft date: January 27, 2013

 14.5. DESTRUCTORS 392

ensures that, once created, an Account object’s balance will never be negative. Notice that the client
provides the required constructor parameters at the point of the object’s declaration:
// Client creating two Account objects
Account acct1("Joe", 2312, 1000.00);
Account acct2("Moe", 2313, 500.29);

Since the Account class contains a constructor definition which requires arguments, it now is impossible
for a client to declare an Account object like
Account acct3; // Illegal, must supply arguments for constructor

If acct is a Account object defined as in Listing 14.3 (bankaccountmethods.cpp), the following code
is illegal:
acct.balance -= 100; // Illegal, balance is private

Clients instead should use the appropriate method call to adjust the balance of an Account object:
Account acct("Joe", 1033, 50.00); // New bank account object
acct.deposit(1000.00); // Add some funds
acct.withdraw(2000.00); // Method should disallow this operation

The operations of depositing and withdrawing funds are the responsibility of the object itself, not the client
code. The attempt to withdraw the $2,000 dollars above would not change the account’s balance, and the
client can check the return value of withdraw to provide appropriate feeback to the user about the error
or take steps correct the situation.
A constructor that specifies no parameters is called a default constructor. If the programmer does not
specify any constructor for a class, the compiler will provide a default constructor that does nothing. If the
programmer defines any constructor for a class, the compiler will not generate a default constructor. See
Section 15.1 for the consequences of this constructor policy.

If you do not define a constructor for your class, the constructor automatically
will create one for you—a default constructor that accepts no parameters. The
compiler-generated constructor does not do anything to affect the state of newly
created instances.
If you define any constructor for your class, the compiler will not provide a default
constructor.

14.5 Destructors

A class constructor ensures that when a client creates an instance of that class the object will begin its life
in a well-defined state. A class destructor takes care of any necessary clean up at the end of an object’s
life. Listing 14.4 (dynamicaccount.cpp) modifies our Account class so that it uses dynamic C strings for
names. The constructor is responsible for the memory allocation, and the destructor is responsible for the
memory clean up.

Listing 14.4: dynamicaccount.cpp

1 #include <iostream>

©2013 Richard L. Halterman Draft date: January 27, 2013

 14.5. DESTRUCTORS 393

2 #include <cstring>
3
4 using namespace std;
5
6 class Account
7 {
8 // String representing the name of the account's owner
9 char *name;
10 // The account number
11 int id;
12 // The current account balance
13 double balance;
14 public:
15 // Initializes a bank account object
16 Account(const char *customer_name, int account_number,
17 double amount):
18 // Allocate space for name (include space for terminating
19 // null character
20 name(new char[strlen(customer_name) + 1]),
21 id(account_number), balance(amount)
22 {
23 cout << "Creating an Account object" << endl;
24 strcpy(name, customer_name);
25 id = account_number;
26 if ( amount < 0 )
27 {
28 cout << "Warning: negative account balance" << endl;
29 balance = 0.0;
30 }
31 }
32
33 // Releases the memory allocated to this account object
34 ∼Account()
35 {
36 delete [] name;
37 cout << "Destroying an Account object" << endl;
38 }
39
40 // Adds amount amt to the account's balance.
41 void deposit(double amt)
42 {
43 balance += amt;
44 }
45
46 // Deducts amount amt from the account's balance,
47 // if possible.
48 // Returns true if successful; otherwise, it returns false.
49 // A call can fail if the withdraw would
50 // cause the balance to fall below zero
51 bool withdraw(double amt)
52 {
53 bool result = false; // Unsuccessful by default
54 if ( balance - amt >= 0 )
55 {
56 balance -= amt;

©2013 Richard L. Halterman Draft date: January 27, 2013

 14.6. DEFINING A NEW NUMERIC TYPE 394

57 result = true; // Success

58 }
59 return result;
60 }
61
62 // Displays information about the account object
63 void display()
64 {
65 cout << "Name: " << name << ", ID: " << id
66 << ", Balance: " << balance << endl;
67 }
68 };
69
70 // Allows a user interact with a bank account database.
71 int main()
72 {
73 // The constructor dynamically allocates the space
74 // for the name string
75 Account acct("Jerry", 193423, 1250.00);
76 acct.display();
77 // The destructor automatically deallocates the
78 // account object's string space
79 }

Like a constructor, a destructor has the same name as its class, except that the destructor’s name begins
with a tilde ( ∼ ) symbol. This tilde symbol distinguishes a destructor from a constructor. Also unlike a
constructor, a destructor may not be overloaded, and a destructor may not accept any parameters.
The output of Listing 14.4 (dynamicaccount.cpp) is

Creating an Account object

Name: Jerry, ID: 193423, Balance: 1250
Destroying an Account object

The constructor prints the first line when the main function initializes the acct object. The acct vari-
able’s display method prints the second line. The destructor prints the last line at the end of main’s
execution when the acct variable goes out of scope.

14.6 Defining a New Numeric Type

C++’s class feature allows us to define our own complete types. We will define a new numeric type that
models mathematical rational numbers. In mathematics, a rational number is defined as the ratio of two
integers, where the second integer must be nonzero. Commonly called a fraction, a rational number’s two
integer components are called numerator and denominator. Rational numbers possess certain properties;
for example, two fractions can have different numerators and denominators but still be considered equal
( 12 = 24 ). Listing 14.5 (simplerational.cpp) shows how to define and use rational numbers.

Listing 14.5: simplerational.cpp

©2013 Richard L. Halterman Draft date: January 27, 2013

 14.6. DEFINING A NEW NUMERIC TYPE 395

1 #include <iostream>
2
3 using namespace std;
4
5 // Models a mathematical rational number
6 class SimpleRational
7 {
8 int numerator;
9 int denominator;
10 public:
11 // Initializes the components of a Rational object
12 SimpleRational(int n, int d): numerator(n), denominator(d)
13 {
14 if ( d == 0 )
15 {
16 // Display error message
17 cout << "Zero denominator error" << endl;
18 exit(1); // Exit the program
19 }
20 }
21
22 // The default constructor makes a zero rational number
23 // 0/1
24 SimpleRational(): numerator(0), denominator(1) {}
25
26 // Allows a client to reassign the numerator
27 void set_numerator(int n)
28 {
29 numerator = n;
30 }
31
32 // Allows a client to reassign the denominator.
33 // Disallows an illegal fraction (zero denominator).
34 void set_denominator(int d)
35 {
36 if ( d != 0 )
37 denominator = d;
38 else
39 {
40 // Display error message
41 cout << "Zero denominator error" << endl;
42 exit(1); // Exit the program
43 }
44 }
45
46 // Allows a client to see the numerator's value.
47 int get_numerator()
48 {
49 return numerator;
50 }
51
52 // Allows a client to see the denominator's value.
53 int get_denominator()
54 {
55 return denominator;

©2013 Richard L. Halterman Draft date: January 27, 2013

 14.7. ENCAPSULATION 396

56 }
57 };
58
59 int main()
60 {
61 SimpleRational fract(1, 2); // The fraction 1/2
62 cout << "The fraction is " << fract.get_numerator()
63 << "/" << fract.get_denominator() << endl;
64 fract.set_numerator(19);
65 fract.set_denominator(4);
66 cout << "The fraction is " << fract.get_numerator()
67 << "/" << fract.get_denominator() << endl;
68 }

The SimpleRational class defines a new numeric type that C++ does not natively provide—the rational
number type. (It is named SimpleRational because it is our first cut at a rational number class; a better
version is to come in Listing 16.1 (rational.cpp).) This SimpleRational class defines two overloaded
constructors. One constructor accepts the numerator and denominator values from the client. The other
constructor allows a client to declare a SimpleRational object as
SimpleRational frac;

without supplying the initial numerator and denominator values. This default constructor assigns 10 to the
object. Both constructors ensure that a SimpleRational object’s denominator will not be zero.
Our new numeric type certainly leaves a lot to be desired. We cannot display one of our rational number
objects with cout very conveniently. We cannot use the standard arithmetic operators like + or ”*”, and
we cannot compare two rational numbers using == or <. In Section 16.1 we address these shortcomings.

14.7 Encapsulation

When developing complex systems, allowing indiscriminate access to an object’s internals can be disas-
trous. It is all too easy for a careless, confused, or inept programmer to change an object’s state in such a
way as to corrupt the behavior of the entire system. A malicious programmer may intentionally tweak one
or more objects to sabotage the system. In either case, if the software system controls a medical device or
military missile system, the results can be deadly.
C++ provides several ways to protect the internals of an object from the outside world, but the simplest
strategy is the one we have been using: fields and methods, generically referred to as class members, can
be qualified as either public or private.
The compiler enforces the inaccessibility of private members. In Listing 14.5 (simplerational.cpp), for
example, client code cannot directly modify the denominator instance variable of a Rational object
making it zero. A client may influence the values of numerator and denominator only via methods
provided by the class designer.
Accessibility rules, also called visibility rules or permissions, determine what parts of a class and/or
object are accessible to the outside world. C++ provides a great deal of flexibility in assigning access
permissions, but some general principles exist that, if followed, foster programs that are easier to build and
extend.

• In general, fields should be private. Clients should not be able to arbitrarily change the state of an
object. Allowing such might allow client code to put an object into an undefined state (for example,

©2013 Richard L. Halterman Draft date: January 27, 2013

14.7. ENCAPSULATION 397

changing the denominator of a fraction to zero). An object’s state should only change as a result of
calling the object’s methods.
The built-in primitive types like int and double offer no protection from client access. One ex-
ception to the private fields rule applies to simple objects that programmers naturally would treat as
primitive types. Recall the geometric Point class found in Listing 14.1 (mathpoints.cpp). The x
and y fields of a point object safely may assume any legitimate floating-point value, and it may be
reasonable in some applications for clients to treat a Point object as a primitive type.

• Methods that provide a service to client code should be part of the public section of the class. The
author of the class must ensure that the public methods cannot place the object into an illegal state.
For example, the method
void set_denominator(int d)
{
denominator = d;
}

if part of the Rational class would permit client code to sabotage a valid fraction with a simple
statement:
fract.set_denominator(0);

• Methods that assist the service methods but that are not meant to be used by the outside world should
be in the private section of the class. This allows a public method to be decomposed into sim-
pler, perhaps more coherent activities without the threat of client code accessing these more primitive
methods. These private methods are sometimes called helper methods or auxiliary methods.

Why would a programmer intentionally choose to limit access to parts of an object? Restricting access
obviously limits the client’s control over the objects it creates. While this may appear to be a disadvantage
at first glance, this access restriction actually provides a number of advantages:

• Flexibility in implementation. A class conceptually can be separated into two parts:

– The class interface—the public part. Clients see and can use the public parts of an object.
The public methods and public variables of a class constitute the interface of the class. A class’s
interface specifies what it does.
– The class implementation—the hidden part. Clients cannot see any private methods or pri-
vate variables. Since this private information is invisible to clients, class developers are free to
do whatever they want with the private parts of the class. A class’s implementation specifies
how it accomplishes what it needs to do.

We would like our objects to be black boxes: clients shouldn’t need to know how the objects work
but merely rely on what objects can do.

A good rule of thumb in class design is this: make data private, and make
methods that provide a service to clients public.

©2013 Richard L. Halterman Draft date: January 27, 2013

14.7. ENCAPSULATION 398

Many real-world objects follow this design philosophy. Consider a digital wristwatch. Its display
gives its user the current time and, perhaps, date. It can produce different output in different modes;
for examples, elapsed time in stopwatch mode or wake up time in alarm mode. It presents to its user
only a few buttons for changing modes, starting and stopping stopwatches, and setting the time. How
it does what is does is irrelevant to most users; most users are concerned with what it does. Its user
risks great peril by opening the watch and looking at its intricate internal details. The user is meant
to interact with the watch only through its interface—the display and buttons.
Similarly, an automobile presents an accelerator pedal to its user. The user knows that pushing the
pedal makes the car go faster. That the pedal is connected to the fuel injection system (and possibly
other systems, like cruise control) through a cable, wire, or other type of linkage is of concern only
to the automotive designer or mechanic. Most drivers prefer to be oblivious to the under-the-hood
details.
Changing the interface of a class can disturb client code that already has been written to use objects
of that class. For example, what if the maintainers of the Rational class decide that Rational
objects should be immutable; that is, once a Rational object is created the client cannot adjust the
numerator or denominator values. The set_numerator and set_denominator would
therefore disappear. Unfortunately, both of these methods are public and thus part of Rational’s
interface to client. Existing client code may be using these methods, and removing them, making
them private, altering the types of their parameters or any other changes to the interface would ren-
der existing client code incompatible. Client code that has been written to use set_numerator
according to its original interface no longer will be correct. We say the change in Rational’s
interface breaks the client code. Class
Class authors have no flexibility to alter the interface of a class once the class has been released for
clients to use. On the other hand, altering the private information in a class will not break existing
client code that uses that class, since private class information is invisible to clients. When the
private parts of a class change, clients need only recompile their code; client programmers do not
need to modify their source code. A class, therefore, becomes less resilient to change as more of
its components become exposed to clients. To make classes as flexible as possible, which means
maximizing the ability to make improvements to the class in the future, hide as much information as
possible from clients.
• Reducing programming errors. Parts of a class that are private cannot be misused by client code
since the client cannot see the private parts of a class. Properly restricting client access can make it
impossible for client code to put objects into an undefined state. In fact, if a client can coax an object
into an illegal state via the class interface, then the design and/or implementation of the class is faulty.
As an example, if a client can somehow make a Rational object’s denominator zero, then one of
the methods must be faulty. Clients should never be able to place an object into an illegal state.
• Hiding complexity. Objects can provide a great deal of functionality. Even though a class may
provide a fairly simple interface to clients, the services it provides may require a significant amount
of complex code to accomplish their tasks. One of the challenges of software development is dealing
with the often overwhelming complexity of the task.
It is difficult, if not impossible, for one programmer to be able to comprehend at one time all the de-
tails of a large software system. Classes with well-designed interfaces and hidden implementations
provide a means to reduce this complexity. Since private components of a class are hidden, their de-
tails cannot contribute to the complexity the client programmer must manage. The client programmer
needs not be concerned with exactly how an object works, but the details that make the object work
are present nonetheless. The trick is exposing details only when necessary:
– Class designer. The class designer must be concerned with the hidden implementation details
of the class. The class designer usually does not have to worry about the context in which

©2013 Richard L. Halterman Draft date: January 27, 2013

14.8. SUMMARY 399

the class will be used because the class may be used in many different contexts. From the
perspective of the class designer, the complexity of the client code that may use the class is
therefore eliminated.
– Applications developer using a class. The developer of the client code must be concerned with
the details of the application code being developed. The application code will use objects. The
hidden details of the class these objects represent are of no concern to the client developers.
From the perspective of the client code designer, therefore, the complexity of the code that
makes the objects work is eliminated.

This concept of information hiding is called encapsulation. Details are exposed to particular parties
only when appropriate. In sum, the proper use of encapsulation results in

• software that is more flexible and resilient to change,

• software that is more robust and reliable, and

• software development that is more easily comprehended and managed.

Finally, the C++ encapsulation model has its limits. It is not possible to protect an object from code
within itself. Any method within a class has full access to any member defined within that class. If it
appears that parts of a class should be protected from some of its methods, the class should be split up into
multiple classes with suitable restrictions among the resulting component classes.

14.8 Summary

• The class reserved word introduces a programmer-defined type.

• A semicolon (;) must follow the closing curly brace of a class declaration.

• Variables of a class are called objects or instances of that class.

• The public label within a class specifies the parts of objects of that class that are visible to code
outside of that class.

• The parts of a class that are not in the public section are by default private and thus inaccessible to
code outside of the class. The private label explicitly declares a section of a class to be private.
The public and private labels may appear multiple times within a class declaration allowing
programmers to order class members as they chose without being forced to put all the private mem-
bers in one place.

• The dot (.) operator is used to access elements of an object.

• A data member of a class is known as a field. Equivalent terms include data member, instance
variable, and attribute.

• A function defined in a class that operates on objects of that class is called a method. Equivalent
terms include member function and operation.

• A constructor represents code that executes automatically when an object comes into existence. A
constructor has the same name as the class and no specified return type (not even void).

©2013 Richard L. Halterman Draft date: January 27, 2013

14.9. EXERCISES 400

• A constructor initialization list assigns initial values to an object’s fields. The initialization list ap-
pears after the closing parenthesis of the constructor’s parameter list but before the opening curly
brace of the constructor’s body.

• A destructor represents code that executes automatically when an object goes out of scope. Destruc-
tors are useful for cleaning up any renewable resources (like dynamically allocated heap memory)
that the object may have allocated during its lifetime.

• Encapsulation and data hiding offers several benefits to programmers:

– Flexibility—class authors are free to change the private details of a class. Existing client code
need not be changed to work with the new implementation.
– Reducing programming errors—if client code cannot touch directly the hidden details of an
object, the internal state of that object is completely under the control of the class author. With
a well-designed class, clients cannot place the object in an ill-defined state (thus leading to
incorrect program execution).
– Hiding complexity—the hidden internals of an object might be quite complex, but clients cannot
see and should not be concerned with those details. Clients need to know what an object can
do, not how it accoplishes the task.

14.9 Exercises

1. Given the definition of the Rational number class in Listing 14.5 (simplerational.cpp), complete
the function named add:
Rational add(const Rational& r1, const Rational& r2)
{
// Details go here
}

that returns the rational number representing the sum of its two parameters.

2. Given the definition of the geometric Point class in Listing 15.3 (point.cpp), complete the function
named distance:
double distance(const Point& r1, const Point& r2)
{
// Details go here
}

that returns the distance between the two points passed as parameters.

3. Given the definition of the Rational number class in Section 14.3, complete the following function
named reduce:
Rational reduce(const Rational& r)
{
// Details go here
}

that returns the rational number that represents the parameter reduced to lowest terms; for example,
the fraction 10/20 would be reduced to 1/2.

©2013 Richard L. Halterman Draft date: January 27, 2013

14.9. EXERCISES 401

4. Given the definition of the Rational number class in Section 14.3, complete the following function
named reduce:
void reduce(Rational& r)
{
// Details go here
}

that uses call by reference to reduce to lowest terms the parameter passed to it; for example, the
fraction 10/20 would be reduced to 1/2.

5. What is the purpose of a class constructor?

6. May a class constructor be overloaded?

7. What are the consequences of declaring a method to be const?

8. Why are const methods necessary?

9. Given the definition of the Rational number class in Section 14.3, complete the following method
named reduce:
class Rational
{
// Other details omitted here ...

// Returns an object of the same value reduced

// to lowest terms
Rational reduce() const
{
// Details go here
}
};

that returns the rational number that represents the object reduced to lowest terms; for example, the
fraction 10/20 would be reduced to 1/2.

10. Given the definition of the Rational number class in Section 14.3, complete the following method
named reduce:
class Rational
{
// Other details omitted here ...

// Reduces the object to lowest terms

void reduce()
{
// Details go here
}
};

that reduces the object on whose behalf the method is called to lowest terms; for example, the fraction
10/20 would be reduced to 1/2.

11. Given the definition of the geometric Point class in Section 14.2 add a method named distance:

©2013 Richard L. Halterman Draft date: January 27, 2013

14.9. EXERCISES 402

class Point
{
// Other details omitted

// Returns the distance from this point to the

// parameter p
double distance(const Point& p)
{
// Details go here
}
};

that returns the distance between the point on whose behalf the method is called and the parameter p.
12. Consider the following C++ code:
#include <iostream>

using namespace std;

class IntPoint
{
public:
int x;
int y;
IntPoint(int x, int y): x(x), y(y) {}
};

class Rectangle
{
IntPoint corner; // Location of the rectangle's lower-left corner
int width; // The rectangle's width
int height; // The rectangle's width
public:
Rectangle(IntPoint pt, int w, int h):
corner((pt.x < -100) ? -100 : (pt.x > 100 ? 100 : pt.x),
(pt.y < -100) ? -100 : (pt.y > 100 ? 100 : pt.y)),
width((w < 0) ? 0 : w), height((h < 0) ? 0 : h) {}

int perimeter() const

{
return 2*width + 2*height;
}

int area() const

{
return width * height;
}

int get_width() const

{
return width;

©2013 Richard L. Halterman Draft date: January 27, 2013

14.9. EXERCISES 403

int get_height() const

{
return height;
}

// Returns true if rectangle r overlaps this

// rectangle object.
bool intersect(const Rectangle& r) const
{
// Details omitted
}

// Returns the length of a diagonal rounded to the nearest

// integer.
int diagonal() const
{
// Details omitted
}

// Returns the geometric center of the rectangle with

// the (x,y) coordinates rounded to the nearest integer.
IntPoint center() const
{
// Details omitted
}

bool is_inside(const IntPoint& pt) const

{
// Details omitted
}
};

int main()
{
Rectangle rect1(IntPoint(2, 3), 5, 7),
rect2(IntPoint(2, 3), 1, 3),
rect3(IntPoint(2, 3), 15, 3),
rect4(IntPoint(2, 3), 5, 3);
cout << rect1.get_width() << endl;
cout << rect1.get_height() << endl;
cout << rect2.get_width() << endl;
cout << rect2.get_height() << endl;
cout << rect3.get_width() << endl;
cout << rect3.get_height() << endl;
cout << rect4.get_width() << endl;
cout << rect4.get_height() << endl;
cout << rect1.get_perimeter() << endl;
cout << rect1.get_area() << endl;

©2013 Richard L. Halterman Draft date: January 27, 2013

14.9. EXERCISES 404

cout << rect2.get_perimeter() << endl;

cout << rect2.get_area() << endl;
cout << rect3.get_perimeter() << endl;
cout << rect3.get_area() << endl;
cout << rect4.get_perimeter() << endl;
cout << rect4.get_area() << endl;
}

(a) What does the program print?

(b) With regard to a Rectangle object’s lower-left corner, what are the minimum and maximum
values allowed for the x coordinate? What are the minimum and maximum values allowed for
the y coordinate?
(c) What is a Rectangle object’s minimum and maximum width?
(d) What is a Rectangle object’s minimum and maximum height?
(e) What happens when a client attempts to create a Rectangle object with parameters that are
outside the acceptable ranges?
(f) Implement the diagonal method.
(g) Implement the center method.
(h) Implement the intersect method.
(i) Implement the is_inside method.
(j) Complete the following function named corner:
IntPoint corner(const Rectangle& r)
{
// Details go here
};
that returns the lower-left corner of the Rectangle object r passed to it. You may not modify
the Rectangle class.
(k)

13. Develop a Circle class that, like the Rectangle class above, provides methods to compute
perimeter and area. The Rectangle instance variables are not appropriate for circles; specifi-
cally, circles do have corners, and there is no need to specify a width and height. A center point and
a radius more naturally describe a circle. Build your Circle class appropriately.
14. Given the Rectangle and Circle classes from questions above, write an encloses function:
// Returns true if rectangle rect is large enough to
// completely enclose circle circ
bool encloses(const Rectangle& rect, const Circle& circ)
{
// Details omitted
}

so that it returns true if circle circ’s dimensions would allow it to fit completely within rectangle
rect. If circ is too big, the function returns false. The positions of rect and circ do not
influence the result.
15. Consider the following C++ code:

©2013 Richard L. Halterman Draft date: January 27, 2013

14.9. EXERCISES 405

class Widget
{
int value;
public:
Widget();
Widget(int v);
int get();
void bump();
};

Widget::Widget()
{
value = 40;
}

Widget::Widget(int v)
{
if ( v >= 40 )
value = v;
else
value = 0;
}

int Widget::get() const

{
return value;
}

void Widget::bump()
{
if ( value < 50 )
value++;
}

int main()
{
Widget w1, w2(5);
cout << w1.get() << endl;
cout << w2.get() << endl;
w1.bump(); w2.bump();
cout << w1.get() << endl;
cout << w2.get() << endl;
for ( int i = 0; i < 20; i++ )
{
w1.bump();
w2.bump();
}
cout << w1.get() << endl;
cout << w2.get() << endl;
}

©2013 Richard L. Halterman Draft date: January 27, 2013

14.9. EXERCISES 406

(a) What does the program print?

(b) If wid is a Widget object, what is the minimum value the expression wid.get() can return?
(c) If wid is a Widget object, what is the maximum value the expression wid.get() can return?

©2013 Richard L. Halterman Draft date: January 27, 2013

 407

Chapter 15

More on Objects

15.1 Pointers to Objects and Object Arrays

Given the Point class from Section 14.2, the statement

Point pt;

declares the variable pt to be a Point object. As with primitive data, pointers can be used with objects:
Point pt;
Point *p_pt;

Here, p_pt is a pointer to a Point object. Before we use the pointer it must be initialized to point to a
valid object. We can assign the pointer to refer to an existing object, as in
p_pt = &pt;

or use the new operator to dynamically allocate an object from the heap:
p_pt = new Point;

If the class has a constructor that that accepts parameters, we need to provide the arguments when using
new. Recall the Account class from Section 14.4. Given the declarations
Account acct("Joe", 3143, 90.00);
Account *acct_ptr;

before we use acct_ptr it must point to a valid object. As in the Point class example, we can assign
the pointer to refer to an existing object, as in
acct_ptr = &acct;

or use the new operator to dynamically allocate an object from the heap:
acct_ptr = new Account("Joe", 400, 1300.00);

Note that we include the arguments expected by the Account class constructor. This statement allocates
memory for one Account object and executes the constructor’s code to initialize the newly created ob-
ject. Figure 15.1 illustrates a pointer to an account object. As with any dynamically allocated intity, a

©2013 Richard L. Halterman Draft date: January 27, 2013

15.1. POINTERS TO OBJECTS AND OBJECT ARRAYS 408

acct_ptr
name 'J' 'o' 'e' '\n'
0 1 2 3
id 400

balance 1300.00

Figure 15.1: A pointer to an Account object

programmer must be careful to use the delete operator to deallocate any objects created via new.
C++ supports arrays of objects, but they present special challenges. First, consider a simple situation.
The Point class defines no constructor, so the following code is valid:
Point pts[100]; // Okay

The compiler happily generates code that at run time will allocate enough space for 100 Point objects.
No special initializations are needed since Point has no constructor. What if a class defines a constructor
that accepts arguments and does not supply also a constructor that requires no arguments? Consider the
Account class. The following statement is illegal:
Account accts[100]; // Illegal, the Account class has no default
// constructor

When creating the space for the accts elements, C++ expects a default constructor properly to properly
initialize all of the array’s elements. The only constructor in the Account class requires arguments, and
so the compiler refuses to accept such a declaration. The compiler cannot ensure that the array’s elements
are properly initialized before the programmer begins using the array.
One solution uses an array of pointers, as in
Account *accts[100]; // An array of account pointers

Note that this does not create any Account objects. The programmer subsequently must iterate through
the array and use new to create individually each account element. An example of this would be
Account *accts[100]; // An array of account pointers
for ( int i = 0; i < 100; i++ )
{
// Get information from the user
cin >> name >> id >> amount;
// Create the new account object
accts[i] = new Account(name, id, amount);
}

The dot operator syntax to access a field of a object through a pointer is a bit awkward:

©2013 Richard L. Halterman Draft date: January 27, 2013

15.1. POINTERS TO OBJECTS AND OBJECT ARRAYS 409

Point *p = new Point;

(*p).x = 253.7;
(*p).y = -00.5;

The parentheses are required since the dot (.) has higher precedence than the pointer dereferencing operator
(*). Without the parentheses, the statement would be evaluated as if the parentheses were placed as shown
here

*(p.x) = 253.7; // Error

The variable p is not a Point object, but a pointer to a Point object, so it must be dereferenced with
the * operator before applying the . operator. C++ provides a simpler syntax to access fields of an object
through a pointer. The structure pointer operator eliminates the need for the parentheses:
Point *p = new Point;
p->x = 253.7;
p->y = -00.5;

The pair of symbols -> are treated as one operator (no space in between is allowed), and the operator is
meant to look like an arrow pointing right. There is no associated left pointing arrow in C++.
You can use the -> operator to access the methods of an object referenced by a pointer:
Account *acct_ptr = new Account("Joe", 400, 1300.00);
if ( acct_ptr->withdraw(10.00) )
cout << "Withdrawal successful" << endl;
else
cout << "*** Insufficient funds ***" << endl;

Recall from Section 11.8 that local variables and parameters declared within a function are automat-
ically allocated on the stack when the function begins executing and automatically deallocated when the
function completes and returns. The programmer does not have to explicitly allocate and deallocate the
space for local variables. Global variables and static local variables are allocated in static memory
when the program begins executing and are deallocated when the program is finished. The programmer
does not have to explicitly allocate and deallocate the space for global and static local variables. Mem-
ory allocated from the heap via new is not deallocated automatically. It is the programmer’s responsibility
to manually deallocate the memory when it is no longer needed. The delete operator with no [] deco-
ration frees up a single dynamically allocated memory:
// acct_ptr points to an Account object previously allocated via new
delete acct_ptr;

The following function has a logic error because it fails to free up memory allocated with new:
void faulty_func()
{
Account *acct_ptr = new Account("Joe", 400, 1300.00);
if ( acct_ptr->withdraw(10.00) )
cout << "Withdrawal successful" << endl;
else
cout << "*** Insufficient funds ***" << endl;
acct_ptr->display();
}

©2013 Richard L. Halterman Draft date: January 27, 2013

15.2. THE THIS POINTER 410

Here, acct_ptr is a local variable, so it occupies space on the stack. It represents a simple address,
a number. acct_ptr is automatically deallocated when faulty_func completes. The problem is
acct_ptr points to memory that was allocated with new, and this memory is not freed up within the
function. The pointer acct_ptr is the only way to get to that memory, so when the function is finished,
that memory is lost for the life of the program. The condition is known as a memory leak. If the program
runs to completion quickly, the leak may go undetected. In a longer running program such as a web server,
memory leaks can cause the program to crash after a period of time. The problem arises when code that
leaks memory is executed repeatedly and eventually all of available memory becomes used up.
The corrected function would be written
void faulty_func()
{
Account *acct_ptr = new Account("Joe", 400, 1300.00);
if ( acct_ptr->withdraw(10.00) )
cout << "Withdrawal successful" << endl;
else
cout << "*** Insufficient funds ***" << endl;
acct_ptr->display();
delete acct_ptr;
}

Recall (see 10.6) that in C++ there is one literal value to which a pointer of any type may be assigned—
nullptr:
Account *p_rec = nullptr; // p_rec is null

A pointer with the value nullptr is interpreted to mean a pointer that is pointing to nothing. An attempt
to delete a null pointer is legal and does nothing.
Pointers to objects are commonly used to build elaborate linked data structures. Section 16.4 describes
one such data structure.

15.2 The this Pointer

Recall our Counter class from Section 14.3, reproduced here:

class Counter
{
int count;
public:
// Allow clients to reset the counter to zero
void clear()
{
count = 0;
}

// Allow clients to increment the counter

void inc()
{
count++;
}

©2013 Richard L. Halterman Draft date: January 27, 2013

15.2. THE THIS POINTER 411

// Return a copy of the counter's current value

int get()
{
return count;
}
};

In Section 14.3 we saw that an expression such as ctr1.clear() passes the address of ctr1 as a secret
parameter during the call to the clear method. This is how the method determines which count field to
reset—ctr1.clear() resets ctr1’s count field, and ctr2.clear() would reset ctr2’s count
field.
Within a method definition a programmer may access this secret parameter via the reserved work this.
The this expression represents a pointer to the object upon which the method was called. The above
Counter class may expressed as
// Uses this directly
class Counter
{
int count;
public:
// Allow clients to reset the counter to zero
void clear()
{
this->count = 0;
}

// Allow clients to increment the counter

void inc()
{
this->count++;
}

// Return a copy of the counter's current value

int get()
{
return this->count;
}
};

Within a method of the Counter class, this->count is an alternate way of writing just count. Some
programmers always use the this pointer as shown here to better communicate to human readers that the
count variable has to be a field, and it cannot be a local or global variable.
The this pointer is handy when a method parameter has the same name as a field:
class Point
{
double x;
double y;
private:
void set_x(double x)

©2013 Richard L. Halterman Draft date: January 27, 2013

15.3. CONST METHODS 412

{
// Assign the parameter's value to the field
this->x = x;
}
// Other details omitted . . .
};

In the set_x method the parameter x has the same name as field x. This is legal in C++; a method
parameter or local variable of a method may have the same name as a field within that class. The problem
is that the local variable or parameter hides the access to the field. Any unqualified use of the name x refers
to the parameter x, not the field x. One solution would be to name the parameter something else: perhaps
_x or x_param. A strong argument can be made, though, that x is the best name for the field, and x is
also the best name for the parameter to the set_x method. To get access to the field in this case, use the
this pointer. Since this is a reserved word, the expression this->x cannot be mistaken for anything
other the x field of the object upon which the method was invoked.
Another use of the this pointer involves passing the current object off to another function or method.
Suppose a global function named log exists that accepts a Counter object as shown here:
void log(Counter c)
{
// Details omitted . . .
}

If within the clear method we wish to call the log function passing it the object on whose behalf the
clear method was executing, we could write
class Counter
{
// . . .
public:
void clear()
{
count = 0;
// Pass this object off to the log function
log(*this);
}
// . . .
};

We pass the expression *this as the actual parameter to log because the log function expects an object
(or object reference), not a pointer to an object. Remember the syntax of pointers: If this is a pointer,
*this is the object this to which this points.
Since this serves as the secret parameter passed to methods, it is illegal to use the this expression
outside of the body of a method.

15.3 const Methods

Given the SimpleRational class in Listing 14.5 (simplerational.cpp), the following code produces a
compiler error:

©2013 Richard L. Halterman Draft date: January 27, 2013

15.3. CONST METHODS 413

const SimpleRational fract(1, 2); // Constant fraction 1/2

fract.set_numerator(2); // Error, cannot change a constant

This behavior is desirable, since the set_numerator method can change the state of a SimpleRational
object, and our fract object is supposed to be constant. Unfortunately, this correct behavior is accidental.
The compiler does not analyze our methods to determine exactly what they do. Consider the following code
that also will not compile:
const SimpleRational fract(1, 2); // Constant fraction 1/2
cout << fract.get_numerator(); // Error!

Since the get_numerator method does not modify a SimpleRational object, we would expect that
invoking it on a constant object should be acceptable, but the compiler rejects it. Again, the compiler cannot
understand what set_numerator is supposed to do; specifically, it cannot determine that a method will
not change the state of an object. The programmer must supply some additional information to help the
compiler.
If a method is not supposed to change the state of an object, that method should be declared const. In
the SimpleRational class, the methods get_numerator and get_denominator simply return,
respectively, copies of the fraction’s numerator and denominator. Neither method is intended to modify any
instance variables. If we look at the code for those methods, we see that indeed neither method changes
anything about a SimpleRational object. What if the programmer made a mistake—perhaps a spurious
copy and paste error—and the statement
numerator = 0;

made its way into the get_numerator method? Unfortunately, the way things stand now, the compiler
cannot detect this error, and get_numerator will contain a serious logic error.
We can remove the possibility of such an error by declaring get_numerator (and get_denominator)
const:
class SimpleRational
{
public:
/* ... stuff omitted ... */
int get_numerator() const
{
return numerator;
}
int get_denominator() const
{
return denominator;
}
/* ... other stuff omitted ... */
};

The const keyword goes after the closing parenthesis of the parameter list and before the opening curly
brace of the methods body. If we accidentally attempt to reassign an instance variable in a const method,
the compiler will report an error.
Declaring a method const is not merely a good defensive programming strategy used by a class
developer. Methods declared to be const can be called with const objects, while it is illegal to in-
voke a non-const method with a const object. With the new const version of SimpleRational’s
get_numerator method, the following code

©2013 Richard L. Halterman Draft date: January 27, 2013

 15.4. SEPARATING METHOD DECLARATIONS AND DEFINITIONS 414

const Rational fract(1, 2); // Constant fraction 1/2

fract.set_numerator(2); // Error, cannot change a constant

is still illegal, since Rational::set_numerator is not const, but

const Rational fract(1, 2); // Constant fraction 1/2
cout << fract.get_numerator(); // OK with const get_numerator

now is legal.
Any method that has no need to change any field within the object should be declared const. Any
method that is supposed to change a field in an object should not be declared const. A const method can
be used with both const and non-const objects, but a non-const method cannot be used with const
objects. For maximum flexibility, always declare a method to be const unless doing so would prevent the
method from doing what it is supposed to do.

15.4 Separating Method Declarations and Definitions

It is common in larger C++ projects to separate a method implementation from its declaration. A simple
example illustrates how to do this; consider the class MyClass:
class MyClass
{
public:
void my_method() const
{
cout << "Executing \"my_method\"" << endl;
}
};

This version of MyClass uses what is known as an inline method definition; that is, the method my_method
is defined completely with a body in the same place as it is declared. Compare the inline version to the fol-
lowing equivalent representation split across two files: myclass.h and myclass.cpp. The code in myclass.h
is shown in Listing 15.1 (myclass.h).

Listing 15.1: myclass.h

1 // File myclass.h
2
3 class MyClass
4 {
5 public:
6 void my_method() const; // Method declaration
7 };

The code in myclass.cpp is shown in Listing 15.2 (myclass.cpp).

Listing 15.2: myclass.cpp

1 // File myclass.cpp
2
3 // Include the class declaration
4 #include "myclass.h"

©2013 Richard L. Halterman Draft date: January 27, 2013

 15.4. SEPARATING METHOD DECLARATIONS AND DEFINITIONS 415

5 #include <iostream>
6
7 // Method definition
8 void MyClass::my_method() const
9 {
10 std::cout << "Executing \"my_method\"" << std::endl;
11 }

No body for my_method appears in the declaration of the class MyClass in myclass.h; instead, the
method implementation appears elsewhere, in myclass.cpp.
Without the prefix MyClass:: the definition of my_method in myclass.cpp simply would be global
function definition like all the ones we have seen since Chapter 9. The class name and scope resolution
operator (::) binds the definition of the method to the class to which is belongs.
The .h header file would be #included by all the source files that need to use MyClass objects; the
myclass.cpp file would be compiled separately and linked into the rest of the project’s .cpp files.
A Point class written inline as
class Point
{
double x;
double y;
public:
Point(double x, double y): x(x), y(y) {}
double get_x() const { return x; }
double get_y() const { return y; }
double set_x(double v) { x = v; }
double set_y(double v) { y = v; }
};

could be split up into the declaration:

class Point
{
double x;
double y;
public:
Point(double x, double y); // No constructor implementation
double get_x() const; // and no method
double get_y() const; // implementations`
double set_x(double v);
double set_y(double v);
};

and method implementations:

// Point constructor
Point::Point(double x, double y): x(x), y(y) {}

// Get the x value

double Point::get_x() const
{

©2013 Richard L. Halterman Draft date: January 27, 2013

15.4. SEPARATING METHOD DECLARATIONS AND DEFINITIONS 416

return x;
}

// Get the y value

double Point::get_y() const
{
return y;
}

// Set the x value

double Point::set_x(double v)
{
x = v;
}

// Set the y value

double Point::set_y(double v)
{
y = v;
}

The class name prefix such as Point:: is necessary not only so a method definition such as get_x can be
distinguished from a global function definition, but also to distinguish it from a different class that contains
a method with the same name and parameter types. A method signature for a method is just like a global
function signature, except a method signature includes that class name. Each of the following represent
distinct signatures:

• get_x() is the signature for a global function named get_x that accepts no parameters.
• Point::get_x() is the signature for a method of the Point class named get_x() that accepts
no parameters.
• LinearEquation::get_x() is the signature for a method of the LinearEquation class
named get_x() that accepts no parameters.

Recall from Section 10.3 that the return type is not part of a function’s signature. The same is true for
methods.
Many C++ programmers avoid the inline style of class declarations and use the separate class declaration
and method definition files for several reasons:

• If the class is meant to be used in multiple programs, the compiler must recompile the methods each
time the header file is #included by some C++ source file. When the method declarations and
definitions are separate the compiler can compile the code for the definitions once, and this compiled
code can be linked with new client code that uses it.
• Client programmers can look at the contents of header files. If method definitions are inlined, client
programmers can see exactly how the methods work. This can be a disadvantage; for example, a
client programmer may make assumptions about how fast a method takes to execute or the particular
order in which a method processes data in a vector. These assumptions can influence how the client
code calls the method. If the class maintainer changes the implementation of the method, the client
programmer’s previous assumptions may be invalid. A certain ordering of the data that before the

©2013 Richard L. Halterman Draft date: January 27, 2013

 15.4. SEPARATING METHOD DECLARATIONS AND DEFINITIONS 417

change resulted in faster processing may now be slower. An improvement in a graphics processing
algorithm may cause an animation to run too quickly when the method is rewritten. For the class
designer’s maximum flexibility, client programmers should not be able to see the details of a method’s
implementation because then they cannot form such assumptions.
Client programmers need to know what the method does, not how it accomplishes it.
To enforce this hidden method implementation:
1. Separate the method declarations from their definitions. Put the class declaration in the header
file and the method definitions in a .cpp file.
2. Compile the .cpp file into an object file.
3. Provide the client the .h file and the compiled object file but not the source file containing the
method definitions.
4. The client code can now be compiled by including the .h file and linked with the object file,
but the client programmer has no access to the source code of the method definitions.
• Some programmers find the inline style difficult since it provides too much detail. It complicates
determining what objects of the class are supposed to do because the how is sprinkled throughout the
class declaration.

Despite the disadvantages mentioned above, inline methods are sometimes appropriate. For simple
classes defined in the same file that uses them, the inline style is handy, as Listing 15.3 (point.cpp) shows.

Listing 15.3: point.cpp

1 #include <iostream>
2
3 class Point
4 {
5 double x;
6 double y;
7 public:
8 Point(double x, double y): x(x), y(y) {}
9 double get_x() const { return x; }
10 double get_y() const { return y; }
11 double set_x(double v) { x = v; }
12 double set_y(double v) { y = v; }
13 };
14
15 double dist(const Point& pt1, const Point& pt2)
16 {
17 // Compute distance between pt1 and pt2 and return it
18 // This is a function stub; add the actual code later
19 return 0.0; // Just return zero for now
20 }
21
22 int main()
23 {
24 Point p1(2.5, 6), p2(0.0, 0.0);
25 std::cout << dist(p1, p2) << std::endl;
26 }

Here the Point class is very simple, and the constructor and method implementations are obvious. The
inline structure makes more sense in this situation.

©2013 Richard L. Halterman Draft date: January 27, 2013

 15.4. SEPARATING METHOD DECLARATIONS AND DEFINITIONS 418

Listing 15.4 (trafficlight.h) contains the interface for a class used to create objects that simulate stop-
caution-go traffic signals.

Listing 15.4: trafficlight.h

1 class Trafficlight
2 {
3 public:
4 // Class constants available to clients
5 enum SignalColor { Red, Green, Yellow };
6 private:
7 SignalColor color; // The light's current color: Red, Green, or Yellow
8 public:
9
10 Trafficlight(SignalColor initial_color);
11 void change();
12 SignalColor get_color() const;
13 };

Notice the Trafficlight class defines a public enumerated type (SignalColor) that clients may
use. Because the enumerated type is defined inside the class, clients must use the fully-qualified name
Trafficlight::SignalColor when declaring variables or parameters of this type.
The traffic signal implementation code in Listing 15.5 (trafficlight.cpp) defines the Trafficlight
methods.

Listing 15.5: trafficlight.cpp

1 #include "trafficlight.h"
2
3 // Ensures a traffic light object is in the state of
4 // red, green, or yellow. A rogue value makes the
5 // traffic light red
6 Trafficlight::Trafficlight(Trafficlight::SignalColor initial_color)
7 {
8 switch ( initial_color )
9 {
10 case Red:
11 case Green:
12 case Yellow:
13 color = initial_color;
14 break;
15 default:
16 color = Red; // Red by default, just in case
17 }
18 }
19
20 // Ensures the traffic light's signal sequence
21 void Trafficlight::change()
22 {
23 // Red --> green, green --> yellow, yellow --> red
24 // Treat enum type as int, compute successor color, then
25 // covert back to enum type.
26 color = static_cast<SignalColor>((color + 1) % 3);
27 }
28

©2013 Richard L. Halterman Draft date: January 27, 2013

 15.4. SEPARATING METHOD DECLARATIONS AND DEFINITIONS 419

29 // Returns the light's current color so a client can

30 // act accordingly
31 Trafficlight::SignalColor Trafficlight::get_color() const
32 {
33 return color;
34 }

The code within Listing 15.5 (trafficlight.cpp) must #include the trafficlight.h header so the compiler is
exposed to the Trafficlight class declaration; otherwise, when compiling trafficlight.cpp the compiler
would not know whether or not the programmer was writing the Trafficlight class method imple-
mentations correct. If the method definition of TrafficLight::change in trafficlight.cpp specified
parameters but its declaration within trafficlight.h did not, that would be a problem. Furthermore, without
including trafficlight.h, the type SignalColor would be an undefined type within trafficlight.cpp.
Notice that outside the class declaration in Listing 15.4 (trafficlight.h) must prefix any members the
TrafficLight class with TrafficLight::. Without this prefix the compiler would treat the iden-
tifiers as globals. TrafficLight::SignalColor is the type declared in the TrafficLight class,
but SignalColor by itself is treated as a global type (that does not exist).
Listing 15.6 (trafficmain.cpp) shows how a client could use the Trafficlight class.

Listing 15.6: trafficmain.cpp

1 #include <iostream>
2 #include "trafficlight.h"
3
4 using namespace std;
5
6 void print(Trafficlight lt)
7 {
8 Trafficlight::SignalColor color = lt.get_color();
9 cout << "+-----+" << endl;
10 cout << "| |" << endl;
11 if ( color == Trafficlight::Red )
12 cout << "| (R) |" << endl;
13 else
14 cout << "| ( ) |" << endl;
15 cout << "| |" << endl;
16 if ( color == Trafficlight::Yellow )
17 cout << "| (Y) |" << endl;
18 else
19 cout << "| ( ) |" << endl;
20 cout << "| |" << endl;
21 if ( color == Trafficlight::Green )
22 cout << "| (G) |" << endl;
23 else
24 cout << "| ( ) |" << endl;
25 cout << "| |" << endl;
26 cout << "+-----+" << endl;
27 }
28
29 int main()
30 {
31 Trafficlight light(Trafficlight::Green);
32 while ( true )

©2013 Richard L. Halterman Draft date: January 27, 2013

 15.4. SEPARATING METHOD DECLARATIONS AND DEFINITIONS 420

33 {
34 print(light);
35 light.change();
36 cin.get();
37 }
38 }

Listing 15.6 (trafficmain.cpp) first prints

+-----+
| |
| ( ) |
| |
| ( ) |
| |
| (G) |
| |
+-----+

When the user presses the Enter key the program prints

+-----+
| |
| ( ) |
| |
| (Y) |
| |
| ( ) |
| |
+-----+

a second Enter press prints

+-----+
| |
| (R) |
| |
| ( ) |
| |
| ( ) |
| |
+-----+

A third Enter press completes the cycle:

©2013 Richard L. Halterman Draft date: January 27, 2013

15.5. PREVENTING MULTIPLE INCLUSION 421

+-----+
| |
| ( ) |
| |
| ( ) |
| |
| (G) |
| |
+-----+

The program’s execution continues in this manner within its infinite loop until the user presses Ctrl C .

15.5 Preventing Multiple Inclusion

C++ follows the one definition rule, which means a variable, function, class, instance variable, or method
may not have more than one definition in the same context. The code within the following function defini-
tion violates the one definition rule:
int sum_and_display(const vector<int>& list)
{
int s = 0, size = list.size();
// Add up the values in the list
int i = 0;
while ( i < size )
{
s += list[i];
i++;
}
// Print the contents of the list
int i = 0; // Illegal, variable i already defined above
while ( i < size )
{
cout << setw(5) << list[i] << endl;
i++;
}
cout << "--------" << endl;
cout << setw(5) << s << endl;
}

The line
int s = 0, size = list.size();

is not a problem even though the identifier size appears twice. The first appearance denotes the local
variable named size, and the second use is a call to the size method of the std::vector class. Only
the variable size is being defined (declared) here; the size method is defined in the std::vector
class.

©2013 Richard L. Halterman Draft date: January 27, 2013

15.5. PREVENTING MULTIPLE INCLUSION 422

The variable i has two distinct definitions in the same context, so its redeclaration right before the
display code is an error.
Like a variable, a class may have only one definition. When we build general purpose classes that can
be used widely in our programs we must take care that we do not violate the one definition rule. In fact, we
can violate the one definition rule accidentally even if we define a class only once!
To see how we can accidentally violate the one definition rule, consider the following simple example.
Recall Listing 16.2 (stopwatch.h) and Listing 16.3 (stopwatch.cpp) which provided stopwatch objects we
can use for timing program execution. Next, consider the following two header files defining objects that
themselves use Stopwatch objects, widget.h:
// File widget.h

#include "Stopwatch.h"

class Widget
{
Stopwatch clock;
public:
// Other stuff omitted
};

and gadget.h:
// File gadget.h

#include "Stopwatch.h"

class Gadget
{
Stopwatch timer;
public:
// Other stuff omitted
};

The preprocessor directive

#include "Stopwatch.h"

is required in the file widget.h, because the compiler must be acquainted with the type Stopwatch when
the instance variable clock is declared. Similarly, the Stopwatch.h header must be included in gadget.h
in order for the compiler to accept the declaration of the timer field.
Client code that wishing to use Widget objects must use the appropriate #include statement:
#include "widget.h"

int main()
{
Widget myWidget;
// Use the myWidget object
}

Similarly, client code that uses Gadget objects must include at the top of its file:

©2013 Richard L. Halterman Draft date: January 27, 2013

15.5. PREVENTING MULTIPLE INCLUSION 423

#include "gadget.h"

int main()
{
Gadget myGadget;
// Use the myGadget object
}

Sometimes, however, a program may need to use both Widget objects and Gadget objects. Since the
widget.h header file does not know anything about the Gadget class and the gadget.h header file has
no information about Widgets, we must include both header files in code that uses both classes. The
appropriate include directives would be
#include "widget.h"
#include "gadget.h"

int main()
{
Widget myWidget;
Gadget myGadget;
// Use the myWidget and myGadget objects
}

Including one of the header files without the other is insufficient since the compiler must be able to check if
the client is using both types correctly. This client code, however, will not compile. In this case the problem
is with 16.2, Stopwatch.h. Including both widget.h and gadget.h includes the definition of the Stopwatch
class twice, so the compiler sees two definitions Stopwatch. Even though the two definitions are identi-
cal, the one definition rule is violated so the compiler generates an error.
This problem is known as multiple inclusion. Multiple inclusion is a problem because the programmer
may have a legitimate need for both types of objects within the same program. Fortunately, the solution is
simple. The standard way to prevent multiple inclusion is to wrap a class definition as follows:
#ifndef _STOPWATCH_H_
#define _STOPWATCH_H_

#include <ctime>

class Stopwatch
{
clock_t start_time;
clock_t end_time;
bool running;
public:
Stopwatch();
void start();
void stop();
double elapsed() const;
};

#endif

©2013 Richard L. Halterman Draft date: January 27, 2013

15.6. OVERLOADED OPERATORS 424

Operator Class Function Example Chapter

operator<< std::ostream Right-hand operand is sent to the stream specified cout << x; Chapter 2
by the left-hand operand
operator>> std::istream Right-hand operand is extracted from the stream cin >> x; Chapter 4
specified by the left-hand operand
operator[] std::vector Right-hand operand (the integer within the square x = v[2]; Section 13.2
brackets) is used to locate an element within the left-
hand operand (a vector)
operator== std::string Right-hand operand (a string) is compared with if ( word == "Please") proceed(); Chapter 13
the left-hand operator (also a string) to determine
if they contain exactly the same characters in exactly
the same order

Table 15.1: Some Overloaded Operators for Objects

Do not use semicolons at the end of the lines beginning with #ifndef, #define, and #endif
because these are preprocessor directives not C++ statements. The identifier following the #ifndef and
#define preprocessor directives should be related to the file name of the header file. The convention is
to use all capital letters with underscores in place of dots as shown above. Putting underscores before and
after the header file name further makes it less likely that name will be used elsewhere within the program.
If the header file is named myheader.h, the preprocessor wrapper would be
#ifndef _MYHEADER_H_
#define _MYHEADER_H_

/* Declare your classes here */

#endif

The net effect of the #ifndef/#define/#endif directives is that the preprocessor will not include the
header file more than once when it is processing a C++ source file. This means the compiler will see the
class definition exactly once, thus satisfying the one definition rule.
The #ifndef...#define...#endif preprocessor directives should be used to wrap the class decla-
ration in the header file and do not appear in the .cpp file containing the method definitions. Since you
cannot always predict how widespread the use of a class will become, it is good practice to use this prepro-
cessor trick for all general purpose classes you create. By doing so you will avoid the problem of multiple
inclusion.

15.6 Overloaded Operators

Many of the C++ operators can be overloaded to work with programmer-defined types. Table 15.1 shows
some operators that have been defined to work with several of the standard object classes.
We can define how specific operators work for the types we devise. Operators for classes are expressed
as either global functions or member functions.

15.6.1 Operator Functions

Consider a simple mathematical point class:

class Point
{

©2013 Richard L. Halterman Draft date: January 27, 2013

15.6. OVERLOADED OPERATORS 425

public:
double x;
double y;
};

Suppose we wish to define the operation of addition on points as follows:

(x1 , y1 ) + (x2 , y2 ) = (x1 + y1 , x2 + y2 )

Thus, the x coordinate of the sum of two points is the sum of the x coordinates, and the y coordinate is the
sum of the y coordinates.
We can overload the addition operator to work on Point objects with the following global function:
Point operator+(const Point& p1, const Point& p2)
{
Point result;
result.x = p1.x + p2.x;
result.y = p1.y + p2.y;
return result;
}

With a slightly more sophisticated Point definition:

class Point
{
double x; // Fields are now private
double y;
public:
// Add a constructor
Point(double _x, double _y);
// Add some methods to access the private data members
double getX() const;
double getY() const;
};

with the expected constructor and method implementations:

// Initialize fields
Point::Point(double _x, double _y): x(_x), y(_y) {}

double Point::getX() const

{
return x; // Return a copy of the x member
}

double Point::getY() const

{
return y; // Return a copy of the y member
}

the operator function could be written in one line:

©2013 Richard L. Halterman Draft date: January 27, 2013

15.6. OVERLOADED OPERATORS 426

Point operator+(const Point& p1, const Point& p2)

{
return Point(p1.getX() + p2.getX(), p1.getY() + p2.getY());
}

The output stream << operator is often overridden for custom classes. The cout object (and, therefore,
the ostream class) has overloaded methods named operator<< that allow the primitive types like
integers and floating-point numbers to be printed. If we create a new type, such as Stopwatch, the
ostream class has no operator<<method built in to handle objects of our new type. In order to use
cout’s << operator with programmer-defined objects we must define a global operator function of the
form:

ostream& operator<<(ostream& os, const X& x)

where X represents a programmer-defined type. Notice that the function returns type ostream&. This is
because the first parameter is also an ostream&, and the function returns the same object that was passed
into it. If x is a programmer-defined type, the expression

operator<<(cout, x)

thus evaluates to a reference to cout. This process of returning the object that was passed allows the <<
operator to be chained, as in

cout << x << y << endl;

This is the beautified syntax for

operator<<(operator<<(operator<<(operator<<(cout, x), y)), endl)

If you examine this expression carefully, you will see that the first argument to all the calls of operator<<
is simply cout.
For our enhanced Point class, operator<< would look like:

ostream& operator<<(ostream& os, const Point& pt)

{
os << '(' << pt.getX() << ',' << pt.getY() << ')';
return os;
}

The function simply returns the same ostream object reference that was passed into it via the first param-
eter.
Given the above definitions of operator+ and operator<<, clients can write code such as

Point my_point(1, 2), your_point(0.45, 0);

cout << "Point 1: " << my_point << endl;
cout << "Point 2: " << my_point << endl;
cout << my_point << " + " << your_point
<< " = " << my_point + your_point << endl;

which displays

©2013 Richard L. Halterman Draft date: January 27, 2013

15.6. OVERLOADED OPERATORS 427

Point 1: (1.0, 2.0)

Point 2: (0.45, 0.0)
(1.0, 2.0) + (0.45, 0.0) = (1.45, 2.0)

When class developers provide such an operator<< function, clients can print out objects just as easily
as printing the built in types.
The << operator is not overloaded for the std::vector class, but we now can easily do it ourselves
as needed. For a vector of integers, the following function
ostream& operator<<(ostream& os, const vector<int>& vec)
{
os << '{';
vector<int>::const_iterator iter = vec.begin(),
end = vec.end();
if ( iter != end )
{
os << *iter;
iter++;
while ( iter != end )
{
os << ',' << *iter;
iter++;
}
}
os << '}';
return os;
}

provides the necessary functionality. Given this operator<< function, clients can display vectors as they
would a built in type:
vector<int> list(10, 5);
cout << list << endl;

This code fragment prints

{5,5,5,5,5,5,5,5,5,5}

The statement
vector<int>::const_iterator iter = vec.begin(),
end = vec.end();

declares and initializes two vector iterator objects. Using C++ś type inference capability (see Section 3.9),
we can write this statement more simply:
auto iter = vec.begin(),
end = vec.end();

©2013 Richard L. Halterman Draft date: January 27, 2013

15.6. OVERLOADED OPERATORS 428

The non-iterator version of operator<< for vectors would look like:

ostream& operator<<(ostream& os, const vector<int>& vec)
{
os << '{';
int n = vec.size();
if ( n != 0 )
{
os << vec[0]
for ( int i = 1; i < n; i++ )
os << ',' << vec[i];
}
os << '}';
return os;
}

We can redefine the == operator for vectors of integers as:

bool operator==(const vector<int>& v1, const vector<int>& v2)
{
int n = v1.size();
if ( n == v2.size() ) // Vectors must be the same size
{ // to be equal
for ( int i = 0; i < n; i++ )
if ( v1[i] != v2[i] ) // Find a non-match?
return false;
return true; // All match
}
return false; // Different sizes, so not equal
}

Once we have the operator== function, the operator!= is easy:

bool operator!=(const vector<int>& v1, const vector<int>& v2)
{ // Uses operator== function defined above
return !(v1 == v2);
}

15.6.2 Operator Methods

A class may define operator methods. A method for a unary operator accepts no parameters, and a method
for a binary operator accepts only one parameter. The missing parameter is the object upon which the
operator is applied. To see how this works, consider an enhanced rational class:
class EnhancedRational
{
int numerator;
int denominator;
public:
/* Other details omitted */

©2013 Richard L. Halterman Draft date: January 27, 2013

15.7. STATIC MEMBERS 429

// Unary plus returns the double-precision floating-point

// equivalent of the rational number.
double operator+() const
{
return static_cast<double>(numerator)/denominator;
}

// Binary plus returns the rational number that results

// from adding another rational number (the parameter) to
// this object.
Rational operator+(Rational other) const
{
int den = denominator * other.denominator,
num = numerator * other.denominator
+ other.numerator * denominator;
Rational result(num, den);
return result;
}
};

The following code fragment shows how clients can use these operator methods:
Rational fract1(1, 2), fract2(1, 3), fract3;
double value = +fract1; // Assigns 0.5
fract3 = fract1 + fract2; // fract3 is 5/6

Remember that the statement

fract3 = fract1 + fract2;

is syntactic sugar for the equivalent statement

fract3 = fract1.operator+(fract2);

Here we see how the righthand operand of the plus operator, fract2, becomes the single parameter to the
binary operator+ method of EnhancedRational.
The statement
double value = +fract1;

is the syntactically sugared version of

double value = fract1.operator+();

While programmers may change the meaning of many C++ operators in the context of objects, the
precedence and associativity of all C++ operators are fixed. In the Rational class, for example, it is
impossible to enable the + operator to have a higher precedence than the * in the context of Rational
objects.

15.7 static Members

Variables declared in a class declaration are known as instance variables because each instance (object)
of that class maintains its own copy of the variables. This allows us, for example, to create two different

©2013 Richard L. Halterman Draft date: January 27, 2013

 15.7. STATIC MEMBERS 430

Rational number objects, and changing the numerator of one Rational object will not affect the
numerator of the other object.
Sometimes it is convenient to have variables or constants that are shared by all objects of a class. Global
variables and constants certainly will work, but globals are not tied to any particular class. C++ uses the
static keyword within a class to specify that all objects of that class share a field; for example,
class Widget
{
int value;
static int quantity;
};

Each individual Widget object has its own value variable, but only one quantity variable exists and
is shared by all Widget objects.
One unusual thing about static fields is that must be defined outside of the class declaration. For the
Widget class above, the statement
int Widget::quantity;

must appear somewhere in the source code outside of the class declaration.
Consider a widget factory in which each widget object must have a unique serial number. Serial numbers
are sequential, and a new widget object’s serial number is one larger than the widget produced immediately
before. Important for warranty claims, a client should not be able to alter a serial number of a widget
object. The Widget class in Listing 15.7 (serialnumber.cpp) shows how a static variable can be used
to manage serial numbers for widget objects.

Listing 15.7: serialnumber.cpp

1 #include <iostream>
2
3 using namespace std;
4
5 class Widget
6 {
7 // All Widget objects share serial_number_source
8 static int serial_number_source;
9 // Each Widget object is supposed to have its own
10 // unique serial_number
11 int serial_number;
12 public:
13 // A Widget object's serial number is initialized from
14 // the shared serial_number_source variable which is
15 // incremented each time a Widget object is created
16 Widget(): serial_number(serial_number_source++) {}
17 // Client's can look at the serial number but not touch
18 int get_serial_number() const
19 {
20 return serial_number;
21 }
22 };
23
24 // Initialize the initial serial number; the first
25 // Widget made will have serial number 1

©2013 Richard L. Halterman Draft date: January 27, 2013

 15.7. STATIC MEMBERS 431

26 int Widget::serial_number_source = 1;
27
28 // Make some widgets and check their serial numbers
29 int main()
30 {
31 Widget w1, w2, w3, w4;
32 cout << "w1 serial number = " << w1.get_serial_number() << endl;
33 cout << "w2 serial number = " << w2.get_serial_number() << endl;
34 cout << "w3 serial number = " << w3.get_serial_number() << endl;
35 cout << "w4 serial number = " << w4.get_serial_number() << endl;
36 }

The output of Listing 15.7 (serialnumber.cpp) is

w1 serial number = 1
w2 serial number = 2
w3 serial number = 3
w4 serial number = 4

Each Widget object has its own serial_number variable, but they all have access in the constructor to
the shared serial_number_source variable. The executing program initializes serial_number_source
one time at the beginning of the program’s execution before it calls the main function. This means
serial_number_source is properly initialized to 1 before the program creates any Widget objects.
C++ programmers often use class static fields to provide public constants available to clients. Con-
sider Listing 15.8 (trafficsignal.h) that models a simple traffic light a little differently from Listing 15.4
(trafficlight.h).

Listing 15.8: trafficsignal.h

1 #define _TRAFFICSIGNAL_H_
2
3 class TrafficSignal
4 {
5 int color; // The light's current color: RED, GREEN, or YELLOW
6 public:
7 // Class constants available to clients
8 static const int RED = 0;
9 static const int GREEN = 1;
10 static const int YELLOW = 2;
11
12 TrafficSignal(int initial_color);
13 void change();
14 int get_color() const;
15 };
16 #endif

The state of a traffic light object—which of its lamps is illuminated—is determined by an integer value:
0 represents red, 1 stands for green, and 2 means yellow. It is much more convenient for clients to use
the symbolic constants RED, GREEN, and YELLOW than to try to remember which integer values stand for
which colors. These constants are public, so clients can freely access them, but, since they are constants,
clients cannot alter their values.

©2013 Richard L. Halterman Draft date: January 27, 2013

 15.7. STATIC MEMBERS 432

Listing 15.9 (trafficsignal.cpp) provides the implementation of the methods of TrafficSignal class.

Listing 15.9: trafficsignal.cpp

1 #include "trafficsignal.h"
2
3 //const int Trafficlight::RED = 0,
4 // Trafficlight::GREEN = 1,
5 // Trafficlight::YELLOW = 2;
6
7 // Ensures a traffic light object is in the state of
8 // red, green, or yellow. A rogue integer value makes the
9 // traffic light red
10 TrafficSignal::TrafficSignal(int initial_color)
11 {
12 switch ( initial_color )
13 {
14 case RED:
15 case GREEN:
16 case YELLOW:
17 color = initial_color;
18 break;
19 default:
20 color = RED; // Red by default
21 }
22 }
23
24 // Ensures the traffic light's signal sequence
25 void TrafficSignal::change()
26 {
27 // Red --> green, green --> yellow, yellow --> red
28 color = (color + 1) % 3;
29 }
30
31 // Returns the light's current color so a client can
32 // act accordingly
33 int TrafficSignal::get_color() const
34 {
35 return color;
36 }

A client can create a traffic light object as

TrafficSignal light(TrafficSignal::YELLOW);

This statement makes an initially yellow traffic light. Since the RED, GREEN, and YELLOW public fields are
constants, clients cannot use them to control the state of a traffic light object. The name Trafficlight::YELLOW
will not conflict with any other globally available name.
It may not be obvious, but the color constants in the TrafficSignal class must be declared static.
To see why, consider a minor change to TrafficSignal in the class we will call TrafficSignalAlt:
class TrafficSignalAlt
{
int color; // The light's current color: RED, GREEN, or YELLOW
public:

©2013 Richard L. Halterman Draft date: January 27, 2013

15.8. CLASSES VS. STRUCTS 433

// Constant instance fields available to clients

const int RED; // Note: NOT static
const int GREEN;
const int YELLOW;

TrafficSignalAlt(int initial_color);
void change();
int get_color() const;
};

The TrafficSignalAlt class is identical to TrafficSignal, except that the color constants are con-
stant instance fields instead of constant class (static) fields. Observe that just like the TrafficSignal
class, the TrafficSignalAlt class has no default constructor. Recall from Section 14.4 that a default
constructor accepts no arguments, and that if a programmer provides any constructor for a class, the com-
piler will not automatically provide a default constructor. This means the client cannot write code such
as
TrafficSignal light;

or
TrafficSignalAlt light2;

During the object’s creation the client must provide an integer argument representing a traffic light color.
If RED, GREEN, and YELLOW are constant instance variables (that is, constant non-static fields), every
TrafficSignalAlt object has its own copy of the fields, and the RED, GREEN, and YELLOW fields
cannot exist outside of any traffic light object. This leads to a chicken-and-egg problem—how can we
create the first TrafficSignalAlt object using the symbolic constants RED, GREEN, or YELLOW?
These constants do not exist unless we have a traffic light object, yet we need a traffic light object to have
any of these constants!
During a program’s execution, static class fields are created before main begins executing. This
means any data pertaining to a class that must exist before any object of that class is created must be
declared static. Static class variables exist outside of any instance of that class.
C++ allows methods to be declared static. A static method executes on behalf of the class, not
an instance of the class. This means that a static method may not access any instance variables (that is
non-static fields) of the class, nor may they call other non-static methods. Since a static method
executes on behalf of the class, it has no access to the fields of any particular instance of that class. That
explains the restriction against static methods accessing non-static data members. Since a non-
static method may access instance variables of an object upon which it is called, a static method may
not call a non-static method and thus indirectly have access to instance variables. The restriction goes
only one way—any class method, static or non-static, may access a static data member or call a
static method.

15.8 Classes vs. structs

All members of a C++ class are private by default. The class

class Point
{
public:

©2013 Richard L. Halterman Draft date: January 27, 2013

15.9. SUMMARY 434

double x;
double y;
};

must use the public label so clients can access a Point object’s fields. The struct keyword is exactly
like the class keyword, except that the default access to members is public:
struct Point
{
double x; // These fields now are public
double y;
};

The C language supports the struct feature, but not the class keyword. C structs do not support
methods, constructors, and destructors. In C++ (unlike in C), a struct can contain methods, constructors,
and destructors. By default, members in structs are public, but the private and public labels can
be applied as in a class to fine tune client access.
Despite their similarities, C++ programmers favor classes over structs for programmer-defined
types with methods. The struct construct is useful for declaring simple composite data types that are
meant to be treated like primitive types. Programmers can manipulate directly integers, for example. Inte-
gers do not have methods or any hidden parts. Likewise, a geometric point object consists of two coordinates
that can assume any valid floating-point values. It makes sense to allow client code to manipulate directly
the coordinates, rather than forcing clients to use methods like set_x and set_y. On the other hand, it is
unwise to allow clients to modify directly the denominator of a Rational object, since a fraction with a
zero denominator is undefined.

In C++, by default everything in an object defined by a struct is accessible

to clients that use that object. In contrast, clients have no default access to the
internals of an object that is an instance of a class. The default member access
for struct instances is public, and the default member access for class
instances is private.

The struct feature is, in some sense, redundant. It is a carryover from the C programming language.
By retaining the struct keyword, however, C++ programs can use C libraries that use C structs. Any
C++ program that expects to use a C library using a struct must restrict its struct definitions to the
limited form supported by C. Such struct definitions may not contain non-public members, methods,
constructors, etc.

15.9 Summary

1. While the dot (.) operator is used to access a member of an object, the arrow (->) operator is used
to access a member of an object via a pointer.

2. A method declared const may not modify the value of an instance variable within an object. The
compiler enforces this restriction.

3. The class interface includes instance variables declarations and method prototypes. The method
implementations can appear in a separate file.

©2013 Richard L. Halterman Draft date: January 27, 2013

15.10. EXERCISES 435

4. Method implementations can appear in class declarations so that a class interface and implementation
is combined in one place.
5. A non-const method may not be called with a const object. Any public method can be called
with a non-const object.

6. Operators may be overloaded for programmer-defined types. Overloaded operators may be global
functions and methods.
7. A programmer-defined operator obeys the same precedence and associativity rules as the built-in
operator it overloads.

8. Unlike an instance variable, a static field is shared by all objects of their class.
9. Unlike instance variables, a static field must be declared outside of the class declaration.
10. A static method executes on behalf of the class, not an instance of the class.
11. A static method may not access any instance variables of the class, nor may they call other non-
static methods.
12. Any class method, static or non-static, may access a static data member or call a static
method.

15.10 Exercises

1. Exercise

©2013 Richard L. Halterman Draft date: January 27, 2013

15.10. EXERCISES 436

©2013 Richard L. Halterman Draft date: January 27, 2013

 437

Chapter 16

Building some Useful Classes

This chapter uses the concepts from the past few chapters to build some complete, practical classes.

16.1 A Better Rational Number Class

Listing 16.1 (rational.cpp) enhances the SimpleRational class (Listing 14.5 (simplerational.cpp)) pro-
viding a more complete type.

Listing 16.1: rational.cpp

1 #include <iostream>
2
3 using namespace std;
4
5 class Rational
6 {
7 int numerator;
8 int denominator;
9
10 // Compute the greatest common divisor (GCD) of two integers
11 static int gcd(int m, int n) {
12 if ( n == 0 ) {
13 return m;
14 } else {
15 return gcd(n, m % n);
16 }
17 }
18 // Compute the least common multiple (LCM) of two integers
19 static int lcm(int m, int n) {
20 return m * n /gcd(m, n);
21 }
22
23 public:
24 Rational(int n, int d): numerator(n), denominator(d)
25 {
26 if ( d == 0 )

©2013 Richard L. Halterman Draft date: January 27, 2013

 16.1. A BETTER RATIONAL NUMBER CLASS 438

27 { // Disallow an undefined fraction

28 cout << "*****Warning---Illegal Rational" << endl;
29 numerator = 0; // Make up a reasonable default fraction
30 denominator = 1;
31 }
32 }
33
34 // Default fraction is 0/1
35 Rational(): numerator(0), denominator(1) {}
36
37 int get_numerator() const
38 {
39 return numerator;
40 }
41
42 int get_denominator() const
43 {
44 return denominator;
45 }
46
47 Rational reduce() const
48 {
49 // Find the factor that numerator and denominator have in common . . .
50 int factor = gcd(numerator, denominator);
51 // . . . then divide it out in the new fraction
52 return Rational(numerator/factor, denominator/factor);
53 }
54
55 // Equal fractions have identical numerators and denominators
56 bool operator==(const Rational& fract) const
57 {
58 // First, find the reduced form of this fraction and the parameter . . .
59 Rational f1 = reduce(),
60 f2 = fract.reduce();
61 // . . . then see if their components match.
62 return (f1.numerator == f2.numerator)
63 && (f1.denominator == f2.denominator);
64 }
65
66 // Compute the sum of fract and the current rational number
67 Rational operator+(const Rational& fract) const
68 {
69 // Find common denominator
70 int commonDenominator = lcm(denominator, fract.denominator);
71 // Add the adjusted numerators
72 int newNumerator = numerator * commonDenominator/denominator
73 + fract.numerator * commonDenominator/fract.denominator;
74 return Rational(newNumerator, commonDenominator);
75 }
76
77 // Compute the product of fract and the current rational number
78 Rational operator*(const Rational& fract) const
79 {
80 return Rational(numerator * fract.numerator,
81 denominator * fract.denominator).reduce();

©2013 Richard L. Halterman Draft date: January 27, 2013

 16.1. A BETTER RATIONAL NUMBER CLASS 439

82 }
83 };
84
85 // Allow a Rational object to be displayed in a nice
86 // human-readable form.
87 ostream& operator<<(ostream& os, const Rational& r)
88 {
89 os << r.get_numerator() << "/" << r.get_denominator();
90 return os;
91 }
92
93 int main()
94 {
95 Rational f1(1, 2), f2(1, 3);
96 cout << f1 << " + " << f2 << " = " << (f1 + f2) << endl;
97 cout << f1 << " * " << f2 << " = " << (f1 * f2) << endl;
98 }

Listing 16.1 (rational.cpp) produces

1/2 + 1/3 = 5/6

1/2 * 1/3 = 1/6

This rational number type has some notable features:

• Constructors. The overloaded constructors permit convenient initialization and make it impossible
to create an undefined fraction.

• Private static methods. Rational provides two private static methods: gcd and lcm.
The algorithm for the recursive gcd (greatest common divisor) method was introduced in Sec-
tion 10.4. The lcm (least common multiple) method is derived from the mathematical relationship:

gcd(m, n) × lcm(m, n) = m × n

These two methods are declared private because they are not meant to be used directly by client
code. Greatest common divisor and least common multiple are concepts from number theory of which
Rational clients have no direct need. Client code expects functionality typical of rational num-
bers, such as addition and reduction; these two private methods are used by other, public, methods
that provide functionality more closely related to rational numbers. These two private methods are
static methods because they do not use instance variables. An object is not required to compute
the greatest common divisor of two integers. It is legal for gcd and lcm to be instance methods, but
instance methods should be used only where necessary, since they have the power to alter the state
of an object. Faulty coding that accidentally modifies an instance variable can be difficult to track
down. If a class method is used, however, the compiler can spot any attempt to modify an instance
variable immediately.

• Public instance methods. None of the instance methods (operator==, reduce, operator+,
and operator*) modify the state of the object upon which they are invoked. Thus, the Rational
class still produces immutable objects. The methods operator+, operator*, and reduce use
the private helper methods to accomplish their respective tasks.

©2013 Richard L. Halterman Draft date: January 27, 2013

16.2. STOPWATCH 440

• Global << operator. operator<< is a global function that allows a Rational object to be sent
to the cout object to be displayed as conveniently as a built-in type.

16.2 Stopwatch

The linear search vs. binary search comparison program (Listing 12.5 (searchcompare.cpp)) accessed the
system clock in order to time the execution of a section of code. The program used the clock function
from the standard C library and an additional variable to compute the elapsed time. The following skeleton
code fragment:
clock_t seconds = clock(); // Record starting time

/*
* Do something here that you wish to time
*/

clock_t other = clock(); // Record ending time

cout << static_cast<double>(other - seconds)/CLOCKS_PER_SEC
<< " seconds" << endl;

certainly works and can be adapted to any program, but it has several drawbacks:

• A programmer must take care to implement the timing code correctly for each section of code to be
timed. This process is error prone:

– clock_t is a specialized type that is used infrequently. It is not obvious from its name that
clock_t is equivalent to an unsigned integer, so a programmer may need to consult a library
reference to ensure its proper use.
– The programmer must specify the correct arithmetic:
(other - seconds)/CLOCKS_PER_SEC

– The type cast to double of the time difference is necessary but easily forgotten or applied
incorrectly. If a programmer incorrectly applies parentheses as so
static_cast<double>((other - seconds)/CLOCKS_PER_SEC)

the result will lose precision. Worse yet, the following parenthetical grouping
static_cast<double>(other) - seconds/CLOCKS_PER_SEC

will result in an incorrect value, since division has precedence over subtraction.

• The timing code is supplemental to the actual code that is being profiled, but it may not be imme-
diately obvious by looking at the complete code which statements are part of the timing code and
which statements are part of the code to be timed.

Section 10.5 offered a solution to the above shortcomings of using the raw types, constants, and func-
tions available from the C time library. Listing 10.13 (timermodule.cpp) hides the details of the C time li-
brary and provides a convenient functional interface to callers. Unfortunately, as mentioned in Section 10.5,
the functional approach has a serious limitation. The code in Listing 10.13 (timermodule.cpp) uses global

©2013 Richard L. Halterman Draft date: January 27, 2013

 16.2. STOPWATCH 441

variables to maintain the state of the timer. There is only one copy of each global variable. This means pro-
grammers using the timer functions cannot independently measure the elapsed time of overlapping events;
for example, you cannot measure how long it takes for a function to execute and simultaneously measure
how long a section of code within that function takes to execute.
A programmer could time multiple, simultaneous activites by using the raw C library clock function
directly, but then we are back to where we began: messy, potentially error-prone code.
Objects, of course, provide a solution. Consider the following client code that uses a stopwatch object
to keep track of the time:
Stopwatch timer; // Declare a stopwatch object

timer.start(); // Start timing

/*
* Do something here that you wish to time
*/

timer.stop(); // Stop the clock

cout << timer.elapsed() << " seconds" << endl;

This code using a Stopwatch object is as simple as the code that uses the timer functions from List-
ing 10.13 (timermodule.cpp). As an added benefit, a developer can think of a Stopwatch object as if it is
a real physical stopwatch object: push a button to start the clock (call the start method), push a button
to stop the clock (call the stop method), and then read the elapsed time (use the result of the elapsed
method). What do you do if you need to time two different things at once? You use two stopwatches, of
course, so a programmer would declare and use two Stopwatch objects. Since each object maintains its
own instance variables, each Stopwatch object can keep track of its own elapsed time independently of
all other active Stopwatch objects.
Programmers using a Stopwatch object in their code are much less likely to make a mistake because
the details that make it work are hidden and inaccessible. With objects we can wrap all the messy details
of the timing code into a convenient package. Given our experience designing our own types though C++
classes, we now are adequately equipped to implement such a Stopwatch class. 16.2 provides the header
file defining the structure and capabilities of our Stopwatch objects.

Listing 16.2: stopwatch.h

1 #ifndef _STOPWATCH_H_DEFINED_
2 #define _STOPWATCH_H_DEFINED_
3
4 #include <ctime>
5
6 class Stopwatch
7 {
8 clock_t start_time;
9 bool running;
10 double elapsed_time;
11 public:
12 Stopwatch();
13 void start(); // Start the timer
14 void stop(); // Stop the timer
15 void reset(); // Reset the timer
16 double elapsed() const; // Reveal the elapsed time

©2013 Richard L. Halterman Draft date: January 27, 2013

 16.2. STOPWATCH 442

17 bool is_running() const; // Is the stopwatch currently running?

18 };
19
20 #endif

From this class declaration we see that when clients create a Stopwatch object a constructor is available
to take care of any initialization details. Four methods are available to clients: start, stop, reset, and
elapsed. The reset method is included to set the clock back to zero to begin a new timing. Note that
the “messy” detail of the clock_t variable is private and, therefore, clients cannot see or affect its value
within a Stopwatch object.
This Stopwatch class (Listing 16.2 (stopwatch.h)) addresses the weaknesses of the non-object-oriented
approach noted above:

• The timing code can be implemented in methods of the Stopwatch class. Once the methods are
correct, a programmer can use Stopwatch objects for timing the execution of sections of code
without worrying about the details of how the timing is actually done. Client code cannot introduce
errors in the timing code if the timing code is hidden within the Stopwatch class.

• The details of the timing code no longer intertwine with the code to be timed, since the timing code
is located in the Stopwatch class. This makes it easier for programmers to maintain the code they
are timing.

• The Stopwatch class provides a convenient interface for the programmer that replaces the lower-
level details of calling system time functions.

Listing 16.3 (stopwatch.cpp) provides the Stopwatch implementation.

Listing 16.3: stopwatch.cpp

1 #include <iostream>
2 #include "Stopwatch.h"
3
4 using namespace std;
5
6 // Creates a Stopwatch object
7 // A newly minted object is not running and is in a "reset" state
8 Stopwatch::Stopwatch(): start_time(0), running(false), elapsed_time(0.0) {}
9
10 // Starts the stopwatch to begin measuring elapsed time.
11 // Starting a stopwatch that already is running has no effect.
12 void Stopwatch::start()
13 {
14 if ( !running )
15 {
16 running = true; // Set the clock running
17 start_time = clock(); // Record start time
18 }
19 }
20
21 // Stops the stopwatch. The stopwatch will retain the
22 // current elapsed time, but it will not measure any time
23 // while it is stopped.
24 // If the stopwatch is already stopped, the method has

©2013 Richard L. Halterman Draft date: January 27, 2013

 16.2. STOPWATCH 443

25 // no effect.
26 void Stopwatch::stop()
27 {
28 if ( running )
29 {
30 clock_t stop_time = clock(); // Record stop time
31 running = false;
32 // Accumulate elapsed time since start
33 elapsed_time += static_cast<double>((stop_time - start_time))
34 /CLOCKS_PER_SEC;
35 }
36 }
37
38 // Reports the cummulative time in seconds since the
39 // stopwatch was last reset.
40 // This method does not affect the state of the stopwatch.
41 double Stopwatch::elapsed() const
42 {
43 if ( running ) // Compute time since last reset
44 {
45 clock_t current_time = clock(); // Record current time
46 // Add time from previous elapsed to the current elapsed
47 // since the latest call to the start method.
48 return elapsed_time
49 + static_cast<double>((current_time - start_time))
50 /CLOCKS_PER_SEC;
51 }
52 else // Timer stopped; elapsed already computed in the stop method
53 return elapsed_time;
54 }
55
56 // Returns the stopwatch's status (running or not) to the client.
57 // This method does not affect the state of the stopwatch.
58 bool Stopwatch::is_running() const
59 {
60 return running;
61 }
62
63 // Resets the stopwatch so a subsequent start begins recording
64 // a new time. Stops the stopwatch if it currently is running.
65 void Stopwatch::reset()
66 {
67 running = false;
68 elapsed_time = 0.0;
69 }

Note that our design allows a clients to see the running time of a Stopwatch object without needing to
stop it. An alternate design might print an error message and perhaps exit the program’s execution if a client
attempts to see the elapsed time of a running stopwatch.
Some aspects of the Stopwatch class are notable:

• Stopwatch objects use three instance variables:

– The start_time instance variable records the time when the client last called the start

©2013 Richard L. Halterman Draft date: January 27, 2013

 16.2. STOPWATCH 444

method.
– The elapsed_time instance variable keeps track of the time since the latest call to the
reset method.
– running: This Boolean variable indicates whether or not the clock is running.

• The constructor sets the initial values of the instance variables start_time, elapsed_time,
and running. Actually, these initializations really are not necessary, since zero is automatically
assigned to numeric instance variables and false is automatically assigned to Boolean instance
variables (see Section 14.2).

• The start method notes the system time after the assignment to running. If these two statements
were reversed, the elapsed time would include the time to do the assignment to running. The
elapsed time should as closely as possible just include the statements in the client code between the
start and stop method calls.
Notice that start_time is not assigned if it currently is not zero. This allows a client to stop the
stopwatch and restart it later without losing an earlier segment of time.

• In the stop method, the system time is noted before the assignment to running so the elapsed
time does not include the assignment to running. This provides a more accurate accounting of the
client code execution time.
The stop method computes the accumulated elapsed time.

• The elapsed method either returns the elapsed time computed by the stop method or computes
the current running time without altering the elapsed_time variable. Clients should avoid calling
elapsed when a Stopwatch object is running since doing so would interfere with the accurate
timing of client code execution.

Compare the main function of Listing 12.5 (searchcompare.cpp) to that of Listing 16.4 (bettersearchcompare.cpp):

Listing 16.4: bettersearchcompare.cpp

1 #include <iostream>
2 #include <iomanip>
3 #include <ctime>
4 #include <vector>
5 #include "Stopwatch.h"
6
7 using namespace std;
8
9 /*
10 * binary_search(v, seek)
11 * Returns the index of element seek in vector v;
12 * returns -1 if seek is not an element of v
13 * v is the vector to search; v's contents must be
14 * sorted in ascending order.
15 * seek is the element to find
16 */
17 int binary_search(const vector<int>& v, int seek)
18 {
19 int first = 0, // Initially the first element in vector
20 last = v.size() - 1, // Initially the last element in vector
21 mid; // The middle of the vector
22 while ( first <= last )

©2013 Richard L. Halterman Draft date: January 27, 2013

 16.2. STOPWATCH 445

23 {
24 mid = first + (last - first + 1)/2;
25 if ( v[mid] == seek )
26 return mid; // Found it
27 else if ( v[mid] > seek )
28 last = mid - 1; // continue with 1st half
29 else // v[mid] < seek
30 first = mid + 1; // continue with 2nd half
31 }
32 return -1; // Not there
33 }
34
35 // This version requires vector v to be sorted in
36 // ascending order.
37 /*
38 * linear_search(v, seek)
39 * Returns the index of element seek in vector v;
40 * returns -1 if seek is not an element of a
41 * v is the vector to search; v's contents must be
42 * sorted in ascending order.
43 * seek is the element to find
44 */
45 int linear_search(const vector<int>& v, int seek)
46 {
47 size_t n = v.size();
48 for ( size_t i = 0; i < n && v[i] <= seek; i++ )
49 if ( v[i] == seek )
50 return i; // Return position immediately
51 return -1; // Element not found
52 }
53
54 int main()
55 {
56 const size_t SIZE = 30000;
57 vector<int> list(SIZE);
58
59 Stopwatch timer;
60
61 // Ensure the elements are ordered low to high
62 for ( size_t i = 0; i < SIZE; i++ )
63 list[i] = i;
64 // Search for all the elements in list using linear search
65 timer.start();
66 for ( size_t i = 0; i < SIZE; i++ )
67 linear_search(list, i);
68 // Print the elapsed time
69 timer.stop();
70 cout << "Linear elapsed: " << timer.elapsed() << " seconds" << endl;
71 // Prepare for a new timing
72 timer.reset();
73 // Search for all the elements in list using binary search
74 timer.start();
75 for ( size_t i = 0; i < SIZE; i++ )
76 binary_search(list, i);
77 // Print the elapsed time

©2013 Richard L. Halterman Draft date: January 27, 2013

 16.3. SORTING WITH LOGGING 446

78 timer.stop();
79 cout << "Binary elapsed: " << timer.elapsed() << " seconds" << endl;
80 }

This new, object-oriented version is simpler and more readable.

The design of the Stopwatch class allows clients to create multiple Stopwatch instances, and each
instance will keep track of its own time. In practice when profiling executing programs, such generality
usually is unnecessary. Rarely do developers need to time overlapping code, so one timer object per program
usually is enough. Multiple sections of code can be checked with the same Stopwatch object; simply
start it, stop it, check the time, and then reset it and start it again when another section of code is to be
timed.

16.3 Sorting with Logging

Section 12.2 shows how to use function pointers to customize the ordering that selection sort performs on
an array of integers. The selection_sort function in Listing 12.2 (flexiblesort.cpp) accepts a function
pointer parameter in addition to the array and the array’s size. The function pointer points to a function
that accepts two integer parameters and returns true or false. The function is supposed to use some kind of
ordering rule to determine if its first integer parameter precedes its second integer parameter.
Suppose we wish to analyze the number of comparisons and the number of swaps the sort function
performs on a given array with a particular ordering strategy. One way to do this is have the sort function
itself keep track of the number of times it calls the comparison function and swap function and return this
information when it finishes. To do so we would have to define an object to hold the two pieces of data
(comparisons and swaps) since a function can return only one value, not two. Also if we do this, we must
significantly alter the code of the sort algorithm itself. We would prefer to keep the sort algorithm focused
on its task of sorting and remain uncluttered from this additional logging code.
If instead of passing a function pointer to our sort function we pass a specially crafted object, we can
design our object to do whatever we want. Specifically, we can design our special object to perform the
necessary comparisons and keep track of how many comparisons it performs. We could let the object do
the swap, and it could log the swaps it performs.
Listing 16.5 (loggingflexiblesort.cpp) is a variation of Listing 12.2 (flexiblesort.cpp) that uses a compar-
ison object instead of a comparison function. Also, for variety, Listing 16.5 (loggingflexiblesort.cpp) sorts a
vector instead of a raw array.

Listing 16.5: loggingflexiblesort.cpp

1 #include <iostream>
2 #include <vector>
3
4 using namespace std;
5
6 /*
7 * Comparer objects manage the comparisons and element
8 * interchanges on the selection sort function below.
9 */
10 class Comparer
11 {
12 // Keeps track of the number of comparisons
13 // performed

©2013 Richard L. Halterman Draft date: January 27, 2013

 16.3. SORTING WITH LOGGING 447

14 int compare_count;
15 // Keeps track of the number of swaps performed
16 int swap_count;
17 // Function pointer directed to the function to
18 // perform the comparison
19 bool (*comp)(int, int);
20 public:
21 // The client must initialize a Comparer object with a
22 // suitable comparison function.
23 Comparer(bool (*f)(int, int)):
24 compare_count(0), swap_count(0), comp(f) {}
25
26 // Resets the counters to make ready for a new sort
27 void reset()
28 {
29 compare_count = swap_count = 0;
30 }
31
32 // Method that performs the comparison. It delegates
33 // the actual work to the function pointed to by comp.
34 // This method logs each invocation.
35 bool compare(int m, int n)
36 {
37 compare_count++;
38 return comp(m, n);
39 }
40
41 // Method that performs the swap.
42 // Interchange the values of
43 // its parameters a and b which are
44 // passed by reference.
45 // This method logs each invocation.
46 void swap(int& m, int& n)
47 {
48 swap_count++;
49 int temp = m;
50 m = n;
51 n = temp;
52 }
53
54 // Returns the number of comparisons this object has
55 // performed since it was created.
56 int comparisons() const
57 {
58 return compare_count;
59 }
60
61 // Returns the number of swaps this object has
62 // performed since it was created.
63 int swaps() const
64 {
65 return swap_count;
66 }
67 };
68

©2013 Richard L. Halterman Draft date: January 27, 2013

 16.3. SORTING WITH LOGGING 448

69 /*
70 * selection_sort(a, compare)
71 * Arranges the elements of vector a in an order determined
72 * by the compare object.
73 * a is a vector of ints.
74 * compare is a function that compares the ordering of
75 * two integers.
76 * The contents of a are physically rearranged.
77 */
78 void selection_sort(vector<int>& a, Comparer& compare)
79 {
80 int n = a.size();
81 for ( int i = 0; i < n - 1; i++ )
82 {
83 // Note: i,small, and j represent positions within a
84 // a[i], a[small], and a[j] represents the elements at
85 // those positions.
86 // small is the position of the smallest value we've seen
87 // so far; we use it to find the smallest value less
88 // than a[i]
89 int small = i;
90 // See if a smaller value can be found later in the array
91 for ( int j = i + 1; j < n; j++ )
92 if ( compare.compare(a[j], a[small]) )
93 small = j; // Found a smaller value
94 // Swap a[i] and a[small], if a smaller value was found
95 if ( i != small )
96 compare.swap(a[i], a[small]);
97 }
98 }
99
100 /*
101 * print
102 * Prints the contents of an integer vector
103 * a is the vector to print.
104 * a is not modified.
105 */
106 void print(const vector<int>& a)
107 {
108 int n = a.size();
109 cout << '{';
110 if ( n > 0 )
111 {
112 cout << a[0]; // Print the first element
113 for ( int i = 1; i < n; i++ )
114 cout << ',' << a[i]; // Print the rest
115 }
116 cout << '}';
117 }
118
119 /*
120 * less_than(a, b)
121 * Returns true if a < b; otherwise, returns
122 * false.
123 */

©2013 Richard L. Halterman Draft date: January 27, 2013

 16.3. SORTING WITH LOGGING 449

124 bool less_than(int a, int b)

125 {
126 return a < b;
127 }
128
129 /*
130 * greater_than(a, b)
131 * Returns true if a > b; otherwise, returns
132 * false.
133 */
134 bool greater_than(int a, int b)
135 {
136 return a > b;
137 }
138
139 int main()
140 {
141 // Make a vector of integers from an array
142 int a[] = { 23, -3, 4, 215, 0, -3, 2, 23, 100, 88, -10 };
143 int len = (sizeof a)/(sizeof a[0]);
144 vector<int> original(a, a + len);
145
146 // Make a working copy of the original vector
147 vector<int> working = original;
148 cout << "Before: ";
149 print(working);
150 cout << endl;
151 Comparer lt(less_than), gt(greater_than);
152 selection_sort(working, lt);
153 cout << "Ascending: ";
154 print(working);
155 cout << " (" << lt.comparisons() << " comparisons, "
156 << lt.swaps() << " swaps)" << endl;
157 cout << "----------------------------" << endl;
158 // Make another copy of the original vector
159 working = original;
160 cout << "Before: ";
161 print(working);
162 cout << endl;
163 selection_sort(working, gt);
164 cout << "Descending: ";
165 print(working);
166 cout << " (" << gt.comparisons() << " comparisons, "
167 << gt.swaps() << " swaps)" << endl;
168 cout << "----------------------------" << endl;
169 // Sort a sorted vector
170 cout << "Before: ";
171 print(working);
172 cout << endl;
173 // Reset the greater than comparer so we start counting at
174 // zero
175 gt.reset();
176 selection_sort(working, gt);
177 cout << "Descending: ";
178 print(working);

©2013 Richard L. Halterman Draft date: January 27, 2013

 16.4. LINKED LISTS 450

179 cout << " (" << gt.comparisons() << " comparisons, "
180 << gt.swaps() << " swaps)" << endl;
181 }

Notice that a Comparison object wraps a comparison function pointer, contains a swap method, and
maintains two integer counters. The comparison object passed to the sort routine customizes the sort’s
behavior (via its function pointer) and keeps track of the number of comparisons and swaps it performs
(via its integer counters). As in Listing 12.2 (flexiblesort.cpp), the basic structure of the sorting algorithm
remains the same regardless of the ordering determined by the comparison object.
The output of Listing 16.5 (loggingflexiblesort.cpp) is

Before: {23,-3,4,215,0,-3,2,23,100,88,-10}
Ascending: {-10,-3,-3,0,2,4,23,23,88,100,215} (55 comparisons, 7 swaps)
----------------------------
Before: {23,-3,4,215,0,-3,2,23,100,88,-10}
Descending: {215,100,88,23,23,4,2,0,-3,-3,-10} (55 comparisons, 5 swaps)
----------------------------
Before: {215,100,88,23,23,4,2,0,-3,-3,-10}
Descending: {215,100,88,23,23,4,2,0,-3,-3,-10} (55 comparisons, 0 swaps)

We see from the results that the number of comparisons is dictated by the algorithm itself, but the number of
element swaps depends on the ordering of the elements and the nature of the comparison. Sorting an already
sorted array with selection sort does not reduce the number of comparisons the function must perform, but,
as we can see, it requires no swaps.

16.4 Linked Lists

An object in C++ can hold just about any type of data, but there are some limitations. Consider the following
struct definition:
struct Node
{
int data;
Node next; // Error, illegal self reference
};

(Here we use a struct instead of a class since we will consider a Node object a primitive data type
that requires no special protection from clients.) How much space should the compiler set aside for a Node
object? A Node contains an integer and a Node, but this contained Node field itself would contain an
integer and a Node, and the recursive containment would go on forever. Such a structure understandably is
illegal in C++, and the compiler will issue an error. You are not allowed to have a class or struct field
of the same type within the class or struct being defined.
Another object definition looks similar, but it is a legal structure:
struct Node
{
int data;

©2013 Richard L. Halterman Draft date: January 27, 2013

 16.4. LINKED LISTS 451

Node *next; // Self reference via pointer is legal

};

The reason this second version is legal is because the compiler now can compute the size of a Node object.
A pointer is simply a memory address under the hood, so all pointer variables are the same size regardless
of their declared type. The pointer solves the infinitely recursive containment problem.
This ability of a object to refer to an object like itself is not merely an interesting curiosity; it has
practical applications. Suppose we wish to implement a sequence structure like a vector. We can use the
self-referential structure defined above to build a list of Node objects linked together via pointers. The
following declaration
Node *n1, *n2, *n2, *n4;

specifies four node pointers that will reference elements in our list. We can build a list containing the
elements 23, 10, 0, and 3 as follows
// Pointers to node objects
Node *n1, *n2, *n2, *n4;
// Create the list nodes
n1 = new Node;
n2 = new Node;
n3 = new Node;
n4 = new Node;
// Assign the data fields
n1->data = 23;
n2->data = 10;
n3->data = 0;
n4->data = 3;
// Link the nodes together
n1->next = n2;
n2->next = n3;
n3->next = n4;
n4->next = nullptr;

The last assignment of the null pointer (nullptr) indicates that node n4 has no other nodes following it.
This sequence of code produces the null-terminated linked list illustrated in Figure 16.1 In Figure 16.1 the
× symbol within the next pointer field represents the null pointer.
The hard-coded list in Figure 16.1 does not exploit the dynamic nature of linked lists. Armed with our
knowledge of C++ classes, encapsulation, and methods, we can build a client-friendly, dynamic linked list
type.
The code found in Listing 16.6 (intlist.h), Listing 16.7 (intlist.cpp), and Listing 16.8 (listmain.cpp)
demonstrates the power of linked lists. Listing 16.6 (intlist.h) is the header file for a simple integer linked
list class.

Listing 16.6: intlist.h

1
2 class IntList
3 {
4 /*
5 * An object that holds an element in a linked list.
6 * This type is local to the IntList class and inaccessible

©2013 Richard L. Halterman Draft date: January 27, 2013

 16.4. LINKED LISTS 452

Figure 16.1: A null-terminated linked list

7 * to outside this class.

8 */
9 struct Node
10 {
11 int data; // A data element of the list
12 Node *next; // The node that follows this one in the list
13 Node(int d); // Constructor
14 };
15
16 Node *head; // Points to the first item in the list
17
18 // Private helper methods
19
20 /*
21 * insert(p, n)
22 * Inserts n onto the back of the
23 * list pointed to by p.
24 * p points to a node in a linked list
25 * n is the element to insert.
26 * Returns a pointer to the list that
27 * contains n.
28 */
29 Node *insert(Node *p, int n);
30
31 /*
32 * print(p)
33 * Prints the contents of a linked list of integers.
34 * p points to a linked list of nodes.
35 */
36 void print(Node *p) const;
37
38 /*
39 * Returns the length of the linked list pointed to by p.
40 */
41 int length(Node *p) const;

©2013 Richard L. Halterman Draft date: January 27, 2013

 16.4. LINKED LISTS 453

42
43 /*
44 * dispose(p)
45 * Deallocate the memory held by the list pointed to by p.
46 */
47 void dispose(Node *p);
48
49 public:
50 /*
51 * The constructor makes an initially empty list
52 */
53 IntList();
54
55 /*
56 * The destructor frees up the memory held by the list
57 */
58 ∼IntList();
59
60 /*
61 * insert(n)
62 * Inserts n onto the back of the list.
63 * n is the element to insert.
64 */
65 void insert(int n);
66
67 /*
68 * print()
69 * Prints the contents of the linked list of integers.
70 * /
71 void print() const;
72
73 /*
74 * Returns the length of the linked list.
75 */
76 int length() const;
77
78 /*
79 * clear()
80 * Removes all the elements in the linked list.
81 */
82 void clear();
83 };

The Node class is declared within the IntList class. We say that Node is a nested class. Since the
declaration of Node appears in the private section of IntList, Node is a type known only to code within
the IntList class. The complete name of the Node type is IntList::Node.
Notice that the IntList class has several private methods in addition to its public methods. Code out-
side the class cannot execute these directly the private methods. These private methods are helper methods
that the public methods invoke to accomplish their tasks. We say that a public method delegates the work
to its private helper methods. Why is this delegation necessary here? The private methods use recursion
that requires a parameter of type IntList::Node which is unknown outside the IntList class. A
client is therefore unable to use the private methods directly, even it they were made public. The public
methods do not expose to the client any details about the class’s implementation; specifically they keep the

©2013 Richard L. Halterman Draft date: January 27, 2013

16.4. LINKED LISTS 454

IntList::Node type and the head instance variable hidden from clients.
Observe that the overloaded print and length methods (both private and public) of the IntList
class are declared const. Neither printing a list nor requesting its length should modify an IntList
object. Clients can, therefore, use the print and length methods with a constant IntList object. An
attempt to use insert or clear on a constant IntList object will yield a compiler error. The error
here makes sense, because insert definitely will modify a list object, and clear potentially will modify
a list object (clear will not modify an empty list).
A the insert method of the IntList class will use new to dynamically allocate Node objects when
elements are added to the collection. The IntList class destructor:
∼IntList();

will be responsible for deallocating the list nodes when an IntList object goes out of scope. The de-
structor simply will call the clear method to clean up the memory held by the list.
Why not just let the client call clear directly to free up the list without depending on the destructor?
A client may forget to call clear when finished with a list object. The resulting memory leak could cause
a long-running program to run out of memory. With the destructor, the client programmer needs not worry
about deallocating the list since the destructor does it automatically; for example, consider the following
code:
void f()
{
IntList my_list; // Constructor called here
// Add some numbers to the list
my_list.insert(22);
my_list.insert(5);
my_list.insert(-44);
// Print the list
my_list.print();
} // Destructor for my_list called when the function is finished

The variable my_list is local to function f. When function f finishes executing the variable my_list
goes out of scope. At this point the IntList destructor executes on behalf of my_list. The destruc-
tor ensures that my_list’s dynamically allocated memory is released. In the context of a pointer to
IntList, as in
void f2()
{
IntList *lptr; // Pointer, constructor NOT called yet
lptr = new IntList; // Constructor called here
// Add some numbers to the list
lptr->insert(22);
lptr->insert(5);
lptr->insert(-44);
// Print the list
lptr->print();
delete lptr; // Destructor called here
}

lptr is not an object; it is a pointer to an object. Declaring lptr does not create an object, and, therefore,
the constructor does not execute. The statement

©2013 Richard L. Halterman Draft date: January 27, 2013

 16.4. LINKED LISTS 455

lptr = new IntList;

does create an object, and so the IntList constructor executes on behalf of the object to which lptr
points. When lptr goes out of scope at the end of function f2’s execution, the stack variable lptr goes
away, but since it is a pointer, not an object, no destructor code executes. The client must explicitly free up
memory with delete:
delete lptr; // Destructor called here

unless the programmer intends for the function to return the pointer to the client (in which case the client is
responsible for calling delete when finished with the object.
Listing 16.7 (intlist.cpp) implements the methods declared in 16.6.

Listing 16.7: intlist.cpp

1 // intlist.cpp
2
3 #include "intlist.h"
4 #include <iostream>
5
6 // Private IntList operations
7
8 /*
9 * Node constructor
10 */
11 IntList::Node::Node(int n): data(n), next(nullptr) {}
12
13 /*
14 * insert(p, n)
15 * Inserts n onto the back of the
16 * list pointed to by p.
17 * p points to a node in a linked list
18 * n is the element to insert.
19 * Returns a pointer to the list that
20 * contains n.
21 */
22 IntList::Node *IntList::insert(IntList::Node *p, int n)
23 {
24 if ( p ) // Is p non-null?
25 p->next = insert(p->next, n);
26 else // p is null, make a new node
27 p = new IntList::Node(n);
28 return p;
29 }
30
31 /*
32 * Returns the length of the linked list pointed to by p.
33 */
34 int IntList::length(IntList::Node *p) const
35 {
36 if ( p )
37 return 1 + length(p->next); // 1 + length of rest of list
38 else
39 return 0; // Empty list has length zero
40 }

©2013 Richard L. Halterman Draft date: January 27, 2013

 16.4. LINKED LISTS 456

41
42 /*
43 * print(p)
44 * Prints the contents of a linked list of integers.
45 * p points to a linked list of nodes.
46 * /
47 void IntList::print(IntList::Node *p) const
48 {
49 while ( p ) // While p is not null
50 {
51 std::cout << p->data << " "; // Print current node's data
52 p = p->next; // Move to next node
53 }
54 std::cout << std::endl;
55 }
56
57 /*
58 * dispose(p)
59 * Deallocate the memory held by the list pointed to by p.
60 */
61 void IntList::dispose(IntList::Node *p)
62 {
63 if ( p )
64 {
65 dispose(p->next); // Free up the rest of the list
66 delete p; // Deallocate this node
67 }
68 }
69
70 // Public IntList operations
71
72 /*
73 * The constructor makes an initially empty list.
74 * The list is empty when head is null.
75 */
76 IntList::IntList(): head(nullptr) {}
77
78 /*
79 * The destructor frees up the memory held by the list
80 */
81 IntList::∼IntList()
82 {
83 clear(); // Deallocate space held by the list nodes
84 }
85
86 /*
87 * insert(n)
88 * Inserts n onto the back of the list.
89 * n is the element to insert.
90 */
91 void IntList::insert(int n)
92 {
93 head = insert(head, n); // Delegate work to private helper method
94 }
95

©2013 Richard L. Halterman Draft date: January 27, 2013

 16.4. LINKED LISTS 457

96 /*
97 * print()
98 * Prints the contents of the linked list of integers.
99 */
100 void IntList::print() const
101 {
102 print(head); // Delegate work to private helper method
103 }
104
105 /*
106 * Returns the length of the linked list.
107 */
108 int IntList::length() const
109 {
110 return length(head); // Delegate work to private helper method
111 }
112
113 /*
114 * clear()
115 * Removes all the elements in the linked list.
116 */
117 void IntList::clear()
118 {
119 dispose(head); // Deallocate space for all the nodes
120 head = nullptr; // Null head signifies list is empty
121 }

Since the code in Listing 16.7 (intlist.cpp) appears outside of the class declaration, any use of the Node type
requires its full name: IntList::Node. Note the :: use in the Node constructor:
IntList::Node::Node(int n): data(n), next(nullptr) {}

Three private methods in Listing 16.7 (intlist.cpp) (insert, length, and dispose) are recursive.
The print could be a recursive method, but for variety it uses iteration.
Listing 16.8 (listmain.cpp) provides some sample client code that uses a linked list.

Listing 16.8: listmain.cpp

1 // list_main.cpp
2
3 #include "intlist.h"
4 #include <iostream>
5
6 using namespace std;
7
8 int main()
9 {
10 bool done = false;
11 char command;
12 int value;
13 IntList list;
14
15 while ( !done )
16 {
17 cout << "I)nsert <item> P)rint L)ength E)rase Q)uit >>";

©2013 Richard L. Halterman Draft date: January 27, 2013

 16.4. LINKED LISTS 458

18 cin >> command;

19 switch ( command )
20 {
21 case 'I': // Insert a new element into the list
22 case 'i':
23 if ( cin >> value )
24 list.insert(value);
25 else
26 done = true;
27 break;
28 case 'P': // Print the contents of the list
29 case 'p':
30 list.print();
31 break;
32 case 'L': // Print the list's length
33 case 'l':
34 cout << "Number of elements: " << list.length() << endl;
35 break;
36 case 'E': // Erase the list
37 case 'e':
38 list.clear();
39 break;
40 case 'Q': // Exit the loop (and the program)
41 case 'q':
42 done = true;
43 break;
44 }
45 }
46 }

The client code in Listing 16.8 (listmain.cpp) allows a user to interactively add items to a list, print the list,
determine the list’s size, and clear the list. Observe that the client code does not use pointers at all. All
the pointer manipulations are hidden within the IntList class. The pointer data and code and the Node
struct itself is private within IntList, out of the reach of programmers who use these linked lists. Pointer
programming can be tricky, and it is easy to introduce subtle, hard to find bugs; thus, encapsulation once
again results in easier and more robust application development.
As shown in the IntList methods, a pointer variable by itself can be used as a condition within a
conditional statement or loop. A null pointer is interpreted as false, and any non-null pointer is true.
This means if p is a pointer, the statement
if ( p )
/* Do something */

is a shorthand for
if ( p != nullptr )
/* Do something */

and the statement

if ( !p )
/* Do something */

is a shorthand for

©2013 Richard L. Halterman Draft date: January 27, 2013

16.4. LINKED LISTS 459

if ( p == nullptr )
/* Do something */

Most C and C++ programmers use the shorter tests.

In order to better understand how the recursive methods work, think about the structure of a Node
object with this interpretation: A Node object holds a data item and a pointer to rest of the list that follows.
A pointer to a Node is either null or non-null. A null pointer represents the empty list. A non-null pointer
represents a non-empty list consisting of two parts: the first element in the list and the rest of the list. If the
next field of a Node object is null, the rest of the list is empty.
Armed with this view of lists we can now examine the behavior of the linked list methods in more detail:

• insert. To understand how the recursive private insert method works we must think recursively.
A list is either empty or non-empty. Inserting a new item onto a list is simple:

– If the list is empty, make a new list with one element in it. The single element in the resulting
list is the item you want to insert. Symbolically (not in C++), we can let ∅ stand for the empty
list and x → ∅ stand for the list containing just the element x. We therefore can represent the
insertion into an empty list as:
insert(∅, x) = x → ∅;
– If the list is not empty, merely insert the new item into the rest of the list. Notice the recursive
description. Symbolically, we can express this as
insert(x →rest of list, a) = x →insert(rest of list, a)

To see how insert works, we will build a list with three elements. First we will insert 1. The list
variable in main is initialized to null, so the first time main calls insert it passes nullptr as
the first parameter, and 1 as the second parameter. Let ∅ represent the empty list, so we are in effect
making the call

list = insert(∅, 1);

Within insert, since the parameter p is null, the condition

if ( p ) // Is p non-null?

is not met, so the statements within the else body are executed. Those statements reassign p to
point to a newly created node holding 1. The node’s next pointer is null, so p points to a list holding
just one element. We can represent this list as 1→ ∅. p is returned, and main assigns the result to
list, so list now points to a linked list containing the single element 1. list now is the list
1→ ∅. Observe that this first call to insert results in no recursive call.
We can symbolize this insertion of 1 into a empty list as

list = insert(∅, 1) = 1→ ∅

The second call to insert, which produces a list holding two elements, is more interesting. This
time we will insert 2 into our list. Again, list is passed as the first parameter and 2 is passed as the
second, but this time list is non-null. Within the method, the condition
if ( p ) // Is p non-null?

is satisfied, so the recursive call takes place:

p->next = insert(p->next, n);

©2013 Richard L. Halterman Draft date: January 27, 2013

16.4. LINKED LISTS 460

Note that p->next is assigned, not p itself. The p->next field represents the rest of the list, so we
now have a new problem to solve: insert 2 into the remainder of the list. At this point the rest of the
list is empty so the recursive call simply creates a new list containing 2 (that is, 2→ ∅) and returns it
to be assigned to p->next. p itself is not reassigned, so it is pointing to the same node as before,
but that node’s next field was reassigned to point 2→ ∅. p thus becomes 1→2→ ∅ and is returned
and assigned to list in main. list is now 1→2→ ∅.
We can symbolize this insertion of 2 into the list 1→ ∅ as

list = insert(1→ ∅, 2) = 1→ insert(∅, 2) = 1→2→ ∅

Finally, we will insert 3 into list. The process unfolds as follows:

list = insert(1→2→ ∅, 3)
= 1→ insert(2→ ∅, 3)
= 1→2→ insert(∅, 3)
= 1→2→3→ ∅

• length. Lists are either empty or non-empty. The length of the empty list is zero. The list of
a non-empty list is at least one, because a non-empty list contains at least one element. Writing a
recursive list length method is as simple as this:

– If the list is empty, its length is zero. Symbolically, we can write length(∅) = 0.
– If the list is non-empty, its length is one (counting its first element) plus the length of the rest
of the list.

This English description translates directly into C++ code.

Symbolically, we can write length(x→rest) = 1 + length(rest).
We can visualize the recursion process for determining the length of 2→10→7→∅ as:
length(2→10→7→∅) = 1 + length(10→7→∅)
= 1 + 1 + length(7→∅)
= 1 + 1 + 1 + length(∅)
= 1+1+1+0
= 3

• print. The print method is iterative, so the loop continues while p is non-null. Each time through
the loop p is updated to point to the next node in the list. Eventually p will point to a node with a
next pointer equal to nullptr. After printing the data in this last node, p becomes null, and the
loop terminates.

• dispose. The dispose method behaves similarly to other two recursive methods. Notice, how-
ever, that it makes the recursive call deleting the nodes in the rest of the list before it deletes the
current node. As we saw in 11.11 it is a logic error to attempt to access data via a pointer after that
data has been deallocated via delete. This means the code in dispose should not be written as
if ( p )
{ // Logic error! Do not do it this way!
delete p; // Deallocate this node first
dispose(p->next); // Then free up the rest of the list
}

because p->next is using p after it has been deleted.

©2013 Richard L. Halterman Draft date: January 27, 2013

 16.5. AUTOMATING TESTING 461

16.5 Automating Testing

We know that a clean compile does not imply that a program will work correctly. We can detect errors
in our code as we interact with the executing program. The process of exercising code to reveal errors or
demonstrate the lack thereof is called testing. The informal testing that we have done up to this point has
been adequate, but serious software development demands a more formal approach. We will see that good
testing requires the same skills and creativity as programming itself.
Until recently testing was often an afterthought. Testing was not seen to be as glamorous as designing
and coding. Poor testing led to buggy programs that frustrated users. Also, tests were written largely after
the program’s design and coding were complete. The problem with this approach is major design flaws may
not be revealed until late in the development cycle. Changes late in the development process are invariably
more expensive and difficult to deal with than changes earlier in the process.
Weaknesses in the standard approach to testing led to a new strategy: test-driven development. In test-
driven development the testing is automated, and the design and implementation of good tests is just as
important as the design and development of the actual program. In pure TDD, tests are developed before
any application code is written, and any application code produced is immediately subjected to testing.
16.9 defines the structure of a rudimentary test object.

Listing 16.9: tester.h

1 #define _TESTER_H_
2
3 #include <vector>
4 #include <string>
5
6 using std::vector;
7 using std::string;
8
9 class Tester
10 {
11 int error_count;
12 int total_count;
13 bool equals(double d1, double d2, double tolerance) const;
14 public:
15 Tester();
16 void check_equals(const string& msg, int expected, int actual);
17 void check_equals(const string& msg, double expected,
18 double actual, double tolerance);
19 void check_equals(const string& msg,
20 const vector<int>& expected,
21 const vector<int>& actual);
22 void report_results() const;
23 };
24
25 #endif

A simple test object keeps track of the number of tests performed and the number of failures. The client
uses the test object to check the results of a computation against a predicted result. Notice that the equals
method, which checks for the equality of two double-precision floating-point numbers is private, as it is
meant to be used internally by the other methods within the class. The equals method works the same
way as the equals function we examined in Listing 9.16 (floatequals.cpp).

©2013 Richard L. Halterman Draft date: January 27, 2013

 16.5. AUTOMATING TESTING 462

Listing 16.10 (tester.cpp) implements the Tester methods.

Listing 16.10: tester.cpp

1 #include <iostream>
2 #include <cmath>
3 #include "tester.h"
4
5 using namespace std;
6
7 Tester::Tester(): error_count(0), total_count(0)
8 {
9 cout << "+---------------------------------------" << endl;
10 cout << "| Testing " << endl;
11 cout << "+---------------------------------------" << endl;
12 }
13
14 // d1 and d2 are "equal" if their difference is less than
15 // a specified tolerance
16 bool Tester::equals(double d1, double d2, double tolerance) const
17 {
18 return d1 == d2 || abs(d1 - d2) < tolerance;
19 }
20
21 void Tester::check_equals(const string& msg, int expected, int actual)
22 {
23 cout << "[" << msg << "] ";
24 total_count++; // Count this test
25 if ( expected == actual )
26 cout << "OK" << endl;
27 else
28 {
29 error_count++; // Count this failed test
30 cout << "*** Failed! Expected: " << expected
31 << ", actual: " << actual << endl;
32 }
33 }
34
35 void Tester::check_equals(const string& msg, double expected,
36 double actual, double tolerance)
37 {
38 cout << "[" << msg << "] ";
39 total_count++; // Count this test
40 if ( equals(expected, actual, tolerance) )
41 cout << "OK" << endl;
42 else
43 {
44 error_count++; // Count this failed test
45 cout << "*** Failed! Expected: " << expected
46 << ", actual: " << actual << endl;
47 }
48 }
49
50 bool vector_equals(const vector<int>& a, const vector<int>& b)
51 {
52 if ( a.size() != b.size() )

©2013 Richard L. Halterman Draft date: January 27, 2013

 16.5. AUTOMATING TESTING 463

53 return false;
54 for ( size_t i = 0; i < a.size(); i++ )
55 if ( a[i] != b[i] )
56 return false;
57 return true;
58 }
59
60 void print_vector(const vector<int>& a)
61 {
62 cout << "{";
63 auto iter = a.begin(), end = a.end();
64 if ( iter != end) // Do we need to print anything?
65 {
66 cout << *iter; // Print first element
67 iter++; // Increment iterator
68 while ( iter != end ) // Continue to print any elements
69 { // that follow
70 cout << "," << *iter;
71 iter++;
72 }
73 }
74 cout << "}";
75 }
76
77 void Tester::check_equals(const string& msg, const vector<int>& expected,
78 const vector<int>& actual)
79 {
80 cout << "[" << msg << "] ";
81 total_count++; // Count this test
82 if ( vector_equals(expected, actual) )
83 cout << "OK" << endl;
84 else
85 {
86 error_count++; // Count this failed test
87 cout << "*** Failed! Expected: ";
88 print_vector(expected);
89 cout << " Actual: ";
90 print_vector(actual);
91 cout << endl;
92 }
93 }
94
95 void Tester::report_results() const
96 {
97 cout << "+--------------------------------------" << endl;
98 cout << "| " << total_count << " tests run, "
99 << total_count - error_count << " passed, "
100 << error_count << " failed" << endl;
101 cout << "+--------------------------------------" << endl;
102
103 }

Listing 16.11: testvectorstuff.cpp

1 #include <iostream>

©2013 Richard L. Halterman Draft date: January 27, 2013

 16.5. AUTOMATING TESTING 464

2 #include <vector>
3 #include "tester.h"
4
5 using namespace std;
6
7 // sort has a bug (it does not do anyting)
8 void sort(vector<int>& vec)
9 {
10 // Not yet implemented
11 }
12
13 // sum has a bug (misses first element)
14 int sum(const vector<int>& vec)
15 {
16 int total = 0;
17 for ( size_t i = 1; i < vec.size(); i++ )
18 total += vec[i];
19 return total;
20 }
21
22 // Handy utility function to make an integer vector of size 3
23 vector<int> make_vector(int n0, int n1, int n2)
24 {
25 vector<int> v;
26 v.push_back(n0);
27 v.push_back(n1);
28 v.push_back(n2);
29 return v;
30 }
31
32 int main()
33 {
34 Tester t; // Declare a test object
35 // Some test cases to test sort
36 vector<int> vec = make_vector(4, 2, 3);
37 sort(vec);
38 t.check_equals("Sort test #1", make_vector(2, 3, 4), vec);
39 vec = make_vector(2, 3, 4);
40 sort(vec);
41 t.check_equals("Sort test #2", make_vector(2, 3, 4), vec);
42 // Some test cases to test sum
43 t.check_equals("Sum test #1", sum(make_vector(0, 3, 4)), 7);
44 t.check_equals("Sum test #2", sum(make_vector(-3, 0, 5)), 2);
45 }

The program’s output is

©2013 Richard L. Halterman Draft date: January 27, 2013

16.6. SUMMARY 465

+---------------------------------------
| Testing
+---------------------------------------
[Sort test #1] *** Failed! Expected: {2,3,4} Actual: {4,2,3}
[Sort test #2] OK
[Sum test #1] OK
[Sum test #2] *** Failed! Expected: 5, actual: 2

Notice that the sort function has yet to be implemented, but we can test it anyway. The first test is
bound to fail. The second test checks to see if our sort function will not disturb an already sorted vector,
and we pass this test with no problem. This is an example of coincidental correctness.
In the sum function, the programmer was careless and used 1 as the beginning index for the vector.
Notice that the first test does not catch the error, since the element in the zeroth position (zero) does not
affect the outcome. A tester must be creative and devious to try and force the code under test to demonstrate
its errors.

16.6 Summary

• The object-oriented features of C++ allow developers to design a variety of useful classes to make
programming tasks easier.
• Sophisticated computation can be packaged within objects that are relatively easy for clients to use.

16.7 Exercises

1. Recall the private insert method from Listing 16.7 (intlist.cpp) and consider the following new
insert function which contains only a couple of minor modifications to the original version:
IntList::Node *IntList::insert(Node *p, int n)
{
// Check that we are not at the end of the list and that the current list
// element is less than the new item to add.
if ( p && p->data < n )
p->next = insert(p->next, n); // Insert into rest of list
else
{ // Insert new item at the front of the list
Node *temp = new Node;
temp->data = n;
temp->next = p;
p = temp;
}
return p;
}

(a) Without executing the code, can you predict how this new version behaves differently, if at all,
from the original version?

©2013 Richard L. Halterman Draft date: January 27, 2013

16.7. EXERCISES 466

(b) Replace the insert function in Listing 16.7 (intlist.cpp) with this new version to check your
prediction. Can you explain why this new insert function behaves the way it does?

©2013 Richard L. Halterman Draft date: January 27, 2013

 467

Chapter 17

Inheritance and Polymorphism

In Chapter 16 we saw how it is possible to design classes from which clients can produce objects that
exhibit somewhat sophisticated behavior. The Rational, Stopwatch, LinkedList, and Tester
classes each were each built from scratch. C++ provides a way to create a class from an existing class in
a process known as inheritance. Through this process the new class inherits all the characteristics of the
existing class, and the class developer can extend and customize the inherited functionality.

17.1 I/O Stream Inheritance

Recall from Section 13.4 that writing data to a file is almost as easy as printing it on the screen. Once an
ofstream object is set up, we can use the << operator in the same way we use it with the cout object:
ofstream fout;
int x = 10;
fout.open("myfile.dat");
if ( fout.good() ) // Make sure the file was opened properly
{
fout << "x = " << x << endl;
fout.close();
}
else
cout << "Unable to write to the file \"myfile.dat\"" << endl;

Section 15.6 showed how operator<< may be overloaded for a programmer-defined type. For a
vector of integers we can write the function
ostream& operator<<(ostream& os, const vector<int>& vec)
{
os << '{';
int n = vec.size();
if ( n > 0 ) // Is the vector non-empty?
{
os << vec[0]; // Send first element to the strean
for ( int i = 1; i < n; i++ )

©2013 Richard L. Halterman Draft date: January 27, 2013

 17.1. I/O STREAM INHERITANCE 468

os << ',' << vec[i]; // Send remaining elements

}
os << '}';
return os;
}

Can we easily adapt our << operator so that we can use it with ofstream objects as well?
The answer, perhaps surprisingly, is we do not have to adapt our operator<< function; it works as
is with ofstream objects. How can this be, since the cout object has type ostream which is not the
same class as ofstream?
C++ allows some automatic conversions among the built in numeric types; for example, an int is
widened to a double, and a double is truncated to an int. bools and ints are freely interconvertible.
These types are built into the language, so standard conversions apply. When it comes to programmer-
defined types, however, the compiler is unable automatically to generate code to convert from one type to
another. Most of the time it would make no sense to do so—imagine attempting to convert an object of type
Stopwatch (16.2) to an object of type Tester (16.9).
So, how is it that an object of type ofstream can be converted automatically to an object of type
ostream? You might think that it is because they are both part of the standard library, so the conversion
has been programmed into them. The fact is, no conversion takes place! Any object of type ofstream
object is automatically treated as if it were an ostream object, because the ostream and ofstream
classes are related in a special way. The ofstream class is derived from the ostream class. We say
that ostream is the base class and ofstream is the derived class. Sometimes the term superclass is
used for the base class, in which case the derived class is called the subclass. The base/derived class
terminology is used interchangeably with super/subclass, although the C++ community tends to prefer the
terms base/derived class to super/subclass. The process is known as inheritance, because the derived class
inherits all the characteristics of its base class. The terms derivation, subclassing and specialization are
sometimes used in place of the term inheritance.
As a consequence of the derivation process, an instance of the derived class can be treated as if it were
an instance of the base class. This allows us to write a function like
void process(const ostream& os)
{
// Details omitted ...
}

with the following perfectly legal client code fragment:

ofstream fout;
fout.open("myfile.dat");
if ( fout.good() )
{
process(fout); // OK, an ofstream object is an ostream object
process(cout); // OK, cout is exactly an ostream object
fout.close();
}

Listing 17.1: printstreams.h

1 #include <fstream>
2

©2013 Richard L. Halterman Draft date: January 27, 2013

 17.2. INHERITANCE MECHANICS 469

3 using namespace std;

4
5 void print(ostream& os, int n)
6 {
7 os.width(10); // Right justified in 10 spaces
8 os.fill('*'); // Fill character is *
9 os << n << endl;
10 }
11
12 int main()
13 {
14 // Pretty print to screen
15 print(cout, 35);
16
17 // Pretty print to text file
18 ofstream fout("temp.data");
19 if ( fout.good() )
20 {
21 print(fout, 36);
22 fout.close();
23 }
24 }

Observe that the print function expects a client to pass an ostream reference as the first parameter, but
the function readily accepts both ostream and ofstream object references.
This ability to allow a subclass object to be used in any context that expects a superclass object is known
as the Liskov Substitution Principle, after computer scientist Barbara Liskov.
The term is a has a special meaning in the context of inheritance. Suppose we have class B and class D
derived from B. B is the base class and D is the derived class. Since we can treat an instance of a derived
class as if it were an instance of its base class, if we declared a D object as
D d; // d is a D object

we can say d is a D (as it is declared), and we also can say d is a B.

17.2 Inheritance Mechanics

The developers of the ofstream class did not begin with a blank slate. The ostream class existed first,
and the developers specified the ofstream class in such a way so any ofstream object would be treated
as a specific kind of ostream. In this section we will examine a very simple example that illustrates the
mechanics of class inheritance in C++.
Suppose we have class B defined as shown here:
class B
{
// Details omitted
};

To derive a new class D from B we use the following syntax:

class D: public B

©2013 Richard L. Halterman Draft date: January 27, 2013

17.2. INHERITANCE MECHANICS 470

{
// Details omitted
};

Here B is the base class and D is the derived class. B is the pre-existing class, and D is the new class based
on B.
To see how inheritance works, consider classes B and D with additional detail:
class B
{
// Other details omitted
public:
void f();
};

void B::f()
{
std::cout << "In function 'f'" << std::endl;
}

class D: public B
{
// Other details omitted
public:
void g();
};

void D::g()
{
std::cout << "In function 'g'" << std::endl;
}

The client code

B myB;
D myD;
myB.f();
myD.f();
myD.g();

prints

In function ’f’
In function ’f’
In function ’g’

Even though the source code for class D does not explicitly show the definition of a method named f, it has
such a method that it inherits from class B.

©2013 Richard L. Halterman Draft date: January 27, 2013

17.2. INHERITANCE MECHANICS 471

Note that inheritiance works in one direction only. Class D inherits method f from class B, but class B
cannot inherit D’s g method. Given the definitions of classes B and D above, the following code is illegal:
B myB;
myB.g(); // Illegal, a B object is NOT a D object

If we omit the word public from class D’s definition, as in

class D: B
{
// Details omitted
};
all the public members of B inherited by D objects will be private by default; for
example, if base class B looks like the following:
class B
{
public:
void f();
};

void B::f()
{
std::cout << "In function 'f'" << std::endl;
}
the following code is not legal:
D myD;
myD.f(); // Illegal, method f now is private!
This means a client may not treat a D object exacty as if it were a B object. This
violates the Liskov Substitution Principle, and the is a relationship does not exist.
While this private inheritance is useful in rare situations, the majority of object-
oriented software design uses public inheritance. C++ is one of the few object-
oriented languages that supports private inheritance.

There is no limit to the number of classes a developer may derive from a single base class. In the
following code:
class D1: public B { /* Details omitted */ };
class D2: public B { /* Details omitted */ };
class D3: public B { /* Details omitted */ };

the classes D1, D2, and D3 are all derived from class B.
A developer may derive a class from more than one base class in a process known as multiple inheri-
tance. In the following code:
class B1 { /* Details omitted */ };
class B2 { /* Details omitted */ };

class D: public B1, public B2 { /* Details omitted */ };

©2013 Richard L. Halterman Draft date: January 27, 2013

 17.3. USES OF INHERITANCE 472

class D has two base classes, B1 an B2. If a particular function or method expects a B1 object as an
argument, the caller may pass a D object; similarly, any context that requires a B2 object will accept a D
object. In object-oriented design, multiple inheritance is not as common as single inheritance (one base
class).
The next section provides a simple example that shows how inheritance works.

17.3 Uses of Inheritance

Inheritance is a design tool that allows developers to take an existing class and produce a new class that
provides enhanced behavior or different behavior. The enhanced or new behavior does not come at the
expense of existing code; that is, when using inheritance programmers do not touch any source code in
the base class. Also, developers can leverage existing code (in the base class) without duplicating it in the
derived classes.
Listing 17.2 (fancytext.cpp) provides a complete program with simple classes that demonstrate how we
can use inheritance to enhance the behavior of unadorned text objects.

Listing 17.2: fancytext.cpp

1 #include <string>
2 #include <iostream>
3
4 using namespace std;
5
6 // Base class for all Text derived classes
7 class Text
8 {
9 string text;
10 public:
11 // Create a Text object from a client-supplied string
12 Text(const string& t): text(t) {}
13
14 // Allow clients to see the text field
15 virtual string get() const
16 {
17 return text;
18 }
19
20 // Concatenate another string onto the
21 // back of the existing text
22 virtual void append(const string& extra)
23 {
24 text += extra;
25 }
26 };
27
28 // Provides minimal decoration for the text
29 class FancyText: public Text
30 {
31 string left_bracket;
32 string right_bracket;
33 string connector;

©2013 Richard L. Halterman Draft date: January 27, 2013

 17.3. USES OF INHERITANCE 473

34 public:
35 // Client supplies the string to wrap plus some extra
36 // decorations
37 FancyText(const string& t, const string& left,
38 const string& right, const string& conn):
39 Text(t), left_bracket(left),
40 right_bracket(right), connector(conn) {}
41
42 // Allow clients to see the decorated text field
43 string get() const override
44 {
45 return left_bracket + Text::get() + right_bracket;
46 }
47
48 // Concatenate another string onto the
49 // back of the existing text, inserting the connector
50 // string
51 void append(const string& extra) override
52 {
53 Text::append(connector + extra);
54 }
55 };
56
57 // The text is always the word FIXED
58 class FixedText: public Text
59 {
60 public:
61 // Client does not provide a string argument; the
62 // wrapped text is always "FIXED"
63 FixedText(): Text("FIXED") {}
64
65 // Nothing may be appended to a FixedText object
66 void append(const string&) override
67 {
68 // Disallow concatenation
69 }
70 };
71
72 int main()
73 {
74 Text t1("plain");
75 FancyText t2("fancy", "<<", ">>", "::");
76 FixedText t3;
77 cout << t1.get() << endl;
78 cout << t2.get() << endl;
79 cout << t3.get() << endl;
80 cout << "-------------------------" << endl;
81 t1.append("A");
82 t2.append("A");
83 t3.append("A");
84 cout << t1.get() << endl;
85 cout << t2.get() << endl;
86 cout << t3.get() << endl;
87 cout << "-------------------------" << endl;
88 t1.append("B");

©2013 Richard L. Halterman Draft date: January 27, 2013

 17.3. USES OF INHERITANCE 474

89 t2.append("B");
90 t3.append("B");
91 cout << t1.get() << endl;
92 cout << t2.get() << endl;
93 cout << t3.get() << endl;
94 }

The output of Listing 17.2 (fancytext.cpp) is

plain
<<fancy>>
FIXED
-------------------------
plainA
<<fancy::A>>
FIXED
-------------------------
plainAB
<<fancy::A::B>>
FIXED

In Listing 17.2 (fancytext.cpp), Text serves as the base class for two other classes, FancyText and
FixedText. A Text object wraps a std::string object, and clients cannot get to the string object
itself because the string is private to the Text class. The clients may see the string via the get method, and
can modify the wrapped string only in a limited way via the append method. Listing 17.2 (fancytext.cpp)
contains two new keywords:

• The virtual keyword appears in front of the definitions of the get and append methods in the
Text class. The virtual specifier indicates that the designer of the Text class intends for derived
classes to be able to customize the behavior of the get and append methods.
• The override keyword appears at the end of the method definition headers for the get and
append methods in the FancyText class and the append method for the FixedText. This
means the exact behavior of these methods will be different in some way from their implementation
in the Text class. The FixedText class overrides only the append method; it inherits the get
method without alteration.

Consider the FixedText::get method:

class FancyText: public Text
{
// . . .
public:
// . . .
string get() const override
{
return left_bracket + Text::get() + right_bracket;
}
};

©2013 Richard L. Halterman Draft date: January 27, 2013

17.3. USES OF INHERITANCE 475

The FancyText class alters the way get works. We say the FancyText class overrides its inher-
ited get method. The keyword override emphasizes the fact that the code in the get method in
FancyText intends to do something differently from the code in Text’s get method. In this case
the FancyText::get method builds a string result by concatenating three other strings: the first string
is a front bracketing string, the second is the wrapped string, and the third string is a trailing bracketing
string. Notice that the second string is obtained with the expression
Text::get()

This calls the base class version of the get method. We say that the FancyText::get method delegates
some of its work to the base class version, Text::get. Notice that the statement
return left_bracket + text + right_bracket; // Illegal

would be illegal, because text is private to the base class. The member specifier private means inac-
cessible outside of the class, period. Derived classes have no special privilege in this regard. In order to
perform the delegation we must use Text::get() and not simply get(); the unqualified expression
get() is equivalent to this->get, which calls FancyText::get. This means invoking the unquali-
fied get within the definition of FancyText::get is a recursive call which is not what we want.
A method declared virtual in a base class is automatically virtual in its derived classes. You may repeat
the virtual specifier in derived classes, but the use of override makes this redundant.

The override keyword was added to the language in C++11. Prior to C++11
when a method in a derived class had the same signature as a virtual method in
its base class, the method implicitly overrode its base class version. The problem
was that a programmer could intend to override a method in the derived class but
get the signature wrong. The resulting method overloaded the original method
rather than overriding it. If a programmer uses the override specifier and uses
a signature that does not match the base class version, the compiler will report
an error. The override specifier provides a way for programmers to explicitly
communicate their intentions.
The override keyword is a context-sensitive keyword, meaning it is a keyword
only when appearing as it does here in the declaration of a method header. In other
contexts it behaves like an identifier.

The FixedText class does not change how the get method works for FixedText objects; thus, the
FixedText::get method does add any special decorations.
The FancyText::append method overrides the inherited append method by inserting a special
separator string in between the existing wrapped text string and the string to append. Like FancyText::get,
the FancyText::append method delegates the actual concatenation operation to its base class version
because code within the FancyText class cannot directly influence its text field.
The FixedText::append method does not allow the existing wrapped text string to be modified.
The constructor for Text requires a single std::string parameter. The constructor for FancyText
requires four string arguments:
class FancyText: public Text
{
// . . .

©2013 Richard L. Halterman Draft date: January 27, 2013

17.3. USES OF INHERITANCE 476

public:
FancyText(const string& t, const string& left,
const string& right, const string& conn):
Text(t), left_bracket(left),
right_bracket(right), connector(conn) {}
// . . .
};

We want to assign the constructor’s first parameter, t, to the inherited member text, but text is private in
the base class. This means the FancyText constructor cannot initialize it directly. Since the constructor
of its base class knows what to do with this parameter, the first expression in the constructor initialization
list (the part between the : and the {):
... Text(t) ...

explicitly calls the base class constructor, passing it t. This base class initialization expression must be
the first thing in the initialization list because C++ requires that the parts of an object inherited from its
base class be initialized before any new parts (added by the derived class) are initialized. The next three
expressions in the initialization list:
... left_bracket(left), right_bracket(right), connector(conn) ...

initialize the left_bracket, right_bracket, and connector fields as usual. The body of the
constructor is empty as no other initialization work is necessary.
The FixedText class is somewhat simpler than the FancyText class. It adds no fields and does
not override the get method. Its constructor accepts no parameters because clients are not permitted to
determine the contents of the wrapped string—the text field is always the word FIXED. It inherits the
text field and get method (as is) from the Text class.
Given the Text and FancyText classes as defined in Listing 17.2 (fancytext.cpp), the following client
code is legal:
Text t1("ABC");
FancyText t2("XYZ", "[", "]", ":");
cout << t1.get() << " " << t2.get() << endl;
t1 = t2;
cout << t1.get() << " " << t2.get() << endl;

Since t2’s declared type is FancyText, by virtue of inheritiance, t2 is a Text object as well. The above
code fragment prints

ABC <<XYZ>>
XYZ <<XYZ>>

Notice that the assignment

t1 = t2;

copied into object t1 only the fields that FancyText objects have in common with Text objects; that is,
the text field. Since t1 is a plain Text object, it does not have the left_bracket, right_bracket,

©2013 Richard L. Halterman Draft date: January 27, 2013

17.3. USES OF INHERITANCE 477

and connector fields capable of storing the additional data contained in a FancyText object. This pro-
cess of losing derived class data when assigning to a base class instance is known as object slicing. The
parts that will not fit into the base class instance are “sliced” off. Figure 17.1 illustrates object slicing. Note

Text t1("plain");
FancyText t2("Fancy", "<<", ">>", "::");

t1 t2
text plain text fancy

left_bracket << Before

right_bracket >>
Assignment
connector ::

t1 = t2;
t1 t2
text fancy text fancy

left_bracket << After

right_bracket >> Assignment
connector ::

Figure 17.1: Object slicing during assignment of a derived class instance to a base class variable

that the assignment in the other direction:

t2 = t1; // Illegal

is not possible, because even though any FancyText object is a Text object, we cannot say any Text
object is a FancyText object (it could be a FixedText object or just a Text object without any special
decoration). Figure 17.2 shows how such an attempted assignment would be meaningless because the base
class instance has missing information needed by the derived class instance. Failure to properly assign all
the fields in the derived class instance would produce an object that is not well defined, so C++ does not
allow the assignment.
FixedText class instances do not contain any additional data that plain Text instances do not have,
but the assignment rules remain the same:
Text t1("ABC");
FixedText t3;
cout << t1.get() << " " << t3.get() << endl;
t1 = t3;
cout << t1.get() << " " << t3.get() << endl;

©2013 Richard L. Halterman Draft date: January 27, 2013

17.3. USES OF INHERITANCE 478

Text t1("plain");
FancyText t2("Fancy", "<<", ">>", "::");

t1 t2
text plain text fancy

left_bracket << Before

right_bracket >> Assignment
connector ::

t2 = t1;
t1 t2
text plain text fancy

left_bracket ?? After
Assignment
right_bracket ??
connector ??

Figure 17.2: Assigning a base class instance to a derived class variable is not possible and is illegal in C++

is legal and produces

ABC FIXED
FIXED FIXED

but the statement

t3 = t1;

is illegal and will not compile.

It always is legal to assign a derived class instance to a variable of a base type.

This is because a derived class instance is a specific kind of base class instance.
In contrast, it never is legal to assign a base class instance to a variable of a derived
type. This is because the is a relationship is only one directional, from a derived
class to its base class.

©2013 Richard L. Halterman Draft date: January 27, 2013

17.4. POLYMORPHISM 479

The Text, FancyText, and FixedText classes form a small class hierarchy. The relationships in
this class hierarchy can be represented in a graphical form using the Unified Modeling Language (UML).
Figure 17.3 shows the UML diagram for our simple Text class hierarchy. A rectangle represents a class.

Text

FancyText FixedText

Figure 17.3: The Unified Modeling Language class hierarchy diagram

Text, the base class, appears at the top of the Figure 17.3. The two derived classes appear below Text.
The inheritance arrow points from the derived classes to the base class. The arrow represents the is a
relationship which flows upward. This visual layout and level of detail provided by the UML diagram more
succinctly communicates the relationships amongst the classes than does the C++ source code.
The UML is a large, complex graphical language that may be used to model many facets of the software
development process. More information about the UML is available at http://www.uml.org.

17.4 Polymorphism

Recall the Text class hierarchy from Listing 17.2 (fancytext.cpp) illustrated in Figure 17.3. Consider the
following client code fragment:

Text t1("ABC");
FancyText t2("XYZ", "[", "]", ":");
cout << t1.get() << " " << t2.get() << endl;

What is the nature of the machine code generated by the expression t1.get()? The compiler translates
this higher-level expression into the machine language instruction that causes the program’s execution to
jump to another part of the compiled code. Where does the program’s execution jump to? The variable
t1’s declared type is Text, the program’s execution jumps to the compiled Text::get method (also the
address of t1 is passed as the secret this parameter to the method so the compiled code can access t1’s
text field).
For the expression t2.get(), the program’s execution jumps to FancyText’s compiled get method,
passing the address of object t2 as the secret parameter.

©2013 Richard L. Halterman Draft date: January 27, 2013

17.4. POLYMORPHISM 480

When the compiler can determine which method to execute based on the declared type of an object,
the process is known as static binding or early binding. Static binding is used in all cases for non-virtual
methods and in the cases we have seen so far for virtual methods.
The situation is different if we use pointers to objects instead of the objects themselves. Consider the
following code:
Text t1("ABC");
FancyText t2("XYZ", "[", "]", ":");
cout << t1.get() << " " << t2.get() << endl;
Text *p1, *p2;
p1 = &t1;
p2 = &t2;

The variables p1 and p2 are declared to be pointers to Text objects. The variable t1 is a Text object, so
the assignment
p1 = &t1;

make perfect sense. The variable t2 has a declared type of FancyText, but a FancyText object is a
Text object, so the assignment
p2 = &t2;

is legal since p2 is indeed pointing to a special kind of Text object.

What does the following code fragment print?
Text t1("ABC");
FancyText t2("XYZ", "[", "]", ":");
cout << t1.get() << " " << t2.get() << endl;
Text *p1, *p2;
p1 = &t1;
p2 = &t2;
cout << p1->get() << " " << p2->get() << endl;

This code displays

ABC <<XYZ>>
ABC <<XYZ>>

Even though p2’s declared type is “pointer to a Text object” not “pointer to a FancyText object,” the
expression p2->get() calls FancyText::get, not Text::get. How does the compiler determine
which method to call in this case? The answer may be surprising: The compiler does not determine which
method to call!
In the case of a virtual method invoked via a pointer, the running program, not the compiler, determines
exactly which code to execute. The process is known as dynamic binding or late binding. Static binding
is relatively easy to understand: the method to execute depends on the declared type of the variable upon
which the method is invoked. The compiler keeps track of the declared type of every variable, so the choice
is easy. Inheritance and the is a relationship make things more complicated. In the example above, p2’s
declared type is ”pointer to Text.” If the compiler were given the authority to select the method to call for

©2013 Richard L. Halterman Draft date: January 27, 2013

17.4. POLYMORPHISM 481

the expression p2->get(), its only choice would be Text::get; however, p2 actually is pointing to a
FancyText object. How does the executing program know which code to execute?

Text *p1, *p2, *p3, *p4;

p1 = new Text("ABC");
p2 = new FancyText("XYZ", "[", "]", "-");
p3 = new FixedText;
p4 = new FancyText("1", "{", "}", ",");

text ABC

vtable_ptr Text vtable

Text::get
p1 0 1 Text::append
text XYZ

left_bracket [

right_bracket ]
p2
connector — FancyText vtable
vtable_ptr FancyText::get
0 1 FancyText::append
p3
text FIXED

vtable_ptr

FixedText vtable
p4 text 1

left_bracket { 0 1 FixedText::append
right_bracket }
connector ,

vtable_ptr

Figure 17.4: Several objects and their associated vtables. You will not find any evidence in the C++ source
code of the dashed pointers and vtables shown in this diagram. When virtual methods are involved the
compiler produces executable code that builds these hidden data structures behind the scenes. When a
client calls a virtual method via a pointer to an object, the run-time environment locates the exact method
to execute by following the vtable pointer stored in the object itself. Observe that since the FixedText class
does not override the inherited get method, its vtable references the same get code as that of Text’s vtable.

We know that every instance of a class contains its own copy of the fields declared within the class. In
the code
Text word1("Wow"), word2("Wee");

the word1 object’s text field in the std::string object representing the string "Wow", and word2’s
text field is "Wee". In reality, if a class contains at least one virtual method, all instances of that class
will contain one extra “hidden” field, a pointer to an array of virtual method pointers. This array of method
pointers commonly is called the vtable. One vtable exists for each class, but all instances of that class must
store a pointer to that shared vtable. The compiler assigns an index to each virtual method in a class. In our
example, the Text class contains two virtual methods, so the index of get might be 0, and the index of
append might be 1.

©2013 Richard L. Halterman Draft date: January 27, 2013

 17.4. POLYMORPHISM 482

Figure 17.4 illustrates a typical scenario with vtables. In this case the compiler translates the expression
p1->get() into the machine language equivalent of p1->vtable[0](), which invokes the correct get
method for the instance pointed to by p1. Similarly for the append method, p1->append("suffix")
internally becomes p1->vtable[1]("suffix"). If p1 points to a Text instance, p1->vtable points to
the vtable array for the Text class. If p1 points instead to a FancyText instance, p1->vtable points to
the vtable array of the FancyText class. In either case the correct method is selected during the program’s
execution.
You may argue that perhaps the compiler could be made to be able to determine in all situations which
virtual method to call. In the code from above:
Text t1("ABC");
FancyText t2("XYZ", "[", "]", ":");
cout << t1.get() << " " << t2.get() << endl;
Text *p1, *p2;
p1 = &t1;
p2 = &t2;
cout << p1->get() << " " << p2->get() << endl;

cannot the compiler deduce p2’s exact type since p2 is assigned to point to t2, and it knows t2’s type is
FancyText?
Such extended analysis capabilities would further complicate the compiler, and C++ compilers are al-
ready very complicated pieces of software. Attempting to add this ability would prove futile because the
compiler cannot in general always determine which method to invoke. To see why, consider the following
code fragment:
Text *p;
if ( std::rand() % 2 == 0 )
p = new Text("ABC");
else
p = new FancyText("XYZ", "[", "]", ":");
cout << p->get() << endl;

This code generates a pseudorandom number at run time. If it generates an even number, p will be directed
to a plain Text object; otherwise, p will point to a FancyText object. Which method will this code call,
Text::get or FancyText::get? Since the code generates the pseudorandom number at run time, the
compiler is powerless to determine the exact type of the object to which p points; thus, the compiler cannot
determine which get method to call. The compiler can, however, generate code that jumps to the method
found at the address stored in p’s associated vtable at the index corresponding to the get method.
Listing 17.3 (vtablesize.cpp) demonstrates that the vtable pointer does occupy space within an object
that contains a virtual method.

Listing 17.3: vtablesize.cpp

1 #include <iostream>
2
3 using namespace std;
4
5 // A class with no virtual methods
6 class NoVTable
7 {
8 int data;
9 public:

©2013 Richard L. Halterman Draft date: January 27, 2013

 17.4. POLYMORPHISM 483

10 void set(int d) { data = d; }

11 int get() { return data; }
12 };
13
14 // A class with virtual methods
15 class HasVTable
16 {
17 int data;
18 public:
19 virtual void set(int d) { data = d; }
20 virtual int get() { return data; }
21 };
22
23 int main()
24 {
25 NoVTable no_vtable;
26 HasVTable has_vtable;
27 cout << "no_vtable size = " << sizeof no_vtable << endl;
28 cout << "has_vtable size = " << sizeof has_vtable << endl;
29 }

The output of Listing 17.3 (vtablesize.cpp) is

no_vtable size = 4
has_vtable size = 8

A NoVTable requires four bytes for its integer field, but a HasVTable object occupies eight bytes—four
bytes for its integer field and four bytes for its secret vtable pointer.
Dynamic binding enables a powerful technique in object-oriented programming called polymorphism.
A polymorphic method behaves differently depending on the actual type of object upon which it is invoked.
Consider Listing 17.4 (polymorphicarray.cpp).

Listing 17.4: polymorphicarray.cpp

1 #include <string>
2 #include <iostream>
3
4 using namespace std;
5
6 // Base class for all Text derived classes
7 class Text
8 {
9 string text;
10 public:
11 Text(const string& t): text(t) {}
12 virtual string get() const
13 {
14 return text;
15 }
16 };
17

©2013 Richard L. Halterman Draft date: January 27, 2013

 17.4. POLYMORPHISM 484

18 // Provides minimal decoration for the text

19 class FancyText: public Text
20 {
21 string left_bracket;
22 string right_bracket;
23 string connector;
24 public:
25 FancyText(const string& t, const string& left,
26 const string& right, const string& conn):
27 Text(t), left_bracket(left), right_bracket(right),
28 connector(conn) {}
29 string get() const override
30 {
31 return left_bracket + Text::get() + right_bracket;
32 }
33 };
34
35 // The text is always the word FIXED
36 class FixedText: public Text
37 {
38 public:
39 FixedText(): Text("FIXED") {}
40 };
41
42 int main()
43 {
44 Text *texts[] = {new Text("Wow"), new FancyText("Wee", "[", "]", "-"),
45 new FixedText, new FancyText("Whoa", ":", ":", ":") };
46 for ( auto t : texts )
47 cout << t->get() << endl;
48 }

We know from Section 11.1 that an array is a collection of homogeneous elements. Homogeneous means
the elements in the array must all be of the same type. Homogeneity takes on a deeper meaning when
inheritance and the is a relationship is involved. In Listing 17.4 (polymorphicarray.cpp) the declared type
of the texts array is Text *. With inheritance, not only can texts hold pointers to simple Text
objects, it also simultaneously can hold pointers to FixedText and FancyText objects. Listing 17.4
(polymorphicarray.cpp) prints

Wow
[Wee]
FIXED
:Whoa:

As we can see, the expression t->get() in the main function is polymorphic; the actual get method
invoked—Text::get, FancyText::get, or FixedText::get—depends on the exact type of t.
As t assumes the value of each element in the array during the loop’s execution, the exact type of object
that t points to varies. Even though all the elements of the texts array are pointers to Text objects,
only one of the elements points to a pure Text object; the rest of the elements point to FancyText or
FixedText objects.

©2013 Richard L. Halterman Draft date: January 27, 2013

17.5. PROTECTED MEMBERS 485

Why must we use pointers to objects rather than the objects themselves to achieve polymorphism?
Remember that a pointer stores a memory address (see Section 10.6). All pointers, no matter what type
they point to, are all the same size (4 bytes on 32-bit systems and 8 bytes on 64-bit systems). Text objects
and FancyText objects are not the same size (see Figure 17.1 for a conceptual picture); FancyText
objects are bigger, containing three extra string fields. All the elements of an array must be the same size.
If we made texts an array of Text objects rather than an array of pointers to Text objects, when we
assign a FancyText object to an element in the texts array, the assignment would slice the extra fields
in the FancyText object. Pointer assignment avoids the slicing problem.
The main reason for using pointers is that C++ uses static binding for all methods (virtual and non-
virtual) invoked on behalf of objects; the compiler chooses the method based on the declared type of the
object. In contrast, C++ uses dynamic binding for virtual method calls made via pointers to objects; the
exact type of the object determines the method selection.

A polymorphic method in C++ requires four key ingredients:

1. The method must appear in a class that is part of an inheritance hierarchy.

2. The method must declared virtual in the base class at the top of the
hierarchy.
3. Derived classes override the behavior of the inherited virtual methods as
needed.
4. Clients must invoke the method via a pointer to an object, not directly
through the object itself.

In summary, polymorphism requires inheritance and the is a relationship. The base class defines the
method signature, and the derived classes override the method with their own custom behavior.
[adapter design pattern example]

17.5 Protected Members

As a more practical, yet still relatively simple example, suppose we need to model two-dimensional shape
objects, such as rectangles, squares, triangles, ellipses, and circles. Our application will determine how to
pack parts into a containing area, so it is important to be able to determine an individual shape object’s area.
Figure 17.5 illustrates packing some two-dimensional shapes into a two-dimensional container. We thus
want to determine the minimum area of the container that can hold a given collection of parts. A real-world
application could be computing the size of the smallest circuit board that can hold a collection of electronic
components. Our program will be much simpler and give us only a rough approximation; it does not take
into account the geometry and orientation of the individual shapes but merely computes the total area of the
components. The actual answer generally will be larger than the result computed by our program.
An interesting problem arises if a method in a derived class needs to use an inherited variable. If the
variable is declared private in the base class, a derived class method cannot access it. The private
specifier means “not available outside the class, period.” We know it is unwise in general to make instance
variables public, since that breaks encapsulation. Since encapsulation is a desirable property, C++ provides
a third level of access protection within a class—protected. A protected member, whether it be data

©2013 Richard L. Halterman Draft date: January 27, 2013

 17.5. PROTECTED MEMBERS 486

Figure 17.5: Packing Two-dimensional Shapes into a Two-dimensional Container

or a method, is inaccessible to all code outside the class, except for code within a derived class. The use of
protected is illustrated in the shapes code.
We begin by defining the base class for all shape objects: Shape. Listing 17.5 (shape.h) provides the
code for Shape.

Listing 17.5: shape.h

1
2 #ifndef _SHAPE_H_
3 #define _SHAPE_H_
4
5 /*
6 * Shape is the base class for all shapes
7 */
8 class Shape
9 {
10 public:
11 // Longest distance across the shape
12 virtual double span() const = 0;
13 // The shape's area

©2013 Richard L. Halterman Draft date: January 27, 2013

 17.5. PROTECTED MEMBERS 487

14 virtual double area() const = 0;

15 };
16
17 #endif

The “assignment to zero” of the two virtual methods in the Shape class make them special; they are called
pure virtual methods or pure virtual functions. This special syntax signifies that these methods have no
implementations. It is not possible to provide a body for these methods. A class that contains at least one
pure virtual method is an abstract class. It is not possible to create an instance of an abstract class. The
compiler enforces this restriction; the following statement;
Shape myShape; // Illegal

is illegal. A non-abstract class is called a concrete class. All the classes we have seen to this point except
for Shape have been concrete classes.
An abstract class represents an abstract concept. Shape is an abstract concept. We can have circles,
rectangles, and lots of other kinds of shapes, but can we have something that is “just a shape” without being
a particular kind of shape? Even though we cannot create instances of abstract classes, as we will see,
abstract classes can be used to organize a class hierarchy.
We can derive a concrete class from our Shape class as shown in Listing 17.6 (rectangle.h).

Listing 17.6: rectangle.h

1
2 #ifndef _RECTANGLE_H_
3 #define _RECTANGLE_H_
4
5 #include "shape.h"
6
7 class Rectangle: public Shape
8 {
9 protected:
10 double length;
11 double width;
12 public:
13 Rectangle(double len, double wid);
14 // Length of the longer side
15 double span() const override;
16 double area() const override;
17 };
18
19 #endif

Note the new access specifier: protected. A protected member is like a private member, except methods
within derived classes will have complete access to the inherited member. Listing 17.7 (rectangle.cpp)
provides the implementation of Rectangle’s methods.

Listing 17.7: rectangle.cpp

1 // File rectangle.cpp
2
3 #include "rectangle.h"
4 #include <algorithm> // For max function

©2013 Richard L. Halterman Draft date: January 27, 2013

 17.5. PROTECTED MEMBERS 488

5
6 // Generally for rectangles length >= width, but the
7 // constructor does not enforce this.
8 Rectangle::Rectangle(double len, double wid): length(len), width(wid) {}
9
10 // Length of the longer side--determine which is longer
11 double Rectangle::span() const
12 {
13 return std::max(length, width);
14 }
15
16 double Rectangle::area() const
17 {
18 return length * width;
19 }

From mathematics we know that a square is a special kind of rectangle, so Listing 17.8 (square.h) and
Listing 17.9 (square.cpp) specify a Square class.

Listing 17.8: square.h

1
2 #ifndef _SQUARE_H_
3 #define _SQUARE_H_
4
5 #include "rectangle.h"
6
7 // A square is a special case of a rectangle
8
9 class Square: public Rectangle
10 {
11 public:
12 // Length and width are equal in a square, so specify the
13 // length of only one side
14 Square(double side);
15 // The inherited methods work as is; no need to
16 // override them.
17 };
18
19 #endif

Listing 17.9: square.cpp

1 // File square.cpp
2
3 #include "square.h"
4
5 // Defer the work of initialization to the base class constructor
6 Square::Square(double side): Rectangle(side, side) {}

Next, we add a triangle shape (Listing 17.10 (triangle.h) and Listing 17.11 (triangle.cpp)).

Listing 17.10: triangle.h

1

©2013 Richard L. Halterman Draft date: January 27, 2013

 17.5. PROTECTED MEMBERS 489

2 #ifndef _TRIANGLE_H_
3 #define _TRIANGLE_H_
4
5 #include "shape.h"
6
7 class Triangle: public Shape
8 {
9 protected:
10 double side1; // Triangles have three sides
11 double side2;
12 double side3;
13 public:
14 Triangle(double s1, double s2, double s3);
15 double span() const override;
16 double area() const override;
17 };
18
19 #endif

Listing 17.11: triangle.cpp

1 // File triangle.cpp
2
3 #include "triangle.h"
4 #include <algorithm> // For max function
5
6 Triangle::Triangle(double s1, double s2, double s3):
7 side1(s1), side2(s2), side3(s3) {}
8
9 // The span of a triangle is the length of the longest side
10 double Triangle::span() const
11 {
12 return std::max(side1, std::max(side2, side3));
13 }
14
15 // Not having the base and height of the triangle explicitly, we
16 // use Heron's formula to compute the area
17 double Triangle::area() const
18 {
19 // Compute semiperimeter
20 double s = (side1 + side2 + side3)/2;
21 // Compute area using Heron's formula
22 return s*(s - side1)*(s - side2)*(s - side3);
23 }

An ellipse is a simple curved shape that we can add to our class hierarchy. Listing 17.12 (ellipse.h) and
Listing 17.13 (ellipse.cpp) define the ellipse class.

Listing 17.12: ellipse.h

1
2 #ifndef _ELLIPSE_H_
3 #define _ELLIPSE_H_
4
5 #include "shape.h"

©2013 Richard L. Halterman Draft date: January 27, 2013

 17.5. PROTECTED MEMBERS 490

6
7 class Ellipse: public Shape
8 {
9 protected:
10 double major_radius; // The longer radius of the ellipse
11 double minor_radius; // The shorter radius of the ellipse
12 public:
13 Ellipse(double major, double minor);
14 double span() const override;
15 double area() const override;
16 };
17
18 #endif

Listing 17.13: ellipse.cpp

1 // File ellipse.cpp
2
3 #include "ellipse.h"
4 #include <algorithm> // For max function
5
6 // PI is local to this file
7 static const double PI = 3.14159;
8
9 // Note: This constructor does not enforce
10 // major_axis >= minor_axis
11 Ellipse::Ellipse(double major, double minor):
12 major_radius(major), minor_radius(minor) {}
13
14 // Greatest distance across is the length of the longer radius
15 double Ellipse::span() const
16 {
17 return std::max(major_radius, minor_radius);
18 }
19
20 double Ellipse::area() const
21 {
22 return PI * major_radius * minor_radius;
23 }

A circle is just an ellipse with equal major and minor radii. Listing 17.14 (circle.h) and Listing 17.15
(circle.cpp) define the Circle class.

Listing 17.14: circle.h

1
2 #ifndef _CIRCLE_H_
3 #define _CIRCLE_H_
4
5 #include "ellipse.h"
6
7 // A circle is a special case of an ellipse
8
9 class Circle: public Ellipse
10 {

©2013 Richard L. Halterman Draft date: January 27, 2013

 17.5. PROTECTED MEMBERS 491

11 public:
12 // In a circle the major and minor radii are the same, so
13 // we need specify only one value when creating a circle.
14 Circle(double radius);
15 // Inherited methods work as is, no need to override
16 };
17
18 #endif

Listing 17.15: circle.cpp

1 // File circle.cpp
2
3 #include "circle.h"
4
5 // PI is local to this file
6 static const double PI = 3.14159;
7
8 Circle::Circle(double radius): Ellipse(radius, radius) {}

These shape classes form the class hierachy shown in Figure 17.6.

Shape

Rectangle Triangle Ellipse

Square Circle

Figure 17.6: Inheritance hierarchy of shape classes

Listing 17.16: testshapes.cpp

1 // File testshapes.cpp
2
3 #include <iostream>
4 #include <vector>
5

©2013 Richard L. Halterman Draft date: January 27, 2013

 17.5. PROTECTED MEMBERS 492

6 #include "rectangle.h"
7 #include "triangle.h"
8 #include "circle.h"
9 #include "ellipse.h"
10
11 using namespace std;
12
13 int main()
14 {
15 Rectangle rect(3, 4);
16 Circle circ(4.5);
17 Triangle tri(3, 4, 5);
18 Ellipse elli(3, 4);
19 vector<Shape *> shape_list;
20
21 shape_list.push_back(&circ);
22 shape_list.push_back(&tri);
23 shape_list.push_back(&rect);
24 shape_list.push_back(&elli);
25
26 int n = shape_list.size();
27 double area_total = 0.0, max_span = 0.0;
28
29 for ( int i = 0; i < n; i++ )
30 {
31 // Examine the area each shape
32 cout << "Area = " << shape_list[i]->area() << endl;
33 // Accumulate the areas of all the shapes
34 area_total += shape_list[i]->area();
35 // Account for the longest object
36 if ( max_span < shape_list[i]->span() )
37 max_span = shape_list[i]->span();
38 }
39 // Report the total area of all the shapes combined
40 cout << "Total shape area is " << area_total << endl;
41 // Report the minimum length of the container
42 cout << "Longest shape is " << max_span << endl;
43 }

Observe that neither Square nor Circle needed access to the protected fields of their base classes.
Consider Listing 17.17 (drawablerectangle.cpp) that derives a new class from Rectangle to allow clients
to draw rectangle objects in a console window using text graphics.

Listing 17.17: drawablerectangle.cpp

1 #include <iostream>
2 #include "rectangle.h"
3
4 int round(double d)
5 {
6 return static_cast<int>(d + 0.5);
7 }
8
9 class DrawableRectangle: public Rectangle
10 {

©2013 Richard L. Halterman Draft date: January 27, 2013

 17.5. PROTECTED MEMBERS 493

11 public:
12 // Delegate construction to the base class
13 DrawableRectangle(double length, double width):
14 Rectangle(length, width) {}
15
16 // Draw a rectangle using text graphics
17 void draw() const
18 {
19 // Access the inherited protected fields
20 int rows = round(length),
21 columns = round(width);
22 // Draw the rectangle
23 for ( int r = 0; r < rows; r++ )
24 {
25 for ( int c = 0; c < columns; c++ )
26 std::cout << '#';
27 std::cout << std::endl;
28 }
29 }
30 };
31
32 int main()
33 {
34 DrawableRectangle rec1(3, 2),
35 rec2(10, 5),
36 rec3(4, 8);
37 rec1.draw();
38 std::cout << "----------------" << std::endl;
39 rec2.draw();
40 std::cout << "----------------" << std::endl;
41 rec3.draw();
42 }

Listing 17.17 (drawablerectangle.cpp) displays

©2013 Richard L. Halterman Draft date: January 27, 2013

17.6. FINE TUNING INHERITANCE 494

##
##
##
----------------
#####
#####
#####
#####
#####
#####
#####
#####
#####
#####
----------------
########
########
########
########

The DrawableRectangle::draw method needs access to the fields length and width to be able to
draw the rectangle. It is important to remember that every DrawableRectangle object contains these
fields because it inherits them from Rectangle; however, if the length and width fields were declared
private instead of protected in Rectangle, DrawableRectangle::draw would not be able to access
these fields.

A class supports three levels of protection for its members:

1. private: This is the default. Private members are accessible to any code
within the class itself but are inaccessible to code outside of the class.
2. protected: Protected members are accessible to code within the class it-
self and to code within derived classes that use public inheritance. Protected
members are inaccessible to other (non-derived) classes.
3. public: Public members are accessible to code anywhere.

In sum, private means not accessible outside the class, period. Public means open
to everyone. Protected means “public” to subclasses and “private” to all other
classes.

17.6 Fine Tuning Inheritance

Recall the flexible sorting code found in Listing 16.5 (loggingflexiblesort.cpp). It uses objects to enable the
following functionality:

©2013 Richard L. Halterman Draft date: January 27, 2013

 17.6. FINE TUNING INHERITANCE 495

• The ordering imposed by a selection sort function can be varied by using custom comparison func-
tions.
• The object used to determine the sort’s ordering collects data about the number of comparisons and
element interchanges the sort function performs.

Listing 17.20 (polymorphicsort.cpp) provides a slight variation of Listing 16.5 (loggingflexiblesort.cpp).

Besides splitting the code up into multiple source files, it adds two virtual methods that enable future
developers to customize the compare and swap methods in derived classes.
Listing 17.18 (comparer.h) contains the declaration of the Comparer class.

Listing 17.18: comparer.h

1 #ifndef _COMPARER_H_
2 #define _COMPARER_H_
3 /*
4 * Comparer objects manage the comparisons and element
5 * interchanges on the selection sort function below.
6 */
7 class Comparer
8 {
9 // The object's data is private, so it is inaccessible to
10 // clients and derived classes
11
12 // Keeps track of the number of comparisons
13 // performed
14 int compare_count;
15
16 // Keeps track of the number of swaps performed
17 int swap_count;
18
19 // Function pointer directed to the function to
20 // perform the comparison
21 bool (*comp)(int, int);
22
23 protected:
24 // Method that actually performs the comparison
25 // Derived classes may override this method
26 virtual bool compare_impl(int m, int n);
27
28 // Method that actually performs the swap
29 // Derived classes may override this method
30 virtual void swap_impl(int& m, int& n);
31
32 public:
33 // The client must initialize a Comparer object with a
34 // suitable comparison function.
35 Comparer(bool (*f)(int, int));
36
37 // Resets the counters to make ready for a new sort
38 void reset();
39
40 // Method that performs the comparison. It delegates
41 // the actual work to the function pointed to by comp.
42 // This method logs each invocation.

©2013 Richard L. Halterman Draft date: January 27, 2013

 17.6. FINE TUNING INHERITANCE 496

43 bool compare(int m, int n);

44
45 // Method that performs the swap.
46 // Interchange the values of
47 // its parameters a and b which are
48 // passed by reference.
49 // This method logs each invocation.
50 void swap(int& m, int& n);
51
52 // Returns the number of comparisons this object has
53 // performed since it was created.
54 int comparisons() const;
55
56 // Returns the number of swaps this object has
57 // performed since it was created.
58 int swaps() const;
59 };
60 #endif

Listing 17.19 (comparer.cpp) provides the implementation of the Comparer class methods.

Listing 17.19: comparer.cpp

1 #include "comparer.h"
2
3 // Method that actually performs the comparison
4 // Derived classes may override this method
5 bool Comparer::compare_impl(int m, int n)
6 {
7 return comp(m, n);
8 }
9
10 // Method that actually performs the swap
11 // Derived classes may override this method
12 void Comparer::swap_impl(int& m, int& n)
13 {
14 int temp = m;
15 m = n;
16 n = temp;
17 }
18
19 // The client must initialize a Comparer object with a
20 // suitable comparison function.
21 Comparer::Comparer(bool (*f)(int, int)):
22 compare_count(0), swap_count(0), comp(f) {}
23
24 // Resets the counters to make ready for a new sort
25 void Comparer::reset()
26 {
27 compare_count = swap_count = 0;
28 }
29
30 // Method that performs the comparison. It delegates
31 // the actual work to the function pointed to by comp.
32 // This method logs each invocation.

©2013 Richard L. Halterman Draft date: January 27, 2013

 17.6. FINE TUNING INHERITANCE 497

33 bool Comparer::compare(int m, int n)

34 {
35 compare_count++;
36 return compare_impl(m, n);
37 }
38
39 // Method that performs the swap.
40 // Interchange the values of
41 // its parameters a and b which are
42 // passed by reference.
43 // This method logs each invocation.
44 void Comparer::swap(int& m, int& n)
45 {
46 swap_count++;
47 swap_impl(m, n);
48 }
49
50 // Returns the number of comparisons this object has
51 // performed since it was created.
52 int Comparer::comparisons() const
53 {
54 return compare_count;
55 }
56
57 // Returns the number of swaps this object has
58 // performed since it was created.
59 int Comparer::swaps() const
60 {
61 return swap_count;
62 }

Notice that even though the virtual keyword appears before the method declarations in comparer.h, it
does not appear before the method implementations in comparer.cpp.
Listing 17.20 (polymorphicsort.cpp) provides the client code that tests the new Comparer class.

Listing 17.20: polymorphicsort.cpp

1 #include <iostream>
2 #include <vector>
3 #include "comparer.h"
4
5 using namespace std;
6
7 /*
8 * selection_sort(a, compare)
9 * Arranges the elements of vector a in an order determined
10 * by the compare object.
11 * a is a vector of ints.
12 * compare is a function that compares the ordering of
13 * two integers.
14 * The contents of a are physically rearranged.
15 */
16 void selection_sort(vector<int>& a, Comparer& compare)
17 {
18 int n = a.size();

©2013 Richard L. Halterman Draft date: January 27, 2013

 17.6. FINE TUNING INHERITANCE 498

19 for ( int i = 0; i < n - 1; i++ )

20 {
21 // Note: i,small, and j represent positions within a
22 // a[i], a[small], and a[j] represents the elements at
23 // those positions.
24 // small is the position of the smallest value we've seen
25 // so far; we use it to find the smallest value less
26 // than a[i]
27 int small = i;
28 // See if a smaller value can be found later in the array
29 for ( int j = i + 1; j < n; j++ )
30 if ( compare.compare(a[j], a[small]) )
31 small = j; // Found a smaller value
32 // Swap a[i] and a[small], if a smaller value was found
33 if ( i != small )
34 compare.swap(a[i], a[small]);
35 }
36 }
37
38 /*
39 * print
40 * Prints the contents of an integer vector
41 * a is the vector to print.
42 * a is not modified.
43 * /
44 void print(const vector<int>& a)
45 {
46 int n = a.size();
47 cout << '{';
48 if ( n > 0 )
49 {
50 cout << a[0]; // Print the first element
51 for ( int i = 1; i < n; i++ )
52 cout << ',' << a[i]; // Print the rest
53 }
54 cout << '}';
55 }
56
57 /*
58 * less_than(a, b)
59 * Returns true if a < b; otherwise, returns
60 * false.
61 */
62 bool less_than(int a, int b)
63 {
64 return a < b;
65 }
66
67 /*
68 * greater_than(a, b)
69 * Returns true if a > b; otherwise, returns
70 * false.
71 */
72 bool greater_than(int a, int b)
73 {

©2013 Richard L. Halterman Draft date: January 27, 2013

 17.6. FINE TUNING INHERITANCE 499

74 return a > b;
75 }
76
77 int main()
78 {
79 // Make a vector of integers from an array
80 int a[] = { 23, -3, 4, 215, 0, -3, 2, 23, 100, 88, -10 };
81 int len = (sizeof a)/(sizeof a[0]);
82 vector<int> original(a, a + len);
83
84 // Make a working copy of the original vector
85 vector<int> working = original;
86 cout << "Before: ";
87 print(working);
88 cout << endl;
89 Comparer lt(less_than), gt(greater_than);
90 selection_sort(working, lt);
91 cout << "Ascending: ";
92 print(working);
93 cout << " (" << lt.comparisons() << " comparisons, "
94 << lt.swaps() << " swaps)" << endl;
95 cout << "----------------------------" << endl;
96 // Make another copy of the original vector
97 working = original;
98 cout << "Before: ";
99 print(working);
100 cout << endl;
101 selection_sort(working, gt);
102 cout << "Descending: ";
103 print(working);
104 cout << " (" << gt.comparisons() << " comparisons, "
105 << gt.swaps() << " swaps)" << endl;
106 cout << "----------------------------" << endl;
107 // Sort a sorted vector
108 cout << "Before: ";
109 print(working);
110 cout << endl;
111 // Reset the greater than comparer so we start counting at
112 // zero
113 gt.reset();
114 selection_sort(working, gt);
115 cout << "Descending: ";
116 print(working);
117 cout << " (" << gt.comparisons() << " comparisons, "
118 << gt.swaps() << " swaps)" << endl;
119 }

The functions in Listing 17.20 (polymorphicsort.cpp) are identical to the functions in Listing 16.5 (loggingflexiblesort.cpp),
and, not surprisingly, the output of Listing 17.20 (polymorphicsort.cpp) is identical to the output of List-
ing 16.5 (loggingflexiblesort.cpp):

©2013 Richard L. Halterman Draft date: January 27, 2013

 17.6. FINE TUNING INHERITANCE 500

Before: {23,-3,4,215,0,-3,2,23,100,88,-10}
Ascending: {-10,-3,-3,0,2,4,23,23,88,100,215} (55 comparisons, 7 swaps)
----------------------------
Before: {23,-3,4,215,0,-3,2,23,100,88,-10}
Descending: {215,100,88,23,23,4,2,0,-3,-3,-10} (55 comparisons, 5 swaps)
----------------------------
Before: {215,100,88,23,23,4,2,0,-3,-3,-10}
Descending: {215,100,88,23,23,4,2,0,-3,-3,-10} (55 comparisons, 0 swaps)

The question arises: What advantage does the new Comparer class have over the original one?
The design of this new Comparer class is interesting:

• Since its counter fields are private, only methods in the Comparer class itself can access a Comparer
object’s data.
• The two virtual methods, compare_impl and swap_impl, are protected, so clients cannot access
them directly. Derived classes, however, can see them and override them. The suffix _impl stands
for “implementation,” so compare_impl represents the implementation details of the compare
method and swap_impl represents the implementation details of the swap method.
• The public compare and swap methods both delegate part of their work to the protected compare_impl
and swap_impl methods.
• The compare and swap methods are not declared virtual, so derived classes cannot override
them.
• The compare and swap methods manage the compare_count and swap_count counter fields.
Since derived classes cannot see these fields, there is nothing that a designer of a derived class can do
when overriding compare_impl or swap_impl to disturb the correct accounting of the number
of times a client calls compare or swap.
• The comparisons and swaps methods that report the results to the client are non-virtual, so
derived classes may not override their behavior.

Observe that the designer of the Comparer class allows the nature of the comparisons and swaps in derived
classes to be flexible, but it is rigid on the enforcement of how the accounting is performed and reported.
The proper use of protected and private specifiers in a base class as shown in Comparer affords
class designers a great deal of control over exactly what derived class designers can do. Derived classes
may adapt some behaviors, but other behaviors are non-negotiable.
What kind of customization would a programmer want to do to the Comparer class beyond chang-
ing how the comparison is performed? Consider the LogComparer class declared in Listing 17.21
(logcomparer.h).

Listing 17.21: logcomparer.h

1 #ifndef _LOGCOMPARER_H_
2 #define _LOGCOMPARER_H_
3
4 #include <fstream>
5 #include <string>

©2013 Richard L. Halterman Draft date: January 27, 2013

 17.6. FINE TUNING INHERITANCE 501

6 #include "comparer.h"
7
8 /*
9 * Comparer objects manage the comparisons and element
10 * interchanges on the selection sort function below.
11 */
12 class LogComparer: public Comparer
13 {
14 // Output stream to which logging messages are directed
15 std::ofstream fout;
16
17 protected:
18 // Method that actually performs the comparison
19 bool compare_impl(int m, int n) override;
20
21 // Method that actually performs the swap
22 void swap_impl(int& m, int& n) override;
23
24 public:
25 // The client must initialize a LogComparer object with a
26 // suitable comparison function and the file name of a text
27 // file to which the object will direct logging messages
28 LogComparer(bool (*f)(int, int), const std::string& filename);
29
30 // The destructor must close the log file
31 ∼LogComparer();
32 };
33
34 #endif

The LogComparer class overrides the compare_impl and swap_impl methods. Instances of the
LogComparer class use an ofstream object to record logging information to a text file.
The implementation of the LogComparer methods can be found in Listing 17.22 (logcomparer.cpp).

Listing 17.22: logcomparer.cpp

1 #include "logcomparer.h"
2 #include <cstdlib>
3 #include <iostream>
4
5 using namespace std;
6
7 // Method that actually performs the comparison
8 // Derived classes may override this method
9 bool LogComparer::compare_impl(int m, int n)
10 {
11 fout << "Comparing " << m << " to " << n << std::endl;
12 // Base class method does the comparision
13 return Comparer::compare_impl(m, n);
14 }
15
16 // Method that actually performs the swap
17 // Derived classes may override this method
18 void LogComparer::swap_impl(int& m, int& n)
19 {

©2013 Richard L. Halterman Draft date: January 27, 2013

 17.6. FINE TUNING INHERITANCE 502

20 fout << "Swapping " << m << " and " << n << endl;
21 int temp = m;
22 m = n;
23 n = temp;
24 }
25
26 // The client must initialize a LogComparer object with a
27 // suitable comparison function and the file name of the
28 // text file to receive logging messages.
29 LogComparer::LogComparer(bool (*f)(int, int), const string& filename):
30 Comparer(f)
31 {
32 fout.open(filename.c_str());
33 if ( !fout.good() )
34 {
35 fout.close();
36 cout << "Could not open log file " << filename
37 << " for writing" << endl;
38 exit(1); // Terminate the program
39 }
40 }
41
42 // Destructor closes the log file
43 LogComparer::∼LogComparer()
44 {
45 fout.close();
46 }

The constructor attempts to open a file output stream to write logging information to a text file. If the
constructor cannot for any reason open the text file for writing, it terminates the program’s execution. The
destructor closes the file stream object when the LogComparer object’s life is over.
The two overridden methods, compare_impl and swap_impl, write text to the log file. Both
methods use the LogComparer object’s fout field rather than the cout console output stream object.
Notice that even though the override keyword appears after the method declarations in logcom-
parer.h, it does not appear after the method implementations in logcomparer.cpp.
The client code in Listing 17.23 (loggingsort.cpp) uses a LogComparer object to create a text file
named sort.log.

Listing 17.23: loggingsort.cpp

1 #include <iostream>
2 #include <vector>
3 #include "logcomparer.h"
4
5 using namespace std;
6
7 /*
8 * selection_sort(a, compare)
9 * Arranges the elements of vector a in an order determined
10 * by the compare object.
11 * a is a vector of ints.
12 * compare is a function that compares the ordering of
13 * two integers.

©2013 Richard L. Halterman Draft date: January 27, 2013

 17.6. FINE TUNING INHERITANCE 503

14 * The contents of a are physically rearranged.

15 * /
16 void selection_sort(vector<int>& a, Comparer& compare)
17 {
18 int n = a.size();
19 for ( int i = 0; i < n - 1; i++ )
20 {
21 // Note: i,small, and j represent positions within a
22 // a[i], a[small], and a[j] represents the elements at
23 // those positions.
24 // small is the position of the smallest value we've seen
25 // so far; we use it to find the smallest value less
26 // than a[i]
27 int small = i;
28 // See if a smaller value can be found later in the array
29 for ( int j = i + 1; j < n; j++ )
30 if ( compare.compare(a[j], a[small]) )
31 small = j; // Found a smaller value
32 // Swap a[i] and a[small], if a smaller value was found
33 if ( i != small )
34 compare.swap(a[i], a[small]);
35 }
36 }
37
38 /*
39 * print
40 * Prints the contents of an integer vector
41 * a is the vector to print.
42 * a is not modified.
43 */
44 void print(const vector<int>& a)
45 {
46 int n = a.size();
47 cout << '{';
48 if ( n > 0 )
49 {
50 cout << a[0]; // Print the first element
51 for ( int i = 1; i < n; i++ )
52 cout << ',' << a[i]; // Print the rest
53 }
54 cout << '}';
55 }
56
57 /*
58 * less_than(a, b)
59 * Returns true if a < b; otherwise, returns
60 * false.
61 */
62 bool less_than(int a, int b)
63 {
64 return a < b;
65 }
66
67 int main()
68 {

©2013 Richard L. Halterman Draft date: January 27, 2013

 17.6. FINE TUNING INHERITANCE 504

69 // Make a vector of integers from an array

70 int a[] = { 23, -3, 4, 215, 0, -3, 2 };
71 int len = (sizeof a)/(sizeof a[0]);
72 vector<int> vec(a, a + len);
73
74 // Make a working copy of the original vector
75 cout << "Before: ";
76 print(vec);
77 cout << endl;
78 LogComparer lt(less_than, "sort.log");
79 selection_sort(vec, lt);
80 cout << "Ascending: ";
81 print(vec);
82 cout << " (" << lt.comparisons() << " comparisons, "
83 << lt.swaps() << " swaps)" << endl;
84 }

Listing 17.23 (loggingsort.cpp) prints to the console

Before: {23,-3,4,215,0,-3,2}
Ascending: {-3,-3,0,2,4,23,215} (21 comparisons, 4 swaps)

and writes the file sort.log the text

Comparing -3 to 23
Comparing 4 to -3
Comparing 215 to -3
Comparing 0 to -3
Comparing -3 to -3
Comparing 2 to -3
Swapping 23 and -3
Comparing 4 to 23
Comparing 215 to 4
Comparing 0 to 4
Comparing -3 to 0
Comparing 2 to -3
Swapping 23 and -3
Comparing 215 to 4
Comparing 0 to 4
Comparing 23 to 0
Comparing 2 to 0
Swapping 4 and 0
Comparing 4 to 215
Comparing 23 to 4
Comparing 2 to 4
Swapping 215 and 2
Comparing 23 to 4
Comparing 215 to 4
Comparing 215 to 23

©2013 Richard L. Halterman Draft date: January 27, 2013

17.7. SUMMARY 505

If during development the selection sort function has a problem, the programmer can review the contents
of the log file to examine how the sort progresses and perhaps determine where the problem lies.
The design of the Comparer class provides a overall structure that allows inheriting classes to fine tune
the details without disturbing the overarching framework. Inheritance and polymorphism allow us to add
functionality to an existing code base. Virtual methods provide extension points by which derived classes
can add custom behavior.

17.7 Summary

• A new class can be derived from an existing class via inheritance.

• Other names for inheritance include subclassing, derivation, and specialization.
• In the code
class Gadget: public Widget
{
// Details omitted
};

– Widget is the base class (also called the superclass), and Gadget is the derived class (also
called the subclass).
– Gadget inherits all the instance variables and methods from Widget.
– Code within the Gadget class can see and use the public and protected members inherited
from Widget.

• The file streams ofstream and ifstream are subclasses of, respectively, ostream and istream.
Because of the is a relationship imposed by inheritance, an ofstream object can be used in any
context that expects an ostream object reference or pointer, and ifstream objects can used where
an istream object is required.
• The specifier protected works like “private” as far as most code outside the class is concerned,
but it works like “public” to code within derived classes. Subclass methods can see and manipu-
late inherited protected data and call inherited protected methods, but subclasses cannot directly see
inherited private data and cannot call directly inherited private methods.
• Polymorphism in object-oriented languages is a process by which the code executed during a method
invocation is determined at run time, not compile time.
• Polymorphism in C++ requires three key ingredients:
– inheritance—a derived class overrides a method in its base class
– virtual methods—an overridden virtual method can use dynamic binding (code to execute
is determined at run time) rather than static binding (compiler determines the code to execute)
– method calls through pointers or references— a pointer’s declared type is of the base class, but
the actual type of the object to which it points may be a derived class. The is a relationship
enables the process to work correctly.
• A derived class may override a public method that is declared virtual in the base class. Such virtual
methods enable polymorphism.

©2013 Richard L. Halterman Draft date: January 27, 2013

17.8. EXERCISES 506

• A pure virtual method (or pure virtual function) has no defined body and it indicated by the ”= 0”
syntax.

• A class that contains at least one pure virtual function is known as an abstract class. It is illegal to
attempt to create an instance of an abstract class. Abstract classes serve to organize hierarchies of
classes related through inheritance. Classes that are not abstract are called concrete classes.

17.8 Exercises

1. Overload the operator<< function so it accepts an ostream reference and a two-dimensional

vector of Boolean values. The function should print the contents of the 2D vector in a table; for
example, a four-row, five column print out might look like:

true false true true true

false true false false false
false false false false false
true false true true true

Note that it prints the words true and false, not simply 0 or 1. Once you have tested your function
using the cout object, demonstrate that you can save the contents of any 2D Boolean vector to a text
file.

2. Consider the following C++ code:

class Widget
{
public:
virtual int f() { return 1; }
};

class Gadget: public Widget

{
public:
virtual int f() { return 2; }
};

class Gizmo: public Widget

{
public:
virtual int f() { return 3; }
};

void do_it(Widget *w)

{
cout << w->f() << " ";
}

int main()
{

©2013 Richard L. Halterman Draft date: January 27, 2013

17.8. EXERCISES 507

vector<Widget *> widgets;

Widget wid;
Gadget gad;
Gizmo giz;
widgets.push_back(&wid);
widgets.push_back(&gad);
widgets.push_back(&giz);
for ( size_t i = 0; i < widgets.size(); i++ )
do_it(widgets[i]);
}

(a) What does the program print?

(b) Would the program still compile and run if the f method within the Widget class were a pure
virtual function?
(c) How would the program run differently if the virtual keyword were removed from all the
code?
(d) Would the program behave the same if the virtual keyword were removed from all the
classes except for Widget?

3. Recall the Comparer class in Section ?? that we used as a base class for custom comparer ob-
jects that changed the way the selection sort function ordered the elements of a vector. Derive
a new Comparer class called PriorityComparer that can be used in conjunction with the
selection_sort function to arrange the elements in an integer vector as described below.
The PriorityComparer constructor requires a single integer argument. Elements in the vector
with this value should appear before all other elements in the vector. The remaining vector elements
should appear in ascending order. For example, if a vector contains 4,3,2,3,10,3,2,4,2, and the special
priority value is 3, the selection sort when provided with such a priority comparer object would sort
the vector into 3,3,3,2,2,2,4,4,10. Notice that all the elements equal to 3 were moved to the front of
the vector, and all other elements follow in non-descending order. If the priority element does not
appear in the vector, all the elements are arranged in ascending order as usual.
Your PriorityComparer class should work with the selection_sort function found in Sec-
tion ??.
4. Recall the Comparer class in Section ?? that we used as a base class for custom comparer ob-
jects that changed the way the selection sort function ordered the elements of a vector. Subclass
Comparer to make new comparer class named NegPosComparer that can be used to create ob-
jects that allow the selection sort function to arrange the elements of an integer vector is this way:
• all positive numbers appear before any negative number or zero
• all negative numbers appear after any positive number or zero
• any zeros, if present, appear after all the positive numbers and before all the negative numbers
• the positive numbers are arranged in ascending order
• the negative numbers are arranged in descending order
Note that the ordering of positive and negative values is different!
For example, if the vector initially held -5,4,-3,2,0, after passing it and a pointer to your subclass
object to the selection sort function, the vector would contain 2,4,0,-3,-5.

©2013 Richard L. Halterman Draft date: January 27, 2013

17.8. EXERCISES 508

©2013 Richard L. Halterman Draft date: January 27, 2013

 509

Chapter 18

Generic Programming

CAUTION! CHAPTER UNDER CONSTRUCTION

The flexible sorting examples in Chapters 12, 16, and 17 allow us to arrange the elements in integer
arrays and vectors in creative ways and perform other interesting activities such as logging. Those exam-
ples demonstrate the power of function pointers, inheritance, and polymorphism. As flexible and powerful
as these techniques are, in isolation they all contain one major limitation: they deal with arrays or vec-
tors of integers only. Consider the Comparer class from Listing 17.18 (comparer.h) and Listing 17.19
(comparer.cpp), and the “flexible” selection_sort function from Listing 17.23 (loggingsort.cpp). The
selection_sort function accepts a vector of integers and a Comparer reference. What if we need
to sort a vector of double-precision floating-point numbers or a vector of string objects? All the nice
flexible code we spent time developing to be as flexible as possible cannot handle this seeming minor
variation.
In this chapter we look at C++’s template mechanism that enables programmers to develop truly generic
algorithms and data structures. We will see how the standard C++ library has embraced the template tech-
nology to provide a wealth of generic algorithms and data structures that greatly assist developers in the
construction of quality software.

18.1 Function Templates

Consider the following comparison function found in Listings 12.2, 16.5, and 17.23:
/*
* less_than(a, b)
* Returns true if a < b; otherwise, returns
* false.
* /
bool less_than(int a, int b)
{
return a < b;
}

Listing 18.1 (testlessthan.cpp) tests the less_than function with various arguments.

©2013 Richard L. Halterman Draft date: January 27, 2013

 18.1. FUNCTION TEMPLATES 510

Listing 18.1: testlessthan.cpp

1 #include <iostream>
2
3 using namespace std;
4
5 /*
6 * less_than(a, b)
7 * Returns true if a < b; otherwise, returns
8 * false.
9 */
10 bool less_than(int a, int b)
11 {
12 return a < b;
13 }
14
15 int main()
16 {
17 cout << less_than(2, 3) << endl;
18 cout << less_than(2.2, 2.7) << endl;
19 cout << less_than(2.7, 2.2) << endl;
20 }

The compiler generates warnings for the last two statements. The less_than function expects two in-
teger arguments, but the calling code sends two double-precision floating-point values. The automatic
conversion from double to int can lose information, hence the warnings. The output of Listing 18.1
(testlessthan.cpp) shows that these warnings should be taken seriously:

1
0
0

Obviously 2 < 3 and 2.7 ≮ 2.2, but 2.2 ≮ 2.7? The automatic double to int conversion truncates, so
the less_than functions sees both 2.2 and 2.7 as 2, and clearly 2 ≮ 2.
The situation is even worse for the following code:
string word1 = "ABC", word2 = "XYZ";
cout << (word1 < word2) << endl;
cout << (word2 < word1) << endl;
// The next statement will not compile
//cout << less_than(word1, word2) << endl;

The C++ standard library provides a global operator< function that compares two std::string ob-
jects lexicographically. The first two cout statements produce the expected output:

1
0

©2013 Richard L. Halterman Draft date: January 27, 2013

 18.1. FUNCTION TEMPLATES 511

but the last line that is commented out will not compile with the commenting symbols removed. This
is because the C++ library does not provide a standard conversion from std::string to int, and
less_than requires two int parameters.
There is a way to express the less_than function that will enable it to work correcty in all the
situations above. To do so we must use a C++ template function. Listing 18.2 (templatelessthan.cpp) uses a
template function to create a generic less_than function.

Listing 18.2: templatelessthan.cpp

1 #include <iostream>
2 #include <string>
3
4 using namespace std;
5
6 /*
7 * less_than(a, b)
8 * Returns true if a < b; otherwise, returns
9 * false.
10 */
11 template <typename T>
12 bool less_than(T a, T b)
13 {
14 return a < b;
15 }
16
17 int main()
18 {
19 cout << less_than(2, 3) << endl;
20 cout << less_than(2.2, 2.7) << endl;
21 cout << less_than(2.7, 2.2) << endl;
22
23 string word1 = "ABC", word2 = "XYZ";
24 cout << less_than(word1, word2) << endl;
25 cout << less_than(word2, word1) << endl;
26 }

The output of Listing 18.2 (templatelessthan.cpp) is

1
1
0
1
0

In the less_than function definition:

template <typename T>
bool less_than(T a, T b)
{
return a < b;
}

©2013 Richard L. Halterman Draft date: January 27, 2013

18.1. FUNCTION TEMPLATES 512

template and typename are reserved words, and T is a type parameter. The template keyword
indicates that the function definition that follows is not a normal function definition but rather is a pattern or
template from which the compiler can produce the correct function definition. A function template is also
known as a generic function. The typename keyword indicates that the identifier that follows (in this case
T) is a placeholder for a C++ type name. The type parameter T stands for an actual type to be determined
elsewhere. T is an identifier and can have any name; some programmers prefer more descriptive names
such as CompareType or TypeParam. Single capital letters such as T, S, U, and V are popular type
parameter names among C++ programmers. Just as a regular parameter of a function or method represents
a value, the type parameter represents a type.
The reserved word class may be used interchangeably with the keyword typename is this context:

template <class T>

bool less_than(T a, T b)
{
return a < b;
}

In Listing 18.2 (templatelessthan.cpp) the compiler generates three different less_than functions:
one that accepts two integers, one that accepts two double-precision floating-point values, and one the
accepts two string objects. Although the programmer cannot see them in the source code, the compiler in
effect creates the following literal function definitions:

bool less_than(int a, int b)

{
return a < b;
}

bool less_than(double a, double b)

{
return a < b;
}

bool less_than(string a, string b)

{
return a < b;
}

When the compiler generates an actual function definition from a template function, the process is called
template instantiation, not to be confused with an object instantiation from a class.
The compiler instantiates functions only as needed. If Listing 18.2 (templatelessthan.cpp) did not
contain the std::string objects, when processing the source code the compiler would not instantiate
the version of less_than that accepts two std::string objects.
For efficiency purposes, the better way to write less_than is

template <typename T>

bool less_than(const T& a, const T& b)
{
return a < b;
}

©2013 Richard L. Halterman Draft date: January 27, 2013

 18.1. FUNCTION TEMPLATES 513

Notice that parameters a and b are passed by reference, not by value. This means the caller passes only the
addresses of objects a and b and does not need to make copies to send to the function. This can make a big
difference if T represents large objects. Since the parameters are const references, the function cannot
change the state of the parameters; therefore, the function offers call-by-value safety without the overhead
of copying the parameters.
The compiler can instantiate an actual function from a template function only if the type substituted for
the template parameter supports its use within the function. Consider the following code:
vector<int> vec1, vec2;
vec1.push_back(3);
vec1.push_back(5);
vec2.push_back(-9);
vec2.push_back(10);
if ( less_than(vec1, vec2 ) // <-- Error!
cout << "vec1 is less than vec2" << endl;

The expression less_than(vec1, vec2) is illegal because no operator< exists to compare two
vector objects; thus, the compiler cannot produce the machine language that represents the expression
a < b when a and b are std::vectors. Ultimately, therefore, the compiler cannot instantiate an
actual less_than function that accepts two vector objects.
All current C++ compilers share one weakness with template processing: poor human-readable error
messages. This is because the compiler bases its error messages on its processing of the instantiated code,
not the original un-instantiated code. Since programmers see only the un-instantiated source, deciphering
the compiler’s error messages is more challenging than usual.
Listing 18.3 (templateflexsort.cpp) increases the flexibility of our flexible sorting program from List-
ing 12.2 (flexiblesort.cpp).

Listing 18.3: templateflexsort.cpp

1 #include <iostream>
2 #include <string>
3 #include <algorithm>
4
5 using namespace std;
6
7 /*
8 * less_than(a, b)
9 * Returns true if a < b; otherwise, returns
10 * false.
11 */
12 template <typename T>
13 bool less_than(const T& a, const T& b)
14 {
15 return a < b;
16 }
17
18 /*
19 * greater_than(a, b)
20 * Returns true if a > b; otherwise, returns
21 * false.
22 */
23 template <typename T>

©2013 Richard L. Halterman Draft date: January 27, 2013

 18.1. FUNCTION TEMPLATES 514

24 bool greater_than(const T& a, const T& b)

25 {
26 return a > b;
27 }
28
29 /*
30 * selection_sort(a, n, compare)
31 * Arranges the elements of a in an order determined
32 * by the compare function.
33 * a is an array
34 * n is the array's length.
35 * compare is a function that compares the ordering of
36 * two types that support operator<.
37 * The contents of a is physically rearranged.
38 */
39 template <typename T>
40 void selection_sort(T a[], int n, bool (compare)(const T&, const T&))
41 {
42 for ( int i = 0; i < n - 1; i++ )
43 {
44 // Note: i,small, and j represent positions within a
45 // a[i], a[small], and a[j] represents the elements at
46 // those positions.
47 // small is the position of the smallest value we've seen
48 // so far; we use it to find the smallest value less
49 // than a[i]
50 int small = i;
51 // See if a smaller value can be found later in the array
52 for ( int j = i + 1; j < n; j++ )
53 if ( compare(a[j], a[small]) )
54 small = j; // Found a smaller value
55 // Swap a[i] and a[small], if a smaller value was found
56 if ( i != small )
57 swap(a[i], a[small]); // std:swap from algorithm header
58 }
59 }
60
61
62 /*
63 * print
64 * Prints the contents of an array
65 * a is the array to print.
66 * n is the array's length.
67 * a is not modified.
68 */
69 template <typename T>
70 void print(const T a[], int n)
71 {
72 cout << '{';
73 if ( n > 0 )
74 {
75 cout << a[0]; // Print the first element
76 for ( int i = 1; i < n; i++ )
77 cout << ',' << a[i]; // Print the rest
78 }

©2013 Richard L. Halterman Draft date: January 27, 2013

 18.1. FUNCTION TEMPLATES 515

79 cout << '}';

80 }
81
82
83 int main()
84 {
85 int list[] = { 23, -3, 4, 215, 0, -3, 2, 23, 100, 88, -10 };
86 int len = (sizeof list)/(sizeof list[0]);
87 cout << "Original: ";
88 print(list, len);
89 cout << endl;
90 selection_sort(list, len, less_than<int>);
91 cout << "Ascending: ";
92 print(list, len);
93 cout << endl;
94 selection_sort(list, len, greater_than<int>);
95 cout << "Descending: ";
96 print(list, len);
97 cout << endl;
98 cout << "-------------------------------" << endl;
99 string words[] = { "tree", "girl", "boy", "apple", "dog", "cat", "bird" };
100 len = (sizeof words)/(sizeof words[0]);
101 cout << "Original: ";
102 print(words, len);
103 cout << endl;
104 selection_sort(words, len, less_than<string>);
105 cout << "Ascending: ";
106 print(words, len);
107 cout << endl;
108 selection_sort(words, len, greater_than<string>);
109 cout << "Descending: ";
110 print(words, len);
111 cout << endl;
112 }

Notice that all the functions except for main are generic functions. The selection_sort function is the
most interesting one. It accepts a generic array and a function pointer that accepts two generic parameters
of the same type. In one of its invocations in main:
selection_sort(list, len, greater_than<int>);

the expression greater_than<int> is an explicit template instantiation, and it tells the compiler we
want to send to the selection_sort function a pointer to the greater_than function instantiated
with integer parameters. This explicit instantiation is required because the compiler cannot instantiate the
greater_than function without an actual call to the function, and a pointer to a function is not a function
invocation. At this point in the program we are not calling the greater_than function but only passing
a pointer to it. Calling it would provide the necessary type information to the compiler via the actual
parameters used, but a pointer to a function does not contain parameter information. The compiler needs
the assistance of this explicit template instantiation.
The call to swap within selection_sort is actually invoking the std::swap function from the
standard library. The #include <algorithms> directive brings in the declaration for std::swap,
which itself is a template function.
As Listing 18.4 (templatescale.cpp) shows, a template can accept non-type parameters.

©2013 Richard L. Halterman Draft date: January 27, 2013

 18.2. CLASS TEMPLATES 516

Listing 18.4: templatescale.cpp

1 #include <iostream>
2 using namespace std;
3
4 template <int N>
5 int scale(int value)
6 {
7 return value * N;
8 }
9
10 int main()
11 {
12 cout << scale<3>(5) << endl;
13 cout << scale<4>(10) << endl;
14 }

Note that in Listing 18.4 (templatescale.cpp) the typename keyword is missing, and in its place is the
actual type int. This means parameter N is a non-type template parameter.
The types of non-type template parameters permitted by C++ are very limited. Integral types (int,
unsigned, long, etc.) are acceptable, as are pointers and references. Other types such as floating-point
and programmer-defined types may not be non-type template parameters.

18.2 Class Templates

CAUTION! SECTION UNDER CONSTRUCTION

Recall the IntList linked list class from Listings 16.6 and 16.7. It implemented a linked list of
integer elements. Suppose we instead would like to make a linked list of std::string objects. Without
C++ templates, we would have rewrite the code to make a StringList class. Note that the code in the
IntList and StringList classes would be almost identical; the only major difference would be the
type of elements the list can hold. C++ provides a class template feature that allows us to build generic
container classes. Listing 18.5 (llist.h) which defines a such a generic linked list in its LList class.

Listing 18.5: llist.h

1 // llist.h
2 #include <iostream>
3
4 template <typename T>
5 class LList
6 {
7 /*
8 * An object that holds an element in a linked list.
9 * This type is local to the IntList class and inaccessible
10 * to outside this class.
11 */
12 struct Node
13 {
14 T data; // A data element of the list
15 Node *next; // The node that follows this one in the list
16 // Constructor

©2013 Richard L. Halterman Draft date: January 27, 2013

 18.2. CLASS TEMPLATES 517

17 Node(const T& d): data(d), next(nullptr) {}

18 };
19
20 Node *head; // Points to the first item in the list
21
22 // Private helper methods
23
24 /*
25 * insert(p, elem)
26 * Inserts elem onto the back of the
27 * list pointed to by p.
28 * p points to a node in a linked list
29 * elem is the element to insert.
30 * Returns a pointer to the list that
31 * contains elem.
32 */
33 Node *insert(Node *p, const T& elem)
34 {
35 if ( p ) // Is p non-null?
36 p->next = insert(p->next, elem);
37 else // p is null, make a new node
38 p = new LList::Node(elem);
39 return p;
40 }
41
42 /*
43 * print(p)
44 * Prints the contents of a linked list of integers.
45 * p points to a linked list of nodes.
46 */
47 void print(Node *p) const
48 {
49 while ( p ) // While p is not null
50 {
51 std::cout << p->data << " "; // Print current node's data
52 p = p->next; // Move to next node
53 }
54 std::cout << std::endl;
55 }
56
57
58 /*
59 * Returns the length of the linked list pointed to by p.
60 */
61 int length(Node *p) const
62 {
63 if ( p )
64 return 1 + length(p->next); // 1 + length of rest of list
65 else
66 return 0; // Empty list has length zero
67 }
68
69 /*
70 * dispose(p)
71 * Deallocate the memory held by the list pointed to by p.

©2013 Richard L. Halterman Draft date: January 27, 2013

 18.2. CLASS TEMPLATES 518

72 */
73 void dispose(Node *p)
74 {
75 if ( p )
76 {
77 dispose(p->next); // Free up the rest of the list
78 delete p; // Deallocate this node
79 }
80 }
81
82 public:
83 /*
84 * The constructor makes an initially empty list
85 */
86 LList(): head(nullptr) {}
87
88 /*
89 * The destructor frees up the memory held by the list
90 */
91 ∼LList()
92 {
93 clear(); // Deallocate space held by the list nodes
94 }
95
96 /*
97 * insert(elem)
98 * Inserts elem onto the back of the list.
99 * elem is the element to insert.
100 * /
101 void insert(const T& elem)
102 {
103 head = insert(head, elem); // Delegate work to private helper method
104 }
105
106 /*
107 * print()
108 * Prints the contents of the linked list of integers.
109 */
110 void print() const
111 {
112 print(head); // Delegate work to private helper method
113 }
114
115 /*
116 * Returns the length of the linked list.
117 */
118 int length() const
119 {
120 return length(head); // Delegate work to private helper method
121 }
122
123 /*
124 * clear()
125 * Removes all the elements in the linked list.
126 */

©2013 Richard L. Halterman Draft date: January 27, 2013

 18.2. CLASS TEMPLATES 519

127 void clear()

128 {
129 dispose(head); // Deallocate space for all the nodes
130 head = nullptr; // Null head signifies list is empty
131 }
132 };

The code in Listing 18.6 (templatelistmain.cpp) contains only trivial changes to just two lines of the
code from Listing 16.8 (listmain.cpp).

Listing 18.6: templatelistmain.cpp

1 // templatelistmain.cpp
2
3 #include "llist.h" // Was ==> #include "intlist.h"
4 #include <iostream>
5
6 using namespace std;
7
8 int main()
9 {
10 bool done = false;
11 char command;
12 int value;
13 LList<int> list; // Was ==> IntList list;
14
15 while ( !done )
16 {
17 cout << "I)nsert <item> P)rint L)ength E)rase Q)uit >>";
18 cin >> command;
19 switch ( command )
20 {
21 case 'I': // Insert a new element into the list
22 case 'i':
23 if ( cin >> value )
24 list.insert(value);
25 else
26 done = true;
27 break;
28 case 'P': // Print the contents of the list
29 case 'p':
30 list.print();
31 break;
32 case 'L': // Print the list's length
33 case 'l':
34 cout << "Number of elements: " << list.length() << endl;
35 break;
36 case 'E': // Erase the list
37 case 'e':
38 list.clear();
39 break;
40 case 'Q': // Exit the loop (and the program)
41 case 'q':
42 done = true;
43 break;

©2013 Richard L. Halterman Draft date: January 27, 2013

 18.2. CLASS TEMPLATES 520

44 }
45 }
46 }

Listing 18.7 (genericloggingflexiblesort.cpp) uses template functions and template classes to make our
logging flexible sort code truly generic. It also uses inheritance to derive a class from a template class.

Listing 18.7: genericloggingflexiblesort.cpp

1 #include <iostream>
2 #include <vector>
3 #include <cstdlib>
4 #include <fstream>
5 #include <string>
6
7 using namespace std;
8
9 /*
10 * Comparer objects manage the comparisons and element
11 * interchanges on the selection sort function below.
12 */
13 template <typename T>
14 class Comparer
15 {
16 // The object's data is private, so it is inaccessible to
17 // clients and derived classes
18
19 // Keeps track of the number of comparisons
20 // performed
21 int compare_count;
22
23 // Keeps track of the number of swaps performed
24 int swap_count;
25
26 // Function pointer directed to the function to
27 // perform the comparison
28 bool (*comp)(const T&, const T&);
29
30 protected:
31 // Method that actually performs the comparison
32 // Derived classes may override this method
33 virtual bool compare_impl(const T& m, const T& n);
34
35 // Method that actually performs the swap
36 // Derived classes may override this method
37 virtual void swap_impl(T& m, T& n);
38
39 public:
40 // The client must initialize a Comparer object with a
41 // suitable comparison function.
42 Comparer(bool (*f)(const T&, const T&));
43
44 // Resets the counters to make ready for a new sort
45 void reset();
46

©2013 Richard L. Halterman Draft date: January 27, 2013

 18.2. CLASS TEMPLATES 521

47 // Method that performs the comparison. It delegates

48 // the actual work to the function pointed to by comp.
49 // This method logs each invocation.
50 bool compare(const T& m, const T& n);
51
52 // Method that performs the swap.
53 // Interchange the values of
54 // its parameters a and b which are
55 // passed by reference.
56 // This method logs each invocation.
57 void swap(T& m, T& n);
58
59 // Returns the number of comparisons this object has
60 // performed since it was created.
61 int comparisons() const;
62
63 // Returns the number of swaps this object has
64 // performed since it was created.
65 int swaps() const;
66 };
67
68
69 // Method that actually performs the comparison
70 // Derived classes may override this method
71 template <typename T>
72 bool Comparer<T>::compare_impl(const T& m, const T& n)
73 {
74 return comp(m, n);
75 }
76
77 // Method that actually performs the swap
78 // Derived classes may override this method
79 template <typename T>
80 void Comparer<T>::swap_impl(T& m, T& n)
81 {
82 T temp = m;
83 m = n;
84 n = temp;
85 }
86
87 // The client must initialize a Comparer object with a
88 // suitable comparison function.
89 template <typename T>
90 Comparer<T>::Comparer(bool (*f)(const T&, const T&)):
91 compare_count(0), swap_count(0), comp(f) {}
92
93 // Resets the counters to make ready for a new sort
94 template <typename T>
95 void Comparer<T>::reset()
96 {
97 compare_count = swap_count = 0;
98 }
99
100 // Method that performs the comparison. It delegates
101 // the actual work to the function pointed to by comp.

©2013 Richard L. Halterman Draft date: January 27, 2013

 18.2. CLASS TEMPLATES 522

102 // This method logs each invocation.

103 template <typename T>
104 bool Comparer<T>::compare(const T& m, const T& n)
105 {
106 compare_count++;
107 return compare_impl(m, n);
108 }
109
110 // Method that performs the swap.
111 // Interchange the values of
112 // its parameters a and b which are
113 // passed by reference.
114 // This method logs each invocation.
115 template <typename T>
116 void Comparer<T>::swap(T& m, T& n)
117 {
118 swap_count++;
119 swap_impl(m, n);
120 }
121
122 // Returns the number of comparisons this object has
123 // performed since it was created.
124 template <typename T>
125 int Comparer<T>::comparisons() const
126 {
127 return compare_count;
128 }
129
130 // Returns the number of swaps this object has
131 // performed since it was created.
132 template <typename T>
133 int Comparer<T>::swaps() const
134 {
135 return swap_count;
136 }
137
138
139 /*
140 * selection_sort(a, compare)
141 * Arranges the elements of vector a in an order determined
142 * by the compare object.
143 * a is a generic vector.
144 * compare is a function that compares the ordering of
145 * two integers.
146 * The contents of a are physically rearranged.
147 */
148 template <typename T>
149 void selection_sort(vector<T>& a, Comparer<T>& compare)
150 {
151 int n = a.size();
152 for ( int i = 0; i < n - 1; i++ )
153 {
154 // Note: i,small, and j represent positions within a
155 // a[i], a[small], and a[j] represents the elements at
156 // those positions.

©2013 Richard L. Halterman Draft date: January 27, 2013

 18.2. CLASS TEMPLATES 523

157 // small is the position of the smallest value we've seen

158 // so far; we use it to find the smallest value less
159 // than a[i]
160 int small = i;
161 // See if a smaller value can be found later in the array
162 for ( int j = i + 1; j < n; j++ )
163 if ( compare.compare(a[j], a[small]) )
164 small = j; // Found a smaller value
165 // Swap a[i] and a[small], if a smaller value was found
166 if ( i != small )
167 compare.swap(a[i], a[small]);
168 }
169 }
170
171 /*
172 * print
173 * Prints the contents of an integer vector
174 * a is the vector to print.
175 * a is not modified.
176 */
177 template <typename T>
178 void print(const vector<T>& a)
179 {
180 int n = a.size();
181 cout << '{';
182 if ( n > 0 )
183 {
184 cout << a[0]; // Print the first element
185 for ( int i = 1; i < n; i++ )
186 cout << ',' << a[i]; // Print the rest
187 }
188 cout << '}';
189 }
190
191 /*
192 * less_than(a, b)
193 * Returns true if a < b; otherwise, returns
194 * false.
195 */
196 template <typename T>
197 bool less_than(const T& a, const T& b)
198 {
199 return a < b;
200 }
201
202 /*
203 * greater_than(a, b)
204 * Returns true if a > b; otherwise, returns
205 * false.
206 */
207 template <typename T>
208 bool greater_than(const T& a, const T& b)
209 {
210 return a > b;
211 }

©2013 Richard L. Halterman Draft date: January 27, 2013

 18.2. CLASS TEMPLATES 524

212
213
214 /*
215 * Comparer objects manage the comparisons and element
216 * interchanges on the selection sort function below.
217 */
218 template <typename T>
219 class LogComparer: public Comparer<T>
220 {
221 // Output stream to which logging messages are directed
222 std::ofstream fout;
223
224 protected:
225 // Method that actually performs the comparison
226 bool compare_impl(const T& m, const T& n) override;
227
228 // Method that actually performs the swap
229 void swap_impl(T& m, T& n) override;
230
231 public:
232 // The client must initialize a LogComparer object with a
233 // suitable comparison function and the file name of a text
234 // file to which the object will direct logging messages
235 LogComparer(bool (*f)(const T&, const T&),
236 const std::string& filename);
237
238 // The destructor must close the log file
239 ∼LogComparer();
240 };
241
242
243
244 // Method that actually performs the comparison
245 // Derived classes may override this method
246 template <typename T>
247 bool LogComparer<T>::compare_impl(const T& m, const T& n)
248 {
249 fout << "Comparing " << m << " to " << n << std::endl;
250 // Base class method does the comparision
251 return Comparer::compare_impl(m, n);
252 }
253
254 // Method that actually performs the swap
255 // Derived classes may override this method
256 template <typename T>
257 void LogComparer<T>::swap_impl(T& m, T& n)
258 {
259 fout << "Swapping " << m << " and " << n << endl;
260 T temp = m;
261 m = n;
262 n = temp;
263 }
264
265 // The client must initialize a LogComparer object with a
266 // suitable comparison function and the file name of the

©2013 Richard L. Halterman Draft date: January 27, 2013

 18.2. CLASS TEMPLATES 525

267 // text file to receive logging messages.

268 template <typename T>
269 LogComparer<T>::LogComparer(bool (*f)(const T&, const T&),
270 const string& filename): Comparer<T>(f)
271 {
272 fout.open(filename.c_str());
273 if ( !fout.good() )
274 {
275 fout.close();
276 cout << "Could not open log file " << filename
277 << " for writing" << endl;
278 exit(1); // Terminate the program
279 }
280 }
281
282 // Destructor closes the log file
283 template <typename T>
284 LogComparer<T>::∼LogComparer()
285 {
286 fout.close();
287 }
288
289
290 int main()
291 {
292 // Make a vector of integers from an array
293 int a[] = { 23, -3, 4, 215, 0, -3, 2 };
294 int len = (sizeof a)/(sizeof a[0]);
295 vector<int> vec(a, a + len);
296
297 // Make a working copy of the original vector
298 cout << "Before: ";
299 print(vec);
300 cout << endl;
301 LogComparer<int> lt(less_than<int>, "upsort.log");
302 selection_sort(vec, lt);
303 cout << "Ascending: ";
304 print(vec);
305 cout << " (" << lt.comparisons() << " comparisons, "
306 << lt.swaps() << " swaps)" << endl;
307 LogComparer<int> gt(greater_than<int>, "downsort.log");
308 selection_sort(vec, gt);
309 cout << "Ascending: ";
310 print(vec);
311 cout << " (" << gt.comparisons() << " comparisons, "
312 << gt.swaps() << " swaps)" << endl;
313
314 cout << "--------------------------------------" << endl;
315
316 // Make a vector of string objects from an array
317 string words[] = { "tree", "girl", "boy", "apple", "dog", "cat", "bird" };
318 len = (sizeof words)/(sizeof words[0]);
319
320 vector<string> w(words, words + len);
321

©2013 Richard L. Halterman Draft date: January 27, 2013

 18.3. THE STANDARD TEMPLATE LIBRARY 526

322 // Make a working copy of the original vector

323 cout << "Before: ";
324 print(w);
325 cout << endl;
326 LogComparer<string> wlt(less_than<string>, "upwords.log");
327 selection_sort(w, wlt);
328 cout << "Ascending: ";
329 print(w);
330 cout << " (" << wlt.comparisons() << " comparisons, "
331 << wlt.swaps() << " swaps)" << endl;
332 LogComparer<string> wgt(greater_than<string>, "downwords.log");
333 selection_sort(w, wgt);
334 cout << "Ascending: ";
335 print(w);
336 cout << " (" << wgt.comparisons() << " comparisons, "
337 << wgt.swaps() << " swaps)" << endl;
338 }

The code in Listing 18.7 (genericloggingflexiblesort.cpp) truly is generic:

• Clients may customize the element ordering.

• Clients may customize the behavior of the comparison and swapping procedures.
• Clients may process vectors that contain elements of any type that supports the < and = operators.

18.3 The Standard Template Library

CAUTION! SECTION UNDER CONSTRUCTION

The C++ Standard Template Library, or STL, contains a number of generic functions and classes. We
have seen the std::vector class and std::swap function. Our LList class from Listing 18.5 (llist.h)
is really re-inventing the wheel because std::List implements a doubly-linked list with pointers con-
necting a node to its successor and pointers connecting the node to its predecessor.

CAUTION! SECTION UNDER CONSTRUCTION

18.4 Iterators

CAUTION! SECTION UNDER CONSTRUCTION

An iterator is an object that behaves like a pointer. As such it overloads operator* and operator++,
and some iterator types provide additional pointer arithmetic capabilities. An STL container class such as
vector or list define their own iterator types that enable clients to traverse their elements as easily as
using a pointer to traverse an array.
Listing 18.8 (veciter.cpp) uses an iterator to print the elements of a vector of integers.

Listing 18.8: veciter.cpp

©2013 Richard L. Halterman Draft date: January 27, 2013

 18.4. ITERATORS 527

1 #include <iostream>
2 #include <vector>
3 #include <cstdlib>
4
5 int main()
6 {
7 std::vector<int> vec;
8 // Populate the vector
9 for ( int i = 0; i < 10; i++ )
10 vec.push_back(rand() % 100);
11 // Print the elements
12 for ( vector<int>::iterator iter = vec.begin();
13 iter != vec.end(); iter++ )
14 cout << *iter << ' ';
15 cout << endl;
16 }

The instantiated vector<int> class has a public nested type (a class) named iterator. This is similar
to our LList::Node struct in Listing 18.5 (llist.h)—the Iterator class is defined within the defi-
nition of the vector class. The iterator class, unlike our Node struct, is public. The expression
vector<int>::iterator represents a type name, with which we can declare variables; for example:
vector<int>::iterator myIter;

As we see in Listing 18.8 (veciter.cpp), a client may obtain an iterator object from a particular vector by
calling the vector’s begin or end method. The begin method returns an iterator that “points” to the first
element in the vector, while the end method returns an iterator that correcponds to the position just past
the end of the vector.
The type name vector<int>::iterator is somewhat unwieldy, and we can use the auto key-
word to simplify the code as shown in Listing 18.9 (veciter2.cpp).

Listing 18.9: veciter2.cpp

1 #include <iostream>
2 #include <vector>
3 #include <cstdlib>
4
5 int main()
6 {
7 std::vector<int> vec;
8 // Populate the vector
9 for ( int i = 0; i < 10; i++ )
10 vec.push_back(rand() % 100);
11 // Print the elements
12 for ( auto iter = vec.begin(); iter != vec.end(); iter++ )
13 cout << *iter << ' ';
14 cout << endl;
15 }

Iterators make range-based for loops possible. Listing 18.10 (vecrangefor.cpp) uses a range-based
for loop to simplify our vector printing code even more.

Listing 18.10: vecrangefor.cpp

©2013 Richard L. Halterman Draft date: January 27, 2013

 18.5. SUMMARY 528

1 #include <iostream>
2 #include <vector>
3 #include <cstdlib>
4
5 int main()
6 {
7 std::vector<int> vec;
8 // Populate the vector
9 for ( int i = 0; i < 10; i++ )
10 vec.push_back(rand() % 100);
11 // Print the elements
12 for ( auto elem : vec )
13 cout << elem << ' ';
14 cout << endl;
15 }

Without iterator objects, we would be unable to traverse a vector using the range-based for loop as shown
in Listing 18.10 (vecrangefor.cpp).
Recall from the technique of using array ranges to traverse the contents of an array. Rather than pass-
ing an array (literally the address of the first element) and its size (number of elements), the range-based
approach passes the array (again, literally a pointer to the first element) and a pointer to position just past
the end of the array. Notice how the pointers in range-based array technique correspond directly to the it-
erator objects returned by the vector begin and end methods. Since iterator objects behave like pointers,
we can write truly generic code that can process arrays and vectors (and any objects that provide pointers.
CAUTION! SECTION UNDER CONSTRUCTION

18.5 Summary

CAUTION! SECTION UNDER CONSTRUCTION

• Add item here.

18.6 Exercises

CAUTION! SECTION UNDER CONSTRUCTION

1. Add item here.

©2013 Richard L. Halterman Draft date: January 27, 2013

 529

Chapter 19

Exception Handling

CAUTION! CHAPTER UNDER CONSTRUCTION

19.1 Summary

CAUTION! SECTION UNDER CONSTRUCTION

• Add item here.

19.2 Exercises

CAUTION! SECTION UNDER CONSTRUCTION

1. Add item here.

©2013 Richard L. Halterman Draft date: January 27, 2013

19.2. EXERCISES 530

©2013 Richard L. Halterman Draft date: January 27, 2013

 531

Appendices

©2013 Richard L. Halterman Draft date: January 27, 2013

 533

Appendix A

Visual C++ Command Line

Development

Visual Studio provides command line tools as an alternative to its integrated development environment.
Some developers prefer the command line tools over the graphical IDE due to their simplicity, especially
when developing simple console applications. The command line tools can be scripted to automate building
and testing programs.
To use the command line tools, select the Visual Studio Command Prompt from the Start menu. A
command console appears similar to the console that the IDE provides when executing a program (see
Section 2.2). Instead of selecting menu items and interacting with dialog boxes, you type commands into the
console window to build and run your program. To edit your program you can use the editor from the IDE,
or you can use a standalone editor such as Notepad++ (http://www.notepad-plus.sourceforge.net)
or Vim (http://www.vim.org). Both Notepad++ and Vim provide color syntax highlighting and folding
like the built-in Visual Studio editor.
In order to use the command prompt version of Visual Studio, it is convenient to first create a folder in
which you will place all your development files; for example, you could create a folder named Code under
your Documents folder. Suppose the full path to this folder is
C:\Users\rick\Documents\Code

Once your development folder is created, you can launch the Visual Studio Command Prompt.
The first command you should issue in the console window is

C:

to ensure you are working on the correct drive. Next issue

cd \Users\rick\Documents\Code

©2013 Richard L. Halterman Draft date: January 27, 2013

 534

The cd command means “change directory.” This command sets the console’s working directory to the
development folder you previously created. If you are working in another folder, you can adjust the path
given in the cd command to your work area.
You are ready to edit your code. If your environment is set up so that your editor program is in the
system path, you can type

notepad++ simple.cpp

If you are using Vim, type

gvim simple.cpp

If the file simple.cpp already exists, you can type

devenv simple.cpp

to launch the Visual Studio editor.

Within the editor type in your program and then save it to a file.
When you have finished creating the source file, we will need to build the executable program. Switch
back to the Visual Studio Command Prompt window. and issue the command

cl /W3 /EHsc /Za simple.cpp

The cl (“cee elle”, not “cee one”) command, which stands for compile and link, preprocesses, compiles,
and links the program. The /W3 switch uses the Level 3 warning level. Level 3 warnings provide more
thorough analysis of your source code and can detect more programming mistakes. If you omit the /W3
switch, by default the compiler operates at warning level 1. Warning level 1 is insufficient for catching
many mistakes made by both beginning and experienced C++ programmers. By default the IDE compiler
uses warning level 3, and the warning level can be changed by adjusting the project’s properties. On the
command line, specifying /W4 (the highest warning level) is just as easy as specifying /W3. It is better to
use /W4 so the compiler can do a more thorough check for potential programmer mistakes. The /EHsc
is required for all the C++ programs we write. The /Za switch makes sure the compiler complies with
standard C++ language rules and does not allow any Microsoft-specific language extensions.
If want to reduce the level of detail in the output produced by the compiler and linker, you can add the
additional switch /nologo:

cl /W4 /EHsc /Za /nologo simple.cpp

©2013 Richard L. Halterman Draft date: January 27, 2013

 535

Notice that we used the enhanced warning level (/W4) here. To see the complete list of compiler options,
many more than you probably will ever need, type

cl /?

In C++ programs consisting of a single source file, the compiler produces an executable file with the
same base name as the source file (simple, in our example), with an exe extension.
Once the program has been compiled successfully, you can run the program by entering the name of the
executable produced. In our example, we would issue the command

simple

The program runs in the same window; no new console window is produced. When the program is finished
executing, no “Press any key to continue” message is printed; you get simply a command prompt ready for
you to enter your next command.
If you modify your source code in your editor, you must remember to recompile your code before you
run it again. In the IDE, if you modify your code, the environment will suggest that you rebuild the project
before running the program.

If the program consists of multiple source files, list all the required files on the
command line. The exe file will be named after the first file listed; for example,
the command

cl /W4 /EHsc /Za /nologo app.cpp list.cpp util.cpp

compiles the three C++ sources files and links the compiled code together to pro-
duce app.exe.

In the Visual Studio command prompt program you can enter the Visual Studio IDE at any time with
the command

devenv

(devenv stands for “development environment.”) If you started Visual Studio in the IDE, not the command
prompt, you can open a command prompt console window at any time by selecting the Command Prompt
item under the Tools menu.

©2013 Richard L. Halterman Draft date: January 27, 2013

 536

©2013 Richard L. Halterman Draft date: January 27, 2013

 537

Appendix B

Developing C++ Programs under

Unix/Linux with the GNU Tools

The tools provided in the GNU Compiler Collection (http://gcc.gnu.org), or GCC for short include
one of the most up-to-date standards compliant C++ compilers available. The GNU tools are free and are
available for the major computing platforms:

• Microsoft Windows (XP, 7, and 8)

Windows-based GCC tools can be found at the MinGW website, http://www.mingw.org.
• Apple Macintosh OS X
Mac-based GCC tools are a part of Apple’s XCode development system which is available from
Apple’s Developer Connection (http://developer.apple.com/tools/download/).
• Linux
Linux-based GCC tools are readily available for all the major Linux distributions. To get them,
simply install the g++ development packages.

The Eclipse Foundation’s CDT (C Development Tools) project (http://www.eclipse.org/cdt).

provides a cross-platform IDE that uses the GCC tools for compiling C++ programs. The Eclipse IDE
runs on Windows, Mac, and Linux.
After using an editor to create a text file containing the code for Listing 2.1 (simple.cpp), save the file
with the name simple.cpp. To create the executable program, issue the command

g++ -Wall -O1 -std=c++11 -o simple simple.cpp

This command has multiple parts:

• g++
This part of the command invokes the GNU C++ build process. This one command runs the pre-
processor, followed by the compiler, and finally the linker to produce the executable program. Even

©2013 Richard L. Halterman Draft date: January 27, 2013

 538

though the preprocessor and linker are also involved here, we generally simply say we are “compiling
the program.”
• -Wall -O1
This part of the command indicates that the compiler should be very strict when it is compiling the
source code. It goes beyond checking that the program is well-formed C++ it also warns about code
constructs that many programmers consider questionable because code of that nature often contains
errors. The -Wall component can be interpreted to mean “warn about as many possible problems
as possible.” The -O1 component (capital oh one, not zero elle) instructs the compiler to perform
some additional analysis and optimization. All C++ programs that you write should be compiled with
these directives because they can catch many careless programmer mistakes that otherwise might go
unnoticed. Use of the -Wall -O1 directives is desirable for all C++ programmers from novice to
expert.
• -c++11
This part of the command indicates that the compiler should conform to the C++ language rules as
specified in the C++11 standard. If this component of the command is missing, the compiler defaults
to C++03 rules, and will reject some newer C++ language features such as type inference with auto
and the enhanced for loop for arrays and vectors.
• -o simple
This part of the command specifies the name of the executable program. The name appears after the
-o part, so our executable program will be named simple. On Microsoft Windows systems the name
of the file produced will be simple.exe. If there are no errors in the source code, the compiler will
produce an executable file named simple (or simple.exe). It can be executed from the command line
with the command

./simple

If you omit this part, the executable will be named a.out under Unix and OS X or a.exe under Win-
dows.

• simple.cpp
This part of the command indicates the C++ source file we wish to compile.

The general form of the command to compile a C++ source file, therefore, is

g++ -Wall -O1 -std=c++11 -o executable source

where executable is the name of the executable program that the compiler is to produce, and source is the
name of the file containing the C++ source code.

©2013 Richard L. Halterman Draft date: January 27, 2013

 539

Appendix C

Introduction to Binary Numbers and

Computer Arithmetic

This appendix provides some additional details about the internal representation of C++’s numeric types.

C.1 The Integer Types

All data are stored internally in binary form. The binary number system is much simpler than the familiar
decimal (base 10) number system because it uses only two digits: 0 and 1. The decimal system uses 10
digits: 0, 1, 2, 3, 4, 5, 6, 7, 8, and 9. Despite the lack of digits, every decimal integer has an equivalent
binary representation. Binary numbers use a place value system not unlike the decimal system. Figure C.1
shows how the familiar base 10 place value system works.

··· 4 7 3 4 0 6
··· 105 104 103 102 101 100
· · · 100,000 10,000 1,000 100 10 1

473, 406 = 4 × 105 + 7 × 104 + 3 × 103 + 4 × 102 + 0 × 101 + 6 × 100

= 400, 000 + 70, 000 + 3, 000 + 400 + 0 + 6
= 473, 406

Figure C.1: The base 10 place value system

With 10 digits to work with, the decimal number system distinguishes place values with powers of 10.
Compare the base 10 system to the base 2 place value system shown in Figure C.2.
Since both binary and decimal numbers share the digits 0 and 1, we will use a subscript of 2 to represent
binary numbers; therefore, 100 represents the decimal value one hundred, while 1002 is the binary number
four. Sometimes to be very clear we will attach a subscript of 10 to a decimal number, as in 10010 .

©2013 Richard L. Halterman Draft date: January 27, 2013

C.1. THE INTEGER TYPES 540

··· 1 0 0 1 1 1
··· 25 24 23 22 21 20
··· 32 16 8 4 2 1

1001112 = 1 × 25 + 0 × 24 + 0 × 23 + 1 × 22 + 1 × 21 + 1 × 20
= 32 + 0 + 0 + 4 + 2 + 1
= 39

Figure C.2: The base 2 place value system

With only two digits to work with, the binary number system distinguishes place values by powers of
two.
In the decimal system, it is easy to add 3 + 8:

3
+ 5
8

The sum 3 + 9 is a little more complicated, as early elementary students soon discover:

3
+ 9

The answer, of course, is 12, but there is no single digit that means 12—it takes two digits, 1 and 2. The
sum is
1
03
+ 09
12
We can say 3 + 9 is 2, carry the 1. The rules for adding binary numbers are shorter and simpler than decimal
numnbers:
02 + 02 = 02
02 + 12 = 12
12 + 02 = 12
12 + 12 = 102
Within a sum 12 + 12 is 02 , carry the 12 . A typical larger sum would be

11
910 = 10012
+ 310 = 112
1210 = 11002

Integers are whole numbers (no fractional parts), both positive and negative. The standard C++ integer
types include int, long, long long, and short. These are distinguished by the number of bits
required to store the type, and, consequently, the range of values that can be represented. Mathematical
integers are infinite, but all of C++’s integer types correspond to finite subsets of mathematical integers. The

©2013 Richard L. Halterman Draft date: January 27, 2013

C.1. THE INTEGER TYPES 541

most commonly used integer type in C++ is int. All ints, regardless of their values, occupy the same
amount of memory and, therefore use the same number of bits. The exact number of bits in an int is
processor specific. A 32-bit processor, for example, is built to manipulate 32-bit integers very efficiently. A
C++ compiler for such a system most likely would use 32-bit ints, while a compiler for a 64-bit machine
would represent ints with 64 bits. On a 32-bit computer, the numbers 4 and 1,320,002,912 both occupy
32 bits of memory.
At the lower level computers deal exclusively with bits, so the sign of an integer must somehow be
encoded as a bit. Two’s complement binary is a specialized binary encoding used by most computer systems
today. This format uses the leftmost bit to indicate the sign of the value: 0 for non-negative and 1 for
negative. To better understand how the integer types work, consider the mythical type nibble (a nibble
is defined formally as four bits, or one-half byte). C++ has no such nibble type, but a nibble’s smaller
range of values makes it easier to explore the properties of integers in a computer system. A nibble uses
four bits of storage, and Table C.1 shows the values that can be represented using only four bits in the two’s
complement binary representation.

Decimal Value Binary Bit String

−8 1000
−7 1001
−6 1010
−5 1011
−4 1100
−3 1101
−2 1110
−1 1111
0 0000
1 0001
2 0010
3 0011
4 0100
5 0101
6 0110
7 0111

Table C.1: The nibble values

The negative values have a one in the leftmost bit position. Notice that there are eight negative numbers
but only seven positive numbers. Using this two’s complement representation, there always will be one
more negative value, regardless of the number of bits used. This is true for all of C++’s integer types since
all commonly used computing systems today use two’s complement binary integer representations.
The bit patterns for the positive numbers make sense; for example:
0110
=⇒ 0 × 8 + 1 × 4 + 1 × 2 + 0 × 1 = 6
8421
The bit patterns for negative numbers appear to be incorrect, but this is because they are represented in
a special way that enables arithmetic to work out correctly. The following steps determine the value of a
two’s complement binary integer:

1. Once you note that the leftmost bit is a one, forget it. It is used to indicate a negative number, but it
is not used to encode its value. For example, consider −5 = 1011. We consider only the bits 011.

©2013 Richard L. Halterman Draft date: January 27, 2013

C.1. THE INTEGER TYPES 542

2. Invert all the remaining bits. This means change all the zeros to ones and all the ones to zeros. 011
becomes 100.
3. Add one to the result:
100
+ 1
101
Since 1012 = 510 , the original two’s complement bit pattern is interpreted as −5.

To see that arithmetic works, try adding 4 + (−7):

410 = 01002
+ −710 = 10012
−310 = 11012
or −3 + 6:
−310 = 11012
+ 610 = 01102
310 = 1 00112
The extra one on the left results from a carry (12 + 12 = 102 ), but five digits cannot be stored in a nibble.
Only the rightmost four digits remain, and they hold the correct answer.
Two important observations:

1. The fact that there are eight 4-bit values that begin with a zero also means that there are eight 4-
bit values that begin with a one (negative values). Since zero begins with a zero (0000) and there
are seven values left that are positive, this explains why there is always one more negative value
than positive values in all the integer types (for example, 32-bit ints range from −2, 147, 483, 648
. . . +2, 147, 483, 647).
2. Seven is the largest nibble. The addition 7 + 1:
0111
+ 0001
1000

results in a nibble that has a one in the first bit position—the value −8!. This explains why
increasing the value of an integer variable can cause it to have a negative value. The largest possible
32-bit int is +2, 147, 483, 647. Add one to +2, 147, 483, 647 and the result is −2, 147, 483, 648, the
smallest possible 32-bit integer. Thus, the smallest value for a given integer type immediately follows
the largest value for that type. In fact, 32-bit two’s complement ints follow the cyclic pattern shown
in Figure C.3. In the figure, an addition moves a value clockwise around the circle, while a subtraction
moves a value counterclockwise around the circle.

Item 1 is more of an intellectual curiosity, but Item 2 has significant consequences for unwary programmers.
Attempting to exceed the maximum limit of a numeric type results in overflow, and attempting to exceed
the minimum limit is called underflow. Integer arithmetic that overflows or underflows produces a valid,
yet incorrect integer result. The compiler does not check that a computation will result in exceeding the
limit of a type because it is impossible to do so in general (consider adding two integer variables whose
values are determined at run time). Also significantly, an overflow or underflow situation does not generate
a run-time error. It is therefore a logic error if a program performs an integral computation that, either as a
final result or an intermediate value, is outside the range of the integer type being used.
When literal numeric values are used in a C++ program, as in

©2013 Richard L. Halterman Draft date: January 27, 2013

C.1. THE INTEGER TYPES 543

‒1 0
1
‒2 2

‒3 3

4
‒4 4

Subtract Add

‒2,147,483,645 2,147,483,645

‒2,147,483,646 2,147,483,646

‒2,147,483,647 2,147,483,647

‒2,147,483,648

Figure C.3: The cyclic nature of two’s complement integers

x = 2;

the number 2 has type int. To treat it as a long, append the letter L suffix:

x = 2; // 2 is an int
x = 2L; // 2 is a long

Lower case elle letter (l) also can be used, but it looks too much like the digit 1 (one) on many displays
and printer fonts, so L is preferable.

All integer literals are int values unless an l or L is appended to the end of the
number; for example, 2 is an int literal, while 2L is a long literal.

©2013 Richard L. Halterman Draft date: January 27, 2013

C.2. THE FLOATING-POINT TYPES 544

C.2 The Floating-Point Types

Floating point types consist of float and double. Floating point numbers can have fractional parts
(decimal places), and the term floating point refers to the fact the decimal point in a number can float
left or right as necessary as the result of a calculation (for example, 2.5 × 3.3 = 8.25, two one-decimal
place values produce a two-decimal place result). As with the integer types, the two floating-point types
are distinguished by the number of bits of storage required and corresponding range of values. The type
float stands for single-precision floating-point, and double stands for double-precision floating-point.
Floating point numbers serve as rough approximations of mathematical real numbers, but as we shall see,
they have some severe limitations compared to actual real numbers.
On most modern computer systems floating-point numbers are stored internally in exponential form
according to the standard adopted by the Institute for Electrical and Electronic Engineers (IEEE 754). In
the decimal system, scientific notation is the most familiar form of exponential notation:

One mole contains 6.023 × 1023 molecules.

Here 6.023 is called the mantissa, and 23 is the exponent.

In the IEEE 754 standard, binary exponential notation is used; that is, the mantissa and exponent are
binary numbers. Single-precision floating-point numbers (type float) are represented with 32 bits:

Mantissa 24 bits
Exponent 7 bits
Sign 1 bit

Total 32 bits

Double-precision floating-point numbers (type double) are represented with 64 bits:

Mantissa 52 bits
Exponent 11 bits
Sign 1 bit

Total 64 bits

The details of the IEEE 754 implementation are beyond the scope of this book, but a simplified example
serves to highlight the limitations of floating-point types in general. Recall the fractional place values in
the decimal system. The place values, from left to right, are

1 1 1 1
... 10, 000 1, 000 100 10 1 • 10 100 1000 10,000 ...
... 104 103 102 101 100 • 10−1 10−2 10−3 10−4 ...

Each place value is one-tenth the place value to its left. Move to the right, divide by ten; move to the
right, multiply by ten. In the binary system, the factor is two instead of ten:

1 1 1 1
... 16 8 4 2 1 • 2 4 8 16 ...
... 24 23 22 21 20 2 −1 2−2 2 −3 2−4 ...
•

As in our nibble example (see Section C.1), consider a a binary exponential number that consists of
fewer bits than either floats or doubles in C++. The mantissa consists of just three bits, all right of the

©2013 Richard L. Halterman Draft date: January 27, 2013

C.2. THE FLOATING-POINT TYPES 545

binary point. The base of the exponent is, of course, two, and a 2-bit exponent is used. Figure C.4 illustrates
such a value. For simplicity, neither the mantissa nor the exponent is negative. Thus, the mantissa ranges

. 21 20 21 2
1 0
1 0
22

−1 −2 −3

Figure C.4: Simplified binary exponential value

from .0002 (01 0) to .1112 , and the exponent ranges from 002 (01 0) to 112 (31 0). The number shown in
Figure C.4 is thus
 
−1 −2 −3 (1×21 +0×20 ) 1 1
(1 × 2 + 0 × 2 + 1 × 2 ) × 2 = + × 22
2 8
5
= ×4
8
= 2.5

The range of these abbreviated floating-point numbers is 0 . . . 7. This range is misleading, however,
since it leaves the impression that any value in between could be represented. 2.5 can be represented in
this scheme, but 2.4 cannot. This floating-point scheme is so simple that we can easily enumerate all the
possible values, as Table C.2 does.

Exponent Values
Mantissa Values 2002 (1) 2012 (2) 2102 (4) 2112 (8)
.0002 (0) 0 0 0 0
.0012 (0.125) 0.125 0.25 0.5 1
.0102 (0.25) 0.25 0.5 1 2
.0112 (0.375) 0.375 0.75 1.5 3
.1002 (0.5) 0.5 1 2 4
.1012 (0.625) 0.625 1.25 2.5 5
.1102 (0.75) 0.75 1.5 3 6
.1112 (0.875) 0.875 1.75 3.5 7

Table C.2: A complete enumeration of the simplified floating-point numbers. Decimal equivalents appear
in parentheses.

These values are plotted on the real number line in Figure C.5.
Several troubling issues are revealed by Table C.2 and Figure C.5:

1. There are many gaps; for example, the value 2.4 is missing and thus cannot be represented exactly
(2.5 is the closest approximation).

©2013 Richard L. Halterman Draft date: January 27, 2013

 C.2. THE FLOATING-POINT TYPES 546

0 1 2 3 4 5 6 7 8
8 eighths
8 fourths
8 halves
8 ones

Figure C.5: A plot of the simplified floating-point numbers on the real number line.

2. Some numbers are duplicated; for example, three different bit patterns represent the value 0.5. This
duplication limits the the number of different values that can be represented by a given number of
bits.
3. They are not uniformly dense. There are more values nearer to zero, and the numbers become sparse
farther away from zero.

The integer types discussed in Section C.1 exhibit none of these weaknesses. All integers in a given
range are present, no two bitstrings represent the same value, and the integers are uniformly distributed
across their specified range. The standard floating-point types provided by C++ use many more bits than
this simplified example, but the same problems shown here still exist to a much smaller degree. This is not
solely a problem of C++’s implementation of floating-point numbers; all computer languages that adhere to
the IEEE 754 standard exhibit these problems. Mathematical real numbers require an infinite amount of
storage to be represented exactly, but any computer’s memory is finite.
Listing C.1 (imprecisedifference.cpp) demonstrates the inexactness of floating-point arithmetic.

Listing C.1: imprecisedifference.cpp

1 #include <iostream>
2 #include <iomanip>
3
4 using namespace std;
5
6 int main()
7 {
8 double d1 = 2000.5;
9 double d2 = 2000.0;
10 cout << setprecision(16) << (d1 - d2) << endl;
11 double d3 = 2000.58;
12 double d4 = 2000.0;
13 cout << setprecision(16) << (d3 - d4) << endl;
14 }

The output of Listing C.1 (imprecisedifference.cpp) is:

0.5
0.5799999999999272

The first subtraction yields the correct answer. We now know that some floating-point numbers have exact

©2013 Richard L. Halterman Draft date: January 27, 2013

 C.2. THE FLOATING-POINT TYPES 547

internal representations while others are only approximations. The exact answer for the second subtraction
should be 0.58, and if we round the reported result to 13 decimal places, the answer matches. Floating-
point arithmetic often produces results that are close approximations of the true answer. What are the
ramifications for programmers of this inexactness of floating-point numbers? Section 9.4.6 shows how the
misuse of floating-point values can lead to logic errors in programs.
In C++ source code, literal numbers that contain decimal places are by default type double. A float
literals have an f or F suffix.

All floating-point literals are double values unless an f or F is appended to the

end of the number; for example, 2.0 is a double literal, while 2.0f is a float
literal.

In C++ a double literal may be assigned to a float variable. The compiler automatically narrows
the double to a float as Listing C.2 (assignfloat.cpp) shows:

Listing C.2: assignfloat.cpp

1 #include <iostream>
2
3 using namespace std;
4
5 int main()
6 {
7 float number;
8 number = 10.0; // OK, double literal assignable to a float
9 }

The statement
number = 10.0;

assigns a double literal (10.0) to a float variable. You instead may explicitly use a float literal as:
number = 10.0f;

©2013 Richard L. Halterman Draft date: January 27, 2013

C.2. THE FLOATING-POINT TYPES 548

©2013 Richard L. Halterman Draft date: January 27, 2013

 549

Appendix D

Bitwise Operators

In addition to the common arithmetic operators introduced in Chapter 4, C++ provides a few other special-
purpose arithmetic operators. These special operators allow programmers to examine or manipulate the
individual bits that make up data values. They are known as the bitwise operators. These operators consist
of &, |, ˆ, ∼ , >>, and <<. Applications programmers generally do not need to use bitwise operators very
often, but bit manipulation is essential in many systems programming tasks.
Consider a typical system which uses 32-bit integers. An integer on this system is a two’s complement
binary number (see Section C.1) containing 32 bits. The bit positions are usually numbered right to left,
starting with zero. Figure D.1 shows how the individual bit positions are often numbered.

31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 9 8 7 6 5 4 3 2 1 0

Figure D.1: The bit positions of a 32-bit C++ integer

The bitwise and operator, &, takes two integer subexpressions and computes an integer result. The
expression e1 & e2 is evaluated as follows:

If bit 0 in both e1 and e2 is 1, then bit 0 in the result is 1; otherwise, bit 0 in the result is 0.
If bit 1 in both e1 and e2 is 1, then bit 1 in the result is 1; otherwise, bit 1 in the result is 0.
If bit 2 in both e1 and e2 is 1, then bit 2 in the result is 1; otherwise, bit 2 in the result is 0.

..
.

If bit 31 in both e1 and e2 is 1, then bit 31 in the result is 1; otherwise, bit 31 in the result is 0.

For example, the expression 13 & 14 evaluates to 12, since:

1310 = 000000000000000000000000000011012
& 1410 = 000000000000000000000000000011102
1210 = 000000000000000000000000000011002
©2013 Richard L. Halterman Draft date: January 27, 2013
 550

Bits 2 and 3 are one for both 13 and 14; thus, bits 2 and 3 in the result must be one.
The bitwise or operator, |, takes two integer subexpressions and computes an integer result. The ex-
pression e1 | e2 is evaluated as follows:

If bit 0 in both e1 and e2 is 0, then bit 0 in the result is 0; otherwise, bit 0 in the result is 1.
If bit 1 in both e1 and e2 is 0, then bit 1 in the result is 0; otherwise, bit 1 in the result is 1.
If bit 2 in both e1 and e2 is 0, then bit 2 in the result is 0; otherwise, bit 2 in the result is 1.

..
.

If bit 31 in both e1 and e2 is 0, then bit 31 in the result is 0; otherwise, bit 31 in the result is 1.

For example, the expression 13 | 14 evaluates to 15, since:

1310 = 000000000000000000000000000011012
| 1410 = 000000000000000000000000000011102
1510 = 000000000000000000000000000011112

Bits 4–31 are zero in both 13 and 14. In bits 0–3 either 13 has a one or 14 has a one; therefore, the result
has ones in bits 0–3 and zeroes everywhere else.
The bitwise exclusive or (often refered to as xor) operator (ˆ) takes two integer subexpressions and
computes an integer result. The expression e1 ˆ e2 is evaluated as follows:

If bit 0 in e1 is the same as bit 0 in e2 , then bit 0 in the result is 0; otherwise, bit 0 in the result is 1.
If bit 1 in e1 is the same as bit 1 in e2 , then bit 1 in the result is 0; otherwise, bit 1 in the result is 1.
If bit 2 in e1 is the same as bit 2 in e2 , then bit 2 in the result is 0; otherwise, bit 2 in the result is 1.

..
.

If bit 31 in e1 is the same as bit 31 in e2 , then bit 31 in the result is 0; otherwise, bit 31 in the result
is 1.

For example, the expression 13 ˆ 14 evaluates to 3, since:

1310 = 000000000000000000000000000011012
ˆ 1410 = 000000000000000000000000000011102
310 = 000000000000000000000000000000112

Bits 0 and 1 differ in 13 and 14, so these bits are one in the result. The bits match in all the other positions,
so these positions must be set to zero in the result.
The bitwise negation operator ( ∼ ) is a unary operator that inverts all the bits of its expression. The
expression ∼ e is evaluated as follows:

If bit 0 in e is 0, then bit 0 in the result is 1; otherwise, bit 0 in the result is 0.

If bit 1 in e is 0, then bit 1 in the result is 1; otherwise, bit 1 in the result is 0.

©2013 Richard L. Halterman Draft date: January 27, 2013

 551

If bit 2 in e is 0, then bit 2 in the result is 1; otherwise, bit 2 in the result is 0.

..
.

If bit 31 in e is 0, then bit 31 in the result is 1; otherwise, bit 31 in the result is 0.

For example, the expression ∼ 13 evaluates to −14, since:

1310 = 000000000000000000000000000011012
negate ↓
−1410 = 111111111111111111111111111100102

The shift operators move all the bits in an integer to the left or right:

• Shift left (<<). The expression x << y shifts all the bits in x to the left y places. Zeros fill vacated
positions. The bits shifted off the left side are discarded. The expression 5 << 2 evaluates to 20,
since 510 = 1012 shifted two places to the left yields 101002 = 2010 . Observe that x << y is equal
to x × 2y .

• Shift right (>>). The expression x >> y shifts all the bits in x to the right y places. What fills the
vacated bits on the left depends on whether the integer is signed or unsigned (for example, int vs.
unsigned):

– For signed values the vacated bit positions are filled with the sign bit (the original leftmost bit).
– For unsigned values the vacated bit positions are filled with zeros.

The bits shifted off the right side are discarded. The expression 5 >> 2 evaluates to 1, since
510 = 1012 shifted two places to the left yields 0012 = 2010 (the original bits in positions 1
and 0 are shifted off the end and lost). Observe that x >> y is equal to x ÷ 2y .

Listing D.1 (bitwiseoperators.cpp) experiments with bitwise operators.

Listing D.1: bitwiseoperators.cpp

1 #include <iostream>
2
3 using namespace std;
4
5 int main()
6 {
7 int x, y;
8 cout << "Please enter two integers: ";
9 cin >> x >> y;
10 cout << x << " & " << y << " = " << (x & y) << endl;
11 cout << x << " | " << y << " = " << (x | y) << endl;
12 cout << x << " ˆ " << y << " = " << (x ˆ y) << endl;
13 cout << "∼" << x << " = " << ∼x << endl;
14 cout << x << " << " << 2 << " = " << (x << 2) << endl;
15 cout << x << " >> " << 2 << " = " << (x >> 2) << endl;
16 }

©2013 Richard L. Halterman Draft date: January 27, 2013

 552

Developers use bitwise operations for a variety of systems-level programming tasks. For example, in a
graphical user interface (GUI), the user generates events by interacting with an application using the mouse
and a keyboard. One event might be clicking a mouse button over a particular graphical element (like a
button) within a window. Information about this event can be stored in an integer. For example, bit 0 may
indicate whether or not the ↑ Shift key was held down when the mouse button was clicked. Bit 1 may be
responsible for the Alt key, bit 3 for the Ctrl key, etc. Thus the integer value 5, which in binary is

00000000000000000000000000000101

means that when the mouse button was clicked both the ↑ Shift and Alt keys were being held down.
This might require a different action on the part of the program than if some other combination of keys (or
none) were being pressed. For example, suppose the int variable key_status holds information about
which keys the user was depressing during the most recent mouse click. If the user presses the ↑ Shift
key during a mouse click, you want to display some warning dialog box. The following code fragment uses
a bitwise and to accomplish the task:
if ( key_status & 1 )
show_dialog("Warning: Operation not supported.");

If the expression key_status & 1 is zero, this means bit 0 is zero in mouseEvent, and, therefore,
the ↑ Shift key was not pressed. If the expression is zero, the condition is interpreted as false and the
running program will not display the dialog box. If on the other hand the expression evaluates to non-zero,
the ↑ Shift key is depressed and the expression is non-zero. Recall from Section 5.1 that any non-zero
condition is considered true, and so the dialog box appears.
Usually the GUI library with have constants defined that can be used to examine or modify particular
bits; for example, given the following constants:
const int SHIFT_DOWN = 1; // This is 0...0001
const int CTRL_DOWN = SHIFT_DOWN << 1; // This is 0...0010
const int ALT_DOWN = CTRL_DOWN << 1; // This is 0...0100

The above code fragment could be written:

if ( key_status & SHIFT_DOWN )
show_dialog("Warning: Operation not supported.");

If you instead wanted to check to see if the user was holding down both the ↑ Shift and Alt keys during
the mouse event, you could write
if ( key_status & (SHIFT_DOWN | ALT_DOWN) )
show_dialog("Warning: Operation not supported.");

Do you see how the expression SHIFT_DOWN | ALT_DOWN means both keys are down simultaneously?
Since we concentrate on applications programming instead of systems programming in this book, we
will have little use for the bitwise operators except for a few isolated situations. It is good to be aware of their
presence, though, since their accidental use may lead to difficult to diagnose compiler messages and logic
errors. As an example, suppose you mean to use && (logical AND) in a particular conditional expression
but an accidental keypress leaves you instead with & (bitwise AND). Because C++ allows integers to be
treated as Boolean expressions (0 is false, and everthing else is true), the mistake may elude the compiler’s
detection. The resulting logic error may be difficult to track down.

©2013 Richard L. Halterman Draft date: January 27, 2013

 553

Assignment Short Cut

x = x & y; x &= y;
x = x & y; x |= y;
x = x ˆ y; x ˆ= y;
x = x << y; x <<= y;
x = x >> y; x >>= y;

Table D.1: The bitwise assignment operators

Like the other C++ arithmetic operators that work on two operands, the bitwise operators &, |, ˆ, <<,
and >> may be combined with assignment to simplify the modification of a variable. Table D.1 lists the
possibilities.

©2013 Richard L. Halterman Draft date: January 27, 2013

 554

Index

abstract class, 487 destructor, 392

actual parameter, 195 dot operator (.), 379
alias, 302 dynamic binding, 480
and, bitwise, 549
and, logical, 86 early binding, 480
array initialization list, 273 encapsulation, 399
assignment, 23 enumerated type, 34
associativity, operator, 50 enumeration type, 34
auxiliary methods, 397 exclusive or, bitwise, 550
explicit template instantiation, 515
binary search, 322
bitwise and, 549 fields, 377
bitwise exclusive or, 550 floating-point types, 544
bitwise left shift, 551 for statment, 155
bitwise negation, 550 function call, 169
bitwise operators, 549 function definition, 189
bitwise or, 550 function invocation, 169
bitwise right shift, 551 function overloading, 238
bitwise shift operators, 551
generic function, 512
bitwise xor, 550
goto statement, 127
break statement, 126
buffer overrun, 292 helper methods, 397
homogeneous, 271
call by reference, 254
call by value, 201 if statement, 78
case label in a switch statement, 149 inheritance, 468
class, 376 instance, 377
class derivation, 468 integer types, 539
class specialization, 468 is-a relationship, 469
class template, 516 iteration, 107
client, 348 iterator, 358, 526
coincidental correctness, 465
compiler, 2 late binding, 480
concrete class, 487 left shift, 551
constructor, 390 linear search, 321
constructor initialization list, 391 linker, 4
continue statement, 129 Liskov Substitution Principle, 469
logical and, 86
data members, 377 logical negation, 86
default constructor, 392 logical not, 86
delegation, method, 475 logical or, 86
derivation, class, 468 loop, 107

©2013 Richard L. Halterman Draft date: January 27, 2013

INDEX 555

loop unrolling, 334 switch statement, 147

syntactic sugar, 350
matrix, 284
member function, 348, 388 template function, 511
memory leak, 298, 410 template instantiation, 512
method, 348 template value parameters, 515
method delegation, 475 this, 410
multiple inheritance, 471 Two’s complement binary, 541
two-dimensional array, 284
negation, bitwise, 550 typedef, 285
negation, logical, 86
nested loop, 117 underflow, 542
nibble, 541
non-type parameters, template, 515 value parameters, template, 515
non-type template parameters, 515
not, logical, 86 while statement, 108

object slicing, 477 xor, bitwise, 550

operator associativity, 50
operator precedence, 50
or, logical, 86
overflow, 542
overloaded function, 238
override, 475

parameter, actual, 195

pass by reference, 254
pass by value, 201
polymorphism, 483
precedence, operator, 50
private inheritance, 471
private members, 386
profiler, 5
protected member, 485
public members, 377
pure virtual methods, 487

ranged-based array parameter passing, 291

rational number, 394
real number, 544
return statement, 194
right shift, 551

shift, bitwise, 551

short-circuit evaluation, 88
specialization, class, 468
static binding, 480
steam manipulator, 114
string concatenation, 351
string objects, 348
struct, 434
subclassing, 468

©2013 Richard L. Halterman Draft date: January 27, 2013