Programming

Abstractions in C++
Eric S. Roberts and Julie Zelenski

This course reader has had an interesting evolutionary history that in some ways mirrors
the genesis of the C++ language itself. Just as Bjarne Stroustup’s first version of C++
was implemented on top of a C language base, this reader began its life as Eric Roberts’s
textbook Programming Abstractions in C (Addison-Wesley, 1998). In 2002-03, Julie
Zelenski updated it for use with the C++ programming language, which we began using
in CS106B and CS106X during that year.
Although the revised text worked fairly well at the outset, CS106B and CS106X have
evolved in recent years so that their structure no longer tracks the organization of the
book. This year, we’re engaged in the process of rewriting the book so that students in
these courses can use it as both a tutorial and a reference. As always, that process takes a
considerable amount to time, and there are likely to be some problems as we update the
reader. At the same time, we’re convinced that the material in CS106B and CS106X is
tremendously exciting and will be able to carry us through a quarter or two of instability,
and we will end up with an even better course in the future.
We want to thank our colleagues at Stanford, several generations of section leaders (with
special thanks to Dan Bentley and Keith Schwarz), and so many students over the
years—all of whom have helped make it so exciting to teach this wonderful material.
 Programming Abstractions in C++
Chapter 1. An Overview of C++ 1
1.1 What is C++? 2
The object-oriented paradigm; The compilation process
1.2 The structure of a C++ program 5
Comments; Library inclusions; Program-level definitions; Function prototypes;
The main program; Function definitions
1.3 Variables, values, and types 9
Naming conventions; Local and global variables; The concept of a data type;
Integer types; Floating-point types; Text types; Boolean type; Simple input and
output
1.4 Expressions 16
Precedence and associativity; Mixing types in an expression; Integer division and
the remainder operator; Type casts; The assignment operator; Increment and
decrement operators; Boolean operators
1.5 Statements 24
Simple statements; Blocks; The if statement; The switch statement; The while
statement; The for statement
1.6 Functions 32
Returning results from functions; Function definitions and prototypes; The
mechanics of the function-calling process; Passing parameters by reference
Summary 38
Review questions 39
Programming exercises 41

Chapter 2. Data Types in C++ 45

2.1 Enumeration types 46
Internal representation of enumeration types; Scalar types
2.2 Data and memory 49
Bits; bytes; and words; Memory addresses
2.3 Pointers 51
Using addresses as data values; Declaring pointer variables; The fundamental
pointer operations
2.4 Arrays 56
Array declaration; Array selection; Effective and allocated sizes; Initialization of
arrays; Multidimensional arrays
2.5 Pointers and arrays 64
The relationship between pointers and arrays
2.6 Records 67
Defining a new structure type; Declaring structure variables; Record selection;
Initializing records; Pointers to records
2.7 Dynamic allocation 71
Coping with memory limitations; Dynamic arrays; Dynamic records

ii
 Summary 74
Review questions 74
Programming exercises 77

Chapter 3. Libraries and Interfaces 85

3.1 The concept of an interface 86
Interfaces and implementations; Packages and abstractions; Principles of good
interface design
3.2 A random number interface 89
The structure of the random.h interface; Constructing a client program; The ANSI
functions for random numbers; The random.cpp implementation
3.3 Strings 98
The data type string; Operations on the string type ; The strutils.h
interface; An aside about C-style strings
3.4 Standard I/O and file streams 105
Data files; Using file streams in C++; Standard streams; Formatted stream output;
Formatted stream input; Single character I/O; Rereading characters from an input
file; Line-oriented I/O
3.5 Other ANSI libraries 112
Summary 113
Review questions 113
Programming exercises 116

Chapter 4. Using Abstract Data Types 123

4.1 The Vector class 125
Specifying the base type of a Vector; Declaring a new Vector object; Operations
on the Vector class; Iterating through the elements of a Vector; Passing a Vector
as a parameter
4.2 The Grid class 131
4.3 The Stack class 133
The structure of the Stack class
4.4 The Queue class 136
Simulations and models; The waiting-line model; Discrete time; Events in
simulated time; Implementing the simulation
4.5 The Map class 146
The structure of the Map class; Using maps in an application; Maps as associative
arrays
4.6 The Lexicon class 151
The structure of the Lexicon class; A simple application of the Lexicon class;
Why are lexicons useful if maps already exist
4.7 The Scanner class 154
Setting scanner options
4.8 Iterators 156
The standard iterator pattern; Iteration order; A simple iterator example;
Computing word frequencies

iii
 Summary 163
Review questions 164
Programming exercises 165

Chapter 5. Introduction to recursion 173

5.1 A simple example of recursion 174
5.2 The factorial function 176
The recursive formulation of Fact; Tracing the recursive process; The recursive
leap of faith
5.3 The Fibonacci function 181
Computing terms in the Fibonacci sequence; Gaining confidence in the recursive
implementation; Recursion is not to blame
5.4 Other examples of recursion 187
Detecting palindromes; Binary search; Mutual recursion
5.5 Thinking recursively 192
Maintaining a holistic perspective; Avoiding the common pitfalls
Summary 194
Review questions 195
Programming exercises 197

Chapter 6. Recursive procedures 201

6.1 The Tower of Hanoi 202
Framing the problem; Finding a recursive strategy; Validating the strategy;
Coding the solution; Tracing the recursive process
6.2 Generating permutations 211
The recursive insight
6.3 Graphical applications of recursion 213
The graphics library; An example from computer art; Fractals
Summary 224
Review questions 225
Programming exercises 226

Chapter 7. Backtracking algorithms 235

7.1 Solving a maze by recursive backtracking 236
The right-hand rule; Finding a recursive approach; Identifying the simple cases;
Coding the maze solution algorithm; Convincing yourself that the solution works
7.2 Backtracking and games 245
The game of nim; A generalized program for two-player games; The minimax
strategy; Implementing the minimax algorithm; Using the general strategy to
solve a specific game
Summary 269
Review questions 270
Programming exercises 271

iv
Chapter 8. Algorithmic analysis 277
8.1 The sorting problem 278
The selection sort algorithm; Empirical measurements of performance; Analyzing
the performance of selection sort
8.2 Computational complexity and big-O notation 282
Big-O notation; Standard simplifications of big-O; Predicting computational
complexity from code structure; Worst-case versus average-case complexity; A
formal definition of big-O
8.3 Recursion to the rescue 288
The power of divide-and-conquer strategies; Merging two vectors; The merge sort
algorithm; The computational complexity of merge sort; Comparing N 2 and N log
N performance
8.4 Standard complexity classes 294
8.5 The Quicksort algorithm 296
Partitioning the vector; Analyzing the performance of Quicksort
8.6 Mathematical induction 301
Summary 304
Review questions 305
Programming exercises 307

Chapter 9. Classes and objects 313

9.1 A simple example of a class definition 314
Defining a Point class; Implementing methods in a class; Constructors and
destructors; The keyword this
9.2 Implementing a specialized version of the Stack class 319
Defining the CharStack interface; Representing the stack data; The advantages of
object encapsulation; Removing the maximum size limitation; Object copying
9.3 Implementing the Scanner class 328
Summary 328
Review questions 334
Programming exercises 335

Chapter 10. Efficiency and Data Representation 339

10.1 The concept of an editor buffer 340
10.2 Defining the buffer abstraction 341
The public interface of the EditorBuffer class; Coding the editor application
10.3 Implementing the editor using arrays 345
Defining the private data representation; Implementing the buffer operations;
Assessing the computational complexity of the array implementation
10.4 Implementing the editor using stacks 352
Defining the private data representation for the stack-based buffer; Implementing
the buffer operations; Comparing computational complexities

v
 10.5 Implementing the editor using linked lists 357
The concept of a linked list; Designing a linked-list data structure; Using a linked
list to represent the buffer; Insertion into a linked-list buffer; Deletion in a linked-
list buffer; Cursor motion in the linked-list representation; Linked-list idioms;
Completing the buffer implementation; Computational complexity of the linked-
list buffer; Doubly linked lists; Time-space tradeoffs
Summary 371
Review questions 372
Programming exercises 373

Chapter 11. Linear Structures 381

11.1 Reimplementing stacks as a template class 382
The interface of a class template
11.2 Reimplementing stacks using linked lists 383
11.3 Implementing queues 391
An array-based implementation of queues; Linked-list representation of queues
11.4 Implementing vectors 404
Supporting insertion and deletion at arbitrary index positions; Implementing
selection brackets; Implementing iterators
Summary 414
Review questions 415
Programming exercises 416

Chapter 12. Implementing Maps 419

12.1 An array-based implementation of the map interface 420
12.2 The advantage of knowing where to look 427
12.3 Hashing 429
Implementing the hash table strategy; Choosing a hash function; Determining the
number of buckets; Using the typename keyword
12.4 Functions as data 438
A general plotting function; Declaring pointers to functions and function
typedefs; Implementing Plot; A generic sorting function
12.5 Mapping functions 444
Mapping over entries in a map; Implementing mapAll; Passing client information
to a callback function; A note on function types and methods
Summary 448
Review questions 449
Programming exercises 450

Chapter 13. Trees 455

13.1 Family trees 456
Terminology used to describe trees; The recursive nature of a tree; Representing
family trees in C++

vi
 13.2 Binary search trees 459
The underlying motivation for using binary search trees; Finding nodes in a
binary search tree; Inserting new nodes in a binary search tree; Tree traversals
13.3 Balanced trees 466
Tree-balancing strategies; Illustrating the AVL idea; Single rotations; Double
rotations; Implementing the AVL algorithm
13.4 Defining a general interface for binary search trees 477
Allowing the client to define the node data; Generalizing the types used for keys;
Removing nodes; Implementing the binary search tree package; Implementing the
map.h interface using binary trees; Using the static keyword
Summary 488
Review questions 489
Programming exercises 492

Chapter 14. Expression Trees 499

14.1 Overview of the interpreter 500
14.2 Understanding the abstract structure of expressions 505
A recursive definition of expressions; Expression trees
14.3 Class hierarchies and inheritance 509
14.4 Defining an inheritance hierarchy for expressions 510
Defining the interface for the expression subclasses
14.5 Implementing the node classes 518
Implementing the methods
14.6 Parsing an expression 522
Parsing and grammars; Parsing without precedence; Adding precedence to the
parser
Summary 528
Review questions 528
Programming exercises 530

Chapter 15. Sets 535

15.1 Sets as a mathematical abstraction 536
Membership; Set operations; Identities on sets
15.2 Designing a set interface 539
Defining the element type; Writing the set interface; Character sets; Using sets to
avoid duplication
15.3 Implementing the set class 544
15.4 Enhancing the efficiency of integer sets 548
Characteristic vectors; Packed arrays of bits; Bitwise operators; Implementing
characteristic vectors using the bitwise operators; Implementing the high-level set
operations; Using a hybrid implementation
Summary 555
Review questions 556
Programming exercises 558

vii
Chapter 16. Graphs 563
16.1 The structure of a graph 564
Directed and undirected graphs; Paths and cycles; Connectivity
16.2 Implementation strategies for graphs 568
Representing connections using an adjacency list; Representing connections using
an adjacency matrix; Representing connections using a set of arcs
16.3 Designing a low-level graph abstraction 571
Using the low-level graph.h interface
16.4 Graph traversals 575
Depth-first search; Breadth-first search
16.5 Defining a Graph class 580
Using classes for graphs, nodes, and arcs; Adopting an intermediate strategy
16.6 Finding minimum paths 589
16.7 An efficient implementation of priority queues 593
Summary 596
Review questions 597
Programming exercises 599

Appendix A. Library Interfaces 607

bst.h 608
cmpfn.h 611
extgraph.h 612
genlib.h 622
graph.h 623
graphics.h 627
grid.h 630
lexicon.h 634
map.h 638
queue.h 642
random.h 644
scanner.h 646
set.h 652
simpio.h 656
sound.h 657
stack.h 658
strutils.h 660
vector.h 662

Index 657

viii
Chapter 1
An Overview of C++

Out of these various experiments come programs. This is

our experience: programs do not come out of the minds of
one person or two people such as ourselves, but out of day-
to-day work.
— Stokely Carmichael and Charles V. Hamilton,
Black Power, 1967
An Overview of C++ –2–

In Lewis Carroll’s Alice’s Adventures in Wonderland, the King asks the White Rabbit to
“begin at the beginning and go on till you come to the end: then stop.” Good advice, but
only if you’re starting from the beginning. This book is designed for a second course in
computer science and therefore assumes that you have already begun your study of
programming. At the same time, because first courses vary considerably in what they
cover, it is difficult to rely on any specific material. Some of you, for example, will
already have experience programming in C or C++. Many of you, however, are coming
from a first course taught in some other language.
Because of this wide disparity in background, the best approach is to adopt the King’s
advice and begin at the beginning. The first three chapters in this text therefore move
quickly through the material I consider to be essential background for the later chapters.
Chapters 1 and 2 discuss C++ in general and may be skimmed if you’ve had experience
with C++. Chapter 3 discusses standard interfaces and some interfaces particular to this
text. By the end of these three chapters, you will be up to speed on the fundamentals of
C++ programming.

1.1 What is C++?

In the early days of computing, programs were written in machine language, which
consists of the primitive instructions that can be executed directly by the machine.
Machine-language programs are difficult to understand, mostly because the structure of
machine language reflects the design of the hardware rather than the needs of
programmers. In the mid-1950s, a group of programmers under the direction of John
Backus at IBM had an idea that profoundly changed the nature of computing. Would it
be possible, they wondered, to write programs that resembled the mathematical formulas
they were trying to compute and have the computer itself translate those formulas into
machine language? In 1955, this team produced the initial version of Fortran (whose
name is an abbreviation of formula translation), which was the first example of a higher-
level programming language. Since that time, many new programming languages have
been invented, as shown in the evolutionary diagram in Figure 1-1.

Figure 1-1 Evolutionary tree of several major programming languages

An Overview of C++ –3–

As Figure 1-1 illustrates, C++ represents the coming together of two branches in the
evolution of programming languages. One of its ancestors is a language called C, which
was designed at Bell Laboratories by Dennis Ritchie in 1972 and then later revised and
standardized by the American National Standards Institute (ANSI) in 1989. But C++ also
descends from another line of languages that have dramatically changed the nature of
modern programming.

The object-oriented paradigm

Over the last decade or so, computer science and programming have gone through
something of a revolution. Like most revolutions—whether political upheavals or the
conceptual restructurings that Thomas Kuhn describes in his 1962 book The Structure of
Scientific Revolutions—this change has been driven by the emergence of an idea that
challenges an existing orthodoxy. Initially, the two ideas compete. For a while, the old
order maintains its dominance. Over time, however, the strength and popularity of the
new idea grows, until it begins to displace the older idea in what Kuhn calls a paradigm
shift. In programming, the old order is represented by the procedural paradigm, in
which programs consist of a collection of procedures and functions that operate on data.
The challenger is the object-oriented paradigm, in which programs are viewed instead
as a collection of data objects that exhibit particular behavior.
The idea of object-oriented programming is not really all that new. The first object-
oriented language was SIMULA, a language for coding simulations designed in 1967 by
the Scandinavian computer scientists Ole-Johan Dahl, Björn Myhrhaug, and Kristen
Nygaard. With a design that was far ahead of its time, SIMULA anticipated many of the
concepts that later became commonplace in programming, including the concept of
abstract data types and much of the modern object-oriented paradigm. In fact, most of
the terminology used to describe object-oriented systems comes from the original 1967
report on SIMULA.
For many years, however, SIMULA mostly just sat on the shelf. Few people paid
much attention to it, and the only place you were likely to hear about it would be in a
course on programming language design. The first object-oriented language to gain any
significant level of recognition within the computing profession was Smalltalk, which
was developed at the Xerox Palo Alto Research Center (more commonly known as Xerox
PARC) in the late 1970s. The purpose of Smalltalk, which is described in the book
Smalltalk-80: The Language and Its Implementation by Adele Goldberg and David
Robson, was to make programming accessible to a wider audience. As such, Smalltalk
was part of a larger effort at Xerox PARC that gave rise to much of the modern user-
interface technology that is now standard on personal-computers.
Despite many attractive features and a highly interactive user environment that
simplifies the programming process, Smalltalk never achieved much commercial success.
The profession as a whole took an interest in object-oriented programming only when the
central ideas were incorporated into variants of C, which had already become an industry
standard. Although there were several parallel efforts to design an object-oriented
language based on C, the most successful was the language C++, which was designed in
the early 1980s by Bjarne Stroustrup at AT&T Bell Laboratories. By making it possible
to integrate object-oriented techniques with existing C code, C++ enabled large
communities of programmers to adopt the object-oriented paradigm in a gradual,
evolutionary way.
Although object-oriented languages are undeniably gaining popularity at the expense
of procedural ones, it would be a mistake to regard the object-oriented and procedural
paradigms as mutually exclusive. Programming paradigms are not so much competitive
An Overview of C++ –4–

as they are complementary. The object-oriented and the procedural paradigm—along

with other important paradigms such as the functional programming style embodied in
LISP—all have important applications in practice. Even within the context of a single
application, you are likely to find a use for more than one approach. As a programmer,
you must master many different paradigms, so that you can use the conceptual model that
is most appropriate to the task at hand.

The compilation process

When you write a program in C++, your first step is to create a file that contains the text
of the program, which is called a source file. Before you can run your program, you
need to translate the source file into an executable form. The first step in that process is
to invoke a program called a compiler, which translates the source file into an object file
containing the corresponding machine-language instructions. This object file is then
combined with other object files to produce an executable file that can be run on the
system. The other object files typically include predefined object files, called libraries,
that contain the machine-language instructions for various operations commonly required
by programs. The process of combining all the individual object files into an executable
file is called linking. The process is illustrated by the diagram shown in Figure 1-2.

Unfortunately, the specific details of the compilation process vary considerably from
one machine to another. There is no way that a general textbook like this can tell you
exactly what commands you should use to run a program on your system. Because those
commands are different for each system, you need to consult the documentation that
comes with the compiler you are using on that machine. The good news, however, is that
the C++ programs themselves will look the same. One of the principal advantages of
programming in a higher-level language like C++ is that doing so often allows you to
ignore the particular characteristics of the hardware and create programs that will run on
many different machines.

Figure 1-2 The compilation process

source file object file
/* File: count.c */ 0100100101011001000
1000010100011101011
#include <stdio.h> 0110100111010101100
#include "genlib.h"

#define N 10

main() compiler
{
int i;
executable file
for (i=1; i<=N; i++) { 0100100101011001000
printf("%d\n", i); 1000010100011101011
} 0110100111010101100
} 1001011010110001011
0100100101001011011
0101101011010100101

linker
other object
files/libraries
1001011010110001011
0100100101001011011
0101101011010100101
An Overview of C++ –5–

1.2 The structure of a C++ program

The best way to get a feeling for the C++ programming language is to look at a sample
program such as the one shown in Figure 1-3. This program generates a table comparing
the values of N 2 and 2N for various values of N—a comparison that will prove to be
important in Chapter 8. The output of the program looks like this:

PowerTable
| 2 | N
N | N | 2
----+-----+------
0 | 0 | 1
1 | 1 | 2
2 | 4 | 4
3 | 9 | 8
4 | 16 | 16
5 | 25 | 32
6 | 36 | 64
7 | 49 | 128
8 | 64 | 256
9 | 81 | 512
10 | 100 | 1024
11 | 121 | 2048
12 | 144 | 4096

As the annotations in Figure 1-3 indicate, the powertab.cpp program is divided into
several components, which are discussed in the next few sections.

Comments
Much of the text in Figure 1-3 consists of English-language comments. A comment is
text that is ignored by the compiler but which nonetheless conveys information to other
programmers. A comment consists of text enclosed between the markers /* and */ and
may continue over several lines. Alternatively, a single-line comment is begun by the
marker // and continues until the end of the line. The powertab.cpp program includes a
comment at the beginning that describes the operation of the program as a whole, one
before the definition of the RaiseIntToPower function that describes what it does, and a
couple of one-line comments that act very much like section headings in English text.

Library inclusions
The lines beginning with #include such as
#include "genlib.h"
#include <iostream>
#include <iomanip>

indicate that the compiler should read in definitions from a header file. The inclusion of
a header file indicates that the program uses facilities from a library, which is a
collection of prewritten tools that perform a set of useful operations. The different
punctuation in these #include lines reflects the fact that the libraries come from different
sources. The angle brackets are used to specify a system library, such as the standard
input/output stream library (iostream) or the stream manipuator library (iomanip) that is
supplied along with C++. The quotation marks are used for private libraries, including
the general library (genlib), which was designed for use with the programs in this text.
Every program in this book will include at least this library most will require other
libraries as well and must contain an #include line for each one.
An Overview of C++ –6–

Figure 1-3 Sample program powertab.cpp

/*
* File: powertab.cpp
* ------------------ program
* This program generates a table comparing values comment
* of the functions n^2 and 2^n.
*/

#include "genlib.h" library

#include <iostream> inclusions
#include <iomanip>

/*
* Constants section
* ---------
comment
* LOWER_LIMIT -- Starting value for the table
* UPPER_LIMIT -- Final value for the table
*/
constant
const int LOWER_LIMIT = 0;
const int UPPER_LIMIT = 12; definitions

/* Private function prototypes */

function
int RaiseIntToPower(int n, int k); prototype
/* Main program */

int main() {
cout << " | 2 | N" << endl;
cout << " N | N | 2" << endl;
cout << "----+-----+------" << endl; main
for (int n = LOWER_LIMIT; n <= UPPER_LIMIT; n++) {
program
cout << setw(3) << n << " |" ;
cout << setw(4) << RaiseIntToPower(n, 2) << " |" ;
cout << setw(5) << RaiseIntToPower(2, n) << endl;
}
return 0;
}

/*
* Function: RaiseIntToPower
* Usage: p = RaiseIntToPower(n, k); function
* --------------------------------- comment
* This function returns n to the kth power.
*/

int RaiseIntToPower(int n, int k) {

local variable
int result;
declarations
result = 1; function
for (int i = 0; i < k; i++) { statements definition
result *= n; forming body
}
of function
return result;
}
An Overview of C++ –7–

Program-level definitions
After the #include lines for the libraries, many programs define constants that apply to
the program as a whole. In the powertab.cpp program, the following lines
const int LOWER_LIMIT = 0;
const int UPPER_LIMIT = 12;

introduce two constants named LOWER_LIMIT and UPPER_LIMIT.

The syntax for declaring a constant resembles that for declaring a variable (discussed
in section 1.3) that includes the type modifier const. The general form is
const type name = value ;

which defines the constant name to be of type type and initialized to value. A constant must
be initialized when it is defined and once initialized, it cannot be assigned a new value or
changed in any way. Attempting to do so will result in a compiler error. After a named
constant is defined, it is available to be used anywhere in the rest of the program. For
example, after encountering the line
const double PI = 3.14159265;

any subsequent use of the name PI refers to the constant value 3.14159265.
Giving symbolic names to constants has several important advantages in terms of
programming style. First, the descriptive names give readers of the program a better
sense of what the constant value means. Second, centralizing such definitions at the top
of the file makes it easier to change the value associated with a name. For example, all
you need to do to change the limits used for the table in the powertab.cpp program is
change the values of the constants. And lastly, a const declaration protects from the
value from any unintended modification.

In addition to constants, programs often define new data types in this section of the
source file, as you will see in Chapter 2.

Function prototypes
Computation in a C++ program is carried out in the context of functions. A function is a
unit of code that (1) performs a specific operation and (2) is identified by name. The
powertab.cpp program contains two functions—main and RaiseIntToPower—which
are described in more detail in the next two sections. The line

int RaiseIntToPower(int n, int k);

is an example of a function prototype, a declaration that tells the compiler the

information it needs to know about a function to generate the proper code when that
function is invoked. This prototype, for example, indicates that the function
RaiseIntToPower takes two integers as arguments and returns an integer as its result.

You must provide the declaration or definition of each function before making any
calls to that function. C++ requires this in order for the compiler to check whether calls to
functions are compatible with the corresponding prototypes and can therefore aid you in
the process of finding errors in your code.
An Overview of C++ –8–

The main program

Every C++ program must contain a function with the name main. This function specifies
the starting point for the computation and is called when the program starts up. When
main has finished its work and returns, execution of the program ends.

The first three statements of the main function in the powertab.cpp program are
sending information to the cout stream to display output on the screen. A few useful
notes about streams are included in the section on “Simple input and output” later in this
chapter and more features are explored in detail in Chapter 3. At this point, you need to
have an informal sense of how to display output to understand any programming example
that communicates results to the user. In its simplest form, you use the insertion operator
<< to put information into the output stream cout . If you insert a string enclosed in
double quotes, it will display that string on the console. You must indicate explicitly that
you want to move on to the next line by inserting the stream manipulator endl. Thus,
the first three lines in main display the header for the table.
The rest of the function main consists of the following code, which is responsible for
displaying the table itself:
for (int n = LOWER_LIMIT; n <= UPPER_LIMIT; n++) {
cout << setw(3) << n << " |" ;
cout << setw(4) << RaiseIntToPower(n, 2) << " |" ;
cout << setw(5) << RaiseIntToPower(2, n) << endl;
}

This code is an example a for loop, which is used to specify repetition. In this case, the
for statement indicates that the body of the loop should be repeated for each of the
values of n from LOWER_LIMIT to UPPER_LIMIT. A section on the detailed structure of
the for loop appears later in the chapter, but the example shown here represents a
common idiomatic pattern that you can use to count between any specified limits.
The body of the loop illustrates an important new feature: the ability to include values
as part of the output display. Rather than just displaying fixed strings, we can display
numeric values and computed results. We can also use stream manipulators to format the
output. Let’s examine the first statement in the body of the loop:
cout << setw(3) << n << " |" ;

This line will display the current value of the variable n followed by a string
containing a space and a vertical bar. The setw(3) that is inserted into the stream just
before n indicates that the stream should format the next value in a field that is three
characters wide. Similarly, the next statement prints the formatted result taken from the
expression RaiseIntToPower(n, 2) . Obtaining the value of the expression requires
making a call on the RaiseIntToPower function, which is discussed in the following
section. The value that RaiseIntToPower returns is displayed as part of the output. So
is the result of the call to RaiseIntToPower(2, n), which supplies the value for the third
column in the table.
The last statement in main is
return 0;

which indicates that the function result is 0. The return value from the main function is
used to communicate the success or failure of the entire program. By convention, a result
of 0 indicates success.
An Overview of C++ –9–

Function definitions
Because large programs are difficult to understand in their entirety, most programs are
broken down into several smaller functions, each of which is easier to understand. In the
powertab.cpp program, the function RaiseIntToPower is responsible for raising an
integer to a power—an operation that is not built into C++ and must therefore be defined
explicitly.
The first line of RaiseIntToPower is the variable declaration
int result;

which introduces a new variable named result capable of holding values of type int,
the standard type used to represent integers. The syntax of variable declarations is
discussed in more detail in the section on “Variables, values, and types” later in this
chapter. For now, all you need to know is that this declaration creates space for an
integer variable that you can then use in the body of the function.
The next line in the body of RaiseIntToPower is
result = 1;

This statement is a simple example of an assignment statement, which sets the variable
on the left of the equal sign to the value of the expression on the right. In this case, the
statement sets the variable result to the constant 1. The next statement in the function is
a for loop that executes its body k times. The repeated code consists of the line
result *= n;

which is a C++ shorthand for the English sentence “Multiply result by n.” Because the
function initializes the value of result to 1 and then multiplies result by n a total of k
times, the variable result ends up with the value nk.
The last statement in RaiseIntToPower is
return result;

which indicates that the function should return result as the value of the function.

1.3 Variables, values, and types

One of the fundamental characteristics of programs is that they manipulate data. To do
so, programs must be able to store data as part of their operation. Moreover, programs
today work with many different kinds of data, including numbers and text, along with
many more sophisticated data structures, such as those introduced in Part 3 of this book.
Learning how to store data of various types is an important part of mastering the basics of
any language, including C++.
Variables
Data values in a program are usually stored in variables. In C++, if you want to use a
variable to hold some information, you must declare that variable before you use it.
Declaring a variable establishes the following properties:
• Name. Every variable has a name, which is formed according to the rules described in
the section entitled “Naming conventions” later in this chapter. You use the name in
the program to refer to the variable and the value it contains.
An Overview of C++ – 10 –

• Type. Each variable in a C++ program is constrained to hold values of a particular

data type. C++ includes several predefined types and also allows you to define new
types of your own, as discussed in Chapter 2.
• Lifetime. Depending on how they are declared, some variables persist throughout the
entire program, while others are created and destroyed dynamically as the program
moves through various levels of function call.
• Scope. The declaration of a variable also controls what parts of the program have
access to the variable, which is called its scope.
The standard syntax for declaring a variable is
type namelist;

where type indicates the data type and namelist is a list of variable names separated by
commas. For example, the function RaiseIntToPower in the powertab.cpp program
contains the line
int result;

which declares the variable result to be of type int.

In C++, the initial contents of a variable are undefined. If you want a variable to have
a particular initial value, you need to initialize it explicitly. One approach is to use an
assignment statement in the body of the function to assign a value to each variable before
you use it. You can, however, include initial values directly in a declaration by writing
an equal sign and a value after a variable name. Thus, the declaration
int result = 0;

is a shorthand for the following code, in which the declaration and assignment are
separate:
int result;
result = 0;

An initial value specified as part of a declaration is called an initializer.

Naming conventions
The names used for variables, functions, types, constants, and so forth are collectively
known as identifiers. In C++, the rules for identifier formation are
1. The name must start with a letter or the underscore character (_).
2. All other characters in the name must be letters, digits, or the underscore. No spaces
or other special characters are permitted in names.
3. The name must not be one of the reserved keywords listed in Table 1-1
Uppercase and lowercase letters appearing in an identifier are considered to be different.
Thus, the name ABC is not the same as the name abc. Identifiers can be of any length, but
C++ compilers are not required to consider any more than the first 31 characters in
determining whether two names are identical. Implementations may impose additional
restrictions on identifiers that are shared between modules.
You can improve your programming style by adopting conventions for identifiers that
help readers identify their function. In this text, names of variables and data types begin
An Overview of C++ – 11 –

Table 1-1 C++ reserved keywords

asm do inline short typeid
auto double int signed typename
bool dynamic_cast long sizeof union
break else mutable static unsigned
case enum namespace static_cast using
catch explicit new struct virtual
char extern operator switch void
class false private template volatile
const float protected this wchar_t
const_cast for public throw while
continue friend register true
default goto reinterpret_cast try
delete if return typedef

with a lowercase letter, such as n1, total, or string. By contrast, function names, such
as RaiseIntToPower, usually begin with an uppercase letter. Moreover, whenever a
name consists of several English words run together, the first letter in each word after the
first is capitalized to make the name easier to read. By tradition, constant names, such as
LOWER_LIMIT are written entirely in uppercase, with underscores between the words.

Local and global variables

Most variables are declared with the body of a function. Such variables are called local
variables. The scope of a local variable extends to the end of the block in which it is
declared. The lifetime of a local variable is the time during which that function is active.
When the function is called, space for each local variable is allocated for the duration of
that function call. When the function returns, all its local variables disappear.
If a variable declaration appears outside any function definition, that declaration
introduces a global variable. The scope of a global variable is the remainder of the file
in which it is declared. Its lifetime continues throughout the entire execution of a
program. Global variables are therefore able to store values that persist across function
calls. Although they have important applications, global variables can easily be
overused. Because global variables can be manipulated by many different functions, it is
harder to keep those functions from interfering with each other. Because of these
dangers, global variables are used infrequently in this text.

The concept of a data type

One of the reasons C++ requires all variables to be declared is that doing so constrains
their contents to values of a particular data type. From a formal perspective, a data type
is defined by two properties: a domain, which is the set of values that belong to that type,
and a set of operations, which defines the behavior of that type. For example, the domain
of the type int includes all integers (. . . –2, –1, 0, 1, 2 . . .) up to the limits established by
the hardware of the machine. The set of operations applicable to values of type int
includes, for example, the standard arithmetic operations like addition and multiplication.
Other types have a different domain and set of operations.
As you will learn in Chapter 2, much of the power of higher-level languages like C++
comes from the fact that you can define new data types from existing ones. To get that
process started, C++ includes several fundamental types that are defined as part of the
language (or, in the case of string, as part of the standard C++ libraries). These types,
which act as the building blocks for the type system as a whole, are called atomic types.
These predefined types are grouped into four categories—integer, floating-point, text, and
Boolean—which are discussed in the sections that follow.
An Overview of C++ – 12 –

Integer types
Although the concept of an integer seems like a simple one, C++ actually includes several
different data types for representing integer values. In most cases, all you need to know
is the type int, which corresponds to the standard representation of an integer on the
computer system you are using. In certain cases, however, you need to be more careful.
Like all data, values of type int are stored internally in storage units that have a limited
capacity. Those values therefore have a maximum size, which limits the range of
integers you can use. To get around this problem, C++ defines three integer types—
short, int, and long—distinguished from each other by the size of their domains.

Unfortunately, the language definition for C++ does not specify an exact range for
these three types. As a result, the range for the different integer types depends on the
machine and the compiler you’re using. On many personal computers, the maximum
value of type int is 32,767, which is rather small by computational standards. If you
wanted, for example, to perform a calculation involving the number of seconds in a year,
you could not use type int on those machines, because that value (31,536,000) is
considerably larger than the largest available value of type int. The only properties you
can rely on are the following:

• The internal size of an integer cannot decrease as you move from short to int to
long. A compiler designer for C++ could, for example, decide to make short and int
the same size but could not make int smaller than short.
• The maximum value of type int must be at least 32,767 (215–1).
• The maximum value of type long must be at least 2,147,483,647 (231–1).

The designers of C++ could have chosen to define the allowable range of type int
more precisely. For example, they could have declared—as the designers of Java did—
that the maximum value of type int would be 231–1 on every machine. Had they done
so, it would be easier to move a program from one system to another and have it behave
in the same way. The ability to move a program between different machines is called
portability, which is an important consideration in the design of a programming
language.
In C++, each of the integer types int , long , and short may be preceded by the
keyword unsigned . Adding u n s i g n e d creates a new data type in which only
nonnegative values are allowed. Because unsigned variables do not need to represent
negative values, declaring a variable to be one of the unsigned types allows it to hold
twice as many positive values. For example, if the maximum value of type int is 32,767,
the maximum value of type unsigned int will be 65,535. C++ allows the type unsigned
int to be abbreviated to unsigned , and most programmers who use this type tend to
follow this practice. Sometimes variables intended to store sizes are declared as
unsigned, because a size will always be nonnegative.
An integer constant is ordinarily written as a string of digits representing a number in
base 10. If the number begins with the digit 0, however, the compiler interprets the value
as an octal (base 8) integer. Thus, the constant 040 is taken to be in octal and represents
the decimal number 32. If you prefix a numeric constant with the characters 0x, the
compiler interprets that number as hexadecimal (base 16). Thus, the constant 0xFF is
equivalent to the decimal constant 255. You can explicitly indicate that an integer
constant is of type long by adding the letter L at the end of the digit string. Thus, the
constant 0L is equal to 0, but the value is explicitly of type long. Similarly, if you use the
letter U as a suffix, the constant is taken to be unsigned.
An Overview of C++ – 13 –

Floating-point types
Numbers that include a decimal fraction are called floating-point numbers, which are
used to approximate real numbers in mathematics. As with integers, C++ defines three
different floating-point types: float, double, and long double. Although ANSI C++
does not specify the exact representation of these types, the way to think about the
difference is that types that appear later in the list allow numbers to be represented with
greater precision but require more memory space. Unless you are doing exacting
scientific calculation, the differences between these types will not matter a great deal. In
keeping with a common convention among C++ programmers, this text uses the type
double as its standard floating-point type.

Floating-point constants in C++ are written with a decimal point. Thus, if 2.0 appears
in a program, the number is represented internally as a floating-point value if the
programmer had written 2, this value would be an integer. Floating-point values can also
be written in a special programmer’s style of scientific notation, in which the value is
represented as a floating-point number multiplied by a integral power of 10. To write a
number using this style, you write a floating-point number in standard notation, followed
immediately by the letter E and an integer exponent, optionally preceded by a + or - sign.
For example, the speed of light in meters per second can be written in C++ as

2.9979E+8

where the E stands for the words times 10 to the power.

Text types
In the early days, computers were designed to work only with numeric data and were
sometimes called number crunchers as a result. Modern computers, however, work less
with numeric data than they do with text data, that is, any information composed of
individual characters that appear on the keyboard and the screen. The ability of modern
computers to process text data has led to the development of word processing systems,
on-line reference libraries, electronic mail, and a wide variety of other useful
applications.
The most primitive elements of text data are individual characters, which are
represented in C++ using the predefined data type char. The domain of type char is the
set of symbols that can be displayed on the screen or typed on the keyboard: the letters,
digits, punctuation marks, spacebar, Return key, and so forth. Internally, these values are
represented inside the computer by assigning each character a numeric code. In most
implementations of C++, the coding system used to represent characters is called ASCII,
which stands for the American Standard Code for Information Interchange. The numeric
values of the characters in the ASCII set are shown in Table 1-2.
Although it is important to know that characters are represented internally using a
numeric code, it is not generally useful to know what numeric value corresponds to a
particular character. When you type the letter A, the hardware logic built into the
keyboard automatically translates that character into the ASCII code 65, which is then
sent to the computer. Similarly, when the computer sends the ASCII code 65 to the
screen, the letter A appears.
You can write a character constant in C++ by enclosing the character in single quotes.
Thus, the constant 'A' represents the internal code of the uppercase letter A. In addition
to the standard characters, C++ allows you to write special characters in a two-character
form beginning with a backward slash (\). These two-character combinations are called
An Overview of C++ – 14 –

Table 1-2 ASCII codes

0 1 2 3 4 5 6 7 8 9
0 \000 \001 \002 \003 \004 \005 \006 \a \b \t
10 \n \v \f \r \016 \017 \020 \021 \022 \023
20 \024 \025 \026 \027 \030 \031 \032 \033 \034 \035
30 \036 \037 space ! " # $ % & '
40 ( ) * + , - . / 0 1
50 2 3 4 5 6 7 8 9 :
60 < = > ? @ A B C D E
70 F G H I J K L M N O
80 P Q R S T U V W X Y
90 Z [ \ ] ^ _ ` a b c
100 d e f g h i j k l m
110 n o p q r s t u v w
120 x y z { | } ~ \177

escape sequences and you can see several listed in the first few rows of Table 1-2. A
few of the more commonly used escape sequences are '\n' the newline character,'\t'
the tab character, and '\\' the backslash character.
Characters are most useful when they are collected together into sequential units. In
programming, a sequence of characters is called a string. Strings make it possible to
display informational messages on the screen. You have already seen strings in the
sample program powertab.cpp. It is important, however, to recognize that strings are
data and that they can be manipulated and stored in much the same way that numbers can.
The standard C++ library defines a string type and operations that manipulate strings.
The details of type string are not important at this point strings are considered in more
detail in Chapter 3. In this chapter, strings are treated as atomic values and used
exclusively to specify text that is displayed directly on the display screen.
You write string constants in C++ by enclosing the characters contained within the
string in double quotes. C++ supports the same escape sequences for strings as for
characters. If two or more string constants appear consecutively in a program, the
compiler concatenates them together. The most important implication of this rule is that
you can break a long string over several lines so that it doesn’t end up running past the
right margin of your program.

Boolean type
In the programs you write, it is often necessary to test a particular condition that affects
the subsequent behavior of your code. Typically, that condition is specified using an
expression whose value is either true or false. This data type—for which the only legal
values are true and false—is called Boolean data, after the mathematician George Boole,
who developed an algebraic approach for working with such values.
In C++, the Boolean type is called bool and its domain consists of the values true and
false. You can declare variables of type bool and manipulate them in the same way as
other data objects.
An Overview of C++ – 15 –

Simple input and output

Before you can write programs that interact with the user, you need to have some way of
accepting input data from the user and displaying results on the screen. In C++, none of
this functionality is provided directly within the language. Instead, all input and output
operations—which are often referred to collectively as I/O operations—are performed
by calling functions provided as part of a library.
Unfortunately, the standard library functions for reading input data from the user,
which are described in Chapter 3, can be a bit complicated in their operation and provide
more power than you need at this point. For most of the programs in this text, it is more
convenient to use functions defined in a simplified I/O library that I designed to make it
easier for beginning students to learn the important concepts of programming without
getting bogged down in extraneous details. To use this library, you need to add the
following line to the library-inclusion section at the beginning of your program:

#include "simpio.h"

The simpio library defines the functions GetInteger , GetLong , GetReal , and
GetLine, which wait for the user to enter a line at the keyboard and then return a value of
type int, long, double, and string, respectively. To let the user know what value is
expected, it is conventional to display a message to the user, which is called a prompt,
before calling the input function. Thus, if you need to request a value from the user for
the integer variable n, you would typically use a pair of statements like this:
cout << "Enter an integer: ";
n = GetInteger();

Output operations in this book use the insertion operator <<. The operand on the left of
the operator is a stream, such as the standard output stream cout. The operand on the
right is the data that you wish to insert into the stream. Several insertions to the same
stream can be chained together as shown here:
cout << "The result is " << val << endl;

Stream manipulators can be used to control the formatting of the output. A manipulator
is inserted into the stream ahead of the value it affects. The manipulator does not print
anything to the stream, but changes the state of the stream such that subsequent insertions
will use the requested formatting. A few of the more common stream manipulators are
shown in Table 1-3. To use these manipulators, you must include the <iomanip>
interface file in the library-inclusion section of your program.
As an example of the use of the simple I/O facilities, the following main program reads
in three floating-point values and displays their average:
int main() {
cout << "This program averages three numbers." << endl;
cout << "1st number: ";
double n1 = GetReal();
cout << "2nd number: ";
double n2 = GetReal();
cout << "3rd number: ";
double n3 = GetReal();
double average = (n1 + n2 + n3) / 3;
cout << "The average is " << average << endl;
return 0;
}
An Overview of C++ – 16 –

Table 1-3 Common output stream maniuplators

setw(n) Sets the field width to n characters. If the value to be formatted is
too short to fill the entire field width, extra space is added. This
stream property is transient it affects only the next value inserted
into the stream.
setprecision(n) Sets the precision to n places. The precision indicates how many
digits should be displayed to the right of the decimal point. This
property is persistent—once set, it applies to all subsequent
values inserted into the stream.
left Sets the justification. For values that are smaller than their field
width, left justification will cause the values to line up along the
left, padding with spaces on the right. This property is persistent.
right Same as left, but to the right, padding with spaces on the left.

1.4 Expressions
Whenever you want a program to perform calculations, you need to write an expression
that specifies the necessary operations in a form similar to that used for expressions in
mathematics. For example, suppose that you wanted to solve the quadratic equation
ax2 + bx + c = 0
As you know from high-school mathematics, this equation has two solutions given by the
formula

–b ± √ b2 – 4ac

x = 2a
The first solution is obtained by using + in place of the ± symbol the second is obtained
by using – instead. In C++, you could compute the first of these solutions by writing the
following expression:
(-b + sqrt(b * b - 4 * a * c)) / (2 * a)

There are a few differences in form—multiplication is represented explicitly by a * ,

division is represented by a / , and the square root function is spelled out—but the
expression nonetheless captures the intent of its mathematical counterpart in a way that is
quite readable, particularly if you’ve written programs in any modern programming
language.
In C++, an expression is composed of terms and operators. A term, such as the
variables a , b , and c or the constants 2 and 4 in the preceding expression, represents a
single data value and must be either a constant, a variable, or a function call. An operator
is a character (or sometimes a short sequence of characters) that indicates a computational
operation. The complete list of operators available in C++ is shown in Table 1-4. The
table includes familiar arithmetic operators like + and - along with several others that
pertain only to types that are introduced in later chapters.

Precedence and associativity

The point of listing all the operators in a single table is to establish how they relate to one
another in terms of precedence, which is a measure of how tightly an operator binds to
its operands in the absence of parentheses. If two operators compete for the same
An Overview of C++ – 17 –

Table 1-4 Complete precedence table for C++ operators

() [] -> . left-associative
unary operators: - ++ -- ! & * ~ (type) sizeof right-associative
* / % left-associative
+ - left-associative
<< >> left-associative
< <= > >= left-associative
== != left-associative
& left-associative
^ left-associative
| left-associative
&& left-associative
|| left-associative
?: right-associative
= op= right-associative
, left-associative

operand, the one that appears higher in the precedence table is applied first. Thus, in the
expression
(-b + sqrt(b * b - 4 * a * c)) / (2 * a)

the multiplications (b * b and 4 * a * c) are performed before the subtraction because *

has a higher precedence than - . It is, however, important to note that the - operator
occurs in two forms. When it is written between two operands, it is a binary operator
representing subtraction. When it is written in front of a single operand, as in is -b, it is a
unary operator representing negation. The precedence of the unary and binary versions
of an operator are different and are listed separately in the precedence table.
If two operators have the same precedence, they are applied in the order specified by
their associativity, which indicates whether that operator groups to the left or to the right.
Most operators in C++ are left-associative, which means that the leftmost operator is
evaluated first. A few operators—primarily the assignment operator, which is discussed
in more detail later in this chapter—are right-associative, which mean that they are
evaluated from right to left. The associativity for each operator appears in Table 1-4.
The quadratic formula illustrates the importance of paying attention to associativity
rules. Consider what would happen if you wrote the expression without the parentheses
around 2 * a, as follows:

(-b + sqrt(b * b - 4 * a * c)) / 2 * a Should be (2 * a)

Without the parentheses, the division operator would be performed first because / and *
have the same precedence and associate to the left.

Mixing types in an expression

In C++, you can write an expression that includes values of different numeric types. If
C++ encounters an operator whose operands are of different numeric types, the compiler
An Overview of C++ – 18 –

automatically converts the operands to a common type by determining which of the two
operand types appears closest to the top in Table 1-5. The result of applying the
operation is always that of the arguments after any conversions are applied. This
convention ensures that the result of the computation is as precise as possible.
As an example, suppose that n is declared as an int, and x is declared as a double.
The expression
n + 1

is evaluated using integer arithmetic and produces a result of type int. The expression
x + 1

however, is evaluated by converting the integer 1 to the floating-point value 1.0 and
adding the results together using double-precision floating-point arithmetic, which results
in a value of type double.
Integer division and the remainder operator
The fact that applying an operator to two integer operands generates an integer result
leads to an interesting situation with respect to the division operator. If you write an
expression like
9 / 4

C++’s rules specify that the result of this operation must be an integer, because both
operands are of type int . When C++ evaluates this expression, it divides 9 by 4 and
discards any remainder. Thus, the value of this expression in C++ is 2, not 2.25.
If you want to compute the mathematically correct result of 9 divided by 4, at least one
of the operands must be a floating-point number. For example, the three expressions
9.0 / 4
9 / 4.0
9.0 / 4.0

each produce the floating-point value 2.25. The decimal fraction is thrown away only if
both operands are of type int. The operation of discarding a decimal fraction is called
truncation.
There is an additional arithmetic operator that computes a remainder, which is
indicated in C++ by the percent sign (%). The % operator returns the remainder when the
first operand is divided by the second, and requires that both operands be of one of the

Table 1-5 Type conversion hierarchy for numeric types

long double most precise

double
float
unsigned long
long
unsigned int
int
unsigned short
short
char least precise
An Overview of C++ – 19 –

integer types. For example, the value of

9 % 4

is 1, since 4 goes into 9 twice, with 1 left over. The following are some other examples
of the % operator:
0 % 4 = 0 19 % 4 = 3
1 % 4 = 1 20 % 4 = 0
4 % 4 = 0 2001 % 4 = 1

The / and % operators turn out to be extremely useful in a wide variety of

programming applications. The % operator, for example, is often used to test whether one
number is divisible by another. For example, to determine whether an integer n is
divisible by 3, you just check whether the result of the expression n % 3 is 0.
It is, however, important to use caution if either or both of the operands to / and %
might be negative, because the results may differ from machine to machine. On most
machines, division truncates its result toward 0, but this behavior is not actually
guaranteed by the ANSI standard. In general, it is good programming practice to avoid
using these operators with negative values.

Type casts
In C++, you can specify explicit conversion by using what is called a type cast, a unary
operator that consists of the desired type followed by the value you wish to convert in
parentheses. For example, if num and den were declared as integers, you could compute
the floating-point quotient by writing
quotient = double(num) / den;

The first step in evaluating the expression is to convert num to a double, after which the
division is performed using floating-point arithmetic as described in the section on
“Mixing types in an expression” earlier in this chapter.
As long as the conversion moves upward in the hierarchy shown in Table 1-5, the
conversion causes no loss of information. If, however, you convert a value of a more
precise type to a less precise one, some information may be lost. For example, if you use
a type cast to convert a value of type double to type int, any decimal fraction is simply
dropped. Thus, the value of the expression
int(1.9999)

is the integer 1.

The assignment operator

In C++, assignment of values to variables is built into the expression structure. The =
operator takes two operands, just like + or *. The left operand must indicate a value that
can change, which is typically a variable name. When the assignment operator is
executed, the expression on the right-hand side is evaluated, and the resulting value is
then stored in the variable that appears on the left-hand side. Thus, if you evaluate an
expression like
result = 1
An Overview of C++ – 20 –

the effect is that the value 1 is assigned to the variable result. In most cases, assignment
expressions of this sort appear in the context of simple statements, which are formed by
adding a semicolon after the expression, as in the line
result = 1;

that appears in the powertab.cpp program. Such statements are often called assignment
statements, although they in fact have no special status in the language definition.
The assignment operator converts the type of the value on the right-hand side so that it
matches the declared type of the variable. Thus, if the variable total is declared to be of
type double, and you write the assignment statement
total = 0;

the integer 0 is converted into a double as part of making the assignment. If n is declared
to be of type int, the assignment
n = 3.14159265;

has the effect of setting n to 3, because the value is truncated to fit in the integer variable.
Even though assignment operators usually occur in the context of simple statements,
they can also be incorporated into larger expressions, in which case the result of applying
the assignment operator is simply the value assigned. For example, the expression
z = (x = 6) + (y = 7)

has the effect of setting x to 6, y to 7, and z to 13. The parentheses are required in this
example because the = operator has a lower precedence than +. Assignments that are
written as part of larger expressions are called embedded assignments.
Although there are contexts in which embedded assignments are extremely convenient,
they often make programs more difficult to read because the assignment is easily
overlooked in the middle of a complex expression. For this reason, this text limits the use
of embedded assignments to a few special circumstances in which they seem to make the
most sense. Of these, the most important is when you want to set several variables to the
same value. C++’s definition of assignment as an operator makes it possible, instead of
writing separate assignment statements, to write a single statement like
n1 = n2 = n3 = 0;

which has the effect of setting all three variables to 0. This statement works because C++
evaluates assignment operators from right to left. The entire statement is therefore
equivalent to
n1 = (n2 = (n3 = 0));

The expression n3 = 0 is evaluated, which sets n3 to 0 and then passes 0 along as the
value of the assignment expression. That value is assigned to n2, and the result is then
assigned to n1. Statements of this sort are called multiple assignments.
As a programming convenience, C++ allows you to combine assignment with a binary
operator to produce a form called a shorthand assignment. For any binary operator op,
the statement
variable op= expression;
An Overview of C++ – 21 –

is equivalent to
variable = variable op (expression);

where the parentheses are required only if the expression contains an operator whose
precedence is lower than that of op. Thus, the statement
balance += deposit;

is a shorthand for
balance = balance + deposit;

which adds deposit to balance.

Because this same shorthand applies to any binary operator in C++, you can subtract
the value of surcharge from balance by writing
balance -= surcharge;

Similarly, you can divide the value of x by 10 using

x /= 10;

or double the value of salary by using

salary *= 2;

Increment and decrement operators

Beyond the shorthand assignment operators, C++ offers a further level of abbreviation for
two particularly common programming operations—adding or subtracting 1 from a
variable. Adding 1 to a variable is called incrementing it subtracting 1 is called
decrementing it. To indicate these operations in an extremely compact form, C++ uses
the operators ++ and --. For example, the statement
x++;

in C++ has the same ultimate effect as

x += 1;

which is itself short for

x = x + 1;
Similarly,
y--;

has the same effect as

y -= 1;
or
y = y - 1;

As it happens, these operators are more intricate than the previous examples would
suggest. To begin with, each of these operators can be written in two ways. The operator
An Overview of C++ – 22 –

can come after the operand to which it applies, as in

x++

or before the operand, as in

++x

The first form, in which the operator follows the operand, is called the postfix form, the
second, the prefix form.
If all you do is execute the ++ operator in isolation—as you do in the context of a
separate statement or a typical for loop like those in the powertab.cpp example—the
prefix and postfix operators have precisely the same effect. You notice the difference
only if you use these operators as part of a larger expression. Then, like all operators, the
++ operator returns a value, but the value depends on where the operator is written
relative to the operand. The two cases are as follows:
x++ Calculates the value of x first, and then increments it. The value
returned to the surrounding expression is the original value before
the increment operation is performed.
++x Increments the value of x first, and then uses the new value as the
value of the ++ operation as a whole.
The -- operator behaves similarly, except that the value is decremented rather than
incremented.
You may wonder why would anyone use such an arcane feature. The ++ and - -
operators are certainly not essential. Moreover, there are not many circumstances in
which programs that embed these operators in larger expressions are demonstrably better
than those that use a simpler approach. On the other hand, ++ and -- are firmly
entrenched in the historical tradition shared by C++ programmers. They are idioms, and
programmers use them frequently. Because these operators are so common, you need to
understand them so that you can make sense of existing code.

Boolean operators
C++ defines three classes of operators that manipulate Boolean data: the relational
operators, the logical operators, and the ?: operator. The ;relational operators are used
to compare two values. C++ defines six relational operators, as follows:
== Equal
!= Not equal
> Greater than COMMON PITFALLS
< Less than When writing programs that
>= Greater than or equal to test for equality, be sure to
<= Less than or equal to use the == operator and
not the single = operator,
When you write programs that test for equality, be careful to use the which signifies assignment.
== operator, which is composed of two equal signs. A single equal This error is extremely
sign is the assignment operator. Since the double equal sign violates common and can lead to
bugs that are very difficult
conventional mathematical usage, replacing it with a single equal sign to find, because the
is a particularly common mistake. This mistake can also be very compiler cannot detect the
difficult to track down because the C++ compiler does not usually error.
catch it as an error. A single equal sign usually turns the expression
An Overview of C++ – 23 –

into an embedded assignment, which is perfectly legal in C++; it just isn’t at all what you
want.
The relational operators can be used to compare atomic data values like integers,
floating-point numbers, Boolean values, and characters. Some of the types supplied in
the libraries, such as string, also can be compared using the relational operators.
In addition to the relational operators, C++ defines three logical operators that take
Boolean operands and combine them to form other Boolean values:
! Logical not (true if the following operand is false)
&& Logical and (true if both operands are true)
|| Logical or (true if either or both operands are true)

These operators are listed in decreasing order of precedence.

Although the operators &&, ||, and ! closely resemble the English words and, or, and
not, it is important to remember that English can be somewhat imprecise when it comes
to logic. To avoid that imprecision, it is often helpful to think of these operators in a
more formal, mathematical way. Logicians define these operators using truth tables,
which show how the value of a Boolean expression changes as the values of its operands
change. The following truth table illustrates the result for each of the logical operators,
given all possible values of the variables p and q:

p q p && q p || q !p
false false false false true
false true false true true
true false false true false
true true true true false

C++ interprets the && and || operators in a way that differs from the interpretation
used in many other programming languages such as Pascal. Whenever a C++ program
evaluates an expression of the form
exp1 && exp2

or
exp1 || exp2

the individual subexpressions are always evaluated from left to right, and evaluation ends
as soon as the answer can be determined. For example, if exp1 is false in the expression
involving && , there is no need to evaluate exp2 since the final answer will always be
false . Similarly, in the example using || , there is no need to evaluate the second
operand if the first operand is true. This style of evaluation, which stops as soon as the
answer is known, is called short-circuit evaluation.
The C++ programming language provides another Boolean operator that can be
extremely useful in certain situations: the ?: operator. (This operator is referred to as
question-mark colon, even though the two characters do not appear adjacent to one
another in the code.) Unlike any other operator in C++, ?: is written in two parts and
requires three operands. The general form of the operation is
(condition) ? exp1 : exp2
An Overview of C++ – 24 –

The parentheses around the condition are not technically required, but C++ programmers
often include them to emphasize the boundaries of the conditional test.
When a C++ program encounters the ?: operator, it first evaluates the condition. If the
condition turns out to be true , exp1 is evaluated and used as the value of the entire
expression if the condition is false , the value is the result of evaluating exp2 . For
example, you can use the ?: operator to assign to max either the value of x or the value of
y, whichever is greater, as follows:

max = (x > y) ? x : y;

1.5 Statements
Programs in C++ are composed of functions, which are made up in turn of statements.
As in most languages, statements in C++ fall into one of two principal classifications:
simple statements, which perform some action, and control statements, which affect
the way in which other statements are executed. The sections that follow review the
principal statement forms available in C++ and give you the fundamental tools you need
to write your own programs.

Simple statements
The most common statement in C++ is the simple statement, which consists of an
expression followed by a semicolon:

expression;

In most cases, the expression is a function call, an assignment, or a variable followed by

the increment or decrement operator.

Blocks
As C++ is defined, control statements typically apply to a single statement. When you
are writing a program, you often want the effect of a particular control statement to apply
to a whole group of statements. To indicate that a sequence of statements is part of a
coherent unit, you can assemble those statements into a block, which is a collection of
statements enclosed in curly braces, as follows:

{
statement1
statement2
. . .
statementn
}

When the C++ compiler encounters a block, it treats the entire block as a single
statement. Thus, whenever the notation statement appears in a pattern for one of the
control forms, you can substitute for it either a single statement or a block. To emphasize
that they are statements as far as the compiler is concerned, blocks are sometimes referred
to as compound statements. In C++, the statements in any block may be preceded by
declarations of variables. In this text, variable declarations are introduced only in the
block that defines the body of a function.
The statements in the interior of a block are usually indented relative to the enclosing
context. The compiler ignores the indentation, but the visual effect is extremely helpful
to the human reader, because it makes the structure of the program jump out at you from
An Overview of C++ – 25 –

the format of the page. Empirical research has shown that indenting three or four spaces
at each new level makes the program structure easiest to see; the programs in this text use
four spaces for each new level. Indentation is critical to good programming, so you
should strive to develop a consistent indentation style in your programs.
The only aspect of blocks that tends to cause any confusion for new students is the role
of the semicolon. In C++, the semicolon is part of the syntax of a simple statement; it
acts as a statement terminator rather than as a statement separator. While this rule is
perfectly consistent, it can cause trouble for people who have previously been exposed to
the language Pascal, which uses a different rule. In practical terms, the differences are:

1. In C++, there is always a semicolon at the end of the last simple statement in a block.
In Pascal, the semicolon is usually not present although most compilers allow it as an
option.
2. In C++, there is never a semicolon after the closing brace of a statement block. In
Pascal, a semicolon may or may not follow the END keyword, depending on context.

The convention for using semicolons in C++ has advantages for program maintenance
and should not cause any problem once you are used to it.

The if statement
In writing a program, you will often want to check whether some condition applies and
use the result of that check to control the subsequent execution of the program. This type
of program control is called conditional execution. The easiest way to express
conditional execution in C++ is by using the if statement, which comes in two forms:

if (condition) statement

if (condition) statement else statement

You use the first form of the if statement when your solution strategy calls for a set of
statements to be executed only if a particular Boolean condition is true. If the condition
is false, the statements that form the body of the if statement are simply skipped. You
use the second form of the if statement for situations in which the program must choose
between two independent sets of actions based on the result of a test. This statement
form is illustrated by the following program, which reads in a number and classifies it as
either even or odd.

int main() {
int n;

cout << "This program classifies a num even or odd." << endl;
cout << "Enter a number: ";
n = GetInteger();
if (n % 2 == 0) {
cout << "That number is even." << endl;
} else {
cout << "That number is odd." << endl;
}
return 0;
}

As with any control statement, the statements controlled by the if statement can be
either a single statement or a block. Even if the body of a control form is a single
An Overview of C++ – 26 –

statement, you are free to enclose it in a block if you decide that doing so improves the
readability of your code. The programs in this book enclose the body of every control
statement in a block unless the entire statement—both the control form and its body—is
so short that it fits on a single line.

The switch statement

The if statement is ideal for those applications in which the program logic calls for a
two-way decision point: some condition is either true or false, and the program acts
accordingly. Some applications, however, call for more complicated decision structures
involving several mutually exclusive cases: in one case, the program should do x; in
another case, it should do y; in a third, it should do z; and so forth. In many applications,
the most appropriate statement to use for such situations is the switch statement, which
has the following syntactic form:
switch (e) {
case c1:
statements
break;
case c2:
statements
break;
. . . more case clauses . . .
default:
statements
break;
}

The expression e is called the control expression. When the program executes a
switch statement, it evaluates the control expression and compares it against the values
c1, c 2 , and so forth,
each of which must be a constant. If one of the constants matches the
value of the control expression, the statements in the associated case clause are executed.
When the program reaches the break statement at the end of the clause, the operations
specified by that clause are complete, and the program continues with the statement that
follows the entire switch statement.
The default clause is used to specify what action occurs if none of the constants
match the value of the control expression. The default clause, however, is optional. If
none of the cases match and there is no default clause, the program simply continues on
with the next statement after the switch statement without taking any action at all. To
avoid the possibility that the program might ignore an unexpected case, it is good
programming practice to include a default clause in every switch statement unless you
are certain you have enumerated all the possibilities, even if the default clause is simply

default:
Error("Unexpected case value");

The Error function is part of the genlib library and provides a uniform way of
responding to errors. This function takes one string parameter, the error message. The
Error function does not return; after the error message is displayed, the program
terminates.
The code pattern I’ve used to illustrate the syntax of the switch statement deliberately
suggests that break statements are required at the end of each clause. In fact, C++ is
defined so that if the break statement is missing, the program starts executing statements
An Overview of C++ – 27 –

from the next clause after it finishes the selected one. While this design can be useful in
some cases, it causes many more problems than it solves. To reinforce the importance of
remembering to exit at the end of each case clause, the programs in this text always
include a break or return statement in each such clause.
The one exception to this rule is that multiple case lines specifying COMMON PITFALLS
different constants can appear together, one after another, before the
same statement group. For example, a switch statement might It is good programming
practice to include a break
include the following code: statement at the end of
every case clause within a
case 1:
switch statement. Doing
case 2:
so will help you avoid
statements
programming errors that
break;
can be extremely difficult to
which indicates that the specified statements should be executed if the find. It is also good
practice to include a
select expression is either 1 or 2. The C++ compiler treats this
default clause unless
construction as two case clauses, the first of which is empty. Because you are sure you have
the empty clause contains no break statement, a program that selects covered all the cases.
the first path simply continues on with the second clause. From a
conceptual point of view, however, you are better off if you think of
this construction as a single case clause representing two possibilities.
The operation of the switch statement is illustrated by the following function, which
computes the number of days for a given month and year:
int MonthDays(int month, int year) {
switch (month) {
case September:
case April:
case June:
case November:
return 30;
case February:
return (IsLeapYear(year)) ? 29 : 28;
default:
return 31;
}
}

The code assumes that there is a function IsLeapYear(year) which tests whether year
is a leap year and that the names of the months have been defined using constants, as
follows:
const int JANUARY = 1;
const int FEBRUARY = 2;
const int MARCH = 3;
const int APRIL = 4;
const int MAY = 5;
const int JUNE = 6;
const int JULY = 7;
const int AUGUST = 8;
const int SEPTEMBER = 9;
const int OCTOBER = 10;
const int NOVEMBER = 11;
const int DECEMBER = 12;
An Overview of C++ – 28 –

The constants in a switch statement must be of integer type or a type that behaves like
an integer. (The actual restriction is that the type must be a scalar type, which is defined
in Chapter 2.) In particular, characters are often used as case constants, as illustrated by
the following function, which tests to see if its argument is a vowel:
bool IsVowel(char ch) {
switch (ch) {
case 'A': case 'E': case 'I': case 'O': case 'U':
case 'a': case 'e': case 'i': case 'o': case 'u':
return true;
default:
return false;
}
}

The while statement

In addition to the conditional statements if and switch, C++ includes several control
statements that allow you to execute some part of the program multiple times to form a
loop. Such control statements are called iterative statements. The simplest iterative
construct in C++ is the while statement, which executes a statement repeatedly until a
conditional expression becomes false. The general form for the while statement looks
like this:
while (conditional-expression) {
statements
}

When a program encounters a while statement, it first evaluates the conditional

expression to see whether it is true or false. If it is false, the loop terminates and the
program continues with the next statement after the entire loop. If the condition is true,
the entire body is executed, after which the program goes back to the beginning of the
loop to check the condition again. A single pass through the statements in the body
constitutes a cycle of the loop.
There are two important principles about the operation of a while loop:
1. The conditional test is performed before every cycle of the loop, including the first. If
the test is false initially, the body of the loop is not executed at all.
2. The conditional test is performed only at the beginning of a loop cycle. If that
condition happens to become false at some point during the loop, the program
doesn’t notice that fact until a complete cycle has been executed. At that point, the
program evaluates the test condition again. If it is still false, the loop terminates.
The operation of the while loop is illustrated by the following function, which
computes the sum of the digits in an integer:
int DigitSum(int n) {
int sum;
sum = 0;
while (n > 0) {
sum += n % 10;
n /= 10;
}
return sum;
}
An Overview of C++ – 29 –

The function depends on the following observations:

• The expression n % 10 always returns the last digit in a positive integer n.
• The expression n / 10 returns a number without its final digit.
The while loop is designed for situations in which there is some test condition that can
be applied at the beginning of a repeated operation, before any of the statements in the
body of the loop are executed. If the problem you are trying to solve fits this structure,
the while loop is the perfect tool. Unfortunately, many programming problems do not fit
easily into the standard while loop structure. Instead of allowing a convenient test at the
beginning of the operation, some problems are structured in such a way that the test you
want to write to determine whether the loop is complete falls most naturally somewhere
in the middle of the loop.
The most common example of such loops are those that read in data from the user until
some special value, or sentinel, is entered to signal the end of the input. When expressed
in English, the structure of the sentinel-based loop consists of repeating the following
steps:

1. Read in a value.
2. If the value is equal to the sentinel, exit from the loop.
3. Perform whatever processing is required for that value.

Unfortunately, there is no test you can perform at the beginning of the loop to determine
whether the loop is finished. The termination condition for the loop is reached when the
input value is equal to the sentinel; in order to check this condition, the program must
first read in some value. If the program has not yet read in a value, the termination
condition doesn’t make sense. Before the program can make any meaningful test, it must
have executed the part of the loop that reads in the input value. When a loop contains
some operations that must be performed before testing for completion, you have a
situation that programmers call the loop-and-a-half problem.
One way to solve the loop-and-a-half problem in C++ is to use the break statement,
which, in addition to its use in the switch statement, has the effect of immediately
terminating the innermost enclosing loop. By using break, it is possible to code the loop
structure for the sentinel problem in a form that follows the natural structure of the
problem:
while (true) {
Prompt user and read in a value.
if (value == sentinel) break;
Process the data value.
}

Note that the

while (true)

line itself seems to introduce an infinite loop because the value of the constant true can
never become false. The only way this program can exit from the loop is by executing
the break statement inside it. The loop-and-a-half strategy is illustrated by the
addlist.cpp program in Figure 1-4, which computes the sum of a list of integers
terminated by the sentinel value 0.
An Overview of C++ – 30 –

Figure 1-4 Program to add a list of integers

/*
* File: addlist.cpp
* -----------------
* This program adds a list of numbers. The end of the
* input is indicated by entering a sentinel value, which
* is defined by setting the value of the constant SENTINEL.
*/

#include <iostream>
#include "genlib.h"
#include "simpio.h"

/*
* Constant: SENTINEL
* ------------------
* This constant defines the value used to terminate the input
* list and should therefore not be a value one would want to
* include as a data value. The value 0 therefore makes sense
* for a program that adds a series of numbers because the
* user can simply skip any 0 values in the input.
*/
const int SENTINEL = 0;

/* Main program */
int main() {
cout << "This program adds a list of numbers." << endl;
cout << "Use " << SENTINEL << " to signal the end." << endl;
int total = 0;
while (true) {
cout << " ? ";
int value = GetInteger();
if (value == SENTINEL) break;
total += value;
}
cout << "The total is " << total << endl;
return 0;
}

There are other strategies for solving the loop-and-a-half problem that involve copying
part of the code outside the loop. However, empirical studies have demonstrated that
students are more likely to write correct programs if they use a break statement to exit
from the middle of the loop than if they are forced to use some other strategy. This
evidence and my own experience have convinced me that using the break statement
inside a while loop is the best solution to the loop-and-a-half problem.

The for statement

One of the most important control statements in C++ is the for statement, which is used
in situations in which you want to repeat an operation a particular number of times. The
general form is
for (init; test; step) {
statements
}
An Overview of C++ – 31 –

which is equivalent to the while statement

init;
while (test) {
statements
step;
}

The operation of the for loop is determined by the three italicized expressions on the
for control line: init, test, and step. The init expression indicates how the for loop should
be initialized and usually declares and initializes the index variable. For example, if you
write
for (int i = 0; . . .

the loop will begin by setting the index variable i to 0. If the loop begins
for (int i = -7; . . .

the variable i will start as -7, and so on.

Note the initialization expression does not have to also declare the index variable. It
may instead just initialize an already declared variable, but typically it is convenient to
place the declaration inside the for loop itself.1
The test expression is a conditional test written exactly like the test in a while
statement. As long as the test expression is true, the loop continues. Thus, the loop
for (int i = 0; i < n; i++)

begins with i equal to 0 and continues as long as i is less than n , which turns out to
represent a total of n cycles, with i taking on the values 0, 1, 2, and so forth, up to the
final value n–1. The loop
for (int i = 1; i <= n; i++)

begins with i equal to 1 and continues as long as i is less than or equal to n. This loop
also runs for n cycles, with i taking on the values 1, 2, and so forth, up to n.
The step expression indicates how the value of the index variable changes from cycle to
cycle. The most common form of step specification is to increment the index variable
using the ++ operator, but this is not the only possibility. For example, one can count
backward by using the -- operator, or count by twos by using += 2 instead of ++.
As an illustration of counting in the reverse direction, the program
int main() {
for (int t = 10; t >= 0; t--) {
cout << t << endl;
}
cout << "Liftoff!" << endl;
return 0;
}

1 The C++ standard states that the scope of an index variable declared in the initialization expression of a
for loop extends to the end of the loop body and no further. Some C++ compilers, such as Visual C++ 6.0,
mistakenly extend the scope to the end of the block enclosing the entire loop.
An Overview of C++ – 32 –

generates the following sample run:

Countdown
10
9
8
7
6
5
4
3
2
1
0
Liftoff!

The expressions init, test, and s t e p in a for statement are each optional, but the
semicolons must appear. If init is missing, no initialization is performed. If test is
missing, it is assumed to be true . If step is missing, no action occurs between loop
cycles.

1.6 Functions
A function consists of a set of statements that have been collected together and given a
name. The act of using the name to invoke the associated statements is known as calling
that function. To indicate a function call in C++, you write the name of the function,
followed by a list of expressions enclosed in parentheses. These expressions, called
arguments, allow the calling program to pass information to the function. For example,
in the p o w e r t a b . c p p program at the beginning of this chapter, the function
RaiseIntToPower took two integer arguments, n and k, which are the values it needs to
know in order to compute nk. If a function requires no information from its caller, it need
not have any arguments, but an empty set of parentheses must appear in the function call.
Once called, the function takes the data supplied as arguments, does its work, and then
returns to the program step from which the call was made. Remembering what the
calling program was doing and being able to get back precisely to that point is one of the
defining characteristics of the function-calling mechanism. The operation of going back
to the calling program is called returning from the function.

Returning results from functions

As they return, functions can send results back to the calling program. Thus, when the
RaiseIntToPower function returns with the statement

return result;

the value of the local variable result is passed back to the main program as the value of
the function. This operation is called returning a value.
Functions can return values of any type. The following function, for example, returns
a value of type bool, which can then be used in conditional expressions:
bool IsLeapYear(int year) {
return ((year % 4 == 0) && (year % 100 != 0))
|| (year % 400 == 0);
}
An Overview of C++ – 33 –

Functions that return Boolean results play an important role in programming and are
called predicate functions.
Functions, however, do not need to return a value at all. A function that does not
return a value and is instead executed for its effect is called a procedure. Procedures are
indicated in the definition of a function by using the reserved word void as the result
type.
The return statement in C++ has two forms. For procedures, you write the statement
as

return;

For functions that return a value, the return keyword is followed by an expression, as
follows:

return expression;

Executing either form of the return statement causes the current function to return
immediately to its caller, passing back the value of the expression, if any, to its caller as
the value of the function.

Function definitions and prototypes

A function definition has the following syntactic form:
result-type name(parameter-list) {
. . . body . . .
}

In this example, result-type is the type of value returned by the function, name is the
function name, and parameter-list is a list of declarations separated by commas, giving the
type and name of each parameter to the function. Parameters are placeholders for the
arguments supplied in the function call and act like local variables except for the fact that
they are given initial values by the calling program. If a function takes no parameters, the
entire parameter list in the function header line is empty. The body of the function is a
block and typically contains declarations for the local variables required by the function.
Before you use a function in a C++ program, you declare it by specifying its prototype.
A prototype has exactly the same form as a function definition, except that the entire
body is replaced by a semicolon. The names of the parameter variables are optional in a
prototype, but supplying them usually helps the reader.

The mechanics of the function-calling process

When you call a function in a program, the following steps occur:

1. The calling program computes values for each argument. Because the arguments are
expressions, this computation can involve operators and other functions, all of which
are evaluated before execution of the new function actually begins.
2. The system creates new space for all the local variables required by the new function,
including the parameter variables. These variables are usually allocated together in a
block, which is called a stack frame.
3. The value of each argument is copied into the corresponding parameter variable. If
there is more than one argument, the arguments are copied into the parameters in
An Overview of C++ – 34 –

order; the first argument is copied into the first parameter, and so forth. If necessary,
type conversions are performed between the argument values and the parameter
variables, as in an assignment statement. For example, if you pass a value of type int
to a function that expects a parameter of type double, the integer is converted into the
equivalent floating-point value before it is copied into the parameter variable.
4. The statements in the function body are executed until a return statement is
encountered or there are no more statements to execute.
5. The value of the return expression, if any, is evaluated and returned as the value of
the function. If the value being returned does not precisely match the result type
declared for the function, a type conversion is performed. Thus, if a return
statement specifies a floating-point value in a function defined to return an int, the
result is truncated to an integer.
6. The stack frame created for this function call is discarded. In the process, all local
variables disappear.
7. The calling program continues, with the returned value substituted in place of the call.

Passing parameters by reference

In C++, whenever you pass a simple variable from one function to another, the function
gets a copy of the calling value. Assigning a new value to the parameter as part of the
function changes the local copy but has no effect on the calling argument. For example,
if you try to implement a function that initializes a variable to zero using the code

void SetToZero(int var) {

var = 0; This function has no effect.
}

the function has no effect whatever. If you call

SetToZero(x);

the parameter var is initialized to a copy of whatever value is stored in x . The

assignment statement
var = 0;

inside the function sets the local copy to 0 but leaves x unchanged in the calling program.
To address this problem, you can change the parameter to a reference parameter by
adding an ampersand to the parameter declaration in the function header. Now the
parameter value will not be copied, instead a reference is made to the original variable.
Changes to the parameter are reflected in the original variable. The new coding is
void SetToZero(int & var) {
var = 0;
}

To use this function, the caller must pass an assignable integer variable. To set x to 0,
for example, you would need to make the following call:
SetToZero(x);

Passing an integer constant such as 3 would be an error because 3 is not an assignable

integer variable.
An Overview of C++ – 35 –

The use of reference parameters makes it possible for functions to change values in the
frame of their caller. This mechanism is referred to as call by reference.
In C++, one of the common uses of call by reference occurs when a function needs to
return more than one value to the calling program. A single result can easily be returned
as the value of the function itself. If you need to return more than one result from a
function, the return value is no longer appropriate. The standard approach to solving the
problem is to turn that function into a procedure and pass values back and forth through
the argument list.
As an example, suppose that you wanted to write a program to solve the quadratic
equation
ax2 + bx + c = 0
but that—because of your desire to practice good programming style—you were
committed to dividing the work of the program into three phases as represented by the
boxes in the following flowchart:

Input phase:
Accept values of
the coefficients
from the user.

Computation phase:
Solve the quadratic
equation for those
coefficients.

Output phase:
Display the roots
of the equation
on the screen.

Decomposing this problem into separate functions that are responsible for each of these
phases is somewhat tricky because several values must be passed from each phase to the
next. Because there are three coefficients, you would like the input phase to return three
values. Similarly, the computation phase must return two values, because a quadratic
equation has two solutions.
Figure 1-5 shows how call by reference makes it possible to decompose the quadratic
equation problem in this way. At each level, parameters that act as input to each function
are passed in the conventional way; parameters that represent output from the function
are passed by reference.

Stepwise refinement
Procedures and functions enable you to divide a large programming problem into smaller
pieces that are individually easy to understand. The process of dividing a problem into
manageable pieces, called decomposition, is a fundamental programming strategy.
Finding the right decomposition, however, turns out to be a difficult task that requires
considerable practice. If you choose the individual pieces well, each one will have
An Overview of C++ – 36 –

Figure 1-5 Implementation of quadeq.cpp that illustrates call by reference

/*
* File: quadeq.cpp
* ----------------
* This program finds roots of the quadratic equation
*
* 2
* a x + b x + c = 0
*
* If a is 0 or if the equation has no real roots, the
* program exits with an error.
*/

#include <iostream>
#include <cmath>
#include "genlib.h"
#include "simpio.h"

/* Private function prototypes */

void GetCoefficients(double & a, double & b, double & c);

void SolveQuadratic(double a, double b, double c,
double & x1, double & x2);
void DisplayRoots(double x1, double x2);

/* Main program */
int main() {
double a, b, c, r1, r2;

GetCoefficients(a, b, c);
SolveQuadratic(a, b, c, r1, r2);
DisplayRoots(r1, r2);
return 0;
}

/*
* Function: GetCoefficients
* Usage: GetCoefficients(a, b, c);
* --------------------------------
* This function is responsible for reading in the coefficients
* of a quadratic equation. The values of the coefficients are
* passed back to the main program in the variables a, b, and c,
* which are reference parameters.
*/
void GetCoefficients(double & a, double & b, double & c) {
cout << "Enter coefficients for the quadratic equation:" << endl;
cout << "a: ";
a = GetReal();
cout << "b: ";
b = GetReal();
cout << "c: ";
c = GetReal();
}
An Overview of C++ – 37 –

/*
* Function: SolveQuadratic
* Usage: SolveQuadratic(a, b, c, x1, x2);
* ---------------------------------------
* This function solves a quadratic equation. The coefficients
* are supplied as the arguments a, b, and c, and the roots are
* returned in x1 and x2, which are reference parameters.
*/

void SolveQuadratic(double a, double b, double c,

double & x1, double & x2) {
if (a == 0) Error("The coefficient a must be nonzero");
double disc = b * b - 4 * a * c;
if (disc < 0) Error("The solutions are complex numbers");
double sqrtDisc = sqrt(disc);
x1 = (-b + sqrtDisc) / (2 * a);
x2 = (-b - sqrtDisc) / (2 * a);
}

/*
* Function: DisplayRoots
* Usage: DisplayRoots(x1, x2);
* ----------------------------
* This function displays the values x1 and x2, which are
* the roots of a quadratic equation.
*/

void DisplayRoots(double x1, double x2) {

if (x1 == x2) {
cout << "There is a double root at " << x1 << endl;
} else {
cout << "The roots are " << x1 << " and " << x2 << endl;
}
}

conceptual integrity as a unit and make the program as a whole much simpler to
understand. But if you choose unwisely, the decomposition can get in your way. There
are no hard-and-fast rules for selecting a particular decomposition; you will learn how to
apply this process through experience.
When you are faced with the task of writing a program, the best strategy is usually to
start with the main program. At this level, you think about the problem as a whole and
then try to identify the major pieces of the entire task. Once you figure out what the big
pieces of the program are, you can define them as independent functions. Since some of
these functions may themselves be complicated, it is often appropriate to decompose
them into still smaller ones. You can continue this process until every piece of the
problem is simple enough to be solved on its own. This process is called top-down
design, or stepwise refinement.

Summary
This chapter is itself a summary, which makes it hard to condense it to a few central
points. Its purpose was to introduce you to the C++ programming language and give you
a crash course in how to write simple programs in that language. This chapter
concentrated on the low-level structure of the language, proceeding in turn through the
An Overview of C++ – 38 –

topics of expressions, statements, and functions. The facilities that C++ offers for
defining new data structures are detailed in Chapter 2.
Important points in this chapter include:
• In the 25 years of its existence, the C++ programming language has become one of the
most widely used languages in the world.
• A typical C++ program consists of comments, library inclusions, program-level
definitions, function prototypes, a function named main that is called when the
program is started, and a set of auxiliary function definitions that work together with
the main program to accomplish the required task.
• Variables in a C++ program must be declared before they are used. Most variables in
C++ are local variables, which are declared within a function and can only be used
inside the body of that function.
• A data type is defined by a domain of values and a set of operations. C++ includes
several predefined types that allow programs to store data of several different types,
such as integers, floating-point numbers, Booleans, and characters. In addition to these
built-in types, the standard library defines the type string, which is treated in this
book as if it were an integral part of the language.
• The easiest way to read input data from the user is to call functions in the simplified
I/O library (simpio ), which defines such functions as GetInteger, GetReal, and
GetLine. To display output on the computer screen, the usual approach is to insert the
values into the standard cout stream using the insertion operator <<.
• Expressions in C++ are written in a form similar to that in most programming
languages, with individual terms connected by operators. A complete list of the
operators available in C++ appears in Table 1-4, which also indicates the relative
precedence of each operator.
• Statements in C++ fall into two classes: simple statements and control statements. A
simple statement consists of an expression—typically an assignment or a function
call—followed by a semicolon. The control statements described in this chapter are
the i f , switch , while , and for statements. The first two are used to express
conditional execution, while the last two are used to specify repetition.
• C++ programs are typically subdivided into several functions. Each function consists
of a sequence of statements that can be invoked by writing the name of the function,
followed by a list of arguments enclosed in parentheses. These arguments are copied
into the corresponding parameter variables inside the function. The function can
return a result to the caller by using the return statement and can share values using
reference parameters.

Review questions
1. What is the difference between a source file and an object file?
2. What characters are used to mark comments in a C++ program?
3. In an #include line, the name of the library header file can be enclosed in either
angle brackets of double quotation marks. What is the difference between the two
forms of punctuation?
4. How would you define a constant called CENTIMETERS_PER_INCH with the value
2.54?
An Overview of C++ – 39 –

5. What is the name of the function that must be defined in every C++ program?
6. What is the purpose of inserting endl into the output stream cout?
7. What four properties are established when you declare a variable?
8. Indicate which of the following are legal variable names in C++:
a. x g. total output
b. formula1 h. aReasonablyLongVariableName
c. average_rainfall i. 12MonthTotal
d. %correct j. marginal-cost
e. short k. b4hand
f. tiny l. _stk_depth
9. What are the two attributes that define a data type?
10. What is the difference between the types short, int, and long?
11. What does ASCII stand for?
12. List all possible values of type bool.
13. What statements would you include in a program to read a value from the user and
store it in the variable x, which is declared as a double?
14. Suppose that a function contains the following declarations:
int i;
double d;
char c;
string s;
Write a statement that displays the values of each of these variables on the screen.
15. Indicate the values and types of the following expressions:
a. 2 + 3 d. 3 * 6.0
b. 19 / 5 e. 19 % 5
c. 19.0 / 5 f. 2 % 7
16. What is the difference between the unary minus and the binary subtraction operator?
17. What does the term truncation mean?
18. Calculate the result of each of the following expressions:
a. 6 + 5 / 4 - 3
b. 2 + 2 * (2 * 2 - 2) % 2 / 2
c. 10 + 9 * ((8 + 7) % 6) + 5 * 4 % 3 * 2 + 1
d. 1 + 2 + (3 + 4) * ((5 * 6 % 7 * 8) - 9) - 10
19. How do you specify a shorthand assignment operation?
20. What is the difference between the expressions ++x and x++?
21. What does the term short-circuit evaluation mean?
An Overview of C++ – 40 –

22. Write out the general syntactic form for each of the following control statements: if,
switch, while, for.

23. Describe in English the general operation of the switch statement.

24. What is a sentinel?
25. What for loop control line would you use in each of the following situations?
a. Counting from 1 to 100
b. Counting by sevens starting at 0 until the number has more than two digits
c. Counting backward by twos from 100 to 0

26. What is a function prototype?

27. In your own words, describe what happens when you call a function in C++.
28. What is meant by the term stepwise refinement?

Programming exercises
1. Write a program that reads in a temperature in degrees Celsius and displays the
corresponding temperature in degrees Fahrenheit. The conversion formula is
9
F = 5 C + 32

2. Write a program that converts a distance in meters to the corresponding English

distance in feet and inches. The conversion factors you need are
1 inch = 0.0254 meters
1 foot = 12 inches
3. According to legend, the German mathematician Karl Friedrich Gauss (1777–1855)
began to show his mathematical talent at a very early age. When he was in
elementary school, Gauss was asked by his teacher to compute the sum of the
numbers between 1 and 100. Gauss is said to have given the answer instantly: 5050.
Write a program that computes the answer to the question Gauss’s teacher posed.
4. Write a program that reads in a positive integer N and then calculates and displays
the sum of the first N odd integers. For example, if N is 4, your program should
display the value 16, which is 1 + 3 + 5 + 7.
5. Write a program that reads in a list of integers from the user until the user enters the
value 0 as a sentinel. When the sentinel appears, your program should display the
largest value in the list, as illustrated in the following sample run:
FindLargest
¬This program finds the largest integer in a list.
Enter 0 to signal the end of the list.
? 17
? 42
? 11
? 19
? 35
? 0
The largest value is 42
An Overview of C++ – 41 –

6. Using the DigitSum function from the section entitled “The while statement” as a
model, write a program that reads in an integer and then displays the number that has
the same digits in the reverse order, as illustrated by this sample run:
ReverseInteger
This program reverses the digits in an integer.
Enter a positive integer: 123456789
The reversed integer is 987654321

To make sure your program can handle integers as large as the one shown in the
example, use the type long instead of int in your program.
7. Greek mathematicians took a special interest in numbers that are equal to the sum of
their proper divisors (a proper divisor of N is any divisor less than N itself). They
called such numbers perfect numbers. For example, 6 is a perfect number because
it is the sum of 1, 2, and 3, which are the integers less than 6 that divide evenly into
6. Similarly, 28 is a perfect number because it is the sum of 1, 2, 4, 7, and 14.
Write a predicate function IsPerfect that takes an integer n and returns true if n
is perfect, and false otherwise. Test your implementation by writing a main
program that uses the IsPerfect function to check for perfect numbers in the range
1 to 9999 by testing each number in turn. When a perfect number is found, your
program should display it on the screen. The first two lines of output should be 6
and 28. Your program should find two other perfect numbers in the range as well.
8. Every positive integer greater than 1 can be expressed as a product of prime
numbers. This factorization is unique and is called the prime factorization. For
example, the number 60 can be decomposed into the factors 2 x 2 x 3 x 5, each of
which is prime. Note that the same prime can appear more than once in the
factorization.
Write a program to display the prime factorization of a number n, as illustrated by
the following sample run:
Factor
Enter number to be factored: 60
2 * 2 * 3 * 5

9. When a floating-point number is converted to an integer in C++, the value is

truncated by throwing away any fraction. Thus, when 4.99999 is converted to an
integer, the result is 4. In many cases, it would be useful to have the option of
rounding a floating-point value to the nearest integer. For a positive floating-point
number x , the rounding operation can be achieved by adding 0.5 to x and then
truncating the result to an integer. If the decimal fraction of x is less than .5, the
truncated value will be the integer less than x ; if the fraction is .5 or more, the
truncated value will be the next larger integer. Because truncation always moves
toward zero, negative numbers must be rounded by subtracting 0.5 and truncating,
instead of adding 0.5.
Write a function Round(x) that rounds a floating-point number x to the nearest
integer. Show that your function works by writing a suitable main program to test it.
An Overview of C++ – 42 –

10. The German mathematician Leibniz (1646–1716) discovered the rather remarkable
fact that the mathematical constant π can be computed using the following
mathematical relationship:
π
– 1 + 1 – 1 + 1 – 1 + . . .
4 ≅ 1 3 5 7 9 11
The formula to the right of the equal sign represents an infinite series; each fraction
represents a term in that series. If you start with 1, subtract one-third, add one-fifth,
and so on, for each of the odd integers, you get a number that gets closer and closer
to the value of π/4 as you go along.
Write a program that calculates an approximation of π consisting of the first
10,000 terms in Leibniz’s series.

11. You can also approximate π by approximating the area bounded by a circular arc.
Consider the following quarter circle:

which has a radius r equal to two inches. From the formula for the area of a circle,
you can easily determine that the area of the quarter circle should be π square inches.
You can also approximate the area computationally by adding up the areas of a series
of rectangles, where each rectangle has a fixed width and the height is chosen so that
the circle passes through the midpoint of the top of the rectangle. For example, if
you divide the area into 10 rectangles from left to right, you get the following
diagram:

The sum of the areas of the rectangles approximates the area of the quarter circle.
The more rectangles there are, the closer the approximation.
For each rectangle, the width w is a constant derived by dividing the radius by the
number of rectangles. The height h, on the other hand, varies depending on the
position of the rectangle. If the midpoint of the rectangle in the horizontal direction
is given by x, the height of the rectangle can be computed using the distance formula
An Overview of C++ – 43 –

h = r 2 – x2
√

The area of each rectangle is then simply h x w.
Write a program to compute the area of the quarter circle by dividing it into 100
rectangles.

12. When you write a check, the dollar amount appears twice: once as a number and once
as English text. For example, if you write a check for $1729, you need to translate
that number to the English text “one thousand seven hundred twenty-nine.” Your task
in this problem is to write a program that reads in integers from the user and writes
out the equivalent value in figures on the next line, stopping when the user enters any
negative number. For example, the following is a sample run of this program:
NumberToText
Enter numbers in figures; use a negative value to stop.
Number: 0
zero
Number: 1
one
Number: 11
eleven
Number: 256
two hundred fifty-six
Number: 1729
one thousand seven hundred twenty-nine
Number: 2001
two thousand one
Number: 12345
twelve thousand three hundred forty-five
Number: 13000
thirteen thousand
Number: -1

The key idea in this exercise is decomposition. The problem is not nearly as hard as
it looks if you break it down into separate procedures that accomplish parts of the
task. Many of these procedures will have a form that looks something like this:
void PrintOneDigit(int d) {
switch (d) {
case 0: cout << "zero"; break;
case 1: cout << "one"; break;
case 2: cout << "two"; break;
case 3: cout << "three"; break;
case 4: cout << "four"; break;
case 5: cout << "five"; break;
case 6: cout << "six"; break;
case 7: cout << "seven"; break;
case 8: cout << "eight"; break;
case 9: cout << "nine"; break;
default: Error("Illegal call to PrintOneDigit");
}
}

In writing your program, you should keep the following points in mind:
• You don’t need to perform any string manipulation. All you have to do is display
the value on the screen, which means that inserting to cout is all you need.
An Overview of C++ – 44 –

• Your program need work only with values up to 999,999, although it should give
the user some kind of error message if a number is outside of its range.
• It is perfectly acceptable for all the letters in the output to be lowercase. The
problem is much harder if you try to capitalize the first word.
• You should remain on the lookout for functions that you can reuse. For example,
printing the number of thousands is pretty much the same as printing out the last
three digits, and you should be able to use the same procedure more than once.
• Several special cases arise in this problem. For example, the number 11 must be
treated differently than 21 or 31, because eleven doesn’t fit the pattern established
by twenty-one and thirty-one.
Chapter 2
Data Types in C++

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

Insensibly one begins to twist facts to suit theories, instead
of theories to suit facts.
— Sherlock Holmes, in Sir Arthur Conan
Doyle’s A Scandal in Bohemia, 1892
Data Types in C++ – 46 –

Chapter 1 of this text is a capsule summary of the features of C++ necessary to code the
algorithmic structure of a program. The algorithmic structure, however, represents only
part of the story. It is equally important to consider the structure of the data.
Like control statements and function calls—each of which can be nested hierarchically
to represent increasingly complex algorithmic structures—data types in a language also
form a hierarchy. The base of the hierarchy is composed of the atomic types that were
introduced in Chapter 1, coupled with a new class of atomic types called enumeration
types that are introduced in the following section. Starting from this base, you can extend
the hierarchy using the following mechanisms for creating new types from existing ones:

• Pointers. A pointer is simply the internal machine address of a value inside the
computer’s memory. C++ allows you to work directly with pointers as data and makes
them part of the type hierarchy, which means that you can define new types whose
domains consist of pointers to values of some existing type.
• Arrays. An array is an ordered collection of data values, each of which has the same
type.
• Records. A record is a collection of data values that represents some logically
coherent whole. The individual components are identified by name rather than by
order and may be of different types.

Each of these types is described in detail in a separate section later in this chapter. For
now, the main point is that you can combine these mechanisms to generate new types at
whatever level of complexity the program requires. You can, for example, create new
types that are pointers to records containing arrays or any other nested structure you
choose. The hierarchy of types in a program defines its data structure.

2.1 Enumeration types

Before moving on to the question of how to create new types from existing ones, it is
important to complete the set of atomic types that form the base of the type hierarchy.
The most common atomic types are the various built-in types described in Chapter 1:
integers, floating-point numbers, characters, and so on. Like many languages, however,
C++ makes it possible to define new atomic types by listing the elements that constitute
their domains. Such types are called enumeration types.
In this text, new enumeration types are defined using the following syntactic form:
enum name { element-list };

where element-list is a list of identifiers, which are called enumeration constants, and
name is the name of the new type. For example, the following enumeration type defines
the four principal compass directions:

enum directionT { North, East, South, West };

Similarly, the following definition introduces the type colorT, which consists of the six
primary and secondary colors available on a standard color monitor:
Data Types in C++ – 47 –

enum colorT {
Red, Green, Blue, Yellow, Cyan, Magenta
};
Once you have defined an enumeration type, you can declare variables of that type just
as you do with any of the built-in types. For example, the declaration

directionT dir;

declares the variable dir to be of type directionT, which means that it can take on any
of the values North, East, South, or West.

Internal representation of enumeration types

The values of an enumeration type are stored internally as integers. When the compiler
encounters a new enumeration type, it ordinarily assigns consecutive integers to the
enumeration constants, starting with the integer 0. Thus, in the directionT example, the
constants have the following values: North = 0, East = 1, South = 2, and West = 3. You
can, however, control the encoding used for enumeration types by writing an equal sign
and an integer constant after any of the element names. For example, after the definition

enum coinT {
Penny = 1,
Nickel = 5,
Dime = 10,
Quarter = 25,
HalfDollar = 50
};

each of the enumeration constants Penny, Nickel, Dime, Quarter, and HalfDollar is
represented internally as its corresponding monetary value. If the value of any
enumeration constant is not specified, the compiler simply adds one to the value of the
previous constant. Thus, in the definition,

enum monthT {
January = 1, February, March, April, May, June,
July, August, September, October, November, December
};

the constant January is represented internally as 1, February as 2, and so on.

Defining an enumeration type is in many ways similar to defining named constants. In
the monthT example, you could achieve a similar effect by making the following
definitions:

const int JANUARY = 1;

const int FEBRUARY = 2;
const int MARCH = 3;
const int APRIL = 4;
const int MAY = 5;
const int JUNE = 6;
const int JULY = 7;
const int AUGUST = 8;
const int SEPTEMBER = 9;
const int OCTOBER = 10;
const int NOVEMBER = 11;
const int DECEMBER = 12;
Data Types in C++ – 48 –

Inside the machine, the two strategies produce the same result: every element of the
enumeration type is represented by an integer code. From the programmer’s point of
view, however, defining separate enumeration types has these advantages:

• The programmer does not need to specify the internal codes explicitly.
• The fact that there is a separate type name often makes the program easier to read
because declarations can use a meaningful type name instead of the general-purpose
designation int.
• A C++ compiler does some rudimentary checking for enumeration types. For example,
an integer value cannot be assigned to an enum variable without a typecast, which
helps to draw attention to possible mistakes such as assigning a value out of range for
the enumeration.
• On many systems, programs that use enumeration types are easier to debug because
the compiler makes the names of the enumeration constants available to the debugger.
Thus, if you ask it about a value of type monthT, a well-designed debugger would be
able to display the value January instead of the integer constant 1.

Scalar types
In C++, enumeration types, characters, and the various representations of integers form a
more general type class called scalar types. When a value of a scalar type is used in a
C++ expression, the compiler automatically converts it to the integer used to represent
that value. The effect of this rule is that the operations you can perform on values of any
scalar type are the same as those for integers.

As an example, suppose that you want to write a function RightFrom(dir) that takes
a directionT and returns the direction you would be facing if you turned 90 degrees
from that starting position. Thus, RightFrom(North) should return East. Because the
directions appear in order as you move right around the compass points, turning right
corresponds arithmetically to adding one to the underlying value, except for
RightFrom(West), which has to generate 0 instead of 4 as the underlying value. As is
often the case with enumerated types that represent a value which is logically cyclical,
you can use the % operator to write a one-line implementation of RightFrom, as follows:

directionT RightFrom(directionT dir) {

return directionT((dir + 1) % 4);
}

C++ allows implicit conversion from enumeration type to integer, since every
enumeration value has a corresponding integer representation. However, there is no
implicit conversion in the other direction because most integer values do not have a
representation in a particular enumeration. In the above example, the enumeration type is
automatically converted to an integer when used in an arithmetic expression. Once you
have computed the resulting integer value, you must use an explicit typecast to return that
value as a directionT.

You can substitute scalar types in any context in which an integer might appear. For
example, a variable of an enumeration type can be used as the control expression in a
switch statement, so that you can define a function DirectionName(dir), which returns
the name of a direction as a string, like this:
Data Types in C++ – 49 –

string DirectionName(directionT dir) {

switch (dir) {
case North: return "North";
case East: return "East";
case South: return "South";
case West: return "West";
default: Error("Illegal direction value");
}
}

You can also use scalar types as index variables in for loops. For example, you can
cycle through each of the four directions using the following loop control line:

for (directionT dir = North; dir <= West; dir = directionT(dir+1))

2.2 Data and memory

Before you can understand C++’s type system in any detail, you need to know how
information is stored inside a computer. Every modern computer contains some amount
of high-speed internal memory that is its principal repository for information. In a typical
machine, that memory is built out of a special integrated-circuit chip called a RAM,
which stands for random-access memory. Random-access memory allows the program to
use the contents of any memory cell at any time. The technical details of how the RAM
chip operates are not important to most programmers. What is important is how the
memory is organized.

Bits, bytes, and words

At the most primitive level, all data values inside the computer are stored in the form of
fundamental units of information called bits. A bit is a unit of memory that has only two
possible states. If you think of the circuitry inside the machine as if it were a tiny light
switch, you might label those states as on and off. Historically, the word bit is a
contraction of binary digit, and it is therefore more common to label those states using
the symbols 0 and 1, which are the two digits used in the binary number system on which
computer arithmetic is based.
Since a single bit holds so little information, the bits themselves do not provide a
particularly convenient mechanism for storing data. To make it easier to store such
traditional types of information as numbers or characters, individual bits are collected
together into larger units that are then treated as integral units of storage. The smallest
such combined unit is called a byte and is large enough to hold a value of type char ,
which typically requires eight individual bits. On most machines, bytes are assembled
into larger structures called words, where a word is usually defined to be the size
required to hold a value of type int. Some machines use two-byte words (16 bits), some
use four-byte words (32 bits), and some use less conventional sizes.

Memory addresses
Within the memory system, every byte is identified by a numeric address. Typically, the
first byte in the computer is numbered 0, the second is numbered 1, and so on, up to the
number of bytes in the machine. For example, if your computer has four megabytes of
memory (which actually means 4 x 220 or 4,194,304 bytes), the addresses of the memory
cells would look like this:
Data Types in C++ – 50 –

1000

1001

1002

1003

4194301

4194302

4194303

Each byte of memory is large enough to hold one character. For example, if you were
to declare the character variable ch, the compiler would reserve one byte of storage for
that variable as part of the current function frame. Suppose that this byte happened to be
at address 1000. If the program then executed the statement

ch = 'A';

the internal representation of the character 'A' would be stored in location 1000. Since
the ASCII code for 'A' is 65, the resulting memory configuration would look like this:

1000 65 ch
1001

1002

1003

1004

1005

1006

1007

In most programming applications, you will have no way of predicting the actual
address at which a particular variable is stored. In the preceding diagram, the variable ch
is assigned to address 1000, but this choice is entirely arbitrary. Whenever your program
makes a function call, the variables within the function are assigned to memory locations,
but you have no way of predicting the addresses of those variables in advance. Even so,
you may find it useful to draw pictures of memory and label the individual locations with
addresses beginning at a particular starting point. These addresses—even though you
choose them yourself—can help you to visualize what is happening inside the memory of
the computer as your program runs.
Values that are larger than a single character are stored in consecutive bytes of
memory. For example, if an integer takes up four bytes on your computer, that integer
Data Types in C++ – 51 –

requires four consecutive bytes of memory and might therefore be stored in the shaded
area in the following diagram:

1000

1001

1002

1003

1004

1005

1006

1007

Data values requiring multiple bytes are identified by the address of the first byte, so the
integer represented by the shaded area is the word stored at address 1000.
When you write a C++ program, you can determine how much memory will be
assigned to a particular variable by using the sizeof operator. The sizeof operator
takes a single operand, which must be a type name enclosed in parentheses or an
expression. If the operand is a type, the sizeof operator returns the number of bytes
required to store a value of that type; if the operand is an expression, sizeof returns the
number of bytes required to store the value of that expression. For example, the
expression

sizeof(int)

returns the number of bytes required to store a value of type int. The expression

sizeof x

returns the number of bytes required to store the variable x.

2.3 Pointers
One of the principles behind the design of C++ was that programmers should have as
much access as possible to the facilities provided by the hardware itself. For this reason,
C++ makes the fact that memory locations have addresses visible to the programmer. A
data item whose value is an address in memory is called a pointer. In many high-level
programming languages, pointers are used sparingly because those languages provide
other mechanisms that eliminate much of the need for pointers; the Java programming
language, for example, hides pointers from the programmer altogether. In C++, pointers
are pervasive, and it is impossible to understand C++ programs without knowing how
pointers work.
In C++, pointers serve several purposes, of which the following are the most
important:

• Pointers allow you to refer to a large data structure in a compact way. Data structures
in a program can become arbitrarily large. No matter how large they grow, however,
the data structures still reside somewhere in the computer’s memory and therefore
have an address. Pointers allow you to use the address as a shorthand for the complete
Data Types in C++ – 52 –

value. Because a memory address typically fits in four bytes of memory, this strategy
offers considerable space savings when the data structures themselves are large.
• Pointers make it possible to reserve new memory during program execution. Up to
now, the only memory you could use in your programs was the memory assigned to
variables that you have declared explicitly. In many applications, it is convenient to
acquire new memory as the program runs and to refer to that memory using pointers.
This strategy is discussed in the section on “Dynamic allocation” later in this chapter.
• Pointers can be used to record relationships among data items. In advanced
programming applications, pointers are used extensively to model connections
between individual data values. For example, programmers often indicate that one
data item follows another in a conceptual sequence by including a pointer to the
second item in the internal representation of the first.

Using addresses as data values

In C++, any expression that refers to an internal memory location capable of storing data
is called an lvalue (pronounced “ell-value”). The l at the beginning of lvalue comes from
the observation that lvalues can appear on the left side of an assignment statement in
C++. For example, simple variables are lvalues because you can write a statement like

x = 1.0;

Many values in C++, however, are not lvalues. For example, constants are not lvalues
because a constant cannot be changed. Similarly, although the result of an arithmetic
expression is a value, it is not an lvalue, because you cannot assign a value to the result of
an arithmetic expression.
The following properties apply to lvalues in C++:

• Every lvalue is stored somewhere in memory and therefore has an address.

• Once it has been declared, the address of an lvalue never changes, even though the
contents of the lvalue may change.
• Depending on their data type, different lvalues require different amounts of memory.
• The address of an lvalue is a pointer value, which can be stored in memory and
manipulated as data.

Declaring pointer variables

As with all other variables in C++, you must declare pointer variables before you use
them. To declare a variable as a pointer, you precede its name with an asterisk (*) in a
standard declaration. For example, the line

int *p;

declares the variable p to be of the conceptual type pointer-to-int. Similarly, the line

char *cptr;

declares the variable cptr to be of type pointer-to- char. These two types—pointer-to-
int and pointer-to-char—are distinct in C++, even though each of them is represented
internally as an address. To use the value at that address, the compiler needs to know
how to interpret it and therefore requires that its type be specified explicitly. The type of
the value to which a pointer points is called the base type of that pointer. Thus, the type
pointer-to-int has int as its base type.
Data Types in C++ – 53 –

It is important to note that the asterisk used to indicate that a variable is a pointer
belongs syntactically with the variable name and not with the base type. If you use the
same declaration to declare two pointers of the same type, you need to mark each of the
variables with an asterisk, as in

int *p1, *p2;

The declaration

int *p1, p2;

declares p1 as a pointer to an integer, but declares p2 as an integer variable.

The fundamental pointer operations

C++ defines two operators that allow you to move back and forth between values and
pointers to those values:

& Address-of
* Value-pointed-to

The & operator takes an lvalue as its operand and returns the memory address in which
that lvalue is stored. The * operator takes a value of any pointer type and returns the
lvalue to which it points. This operation is called dereferencing the pointer. The *
operation produces an lvalue, which means that you can assign a value to a dereferenced
pointer.
The easiest way to illustrate these operators is by example. Consider the declarations
int x, y;
int *p1, *p2;

These declarations allocate memory for four words, two of type int and two of type
pointer-to- int . For concreteness, let’s suppose that these values are stored in the
machine addresses indicated by the following diagram:

1000
x
1004
y
1008
p1
1012
p2

You can assign values to x and y using assignment statements, just as you always have.
For example, executing the assignment statements

x = –42;
y = 163;

results in the following memory state:

Data Types in C++ – 54 –

1000
-42 x
1004
163 y
1008
p1
1012
p2

To initialize the pointer variables p1 and p2, you need to assign values that represent
the addresses of some integer objects. In C++, the operator that produces addresses is the
& operator, which you can use to assign the addresses of x and y to p 1 and p 2 ,
respectively:

p1 = &x;
p2 = &y;

These assignments leave memory in the following state:

1000
-42 x
1004
163 y
1008
1000 p1
1012
1004 p2

The arrows in the diagram are used to emphasize the fact that the values of the variables
p1 and p2 point to the cells indicated by the heads of the arrows. Drawing arrows makes
it much easier to understand how pointers work, but it is important to remember that
pointers are simply numeric addresses and that there are no arrows inside the machine.
To move from a pointer to the value it points to, you use the * operator. For example,
the expression

*p1

indicates the value in the memory location to which p1 points. Moreover, since p1 is
declared as a pointer to an integer, the compiler knows that the expression *p1 must refer
to an integer. Thus, given the configuration of memory illustrated in the diagram, *p1
turns out to be another name for the variable x.
Like the simple variable name x, the expression *p1 is an lvalue, and you can assign
new values to it. Executing the assignment statement

*p1 = 17;

changes the value in the variable x because that is where p1 points. After you make this
assignment, the memory configuration is
Data Types in C++ – 55 –

1000
17 x
1004
163 y
1008
1000 p1
1012
1004 p2

You can see that the value of p1 itself is unaffected by this assignment. It continues to
hold the value 1000 and therefore still points to the variable x.
It is also possible to assign new values to the pointer variables themselves. For
instance, the statement
p1 = p2;

instructs the computer to take the value contained in the variable p2 and copy it into the
variable p1. The value contained in p2 is the pointer value 1004. If you copy this value
into p1, both p1 and p2 point to the variable y, as the following diagram shows:

1000
17 x
1004
163 y
1008
1004 p1
1012
1004 p2

In terms of the operations that occur inside the machine, copying a pointer is exactly the
same as copying an integer. The value of the pointer is simply copied unchanged to the
destination. From a conceptual perspective, the diagram shows that the effect of copying
a pointer is to replace the destination pointer with a new arrow that points to the same
location as the old one. Thus, the effect of the assignment

p1 = p2;

is to change the arrow leading from p1 so that it points to the same memory address as
the arrow originating at p2.
It is important to be able to distinguish the assignment of a pointer from that of a value.
Pointer assignment, such as
p1 = p2;

makes p1 and p2 point to the same location. Value assignment, which is represented by
the statement

*p1 = *p2;

copies the value from the memory location addressed by p2 into the location addressed
by p1.
Data Types in C++ – 56 –

The special pointer NULL

In many pointer applications, it is useful to be able to store in a pointer variable a special
value indicating that the variable does not in fact point to any valid data, at least for the
present. Such a pointer variable is called a null pointer and its value is assigned the
address 0. The special constant called NULL can be used for this purpose.1
It is important not to dereference a null pointer with the * operator. The intent of the
null value is to indicate that the pointer does not point to valid data, so the idea of trying
to find the data associated with a null pointer does not really make sense. Unfortunately,
most compilers do not produce programs that explicitly check for this error. If you
dereference a null pointer, many computers will interpret the 0 value as an address and
return whatever data value is stored at address 0. If you happen to change that value by
performing value assignment through a null pointer, the program can easily crash, giving
no hint of the cause. The same problems can arise if you use pointer variables whose
values have not yet been initialized.
The uses of the null pointer will be introduced in this text as they become relevant to a
particular application. For now, the important thing to remember is that this constant
exists.

2.4 Arrays
An array is a collection of individual data values with two distinguishing characteristics:

1. An array is ordered. You must be able to count off the individual components of an
array in order: here is the first, here is the second, and so on.
2. An array is homogeneous. Every value stored in an array must be of the same type.
Thus, you can define an array of integers or an array of floating-point numbers but
not an array in which the two types are mixed.

From an intuitive point of view, it is best to think of an array as a sequence of boxes, one
box for each data value in the array. Each of the values in an array is called an element.
For example, the following diagram represents an array with five elements:

In C++, each array has two fundamental properties:

• The element type, which is the type of value that can be stored in the elements of the
array
• The array size, which is the number of elements the array contains

Whenever you create a new array in your program, you must specify both the element
type and the array size.

Array declaration
Like any other variable in C++, an array must be declared before it is used. The general
form for an array declaration is

type name[size];

1 The constant NULL is defined in the <cstddef> header file. You may also just use the constant zero.
Data Types in C++ – 57 –

where type is the type of each element in the array, name is the name of the array variable,
and size is a constant indicating the number of elements allocated to the array. For
example, the declaration

int intArray[10];

declares an array named intArray with 10 elements, each of which is of type int. In
most cases, however, you should specify the size as a symbolic constant rather than an
explicit integer so that the array size is easier to change. Thus, a more conventional
declaration would look like this:

const int N_ELEMENTS = 10;

int intArray[N_ELEMENTS];

You can represent this declaration pictorially by drawing a row of ten boxes and giving
the entire collection the name intArray:

intArray

0 1 2 3 4 5 6 7 8 9

Each element in the array is identified by a numeric value called its index. In C++, the
index numbers for an array always begin with 0 and run up to the array size minus one.
Thus, in an array with 10 elements, the index numbers are 0, 1, 2, 3, 4, 5, 6, 7, 8, and 9, as
the preceding diagram shows.

As is the case with any variable, you use the name of an array to indicate to other
readers of the program what sort of value is being stored. For example, suppose that you
wanted to define an array that was capable of holding the scores for a sporting event, such
as gymnastics or figure skating, in which scores are assigned by a panel of judges. Each
judge rates the performance on a scale from 0 to 10, with 10 being the highest score.
Because a score may include a decimal fraction, as in 9.9, each element of the array must
be of some floating-point type, such as double. Thus the declaration

const int N_JUDGES = 5;

double scores[N_JUDGES];

declares an array named scores with five elements, as shown in the following diagram:

scores

0 1 2 3 4

Array selection
To refer to a specific element within an array, you specify both the array name and the
index corresponding to the position of that element within the array. The process of
identifying a particular element within an array is called selection, and is indicated in
C++ by writing the name of the array and following it with the index written in square
brackets. The result is a selection expression, which has the following form:
Data Types in C++ – 58 –

array[index]

Within a program, a selection expression acts just like a simple variable. You can use
it in an expression, and, in particular, you can assign a value to it. Thus, if the first judge
(judge #0, since C++ counts array elements beginning at zero) awarded the contestant a
score of 9.2, you could store that score in the array by writing the assignment statement

scores[0] = 9.2;

The effect of this assignment can be diagrammed as follows:

scores
9.2
0 1 2 3 4

You could then go ahead and assign scores for each of the other four judges using, for
example, the statements

scores[1] = 9.9;
scores[2] = 9.7;
scores[3] = 9.0;
scores[4] = 9.5;

Executing these statements results in the following picture:

scores
9.2 9.9 9.7 9.0 9.5
0 1 2 3 4

In working with arrays, it is essential to understand the distinction between the index
of an array element and the value of that element. For instance, the first box in the array
has index 0, and its value is 9.2. It is also important to remember that you can change the
values in an array but never the index numbers.
The real power of array selection comes from the fact that the index value need not be
constant, but can be any expression that evaluates to an integer or any other scalar type.
In many cases, the selection expression is the index variable of a for loop, which makes
it easy to perform an operation on each element of the array in turn. For example, you
can set each element in the scores array to 0.0 with the following statement:

for (int i = 0; i < N_JUDGES; i++) {

scores[i] = 0.0;
}

Effective and allocated sizes

At the time you write a program, you often will not know exactly how many elements an
array will contain. The number of array elements usually depends on the user’s data.
Some users may require large amounts of array storage, while others need less.
Unfortunately, you can’t simply declare an array as
Data Types in C++ – 59 –

int dataValues[n]; Array bounds must be constant.

where n is a variable whose value changes in response to the needs of the application.
C++ requires that arrays be declared with a constant size.
The usual strategy for solving this problem is to declare an array that is larger than you
need and use only part of it. Thus, instead of declaring the array based on the actual
number of elements—which you often do not know in advance—you define a constant
indicating the maximum number of elements and use that constant in the declaration of
the array. On any given use of the program, the actual number of elements is always less
than or equal to this bound. When you use this strategy, you need to maintain a separate
integer variable that keeps track of the number of values that are actually in use. The size
of the array specified in the declaration is called the allocated size; the number of
elements actively in use is called the effective size.
As an example, suppose that you wanted to change the declaration of the array scores
introduced in the preceding section so that the program would work with any reasonable
number of judges. Since you can’t imagine that the number of judges at a sports event
would ever be larger than 100, you might declare the array like this:
const int MAX_JUDGES = 100;
int scores[MAX_JUDGES];

To keep track of the effective size, you would need to declare an additional variable, as
follows:
int nJudges;

Passing arrays as parameters

Functions in C++ can take entire arrays as parameters. When they do, it is common—
particularly if the allocated and effective sizes might be different—to omit the maximum
bound in the parameter declaration and use empty brackets instead. For example, the
following function takes an array of type double and an integer indicating the effective
size of the array and returns the mean, or arithmetic average, of the elements in the array:
double Mean(double array[], int n) {
double total = 0;
for (int i = 0; i < n; i++) {
total += array[i];
}
return total / n;
}

When a function takes an array argument, the value of that argument is not copied in
the way that simple variables are. Instead, the function always gets a pointer to the array,
which means that the storage used for the parameter array is shared with that of the actual
argument. Changing the value of an element of the parameter array therefore changes the
value of the corresponding element in the argument array.
Data Types in C++ – 60 –

The use of arrays as parameters is illustrated by the gymjudge.cpp program shown in

Figure 2-1, which asks the user to enter the score for each judge and then displays the
average score. Note that the ReadAllScores function depends on the fact that arrays are
passed as pointers. The whole point of the function is to fill up the elements in the array
scores. If ReadAllScores were unable to make changes to the elements of the calling
array, you would be forced to seek a different implementation strategy.

Initialization of arrays
Array variables can be given initial values at the time they are declared. In this case, the
equal sign specifying the initial value is followed by a list of initializers enclosed in curly
braces. For example, the declaration
int digits[10] = { 0, 1, 2, 3, 4, 5, 6, 7, 8, 9 };

declares an array called digits in which each of the 10 elements is initialized to its own
index number. When initializers are provided for an array, it is legal to omit the array
size from the declaration. Thus, you could also write the declaration of digits like this:
int digits[] = { 0, 1, 2, 3, 4, 5, 6, 7, 8, 9 };

When the compiler encounters a declaration of this form, it counts the number of
initializers and reserves exactly that many elements for the array.
In the digits example, there is little advantage in leaving out the array bound. You
know that there are 10 digits and that new digits are not going to be added to this list. For
arrays whose initial values may need to change over the life cycle of the program, having
the compiler compute the array size from the initializers is useful for program
maintenance because it frees the programmer from having to maintain the element count
as the program evolves.
For example, imagine you’re writing a program that requires an array containing the
names of all U.S. cities with populations of over 1,000,000. Taking data from the 1990
census, you could declare and initialize bigCities as a global array using the following
declaration:
string bigCities[] = {
"New York",
"Los Angeles",
"Chicago",
"Houston",
"Philadelphia",
"San Diego",
"Detroit",
"Dallas",
};

When the figures are in from the 2000 census, it is likely that Phoenix and San Antonio
will have joined this list. If they have, you can then simply add their names to the
initializer list. The compiler will expand the array size to accommodate the new values.
Note that a comma follows the last initializer for the bigCities array. This comma is
optional, but it is good programming practice to include it. Doing so allows you to add
new cities without having to change the existing entries in the initializer list.
If you write a program that uses the bigCities array, you will probably need to know
how many cities the list contains. The compiler has this number because it counted the
Data Types in C++ – 61 –

Figure 2-1 Program to average a set of scores

/*
* File: gymjudge.cpp
* ------------------
* This program averages a set of gymnastic scores.
*/

#include <iostream>
#include "genlib.h"
#include "simpio.h"

/* Constants */

const int MAX_JUDGES = 100;

const double MIN_SCORE = 0.0;
const double MAX_SCORE = 10.0;

/* Private function prototypes */

void ReadAllScores(double scores[], int nJudges);

double GetScore(int judge);
double Mean(double array[], int n);

/* Main program */

int main() {
double scores[MAX_JUDGES];
cout << "Enter number of judges: ";
int nJudges = GetInteger();
if (nJudges > MAX_JUDGES) Error("Too many judges");
ReadAllScores(scores, nJudges);
cout << "The average score is " << Mean(scores, nJudges) << endl;
return 0;
}

/*
* Function: ReadAllScores
* Usage: ReadAllScores(scores, nJudges);
* --------------------------------------
* This function reads in scores for each of the judges. The
* array scores must be declared by the caller and must have
* an allocated size that is at least as large as nJudges.
* Because people tend to count starting at 1 rather than 0,
* this program adds 1 to the argument to GetScore, which means
* that the values the user sees will range from 1 to n instead
* of between 0 and n-1.
*/

void ReadAllScores(double scores[], int nJudges) {

for (int i = 0; i < nJudges; i++) {
scores[i] = GetScore(i + 1);
}
}
Data Types in C++ – 62 –

/*
* Function: GetScore
* Usage: score = GetScore(judge);
* -------------------------------
* This function reads in the score for the specified judge.
* The implementation makes sure that the score is in the
* legal range before returning.
*/

double GetScore(int judge) {

while (true) {
cout << "Score for judge #" << judge << ": ";
double score = GetReal();
if (score >= MIN_SCORE && score <= MAX_SCORE) return score;
cout << "That score is out of range. Try again." << endl;
}
}

/*
* Function: Mean
* Usage: mean = Mean(array, n);
* -----------------------------
* This function returns the statistical mean (average) of a
* distribution stored in array, which has effective size n.
*/

double Mean(double array[], int n) {

double total = 0;
for (int i = 0; i < n; i++) {
total += array[i];
}
return total / n;
}

initializers. The question is how to make that information available to the program. In
C++, there is a standard idiom for determining the number of elements in an array whose
size is established by static initialization. Given any array a, the number of elements in a
can be computed using the expression
sizeof a / sizeof a[0]

Described in English, this expression takes the size of the entire array and divides it by
the size of the initial element in the array. Because all elements of an array are the same
size, the result is the number of elements in the array, regardless of the element type.
Thus you could initialize a variable nBigCities to hold the number of cities in the
bigCities array by writing

int nBigCities = sizeof bigCities / sizeof bigCities[0];

Multidimensional arrays
In C++, the elements of an array can be of any type. In particular, the elements of an
array can themselves be arrays. Arrays of arrays are called multidimensional arrays.
The most common form is the two-dimensional array, which is most often used to
represent data in which the individual entries form a rectangular structure marked off into
Data Types in C++ – 63 –

rows and columns. This type of two-dimensional structure is called a matrix. Arrays of
three or more dimensions are also legal in C++ but occur much less frequently.

The following declaration, for example, introduces a 3 x 3 matrix, each of whose

elements is of type double:

double mat[3][3];

Conceptually, the storage for mat forms a two-dimensional structure in which the
individual elements are laid out like this:

mat[0][0] mat[0][1] mat[0][2]

mat[1][0] mat[1][1] mat[1][2]

mat[2][0] mat[2][1] mat[2][2]

Internally, C++ represents the variable mat as an array of three elements, each of
which is an array of three floating-point values. The memory allocated to mat consists of
nine cells arranged in the following form:

mat[0][0]
mat[0] mat[0][1]
mat[0][2]
mat[1][0]
mat[1] mat[1][1]
mat[1][2]
mat[2][0]
mat[2] mat[2][1]
mat[2][2]

In the two-dimensional diagram, the first index is assumed to indicate the row number.
This choice, however, is arbitrary because the two-dimensional geometry of the matrix is
entirely conceptual; in memory, these values form a one-dimensional list. If you want the
first index to indicate the column and the second to indicate the row, you do not need to
change the declaration, only the way in which you select the elements. In terms of the
internal arrangement, however, it is always true that the first index value varies least
rapidly. Thus all the elements of mat[0] appear in memory before any elements of
mat[1].

Multidimensional arrays are passed between functions just as single-dimensional

arrays are. The parameter declaration in the function header looks like the original
declaration of the variable and includes the index information.C++ requires that you
specify the size of each index in a parameter array, except for the first. However, because
leaving out the first index bound makes the declaration unsymmetrical, it is common to
include the array bounds for each index in the declaration of a multidimensional array
parameter.
Data Types in C++ – 64 –

You can also use initializers with multidimensional arrays. To emphasize the overall
structure, the values used to initialize each internal array are usually enclosed in an
additional set of curly braces. For example, the declaration

double identityMatrix[3][3] = {
{ 1.0, 0.0, 0.0 },
{ 0.0, 1.0, 0.0 },
{ 0.0, 0.0, 1.0 }
};

declares a 3 x 3 matrix of doubles and initializes it to contain the following values:

1.0 0.0 0.0

0.0 1.0 0.0

0.0 0.0 1.0

This particular matrix comes up frequently in mathematical applications and is called the
identity matrix.
As in the case of parameters, the declaration of a statically initialized multidimensional
array must specify all index bounds except possibly the first, which can be determined by
counting the initializers. As was true with parameters, however, it is usually best to
specify all the index bounds explicitly when you declare a multidimensional array.

2.5 Pointers and arrays

In C++, arrays and pointers are defined in such a way that a complete understanding of
either topic requires that you understand the other. Arrays, for example, are implemented
internally as pointers. The operations on pointers only make sense if you consider them
in relation to an array. Thus, in order to give you the full picture of how arrays and
pointers work, it is important to consider the two concepts together.
To get a sense of the relationship, consider the simple array declaration
double list[3];

which reserves enough space for three values of type double. Assuming that a double is
eight bytes long, the memory diagram for the array would look like this:
1000

list[0]

1008

list[1]

1016

list[2]
Data Types in C++ – 65 –

Because each of the elements in the array is an lvalue, it has an address that can be
derived using the & operator. For example, the expression
&list[1]

has the pointer value 1008 because the element list[1] is stored at that address.
Moreover, the index value need not be constant. The selection expression
list[i]

is an lvalue, and it is therefore legal to write

&list[i]

which indicates the address of the ith element in list.

Because the address of the ith element in list depends on the value of the variable i,
the C++ compiler cannot compute this address when compiling the program. To
determine the address, the compiler generates instructions that take the base address of
the array and then add the value of i multiplied by the size of each array element in bytes.
Thus, the numeric calculation necessary to find the address of list[i] is given by the
formula
1000 + i x 8
If i is 2, for example, the result of the address calculation is 1016, which matches the
address shown in the diagram for list[2] . Because the process of calculating the
address of an array element is entirely automatic, you don’t have to worry about the
details when writing your programs.

The relationship between pointers and arrays

Among the unusual characteristics of C++, one of the most interesting is that the name of
an array is treated as being synonymous with a pointer to the initial element in that array.
This concept is most easily illustrated by example.
The declaration
int intList[5];

allocates space for an array of five integers, which is assigned storage somewhere inside
the computer’s memory, as illustrated in the following diagram:
2000
intList[0]
2004
intList[1]
2008
intList[2]
2012
intList[3]
2016
intList[4]
Data Types in C++ – 66 –

The name intList represents an array but can also be used directly as a pointer value.
When it is used as a pointer, intList is defined to be the address of the initial element in
the array. For any array arr, the following identity always holds in C++:
arr is defined to be identical to &arr[0]

Given any array name, you can assign its address directly to any pointer variable.
The most common example of this equivalence occurs when an array is passed from
one function to another. For example, suppose that you make the function call
sum = SumIntegerArray(intList, 5);

where the definition of SumIntegerArray is

int SumIntegerArray(int array[], int n) {

int sum = 0;

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

sum += array[i];
}
return sum;
}

The SumIntegerArray function would work exactly the same way if the prototype had
been written as
int SumIntegerArray(int *array, int n)

In this case, the first argument is declared as a pointer, but the effect is the same as in the
preceding implementation, which declared this parameter as an array. The address of the
first element in intList is copied into the formal parameter array and manipulated
using pointer arithmetic. Inside the machine, the declarations are equivalent and the same
operations can be applied in either case.
As a general rule, you should declare parameters in the way that reflects their use. If
you intend to use a parameter as an array and select elements from it, you should declare
that parameter as an array. If you intend to use the parameter as a pointer and dereference
it, you should declare it as a pointer.
The crucial difference between arrays and pointers in C++ comes into play when
variables are originally declared, not when those values are passed as parameters. The
fundamental distinction between the declaration
int array[5];

and the declaration

int *p;

is one of memory allocation. The first declaration reserves five consecutive words of
memory capable of holding the array elements. The second declaration reserves only a
single word, which is large enough to hold a machine address. The implication of this
difference is extremely important to you as a programmer. If you declare an array, you
have storage to work with; if you declare a pointer variable, that variable is not associated
with any storage until you initialize it.
Data Types in C++ – 67 –

Given your current level of understanding, the only way to use a pointer as an array is
to initialize the pointer by assigning the base address of the array to the pointer variable.
If, after making the preceding declarations, you were to write
p = array;

the pointer variable p would then point to the same address used for array, and you could
use the two names interchangeably.
The technique of setting a pointer to the address of an existing array is rather limited.
After all, if you already have an array name, you might as well use it. Assigning that
name to a pointer does not really do you any good. The real advantage of using a pointer
as an array comes from the fact that you can initialize that pointer to new memory that
has not previously been allocated, which allows you to create new arrays as the program
runs. This important programming technique is described in the section on “Dynamic
allocation” later in this chapter.

2.6 Records
To understand the idea of a record, imagine for a moment that you are in charge of the
payroll system for a small company. You need to keep track of various pieces of
information about each employee. For example, in order to print a paycheck, you need to
know the employee’s name, job title, Social Security number, salary, withholding status,
and perhaps some additional data as well. These pieces of information, taken together,
form the employee’s data record.
What do employee records look like? It is often easiest to think of records as entries in
a table. For example, consider the case of the small firm of Scrooge and Marley,
portrayed in Charles Dickens’s A Christmas Carol, as it might appear in this day of
Social Security numbers and withholding allowances. The employee roster contains two
records, which might have the following values:
Name Job title Soc. Sec. # Salary # With.
Ebenezer Scrooge Partner 271-82-8183 250.00 1
Bob Cratchit Clerk 314-15-9265 15.00 7

Each record is broken up into individual components that provide a specific piece of
information about the employee. Each of these components is usually called a field,
although the term member is also used, particularly in the context of C++ programming.
For example, given an employee record, you can talk about the name field or the salary
field. Each of the fields is associated with a type, which may be different for different
fields. The name and title field are strings, the salary field might well be represented as a
floating-point number, and the number of withholding exemptions is presumably an
integer. The Social Security number could be represented as either an integer or a string;
because Social Security numbers are too big to fit within the limits imposed on integers
by many systems, they are represented here as strings.
Even though a record is made up of individual fields, it must have meaning as a
coherent whole. In the example of the employee roster, the fields in the first line of the
table represent a logically consistent set of data referring to Ebenezer Scrooge; those in
the second line refer to Bob Cratchit. The conceptual integrity of each record suggests
that the data for that employee should be collected into a compound data structure.
Moreover, since the individual fields making up that structure are of different types,
arrays are not suited to the task. In cases such as this, you need to define the set of data
for each employee as a record.
Data Types in C++ – 68 –

Defining a new structure type

Creating new records in C++ is conceptually a two-step process.

1. Define a new structure type. Before you declare any variables, you must first define a
new structure type. The type definition specifies what fields make up the record,
what the names of those fields are, and what type of information each field contains.
This structure type defines a model for all objects that have the new type but does not
by itself reserve any storage.
2. Declare variables of the new type. Once you have defined the new type, you can then
declare variables of that type so that you can store actual data values.

The general form for defining a new structure type looks like this:

struct name {
field-declarations
};

where field-declarations are standard variable declarations used to define the fields of the
structure and name indicates the name of the newly defined type. For example, the
following code defines a new structure type called employeeRecordT to represent
employee records:

struct employeeRecordT {
string name;
string title;
string ssnum;
double salary;
int withholding;
};

This definition provides a template for all objects that have the new type
employeeRecordT . Each such object will have five fields, starting with a name field,
which is a string, and continuing through a withholding field, which is an int.

Declaring structure variables

Now that you have defined a new type, the next step is to declare variables of that type.
For example, given the type employeeRecordT, you can declare empRec to be a variable
of that type by writing

employeeRecordT empRec;

If you want to illustrate this variable using a box diagram, you can choose to represent
it in either of two ways. If you take a very general view of the situation—which
corresponds conceptually to looking at the diagram from a considerable distance—what
you see is just a box named empRec:
empRec

If, on the other hand, you step close enough to see the details, you discover that the box
labeled empRec is composed internally of five individual boxes:
Data Types in C++ – 69 –

empRec

name title

ssnum salary withholding

Record selection
Once you have declared the variable empRec by writing

employeeRecordT empRec;

you can refer to the record as a whole simply by using its name. To refer to a specific
field within a record, you write the name of the complete record, followed by a period,
followed by the name of the field. Thus, to refer to the job title of the employee stored in
empRec, you need to write

empRec.title

When used in this context, the period is invariably called a dot, so that you would read
this expression aloud as “empRec dot title.” Selecting a field using the dot operator is
called record selection.

Initializing records
As with any other type of variable, you can initialize the contents of a record variable by
assigning values to its components. The dot operator returns an lvalue, which means that
you can assign values to a record selection expression. For example, if you were to
execute the statements

empRec.name = "Ebenezer Scrooge";

empRec.title = "Partner";
empRec.ssnum = "271-82-8183";
empRec.salary = 250.00;
empRec.withholding = 1;

you would create the employee record for Ebenezer Scrooge used in the earlier examples.
You can also initialize its contents at the time the record is declared, using much the
same syntax as you use to initialize the elements of an array. Initializers for a record are
specified in the order in which they appear in the structure definition. Thus, you could
declare and initialize a record named manager that contains the data for Mr. Scrooge, as
follows:

employeeRecordT manager = {
"Ebenezer Scrooge", "Partner", "271-82-8183", 250.00, 1
};

Pointers to records
Although small records are sometimes used directly in C++, variables that hold structured
data in C++ are often declared to be pointers to records rather than the records
Data Types in C++ – 70 –

themselves. A pointer to a record is usually smaller and more easily manipulated than the
record itself.
Suppose, for example, that you want to declare a variable that points to employee
records. The syntax for such a declaration is the same as that for any other pointer. The
line
employeeRecordT *empPtr;

declares the variable empPtr as a pointer to an object of type employeeRecordT. When

you make this declaration, space is reserved only for the pointer itself. Before using
empPtr, you still need to provide the actual storage for the fields in the complete record.
The best approach is to allocate space for a new record as described in the next section,
on “Dynamic allocation.” For the moment, let’s assume that the earlier declaration
employeeRecordT empRec;

is still around, which means that you can make empPtr point to the empRec record by
writing
empPtr = &empRec;

The conceptual picture of memory now looks like this:

empPtr empRec

name title

ssnum salary withholding

Starting from empPtr , how can you refer to an individual field in the underlying
record? In seeking an answer to this question, it is easy to be misled by your intuition. It
is not appropriate to write, for example,

*empPtr.salary The order of operations is incorrect.

Contrary to what you might have expected, this statement does not select the salary
component of the object to which empPtr points, because the precedence of the operators
in the expression does not support that interpretation. The selection operator takes
precedence over dereferencing, so the expression has the meaningless interpretation
*(empPtr.salary)

rather than the intended

(*empPtr).salary
Data Types in C++ – 71 –

The latter form has the desired effect but is much too cumbersome for everyday use.
Pointers to structures are used all the time. Forcing programmers to include parentheses
in every selection would make records much less convenient. For this reason, C++
defines the operator ->, which combines the operations of dereference and selection into
a single operator. Thus, the conventional way to refer to the salary in the record to which
empPtr points is to write

empPtr->salary

2.7 Dynamic allocation

Up to this point in the text, you have seen two mechanisms for assigning memory to
variables. When you declare a global variable, the compiler allocates memory space for
that variable which persists throughout the entire program. This style of allocation is
called static allocation because the variables are assigned to fixed locations in memory.
When you declare a local variable inside a function, the space for that variable is
allocated on the system stack. Calling the function assigns memory to the variable; that
memory is freed when the function returns. This style of allocation is called automatic
allocation. There is also a third way of allocating memory that permits you to acquire
new memory when you need it and to free it explicitly when it is no longer needed. The
process of acquiring new storage while the program is running is called dynamic
allocation.
When a program is loaded into memory, it usually occupies only a fraction of the
available storage. In most systems, you can allocate some of the unused storage to the
program whenever it needs more memory. For example, if you need space for a new
array while the program is running, you can reserve part of the unallocated memory,
leaving the rest for subsequent allocations. The pool of unallocated memory available to
a program is called the heap.
In C++, you use the new operator to allocate memory from the heap. In its simplest
form, the new operator takes a type and allocates space for a variable of that type located
in the heap. For example, if you want to allocate an integer in the heap, you call

int *ip = new int;

The call to new operator will return the address of a storage location in the heap that has
been set aside to hold an integer.

The new operator can also be used to allocate variables of compound type. To allocate an
employee record in the heap, you could use the call:

employeeRecordT *empPtr = new employeeRecordT;

The new[] operator is a variant that is used to allocate an array in the heap. Within the
square brackets, you specify the number of array elements, as shown in these examples:

int *intList = new int[12];

employeeRecordT *empList = new employeeRecordT[1000];

The address returned by new[] is the base address of a contiguous piece of memory large
enough for the entire array. You can index into dynamic array using ordinary subscript
notation just as you would for a static array.
Data Types in C++ – 72 –

Coping with memory limitations

Although they are getting larger all the time, computer memory systems are finite in size.
As a result, the heap will eventually run out of space. When this occurs, the new operator
will eventually be unable to allocate a block of the requested size. The failure to fulfill an
allocation request is such a serious error that the default behavior is for the program to
immediately halt.

One way to help ensure that you don’t run out of memory is to free any storage you
have allocated when you are finished using it. C++ supplies the operator delete, which
takes a pointer previously allocated by new and returns the memory associated with that
pointer to the heap. If, for example, you determine that you are completely finished using
the storage allocated for ptr, you can free that storage for later reuse by calling

delete ptr;

The delete[] operator is used to free storage that was allocated using the new[]
operator. You do not indicate the number of elements in the square brackets when using
delete[].

int *arr = new int[45];

...
delete[] arr;

Knowing when to free a piece of memory is not always an easy task, particularly as
programs become large. If several parts of a program share some data structure that has
been allocated in the heap, it may not be possible for any single part to recognize that the
memory can be freed. Given the size of memories today, however, you can often allocate
whatever memory you need without ever bothering to free it again. The problem of
limited memory typically becomes critical only when you design an application that
needs to run for a long period of time, such as the operating system on which all the other
facilities of the system depend. In these applications, it is important to free memory
when you no longer need it.
Some languages, including Java, support a system for dynamic allocation that actively
goes through memory to see what parts of it are in use, freeing any storage that is no
longer needed. This strategy is called garbage collection. Garbage-collecting allocators
exist for C++, and it is likely that their use will increase in coming years, particularly as
people become more familiar with their advantages. If it does, the policy of ignoring
deallocation will become reasonable even in long-running applications because you can
rely on the garbage collector to perform the deallocation operations automatically.

For the most part, this text assumes that your applications fall into the class of
problems for which allocating memory whenever you need it is a workable strategy. This
assumption will simplify your life considerably and make it easier for you to concentrate
on algorithmic details.

Dynamic arrays
From a conceptual perspective, an assignment of the form

cp = new char[10];

creates the following configuration in memory:

Data Types in C++ – 73 –

cp

The variable cp points to a set of 10 consecutive bytes that have been allocated in the
heap. Because pointers and arrays are freely interchangeable in C++, the variable now
acts exactly as if it had been declared as an array of 10 characters.
Arrays that you allocate on the heap and reference using a pointer variable are called
dynamic arrays and play a significant role in modern programming. The principal
differences between declared arrays and dynamic arrays are that

• The memory associated with a declared array is allocated automatically as part of the
declaration process. When the frame for the function declaring the array is created, all
the elements of that array are allocated as part of the frame. In the case of a dynamic
array, the actual memory is not allocated until you invoke the new[] operator.
• The size of a declared array must be a constant in the program. In contrast, because its
memory comes from the heap, a dynamic array can be of any size. Moreover, you can
adjust the size of a dynamic array according to the amount of data. If you know you
need an array with precisely N elements, you can reserve just the right amount of
storage.

You can allocate a dynamic array using the new[] operator. For example, if you
wanted to initialize the variable darray to a dynamic array of n values of type double,
you would declare darray using the line
double *darray;

and then execute the following code:

darray = new double[n];

Dynamic records
Dynamic memory allocation is just as useful for records as it is for arrays. If you declare
a pointer to a record, you can allocate memory to store the actual data in the record by
calling new. For example, if the type employeeRecordT is defined as
struct employeeRecordT {
string name;
string title;
string ssnum;
double salary;
int withholding;
};

you can assign space for a newly allocated record to the variable empPtr as follows:
employeeRecordT *empPtr;
empPtr = new employeeRecordT;
empPtr->name = "Eric S. Roberts";
Data Types in C++ – 74 –

Summary
In this chapter, you have learned how to use the data structure definition capabilities of
C++ to create new data types. The data structures presented in this chapter—pointers,
arrays, and records—form the foundation for abstract data types, which are presented in
later in this text. The principal advantage of these structures is that you can use them to
represent data in a way that reflects the real-world structure of an application. Moreover,
by combining pointers, arrays, and records in the right way, you can create hierarchical
structures that allow you to manage data complexity in much the same way that
decomposing a large program into separate functions allows you to manage algorithmic
complexity.
Important points in this chapter include:

• C++ allows you to define new atomic types by listing the elements that comprise the
domain of the type. Such types are called enumeration types and are part of a more
general class called scalar types that also includes characters and integers.
• Data values inside a computer are represented as collections of bits that are organized
into larger structures called bytes and words. Every byte in memory is associated with
a numeric address that can be used to refer to the data contained at that location.
• Addresses of data in memory are themselves data values and can be manipulated as
such by a program. A data value that is the address of some other piece of data is
called a pointer. Pointer variables are declared in C++ by writing an asterisk in front
of the variable name in its declaration line.
• The fundamental operations on pointers are & and *, which indicate the address of a
stored value and the value stored at a particular address, respectively.
• There is a special pointer value called NULL, which is used to indicate that a pointer
does not refer to anything.
• An array is an ordered, homogeneous collection of data values composed of elements
indicated by an index number. In C++, index numbers in an array always begin with
0. Arrays are declared by specifying a constant size in square brackets after the name
of the array. The size specified in the declaration is called the allocated size of the
array and is typically larger than the effective size, which is the number of elements
actually in use.
• When an array is passed as a parameter, the elements of the array are not copied.
Instead, the function is given the address of the actual array. As a result, if a function
changes the values of any elements of an array passed as a parameter, those changes
will be visible to the caller, which is working with the same array.
• A record is a heterogeneous collection of data values that forms a logically consistent
unit.
• The storage for arrays and records can be allocated dynamically from a pool of unused
memory called the heap. The standard operators new and new[] are used to allocate
records and dynamic arrays, respectively.

Review questions
1. Define each of the following terms: pointer, array, record.

2. What type definition would you use to define a new enumeration type polygonT
consisting of the following elements: Triangle , Square , Pentagon , Hexagon,
Octagon? How would you change the definition so that internal representation for
each constant name corresponded to the number of sides for that polygon?
Data Types in C++ – 75 –

3. What three advantages are cited in the text for defining a new enumeration type as
opposed to defining named constants to achieve an equivalent effect?

4. True or false: In C++, you may apply any operation defined on integers to values of
any scalar type.

5. At first glance, the following function looks very much like RightFrom, which is
defined in the section on “Scalar types”:

directionT LeftFrom(directionT dir) {

return directionT((dir - 1) % 4);
}

The RightFrom implementation works fine, but this one has a small bug. Identify
the problem, and then rewrite the function so that it correctly calculates the compass
direction that is 90 degrees to the left of dir.

6. In your own words, explain what the keyword struct does.

7. Define the following terms: bit, byte, word, address.

8. True or false: In C++, a value of type int always requires four bytes of memory.

9. True or false: In C++, a value of type char always requires one byte of memory.

10. What is the purpose of the sizeof operator? How do you use it?

11. What reasons for using pointers are cited in this chapter?

12. What is an lvalue?

13. What are the types of the variables introduced by the following declaration:
int * p1, p2;

14. What declaration would you use to declare a variable named pFlag as a pointer to a
Boolean value?
15. What are the two fundamental pointer operations? Which one corresponds to the
term dereferencing?
16. Explain the difference between pointer assignment and value assignment.
17. Draw diagrams showing the contents of memory after each line of the following
code:
v1 = 10; v2 = 25; p1 = &v1; p2 = &v2;
*p1 += *p2;
p2 = p1;
*p2 = *p1 + *p2;

18. What is the internal representation of the constant NULL?

19. What does the phrase call by reference mean?

Data Types in C++ – 76 –

20. Write array declarations for the following array variables:

a. An array realArray consisting of 100 floating-point values
b. An array inUse consisting of 16 Boolean values
c. An array lines that can hold up to 1000 strings
Remember that the upper bounds for these arrays should be defined as constants to
make them easier to change.

21. Write the variable declaration and for loop necessary to create and initialize the
following integer array:
squares
0 1 4 9 16 25 36 49 64 81 100
0 1 2 3 4 5 6 7 8 9 10

22. What is the difference between allocated size and effective size?

23. Assuming that the base address for rectangular is 1000 and that values of type int
require four bytes of memory, draw a diagram that shows the address of each
element in the array declared as follows:
int rectangular[2][3];

24. Write a variable declaration that you could use to record the state of a chessboard,
which consists of an 8 x 8 array of squares, each of which may contain any one of the
following symbols:
K white king k black king
Q white queen q black queen
R white rook r black rook
B white bishop b black bishop
N white knight n black knight
P white pawn p black pawn
– empty square
Explain how you could initialize this array so that it holds the standard starting
position for a chess game:

r n b q k b n r

p p p p p p p p

- - - - - - - -

- - - - - - - -

- - - - - - - -

- - - - - - - -

P P P P P P P P

R N B Q K B N R
Data Types in C++ – 77 –

25. Assuming that intArray is declared as

int intArray[10];

and that j is an integer variable, describe the steps the computer would take to
determine the value of the following expression:

&intArray[j + 3];

26. True or false: If arr is declared to be an array, then the expressions

arr

and

&arr[0]

are equivalent.

27. Assume that variables of type double take up eight bytes on the computer system
you are using. If the base address of the array doubleArray is 1000, what is the
address value of doubleArray + 5?

28. Describe the distinction between the declarations

int array[5];

and

int *p;

29. What steps are necessary to declare a record variable?

30. If the variable p is declared as a pointer to a record that contains a field called cost,
what is wrong with the expression

*p.cost

as a means of following the pointer from p to its value and then selecting the cost
field? What expression would you write in C++ to accomplish this dereference-and-
select operation?

31. What is the heap?

32. Describe the effect of the following operators: new and delete.

33. What is the purpose of the delete[] operator?

34. What is meant by the term garbage collection?

Programming exercises
1. Define an enumeration type weekdayT whose elements are the days of the week.
Write functions NextDay and PreviousDay that take a value of type weekdayT and
return the day of the week that comes after or before the specified day, respectively.
Data Types in C++ – 78 –

Also write a function IncrementDay(startDay, delta) that returns the day of the
week that comes delta days after startDay. Thus, IncrementDay(Thursday, 4)
should return Monday. Your implementation of IncrementDay should work if the
value of delta is negative, in which case it should proceed backward in time.

2. Write a program that computes the surface area and volume of a cylinder, given the
height (h) and radius of the base (r) as shown in the following diagram:

The formulas for calculating surface area and volume are

A = 2π hr
V = π h r2

In this exercise, design your main program so that it consists of three function
calls: one to read the input data, one to compute the results, and one to display the
answers. When appropriate, use call by reference to communicate data between the
functions and the main program.
3. Because individual judges may have some bias, it is common practice to throw out
the highest and lowest score before computing the average. Modify the
gymjudge.cpp program from Figure 2-1 to discard the highest and lowest scores
before computing the average score.

4. Write a predicate function IsSorted(array, n) that takes an integer array and its
effective size as parameters and returns true if the array is sorted in nondecreasing
order.

5. In the third century B.C., the Greek astronomer Eratosthenes developed an algorithm
for finding all the prime numbers up to some upper limit N. To apply the algorithm,
you start by writing down a list of the integers between 2 and N. For example, if N
were 20, you would begin by writing down the following list:

2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20

Next you circle the first number in the list, indicating that you have found a prime.
You then go through the rest of the list and cross off every multiple of the value you
have just circled, since none of those multiples can be prime. Thus, after executing
the first step of the algorithm, you will have circled the number 2 and crossed off
every multiple of two, as follows:

2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20

From this point, you simply repeat the process by circling the first number in the
list that is neither crossed off nor circled, and then crossing off its multiples. In this
Data Types in C++ – 79 –

example, you would circle 3 as a prime and cross off all multiples of 3 in the rest of
the list, which would result in the following state:

2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20

Eventually, every number in the list will either be circled or crossed out, as shown in
this diagram:

2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20

The circled numbers are the primes; the crossed-out numbers are composites. This
algorithm for generating a list of primes is called the sieve of Eratosthenes.
Write a program that uses the sieve of Eratosthenes to generate a list of the primes
between 2 and 1000.

6. A histogram is a graphical way of displaying data by dividing the data into separate
ranges and then indicating how many data values fall into each range. For example,
given the set of exam scores
100, 95, 47, 88, 86, 92, 75, 89, 81, 70, 55, 80
a traditional histogram would have the following form:
*
*
*
* * *
* * * * * *
0–9 10–19 20–29 30–39 40–49 50–59 60–69 70–79 80–89 90–99 100

The asterisks in the histogram indicate one score in the 40s, one score in the 50s, five
scores in the 80s, and so forth.
When you generate histograms using a computer, however, it is usually much
easier to display them sideways on the page, as in this sample run:
HistogramSideways
0:
10:
20:
30:
40: *
50: *
60:
70: **
80: *****
90: **
100: *

Write a program that reads in an array of integers and then displays a histogram of
those numbers, divided into the ranges 0–9, 10–19, 20–29, and so forth, up to the
range containing only the value 100. Your program should generate output that
looks as much like the sample run as possible.
Data Types in C++ – 80 –

7. Rewrite the histogram program from the preceding exercise so that it displays the
histogram in a more traditional vertical orientation, like this:
Histogram
*
*
*
* * *
* * * * * *
0 10 20 30 40 50 60 70 80 90 100

8. Write a function RemoveZeroElements(array, n) that goes through an array of

integers and eliminates any elements whose value is 0. Because this operation
changes the effective size of the array, RemoveZeroElements should take the
effective size as a reference parameter and adjust it accordingly. For example,
suppose that scores contains an array of scores on an optional exam and that
nScores indicates the effective size of the array, as shown:

nScores
13
scores
65 0 95 0 0 79 82 0 84 94 86 90 0
0 1 2 3 4 5 6 7 8 9 10 11 12

At this point, the statement

RemoveZeroElements(scores, nScores);

should remove the 0 scores, compressing the array into the following configuration:
nScores
8
scores
65 95 79 82 84 94 86 90
0 1 2 3 4 5 6 7 8 9 10 11 12

9. The initial state of a checkers game is shown in the following diagram:

Data Types in C++ – 81 –

The dark squares in the bottom three rows are occupied by red checkers; the dark
squares in the top three rows contain black checkers. The two center rows are
unoccupied.
If you want to store the state of a checkerboard in a computer program, you need a
two-dimensional array indexed by rows and columns. The elements of the array
could be of various different types, but a reasonable approach is to use characters.
For example, you could use the letter r to represent a red checker and the letter b to
represent a black checker. Empty squares could be represented as spaces or hyphens
depending on whether the color of the square was light or dark.
Implement a function InitCheckerboard that initializes a checkerboard array so
that it corresponds to the starting position of a checkers game. Implement a second
function DisplayCheckerboard that displays the current state of a checkerboard on
the screen, as follows:
Checkerboard.sp
b b b b
b b b b
b b b b
- - - -
- - - -
r r r r
r r r r
r r r r

10. Design a function prototype that would allow a single function to find and return
simultaneously both the lowest and highest values in an array of type double.
Implement and test your function as shown in the following sample run:
FindRange
Enter the elements of the array, one per line.
Use -1 to signal the end of the list.
? 67
? 78
? 75
? 70
? 71
? 80
? 69
? 86
? 65
? 54
? 76
? 78
? 70
? 68
? 77
? -1
The range of values is 54-86

11. Write a function IndexArray(n) that returns a pointer to a dynamically allocated

integer array with n elements, each of which is initialized to its own index. For
example, assuming that ip is declared as

int *ip;
Data Types in C++ – 82 –

the statement

ip = IndexArray(10);

should produce the following memory configuration:

ip

0 1 2 3 4 5 6 7 8 9

12. Design a new type called payrollT that is capable of holding the data for a list of
employees, each of which is represented using the employeeT type introduced in the
section on “Dynamic records” at the end of the chapter. The type payrollT should
be a pointer type whose underlying value is a record containing the number of
employees and a dynamic array of the actual employeeT values, as illustrated by the
following data diagram:

Ebenezer Scrooge Bob Cratchit

Partner Clerk

271-82-8183 314-15-9265

250.00 15.00

1 7

After writing the types that define this data structure, write a function GetPayroll
that reads in a list of employees, as shown in the following sample run:

GetPayroll
How many employees: 2
Employee #1:
Name: Ebenezer Scrooge
Title: Partner
SSNum: 271-82-8183
Salary: 250.00
Withholding exemptions: 1
Employee #2:
Name: Bob Cratchit
Title: Clerk
SSNum: 314-15-9265
Salary: 15.00
Withholding exemptions: 7
Data Types in C++ – 83 –

After the input values have been entered, the GetPayroll function should return a
value of type payrollT that matches the structure shown in the diagram.

13. Write a program that generates the weekly payroll for a company whose employment
records are stored using the type payrollT, as defined in the preceding exercise.
Each employee is paid the salary given in the employee record, after deducting taxes.
Your program should compute taxes as follows:

• Deduct $1 from the salary for each withholding exemption. This figure is the
adjusted income. (If the result of the calculation is less than 0, use 0 as the
adjusted income.)
• Multiply the adjusted income by the tax rate, which you should assume is a flat
25 percent.

For example, Bob Cratchit has a weekly income of $15. Because he has seven
dependents, his adjusted income is $15 – (7 x $1), or $8. Twenty-five percent of $8
is $2, so Mr. Cratchit’s net pay is $15 – $2, or $13.
The payroll listing should consist of a series of lines, one per employee, each of
which contains the employee’s name, gross pay, tax, and net pay. The output should
be formatted in columns, as shown in the following sample run:

PrintPayroll
Name Gross Tax Net
----------------------------------------------
Ebenezer Scrooge 250.00 - 62.25 = 187.75
Bob Cratchit 15.00 - 2.00 = 13.00

14. Suppose that you have been assigned the task of computerizing the card catalog
system for a library. As a first step, your supervisor has asked you to develop a
prototype capable of storing the following information for each of 1000 books:

• The title
• A list of up to five authors
• The Library of Congress catalog number
• A list of up to five subject headings
• The publisher
• The year of publication
• Whether the book is circulating or noncirculating

Design the data structures that would be necessary to keep all the information
required for this prototype library database. Given your definition, it should be
possible to write the declaration

libraryT libdata;

and have the variable libdata contain all the information you would need to keep
track of up to 1000 books. Remember that the actual number of books will usually
be less than this upper bound.
Data Types in C++ – 84 –

Write an additional procedure SearchBySubject that takes as parameters the

library database and a subject string. For each book in the library that lists the
subject string as one of its subject headings, SearchBySubject should display the
title, the name of the first author, and the Library of Congress catalog number of the
book.

15. Write a function

int *GetDynamicIntegerArray(int sentinel, int & n);

that returns a dynamically allocated array of integers read in from the user. The
sentinel argument indicates the value used to signal the end of the input. The
second argument is an integer reference parameter, which is used to return the
effective size of the array. Note that it is impossible to know in advance how many
values the user will enter. As a result, the implementation must allocate new array
space as needed.
Chapter 3
Libraries and Interfaces

My library / Was dukedom large enough.

— Shakespeare, The Tempest, c. 1612
Libraries and Interfaces – 86 –

In modern programming, it is impossible to write interesting programs without calling

library functions. In fact, as the science of programming advances, programmers depend
more and more on library functions. Today, it is not at all unusual for 90 percent or more
of a program to consist of library code, with only a few parts specifically tailored for a
particular application. As a programmer, you must understand not only how to write new
code but also how to avoid doing so by making appropriate use of existing libraries.
The main purpose of this chapter is to encourage you to think about libraries in a way
that emphasizes the distinction between the library itself and other programs that make
use of it, which are called its clients. To do so, the chapter focuses on the boundary
between a library and its clients, which is called the interface. An interface provides a
channel of communication but also acts as a barrier that prevents complex details on one
side from affecting the other. Interfaces are central to a modern treatment of libraries,
and it is important—even if you’ve done a lot of programming—for you to understand
libraries from this perspective.

3.1 The concept of an interface

In English, the word interface means a common boundary between two distinct entities.
The surface of a pond, for example, is the interface between the water and the air. In
programming, an interface constitutes a conceptual boundary rather than a physical one:
an interface is the boundary between the implementation of a library and its clients. The
purpose of the interface is to provide each client with the information it needs to use the
library without revealing the details required by the implementation. Thus, it is important
to think of an interface not only as a channel for communication between client and
implementation, but also as a barrier that keeps them separated, as illustrated by the
following diagram:

client implementation

interface

By mediating the communication between the two sides, an interface reduces the
conceptual complexity of the programming process by ensuring that details that lie on
one side of the interface boundary do not escape to complicate the code on the other side.

Interfaces and implementations

In computer science, an interface is a conceptual entity. It consists of an understanding
between the programmer who implements a library and the programmer who uses it,
spelling out the information that is required by both sides. When you write a C++
program, however, you must have some way to represent the conceptual interface as part
of the actual program. In C++, an interface is represented by a header file, which
traditionally has the same name as the file that implements it with the .cpp extension
replaced by .h.
As an example, suppose that you have created a collection of functions that you want
to make available to clients as a library. To do so, you need to create two files: an
interface that you might call mylib.h and the corresponding implementation mylib.cpp.
Libraries and Interfaces – 87 –

The code for each of your functions goes in the mylib.cpp implementation file. The
mylib.h interface contains only the function prototypes, which contain the information
the compiler needs to interpret any calls to those functions. Putting the prototypes in the
interface makes them available to clients and is called exporting those functions.
Although function prototypes are the most common component of an interface,
interfaces can export other definitions as well. In particular, interfaces often export data
types and constants. A single definition exported by an interface is called an interface
entry.
Once you have written the interface and implementation for a library, you—or some
other programmer with whom you are collaborating—can then write separate source files
that act as clients of the mylib.h interface. The relationship between the files
representing the client, interface, and implementation is illustrated in the following
diagram:
client.cpp mylib.h mylib.cpp

A client of a The interface The implemen-

library interface contains only the tation contains
is any source file information the code to make
that makes use about the library the library work,
of definitions that its clients the details of
exported by the need to know. which are not
library. of interest to
clients.

The distinction between the abstract concept of an interface and the header file that
represents it may at first seem subtle. In many ways, the distinction is the same as that
between an algorithm and a program that implements it. The algorithm is an abstract
strategy; the program is the concrete realization of that algorithm. Similarly, in C++,
header files provide a concrete realization of an interface.

Packages and abstractions

The same distinction between a general concept and its programming manifestation also
comes up in the definition of two other terms that are often used in discussions of
interfaces. In computer science, you will often hear the term package used to describe
the software that defines a library. If you are assigned to develop a library, part of your
job consists of producing a .h file to serve as the library interface and a corresponding
.cpp file that provides the underlying implementation. Those files constitute the
package. To get a full understanding of a library, however, you must look beyond the
software. Libraries embody a specific conceptual approach that transcends the package
itself. The conceptual basis of a library is called an abstraction.
The relationship between an abstraction and a package is best illustrated by an
example. The programs in the first two chapters of this text use the insertion operator and
the cout stream from the iostream interface for all output operations. For input
operations, however, those programs use functions exported by the simpio.h interface
such as GetInteger, GetReal, and GetLine. The iostream interface provides functions
for accepting user input, but they are more difficult for beginning programmers to use
than their counterparts in simpio.h. The two libraries embody different approaches to
input operations: the iostream interface emphasizes power and flexibility, while the
simpio.h interface emphasizes simplicity of structure and ease of use. The approach
used in each of these interfaces is part of the abstraction. The associated packages
Libraries and Interfaces – 88 –

implement those abstractions and make them real, in the sense that they can then be used
by programmers.

Principles of good interface design

Programming is hard because programs reflect the complexity of the problems they solve.
As long as we use computers to solve problems of ever-increasing sophistication, the
process of programming will need to keep becoming more sophisticated as well.
Writing a program to solve a large or difficult problem forces you to manage an
enormous amount of complexity. There are algorithms to design, special cases to
consider, user requirements to meet, and innumerable details to get right. To make
programming manageable, you must reduce the complexity of the programming process
as much as possible.
One of the primary purposes of defining new functions is to reduce complexity by
dividing up the entire program into more manageable pieces. Interfaces offer a similar
reduction in programming complexity but at a higher level of detail. A function gives its
caller access to a set of steps that together implement a single operation. An interface
gives its client access to a set of functions that together implement a programming
abstraction. The extent to which the interface simplifies the programming process,
however, depends largely on how well it is designed.
To design an effective interface, you must balance several criteria. In general, you
should try to develop interfaces that are

• Unified. A single interface should define a consistent abstraction with a clear unifying
theme. If a function does not fit within that theme, it should be defined in a separate
interface.
• Simple. To the extent that the underlying implementation is itself complex, the
interface must hide as much of that complexity from the client as possible.
• Sufficient. When clients use an abstraction, the interface must provide sufficient
functionality to meet their needs. If some critical operation is missing from an
interface, clients may decide to abandon it and develop their own, more powerful
abstraction. As important as simplicity is, the designer must avoid simplifying an
interface to the point that it becomes useless.
• General. A well-designed interface should be flexible enough to meet the needs of
many different clients. An interface that performs a narrowly defined set of operations
for one client is not as useful as one that can be used in many different situations.
• Stable. The functions defined in an interface should continue to have precisely the
same structure and effect, even if their underlying implementation changes. Making
changes in the behavior of an interface forces clients to change their programs, which
compromises the value of the interface.

Because it is important to maintain stable interfaces, the designer of an interface must

exercise special care to get it right. As more and more clients start to depend on a
particular interface, the cost of changing that interface increases. In fact, it is quite
common to discover interfaces that cannot be changed at all, even if they have serious
design flaws. Certain functions in the standard libraries, for example, exhibit behaviors
that are widely considered flaws by the C++ programming community. Even so, it is
impossible to change the design of these functions because too many clients depend on
the current behavior.
Libraries and Interfaces – 89 –

Some interface changes, however, are more drastic than others. For example, adding
an entirely new function to an interface is usually a relatively straightforward process,
since no clients already depend on that function. Changing an interface in a way that
requires no changes to existing programs is called extending the interface. If you find
that you need to make evolutionary changes over the lifetime of an interface, it is usually
best to make those changes by extension.

3.2 A random number interface

The easiest way to illustrate the structure of an interface is to provide a simple example.
The libraries for this book and its C-based predecessor include an interface called
random.h that makes it easier for client programs to simulate random processes like
flipping a coin or rolling a die. Getting a computer—which ordinarily operates
deterministically in the sense that running the same program with the same input always
produces the same result—to behave in a random way involves a certain amount of
complexity. For the benefit of client programmers, you want to hide this complexity
behind an interface so that the client has access to the capabilities it needs without having
to confront the complexity of the complete implementation.
The structure of the random.h interface
Figure 3-1 shows the random.h interface, which exports the functions RandomInteger,
RandomReal, RandomChance, and Randomize. For each of these functions, the interface
contains a one-line prototype along with a comment that describes the purpose of the
function from the perspective of the client. As is typical for the interfaces defined in this
text, comments comprise most of the interface and provide all the information clients
need. If the comments in an interface are well designed, clients should be able to rely on
them without having to read the underlying implementation.
After the initial comment that describes the interface, the random.h interface includes
a somewhat cryptic set of lines that is part of the conventional syntax for an interface.
The lines

#ifndef _random_h
#define _random_h

operate in conjunction with the last line of the interface, which is

#endif

These three lines are often referred to as interface boilerplate. When you design your
own interfaces, you should be sure to include similar boilerplate lines, substituting the
name of your own interface for the name random in this example.
The purpose of the interface boilerplate is to prevent the compiler from reading the
same interface many times during a single compilation. The line
#ifndef _random_h

causes the compiler to skip any text up to the #endif line if the symbol _random_h has
been previously defined. When the compiler reads this interface for the first time,
_random_h is undefined, which means that the compiler goes ahead and reads the
contents of the file. The compiler immediately thereafter encounters the definition

#define _random_h
Libraries and Interfaces – 90 –

Figure 3-1 The random.h interface

/*
* File: random.h
* --------------
* This interface provides several functions for generating
* pseudorandom numbers.
*/

#ifndef _random_h
#define _random_h

/*
* Function: RandomInteger
* Usage: n = RandomInteger(low, high);
* ------------------------------------
* This function returns a random integer in the range low to high,
* inclusive.
*/

int RandomInteger(int low, int high);

/*
* Function: RandomReal
* Usage: d = RandomReal(low, high);
* ---------------------------------
* This function returns a random real number in the half-open
* interval [low .. high), meaning that the result is always
* greater than or equal to low but strictly less than high.
*/

double RandomReal(double low, double high);

/*
* Function: RandomChance
* Usage: if (RandomChance(p)) . . .
* ---------------------------------
* The RandomChance function returns true with the probability
* indicated by p, which should be a floating-point number between
* 0 (meaning never) and 1 (meaning always). For example, calling
* RandomChance(.30) returns true 30 percent of the time.
*/

bool RandomChance(double p);

/*
* Function: Randomize
* Usage: Randomize();
* -------------------
* This function initializes the random-number generator so that
* its results are unpredictable. If this function is not called,
* the other functions will return the same values on each run.
*/

void Randomize();

#endif
Libraries and Interfaces – 91 –

which defines the symbol _random_h. If the compiler reads the random.h interface a
second time, the symbol _random_h has already been defined, which means that the
compiler ignores the entire contents of the file on the second pass.
The remainder of the interface consists of function prototypes and their associated
comments. The first prototype is for the function RandomInteger(low, high), which
returns a randomly chosen integer in the range between low and high, inclusive. For
example, calling RandomInteger(1, 6) would return 1, 2, 3, 4, 5, or 6 and could be used
to simulate rolling a die. Similarly, calling RandomInteger(0, 36) returns an integer
between 0 and 36 and could be used to model a European roulette wheel, which is
marked with those 37 numbers.
The function RandomReal(low, high) is conceptually similar to RandomInteger and
returns a floating-point value r subject to the condition that low _ r < high. For example,
calling RandomReal(0, 1) returns a random number that can be as small as 0 but is
always strictly less than 1. In mathematics, a range of real numbers that can be equal to
one endpoint but not the other is called a half-open interval. On a number line, a half-
open interval is marked using an open circle to show that the endpoint is excluded, like
this:

0 1

In text, the standard convention is to use square brackets to indicate closed ends of
intervals and parentheses to indicate open ones, so that the notation [0, 1) indicates the
half-open interval corresponding to this diagram.
The function RandomChance(p) is used to simulate random events that occur with
some fixed probability. To be consistent with the conventions of statistics, a probability
is represented as a number between 0 and 1, where 0 means that the event never occurs
and 1 means that it always does. The RandomChance function returns a Boolean value
that is true with probability p, so that RandomChance(0.75) returns true 75 percent of
the time. You can use RandomChance to simulate flipping a coin, as illustrated by the
following function, which returns either "heads" or "tails":

string FlipCoin() {
if (RandomChance(0.50)) {
return "heads";
} else {
return "tails";
}
}

The last function in the random.h interface requires a little more explanation. Because
computers are deterministic machines, random numbers are usually computed by going
through a deterministic calculation that nonetheless appears random to the user. Random
numbers computed in this way are called pseudorandom numbers. If you take no
special action, the computer always applies the same process to generate its sequence of
random numbers, which means that the results will be the same every time the program is
run. The purpose of the Randomize function is to initialize the internal pseudorandom
number generator so that each run of the program produces a different sequence, which is
what you want if you are writing a program that plays a game.
At first, it may seem hard to understand why a random number package should return
the same values on each run. After all, deterministic behavior of this sort seems to defeat
Libraries and Interfaces – 92 –

the whole purpose of the package. There is, however, a good reason behind this
behavior: programs that behave deterministically are easier to debug. To illustrate this
fact, suppose you have just written a program to play an intricate game, such as
Monopoly™. As is always the case with newly written programs, the odds are good that
your program has a few bugs. In a complex program, bugs can be relatively obscure, in
the sense that they only occur in rare situations. Suppose you are playing the game and
discover that the program is starting to behave in a bizarre way. As you begin to debug
the program, it would be very convenient if you could regenerate the same state and take
a closer look at what is going on. Unfortunately, if the program is running in a
nondeterministic way, a second run of the program will behave differently from the first.
Bugs that showed up the first time may not occur on the second pass.
In general, it is extremely difficult to reproduce the conditions that cause a program to
fail if the program is behaving in a truly random fashion. If, on the other hand, the
program is operating deterministically, it will do the same thing each time it is run. This
behavior makes it possible for you to recreate the conditions under which the problem
occurred. When you write a program that works with random numbers, it is usually best
to leave out the call to Randomize during the debugging phase. When the program seems
to be working well, you can insert a call to Randomize at the beginning of the main
program to make the program change its behavior from one run to the next.

Constructing a client program

Once you know what the interface looks like, you can immediately begin to code
applications that use it. The interface provides all the information that any client needs to
know. From the client’s perspective, the important questions are what the functions in
the library do and how to call them. The details of how they work are important only for
the implementation.
As a simple illustration of how clients can use the random number package, let’s
consider the craps.cpp program shown in Figure 3-2, which simulates the casino game
of craps. Because the program uses the random number library to simulate rolling the
dice, the craps.cpp file needs to include the random.h interface. Moreover, in order to
ensure that the outcome is not the same every time, the program calls Randomize at the
beginning of its operation. The rest of the program is simply an encoding of the rules of
the game, which are outlined in the comments at the beginning of the program.

The ANSI functions for random numbers

Although you can design the craps.cpp program by relying on the interface description,
you cannot run the program until you have an implementation for the random number
library. To make sense of that implementation, you first need to understand the facilities
provided by the standard libraries for generating random numbers. The basic tools are
already provided in the form of a small set of functions exported by the cstdlib
interface. Unfortunately, that interface is not well suited to the needs of clients, largely
because the results returned by the functions differ from machine to machine. One of the
advantages of random.h is that it provides clients with a machine-independent interface
that is considerably easier to use than the underlying facilities on which it is based.

The random.cpp implementation is based on a more primitive random number facility

provided as part of the standard library interface cstdlib. The function on which the
implementation depends is called rand and has the following prototype:

int rand();
Libraries and Interfaces – 93 –

Figure 3-2 A program to play the game of craps

/*
* File: craps.cpp
* ---------------
* This program plays the casino game called craps, which is
* played using a pair of dice. At the beginning of the game,
* you roll the dice and compute the total. If your first roll
* is 7 or 11, you win with what gamblers call a "natural."
* If your first roll is 2, 3, or 12, you lose by "crapping
* out." In any other case, the total from the first roll
* becomes your "point," after which you continue to roll
* the dice until one of the following conditions occurs:
*
* a) You roll your point again. This is called "making
* your point," which wins.
*
* b) You roll a 7, which loses.
*
* Other rolls, including 2, 3, 11, and 12, have no effect
* during this phase of the game.
*/

#include "genlib.h"
#include "random.h"
#include <iostream>

/* Function prototypes */

bool TryToMakePoint(int point);

int RollTwoDice();

/* Main program */

int main() {
Randomize();
cout << "This program plays a game of craps." << endl;
int point = RollTwoDice();
switch (point) {
case 7: case 11:
cout << "That's a natural. You win." << endl;
break;
case 2: case 3: case 12:
cout << "That's craps. You lose." << endl;
break;
default:
cout << "Your point is " << point << "." << endl;
if (TryToMakePoint(point)) {
cout << "You made your point. You win." << endl;
} else {
cout << "You rolled a seven. You lose." << endl;
}
}
return 0;
}
Libraries and Interfaces – 94 –

/*
* Function: TryToMakePoint
* Usage: flag = TryToMakePoint(point);
* ------------------------------------
* This function is responsible for the part of the game
* during which you roll the dice repeatedly until you either
* make your point or roll a 7. The function returns true if
* you make your point and false if a 7 comes up first.
*/

bool TryToMakePoint(int point) {

while (true) {
int total = RollTwoDice();
if (total == point) return true;
if (total == 7) return false;
}
}

/*
* Function: RollTwoDice
* Usage: total = RollTwoDice();
* -----------------------------
* This function rolls two dice and both prints and returns their sum.
*/

int RollTwoDice() {
cout << "Rolling the dice . . ." << endl;
int d1 = RandomInteger(1, 6);
int d2 = RandomInteger(1, 6);
int total = d1 + d2;
cout << "You rolled " << d1 << " and " << d2 << " - that's "
<< total << endl;
return total;
}

Unlike most functions, rand returns a different result each time it is called. The result of
rand is guaranteed to be nonnegative and no larger than the constant RAND_MAX, which is
also defined in the cstdlib interface. Thus, each time rand is called, it returns a
different integer between 0 and RAND_MAX, inclusive.
The value of RAND_MAX depends on the computer system. When you write programs
that work with random numbers, you should not make any assumptions about the precise
value of RAND_MAX. Instead, your programs should be prepared to use whatever value of
RAND_MAX the system defines.

You can get a sense of how rand behaves on your own system by running the program
const int N_TRIALS = 10;
int main() {
cout << "On this computer, RAND_MAX is " << RAND_MAX << endl;
cout << "The first " << N_TRIALS << " calls to rand:" << endl;
for (int i = 0; i < N_TRIALS; i++) {
cout << setw(10) << rand() << endl;
}
return 0;
}
Libraries and Interfaces – 95 –

On the computer in my office, the program generates the following output:

RandTest
On this computer, RAND_MAX is 2147483647
The first 10 calls to rand:
16807
282475249
1622650073
984943658
1144108930
470211272
101027544
1457850878
1458777923
2007237709

You can see that the program is generating integers, all of which are positive and none of
which is greater than 2147483647, which the sample run shows as the value of RAND_MAX
for this machine. Although this number itself seems rather arbitrary, it happens to be
231 – 1, which is the largest value of type int on my computer. Because the numbers are
pseudorandom, you know that there must be some pattern, but it is unlikely that you can
discern one. From your point of view, the numbers appear to be random, because you
don’t know what the underlying pattern is.
The rand function generates each new random value by applying a set of mathematical
calculations to the last value it produced. Because you don’t know what those
calculations are, it is best to think of the entire operation as a black box where old
numbers go in on one side and new pseudorandom numbers pop out on the other. Since,
the first call to rand produces the number 16807, the second call to rand corresponds to
putting 16807 into one end of the black box and having 282475249 appear on the other:

black
16807 282475249
box

Similarly, on the next call to rand, the implementation puts 282475249 into the black
box, which returns 1622650073:

black
282475249 1622650073
box

This same process is repeated on each call to rand. The computation inside the black box
is designed so that (1) the numbers are uniformly distributed over the legal range, and (2)
the sequence goes on for a long time before it begins to repeat.
But what about the first call to rand—the one that returns 16838? The implementation
must have a starting point. There must be an integer, s0 , that goes into the black box and
produces 16807:

black
s0 16807
box

This initial value—the value that is used to get the entire process started—is called a seed
for the random number generator. The ANSI library implementation sets the initial seed
Libraries and Interfaces – 96 –

to a constant value every time a program is started, which is why the library always
generates the same sequence of values. To change the sequence, you need to set the seed
to a different value, which is done by calling the function srand(seed).
The srand function is essential to the implementation of Randomize, which resets the
seed so that the sequence of random numbers is different each time. The usual strategy is
to use the value of the internal system clock as the initial seed. Because the time keeps
changing, the random number sequence will change as well. You can retrieve the current
value of the system clock by calling the function time, which is defined in the library
interface ctime, and then converting the result to an integer. This technique allows you
to write the following statement, which has the effect of initializing the pseudorandom
number generator to some unpredictable point:

srand(int(time(NULL)));

Although it requires only a single line, the operation to set the random seed to an
unpredictable value based on the system clock is relatively obscure. If this line were to
appear in the client program, the client would have to understand the concept of a random
number seed, along with the functions srand and time. To make things simpler for the
client, it is much better to give this operation a simple name like Randomize and make it
part of the random number library. By doing so, the client simply needs to call

Randomize();

which is certainly easier to explain.

The random.cpp implementation

The ANSI functions described in the preceding section provide all the tools you need to
implement the random.h interface. The implementation is contained in a separate source
file called random.cpp which is shown in Figure 3-3.
After the initial comment, the implementation file lists the #include files it needs. The
implementation needs cstdlib and ctime so that it has access to the functions rand ,
srand, and time. Finally, every implementation needs to include its own interface so the
compiler can check the prototypes against the actual definitions.
The rest of the file consists of the functions exported by the interface, along with any
comments that would be useful to the programmers who may need to maintain this
program in the future. Like all other forms of expository writing, comments must be
written with their audience in mind. When you write comments, you must put yourself in
the role of the reader so that you can understand what information that reader will want to
see. Comments in the .cpp file have a different audience than their counterparts in the .h
file. The comments in the implementation are written for another implementer who may
have to modify the implementation in some way. They therefore must explain how the
implementation works and provide any details that later maintainers would want to know.
Comments in the interface, on the other hand, are written for the client. A client should
never have to read the comments inside the implementation. The comments in the
interface should suffice.
The use of comments to explain the operation of the code are best illustrated by the
RandomInteger implementation, which converts a random number in the range 0 to
RAND_MAX into one that lies in the interval
between the parameters low and high. As the
comments indicate, the implementation uses the following four-step process:
Libraries and Interfaces – 97 –

Figure 3-3 The implementation of the random number library

/*
* File: random.cpp
* -----------------
* This file implements the random.h interface.
*/

#include <cstdlib>
#include <ctime>
#include "random.h"

/*
* Function: RandomInteger
* -----------------------
* This function begins by using rand to select an integer
* in the interval [0, RAND_MAX] and then converts it to the
* desired range by applying the following steps:
*
* 1. Normalize the value to a real number in the interval [0, 1)
* 2. Scale the resulting value to the appropriate range size
* 3. Truncate the scaled value to an integer
* 4. Translate the integer to the appropriate starting point
*/

int RandomInteger(int low, int high) {

double d = double(rand()) / (double(RAND_MAX) + 1);
int k = int(d * (high - low + 1));
return low + k;
}

/*
* Function: RandomReal
* --------------------
* The implementation of RandomReal is similar to that
* of RandomInteger, without the truncation step.
*/

double RandomReal(double low, double high) {

double d = double(rand()) / (double(RAND_MAX) + 1);
return low + d * (high - low);
}

/*
* Function: RandomChance
* ----------------------
* This function uses RandomReal to generate a real number
* in the interval [0, 1) and then compares that value to p.
*/

bool RandomChance(double p) {
return RandomReal(0, 1) < p;
}
Libraries and Interfaces – 98 –

/*
* Function: Randomize
* -------------------
* This function operates by setting the random number
* seed to the current time. The srand function is
* provided by the <cstdlib> library and requires an
* integer argument. The time function is exported by
* the <ctime> interface.
*/

void Randomize() {
srand(int(time(NULL)));
}

• Normalization. The first step in the process is to convert the integer result from rand
into a floating-point number d in the half-open interval [0, 1). To do so, all you need to
do is convert the result of rand to a double and then divide it by the number of
elements in the range. Because RAND_MAX is often the largest integer the machine can
hold, it is important to convert it to a double before adding the constant 1, which
ensures that the division produces a value that is strictly less than 1.
• Scaling. The second step consists of multiplying the value d by the size of the desired
range, so that it spans the correct number of integers. Because the desired range
includes both the endpoints, low and high, the number of integers in the range is given
by the expression high - low + 1.
• Truncation. The third step consists of using a type cast to convert the number back to
an integer by throwing away any fraction. This step gives you a random integer with a
lower bound of 0.
• Translation. The final step consists of adding the value low so that the range begins at
the desired lower bound.
The steps in this process are illustrated by the diagram in Figure 3-4, which shows how a
call to RandomInteger(1, 6) converts the result of rand into an integer in the desired
range of 1 to 6.

3.3 Strings
In any programming language, strings are one of the most important data types because
they come up so frequently in applications. Conceptually, a string is just a sequence of
characters. For example, the string "hello, world" is a sequence of 12 characters
including ten letters, a comma, and a space. In C++, the string data type and its
associated operations are defined in the <string> interface, and you must therefore
include this interface in any source file that manipulates string data.
In many ways, string is an ideal example of an abstract data type, which is the topic
of the following chapter. However, given that strings are essential to many applications
and that they are more deeply integrated into the syntax of C++ than the classes described
in Chapter 4, it makes sense to introduce them before the others. In this chapter, you will
look at strings only from a client perspective, which means that you will be able to use
them as a tool without having to understand how their underlying implementation.
The data type string
In Chapter 1, you learned that a data type is defined by two properties: a domain and a set
of operations. If you think about strings as a separate data type, the domain is easy to
Libraries and Interfaces – 99 –

Figure 3-4 Steps required to calculate a random integer

Initial call to r a n d
0 9768 RAND_MAX

Normalization
0 0.3 1

Scaling
0 1.8 6

Truncation
0 1 2 3 4 5 6

Translation
0 1 2 3 4 5 6

identify; the domain of type string is the set of all sequences of characters. The more
interesting problem is to identify an appropriate set of operations. Early versions of C++
and its predecessor language C did not have much support for manipulating strings. The
only facilities provided were fairly low-level operations that required you to be very
familiar with the precise representation of strings and work at a detailed level. The C++
standard introduced the standard string type that presents an abstract view of string and
a richer set of facilities for operating on strings. This allows us to focus on manipulating
strings using their published interface without concern for how strings might be internally
represented.

Operations on the string type

One common way to initialize a string variable is with a string literal, a sequence of
characters enclosed in double quotes:
string str = "Hello";

You can also create strings by concatenating existing strings using the + operator:
string str2 = str + "World";

The addition operator has been overloaded, or given multiple meanings. When + is
applied to numbers, it performs addition, when applied to strings, it performs string
concatenation. It is an error to attempt to add two operands of incompatible type such as a
double and a string.
The shorthand += form can be used to append onto an existing string. Both + and +=
allow concatenation of another string or a single character:
string str = "Stanford";
str += '!';
Libraries and Interfaces – 100 –

As much as possible, operations on strings were defined so as to mimic the behavior

of those operations on the built-in primitive types. For example, assigning one string
variable to another copies the string contents:
str2 = str1;

This assignment overwrites any previous contents of str2 with a copy of the characters
contained in str1 . The variables str1 and str2 remain independent. Changing the
characters in str1 would not cause any changes to str2.
When passing a string as parameter, a copy is made, just as with a primitive type.
Changes to the parameter made within a function are not reflected in the calling function
unless the string parameter is explicitly passed by reference.
Two strings can be compared using the relational operators, ==, !=, <, >, <=, and >=.
These operators compare strings lexicographically, which follows the order given by the
character codes. This ordering means case is significant, so "abc" is not equal to "ABC".
if (str == "quit") ...

You can retrieve the number of characters contained in a string by invoking the
lengthfunction on the string using the dot operator:
int numChars = str.length();

The syntax for calling the length function is different from what you have seen so far.
If length were like the functions in the random.h, you would expect to call it like this:

numChars = length(str); This syntax is illegal.

What makes length different is that it is an integral part of the string class and must
therefore be applied to a particular string object. In the expression
str.length()

the object is the variable str. In the language of the object-oriented paradigm, this object
to the left of the dot is called the receiver, and the syntax is intended to emphasize that
the object stored in str is receiving a request to perform the length operation.
Informally, it makes sense to think of this operation as sending a message to the object
asking for its length and having it respond by providing the requested information.
In C++, functions that are directed to a specific objects were originally called member
functions, but that designation is gradually giving way to the term method, which is
more common in other object-oriented languages. Functions that are independent of an
object—such as those you’ve seen up to now—are called free functions. Other than the
use of dot notation to identify the receiver, the mechanics of calling a method and calling
a free function are pretty much the same.
It is possible to access the ith character from a string str using the notation like that
of array selection, as follows:
str[i]

As with arrays, index numbers in a string begin at 0. Thus, if str contains the string
"Hello", then str[0] has the value 'H', str[1] has the value 'e', and so on.
Libraries and Interfaces – 101 –

The standard idiom for processing every character in a string looks like this:
for (int i = 0; i < str.length(); i++) {
. . . body of loop that manipulates str[i] . . .
}

On each loop cycle, the selection expression str[i] refers to the i th character in the
string. Because the purpose of the loop is to process every character, the loop continues
until i reaches the length of the string. Thus, you can count the number of spaces in a
string using the following function:
int CountSpaces(string str) {
int nSpaces = 0;
for (int i = 0; i < str.length(); i++) {
if (str[i] == ' ') nSpaces++;
}
return nSpaces;
}

As with array selection, selecting a character position that is out of range for the string is
not automatically recognized as an error but will typically lead to incorrect results.
The string interface exports a large number of methods, many of which come up only
in relatively specialized applications. Table 3-1 lists several of the more useful methods.

Table 3-1 Common string methods

str.length() This method returns the number of characters in the
receiver string str .
str.at(pos) Both of these expressions return the character at index pos
str[pos] within the receiver string str. However, if used to access
an invalid position, the at method raises an error whereas
bracket-selection has undefined behavior.
str.substr(pos, len) This method returns a new string made of up l e n
characters starting at pos copied from the receiver string
str . If necessary, len is truncated at the end of str. The
second parameter is optional; if not given, all characters
to the end of the receiver string are assumed. A new
string is returned; the receiver string is unchanged.
str.find(ch, pos) This method searches the receiver string str starting at pos
for the character ch and returns the index of the first
location in which it appears. If ch does not appear, the
function returns the constant string::npos. The second
parameter is optional; if not given, zero is assumed.
str .find(pattern, pos) An overloaded version of f i n d that searches for a
substring pattern instead of a single character.
str.insert(pos, txt) This method copies the characters from txt and inserts
them into the receiver string str starting at pos. Note this
operation destructively modifies the receiver string.
str.replace(pos, count, txt) This method replaces the count characters starting at pos in
the receiver string str with the entire contents of string txt.
If necessary, count is truncated to the end of str. Note this
operation destructively modifies the receiver string.
Libraries and Interfaces – 102 –

The code in Figure 3-5 provides a more extensive example of how you can use the
string interface to perform simple string manipulation. The principal function is
PigLatin, which converts an English word to
Pig Latin by applying the following rules:
• If the word contains no vowels, no translation is done, which means that the translated
word is the same as the original.
• If the word begins with a vowel, the function adds the string "way" to the end of the
original word. Thus, the Pig Latin equivalent of alley is alleyway.
• If the word begins with a consonant, the function extracts the string of consonants up
to the first vowel, moves that collection of consonants to the end of the word, and adds
the string "ay". For example, the Pig Latin equivalent of trash is ashtray.
The implementation of PigLatin uses the substr function to extract pieces of the
original word and string concatenation to reassemble them. The FindFirstVowel
function returns the integer index of the first vowel, or –1 if no vowel appears.

Figure 3-5 Implementation of Pig Latin translation using the string interface

/*
* Function: PigLatin
* Usage: translation = PigLatin(word);
* ------------------------------------
* This function translates a word from English to Pig Latin using
* the rules specified in Chapter 3. The translated word is
* returned as the value of the function.
*/

string PigLatin(string word) {

int vp = FindFirstVowel(word);
if (vp == -1) {
return word;
} else if (vp == 0) {
return word + "way";
} else {
string head = word.substr(0, vp);
string tail = word.substr(vp);
return tail + head + "ay";
}
}

/*
* Function: FindFirstVowel
* Usage: k = FindFirstVowel(word);
* --------------------------------
* This function returns the index position of the first vowel
* in word. If word does not contain a vowel, FindFirstVowel
* returns -1. The code for IsVowel appears in Chapter 1.
*/

int FindFirstVowel(string word) {

for (int i = 0; i < word.length(); i++) {
if (IsVowel(word[i])) return i;
}
return -1;
}
Libraries and Interfaces – 103 –

The principal advantage of the string interface is that strings are treated as abstract
values. When a function needs to return a string to its caller, it does so by returning a
string as its value. Any operations on that string are performed using the operations
exported by the library. The client is free from having to worry about the details of string
representation and allocation.
Even though this text relies extensively on the functions in the string interface, it is
not necessary to understand that code from the implementation side. The whole point of
interfaces is that they protect clients from the complexity of the implementation. As you
continue with your programming career, you will often make use of libraries even though
you have no understanding of their implementation, so it is best to become comfortable
with that process as early as you can.

The strutils.h interface

There are a few minor convenience operations on strings that are used in this text as a
supplement to the standard library. These functions are simple adaptations of existing
standard functionality, but packaged in a way to be more convenient for use. This
interface is called strutils.h, and is presented in its complete form in Figure 3-6.

An aside about C-style strings

Prior to the introduction of the string class into C++, strings were implemented as raw

Figure 3-6 Interface to the string utilities library

/*
* File: strutils.h
* --------------
* The strutils.h file defines the interface for a library of string
* utilities.
*/

#ifndef _strutils_h
#define _strutils_h

#include "genlib.h"

/*
* Function: ConvertToLowerCase
* Usage: s = ConvertToLowerCase(s);
* ---------------------------------
* This function returns a new string with all
* alphabetic characters converted to lower case.
*/

string ConvertToLowerCase(string s);

/*
* Function: ConvertToUpperCase
* Usage: s = ConvertToUpperCase(s);
* ---------------------------------
* This function returns a new string with all
* alphabetic characters converted to upper case.
*/

string ConvertToUpperCase(string s);

Libraries and Interfaces – 104 –

/*
* Function: IntegerToString
* Usage: s = IntegerToString(n);
* ------------------------------
* This function converts an integer into the corresponding
* string of digits. For example, IntegerToString(123)
* returns "123" as a string.
*/

string IntegerToString(int n);

/*
* Function: StringToInteger
* Usage: n = StringToInteger(s);
* ------------------------------
* This function converts a string of digits into an integer.
* If the string is not a legal integer or contains extraneous
* characters, StringToInteger signals an error condition.
*/

int StringToInteger(string s);

/*
* Function: RealToString
* Usage: s = RealToString(d);
* ---------------------------
* This function converts a floating-point number into the
* corresponding string form. For example, calling
* RealToString(23.45) returns "23.45".
*/

string RealToString(double d);

/*
* Function: StringToReal
* Usage: d = StringToReal(s);
* ---------------------------
* This function converts a string representing a real number
* into its corresponding value. If the string is not a
* legal floating-point number or if it contains extraneous
* characters, StringToReal signals an error condition.
*/

double StringToReal(string s);

#endif

arrays of characters with a sentinel value, the null character, used to mark the end of the
string. You can recognize older programs that operate on these C-style strings by their
use of the data type char* or char[]. Working with C-style strings is error-prone
because of the exposed-pointer implementation and issues of allocation. Using the C++
string type frees the programmer from managing string storage and decreases the
probability of difficult pointer bugs. Thus, you should use C++ strings wherever possible.
That said, there are some facilities in the standard libraries that predate the introduction of
the C++ string type and require the use of the older C-style strings. You can obtain the
C-style equivalent from a C++ string using the method c_str:
Libraries and Interfaces – 105 –

string str = ...

char *cstr = str.c_str();

Some of the legacy of C++ is demonstrated by the fact that string literals are in fact C-
style strings. However, they are automatically converted to C++ strings as needed in most
contexts, so you can use a string literal in most places where a C++ string is expected.
Where needed, you can explicitly convert a string literal to a C++ string using a typecast-
like notation:
string str = string("Hello");

3.4 Standard I/O and file streams

The most commonly used library is the I/O stream library. Every program in the text
includes iostream , in order to display output using the cout stream. The iostream
interface is extended in the fstream interface with additional features that make it
possible to read and write files, which in turn enable you to store data values that persist
even after your program completes its operation.

Data files
Whenever you want to store information on the computer for longer than the running
time of a program, the usual approach is to collect the data into a logically cohesive
whole and store it on a permanent storage medium as a file. Ordinarily, a file is stored
using magnetic or optical media, such as on a removable floppy disk, a compact disc, or a
hard disk. The important point is that the permanent data objects you store on the
computer—documents, games, executable programs, source code, and the like—are all
stored in the form of files.
On most systems, files come in a variety of types. For example, in the programming
domain, you work with source files, object files, and executable files, each of which has a
distinct representation. When you use a file to store data for use by a program, that file
usually consists of text and is therefore called a text file. You can think of a text file as a
sequence of characters stored in a permanent medium and identified by a file name. The
name of the file and the characters it contains have the same relationship as the name of a
variable and its contents.
As an example, the following text file contains the first stanza of Lewis Carroll’s
nonsense poem “Jabberwocky,” which appears in Through the Looking Glass:
jabberwocky.txt
'Twas brillig, and the slithy toves
Did gyre and gimble in the wabe;
All mimsy were the borogoves,
And the mome raths outgrabe.

When you look at a file, it is often convenient to regard it as a two-dimensional

structure—a sequence of lines composed of individual characters. Internally, however,
text files are represented as a one-dimensional sequence of characters. In addition to the
printing characters you can see, files also contain the newline character '\n', which
marks the end of each line. The file system also keeps track of the length of the file so
programs that read files can recognize where the file ends.
In many respects, text files are similar to strings. Each consists of an ordered
collection of characters with a specified endpoint. On the other hand, strings and files
differ in several important respects. The most important difference is the permanence of
Libraries and Interfaces – 106 –

the data. A string is stored temporarily in the computer’s memory during the time that a
program runs; a file is stored permanently on a long-term storage device until it is
explicitly deleted. There is also a difference in the way you use data in strings and files.
The individual characters in a string may be accessed in any order by specifying the
appropriate index. The characters in a text file are usually accessed in a sequential
fashion, with a program starting at the beginning of the file and either reading or writing
characters from the start to the end in order.

Using file streams in C++

To read or write a file as part of a C++ program, you must use the following steps:
1. Declare a stream variable. The fstream interface exports two types, ifstream, for
files being read from, and ofstream, for files being written to. Each type keeps track
of the information the system needs to manage file processing activity. Thus, if you
are writing a program that reads an input file and creates a second file for output, you
need to declare two variables, as follows:
ifstream infile;
ofstream outfile;

2. Open the file. Before you can use a stream variable, you need to establish an
association between that variable and an actual file. This operation is called opening
the file and is performed by calling the stream method open. Like the methods seen
previously on string, you invoke it using dot notation on the stream. For example, if
you wanted to read the text of the jabberwocky.txt file, you would execute the
statement
infile.open("jabberwocky.txt");
One minor glitch to note is that the open method expects a C-style string as the
argument. A string literal is acceptable as is, but if you have the filename in a C++
string variable, you will need to pass its C-style equivalent as shown:
string str = ...
infile.open(str.c_str());
If a requested input file is missing, the open operation will fail. You can test the
current state of a stream by using the fail method. As a programmer, you have a
responsibility to check for this error and report it to the user.
if (infile.fail()) Error("Could not open file.");
In some cases, you may want to recover from such an input file failure (for example,
if the user accidentally typed a non-existent filename), and try to open the input
stream once again. In such cases, you must first clear the "fail state" that the input
stream variable will be in after such a failure. You can do this by calling the clear
method on the stream, as exemplified below.
infile.clear();

3. Transfer the data. Once you have opened the data files, you then use the appropriate
stream operations to perform the actual I/O operations. For an input file, these
operations read data from the file into your program; for an output file, the operations
transfer data from the program to the file. To perform the actual transfers, you can
choose any of several strategies, depending on the application. At the simplest level,
you can read or write files character by character. In some cases, however, it is more
convenient to process files line by line. At a still higher level, you can choose to read
Libraries and Interfaces – 107 –

and write formatted data. Doing so allows you to intermix numeric data with strings
and other data types.
4. Close the file. When you have finished all data transfers, you need to indicate that
fact to the file system by calling the stream method close. This operation, called
closing the file, breaks the association between a stream variable and the actual file.
infile.close();

Standard streams
The standard I/O library defines three special identifiers—cin, cout, and cerr—that act
as stream variables and are available to all programs. These variables are referred to as
standard streams. The constant cin designates the standard input stream, which is the
source for user input. The constant cout indicates the standard output stream and
represents the device on which output data is written to the user. The constant cerr
represents the standard error stream and is used to report any error messages the user
should see. Typically, the standard streams all refer to the computer console. When you
read data from cin, the input comes from the keyboard; when you write data to cout or
cerr, the output appears on the screen. You do not open or close these streams; they are
handled automatically.
Formatted stream output
Stream output is usually accomplished by using the insertion operator <<. The operand on
the left is the output stream; the operand on the right is the data being inserted into the
stream. Thus far, you have only written to cout, but any open output stream can be used
as the destination.
outfile << "some text " << num << endl;

The insertion operator will accept any primitive data type on the right. Arrays are
treated as pointers in this context and only the base address is printed. Strings print the
entire sequence of characters. By default, complex types such as structs cannot be printed
with a single insertion operation, you must output each of the struct members one by one.
Several insertions can be chained together by multiple uses of the insertion operator. The
manipulator endl is used to output a newline.
Values are formatted according to the stream’s default rules unless you specify
otherwise. For explicit control, you insert I/O manipulators to set the field width,
alignment, precision, and other features on a per-stream basis. A few common output
manipulators options are described in the section on “Simple input and output” in
Chapter 1. Although the mechanisms for formatted I/O in any programming language can
be quite useful, they also tend to be detail-ridden. I recommend having a reference on
hand for looking up details on as-needed basis rather than memorizing the entirety of the
options.

Formatted stream input

The extraction operator >> provides an input counterpart to the insertion operator << and
allows programs to read in values of the various basic types. Here is a sample usage of
reading an integer from an input stream:
int num;
infile >> num;
Libraries and Interfaces – 108 –

The operand on the left is the input stream; the operand on the right is the variable that
will store the value read from the stream. The previous value of the variable is
overwritten by the next integer read from the stream.
The default setting for an input stream is to skip characters that appear as blank space,
such as the space, tab, and newline characters. When reading the next value, the insertion
operator will skip over any leading blank space and start reading from the first non-blank
character. When extracting a string from an input stream, a sequence of non-blank
characters is read, stopping at the first blank. To illustrate how blank space is handled,
consider the a text file with the following contents:
data.txt
3 45
some text

The code below attempts to extract two integers, a string, and character from this file:
int num1, num2;
string str;
char ch;
infile >> num1 >> num2 >> str >> ch;
After executing this code, num1 would hold the value 3, num2 the value 45, str would be
"some" , and ch would be 't'.

The extraction operator will fail if there is no input remaining or if the next input is not
compatible with the type you are attempting to extract. The stream method fail can be
used to test whether an extraction operation was successful.
if (infile.fail()) ...

If you were attempting to read and sum integers from a file until you reached the end, you
could use a loop such as this one:
int total, num;
total = 0;
while (true) {
infile >> num;
if (infile.fail()) break;
total += num;
}
For a program to be robust, it typically must take care to handle malformed input. This
is especially important when reading from cin, where the human user is prone to typing
errors at the console. To avoid cluttering the code with error-handling, this text does not
use the extraction operator on cin and relies instead on the facilities provided by the
simpio.h interface described in Chapter 1 that have error-handling built-in.

Single character I/O

Most I/O is handled with the insertion and extraction operators, but there are times when
it is more convenient to process files as a sequence of characters or lines. The method get
is used to read a single character from an input stream. Although the idea of get seems
simple enough, there is a confusing aspect in its design. If you look at the formal
definition of get, its prototype looks like this:
int get();
Libraries and Interfaces – 109 –

At first glance, the result type seems odd. The prototype indicates that COMMON PITFALLS
get returns an integer, even though conceptually the function returns a
character. The reason for this design decision is that returning a Remember that get
returns an int, not a
character would make it impossible for a program to detect the end-of- char. If you use a
file mark. There are only 256 possible character codes, and a data file variable of type char to
might contain any of those values. There is no value—or at least no store the result of get,
value of type char—that you could use as a sentinel to indicate the your program will be
end-of-file condition. By extending the definition so that get returns unable to detect the end-
an integer, the implementation can return a value outside the range of of-file condition.
legal character codes to indicate the end-of-file condition. That value
is given the symbolic name of EOF.
For output streams, the stream method put takes just one argument, a single character,
and writes that character to the stream:
outfile.put(ch);

As an example of the use of get and put, you can copy one file to another by calling
the following function:
void CopyFile(istream & infile, ostream & outfile) {
int ch;
while ((ch = infile.get()) != EOF) {
outfile.put(ch);
}
}

The while loop in CopyFile is highly idiomatic and deserves some consideration.
The test expression for the while loop uses embedded assignment to combine the
operations of reading in a character and testing for the end-of-file condition. When the
program evaluates the while condition, it begins by evaluating the subexpression
ch = infile.get()

which reads a character and assigns it to ch . Before executing the loop body, the
program then goes on to make sure the result of the assignment is not EOF . The
parentheses around the assignment are required; without them, the expression would
incorrectly assign to ch the result of comparing the character against EOF.
Note that both stream parameters are passed by reference to the function CopyFile.
Reading from and writing to a stream changes its internal state, and it is essential to pass
streams by reference so that the stream state is consistently maintained throughout the
context of the entire program.

Rereading characters from an input file

When you are reading data from an input file, you will often find yourself in the position
of not knowing that you should stop reading characters until you have already read more
than you need. For example, suppose that you are asked to write a program that copies a
program from one file to another, removing all comments as it does so. As you know, a
comment in C++ begins with the character sequence /* and ends with the sequence */.
A program to remove them must copy characters until it detects the initial /* sequence
and then read characters without copying them until it detects the */ at the end. The only
aspect of this problem that poses any difficulty is the fact that the comment markers are
two characters long. If you are copying the file a character at a time, what do you do
when you encounter a slash? It might be the beginning of a comment, in which case you
Libraries and Interfaces – 110 –

should not copy it to the output file. On the other hand, it might be the division operator.
The only way to determine which of these cases applies is to look at the next character. If
it is an asterisk, you need to ignore both characters and make note of the fact that a
comment is in progress. If it not, however, what you would most like to do is forget that
you ever read that character and then treat it normally on the next cycle of the loop.
The stream interface provides a function that allows you to do just that. The method is
called unget and has the following form:
infile.unget();

The effect of this call is to “push” the last character read back into the input stream so
that it is returned on the next call to get. The C++ libraries only guarantee the ability to
push back one character into the input file, so you should not rely on being able to read
several characters ahead and then push them all back. Fortunately, being able to push
back one character is sufficient in the vast majority of cases.
An implementation of a function CopyRemovingComments that uses unget is shown in
Figure 3-7.

Line-oriented I/O
Because files are usually subdivided into individual lines, it is often useful to read an
entire line of data at a time. The stream function that performs this operation is called
getline . (not to be confused with the similarly-named GetLine function from the
simpio.h interface). Unlike the other stream operations we’ve discussed, getline is not

Figure 3-7 Implementation of CopyRemovingComments

void CopyRemovingComments(ifstream & infile, ofstream & outfile) {

int ch, nch;
bool commentFlag;
commentFlag = false;
while ((ch = infile.get()) != EOF) {
if (commentFlag) {
if (ch == '*') {
nch = infile.get();
if (nch == '/') {
commentFlag = false;
} else {
infile.unget();
}
}
} else {
if (ch == '/') {
nch = infile.get();
if (nch == '*') {
commentFlag = true;
} else {
infile.unget();
}
}
if (!commentFlag) outfile.put(ch);
}
}
}
Libraries and Interfaces – 111 –

a method and is not invoked using dot notation. getline takes two arguments, the input
stream to read from, and a string; both are reference parameters.
getline(infile, str);

The effect of this function is to copy the next line of the file into the string parameter.
getline removes the newline character used to signal the end of the line, so that the
string contains simply the characters on the line. The string will be the empty string if the
next character coming up in the stream is a newline. The getline function will fail if
there are no more lines to be read from the file. As always, you can test whether the
stream is in failure state using the fail method
Some of the most common operations exported by the iostream and fstream
interfaces are summarized in Table 3-3.

Table 3-3 Common operations for the iostream and fstream classes
Call Operations for all streams
file.open(filename) This method attempts to open the named file and attach it to the
receiver stream file. Note that the filename parameter is a C-style
string, not a C++ string object. You can convert a C++ string to
its C-style equivalent with the string method c_str . You can
check whether open fails by calling the fail method.
file.close() This method closes the file attached to the stream file.
file.fail() This method returns a boolean indicating the error state of the
stream file. A true result means that a previous operation on this
stream failed for some reason. Once an error occurs, the error state
must be cleared before any further operations will succeed.
file.clear() This method clears the error state of the stream file. After an error,
you must clear the stream before executing subsequent operations.
Operations for input streams
infile.get() This method reads and returns the next character from the input
stream infile. If there are no more characters, get returns the
constant EOF. Note that the return value is of type int, not char.
infile.unget() This method pushes the last character read back onto the input
stream infile .
getline(infile, str) This function reads the next line from the input stream infile into
the reference parameter str , discarding the newline. Note this is not
a method, but an ordinary free function. (This is a different
function than GetLine from simpio.h.)
infile >> var The stream extraction operation reads a value into var from the
input stream infile. By default, leading whitespace is skipped. If
there is no input remaining or the next input is not compatible
with the type of var, the stream is set into an error state.
Operations for output streams
outfile.put(ch) This method writes the character ch to the output stream outfile.
outfile << expr This stream insertion operations writes the value of expr to the
output stream outfile. The expression can evaluate to any type that
has a defined stream insertion behavior, generally this means all
primitive types and string.
Libraries and Interfaces – 112 –

3.5 Other ANSI libraries

In addition to the string and fstream interfaces described earlier in this chapter, there
are a few other interfaces in the standard library that are important for you to know. The
most important of these are the cctype interface, which exports a set of functions for
working with characters, and the cmath interface, which exports several common
mathematical functions. The most important functions exported by these interfaces are
shown in Tables 3-4 and 3-5, respectively.

Table 3-4 Common functions exported by the cctype interface

isupper(ch) These functions return true if ch is an uppercase letter, a lowercase
islower(ch) letter, or any type of letter, respectively.
isalpha(ch)
isdigit(ch) This function returns true if ch is one of the digit characters.
isalnum(ch) This function returns true if ch is alphanumeric, which means that it
is either a letter or a digit.
ispunct(ch) This function returns true if ch is a punctuation symbol.
isspace(ch) This function returns true if ch is a whitespace character. These
characters are ' ' (the space character), '\t', '\n', '\f', or '\v', all
of which appear as blank space on the screen.
isprint(ch) This function returns true if ch is any printable character, including the
whitespace characters.
toupper(ch) If ch is a letter, these functions return ch converted to the desired case,
tolower(ch) so that tolower('A') returns 'a'; if not, ch is returned unchanged.

Table 3-5 Common functions exported by the math interfaces

fabs(x) This function returns the absolute value of a real number x. (Note: The
function abs, which takes the absolute value of an integer, is exported
by cstdlib)
floor(x) This function returns the floating-point representation of the largest
integer less than or equal to x.
ceil(x) This function returns the floating-point representation of the smallest
integer greater than or equal to x.
fmod(x, y) This function returns the floating-point remainder of x / y .
sqrt(x) This function returns the square root of x.
pow(x, y) This function returns xy.
exp(x) This function returns ex .
log(x) This function returns the natural logarithm of x.
sin(theta) These functions return the sine and cosine of the angle theta, which is
cos(theta) expressed in radians. You can convert from degrees to radians by
multiplying by π / 180.
atan(x) This function returns the trigonometric arctangent of the value x. The
result is an angle expressed in radians between –π/2 and + π /2.
atan2(y, x) This function returns the angle formed between the x-axis and the line
extending from the origin through the point (x , y). As with the other
trigonometric functions, the angle is expressed in radians.
Libraries and Interfaces – 113 –

In some cases, the functions in these interfaces are easy to implement on your own.
Even so, it is good programming practice to use the library functions instead of writing
your own. There are three principal reasons for doing so:
1. Because the library functions are standard, programs you write will be easier for other
programmers to read. Assuming that the programmers are at all experienced in C++,
they will recognize these functions and know exactly what they mean.
2. It is easier to rely on library functions for correctness than on your own. Because the
C++ libraries are used by millions of client programmers, there is considerable
pressure on the implementers to get the functions right. If you rewrite library
functions yourself, the chance of introducing a bug is much larger.
3. The library implementations of functions are often more efficient than those you can
write yourself. Because these libraries are standard and their performance affects
many clients, the implementers have a large incentive to optimize the performance of
the libraries as much as possible.

Summary
In this chapter, you have learned about interfaces, which are the points of connection
between the client of a library abstraction and the corresponding implementation.
Interfaces are one of the central themes of modern programming and will be used
extensively throughout the rest of this text. You have also learned how to use several
specific interfaces including random.h, string, fstream, cctype, and cmath.
Important points in this chapter include:

• An interface provides a channel of communication between a client of a library and its

implementation but also acts as a barrier to keep unnecessary information from
crossing the boundary between the two sides.
• Interfaces in C++ are represented using header files.
• The definitions exported by an interface are called interface entries. The most
common interface entries are function prototypes, constant definitions, and type
definitions. The interface should contain extensive comments for each entry so that
the client can understand how to use that entry but should not expose details that are
relevant only to the implementation.
• A well-designed interface must be unified, simple, sufficient, general, and stable.
Because these criteria sometimes conflict with each other, you must learn to strike an
appropriate balance in your interface design.
• You can use the random.h interface to generate pseudorandom numbers that appear to
be random even though they are generated deterministically. The random.h interface
exports four entries: RandomInteger, RandomReal, RandomChance, and Randomize.
• The interface string provides the string class, an abstraction used for manipulating
sequences of characters. The strutils.h interface adds some convenience operations
on strings.
• The fstream interface exports operations that allow you to read and write data files.
The process of using a data file consists of four steps: declaring a stream variable,
opening the file, transferring data, and closing the file.

Review questions
1. Define the following terms: interface, package, abstraction, implementation, client.
Libraries and Interfaces – 114 –

2. In your own words, describe the difference in perspective between a programmer

who writes the implementation of a library and one who writes programs that are
clients of that library.
3. How are interfaces represented in C++?
4. What are the most common interface entries?
5. What are the five criteria for good interface design listed in this chapter?
6. Why is it important for an interface to be stable?
7. What is meant by the term extending an interface?
8. Why are comments particularly important in interfaces?
9. True or false: The comments in an interface should explain in detail how each of the
exported functions is implemented.
10. If you were defining an interface named magic.h , what would the interface
boilerplate look like? What is the purpose of these lines?
11. Why are the values generated by the rand function called pseudorandom numbers ?
12. How would you use RandomInteger to generate a random four-digit positive
number?
13. Could you use the multiple assignment statement
d1 = d2 = RandomInteger(1, 6);

to simulate the process of rolling two dice?

14. The rand function ordinarily generates the same sequence of random numbers every
time a program is run. What is the reason for this behavior?
15. What is meant by the term seed in the context of random numbers?
16. What four steps are necessary to convert the result of rand into an integer value
between the limits low and high?
17. Why is it not necessary to understand how a string is stored in memory?
18. What idiom can you use to process each character in a string?
19. What is the purpose of the strutils.h interface?
20. Assuming that s1 and s2 are strings, describe the effect of the conditional test in the
following if statement:
if (s1 == s2) . . .
21. If you call s1.replace(0, 1, s2), which string is the receiver?
22. What happens if you attempt to access a character at an invalid index of a string?
23. How are two strings of different cases handled when being compared using the
relational operators on two strings?
Libraries and Interfaces – 115 –

24. What is the result of calling each of the following functions from the string
interface?
string s = "ABCDE", t = "";
a. s.length();
b. t.length();
c. s[2]
d. s + t
e. t += 'a'
f. s.replace(0, 2, "Z")
g. s.substr(0, 3)
h. s.substr(4)
i. s.substr(3, 9)
j. s.substr(3, 3)

25. What is the result of calling each of the following functions?

string s = "ABCDE", t = "Abracadabra";
a. s == "abcde"
b. s == "ABCDE"
c. s == "ABC"
d. s < "abcde"
e. t.find('a', 0)
f. t.find("ra", 3)
g. t.find("cad)
h. t.find("CAD", 1);
i. ConvertToLowerCase("Catch-22")
j. RealToString(3.140)

26. What is the purpose of the types ifstream and ofstream? Is understanding the
underlying structure of these types important to most programmers?
27. The argument to open must be a C-style string. What is the significance of this
requirement?
28. How can you determine if an open operation on a stream was unsuccessful?
29. The iostream interface automatically defines three standard streams. What are their
names? What purpose does each one serve?
30. When you are using the get method, how do you detect the end of a file?
31. Why is the return type of get declared as int instead of char?
32. What is the purpose of the unget method?
33. How can you determine whether an extraction operation on a stream was successful?
34. What are the differences between the functions GetLine and getline?
Libraries and Interfaces – 116 –

35. True or false: It is very worthwhile to memorize all the features of the standard I/O
library since they are used so extensively.
36. What is the result of each of the following calls from cctype:
a. isdigit(7)
b. isdigit('7')
c. isalnum('7')
d. toupper('7')
e. toupper('A')
f. tolower('A')

37. When using the trigonometric functions in the cmath interface, how can you convert
an angle from degrees to radians?

Programming exercises
1. Write a program that repeatedly generates a random real number between 0 and 1
and then displays the average after a certain number of runs, as illustrated by the
following sample run:
AverageRand
This program averages a series of random numbers
between 0 and 1.
How many trials: 10000
The average value after 10000 trials is 0.501493

If the random number generator is working well, the average should get closer to 0.5
as the number of trials increases.
2. Heads. . . .
Heads. . . .
Heads. . . .
A weaker man might be moved to re-examine his faith, if in nothing
else at least in the law of probability.
— Tom Stoppard, Rosencrantz and Guildenstern Are Dead, 1967
Write a program that simulates flipping a coin repeatedly and continues until three
consecutive heads are tossed. At that point, your program should display the total
number of coin flips that were made. The following is one possible sample run of
the program:
ConsecutiveHeads
tails
heads
heads
tails
tails
heads
tails
heads
heads
heads
It took 10 flips to get 3 consecutive heads.
Libraries and Interfaces – 117 –

3. In casinos from Monte Carlo to Las Vegas, one of the most common gambling
devices is the slot machine—the “one-armed bandit.” A typical slot machine has
three wheels that spin around behind a narrow window. Each wheel is marked with
the following symbols: CHERRY, LEMON, ORANGE, PLUM, BELL, and BAR. The window,
however, allows you to see only one symbol on each wheel at a time. For example,
the window might show the following configuration:

BELL ORANGE BAR

If you put a dollar into a slot machine and pull the handle on its side, the wheels spin
around and eventually come to rest in some new configuration. If the configuration
matches one of a set of winning patterns printed on the front of the slot machine, you
get back some money. If not, you’re out a dollar. The following table shows a
typical set of winning patterns, along with their associated payoffs:

BAR BAR BAR pays $250

BELL BELL BELL/BAR pays $20
PLUM PLUM PLUM/BAR pays $14
ORANGE ORANGE ORANGE/BAR pays $10
CHERRY CHERRY CHERRY pays $7
CHERRY CHERRY — pays $5
CHERRY — — pays $2

The notation BELL/BAR means that either a BELL or a BAR can appear in that position,
and the dash means that any symbol at all can appear. Thus, getting a CHERRY in the
first position is automatically good for $2, no matter what appears on the other
wheels. Note that there is never any payoff for the LEMON symbol, even if you
happen to line up three of them.
Write a program that simulates playing a slot machine. Your program should
provide the user with an initial stake of $50 and then let the user play until the money
runs out or the user decides to quit. During each round, your program should take
away a dollar, simulate the spinning of the wheels, evaluate the result, and pay the
user any appropriate winnings. For example, a user might be lucky enough to see the
following sample run:
Slots
Would you like instructions? no
You have $50. Would you like to play? yes
PLUM LEMON LEMON -- You lose
You have $49. Would you like to play? yes
PLUM BAR LEMON -- You lose
You have $48. Would you like to play? yes
BELL LEMON ORANGE -- You lose
You have $47. Would you like to play? yes
CHERRY CHERRY ORANGE -- You win $5
You have $51. Would you like to play? yes
BAR BAR BAR -- You win $250
You have $300. Would you like to play? no

Even though it’s not realistic (and would make the slot machine unprofitable for the
casino), you should assume that the six symbols are equally likely on each wheel.
Libraries and Interfaces – 118 –

4. Implement a function EqualIgnoringCase(str1, str2) that returns true if the

two string parameters contain the same sequence of characters ignoring case
distinctions. Implement this once using the convenience functions from strutils.h
and again without.
5. Implement a function Capitalize(str) that returns a string in which the initial
character is capitalized (if it is a letter) and all other letters are converted so that they
appear in lowercase form. Characters other than letters are not affected. For
example, Capitalize("BOOLEAN") and Capitalize("boolean") should each
return the string "Boolean".
6. A palindrome is a word that reads identically backward and forward, such as level
or noon. Write a predicate function IsPalindrome(str) that returns true if the
string str is a palindrome. In addition, design and write a test program that calls
IsPalindrome to demonstrate that it works.

7. One of the simplest types of codes used to make it harder for someone to read a
message is a letter-substitution cipher, in which each letter in the original message
is replaced by some different letter in the coded version of that message. A
particularly simple type of letter-substitution cipher is a cyclic cipher, in which each
letter is replaced by its counterpart a fixed distance ahead in the alphabet. The word
cyclic refers to the fact that if the operation of moving ahead in the alphabet would
take you past Z, you simply circle back to the beginning and start over again with A.
Using the string functions provided by the string interface, implement a function
EncodeString with the prototype
string EncodeString(string str, int key);
The function returns the new string formed by shifting every letter in str forward
the number of letters indicated by key, cycling back to the beginning of the alphabet
if necessary. For example, if key has the value 4, the letter A becomes E, B becomes
F, and so on up to Z, which becomes D because the coding cycles back to the
beginning. If key is negative, letter values should be shifted toward the beginning of
the alphabet instead of the end.
After you have implemented EncodeString, write a test program that duplicates
the examples shown in the following sample run:
CyclicCipher
This program encodes messages using a cyclic cipher.
To stop, enter 0 as the key.
Enter the key: 4
Enter a message: ABCDEFGHIJKLMNOPQRSTUVWXYZ
Encoded message: EFGHIJKLMNOPQRSTUVWXYZABCD
Enter the key: 13
Enter a message: This is a secret message.
Encoded message: Guvf vf n frperg zrffntr.
Enter the key: -1
Enter a message: IBM-9000
Encoded message: HAL-9000
Enter the key: 0

Note that the coding operation applies only to letters; any other character is included
unchanged in the output. Moreover, the case of letters is unaffected: lowercase
letters come out as lowercase, and uppercase letters come out as uppercase.
Libraries and Interfaces – 119 –

8. Without using the string method s u b s t r , implement your own function

SubString(s, pos, len) , which returns the substring of s , beginning at position
pos and including at most len characters. Make sure that your function correctly
applies the following rules:
a. If pos is negative, it is set to 0 so that it indicates the first character in the string.
b. If len is greater than s.length()– pos, it is set to strlen(s) – pos so that it
stops at last character.
c. If pos is greater than s.length() - 1, SubString returns the empty string.

9. Write a program wc.cpp that reads a file and reports how many lines, words, and
characters appear in it. For the purposes of this program, a word consists of a
consecutive sequence of any characters except whitespace characters. For example,
if the file twinkle.txt contains the following verse from Alice in Wonderland,

Twinkle, twinkle, little bat!

How I wonder what you're at!
Up above the world you fly,
Like a teatray in the sky.

your program should be able to generate the following sample run:

LineWordCharCount
File: twinkle.txt
Lines: 4
Words: 22
Chars: 114

10. In the 1960s, entertainer Steve Allen often played a game called madlibs as part of
his comedy routine. Allen would ask the audience to supply words that fit specific
categories—a verb, an adjective, or a plural noun, for example—and then use these
words to fill in blanks in a previously prepared text that he would then read back to
the audience. The results were usually nonsense, but often very funny nonetheless.
In this exercise, your task is to write a program that plays madlibs with the user.
The text for the story comes from a text file that includes occasional placeholders
enclosed in angle brackets. For example, suppose the input file carroll.txt
contains the following excerpt from Lewis Carroll’s poem “The Walrus and the
Carpenter,” with a few key words replaced by placeholders as shown:

"The time has come," the <animal> said,

"To talk of many things:"
"Of <plural noun>-and ships-and sealing <sticky substance>-
Of <plural vegetable name>-and kings-
And why the <body of water> is boiling hot-
And whether <plural animal name> have wings.”

Your program must read this file and display it on the console, giving the user the
chance to fill in the placeholders with new strings. If Carroll himself had used the
program, he would likely have obtained the following sample run:
Libraries and Interfaces – 120 –

MadLibs
Input file: carroll.txt
animal: Walrus
plural noun: shoes
sticky substance: wax
plural vegetable name: cabbages
body of water: sea
plural animal name: pigs

"The time has come," the Walrus said,

"To talk of many things:"
"Of shoes-and ships-and sealing wax-
Of cabbages-and kings-
And why the sea is boiling hot-
And whether pigs have wings."

Note that the user must provide all the substitutions before any of the text is
displayed. This design complicates the program structure slightly because it is
impossible to display the output text as you go. The simplest strategy is to write the
output to a temporary file first and then copy the contents of the temporary file back
to the screen.
11. Write a program that plays the game of hangman. In hangman, the computer begins
by selecting a secret word at random from a list of possibilities. It then prints out a
row of dashes—one for each letter in the secret word—and asks the user to guess a
letter. If the user guesses a letter that appears in the word, the word is redisplayed
with all instances of that letter shown in the correct positions, along with any letters
guessed correctly on previous turns. If the letter does not appear in the word, the
player is charged with an incorrect guess. The player keeps guessing letters until
either (1) the player has correctly guessed all the letters in the word or (2) the player
has made eight incorrect guesses. A sample run of the hangman program is shown in
Figure 3-8.
To separate the process of choosing a secret word from the rest of the game,
define and implement an interface called randword.h that exports two functions:
InitDictionary and ChooseRandomWord. InitDictionary should take the name
of a data file containing a list of words, one per line, and read it into an array
declared as a static global variable in the implementation. ChooseRandomWord takes
no arguments and returns a word chosen at random from the internally maintained
array.
12. Design and implement an interface called card.h that exports the following interface
entries:

• A type rankT that allows you to represent the rank of a card. The values of type
rankT include the integers between 2 and 10 but should also include the constants
Ace, Jack, Queen, and King.
• A type suitT consisting of the four suits: Clubs, Diamonds, Hearts, and Spades.
• A type cardT that combines a rank and a suit.
• A function NewCard(rank, suit) that creates a cardT from the rank and suit values.
• Two functions, Rank(card) and Suit(card), that allow the client to select the rank
and suit of a cardT value. These functions could easily be replaced by code that
selected the appropriate components of the card, but defining them as functions
means that the client need not pay attention to the underlying structure of the type.
Libraries and Interfaces – 121 –

Figure 3-8 Sample run of the hangman program

Hangman
Welcome to Hangman!
I will guess a secret word. On each turn, you guess
a letter. If the letter is in the secret word, I
will show you where it appears; if not, a part of
your body gets strung up on the scaffold. The
object is to guess the word before you are hung.
The word now looks like this: --------
You have 8 guesses left.
Guess a letter: E
That guess is correct.
The word now looks like this: ------E-
You have 8 guesses left.
Guess a letter: A
There are no A's in the word.
The word now looks like this: ------E-
You have 7 guesses left.
Guess a letter: I
There are no I's in the word.
The word now looks like this: ------E-
You have 6 guesses left.
Guess a letter: O
That guess is correct.
The word now looks like this: -O----E-
You have 6 guesses left.
Guess a letter: S
There are no S's in the word.
The word now looks like this: -O----E-
You have 5 guesses left.
Guess a letter: T
That guess is correct.
The word now looks like this: -O---TE-
You have 5 guesses left.
Guess a letter: R
That guess is correct.
The word now looks like this: -O---TER
You have 5 guesses left.
Guess a letter: N
There are no N's in the word.
The word now looks like this: -O---TER
You have 4 guesses left.
Guess a letter: P
That guess is correct.
The word now looks like this: -O-P-TER
You have 4 guesses left.
Guess a letter: C
That guess is correct.
The word now looks like this: CO-P-TER
You have 4 guesses left.
Guess a letter: M
That guess is correct.
The word now looks like this: COMP-TER
You have 4 guesses left.
Guess a letter: U
That guess is correct.
You guessed the word: COMPUTER
You win.
Libraries and Interfaces – 122 –

• A function CardName(card) that returns a string identifying the card. The result of
CardName begins with a rank indicator (which is one of A, 2, 3, 4, 5, 6, 7, 8, 9, 10,
J, Q, or K), followed by a one-character suit (C, D, H, or S). Note that the result
is usually a two-character string, but contains three characters if the rank is a 10.

13. Using the card.h interface from the preceding exercise, write a program that
initializes a complete deck of 52 cards, shuffles it, and then displays the shuffled
values, as shown in the following sample run:
ShuffleDeck
This program initializes, shuffles, and displays
a deck of playing cards.

AH 10C 5D 4H JS AD KH 3C 4C 2D 6C AC JD
2H KS 9H 5S AS 6S 6D 8S KD 2S 7H 8H 5C
8C QH 4S 9S QS 9D 6H 7S 9C 7D 3H JH 10D
KC 10H 8D 2C 7C QD JC 5H QC 4D 10S 3D 3S

One of the easiest ways to shuffle the contents of an array is to adopt the strategy
represented by the following pseudocode:
for (each position p1 in the array) {
Pick a random position p2 between p1 and the end of the array.
Exchange the values at positions p 1 and p2 .
}
Chapter 4
Using Abstract Data Types

Nothing remained in whose reality she could believe, save

those abstract ideas.
— Virginia Woolf, Night and Day, 1919
Using Abstract Data Types – 124 –

As you know from your programming experience, data structures can be assembled to
form hierarchies. The atomic data types—such as int, char, double, and enumerated
types—occupy the lowest level in the hierarchy. To represent more complex
information, you combine the atomic types to form larger structures. These larger
structures can then be assembled into even larger ones in an open-ended process.
Collectively, these assemblages of information into more complex types are called data
structures.
As you learn more about programming, however, you will discover that particular data
structures are so useful that they are worth studying in their own right. Moreover, it is
usually far more important to know how those structures behave than it is to understand
their underlying representation. For example, even though a string might be represented
inside the machine as an array of characters, it also has an abstract behavior that
transcends its representation. A type defined in terms of its behavior rather than its
representation is called an abstract data type, which is often abbreviated to ADT.
Abstract data types are central to the object-oriented style of programming, which
encourages thinking about data structures in a holistic way.
In this chapter, you will have a chance to learn about seven classes—Vector, Grid,
Stack , Queue , Map , Lexicon , and Scanner —each of which represents an important
abstract data type. For the moment, you will not need to understand how to implement
those classes. In subsequent chapters, you’ll have a chance to explore how each of these
classes can be implementated and to learn about the algorithms and data structures
necessary to make those implementations efficient.
Being able to separate the behavior of a class from its underlying implementation is a
fundamental precept of object-oriented programming. As a design strategy, it offers the
following advantages:
• Simplicity. Hiding the internal representation from the client means that there are
fewer details for the client to understand.
• Flexibility. Because a class is defined in terms of its public behavior, the programmer
who implements one is free to change its underlying private representation. As with
any abstraction, it is appropriate to change the implementation as long as the interface
remains the same.
• Security. The interface boundary acts as a wall that protects the implementation and
the client from each other. If a client program has access to the representation, it can
change the values in the underlying data structure in unexpected ways. Making the
data private in a class prevents the client from making such changes.
The ADT classes used in this book are inspired by and draw much of their structure
from a more advanced set of classes available for C++ called the Standard Template
Library, or STL for short. Although the STL is extremely powerful and provides some
capabilities beyond the somewhat simplified class library covered in this book, it is also
more difficult to understand from both the client and implementation perspectives. One
of the primary advantages of using the simplified class library is that you can easily
understand the entire implementation by the time you finish this book. Understanding the
implementation gives you greater insight into how classes work in general and what
libraries like the Standard Template Library are doing for you behind the scenes.
Experience has shown, however, that you will be able to understand the implementation
of a class more easily if you have first had a chance to use with that class as a client.
Using Abstract Data Types – 125 –

4.1 The Vector class

One of the most valuable classes in the Standard Template Library and the simplified
version of it used in this book is the Vector class, which represents a generalization of
the array concept presented in section 2.4. To use the Vector class, you must include its
interface, just as you would for any of the libraries in Chapter 3. The interfaces for each
of the ADT classes introduced in this chapter is simply the name of the class spelled with
a lowercase initial letter and followed with the extension .h at the end. Every program
that wants to use the Vector class must therefore include the line
#include "vector.h"

Arrays are a fundamental type in almost all programming languages and have been
part of programming language designs since the beginnings of the field. Arrays,
however, have a number of weaknesses that can make using them difficult, such as the
following:
• Arrays are allocated with a particular size that doesn’t change after the array is
allocated.
• Even though arrays have a fixed size, C++ does not in fact make that size available to
the programmer. In most applications, you need to keep track of the effective size of
the array, as discussed in Chapter 2.
• It is impossible to insert new elements into an array or to delete elements without
writing a fair amount of code to shift the existing elements to new index positions.
• Many languages, including both C and C++, make no effort to ensure that the elements
you select are actually present in the array. For example, if you create an array with 25
elements and then try to select the value at index position 50, C++ will not ordinarily
detect this as an error. Instead, the program will blithely go on and look at the memory
addresses at which element 50 would appear if the array were long enough. It would
be far better if arrays in C++ (as they do in Java) implemented bounds checking,
which means that every array access checks to see whether the index is valid.
The Vector class solves each of these problems by reimplementing the array concept
in the form of an abstract data type. You can use the Vector class in place of arrays in
any application, usually with surprisingly few changes in the source code and only a
minor reduction in efficiency. In fact, once you have the Vector class, it’s unlikely that
you will have much occasion to use arrays at all, unless you actually have to implement
classes like Vector, which, not surprisingly, uses arrays in its underlying structure. As a
client of the Vector class, however, you are not interested in that underlying structure
and can leave the array mechanics to the programmers who implement the abstract data
type.
As a client of the Vector class, you are concerned with a different set of issues and
need to answer the following questions:
1. How is it possible to specify the type of object contained in a Vector?
2. How does one create an object that is an instance of the Vector class?
3. What methods are available in the Vector class to implement its abstract behavior?
The next three sections explore the answers to each of these questions in turn.
Using Abstract Data Types – 126 –

Specifying the base type of a Vector

Most of the classes covered in this chapter contain other objects as part of a unified
collection. Such classes are called either container classes or collection classes. In
C++, container classes must specify the type of object they contain by including the type
name in angle brackets following the class name. For example, the class Vector<int>
represents a vector whose elements are integers, Vector<char> specifies a vector whose
elements are single characters, while Vector<string> specifies one in which the
elements are strings. The type enclosed within the angle brackets is called the base type
for the collection and is analogous to the element type of a conventional array.
Classes that include a base-type specification are called parameterized classes in the
object-oriented community. In C++, parameterized classes are more often called
templates, which reflects the fact that C++ compilers treat Vector<int>, Vector<char>,
and Vector<string> as independent classes that share a common structure. The name
Vector acts as a template for stamping out a whole family of classes, in which the only
difference is what type of value can appear as an element of the vector. For now, all you
need to understand is how to use templates; the process of implementing basic templates
is described in Chapter 9.

Declaring a new Vector object

One of the philosophical principles behind abstract data types is that clients should be
able to think of them as if they were built-in primitive types. Thus, just as you would
declare an integer variable by writing a declaration such as
int n;

it ought to be possible to declare a new vector by writing

Vector<int> vec;

In C++, that is precisely what you do. That declaration introduces a new variable named
vec, which is—as the template marker in angle brackets indicates—a vector of integers.

As it happens, however, there is more going on in that declaration than meets the eye.
Unlike declarations of a primitive type, declarations of a new class instance automatically
initialize the object by invoking a special method called a constructor. The constructor
for the Vector class initializes the underlying data structures so that they represent a
vector with no elements, which is called an empty vector, to which you can later add any
elements you need. As a client, however, you have no idea what those underlying data
structures are. From your point of view, the constructor simply creates the Vector object
and leave it ready for you to use.

Operations on the Vector class

Every abstract data type includes a suite of methods that define its behavior The methods
exported by the Vector class appears in Table 4-1. As you can see, the Vector class
includes methods that correspond directly to standard array operations (selecting an
individual element and determining the length) along with new methods that extend the
traditional array behavior (adding, inserting, and removing, elements).
Given that every Vector you create starts out with no elements, one of the first things
you need to learn is how to add new elements to a Vector object. The usual approach is
to invoke the add method, which adds a new element at the end of the Vector . For
example, if vec is an empty array of integers as declared in the preceding section,
executing the code
Using Abstract Data Types – 127 –

Table 4-1 Methods exported by the Vector class

size() Returns the number of elements in the vector.
isEmpty() Returns true if the vector is empty.
getAt(index) Returns the element of the vector that appears at the specified
index position. As a convenience, the Vector class also
makes it possible to select an element using array notation, so
that vec[i] selects the element of vec at index position i.
setAt(index, value) Sets the element at the specified index to the new value.
Attempting to reset an element outside the bounds of the
vector generates an error.
add(value) Adds a new element at the end of the vector.
insertAt(index, value) Inserts the new value before the specified index position.
removeAt(index) Deletes the element at the specified index position.
clear() Removes all elements from the vector.
iterator() Returns an iterator that cycles through the elements of the
vector in turn. Iterators are discussed in section 4.8.

vec.add(10);
vec.add(20);
vec.add(30);

would create a three-element vector containing the values 10, 20, and 30. Conceptually,
you could diagram the resulting structure just as if it were an array:
vec
10 20 30
0 1 2

The major difference between the vector and the array is that you can add additional
elements to the vector. For example, if you subsequently called
vec.add(40);

the vector would expand to make room for the new element, like this:
vec
10 20 30 40
0 1 2 3

The insertAt method allows you to add new elements in the middle of a vector. The
first argument to insertAt is an index number, and the new element is inserted before
that position. For example, if you call
vec.insertAt(2, 25);

the value 25 is inserted before index position 2, as follows:

vec
10 20 25 30 40
0 1 2 3 4
Using Abstract Data Types – 128 –

Internally, the implementation of the Vector class has to take care of moving the values
30 and 40 over one position to make room for the 25. From your perspective as a client,
all of that is handled magically by the class.
The Vector class also lets you remove elements. For example, calling
vec.removeAt(0);

removes the element from position 0, leaving the following values:

vec
20 25 30 40
0 1 2 3

Once again, the implementation takes care of shifting elements to close the hole left by
the deleted value.
Now that you have a way of getting elements into a vector, it would be useful to know
how to examine them once they are there. The counterpart to array selection in the
Vector class is the getAt method, which takes an index number and returns the value in
that index position. For example, given the most recent of vec, calling vec.getAt(2)
would return the value 30. If you were to call vec.getAt(5), the bounds-checking code
in the Vector implementation would signal an error because no such element exists.
Symmetrically, you can change the value of an element by calling the setAt method.
Calling
vec.setAt(3, 35);

would change the value in index position 3 to 35, like this:

vec
20 25 30 35
0 1 2 3

Even though the getAt and setAt methods are relatively simple to use, hardly anyone
in fact calls these methods directly. One of the characteristics of C++ that sets it apart
from most other languages is that classes can override the definition of the standard
operators In particular, C++ allows classes to override the selection operator used to
select elements in an array. This feature makes it possible for the Vector class to support
exactly the same selection syntax as arrays. To select the element at position i, all you
need to write is
vec[i];

To change an element, all you need to do is assign to the selected element. Thus, you can
set element 3 in vec to 35 by writing
vec[3] = 35;

The resulting syntax is marginally shorter but considerably more evocative of the array
operations that the Vector class tries to emulate.

Iterating through the elements of a Vector

As with conventional arrays, one of the most common programming patterns used with
vectors is iteration, in which you cycle through each of the elements in turn. The basic
idiom for doing so is
Using Abstract Data Types – 129 –

for (int i = 0; i < vec.size(); i++) {

loop body
}

Inside the loop body, you can refer to the current element as vec[i].
As an example, the following code writes out the contents of the vector vec as a
comma-separated list enclosed in square brackets:
cout << "[";
for (int i = 0; i < vec.size(); i++) {
if (i > 0) cout << ", ";
cout << vec[i];
}
cout << "]" << endl;

If you were to execute this code given the most recent contents of vec, you would see the
following output on the screen:
PrintVector
[20, 25, 30, 35]

Passing a Vector as a parameter

The code at the end of the preceding section is so useful (particularly when you’re
debugging and need to see what values the vector contains), that it would be worth
defining a function that does it. At one level, encapsulating those lines as a single
function is easy; all you have to do is add the appropriate function header, like this:
void PrintVector(Vector<int> & vec) {
cout << "[";
for (int i = 0; i < vec.size(); i++) {
if (i > 0) cout << ", ";
cout << vec[i];
}
cout << "]" << endl;
}

The header line, however, involves one subtlety that you absolutely have to understand
before you can use the library classes effectively. As described in Chapter 1, the & before
the parameter name indicates that the argument to PrintVector is passed by reference,
which means that the vector in the caller is shared with the vector in the function.
Passing by reference is more efficient than C++’s default model of passing by value,
which specifies that the entire contents of the argument vector must be copied before
passing it along to the function. More importantly, passing by reference makes it
possible for you to write functions that change the contents of a vector. As an example,
the following function adds the elements of an integer array to a vector:
void AddArrayToVector(Vector<int> & vec, int array[], int n) {
for (int i = 0; i < n; i++) {
vec.add(array[i]);
}
}
Using Abstract Data Types – 130 –

If you had left out the ampersand in this header line, the function COMMON PITFALLS
would have no effect at all. The code would happily add the first n
elements from array to a vector, but that vector would be a copy of When you are using the
classes in the template
the one the caller supplied. As soon as AddArrayToVector returned, library, you should always
that copy would go away, leaving the original value unchanged. This pass them by reference.
kind of error is easy to make, and you should learn to look for it when The C++ compiler won’t
your programs go awry. notice if you don’t, but the
results are unlikely to be
The revfile.cpp program in Figure 4-1 shows a complete C++ what you intend.
program that uses V e c t o r to reverse the lines in a file. The
AskUserForInputFile and ReadTextFile in this example will

Figure 4-1 Program to print the lines of a file in reverse order

/*
* File: revfile.cpp
* -----------------
* This program reads in a text file and then displays the lines of
* the file in reverse order.
*/

#include "genlib.h"
#include "simpio.h"
#include "vector.h"
#include <string>
#include <iostream>
#include <fstream>

/* Function prototypes */
void ReadTextFile(ifstream & infile, Vector<string> & lines);
void AskUserForInputFile(string prompt, ifstream & infile);
void PrintReversed(Vector<string> & lines);

/* Main program */
int main() {
ifstream infile;
AskUserForInputFile("Input file: ", infile);
Vector<string> lines;
ReadTextFile(infile, lines);
infile.close();
PrintReversed(lines);
return 0;
}

/*
* Reads an entire file into the Vector<string> supplied by the user.
*/
void ReadTextFile(ifstream & infile, Vector<string> & lines) {
while (true) {
string line;
getline(infile, line);
if (infile.fail()) break;
lines.add(line);
}
}
Using Abstract Data Types – 131 –

/*
* Opens a text file whose name is entered by the user. If the file
* does not exist, the user is given additional chances to enter a
* valid file. The prompt string is used to tell the user what kind
* of file is required.
*/
void AskUserForInputFile(string prompt, ifstream & infile) {
while (true) {
cout << prompt;
string filename = GetLine();
infile.open(filename.c_str());
if (!infile.fail()) break;
cout << "Unable to open " << filename << endl;
infile.clear();
}
}
/*
* Prints the lines from the Vector<string> in reverse order.
*/
void PrintReversed(Vector<string> & lines) {
for (int i = lines.size() - 1; i >= 0; i--) {
cout << lines[i] << endl;
}
}

probably come in handy in a variety of applications, and you might want to keep a copy
of this program around so that you can cut-and-paste these functions into your own code.

4.2 The Grid class

Just as the Vector class simulates—and improves upon—a single-dimensional array, the
Grid class is designed to provide a better implementation of two-dimensional arrays in
C++. As with Vector, the Grid class contains individual elements, which means that you
need to use parameterized type templates to specify the base type. Thus, if you want to
create a two-dimensional grid that contains characters, the appropriate class name would
be Grid<char>.
The discipline for creating a new Grid object is slightly different from that of Vector,
mostly because clients tend to use the Grid class in a different way. Particularly when
you are reading in elements of a vector from a file, as was true in the revfile.cpp
program in Figure 4-1, it seems natural to start with an empty vector and then add new
elements as you go along. For the most part, that’s not the way people use two-
dimensional arrays. In most cases, you know the size of the two-dimensional array that
you want to create. It therefore makes sense to specify the dimensions of a Grid object
when you create it.
In the Grid class, the most commonly used constructor takes two arguments, which
specify the number of rows and the number of columns. When you declare a Grid
variable, you include these arguments in parentheses after the variable, as in a method
call. For example, if you want to declare a grid with three rows and two columns in
which each value is a double, you could do so by issuing the following declaration:
Grid<double> matrix(3, 2);
Using Abstract Data Types – 132 –

Even though the elements of the matrix are created by the constructor, they may not be
initialized in any helpful way. If the elements of the grid are objects, they will be
initialized by calling the default constructor for that class, which is simply the
constructor that takes no arguments. If the elements, however, are of a primitive type like
double, C++ does not initialize them, and their values depend on whatever happened be
in the memory locations to which those variables were assigned. It is therefore a good
programming practice to initialize the elements of a Grid explicitly before you use them.
To initialize the elements of a grid, you need to know what methods are available to
manipulate the values the grid contains. The most common methods in the Grid class are
shown in Table 4-2. As you can see from the table, the Grid class does include getAt
and setAt methods that allow you to work with individual elements, but it is far more
common to use the more familiar bracket-selection syntax. For example, if you want to
set every element in the matrix grid to 0.0, you could do so with the following code:
for (int i = 0; i < matrix.numRows(); i++) {
for (int j = 0; j < matrix.numCols(); j++) {
matrix[i][j] = 0.0;
}
}

Because so many games—including chess, checkers, and go—are played on two-

dimensional boards, the Grid class is often useful in applications that play those games.
One particularly simple grid-based game is tic-tac-toe, which is played on a board with
three rows and three columns, as follows:

Players take turns placing the letters X and O in the empty squares, trying to line up three
identical symbols horizontally, vertically, or diagonally.

Table 4-2 Methods exported by the Grid class

numRows() These methods return the number of rows and the number of
numCols() columns, respectively.
getAt(row, col) Returns the element of the grid that appears at the specified
row and column. As a convenience, the Grid class also makes
it possible to select an element using array notation, so that
grid[row][col] selects the element at the specified location.
setAt(row, col, value) Sets the element at the specified index to the new value.
Attempting to reset an element outside the bounds of the
vector generates an error.
resize(rows, cols) Changes the dimensions of the grid as specified by the rows
and cols parameters. Any previous contents of the grid are
discarded.
iterator() Returns an iterator that makes it easy to cycle through the
elements of the grid.
Using Abstract Data Types – 133 –

If you want to represent a tic-tac-toe board using the classes provided in this chapter,
the obvious approach is to use a Grid with three rows and three columns. Given that
each of the elements contains a character—an X, an O, or a space—the declaration of a
board will presumably look like this:
Grid<char> board(3, 3);

You will have a chance to see an program that plays tic-tac-toe in Chapter 7, but for now,
it is sufficient to look at how you might manipulate a tic-tac-toe board declared in this
way. Figure 4-2, for example, contains the code for checking to see whether a player has
won the game by looking to see whether the same character appears in every cell of a
row, a column, or a diagonal.

4.3 The Stack class

One of the simplest classes in the ADT library is the Stack class, which is used to model
a data structure called a stack, which turns out to be particularly useful in a variety of
programming applications. A stack provides storage for a collection of data values,
subject to the restriction that values must be removed from a stack in the opposite order
from which they were added, so that the last item added to a stack is always the first item
that gets removed.

Figure 4-2 Program to check whether a player has won a tic-tac-toe game

/*
* Checks to see whether the specified player identified by mark
* ('X' or 'O') has won the game. To reduce the number of special
* cases, this implementation uses the helper function CheckLine.
*/
bool CheckForWin(Grid<char> & board, char mark) {
for (int i = 0; i < 3; i++) {
if (CheckLine(board, mark, i, 0, 0, 1)) return true;
if (CheckLine(board, mark, 0, i, 1, 0)) return true;
}
if (CheckLine(board, mark, 0, 0, 1, 1)) return true;
return CheckLine(board, mark, 2, 0, -1, 1);
}

/*
* Checks a line extending across the board in some direction. The
* starting coordinates are given by the row and col parameters.
* The direction of motion is specified by dRow and dCol, which
* show how to adjust the row and col values on each cycle. For
* rows, dRow is always 0; for columns, dCol is 0. For diagonals,
* these values will be +1 or -1 depending on the direction.
*/
bool CheckLine(Grid<char> & board, char mark, int row, int col,
int dRow, int dCol) {
for (int i = 0; i < 3; i++) {
if (board[row][col] != mark) return false;
row += dRow;
col += dCol;
}
return true;
}
Using Abstract Data Types – 134 –

Because of their importance in computer science, stacks have acquired a terminology

of their own. The values stored in a stack are called its elements. Adding a new element
to a stack is called pushing that element; removing the most recent item from a stack is
called popping the stack. Moreover, the order in which stacks are processed is
sometimes called LIFO, which stands for “last in, first out.”
The conventional (and possibly apocryphal) explanation for the words stack, push, and
pop is that they come from the following metaphor. In many cafeterias, plates for the
food are placed in spring-loaded columns that make it easy for people in the cafeteria line
to take the top plate, as illustrated in the following diagram:

When a dishwasher adds a new plate, it goes on the top of the stack, pushing the others
down slightly as the spring is compressed, as shown:

Customers can only take plates from the top of the stack. When they do, the remaining
plates pop back up. The last plate added to the stack is the first one a customer takes.
The primary reason that stacks are important in programming is that nested function
calls behave in a stack-oriented fashion. For example, if the main program calls a
function named F, a stack frame for F gets pushed on top of the stack frame for main.
main
F

Local variables forF

If F calls G, a new stack frame for G is pushed on top of the frame for F.
main
F
G

Local variables forG

Using Abstract Data Types – 135 –

When G returns, its frame is popped off the stack, restoring F to the top of the stack as
shown in the original diagram.

The structure of the Stack class

Like Vector and Grid, Stack is a collection class that requires you to specify the base
type. For example, Stack<int> represents a stack whose elements are integers, and
Stack<string> represents one in which the elements are strings. Similarly, if you had
previously defined the classes Plate and Frame to contain all the information required to
represent a dinner plate and the stack frame for a function, you could represent the stacks
described in the preceding two sections with the parameterized classes Stack<Plate>
and Stack<Frame>. The default constructor for the Stack class creates an empty stack.
The complete list of methods exported by the Stack class appears in Table 4-3.

Stacks and pocket calculators

One interesting applications of stacks is in electronic calculators, where they are used to
store intermediate results of a calculation. Although stacks play a central role in the
operation of most calculators, that role is easiest to see in scientific calculators that
require users to enter expressions in a form called reverse Polish notation, or RPN.
In reverse Polish notation, operators are entered after the operands to which they
apply. For example, to compute the result of the expression
50.0 * 1.5 + 3.8 / 2.0

on an RPN calculator, you would enter the operations in the following order:

50.0 ENTER 1.5 x 3.8 ENTER 2.0 / +

When the ENTER button is pressed, the calculator takes the previous value and pushes it
on a stack. When an operator button is pressed, the calculator first checks whether the
user has just entered a value and, if so, automatically pushes it on the stack. It then
computes the result of applying the operator by
• Popping the top two values from the stack
• Applying the arithmetic operation indicated by the button to these values
• Pushing the result back on the stack
Except when the user is actually typing in a number, the calculator display shows the
value at the top of the stack. Thus, at each point in the operation, the calculator display

Table 4-3 Methods exported by the Stack class

size() Returns the number of elements currently on the stack.
isEmpty() Returns true if the stack is empty.
push(value) Pushes value on the stack so that it becomes the topmost element.
pop() Pops the topmost value from the stack and returns it to the caller.
Calling pop on an empty stack generates an error.
peek() Returns the topmost value on the stack without removing it. As with
pop, calling peek on an empty stack generates an error.
clear() Removes all the elements from a stack.
Using Abstract Data Types – 136 –

and stack contain the values shown:

Buttons 50.0 ENTER 1.5 x 3.8 ENTER 2.0 / +

Display 50.0 50.0 1.5 75.0 3.8 3.8 2.0 1.9 76.9

3.8 3.8 1.9

Stack 50.0 50.0 75.0 75.0 75.0 75.0 75.0 76.9

To implement the RPN calculator described in the preceding section in C++ requires
making some changes in the user-interface design. In a real calculator, the digits and
operations appear on a keypad. In this implementation, it is easier to imagine that the
user enters lines on the console, where those lines take one of the following forms:
• A floating-point number
• An arithmetic operator chosen from the set +, -, *, and /
• The letter Q, which causes the program to quit
• The letter H, which prints a help message
• The letter C, which clears any values left on the stack
A sample run of the calculator program might therefore look like this:
RPNCalc
RPN Calculator Simulation (type H for help)
> 50.0
> 1.5
> *
75
> 3.8
> 2.0
> /
1.9
> +
76.9
> Q

Because the user enters each number on a separate line terminated with the RETURN key,
there is no need for any counterpart to the calculator’s ENTER button, which really serves
only to indicate that a number is complete. The calculator program can simply push the
numbers on the stack as the user enters them. When the calculator reads an operator, it
pops the top two elements from the stack, applies the operator, displays the result, and
then pushes the result back on the stack.
The complete implementation of the calculator application appears in Figure 4-3.

4.4 The Queue class

As you learned in section 4.3, the defining feature of a stack is that the last item pushed is
always the first item popped. As noted in the introduction to that section, this behavior is
often referred to in computer science as LIFO, which is an acronym for the phrase “last
in, first out.” The LIFO discipline is useful in programming contexts because it reflects
the operation of function calls; the most recently called function is the first to return.
Using Abstract Data Types – 137 –

Figure 4-3 Implementation of a calculator that uses reverse Polish notation

/*
* File: rpncalc.cpp
* -----------------
* This program simulates an electronic calculator that uses
* reverse Polish notation, in which the operators come after
* the operands to which they apply.
*/

#include <iostream>
#include <cctype>
#include "genlib.h"
#include "simpio.h"
#include "strutils.h"
#include "stack.h"

/* Private function prototypes */

void ApplyOperator(char op, Stack<double> &operandStack);

void HelpCommand();
void ClearStack(Stack<double> &operandStack);

/* Main program */

int main() {
Stack<double> operandStack;

cout << "RPN Calculator Simulation (type H for help)" << endl;
while (true) {
cout << "> ";
string line = GetLine();
char ch = toupper(line[0]);
if (ch == 'Q') {
break;
} else if (ch == 'C') {
ClearStack(operandStack);
} else if (ch == 'H') {
HelpCommand();
} else if (isdigit(ch)) {
operandStack.push(StringToReal(line));
} else {
ApplyOperator(ch, operandStack);
}
}
return 0;
}
Using Abstract Data Types – 138 –

/*
* Function: ApplyOperator
* Usage: ApplyOperator(op, operandStack);
* ---------------------------------------
* This function applies the operator to the top two elements on
* the operand stack. Because the elements on the stack are
* popped in reverse order, the right operand is popped before
* the left operand.
*/

void ApplyOperator(char op, Stack<double> &operandStack) {

double result;

double rhs = operandStack.pop();

double lhs = operandStack.pop();
switch (op) {
case '+': result = lhs + rhs; break;
case '-': result = lhs - rhs; break;
case '*': result = lhs * rhs; break;
case '/': result = lhs / rhs; break;
default: Error("Illegal operator");
}
cout << result << endl;
operandStack.push(result);
}

/*
* Function: HelpCommand
* Usage: HelpCommand();
* ---------------------
* This function generates a help message for the user.
*/

void HelpCommand() {
cout << "Enter expressions in Reverse Polish Notation," << endl;
cout << "in which operators follow the operands to which" << endl;
cout << "they apply. Each line consists of a number, an" << endl;
cout << "operator, or one of the following commands:" << endl;
cout << " Q -- Quit the program" << endl;
cout << " H -- Display this help message" << endl;
cout << " C -- Clear the calculator stack" << endl;
}

/*
* Function: ClearStack
* Usage: ClearStack(stack);
* -------------------------
* This function clears the stack by popping elements until empty.
*/

void ClearStack(Stack<double> &stack) {

while (!stack.isEmpty()) {
stack.pop();
}
}
Using Abstract Data Types – 139 –

In real-world situations, however, its usefulness is more limited. In human society, our
collective notion of fairness assigns some priority to being first, as expressed in the
maxim “first come, first served.” In programming, the usual phrasing of this ordering
strategy is “first in, first out,” which is traditionally abbreviated as FIFO.
A data structure that stores items using a FIFO discipline is called a queue. The
fundamental operations on a queue—which are analogous to the push and pop operations
for stacks—are called enqueue and dequeue . The enqueue operation adds a new
element to the end of the queue, which is traditionally called its tail. The dequeue
operation removes the element at the beginning of the queue, which is called its head.
The conceptual difference between these structures can be illustrated most easily with
a diagram. In a stack, the client must add and remove elements from the same end of the
internal data structure, as follows:
Push

Stack:

base of top of
stack stack Pop

In a queue, the client adds elements at one end and removes them from the other, like this:
Enqueue

Queue:

head of tail of
D e q u e u e queue queue

As you might expect from the fact that the conceptual are so similar, the structure of
the Queue class looks very much like its Stack counterpart. The list of methods in the
Queue class shown in Table 4-4 bears out that supposition. The only differences are in
the terminology, which reflects the difference in the ordering of the elements.

The queue data structure has many applications in programming. Not surprisingly,
queues turn up in many situations in which it is important to maintain a first-in/first-out
discipline in order to ensure that service requests are treated fairly. For example, if you
are working in an environment in which a single printer is shared among several

Table 4-4 Methods exported by the Queue class

size() Returns the number of elements currently on the queue.
isEmpty() Returns true if the queue is empty.
enqueue(value) Adds value to the tail of the queue.
dequeue() Removes the element at the head of the queue and returns it to the
caller. Calling dequeue on an empty queue generates an error.
peek() Returns the value at the head of the queue without removing it. As
with dequeue , calling peek on an empty queue generates an error.
clear() Removes all the elements from a queue.
Using Abstract Data Types – 140 –

computers, the printing software is usually designed so that all print requests are entered
in a queue. Thus, if several users decide to enter print requests, the queue structure
ensures that each user’s request is processed in the order received.
Queues are also common in programs that simulate the behavior of waiting lines. For
example, if you wanted to decide how many cashiers you needed in a supermarket, it
might be worth writing a program that could simulate the behavior of customers in the
store. Such a program would almost certainly involve queues, because a checkout line
operates in a first-in/first-out way. Customers who have completed their purchases arrive
in the checkout line and wait for their turn to pay. Each customer eventually reaches the
front of the line, at which point the cashier totals up the purchases and collects the
money. Because simulations of this sort represent an important class of application
programs, it is worth spending a little time understanding how such simulations work.
Simulations and models
Beyond the world of programming, there are an endless variety of real-world events and
processes that—although they are undeniably important—are nonetheless too
complicated to understand completely. For example, it would be very useful to know
how various pollutants affect the ozone layer and how the resulting changes in the ozone
layer affect the global climate. Similarly, if economists and political leaders had a more
complete understanding of exactly how the national economy works, it would be possible
to evaluate whether a cut in the capital-gains tax would spur investment or whether it
would exacerbate the existing disparities of wealth and income.
When faced with such large-scale problems, it is usually necessary to come up with an
idealized model, which is a simplified representation of some real-world process. Most
problems are far too complex to allow for a complete understanding. There are just too
many details. The reason to build a model is that, despite the complexity of a particular
problem, it is often possible to make certain assumptions that allow you to simplify a
complicated process without affecting its fundamental character. If you can come up
with a reasonable model for a process, you can often translate the dynamics of the model
into a program that captures the behavior of that model. Such a program is called a
simulation.
It is important to remember that creating a simulation is usually a two-step process.
The first step consists of designing a conceptual model for the real-world behavior you
are trying to simulate. The second consists of writing a program that implements the
conceptual model. Because errors can occur in both steps of the process, maintaining a
certain skepticism about simulations and their applicability to the real world is probably
wise. In a society conditioned to believe the “answers” delivered by computers, it is
critical to recognize that the simulations can never be better than the models on which
they are based.
The waiting-line model
Suppose that you want to design a simulation that models the behavior of a supermarket
waiting line. By simulating the waiting line, you can determine some useful properties of
waiting lines that might help a company make such decisions as how many cashiers are
needed, how much space needs to be reserved for the line itself, and so forth.
The first step in the process of writing a checkout-line simulation is to develop a model
for the waiting line, detailing the simplifying assumptions. For example, to make the
initial implementation of the simulation as simple as possible, you might begin by
assuming that there is one cashier who serves customers from a single queue. You might
then assume that customers arrive with a random probability and enter the queue at the
Using Abstract Data Types – 141 –

end of the line. Whenever the cashier is free and someone is waiting in line, the cashier
begins to serve that customer. After an appropriate service period—which you must also
model in some way—the cashier completes the transaction with the current customer, and
is free to serve the next customer in the queue.
Discrete time
Another assumption often required in a model is some limitation on the level of accuracy.
Consider, for example, the time that a customer spends being served by the cashier. One
customer might spend two minutes; another might spend six. It is important, however, to
consider whether measuring time in minutes allows the simulation to be sufficiently
precise. If you had a sufficiently accurate stopwatch, you might discover that a customer
actually spent 3.14159265 minutes. The question you need to resolve is how accurate
you need to be.
For most models, and particularly for those intended for simulation, it is useful to
introduce the simplifying assumption that all events within the model happen in discrete
integral time units. Using discrete time assumes that you can find a time unit that—for
the purpose of the model—you can treat as indivisible. In general, the time units used in
a simulation must be small enough that the probability of more than one event occurring
during a single time unit is negligible. In the checkout-line simulation, for example,
minutes may not be accurate enough; two customers could easily arrive in the same
minute. On the other hand, you could probably get away with using seconds as the time
unit and discount the possibility that two customers arrive in precisely the same second.
Although the checkout-line example assumes that simulation time is measured in
seconds, in general, there is no reason you have to measure time in conventional units.
When you write a simulation, you can define the unit of time in any way that fits the
structure of the model. For example, you could define a time unit to be five seconds and
then run the simulation as a series of five-second intervals.
Events in simulated time
The real advantage of using discrete time units is not that it makes it possible to work
with variables of type int instead of being forced to use type double . The most
important property of discrete time is that it allows you to structure the simulation as a
loop in which each time unit represents a single cycle. When you approach the problem
in this way, a simulation program has the following form:
for (int time = 0; time < SIMULATION_TIME; time++) {
Execute one cycle of the simulation.
}

Within the body of the loop, the program performs the operations necessary to advance
through one unit of simulated time.
Think for a moment about what events might occur during each time unit of the
checkout-line simulation. One possibility is that a new customer might arrive. Another
is that the cashier might finish with the current customer and go on the serve the next
person in line. These events bring up some interesting issues. To complete the model,
you need to say something about how often customers arrive and how much time they
spend at the cash register. You could (and probably should) gather approximate data by
watching a real checkout line in a store. Even if you collect that information, however,
you will need to simplify it to a form that (1) captures enough of the real-world behavior
to be useful and (2) is easy to understand in terms of the model. For example, your
surveys might show that customers arrive at the line on average once every 20 seconds.
This average arrival rate is certainly useful input to the model. On the other hand, you
Using Abstract Data Types – 142 –

would not have much confidence in a simulation in which customers arrived exactly once
every 20 seconds. Such an implementation would violate the real-world condition that
customer arrivals have some random variability and that they sometimes bunch together.
For this reason, the arrival process is usually modeled by specifying the probability
that an arrival takes place in any discrete time unit instead of the average time between
arrivals. For example, if your studies indicated that a customer arrived once every 20
seconds, the average probability of a customer arriving in any particular second would be
1/20 or 0.05. If you assume that arrivals occur randomly with an equal probability in
each unit of time, the arrival process forms a pattern that mathematicians call a Poisson
distribution.
You might also choose to make simplifying assumptions about how long it takes to
serve a particular customer. For example, the program is easier to write if you assume
that the service time required for each customer is uniformly distributed within a certain
range. If you do, you can use the RandomInteger function from the random.h interface
to pick the service time.

Implementing the simulation

Even though it is longer than the other programs in this chapter, the code for the
simulation program is reasonably easy to write and appears in Figure 4-4. The core of the
simulation is a loop that runs for the number of seconds indicated by the parameter
SIMULATION_TIME. In each second, the simulation performs the following operations:

1. Determine whether a new customer has arrived and, if so, add that person to the
queue.
2. If the cashier is busy, note that the cashier has spent another second with the current
customer. Eventually, the required service time will be complete, which will free the
cashier.
3. If the cashier is free, serve the next customer in the waiting line.
The waiting line itself is represented, naturally enough, as a queue. The value stored in
the queue is the time at which that customer arrived in the queue, which makes it possible
to determine how many seconds that customer spent in line before reaching the head of
the queue.
The simulation is controlled by the following parameters:
• SIMULATION_TIME—This parameter specifies the duration of the simulation.
• ARRIVAL_PROBABILITY—This parameter indicates the probability that a new customer
will arrive at the checkout line during a single unit of time. In keeping with standard
statistical convention, the probability is expressed as a real number between 0 and 1.
• MIN_SERVICE_TIME, MAX_SERVICE_TIME—These parameters define the legal range of
customer service time. For any particular customer, the amount of time spent at the
cashier is determined by picking a random integer in this range.
When the simulation is complete, the program reports the simulation parameters along
with the following results:
• The number of customers served
• The average amount of time customers spent in the waiting line
• The average length of the waiting line
Using Abstract Data Types – 143 –

Figure 4-4 Program to simulate a checkout line

/*
* File: checkout.cpp
* ------------------
* This program simulates a checkout line, such as one you
* might encounter in a grocery store. Customers arrive at
* the checkout stand and get in line. Those customers wait
* in the line until the cashier is free, at which point
* they are served and occupy the cashier for a randomly
* chosed period of time. After the service time is complete,
* the cashier is free to serve the next customer in the line.
*
* The waiting line is represented by a Queue<int> in which the
* integer value stored in the queue is the time unit in which
* that customer arrived. Storing this time makes it possible
* to determine the average waiting time for each customer.
*
* In each unit of time, up to the parameter SIMULATION_TIME,
* the following operations are performed:
*
* 1. Determine whether a new customer has arrived.
* New customers arrive randomly, with a probability
* determined by the parameter ARRIVAL_PROBABILITY.
*
* 2. If the cashier is busy, note that the cashier has
* spent another minute with that customer. Eventually,
* the customer's time request is satisfied, which frees
* the cashier.
*
* 3. If the cashier is free, serve the next customer in line.
* The service time is taken to be a random period between
* MIN_SERVICE_TIME and MAX_SERVICE_TIME.
*
* At the end of the simulation, the program displays the
* parameters and the following computed results:
*
* o The number of customers served
* o The average time spent in line
* o The average number of customers in the line
*/

#include "genlib.h"
#include "random.h"
#include "queue.h"
#include <iostream>
#include <iomanip>

/* Simulation parameters */

const int SIMULATION_TIME = 2000;

const double ARRIVAL_PROBABILITY = 0.10;
const int MIN_SERVICE_TIME = 5;
const int MAX_SERVICE_TIME = 15;
Using Abstract Data Types – 144 –

/* Private function prototypes */

void RunSimulation();
void ReportResults(int nServed, long totalWait, long totalLength);

/* Main program */

int main() {
Randomize();
RunSimulation();
return 0;
}

/*
* Function: RunSimulation
* Usage: RunSimulation();
* -----------------------
* This function runs the actual simulation. In each time unit,
* the program first checks to see whether a new customer arrives.
* Then, if the cashier is busy (indicated by a nonzero value for
* serviceTimeRemaining), the program decrements that variable to
* indicate that one more time unit has passed. Finally, if the
* cashier is free, the simulation serves another customer from
* the queue after recording the waiting time for that customer.
*/

void RunSimulation() {
Queue<int> queue;
int serviceTimeRemaining = 0;
int nServed = 0;
long totalWait = 0;
long totalLength = 0;
for (int t = 0; t < SIMULATION_TIME; t++) {
if (RandomChance(ARRIVAL_PROBABILITY)) {
queue.enqueue(t);
}
if (serviceTimeRemaining > 0) {
serviceTimeRemaining--;
if (serviceTimeRemaining == 0) nServed++;
} else if (!queue.isEmpty()) {
totalWait += t - queue.dequeue();
serviceTimeRemaining =
RandomInteger(MIN_SERVICE_TIME, MAX_SERVICE_TIME);
}
totalLength += queue.size();
}
ReportResults(nServed, totalWait, totalLength);
}
Using Abstract Data Types – 145 –

/*
* Function: ReportResults
* Usage: ReportResults(nServed, totalWait, totalLength);
* ------------------------------------------------------
* This function reports the results of the simulation.
*/

void ReportResults(int nServed, long totalWait, long totalLength) {

cout << "Simulation results given the following parameters:"
<< endl;
cout << fixed << setprecision(2);
cout << " SIMULATION_TIME: " << setw(4)
<< SIMULATION_TIME << endl;
cout << " ARRIVAL_PROBABILITY: " << setw(7)
<< ARRIVAL_PROBABILITY << endl;
cout << " MIN_SERVICE_TIME: " << setw(4)
<< MIN_SERVICE_TIME << endl;
cout << " MAX_SERVICE_TIME: " << setw(4)
<< MAX_SERVICE_TIME << endl;
cout << endl;
cout << "Customers served: " << setw(4) << nServed << endl;
cout << "Average waiting time: " << setw(7)
<< double(totalWait) / nServed << endl;
cout << "Average queue length: " << setw(7)
<< double(totalLength) / SIMULATION_TIME << endl;
}

For example, the following sample run shows the results of the simulation for the
indicated parameter values:
CheckoutLine
Simulation results given the following parameters:
SIMULATION_TIME: 2000
ARRIVAL_PROBABILITY: 0.05
MIN_SERVICE_TIME: 5
MAX_SERVICE_TIME: 15

Customers served: 93
Average waiting time: 4.97
Average queue length: 0.23

The behavior of the simulation depends significantly on the values of its parameters.
Suppose, for example, that the probability of a customer arriving increases from 0.05 to
0.10. Running the simulation with these parameters gives the following results:
CheckoutLine
Simulation results given the following parameters:
SIMULATION_TIME: 2000
ARRIVAL_PROBABILITY: 0.10
MIN_SERVICE_TIME: 5
MAX_SERVICE_TIME: 15

Customers served: 179

Average waiting time: 109.80
Average queue length: 11.61
Using Abstract Data Types – 146 –

As you can see, doubling the probability of arrival causes the average waiting time to
grow from approximately 5 seconds to over a minute and a half, which is obviously a
dramatic increase. The reason for the poor performance is that the arrival rate in the
second run of the simulation means that new customers arrive at the same rate at which
they are served. When this arrival level is reached, the length of the queue and the
average waiting time begin to grow very quickly. Simulations of this sort make it
possible to experiment with different parameter values. Those experiments, in turn, make
it possible to identify potential sources of trouble in the corresponding real-world
systems.

4.5 The Map class

This section introduces another generic collection called a map, which is conceptually
similar to a dictionary. A dictionary allows you to look up a word to find its meaning. A
map is a generalization of this idea that provides an association between an identifying
tag called a key and an associated value, which may be a much larger and more
complicated structure. In the dictionary example, the key is the word you’re looking up,
and the value is its definition.
Maps have many applications in programming. For example, an interpreter for a
programming language needs to be able to assign values to variables, which can then be
referenced by name. A map makes it easy to maintain the association between the name
of a variable and its corresponding value. When they are used in this context, maps are
often called symbol tables, which is just another name for the same concept.

The structure of the Map class

As with the collection classes introduced earlier in this chapter, Map is implemented as a
template class than must be parameterized with its value type. For example, if you want
to simulate a dictionary in which individual words are associated with their definitions,
you can start by declaring a dictionary variable as follows:
Map<string> dictionary;

Similarly, if you use a Map to store values of floating-point variables in a programming

language, you could start with the following definition:
Map<double> symbolTable;

These definitions create empty maps that contain no keys and values. In either case, you
would subsequently need to add key/value pairs to the map. In the case of the dictionary,
you could read the contents from a data file. For the symbol table, you would add new
associations whenever an assignment statement appeared.
It is important to note that the parameter for the Map class specifies the type of the
value, and not the type of the key. In many implementations of collection classes—
including, for example, the one in the Standard Template Library and its counterpart in
the Java collection classes—you can specify the type of the key as well. The Map class
used in this book avoid considerable complexity by insisting that all keys be strings.
Strings are certainly the most common type for keys, and it is usually possible to convert
other types to strings if you want to use them as map keys. For example, if you want to
use integers as keys, you can simply call the IntegerToString function on the integer
version of the key and then use the resulting string for all map operations.
The most common methods used with the Map class appear in Table 4-5. Of these, the
ones that implement the fundamental behavior of the map concept are put and get. The
Using Abstract Data Types – 147 –

Table 4-5 Methods exported by the Map class

size() Returns the number of key/value pairs contained in the map.
isEmpty() Returns true if the map is empty.
put(key, value) Associates the specified key and value in the map. If key has no
previous definition, a new entry is added; if a previous association
exists, the old value is discarded and replace by the new one.
get(key) Returns the value currently associated with key in the map. If
there is no such value, get generates an error.
remove(key) Removes key from the map along with any associated value. If key
does not exist, this call leaves the map unchanged.
containsKey(key) Checks to see whether key is associated with a value. If so, this
method returns true; if not, it returns false.
clear() Removes all the key/value pairs from the map.
iterator() Returns an iterator that makes it easy to cycle through the keys in
the map.

put method creates an association between a key and a value. Its operation is analogous
to assigning a value to a variable in C++: if there is a value already associated with the
key, the old value is replaced by the new one. The get method retrieves the value most
recently associated with a particular key and therefore corresponds to the act of using a
variable name to retrieve its value. If no value appears in the map for a particular key,
calling get with that key generates an error condition. You can check for that condition
by calling the containsKey method, which returns true or false depending on whether
the key exists in the map.
A few simple diagrams may help to illustrate the operation of the Map class in more
detail. Suppose that you have declared the symbolTable variable to be a Map<double>
as you saw earlier in the section. That declaration creates an empty map with no
associations, which can be represented as the following empty box:
symbolTable

Once you have the map, you can use put to establish new associations. For example, if
you were to call
symbolTable.put("pi", 3.14159);

the conceptual effect would be to add an association inside the box between the key "pi"
and the value 3.14159, as follows:
symbolTable
pi = 3.14159

If you then called

symbolTable.put("e", 2.71828);
Using Abstract Data Types – 148 –

a new association would be added between the key "e" and the value 2.71828, like this:
symbolTable
pi = 3.14159
e = 2.71828

You can then use get to retrieve these values. Calling symbolTable.get("pi") would
return the value 3.14159, and calling symbolTable.get("pi") would return 2.71828.
Although it hardly makes sense in the case of mathematical constants, you could
change the values in the map by making additional calls to put. You could, for example,
reset the value associated with "pi" (as an 1897 bill before the Indiana State General
Assembly sought to do) by calling
symbolTable.put("pi", 3.2);

which would leave the map in the following state:

symbolTable
pi = 3.2
e = 2.71828

At this point, calling symbolTable.containsKey("pi") would return true; by contrast,

calling symbolTable.containsKey("x") would return false.

Using maps in an application

If you fly at all frequently, you quickly learn that every airport in the world has a three-
letter code assigned by the International Air Transport Association (IATA). For example,
John F. Kennedy airport in New York City is assigned the three-letter code JFK. Other
codes, however, are considerably harder to recognize. Most web-based travel systems
offer some means of looking up these codes as a service to their customers.
Suppose that you have been asked to write a simple C++ program that reads a three-
letter airport code from the user and responds with the location of that airport. The data
you need is in the form of a text file called AirportCodes.txt, which contains a list of
the several thousand airport codes that IATA has assigned. Each line of the file consists
of a three-letter code, an equal sign, and the location of the airport. If the file were sorted
in descending order by passenger traffic in 2009, as compiled by Airports Council
International, the file would begin with the lines in Figure 4-5.
Given the Map class, the code for this application fits on a single page, as shown in
Figure 4-6. That program makes it possible for the user to see the following sample run:
AirportCodes
Airport code: LHR
LHR is in London, England, United Kingdom
Airport code: SFO
SFO is in San Francisco, CA, USA
Airport code: XXX
There is no such airport code
Airport code:
Using Abstract Data Types – 149 –

Figure 4-5. The first page of the AirportCodes.txt data file

ATL=Atlanta, GA, USA
ORD=Chicago, IL, USA
LHR=London, England, United Kingdom
HND=Tokyo, Japan
LAX=Los Angeles, CA, USA
CDG=Paris, France
DFW=Dallas/Ft Worth, TX, USA
FRA=Frankfurt, Germany
PEK=Beijing, China
MAD=Madrid, Spain
DEN=Denver, CO, USA
JFK=New York, NY, USA
AMS=Amsterdam, Netherlands
LAS=Las Vegas, NV, USA
HKG=Hong Kong, Hong Kong
IAH=Houston, TX, USA
PHX=Phoenix, AZ, USA
BKK=Bangkok, Thailand
SIN=Singapore, Singapore
EWR=Newark, NJ, USA
MCO=Orlando, FL, USA
DTW=Detroit, MI, USA
SFO=San Francisco, CA, USA
NRT=Tokyo, Japan
LGW=London, England, United Kingdom
MSP=Minneapolis, MN, USA
DXB=Dubai, United Arab Emirates
MUC=Munich, Germany
MIA=Miami, FL, USA
CLT=Charlotte, NC, USA
FCO=Rome, Italy
BCN=Barcelona, Spain
SYD=Sydney, New South Wales, Australia
PHL=Philadelphia, PA, USA
CGK=Jakarta, Indonesia
YYZ=Toronto, Ontario, Canada
ICN=Incheon [Seoul], South Korea
SEA=Seattle, WA, USA
CAN=Guangzhou, China
PVG=Shanghai, China
BOS=Boston, MA, USA
KUL=Kuala Lumpur, Malaysia
ORY=Paris, France
MEX=Mexico City, Distrito Federal, Mexico
IST=Istanbul, Turkey
BOM=Bombay (Mumbai), India
LGA=New York, NY, USA
IAD=Washington, DC, USA
MXP=Milan, Italy
STN=London, England, United Kingdom
TPE=Taipei, Taiwan
DEL=Delhi, India
DUB=Dublin, Ireland
PMI=Palma Mallorca, Mallorca Island, Spain
.
.
.
Using Abstract Data Types – 150 –

Figure 4-6 Program to look up three-letter airport codes

/*
* File: airports.cpp
* ------------------
* This program looks up a three-letter airport code in a Map object.
*/

#include "genlib.h"
#include "simpio.h"
#include "strutils.h"
#include "map.h"
#include <iostream>
#include <fstream>
#include <string>

/* Private function prototypes */

void ReadCodeFile(Map<string> & map);

/* Main program */

int main() {
Map<string> airportCodes;
ReadCodeFile(airportCodes);
while (true) {
cout << "Airport code: ";
string code = ConvertToUpperCase(GetLine());
if (code == "") break;
if (airportCodes.containsKey(code)) {
cout << code << " is in " << airportCodes.get(code) << endl;
} else {
cout << "There is no such airport code" << endl;
}
}
return 0;
}

/* Reads the data file into the map */

void ReadCodeFile(Map<string> & map) {

ifstream infile;
infile.open("AirportCodes.txt");
if (infile.fail()) Error("Can't read the data file");
while (true) {
string line;
getline(infile, line);
if (infile.fail()) break;
if (line.length() < 4 || line[3] != '=') {
Error("Illegal data file line: " + line);
}
string code = ConvertToUpperCase(line.substr(0, 3));
map.put(code, line.substr(4));
}
infile.close();
}
Using Abstract Data Types – 151 –

Maps as associative arrays

The Map class overloads the square bracket operators used for array selection so that the
statement
map[key] = value;

acts as a shorthand for

map.put(key, value);

and the expression map[key] returns the value from map associated with key in exactly
the same way that map.get(key) does. While these shorthand forms of the put and get
methods are certainly convenient, using array notation for maps is initially somewhat
surprising, given that maps and arrays seem to be rather different in their structure. If
you think about maps and arrays in the right way, however, they turn out to be more alike
than you might at first suspect.
The insight necessary to unify these two seemingly different structures is that you can
think of arrays as structures that map index positions to elements. For example, the array
scores
9.2 9.9 9.7 9.0 9.5
0 1 2 3 4

used as an example in Chapter 2 maps the key 0 into the value 9.2, the key 1 into 9.9, the
key 2 into 9.7, and so forth. Thus, an array is in some sense just a map with integer keys.
Conversely, you can think of a map as an array that uses strings as index positions, which
is precisely what the overloaded selection syntax for the Map class suggests.
Using array syntax to perform map-like operations is becoming increasingly common
in programming languages beyond the C++ domain. Many popular scripting languages
implement all arrays internally as maps, making it possible use index values that are not
necessarily integers. Arrays implemented using maps as their underlying representation
are called associative arrays.

4.6 The Lexicon class

A lexicon is conceptually a dictionary from which the definitions have been removed.
Given a lexicon, you can only tell whether a word exists; there are no definitions or
values associated with the individual words.

The structure of the Lexicon class

The Lexicon class offers two different forms of the constructor. The default constructor
creates an empty lexicon to which you can add new words. In many applications,
however, it is convenient to provide the constructor with the name of a data file that
contains the words you want to include. For example, if the file EnglishWords.dat
contains a list of all English words, you could use that file to create a lexicon using the
followng declaration:
Lexicon english("EnglishWords.dat");

The implementation of the Lexicon class allows these data files to be in either of two
formats:
1. A text file in which the words appear in any order, one word per line.
Using Abstract Data Types – 152 –

2. A precompiled data file that mirrors the internal representation of the lexicon. Using
precompiled files (such as EnglishWords.dat) is more efficient, both in terms of
space and time.
Unlike the classes presented earlier in this chapter, Lexicon does not require a type
parameter, because a lexicon doesn’t contain any values. It does, of course, contain a set
of words, but the words are always strings.
The methods available in the Lexicon class appear in Table 4-6. The most commonly
used method is containsWord, which checks to see if a word is in the lexicon. Assuming
that you have initialized the variable english so that it contains a lexicon of all English
words, you could see if a particular word exists by writing a test such as the following:
if (english.containsWord(word)) . . .

And because it is useful to make such tests in a variety of applications, you can also
determine whether any English words begin with a particular substring by calling
if (english.containsPrefix(prefix)) . . .

A simple application of the Lexicon class

In many word games, such as the popular Scrabble™ crossword game, it is critical to
memorize as many two letter words as you can, because knowing the two-letter words
makes it easier to attach new words to the existing words on the board. Given that you
have a lexicon containing English words, you could create such a list by generating all
two-letter strings and then using the lexicon to check which of the resulting combinations
are actually words. The code to do so appears in Figure 4-7.
As you will discover in section 4.8, it is also possible to solve this problem by going
through the lexicon and printing out the words whose length is two. Given that there are
more than 100,000 English words in the lexicon and only 676 (26 x 26) combinations of
two letters, the strategy used in Figure 4-7 is probably more efficient.
Table 4-6 Methods exported by the Lexicon class
size() Returns the number of words in the lexicon.
isEmpty() Returns true if the lexicon is empty.
add(word) Adds a new word to the lexicon. If the word is already in
the lexicon, this call has no effect; each word may appear
only once. All words in a lexicon are stored in lower case.
addWordsFromFile(name) Adds all the words in the named file to the lexicon. The
file must either be a text file, in which case the words are
listed on separate lines, or a precompiled data file, in
which the contents of the file match the internal structure
of the lexicon. The addWordsFromFile method can read a
precompiled file only if the lexicon is empty.
containsWord(word) Returns true if word is in the lexicon.
containsPrefix(prefix) Returns true if any of the words in the lexicon start with
the specified prefix.
clear() Removes all the elements from a lexicon.
iterator() Returns an iterator that makes it easy to cycle through the
words in the lexicon.
Using Abstract Data Types – 153 –

Figure 4-7 Program to display all two-letter English words

/*
* File: twoletters.cpp
* --------------------
* This program generates a list of the two-letter words.
*/

#include "genlib.h"
#include "lexicon.h"
#include <iostream>

int main() {
Lexicon english("EnglishWords.dat");
string word = "xx";
for (char c0 = 'a'; c0 <= 'z'; c0++) {
word[0] = c0;
for (char c1 = 'a'; c1 <= 'z'; c1++) {
word[1] = c1;
if (english.containsWord(word)) {
cout << word << endl;
}
}
}
return 0;
}

Why are lexicons useful if maps already exist

If you think about it, a lexicon is in many ways just a simplified version of a map in
which you ignore the values altogether. It would therefore be easy enough to build most
of the Lexicon class on top of the Map class. Adding a word to the lexicon corresponds
to calling put using the word as the key; checking whether a word exists corresponds to
calling containsKey.
Given that the Map class already provides most of the functionality of the Lexicon, it
may seem odd that both classes are included in this book. When we designed the ADT
library, we chose to include a separate Lexicon class for the following reasons:
• Not having to worry about the values in a map makes it possible to implement the
lexicon in a more efficient way, particularly in terms of how much memory is required
to store the data. Given the data structure used in the Lexicon implementation, the
entire English dictionary requires approximately 350,000 bytes. If you were to use a
Map to store the words, the storage requirements would be more than five times greater.
• The underlying representation used in the lexicon makes it possible to check not only
whether a word exists in the lexicon, but also to find out whether any word in the
lexicon starts with a particular set of letters (the containsPrefix method).
• The lexicon representation ensures that the words remain in alphabetical order.
Although these characteristics of the Lexicon class are clearly advantages, it is still
somewhat surprising that these reasons focus on what seem to be implementation details,
particularly in light of the emphasis this chapter places on ignoring such details. By
choosing to include the Lexicon class, those details are not being revealed to the client.
After reading the justification for the Lexicon class, you have no idea why it might be
more efficient than the Map class for storing a list of words, but you might well choose to
take advantage of that fact.
Using Abstract Data Types – 154 –

4.7 The Scanner class

The last of the abstract data types introduced in this chapter is the Scanner class, which
provides a useful tool for dividing up a string into meaningful units that are larger than a
single character. After all, when you read text on a page, you don’t ordinarily pay much
attention to the individual letters. Your eye instead groups letters to form words, which it
then recognizes as independent units. The Scanner class does much the same thing: it
divides an input string into its component tokens, which are ordinarily either
• A sequence of consecutive alphanumeric characters (letters or digits), or
• A single-character string consisting of a space or punctuation mark
For example, if the scanner were initialized to extract tokens from the string
This line contains 10 tokens.

successive calls to the scanner package would return those ten individual tokens as shown
by the boxes on the following line:

This line contains 10 tokens .

Using the Scanner class

When you want to use the Scanner class, you typically go through the following steps:
1. Create a new scanner object. As with the abstract data types introduced earlier in this
chapter, your first responsibility when using a scanner is to declare a new scanner
object, like this:
Scanner scanner;
The scanner object, which is called scanner in this example, keeps track of all the
state information it needs to know the order in which tokens should be delivered. The
methods in the Scanner class all operate on a particular scanner object, which means
that you must always specify scanner as a receiver in any calls that you make.
2. Initialize the input to be scanned. Once you have a scanner instance, you can then
initialize your scanner input by calling
scanner.setInput(str);
where str is the string from which the tokens are scanned. Alternatively, if you want
to read tokens from a file, you can call call
scanner.setInput(infile);
where infile is an ifstream object, as described in section 3.4.
3. Read tokens from the scanner. Once you have initialized the scanner input, you can
process each of its tokens individually by calling scanner.nextToken() for each
token in the input. When all the tokens have been consumed, the method
scanner.hasMoreTokens() returns false, which means that the standard idiom for
iterating through each token in turn looks like this:
while (scanner.hasMoreTokens()) {
string token = scanner.nextToken();
Do something with the token you’ve found.
}
Using Abstract Data Types – 155 –

Because it is often easier to check for a sentinel value, the nextToken method returns
the empty string if you call it after the last token has been read.
You can use the same scanner instance many times by calling setInput for each string
you want to split into tokens, so you don’t need to declare a separate scanner for each
string that you have.
Figure 4-8 offers a simple example of how to use the scanner to create a program that
reports all words in a text file that aren’t in the English lexicon.

Figure 4-8 Program to check spelling in a text file

/*
* File: spellcheck.cpp
* --------------------
* This program checks the spelling of words in an input file.
*/

#include "genlib.h"
#include "simpio.h"
#include "lexicon.h"
#include "scanner.h"
#include <string>
#include <cctype>
#include <iostream>
#include <fstream>

/* Function prototypes */
bool IsAllAlpha(string & str);
void AskUserForInputFile(string prompt, ifstream & infile); (see page 131)

/* Main program */
int main() {
ifstream infile;
Lexicon english("EnglishWords.dat");
Scanner scanner;
AskUserForInputFile("Input file: ", infile);
scanner.setInput(infile);
while (scanner.hasMoreTokens()) {
string word = scanner.nextToken();
if (IsAllAlpha(word) && !english.containsWord(word)) {
cout << word << " is not in the dictionary" << endl;
}
}
infile.close();
return 0;
}

/* Returns true if a string contains only alphabetic characters. */

bool IsAllAlpha(string & str) {
for (int i = 0; i < str.length(); i++) {
if (!isalpha(str[i])) return false;
}
return true;
}
Using Abstract Data Types – 156 –

Setting scanner options

As you can see from Table 4-7, most of the methods in the Scanner class allow clients to
redefine what character sequences count as individual tokens. By default, the scanner
recognizes only two classes of tokens: sequences of alphanumeric characters and single
characters that fall outside the alphanumeric set. Depending on the application, clients
may want to change this interpretation. For example, a C++ compiler has to recognize
numbers like 3.14159265 and quoted strings like "hello, world" as single tokens. A
browser must do the same with the tags like <p> (start a new paragraph) and <b> (switch
to a boldface font) that are part of the Hypertext Markup Language (HTML) in which
most web pages are written. In many contexts, including both the compiler and the web
browser, it is useful to ignore whitespace characters—spaces, tabs, and end-of-line
markers—because these characters serve only to separate tokens and have no semantic
value in themselves.
These capabilities are incorporated into the Scanner class as option settings. You
enable a particular option by calling a method with an argument that specified the
behavior you want. To enhance readability, the Scanner class defines a set of
enumerated type constants whose names describe as closely as possible exactly what the
option does. For example, if you want to ignore whitespace characters, you can call
scanner.setSpaceOption(Scanner::IgnoreSpaces);

The names of the constants used to set each option are described in Table 4-7 along with
the method to which those constants apply.
The Scanner class also exports a method called saveToken that comes in handy in a
variety of applications. This method solves the problem that arises from the fact that you
often don’t know that you want to stop reading a sequence of tokens until you’ve read the
token that follows that sequence. Unless your application is prepared to deal with the
new token at that point in the code, it is convenient to put that token back in the scanner
stream where it can be read again when the program is ready to do so.

4.8 Iterators
The twoletter.cpp program introduced in Figure 4-7 earlier in this chapter computes a
list of all two-letter words by generating every possible combination of two letters and
then looking up each one to see whether that two-letter string appears in the lexicon of
English words. Another strategy that accomplishes the same result is to go through every
word in the lexicon and display the words whose length is equal to 2. To do so, all you
need is some way of stepping through each word in a Lexicon object, one at a time.
Stepping through the elements of a collection class is a fundamental operation that
each class must provide through its interface. Moreover, if the package of collection
classes is well designed, clients should be able to use the same strategy to perform that
operation, no matter whether they are cycling through all elements in a vector or a grid,
all keys in a map, or all words in a lexicon. In most modern software packages, including
the library ADTs used in this book and the Standard Template Library on which those
classes are based, the process of cycling through the elements of a collection is provided
through a class called an iterator. Each abstract collection class in the library—with the
exception of Stack and Queue, for which being able to process the elements out of order
would violate the LIFO or FIFO discipline that defines those type—exports its own
Iterator class, but defines that class so that all iterators behave in the same way. Once
you learn how to use an iterator for one class, you can easily transfer that knowledge to
any of the other classes.
Using Abstract Data Types – 157 –

Table 4-7 Methods exported by the Scanner class

setInput(str) Sets the scanner input source so that new tokens are taken
from the string str. Any unread tokens remaining from the
preceding call to setInput are discarded.
setInput(stream) Sets the scanner input source so that new tokens are taken
from the input stream stream . Opening and closing the
input stream are the responsibility of the client. As with
the string version of this method, any unread tokens from
a previous call are discarded.
hasMoreTokens() Returns true if the scanner contains additional tokens.
nextToken() Reads the next token from the input source (either a string
or a file) and returns it as a string. If no more tokens
exist, nextToken returns the empty string.
saveToken(token) Stores the specified token in the private state of the
scanner so that it will be returned the next time
nextToken is called.
setSpaceOption(option) Allows the client to control whether whitespace
getSpaceOption() characters (spaces, tabs, and ends of lines) are returned as
single-character tokens or skipped entirely. The option
parameter must be one of the following constant names:
Scanner::PreserveSpaces
Scanner::IgnoreSpaces
setNumberOption(option) Allows the client to control whether numeric strings are
getNumberOption() recognized as single tokens. The option parameter must be
one of the following constant names:
Scanner::ScanNumbersAsLetters
Scanner::ScanNumbersAsIntegers
Scanner::ScanNumbersAsReals
By default, digits are considered just like letters; the other
options allow the scanner to read an complete number.
setStringOption(option) Allows the client to control whether quoted strings are
getStringOption() recognized as single tokens. The option parameter must be
one of the following constant names:
Scanner::ScanQuotesAsPunctuation
Scanner::ScanQuotesAsStrings
By default, quotation marks act like any other punctuation
mark; if ScanQuotesAsStrings is in effect, a quoted
string is treated as a single token.
setBracketOption(option) Allows the client to control whether HTML tags enclosed
getBracketOption() in angle brackets (such as <p> or <font>) are recognized
as single tokens. The option parameter must be one of the
following constant names:
Scanner::ScanBracketsAsPunctuation
Scanner::ScanBracketsAsTag
By default, angle brackets act like any other punctuation
mark; if ScanBracketsAsTag is in effect, the entire tag
(including the angle brackets) is treated as a single token.
Using Abstract Data Types – 158 –

The standard iterator pattern

The general pattern for using an iterator is illustrated by the following code fragment,
which iterates over every word in the now-familiar lexicon of English words:
Lexicon::Iterator iter = english.iterator();
while (iter.hasNext()) {
string word = iter.next();
code to work with that word
}

The iterator method in the Lexicon class returns an iterator that provides each word in
the lexicon, one at a time. The iterator class itself is defined as a nested subclass within
Lexicon, so its full name is Lexicon::Iterator. The first line in this example therefore
applies the iterator method to the lexicon stored in the variable english and then
stores the resulting iterator object in the variable iter, which has been suitably declared
with the full name of its type.
Once you have the iterator variable, you then enter a loop that continues as long as the
iterator has more elements to process. The hasNext method returns a Boolean value that
indicates whether any additional elements remain, which is exactly what you need for the
condition in the while loop. Inside the loop, the next method returns the next element in
the collection. In this example, calling iter.next() returns the next word from the
English language lexicon, which is then stored in the string variable word.
The code that needs to go into the body of the loop depends, of course, on what you’re
trying to do. If, for example, you want to list all two-letter English words using the
iterator model, the code to do so will look like this:
Lexicon::Iterator iter = english.iterator();
while (iter.hasNext()) {
string word = iter.next();
if (word.length() == 2) {
cout << word << endl;
}
}

The type of value produced by the next method depends on the class in which the iterator
is created. In the Map and Lexicon classes, next always returns a value of type string.
In the Array and Grid classes, next returns a value whose type matches the base type of
that collection. Thus, an iterator for an Array<int> will produce values of type int, and
an iterator for a Grid<char> will produce values of type char.

The foreach idiom

Even though the iterator pattern introduced in the preceding section takes up just a few
lines of code, I have discovered over the years that the details of the iterator mechanism
get in the way of understanding the purpose of the iterator, which is simply to step
through the elements of a collection, one at a time. If you were trying to express the
algorithmic function of an iterator in English pseudcode, you would want to write
something like this:
For each element in a particular collection {
Process that element
}
Using Abstract Data Types – 159 –

Some languages, most notably C# and Java, define a new syntactic form that expresses
precisely that idea. Unfortunately, the syntax of C++ does not such a facility.
The good news, however, is that it is possible to use the macro-definition capabilities
of the C++ preprocessor to achieve exactly what you would like to see in the language.
Although the implementation details are beyond the scope of this text, the collection
classes that support iteration also define a foreach macro that has the following form:
foreach (string word in english) {
if (word.length() == 2) {
cout << word << endl;
}
}

The advantage of the foreach syntax is not that the code is a couple of lines shorter,
but rather that the revised code tells the reader exactly what is going on. If you compare
the two versions of this loop, I’m sure you’ll immediately recognize just how much
clearer the foreach version is. In my experience, once students have been given the
opportunity to use foreach in their code, they never go back to using the iterator form.
Even so, it is useful to know that the iterator form exists and that the iterator mechanism
is in fact what is going on inside the implementation of foreach.

Iteration order
When you work with iterators or the foreach macro, it is often useful to understand the
order in which the iterator generates each of the individual values. There is no universal
rule. Each container class defines its own policy about iteration order, usually based on
considerations of efficiency. The classes you’ve already seen make the following
guarantees about the order of values:
• The iterator for the Vector class generates the elements in the order of the index
position, so that the element in position 0 comes first, followed by the element in
position 1, and so on, up to the end of the vector. The order in which elements are
returned by the iterator is therefore the same as the order in which elements are
processed by the standard for loop pattern for iterating through an array:
for (int i = 0; i < vec.size(); i++) {
code to process vec[i]
}

• The iterator for the Grid class steps through the elements of row 0 in order, then the
elements of row 1, and so forth. This order is iteration strategy for Grid is thus
analogous to using the following for loop pattern:
for (int row = 0; row < grid.numRows(); row++) {
for (int col = 0; col < grid.numCols(); col++) {
code to process grid[row][col]
}
}
This order, in which the row subscript appears in the outer loop, is called row-major
order.
• The iterator for the Map class makes no guarantees about the order in which the keys
are returned. As you will discover in Chapter 12, the most efficient representation for
storing keys in a map is incompatible with, for example, keeping them in alphabetical
order.
Using Abstract Data Types – 160 –

• The iterator for the Lexicon class always returns words in alphabetical order, with all
words converted to lower case. The ability to process words alphabetically is one of
the principal advantages of the Lexicon class.
When you use an iterator, it is important that you do not modify the contents of the
collection object over which the iteration is performed, because such changes may
invalidate the data structures stored within the iterator. If, for example, you are iterating
over the keys in a map, deleting the current key may make it impossible for the iterator to
figure out how to get to the next key. The implementations of the iterators used in this
text check that the structure has not changed as the iterator proceeds through it, but that
may not be the case for iterators that exist in other packages.
An example of the foreach mechanism
In the discussion of Pig Latin in section 3.3, the words used to illustrate the rules for
forming Pig Latin words were alley and trash. These words have the interesting property
that their Pig Latin forms—alleyway and ashtray—happen to be other English words.
Such words are not all that common; in the lexicon stored in the file EnglishWords.dat,
there are only 27 words with that property out of over 100,000 English words. Given
iterators and the PigLatin function from Figure 3-5, it is easy to write a program that
lists them all:
int main() {
cout << "This program finds words that remain words when "
<< "translated to Pig Latin." << endl;
Lexicon english("EnglishWords.dat");
foreach (string word in english) {
string pig = PigLatin(word);
if (pig != word && english.containsWord(pig)) {
cout << word << " -> " << pig << endl;
}
}
return 0;
}

Computing word frequencies

Computers have revolutionized many fields of academic inquiry, including some in
which the use of such modern tools might at first seem surprising. Over the last few
decades, computer analysis has become central to resolving questions of disputed
authorship. For example, there are plays from the Elizabethan era that might have been
written by Shakespeare, even though they are not part of the traditional canon.
Conversely, several Shakespearean plays that are attributed to Shakespeare have parts
that don’t sound like his other works and may have in fact been written by someone else.
To resolve such questions, Shakespearean scholars often compute the frequency of
particular words that appear in the text and see whether those frequencies match what we
expect to find based on an analysis of Shakespeare’s known works.
Suppose, for example, that you have a text file containing a passage from Shakespeare,
such as the following well-known lines from Act 5 of Macbeth:
macbeth.txt
Tomorrow, and tomorrow, and tomorrow
Creeps in this petty pace from day to day

If you are trying to determine the relative frequency of words in Shakespeare’s writing,
you need to have a program that counts how many times each word appears in the data
Using Abstract Data Types – 161 –

file. Thus, given the file macbeth.txt , your would like your program to produce
something like the following output:
WordFrequency
and 2
creeps 1
day 2
from 1
in 1
pace 1
petty 1
this 1
to 1
tomorrow 3

The code for the word frequency program appears in Figure 4-9. Given the tools you
have at your disposal from the earlier sections in this chapter, the code required to
tabulate word frequencies is quite straightforward. The Scanner class is clearly the right
mechanism for going through the words in the file, just as it was for the spelling checker
in Figure 4-8. To keep track of the mapping between words and their associated counts, a

Figure 4-9 Program to keep track of the frequency of words in a text file

/*
* File: wordfreq.cpp
* ------------------
* This program computes the frequency of words in a text file.
*/
#include "genlib.h"
#include "simpio.h"
#include "map.h"
#include "scanner.h"
#include <string>
#include <cctype>
#include <iostream>
#include <fstream>
#include <iomanip>

/* Private function prototypes */

void CreateFrequencyTable(ifstream & infile, Map<int> & wordCounts);
void DisplayWordCounts(Map<int> & wordCounts);
void AskUserForInputFile(string prompt, ifstream & infile); (see page 131)
bool IsAllAlpha(string & str); (see page 157)
/* Main program */
int main() {
ifstream infile;
Map<int> wordCounts;
AskUserForInputFile(infile);
CreateFrequencyTable(infile, wordCounts);
infile.close();
DisplayWordCounts(wordCounts);
return 0;
}
Using Abstract Data Types – 162 –

/*
* Creates a frequency table that reads through the input file
* and counts how often each word appears. The client supplies
* both the input file stream and the map used to keep track of
* the word count.
*/
void CreateFrequencyTable(ifstream & infile, Map<int> & wordCounts) {
Scanner scanner;
scanner.setInput(infile);
scanner.setSpaceOption(Scanner::IgnoreSpaces);
while (scanner.hasMoreTokens()) {
string word = ConvertToLowerCase(scanner.nextToken());
if (IsAllAlpha(word)) {
if (wordCounts.containsKey(word)) {
wordCounts[word]++;
} else {
wordCounts[word] = 1;
}
}
}
}

/*
* Displays the count for each word in the frequency table.
*/
void DisplayWordCounts(Map<int> & wordCounts) {
foreach (string word in wordCounts) {
cout << left << setw(15) << word
<< right << setw(5) << wordCounts[word] << endl;
}
}

Map<int> is precisely what you need. And when you need to go through the entries in
the map to list the word counts, Iterator provides just the right tool.
The only minor problem with this implementation is that the words don’t appear in
alphabetical order as they did in the proposed sample run created during the design phase.
Because the iterator for the Map class is allowed to produce the keys in any order, they
will ordinarily come out in some jumbled fashion. Given the implementation of the Map
class as it exists, the program happens to produce the output
WordFrequency
pace 1
to 1
day 2
tomorrow 3
petty 1
and 2
creeps 1
from 1
in 1
this 1

but any other order would be equally possible.

Using Abstract Data Types – 163 –

It’s not hard to get the output to come out in alphabetical order. In fact, as you will have
a chance to discover in exercise 16, the library classes in this chapter make it possible to
display this list alphabetically with just a few additional lines of code. It might, however,
be even more useful to present the list in descending order of frequency. To do that, it
will be useful to understand the sorting algorithms presented in Chapter 8.

Summary
This chapter introduced seven C++ classes—Vector, Grid, Stack, Queue, Map, Lexicon,
and Scanner—that form a powerful collection of programming tools. For the moment,
you have looked at these classes only as a client. In subsequent chapters, you will have a
chance to learn more about how they are implemented. Important points in this chapter
include:
• Data structures that are defined in terms of their behavior rather their representation are
called abstract data types. Abstract data types have several important advantages over
more primitive data structures such as arrays and records. These advantages include:
1. Simplicity. The representation of the underlying data representation is not
accessible, which means that there are fewer details for the client to understand.
2. Flexibility. The implementer is free to enhance the underlying representation as
long as the methods in the interface continue to behave in the same way.
3. Security. The interface barrier prevents the client from making unexpected
changes in the internal structure.
• Classes that contain other objects as elements of an integral collection are called
container classes or, equivalently, collection classes. In C++, container classes are
usually defined using a template or parameterized type, in which the type name of the
element appears in angle brackets after the name of the container class. For example,
the class Vector<int> signifies a vector containing values of type int.
• The Vector class is an abstract data type that behaves in much the same fashion as a
one-dimensional array but is much more powerful. Unlike arrays, a Vector can grow
dynamically as elements are added and removed. They are also more secure, because
the implementation of Vector checks to make sure that selected elements exist.
• The Grid class provides a convenient abstraction for working with two-dimensional
arrays.
• The Stack class represents a collection of objects whose behavior is defined by the
property that items are removed from a stack in the opposite order from which they
were added: last in, first out (LIFO). The fundamental operations on a stack are push,
which adds a value to the stack, and pop, which removes and returns the value most
recently pushed.
• The Queue class is similar to the Stack class except for the fact that elements are
removed from a queue in the same order in which they were added: first in, first out
(FIFO). The fundamental operations on a queue are enqueue, which adds a value to
the end of a queue, and dequeue, which removes and returns the value from the front.
• The Map class makes it possible to associate keys with values in a way that makes it
possible to retrieve those associations very efficiently. The fundamental operations
on a map are put , which adds a key/value pair, and get , which returns the value
associated with a particular key.
• The Lexicon class represents a word list. The fundamental operations on a map are
add , which stores a new word in the list, and containsWord, which checks to see
whether a word exists in the lexicon.
Using Abstract Data Types – 164 –

• The Scanner class simplifies the problem of breaking up a string or an input file into
tokens that have meaning as a unit. The fundamental operations on an scanner are
hasMoreTokens, which determines whether more tokens can be read from the scanner,
and nextToken, which returns the next token from the input source.
• Most collection classes define an internal class named Iterator that makes it easy to
cycle through the contents of the collection. The fundamental operations on an iterator
are hasNext, which determines whether more elements exist, and next, which returns
the next element from the collection.

Review questions
1. True or false: An abstract data type is one defined in terms of its behavior rather than
its representation.
2. What three advantages does this chapter cite for separating the behavior of a class
from its underlying implementation?
3. What is the STL?
4. If you want to use the Vector class in a program, what #include line do you need to
add to the beginning of your code?
5. List at least three advantages of the Vector class over the more primitive array
mechanism available in C++.
6. What is meant by the term bounds-checking?
7. What type name would you use to store a vector of Boolean values?
8. True or false: The default constructor for the Vector class creates a vector with ten
elements, although you can make it longer later.
9. What method do you call to determine the number of elements in a Vector?
10. If a Vector object has N elements, what is the legal range of values for the first
argument to insertAt? What about for the first argument to removeAt?
11. What feature of the Vector class makes it possible to avoid explicit use of the getAt
and setAt methods?
12. Why is it important to pass vectors and other collection object by reference?
13. What declaration would you use to initialize a variable called chessboard to an 8 x 8
grid, each of whose elements is a character?
14. Given the chessboard variable from the preceding exercise, how would you assign
the character 'R' (which stands for a white rook in standard chess notation) to the
squares in the lower left and lower right corners of the board?
15. What do the acronyms LIFO and FIFO stand for? How do these terms apply to
stacks and queues?
16. What are the names of the two fundamental operations for a stack?
17. What are the names for the corresponding operations for a queue?
Using Abstract Data Types – 165 –

18. What does the peek operation do in each of the Stack and Queue classes?
19. What are the names for the corresponding operations for a queue?
20. Describe in your own words what is meant by the term discrete time in the context of
a simulation program.
21. True or false: In the Map class used in this book, the keys are always strings.
22. True or false: In the Map class used in this book, the values are always strings.
23. What happens if you call get for a key that doesn’t exist in a map?
24. What are the syntactic shorthand forms for get and put that allow you to treat a map
as an associative array?
25. Why do the libraries for this book include a separate Lexicon class even though it is
easy to implement the fundamental operations of a lexicon using the Map class?
26. What are the two kinds of data files supported by the constructor for the Lexicon
class?
27. What is the purpose of the Scanner class?
28. What options are available for controlling the definition of the tokens recognized by
the Scanner class?
29. What is an iterator?
30. What reason is offered for why there is no iterator for the Stack and Queue class?
31. If you use the iterator method explicitly, what is the standard idiom for using an
iterator to cycle through the elements of a collection?
32. What is the foreach version of ths standard iterator idiom?
33. What is the principal advantage of using foreach in preference to the explicit
iterator-based code?
34. True or false: The iterator for the Map class guarantees that individual keys will be
delivered in alphabetical order.
35. True or false: The iterator for the Lexicon class guarantees that individual words
will be delivered in alphabetical order.
36. What would happen if you removed the call to scanner.setSpaceOption in the
implementation of createFrequencyTable in Figure 4-9? Would the program still
work?

Programming exercises
1. In Chapter 2, exercise 8, you were askd to write a function RemoveZeroElements
that eliminated any zero-valued elements from an integer array. That operation is
much easier in the context of a vector, because the Vector class makes it possible to
add and remove elements dynamically. Rewrite RemoveZeroElements so that the
function header looks like this:
Using Abstract Data Types – 166 –

void RemoveZeroElements(Vector<int> & vec);

2. Write a function
bool ReadVector(ifstream & infile, Vector<double> & vec);
that reads lines from the data file specified by infile, each of which consists of a
floating-point number, and adds them to the vector vec . The end of the vector is
indicated by a blank line or the end of the file. The function should return true if it
successfully reads the vector of numbers; if it encounters the end of the data file
before it reads any values, the function should return false.
To illustrate the operation of this function, suppose that you have the data file
SquareAndCubeRoots.txt
1.0000
1.4142
1.7321
2.0000

1.0000
1.2599
1.4422
1.5874
1.7100
1.8171
1.9129
2.0000

and that you have opened infile as an ifstream on that file. In addition, suppose
that you have declares the variable roots as follows:
Vector<double> roots;
The first call to ReadVector(infile, roots) should return true after initializing
roots so that it contains the four elements shown at the beginning of the file. The
second call would also return true and change the value of roots to contain the
eight elements shown at the bottom of the file. Calling ReadVector a third time
would return false.

3. Given that it is possible to insert new elements at any point, it is not difficult to keep
elements in order as you create a Vector. Using ReadTextFile as a starting point,
write a function
void SortTextFile(ifstream & infile, Vector<string> & lines);
that reads the lines from the file into the vector lines, but keeps the elements of the
vector sorted in lexicographic order instead of the order in which they appear in the
file. As you read each line, you need to go through the elements of the vector you
have already read, find out where this line belongs, and then insert it at that position.

4. The code in Figure 4-2 shows how to check the rows, columns, and diagonals of a
tic-tac-toe board using a single helper function. That function, however, is coded in
such a way that it only works for 3 x 3 boards. As a first step toward creating a
program that can play tic-tac-toe on larger grids, reimplement the CheckForWin and
CheckLine functions so that they work for square grids of any size.
Using Abstract Data Types – 167 –

5. A magic square is a two-dimensional grid of Figure 4-10 Dürer etching with a magic square
integers in which the rows, columns, and
diagonals all add up to the same value. One
of the most famous magic squares appears in
the 1514 engraving “Melencolia I” by
Albrecht Dürer shown in Figure 4-10, in
which a 4 x 4 magic square appears in the
upper right, just under the bell. In Dürer’s
square, which can be read more easily as

16 3 2 13
5 10 11 8
9 6 7 12
4 15 14 1
all four rows, all four columns, and both
diagonals add up to 34.
A more familiar example is the following
3 x 3 magic square in which each of the rows,
columns, and diagonals add up to 15, as
shown:
=15

=15

=15

5
=1
8 1 6 =15 8 1 6 8 1 6
3 5 7 =15 3 5 7 3 5 7
4 9 2 =15 4 9 2 4 9 2
=1
5

Implement a function
bool IsMagicSquare(Grid<int> & square);
that tests to see whether the grid contains a magic square. Note that your program
should work for square grids of any size. If you call IsMagicSquare with a grid in
which the number of rows and columns are different, it should simply return false.

6. In the last several years, a new logic puzzle called Sudoku has become quite popular
throughout the world. In Sudoku, you start with a 9 x 9 grid of integers in which
some of the cells have been filled in with digits between 1 and 9. Your job in the
puzzle is to fill in each of the empty spaces with a digit between 1 and 9 so that each
digit appears exactly once in each row, each column, and each of the smaller 3 x 3
squares. Each Sudoku puzzle is carefully constructed so that there is only one
solution. For example, given the puzzle shown on the left of the following diagram,
the unique solution is shown on the right:
Using Abstract Data Types – 168 –

2 4 5 8 3 9 2 4 6 5 8 1 7
4 1 8 2 7 4 1 8 9 3 6 2 5
6 7 3 9 6 8 5 2 7 1 4 3 9
2 3 9 6 2 5 4 1 3 8 7 9 6
9 6 7 1 8 3 9 6 2 7 1 5 4
1 7 5 3 1 7 6 9 5 4 2 8 3
9 6 8 1 9 6 7 5 8 2 3 4 1
2 9 5 6 4 2 3 7 1 9 5 6 8
8 3 6 9 5 1 8 3 4 6 9 7 2

Although you won’t have learned the algorithmic strategies you need to solve
Sudoku puzzles until later in this book, you can easily write a method that checks to
see whether a proposed solution follows the Sudoku rules against duplicating values
in a row, column, or outlined 3 x 3 square. Write a function
bool CheckSudokuSolution(Grid<int> puzzle);
that performs this check and returns true if the puzzle is a valid solution. Your
program should check to make sure that puzzle contains a 9 x 9 grid of integers and
report an error if this is not the case.

7. Write a program that uses a stack to reverse a sequence of integers read in one per
line from the console, as shown in the following sample run:
ReverseList
Enter a list of integers, ending with 0:
> 10
> 20
> 30
> 40
> 0
Those integers in reverse order are:
40
30
20
10

8. Write a C++ program that checks whether the bracketing operators (parentheses,
brackets, and curly braces) in a string are properly matched. As an example of
proper matching, consider the string
{ s = 2 * (a[2] + 3); x = (1 + (2)); }
If you go through the string carefully, you discover that all the bracketing operators
are correctly nested, with each open parenthesis matched by a close parenthesis, each
open bracket matched by a close bracket, and so on. On the other hand, the
following strings are all unbalanced for the reasons indicated:
(([]) The line is missing a close parenthesis.
)( The close parenthesis comes before the open parenthesis.
{(}) The bracketing operators are improperly nested.
Using Abstract Data Types – 169 –

The reason that this exercise fits in this chapter is that one of the simplest strategies
for implementing this program is to store the unmatched operators on a stack.

9. Bob Dylan’s 1963 song “The Times They Are A-Changin’” contains the following
lines, which are themselves paraphrased from Matthew 19:30:
And the first one now
Will later be last
For the times they are a-changin’
In keeping with this revolutionary sentiment, write a function
void ReverseQueue(Queue<string> & queue);
that reverses the elements in the queue. Remember that you have no access to the
internal representation of the queue and will need to come up with an algorithm,
presumably involving other structures, that accomplishes the task.

10. The checkout-line simulation in Figure 4-4 can be extended to investigate important
practical questions about how waiting lines behave. As a first step, rewrite the
simulation so that there are several independent queues, as is usually the case in
supermarkets. A customer arriving at the checkout area finds the shortest checkout
line and enters that queue. Your revised simulation should calculate the same results
as the simulation in the chapter.

11. As a second extension to the checkout-line simulation, change the program from the
previous exercise so that there is a single waiting line served by multiple cashiers—a
practice that has become more common in recent years. In each cycle of the
simulation, any cashier that becomes idle serves the next customer in the queue. If
you compare the data produced by this exercise and the preceding one, what can you
say about the relative advantages of these two strategies?

12. If waiting lines become too long, customers can easily become frustrated and may
decide to take their business elsewhere. Simulations may make it possible to reduce
the risk of losing customers by allowing managers to determine how many cashiers
are required to reduce the average waiting time below a predetermined threshold.
Rewrite the checkout-line simulation from exercise 11 so that the program itself
determines how many cashiers are needed. To do so, your program must run the
complete simulation several times, holding all parameters constant except the
number of cashiers. When it finds a staffing level that reduces the average wait to an
acceptable level, your program should display the number of cashiers used on that
simulation run.

13. Write a program to simulate the following experiment, which was included in the
1957 Disney film, Our Friend the Atom, to illustrate the chain reactions involved in
nuclear fission. The setting for the experiment is a large cubical box, the bottom of
which is completely covered with an array of 625 mousetraps, arranged to form a
square grid 25 mousetraps on a side. Each of the mousetraps is initially loaded with
two ping-pong balls. At the beginning of the simulation, an additional ping-pong
ball is released from the top of the box and falls on one of the mousetraps. That
mousetrap springs and shoots its two ping-pong balls into the air. The ping-pong
balls bounce around the sides of the box and eventually land on the floor, where they
are likely to set off more mousetraps.
In writing this simulation, you should make the following simplifying
assumptions:
Using Abstract Data Types – 170 –

• Every ping-pong ball that falls always lands on a mousetrap, chosen randomly by
selecting a random row and column in the grid. If the trap is loaded, its balls are
released into the air. If the trap has already been sprung, having a ball fall on it
has no effect.
• Once a ball falls on a mousetrap—whether or not the trap is sprung—that ball
stops and takes no further role in the simulation.
• Balls launched from a mousetrap bounce around the room and land again after a
random number of simulation cycles have gone by. That random interval is
chosen independently for each ball and is always between one and four cycles.
Your simulation should run until there are no balls in the air. At that point, your
program should report how many time units have elapsed since the beginning, what
percentage of the traps have been sprung, and the maximum number of balls in the
air at any time in the simulation.

14. In May of 1844, Samuel F. B. Morse sent the message “What hath God wrought!” by
telegraph from Washington to Baltimore, heralding the beginning of the age of
electronic communication. To make it possible to communicate information using
only the presence or absence of a single tone, Morse designed a coding system in
which letters and other symbols are represented as coded sequences of short and long
tones, traditionally called dots and dashes. In Morse code, the 26 letters of the
alphabet are represented by the following codes:
A • J • S •••
B ••• K • T
C • • L • •• U ••
D •• M V •••
E • N • W •
F •• • O X ••
G • P • • Y •
H •••• Q • Z ••
I •• R • •
If you want to convert from letters to Morse code, you can store the strings for each
letter in an array with 26 elements; to convert from Morse code to letters, the easiest
approach is to use a map.
Write a program that reads in lines from the user and translates each line either to
or from Morse code depending on the first character of the line:
• If the line starts with a letter, you want to translate it to Morse code. Any
characters other than the 26 letters should simply be ignored.
• If the line starts with a period (dot) or a hyphen (dash), it should be read as a
series of Morse code characters that you need to translate back to letters. Each
sequence of dots and dashes is separated by spaces, but any other characters
should be ignored. Because there is no encoding for the space between words, the
characters of the translated message will be run together when your program
translates in this direction.
The program should end when the user enters a blank line. A sample run of this
program (taken from the messages between the Titanic and the Carpathia in 1912)
might look like this:
Using Abstract Data Types – 171 –

MorseCode
Morse code translator
> SOS TITANIC
... --- ... - .. - .- -. .. -.-.
> WE ARE SINKING FAST
.-- . .- .-. . ... .. -. -.- .. -. --. ..-. .- ... -
> .... . .- -.. .. -. --. ..-. --- .-. -.-- --- ..-
HEADINGFORYOU
>

15. In Chapter 3, exercise 6, you were asked to write a function IsPalindrome that
checks whether a word is a palindrome, which means that it reads identically
forward and backward. Use that function together with the lexicon of English words
to print out a list of all words in English that are palindromes.

16. As noted in the chapter, it is actually rather easy to change the wordfreq.cpp
program from Figure 4-8 so that the words appear in alphabetical order. The only
thing you need to do is think creatively about the tools that you already have. Make
the necessary modifications to the program to accomplish this change.

17. As noted in section 4.5, a map is often called a symbol table when it is used in the
context of a programming language, because it is precisely the structure you need to
store variables and their values. For example, if you are working in an application in
which you need to assign floating-point values to variable names, you could do so
using a map declared as follows:
Map<double> symbolTable;
Write a C++ program that declares such a symbol table and then reads in
command lines from the user, which must be in one of the following forms:
• A simple assignment statement of the form
var = number
This statement should store the value represented by the token number in the
symbol table under the name var. Thus, if the user were to enter
pi = 3.14159
the string pi should be assigned a value of 3.14159 in symbolTable.
• The name of a variable alone on a line. When your program reads in such a line,
it should print out the current value in the symbol table associated with that name.
Thus, if pi has been defined as shown in the preceding example, the command
pi
should display the value 3.14159.
• The command list, which is interpreted by the program as a request to display
all variable/value pairs currently stored in the symbol table, not necessarily in any
easily discernable order.
• The command quit, which should exit from the program.
Once you have implemented each of these command forms, your program should be
able to produce the following sample run:
Using Abstract Data Types – 172 –

SymbolTableTest
> pi = 3.14159
> e = 2.71828
> x = 2.00
> pi
3.14159
> e
2.71828
> list
e = 2.71828
x = 2
pi = 3.14159
> x = 42
> list
e = 2.71828
x = 42
pi = 3.14159
> a = 1.5
> list
e = 2.71828
x = 42
pi = 3.14159
a = 1.5
> quit

18. Rewrite the RPN calculator from Figure 4-3 so that it uses the Scanner class to read
its input tokens from a single line, as illustrated by the following sample run:
RPNCalc
RPN Calculator Simulation (type H for help)
> 1 2 3 * +
7
> 50.0 1.5 * 3.8 2.0 / +
76.9
> quit
Chapter 5
Introduction to Recursion

And often enough, our faith beforehand in a certain result

is the only thing that makes the result come true.
— William James, The Will To Believe, 1897
Introduction to Recursion – 174 –

Most algorithmic strategies used to solve programming problems have counterparts

outside the domain of computing. When you perform a task repeatedly, you are using
iteration. When you make a decision, you exercise conditional control. Because these
operations are familiar, most people learn to use the control statements for, while, and
if with relatively little trouble.

Before you can solve many sophisticated programming tasks, however, you will have
to learn to use a powerful problem-solving strategy that has few direct counterparts in the
real world. That strategy, called recursion, is defined as any solution technique in which
large problems are solved by reducing them to smaller problems of the same form. The
italicized phrase is crucial to the definition, which otherwise describes the basic strategy
of stepwise refinement. Both strategies involve decomposition. What makes recursion
special is that the subproblems in a recursive solution have the same form as the original
problem.
If you are like most beginning programmers, the idea of breaking a problem down into
subproblems of the same form does not make much sense when you first hear it. Unlike
repetition or conditional testing, recursion is not a concept that comes up in day-to-day
life. Because it is unfamiliar, learning how to use recursion can be difficult. To do so,
you must develop the intuition necessary to make recursion seem as natural as all the
other control structures. For most students of programming, reaching that level of
understanding takes considerable time and practice. Even so, learning to use recursion is
definitely worth the effort. As a problem-solving tool, recursion is so powerful that it at
times seems almost magical. In addition, using recursion often makes it possible to write
complex programs in simple and profoundly elegant ways.

5.1 A simple example of recursion

To gain a better sense of what recursion is, let’s imagine you have been appointed as the
funding coordinator for a large charitable organization that is long on volunteers and
short on cash. Your job is to raise $1,000,000 in contributions so the organization can
meet its expenses.
If you know someone who is willing to write a check for the entire $1,000,000, your
job is easy. On the other hand, you may not be lucky enough to have friends who are
generous millionaires. In that case, you must raise the $1,000,000 in smaller amounts. If
the average contribution to your organization is $100, you might choose a different tack:
call 10,000 friends and ask each of them for $100. But then again, you probably don’t
have 10,000 friends. So what can you do?
As is often the case when you are faced with a task that exceeds your own capacity, the
answer lies in delegating part of the work to others. Your organization has a reasonable
supply of volunteers. If you could find 10 dedicated supporters in different parts of the
country and appoint them as regional coordinators, each of those 10 people could then
take responsibility for raising $100,000.
Raising $100,000 is simpler than raising $1,000,000, but it hardly qualifies as easy.
What should your regional coordinators do? If they adopt the same strategy, they will in
turn delegate parts of the job. If they each recruit 10 fundraising volunteers, those people
will only have to raise $10,000. The delegation process can continue until the volunteers
are able to raise the money on their own; because the average contribution is $100, the
volunteer fundraisers can probably raise $100 from a single donor, which eliminates the
need for further delegation.
Introduction to Recursion – 175 –

If you express this fundraising strategy in pseudocode, it has the following structure:

void CollectContributions(int n) {
if (n <= 100) {
Collect the money from a single donor.
} else {
Find 10 volunteers.
Get each volunteer to collect n/10 dollars.
Combine the money raised by the volunteers.
}
}

The most important thing to notice about this pseudocode translation is that the line

Get each volunteer to collect n/10 dollars.

is simply the original problem reproduced at a smaller scale. The basic character of the
task—raise n dollars—remains exactly the same; the only difference is that n has a
smaller value. Moreover, because the problem is the same, you can solve it by calling the
original function. Thus, the preceding line of pseudocode would eventually be replaced
with the following line:

CollectContributions(n / 10);

It’s important to note that the CollectContributions function ends up calling itself if
the contribution level is greater than $100. In the context of programming, having a
function call itself is the defining characteristic of recursion.
The structure of the CollectContributions procedure is typical of recursive
functions. In general, the body of a recursive function has the following form:

if (test for simple case) {

Compute a simple solution without using recursion.
} else {
Break the problem down into subproblems of the same form.
Solve each of the subproblems by calling this function recursively.
Reassemble the solutions to the subproblems into a solution for the whole.
}

This structure provides a template for writing recursive functions and is therefore called
the recursive paradigm. You can apply this technique to programming problems as
long as they meet the following conditions:

1. You must be able to identify simple cases for which the answer is easily determined.
2. You must be able to identify a recursive decomposition that allows you to break any
complex instance of the problem into simpler problems of the same form.
The CollectContributions example illustrates the power of recursion. As in any
recursive technique, the original problem is solved by breaking it down into smaller
subproblems that differ from the original only in their scale. Here, the original problem is
to raise $1,000,000. At the first level of decomposition, each subproblem is to raise
$100,000. These problems are then subdivided in turn to create smaller problems until
the problems are simple enough to be solved immediately without recourse to further
subdivision. Because the solution depends on dividing hard problems into simpler ones,
recursive solutions are often called divide-and-conquer strategies.
Introduction to Recursion – 176 –

5.2 The factorial function

Although the CollectContributions example illustrates the concept of recursion, it
gives little insight into how recursion is used in practice, mostly because the steps that
make up the solution, such as finding 10 volunteers and collecting money, are not easily
represented in a C++ program. To get a practical sense of the nature of recursion, you
need to consider problems that fit more easily into the programming domain.
For most people, the best way to understand recursion is to start with simple
mathematical functions in which the recursive structure follows directly from the
statement of the problem and is therefore easy to see. Of these, the most common is the
factorial function—traditionally denoted in mathematics as n!—which is defined as the
product of the integers between 1 and n. In C++, the equivalent problem is to write an
implementation of a function with the prototype
int Fact(int n);

that takes an integer n and returns its factorial.

As you probably discovered in an earlier programming course, it is easy to implement
the Fact function using a for loop, as illustrated by the following implementation:
int Fact(int n) {
int product;

product = 1;
for (int i = 1; i <= n; i++) {
product *= i;
}
return product;
}

This implementation uses a for loop to cycle through each of the integers between 1 and
n. In the recursive implementation this loop does not exist. The same effect is generated
instead by the cascading recursive calls.
Implementations that use looping (typically by using for and while statements) are
said to be iterative. Iterative and recursive strategies are often seen as opposites because
they can be used to solve the same problem in rather different ways. These strategies,
however, are not mutually exclusive. Recursive functions sometimes employ iteration
internally, and you will see examples of this technique in Chapter 6.

The recursive formulation of Fact

The iterative implementation of Fact , however, does not take advantage of an
important mathematical property of factorials. Each factorial is related to the factorial of
the next smaller integer in the following way:

n! = n x (n – 1)!

Thus, 4! is 4 x 3!, 3! is 3 x 2!, and so on. To make sure that this process stops at some
point, mathematicians define 0! to be 1. Thus, the conventional mathematical definition
of the factorial function looks like this:
Introduction to Recursion – 177 –

 1 if n = 0

n! = 
 n x (n – 1)! otherwise

This definition is recursive, because it defines the factorial of n in terms of the factorial of
n – 1. The new problem—finding the factorial of n – 1—has the same form as the
original problem, which is the fundamental characteristic of recursion. You can then use
the same process to define (n – 1)! in terms of (n – 2)!. Moreover, you can carry this
process forward step by step until the solution is expressed in terms of 0!, which is equal
to 1 by definition.
From your perspective as a programmer, the practical impact of the mathematical
definition is that it provides a template for a recursive implementation. In C++, you can
implement a function Fact that computes the factorial of its argument as follows:
int Fact(int n) {
if (n == 0) {
return 1;
} else {
return n * Fact(n - 1);
}
}

If n is 0, the result of Fact is 1. If not, the implementation computes the result by calling
Fact(n - 1) and then multiplying the result by n. This implementation follows directly
from the mathematical definition of the factorial function and has precisely the same
recursive structure.

Tracing the recursive process

If you work from the mathematical definition, writing the recursive implementation of
Fact is straightforward. On the other hand, even though the definition is easy to write,
the brevity of the solution may seem suspicious. When you are learning about recursion
for the first time, the recursive implementation of Fact seems to leave something out.
Even though it clearly reflects the mathematical definition, the recursive formulation
makes it hard to identify where the actual computational steps occur. When you call
Fact , for example, you want the computer to give you the answer. In the recursive
implementation, all you see is a formula that transforms one call to Fact into another one.
Because the steps in that calculation are not explicit, it seems somewhat magical when
the computer gets the right answer.
If you follow through the logic the computer uses to evaluate any function call,
however, you discover that no magic is involved. When the computer evaluates a call to
the recursive Fact function, it goes through the same process it uses to evaluate any other
function call. To visualize the process, suppose that you have executed the statement

f = Fact(4);

as part of the function main. When main calls Fact, the computer creates a new stack
frame and copies the argument value into the formal parameter n. The frame for Fact
temporarily supersedes the frame for main, as shown in the following diagram:
Introduction to Recursion – 178 –

main
Fact

→ if (n == 0) {
n return (1);
4 } else {
return (n * Fact(n - 1));
}

In the diagram, the code for the body of Fact is shown inside the frame to make it easier
to keep track of the current position in the program, which is indicated by an arrow. In
the current diagram, the arrow appears at the beginning of the code because all function
calls start at the first statement of the function body.
The computer now begins to evaluate the body of the function, starting with the if
statement. Because n is not equal to 0, control proceeds to the else clause, where the
program must evaluate and return the value of the expression

n * Fact(n - 1)

Evaluating this expression requires computing the value of Fact(n - 1) , which

introduces a recursive call. When that call returns, all the program has to do is to
multiply the result by n . The current state of the computation can therefore be
diagrammed as follows:

main
Fact

if (n == 0) {
n return (1);
4 } else {
return (n * Fact(n - 1));
}
↵

As soon as the call to Fact(n - 1) returns, the result is substituted for the expression
underlined in the diagram, allowing computation to proceed.
The next step in the computation is to evaluate the call to Fact(n - 1), beginning with
the argument expression. Because the current value of n is 4, the argument expression
n - 1 has the value 3. The computer then creates a new frame for Fact in which the
formal parameter is initialized to this value. Thus, the next frame looks like this:

main
Fact
Fact

→ if (n == 0) {
n return (1);
3 } else {
return (n * Fact(n - 1));
}
Introduction to Recursion – 179 –

There are now two frames labeled Fact . In the most recent one, the computer is just
starting to calculate Fact(3). In the preceding frame, which the newly created frame
hides, the Fact function is awaiting the result of the call to Fact(n - 1).
The current computation, however, is the one required to complete the topmost frame.
Once again, n is not 0, so control passes to the else clause of the if statement, where the
computer must evaluate Fact(n - 1) . In this frame, however, n is equal to 3, so the
required result is that computed by calling Fact(2). As before, this process requires the
creation of a new stack frame, as shown:

main
Fact
Fact
Fact

→ if (n == 0) {
n return (1);
2 } else {
return (n * Fact(n - 1));
}

Following the same logic, the program must now call Fact(1), which in turn calls
Fact(0), thereby creating two new stack frames. The resulting stack configuration looks
like this:

main
Fact
Fact
Fact
Fact
Fact

→ if (n == 0) {
n return (1);
0 } else {
return (n * Fact(n - 1));
}

At this point, however, the situation changes. Because the value of n is 0, the function
can return its result immediately by executing the statement

return 1;

The value 1 is returned to the calling frame, which resumes its position on top of the
stack, as shown:
Introduction to Recursion – 180 –

main
Fact
Fact
Fact
Fact

→ if (n == 0) {
n return (1);
1 } else {
return (n * Fact(n - 1));
}

↵
1

From this point, the computation proceeds back through each of the recursive calls,
completing the calculation of the return value at each level. In this frame, for example,
the call to Fact(n - 1) can be replaced by the value 1, so that the result at this level can
be expressed as follows:

return n * 1 );

In this stack frame, n has the value 1, so the result of this call is simply 1. This result gets
propagated back to its caller, which is represented by the top frame in the following
diagram:
main
Fact
Fact
Fact

→ if (n == 0) {
n return (1);
2 } else {
return (n * Fact(n - 1));
}
↵

Because n is now 2, evaluating the return statement causes the value 2 to be passed back
to the previous level, as follows:
main
Fact
Fact

→ if (n == 0) {
n return (1);
3 } else {
return (n * Fact(n - 1));
}
↵

At this stage, the program returns 3 x 2 to the previous level, so that the frame for the
initial call to Fact looks like this:
Introduction to Recursion – 181 –

main
Fact

→ if (n == 0) {
n return (1);
4 } else {
return (n * Fact(n - 1));
}

↵
6

The final step in the calculation process consists of calculating 4 x 6 and returning the
value 24 to the main program.

The recursive leap of faith

The point of the long Fact(4) example in the preceding section is to show you that the
computer treats recursive functions just like all other functions. When you are faced with
a recursive function, you can—at least in theory—mimic the operation of the computer
and figure out what it will do. By drawing all the frames and keeping track of all the
variables, you can duplicate the entire operation and come up with the answer. If you do
so, however, you will usually find that the complexity of the process ends up making the
problem much harder to understand.
When you try to understand a recursive program, you must be able to put the
underlying details aside and focus instead on a single level of the operation. At that level,
you are allowed to assume that any recursive call automatically gets the right answer as
long as the arguments to that call are simpler than the original arguments in some respect.
This psychological strategy—assuming that any simpler recursive call will work
correctly—is called the recursive leap of faith. Learning to apply this strategy is
essential to using recursion in practical applications.
As an example, consider what happens when this implementation is used to compute
Fact(n) with n equal to 4. To do so, the recursive implementation must compute the
value of the expression

n * Fact(n - 1)

By substituting the current value of n into the expression, you know that the result is

4 * Fact(3)

Stop right there. Computing Fact(3) is simpler than computing Fact(4). Because it is
simpler, the recursive leap of faith allows you to assume that it works. Thus, you should
assume that the call to Fact(3) will correctly compute the value of 3!, which is 3 x 2 x 1,
or 6. The result of calling Fact(4) is therefore 4 x 6, or 24.
As you look at the examples in the rest of this chapter, try to focus on the big picture
instead of the morass of detail. Once you have made the recursive decomposition and
identified the simple cases, be satisfied that the computer can handle the rest.

5.3 The Fibonacci function

In a mathematical treatise entitled Liber Abbaci published in 1202, the Italian
mathematician Leonardo Fibonacci proposed a problem that has had a wide influence on
many fields, including computer science. The problem was phrased as an exercise in
Introduction to Recursion – 182 –

population biology—a field that has become increasingly important in recent years.
Fibonacci’s problem concerns how the population of rabbits would grow from generation
to generation if the rabbits reproduced according to the following, admittedly fanciful,
rules:

• Each pair of fertile rabbits produces a new pair of offspring each month.
• Rabbits become fertile in their second month of life.
• Old rabbits never die.

If a pair of newborn rabbits is introduced in January, how many pairs of rabbits are there
at the end of the year?
You can solve Fibonacci’s problem simply by keeping a count of the rabbits at each
month during the year. At the beginning of January, there are no rabbits, since the first
pair is introduced sometime in that month, which leaves one pair of rabbits on February
1. Since the initial pair of rabbits is newborn, they are not yet fertile in February, which
means that the only rabbits on March 1 are the original pair of rabbits. In March,
however, the original pair is now of reproductive age, which means that a new pair of
rabbits is born. The new pair increases the colony’s population—counting by pairs—to
two on April 1. In April, the original pair goes right on reproducing, but the rabbits born
in March are as yet too young. Thus, there are three pairs of rabbits at the beginning of
May. From here on, with more and more rabbits becoming fertile each month, the rabbit
population begins to grow more quickly.

Computing terms in the Fibonacci sequence

At this point, it is useful to record the population data so far as a sequence of terms,
indicated here by the subscripted value ti , each of which shows the number of rabbit pairs
at the beginning of the ith month from the start of the experiment on January 1. The
sequence itself is called the Fibonacci sequence and begins with the following terms,
which represent the results of our calculation so far:

t0 t1 t2 t3 t4
0 1 1 2 3

You can simplify the computation of further terms in this sequence by making an
important observation. Because rabbits in this problem never die, all the rabbits that were
around in the previous month are still around. Moreover, all of the fertile rabbits have
produced a new pair. The number of fertile rabbit pairs capable of reproduction is simply
the number of rabbits that were alive in the month before the previous one. The net effect
is that each new term in the sequence must simply be the sum of the preceding two.
Thus, the next several terms in the Fibonacci sequence look like this:

t0 t1 t2 t3 t4 t5 t6 t7 t8 t9 t10 t11 t12

0 1 1 2 3 5 8 13 21 34 55 89 144

The number of rabbit pairs at the end of the year is therefore 144.
From a programming perspective, it helps to express the rule for generating new terms
in the following, more mathematical form:

tn = tn–1 + tn–2
Introduction to Recursion – 183 –

An expression of this type, in which each element of a sequence is defined in terms of

earlier elements, is called a recurrence relation.

The recurrence relation alone is not sufficient to define the Fibonacci sequence.
Although the formula makes it easy to calculate new terms in the sequence, the process
has to start somewhere. In order to apply the formula, you need to have at least two
terms in hand, which means that the first two terms in the sequence—t0 and t1—must be
defined explicitly. The complete specification of the terms in the Fibonacci sequence is
therefore
 n if n is 0 or 1

tn = 
 tn–1 + tn–2 otherwise

This mathematical formulation is an ideal model for a recursive implementation of a

function Fib(n) that computes the nth term in the Fibonacci sequence. All you need to
do is plug the simple cases and the recurrence relation into the standard recursive
paradigm. The recursive implementation of Fib(n) is shown in Figure 5-1, which also
includes a test program that displays the terms in the Fibonacci sequence between two
specified indices.

Gaining confidence in the recursive implementation

Now that you have a recursive implementation of the function Fib, how can you go about
convincing yourself that it works? You can always begin by tracing through the logic.
Consider, for example, what happens if you call Fib(5). Because this is not one of the
simple cases enumerated in the if statement, the implementation computes the result by
evaluating the line
return Fib(n - 1) + Fib(n - 2);

which is in this case equivalent to

return Fib(4) + Fib(3);

At this point, the computer calculates the result of Fib(4) , adds that to the result of
calling Fib(3), and returns the sum as the value of Fib(5).
But how does the computer go about evaluating Fib(4) and Fib(3)? The answer, of
course, is that it uses precisely the same strategy. The essence of recursion is to break
problems down into simpler ones that can be solved by calls to exactly the same function.
Those calls get broken down into simpler ones, which in turn get broken down into even
simpler ones, until at last the simple cases are reached.
On the other hand, it is best to regard this entire mechanism as irrelevant detail.
Remember the recursive leap of faith. Your job at this level is to understand how the call
to Fib(5) works. In the course of walking though the execution of that function, you
have managed to transform the problem into computing the sum of Fib(4) and Fib(3).
Because the argument values are smaller, each of these calls represents a simpler case.
Applying the recursive leap of faith, you can assume that the program correctly computes
each of these values, without going through all the steps yourself. For the purposes of
validating the recursive strategy, you can just look the answers up in the table. Fib(4) is
3 and Fib(3) is 2, so the result of calling Fib(5) is 3 + 2, or 5, which is indeed the
correct answer. Case closed. You don’t need to see all the details, which are best left to
the computer.
Introduction to Recursion – 184 –

Figure 5-1 Recursive implementation of the Fibonacci function

/*
* File: fib.cpp
* -------------
* This program lists the terms in the Fibonacci sequence with
* indices ranging from MIN_INDEX to MAX_INDEX.
*/

#include "genlib.h"
#include <iostream>

/*
* Constants
* ---------
* MIN_INDEX -- Index of first term to generate
* MAX_INDEX -- Index of last term to generate
*/

const int MIN_INDEX = 0;

const int MAX_INDEX = 12;

/* Private function prototypes */

int Fib(int n);

/* Main program */

int main() {
cout << "This program lists the Fibonacci sequence." << endl;
for (int i = MIN_INDEX; i <= MAX_INDEX; i++) {
cout << "Fib(" << i << ")";
if (i < 10) cout << " ";
cout << " = " << Fib(i) << endl;
}
return 0;
}

/*
* Function: Fib
* Usage: t = Fib(n);
* ------------------
* This function returns the nth term in the Fibonacci sequence
* using a recursive implementation of the recurrence relation
*
* Fib(n) = Fib(n - 1) + Fib(n - 2)
*/

int Fib(int n) {
if (n < 2) {
return n;
} else {
return Fib(n - 1) + Fib(n - 2);
}
}
Introduction to Recursion – 185 –

Efficiency of the recursive implementation

If you do decide to go through the details of the evaluation of the call to Fib(5),
however, you will quickly discover that the calculation is extremely inefficient. The
recursive decomposition makes many redundant calls, in which the computer ends up
calculating the same term in the Fibonacci sequence several times. This situation is
illustrated in Figure 5-2, which shows all the recursive calls required in the calculation of
Fib(5) . As you can see from the diagram, the program ends up making one call to
Fib(4), two calls to Fib(3), three calls to Fib(2), five calls to Fib(1), and three calls to
Fib(0) . Given that the Fibonacci function can be implemented efficiently using
iteration, the enormous explosion of steps required by the recursive implementation is
more than a little disturbing.

Recursion is not to blame

On discovering that the implementation of Fib(n) given in Figure 5-1 is highly
inefficient, many people are tempted to point their finger at recursion as the culprit. The
problem in the Fibonacci example, however, has nothing to do with recursion per se but
rather the way in which recursion is used. By adopting a different strategy, it is possible
to write a recursive implementation of the Fib function in which the large-scale
inefficiencies revealed in Figure 5-2 disappear completely.
As is often the case when using recursion, the key to finding a more efficient solution
lies in adopting a more general approach. The Fibonacci sequence is not the only
sequence whose terms are defined by the recurrence relation
tn = tn–1 + tn–2
Depending on how you choose the first two terms, you can generate many different
sequences. The traditional Fibonacci sequence

Figure 5-2 Steps in the calculation of Fib(5)

Fib(5)

Fib(4) Fib(3)

Fib(3) Fib(2) Fib(2) Fib(1)

Fib(2) Fib(1) Fib(1) Fib(0) Fib(1) Fib(0)

1 1 0 1 0

Fib(1) Fib(0)

1 0
Introduction to Recursion – 186 –

0, 1, 1, 2, 3, 5, 8, 13, 21, 34, 55, 89, 144, . . .

comes from defining t0 = 0 and t1 = 1. If, for example, you defined t0 = 3 and t1 = 7, you
would get this sequence instead:
3, 7, 10, 17, 27, 44, 71, 115, 186, 301, 487, 788, 1275, . . .
Similarly, defining t0 = –1 and t1 = 2 gives rise to the following sequence:
–1, 2, 1, 3, 4, 7, 11, 18, 29, 47, 76, 123, 199, . . .
These sequences all use the same recurrence relation, which specifies that each new term
is the sum of the previous two. The only way the sequences differ is in the choice of the
first two terms. As a general class, the sequences that follow this pattern are called
additive sequences.
This concept of an additive sequence makes it possible to convert the problem of
finding the n th term in the Fibonacci sequence into the more general problem of finding
the n th term in an additive sequence whose initial terms are t0 and t 1 . Such a function
requires three arguments and might be expressed in C++ as a function with the following
prototype:
int AdditiveSequence(int n, int t0, int t1);

If you had such a function, it would be easy to implement Fib using it. All you would
need to do is supply the correct values of the first two terms, as follows:
int Fib(int n) {
return AdditiveSequence(n, 0, 1);
}

The body consists of a single line of code that does nothing but call another function,
passing along a few extra arguments. Functions of this sort, which simply return the
result of another function, often after transforming the arguments in some way, are called
wrapper functions. Wrapper functions are extremely common in recursive
programming. In most cases, a wrapper function is used—as it is here—to supply
additional arguments to a subsidiary function that solves a more general problem.
From here, the only remaining task is to implement the function AdditiveSequence.
If you think about this more general problem for a few minutes, you will discover that
additive sequences have an interesting recursive character of their own. The simple case
for the recursion consists of the terms t0 and t 1 , whose values are part of the definition of
the sequence. In the C++ implementation, the value of these terms are passed as
arguments. If you need to compute t0, for example, all you have to do is return the
argument t0.
But what if you are asked to find a term further down in the sequence? Suppose, for
example, that you want to find t6 in the additive sequence whose initial terms are 3 and 7.
By looking at the list of terms in the sequence
t0 t1 t2 t3 t4 t5 t6 t7 t8 t9
3 7 10 17 27 44 71 115 186 301 ...
you can see that the correct value is 71. The interesting question, however, is how you
can use recursion to determine this result.
Introduction to Recursion – 187 –

The key insight you need to discover is that the nth term in any additive sequence is
simply the n–1st term in the additive sequence which begins one step further along. For
example, t6 in the sequence shown in the most recent example is simply t5 in the additive
sequence
t0 t1 t2 t3 t4 t5 t6 t7 t8
7 10 17 27 44 71 115 186 301 ...
that begins with 7 and 10.
This discovery makes it possible to implement the function AdditiveSequence as
follows:
int AdditiveSequence(int n, int t0, int t1) {
if (n == 0) return t0;
if (n == 1) return t1;
return AdditiveSequence(n - 1, t1, t0 + t1);
}

If you trace through the steps in the calculation of Fib(5) using this technique, you
will discover that the calculation involves none of the redundant computation that
plagued the earlier recursive formulation. The steps lead directly to the solution, as
shown in the following diagram:
Fib(5)
= AdditiveSequence(5, 0, 1)
= AdditiveSequence(4, 1, 1)
= AdditiveSequence(3, 1, 2)
= AdditiveSequence(2, 2, 3)
= AdditiveSequence(1, 3, 5)
= 5

Even though the new implementation is entirely recursive, it is comparable in efficiency

to the standard iterative version of the Fibonacci function.

5.4 Other examples of recursion

Although the factorial and Fibonacci functions provide excellent examples of how
recursive functions work, they are both mathematical in nature and may therefore convey
the incorrect impression that recursion is applicable only to mathematical functions. In
fact, you can apply recursion to any problem that can be decomposed into simpler
problems of the same form. It is useful to consider a few additional examples, including
several that are far less mathematical in their character.

Detecting palindromes
A palindrome is a string that reads identically backward and forward, such as "level"
or "noon". Although it is easy to check whether a string is a palindrome by iterating
through its characters, palindromes can also be defined recursively. The insight you need
to do so is that any palindrome longer than a single character must contain a shorter
palindrome in its interior. For example, the string "level" consists of the palindrome
"eve" with an "l" at each end. Thus, to check whether a string is a palindrome—
assuming the string is sufficiently long that it does not constitute a simple case—all you
need to do is
Introduction to Recursion – 188 –

1. Check to see that the first and last characters are the same.
2. Check to see whether the substring generated by removing the first and last characters
is itself a palindrome.

If both conditions apply, the string is a palindrome.

The only other question you must consider before writing a recursive solution to the
palindrome problem is what the simple cases are. Clearly, any string with only a single
character is a palindrome because reversing a one-character string has no effect. The
one-character string therefore represents a simple case, but it is not the only one. The
empty string—which contains no characters at all—is also a palindrome, and any
recursive solution must operate correctly in this case as well.
Figure 5-3 contains a recursive implementation of the predicate function
IsPalindrome(str) that returns true if and only if the string str is a palindrome. The
function first checks to see whether the length of the string is less than 2. If it is, the
string is certainly a palindrome. In not, the function checks to make sure that the string
meets both of the criteria listed earlier.
This implementation in Figure 5-3 is somewhat inefficient, even though the recursive
decomposition is easy to follow. You can write a more efficient implementation of
IsPalindrome by making the following changes:

• Calculate the length of the argument string only once. The initial implementation
calculates the length of the string at every level of the recursive decomposition, even
though the structure of the solution guarantees that the length of the string decreases by
two on every recursive call. By calculating the length of the string at the beginning
and passing it down through each of the recursive calls, you can eliminate many calls
to the length method. To avoid changing the prototype for IsPalindrome, you need
Figure 5-3 Recursive implementation of IsPalindrome

/*
* Function: IsPalindrome
* Usage: if (IsPalindrome(str)) . . .
* -----------------------------------
* This function returns true if and only if the string is a.
* palindrome. This implementation operates recursively by noting
* that all strings of length 0 or 1 are palindromes (the simple
* case) and that longer strings are palindromes only if their first
* and last characters match and the remaining substring is a
* palindrome.
*/

bool IsPalindrome(string str) {

int len = str.length();
if (len <= 1) {
return true;
} else {
return (str[0] == str[len - 1]
&& IsPalindrome(str.substr(1, len - 2)));
}
}
Introduction to Recursion – 189 –

to define IsPalindrome as a wrapper function and have it pass the information to a

second recursive function that does all the actual work.
• Don’t make a substring on each call. Instead of calling substr to make copy of the
interior of the string, you can pass the first and last position for the substring as
parameters and allow those positions to define the subregion of the string being
checked.

The revised implementation of IsPalindrome appears in Figure 5-4.

Binary search
When you work with arrays or vectors, one of the most common algorithmic operations
consists of searching the array for a particular element. For example, if you were
working with arrays of strings, it would be extremely useful to have a function

Figure 5-4 More efficient implementation of IsPalindrome

/*
* Function: IsPalindrome
* Usage: if (IsPalindrome(str)) . . .
* -----------------------------------
* This function returns true if and only if the character string
* str is a palindrome. This level of the implementation is
* just a wrapper for the CheckPalindrome function, which
* does the real work.
*/

bool IsPalindrome(string str) {

return CheckPalindrome(str, 0, str.length() - 1);
}

/*
* Function: CheckPalindrome
* Usage: if (CheckPalindrome(str, firstPos, lastPos)) . . .
* ----------------------------------------------------------
* This function returns true if the characters from firstPos
* to lastPos in the string str form a palindrome. The
* implementation uses the recursive insight that all
* strings of length 0 or 1 are palindromes (the simple
* case) and that longer strings are palindromes only if
* their first and last characters match and the remaining
* substring is a palindrome. Recursively examining the
* interior substring is performed by adjusting the indexes
* of the range to examine. The interior substring
* begins at firstPos+1 and ends at lastPos-1.
*/

bool CheckPalindrome(string str, int firstPos, int lastPos) {

if (firstPos >= lastPos) {
return true;
} else {
return (str[firstPos] == str[lastPos]
&& CheckPalindrome(str, firstPos + 1, lastPos - 1));
}
}
Introduction to Recursion – 190 –

int FindStringInArray(string key, string array[], int n);

that searches through each of the n elements of array, looking for an element whose
value is equal to key. If such an element is found, FindStringInArray returns the index
at which it appears (if the key appears more than once in the array, the index of any
matching is fine). If no matching element exists, the function returns –1.
If you have no specific knowledge about the order of elements within the array, the
implementation of FindStringInArray must simply check each of the elements in turn
until it either finds a match or runs out of elements. This strategy is called the linear
search algorithm, which can be time-consuming if the arrays are large. On the other
hand, if you know that the elements of the array are arranged in alphabetical order, you
can adopt a much more efficient approach. All you have to do is divide the array in half
and compare the key you’re trying to find against the element closest to the middle of the
array, using the order defined by the ASCII character codes, which is called
lexicographic order. If the key you’re looking for precedes the middle element, then the
key—if it exists at all—must be in the first half. Conversely, if the key follows the
middle element in lexicographic order, you only need to look at the elements in the
second half. This strategy is called the binary search algorithm. Because binary search
makes it possible for you to discard half the possible elements at each step in the process,
it turns out to be much more efficient than linear search for sorted arrays.
The binary search algorithm is also a perfect example of the divide-and-conquer
strategy. It is therefore not surprising that binary search has a natural recursive
implementation, which is shown in Figure 5-5. Note that the function
FindStringInSortedArray is implemented as a wrapper, leaving the real work to the
recursive function BinarySearch, which takes two indices—low and high—that limit
the range of the search.
The simple cases for BinarySearch are
1. There are no elements in the active part of the array. This condition is marked by the
fact that the index low is greater than the index high, which means that there are no
elements left to search.
2. The middle element (or an element to one side of the middle if the array contains an
even number of elements) matches the specified key. Since the key has just been
found, FindStringInSortedArray can simply return the index of the middle value.
If neither of these cases applies, however, the implementation can simplify the problem
by choosing the appropriate half of the array and call itself recursively with an updated
set of search limits.

Mutual recursion
In each of the examples considered so far, the recursive functions have called themselves
directly, in the sense that the body of the function contains a call to itself. Although most
of the recursive functions you encounter are likely to adhere to this style, the definition of
recursion is actually somewhat broader. To be recursive, a function must call itself at
some point during its evaluation. If a function is subdivided into subsidiary functions, the
recursive call can actually occur at a deeper level of nesting. For example, if a function ƒ
calls a function g, which in turn calls ƒ, the function calls are still considered to be
recursive. Because the functions ƒ and g call each other, this type of recursion is called
mutual recursion.
Introduction to Recursion – 191 –

Figure 5-5 Divide-and-conquer implementation of binary search

/*
* Function: FindStringInSortedArray
* Usage: index = FindStringInSortedArray(key, array, n);
* ------------------------------------------------------
* This function searches the array looking for the specified
* key. The argument n specifies the effective size of the
* array, which must be sorted according to lexicographic
* order. If the key is found, the function returns the
* index in the array at which that key appears. (If the key
* appears more than once in the array, any of the matching
* indices may be returned). If the key does not exist in
* the array, the function returns -1. In this implementation,
* FindStringInSortedArray is simply a wrapper; all the work
* is done by the recursive function BinarySearch.
*/

int FindStringInSortedArray(string key, string array[], int n) {

return BinarySearch(key, array, 0, n - 1);
}

/*
* Function: BinarySearch
* Usage: index = BinarySearch(key, array, low, high);
* ---------------------------------------------------
* This function does the work for FindStringInSortedArray.
* The only difference is that BinarySearch takes both the
* upper and lower limit of the search.
*/

int BinarySearch(string key, string array[], int low, int high) {

if (low > high) return -1;
int mid = (low + high) / 2;
if (key == array[mid]) return mid;
if (key < array[mid]) {
return BinarySearch(key, array, low, mid - 1);
} else {
return BinarySearch(key, array, mid + 1, high);
}
}

As a simple example, let’s investigate how to use recursion to test whether a number is
even or odd. If you limit the domain of possible values to the set of natural numbers,
which are defined simply as the set of nonnegative integers, the even and odd numbers
can be characterized as follows:

• A number is even if its predecessor is odd.

• A number is odd if is not even.
• The number 0 is even by definition.

Even though these rules seem simplistic, they constitute the basis of an effective, if
inefficient, strategy for distinguishing odd and even numbers. A mutually recursive
implementation of the predicate functions IsEven and IsOdd appears in Figure 5-6.
Introduction to Recursion – 192 –

Figure 5-6 Mutually recursive definitions of IsEven and IsOdd

/*
* Function: IsEven
* Usage: if (IsEven(n)) . . .
* ---------------------------
* This function returns true if n is even. The number 0
* is considered even by definition; any other number is
* even if its predecessor is odd. Note that this function
* is defined to take an unsigned argument and is therefore
* not applicable to negative integers.
*/

bool IsEven(unsigned int n) {

if (n == 0) {
return true;
} else {
return IsOdd(n - 1);
}
}

/*
* Function: IsOdd
* Usage: if (IsOdd(n)) . . .
* --------------------------
* This function returns true if n is odd, where a number
* is defined to be odd if it is not even. Note that this
* function is defined to take an unsigned argument and is
* therefore not applicable to negative integers.
*/

bool IsOdd(unsigned int n) {

return !IsEven(n);
}

5.5 Thinking recursively

For most people, recursion is not an easy concept to grasp. Learning to use it effectively
requires considerable practice and forces you to approach problems in entirely new ways.
The key to success lies in developing the right mindset—learning how to think
recursively. The remainder of this chapter is designed to help you achieve that goal.

Maintaining a holistic perspective

In Chapter 2 of The Art and Science of C, I devote one section to the philosophical
concepts of holism and reductionism. Simply stated, reductionism is the belief that the
whole of an object can be understood merely by understanding the parts that make it up.
Its antithesis is holism, the position that the whole is often greater than the sum of its
parts. As you learn about programming, it helps to be able to interleave these two
perspectives, sometimes focusing on the behavior of a program as a whole, and at other
times delving into the details of its execution. When you try to learn about recursion,
however, this balance seems to change. Thinking recursively requires you to think
holistically. In the recursive domain, reductionism is the enemy of understanding and
invariably gets in the way.
To maintain the holistic perspective, you must become comfortable adopting the
recursive leap of faith, which was introduced in its own section earlier in this chapter.
Introduction to Recursion – 193 –

Whenever you are writing a recursive program or trying to understand the behavior of
one, you must get to the point where you ignore the details of the individual recursive
calls. As long as you have chosen the right decomposition, identified the appropriate
simple cases, and implemented your strategy correctly, those recursive calls will simply
work. You don’t need to think about them.
Unfortunately, until you have had extensive experience working with recursive
functions, applying the recursive leap of faith does not come easily. The problem is that
it requires to suspend your disbelief and make assumptions about the correctness of your
programs that fly in the face of your experience. After all, when you write a program, the
odds are good—even if you are an experienced programmer—that your program won’t
work the first time. In fact, it is quite likely that you have chosen the wrong
decomposition, messed up the definition of the simple cases, or somehow messed things
up trying to implement your strategy. If you have done any of these things, your
recursive calls won’t work.
When things go wrong—as they inevitably will—you have to remember to look for the
error in the right place. The problem lies somewhere in your recursive implementation,
not in the recursive mechanism itself. If there is a problem, you should be able to find it
by looking at a single level of the recursive hierarchy. Looking down through additional
levels of recursive calls is not going to help. If the simple cases work and the recursive
decomposition is correct, the subsidiary calls will work correctly. If they don’t, there is
something you need to fix in the definition of the recursive function itself.
Avoiding the common pitfalls
As you gain experience with recursion, the process of writing and debugging recursive
programs will become more natural. At the beginning, however, finding out what you
need to fix in a recursive program can be difficult. The following is a checklist that will
help you identify the most common sources of error.

• Does your recursive implementation begin by checking for simple cases? Before you
attempt to solve a problem by transforming it into a recursive subproblem, you must
first check to see if the problem is so simple that such decomposition is unnecessary.
In almost all cases, recursive functions begin with the keyword if. If your function
doesn’t, you should look carefully at your program and make sure that you know what
you’re doing.1
• Have you solved the simple cases correctly? A surprising number of bugs in recursive
programs arise from having incorrect solutions to the simple cases. If the simple cases
are wrong, the recursive solutions to more complicated problems will inherit the same
mistake. For example, if you had mistakenly defined Fact(0) as 0 instead of 1,
calling Fact on any argument would end up returning 0.
• Does your recursive decomposition make the problem simpler? For recursion to work,
the problems have to get simpler as you go along. More formally, there must be some
metric—a standard of measurement that assigns a numeric difficulty rating to the
problem—that gets smaller as the computation proceeds. For mathematical functions
like Fact and Fib , the value of the integer argument serves as a metric. On each
recursive call, the value of the argument gets smaller. For the IsPalindrome function,
the appropriate metric is the length of the argument string, because the string gets
shorter on each recursive call. If the problem instances do not get simpler, the

1 At times, as in the case of the IsPalindrome implementation, it may be necessary to perform some
calculations prior to making the simple-case test. The point is that the simple-case test must precede any
recursive decomposition.
Introduction to Recursion – 194 –

decomposition process will just keep making more and more calls, giving rise to the
recursive analogue of the infinite loop, which is called nonterminating recursion.
• Does the simplification process eventually reach the simple cases, or have you left out
some of the possibilities? A common source of error is failing to include simple case
tests for all the cases that can arise as the result of the recursive decomposition. For
example, in the IsPalindrome implementation presented in Figure 5-3, it is critically
important for the function to check the zero-character case as well as the one-character
case, even if the client never intends to call IsPalindrome on the empty string. As the
recursive decomposition proceeds, the string arguments get shorter by two characters
at each level of the recursive call. If the original argument string is even in length, the
recursive decomposition will never get to the one-character case.
• Do the recursive calls in your function represent subproblems that are truly identical
in form to the original? When you use recursion to break down a problem, it is
essential that the subproblems be of the same form. If the recursive calls change the
nature of the problem or violate one of the initial assumptions, the entire process can
break down. As several of the examples in this chapter illustrate, it is often useful to
define the publicly exported function as a simple wrapper that calls a more general
recursive function which is private to the implementation. Because the private
function has a more general form, it is usually easier to decompose the original
problem and still have it fit within the recursive structure.
• When you apply the recursive leap of faith, do the solutions to the recursive
subproblems provide a complete solution to the original problem? Breaking a
problem down into recursive subinstances is only part of the recursive process. Once
you get the solutions, you must also be able to reassemble them to generate the
complete solution. The way to check whether this process in fact generates the
solution is to walk through the decomposition, religiously applying the recursive leap
of faith. Work through all the steps in the current function call, but assume that every
recursive call generates the correct answer. If following this process yields the right
solution, your program should work.

Summary
This chapter has introduced the idea of recursion, a powerful programming strategy in
which complex problems are broken down into simpler problems of the same form. The
important points presented in this chapter include:

• Recursion is similar to stepwise refinement in that both strategies consist of breaking a

problem down into simpler problems that are easier to solve. The distinguishing
characteristic of recursion is that the simpler subproblems must have the same form as
the original.
• In C++, recursive functions typically have the following paradigmatic form:

if (test for simple case) {

Compute a simple solution without using recursion.
} else {
Break the problem down into subproblems of the same form.
Solve each of the subproblems by calling this function recursively.
Reassemble the solutions to the subproblems into a solution for the whole.
}

• To use recursion, you must be able to identify simple cases for which the answer is
easily determined and a recursive decomposition that allows you to break any complex
instance of the problem into simpler problems of the same type.
Introduction to Recursion – 195 –

• Recursive functions are implemented using exactly the same mechanism as any other
function call. Each call creates a new stack frame that contains the local variables for
that call. Because the computer creates a separate stack frame for each function call,
the local variables at each level of the recursive decomposition remain separate.
• Before you can use recursion effectively, you must learn to limit your analysis to a
single level of the recursive decomposition and to rely on the correctness of all simpler
recursive calls without tracing through the entire computation. Trusting these simpler
calls to work correctly is called the recursive leap of faith.
• Mathematical functions often express their recursive nature in the form of a recurrence
relation, in which each element of a sequence is defined in terms of earlier elements.
• Although some recursive functions may be less efficient than their iterative
counterparts, recursion itself is not the problem. As is typical with all types of
algorithms, some recursive strategies are more efficient than others.
• In order to ensure that a recursive decomposition produces subproblems that are
identical in form to the original, it is often necessary to generalize the problem. As a
result, it is often useful to implement the solution to a specific problem as a simple
wrapper function whose only purpose is to call a subsidiary function that handles the
more general case.
• Recursion need not consist of a single function that calls itself but may instead involve
several functions that call each other in a cyclical pattern. Recursion that involves
more than one function is called mutual recursion.
• You will be more successful at understanding recursive programs if you can maintain a
holistic perspective rather than a reductionistic one.

Thinking about recursive problems in the right way does not come easily. Learning to
use recursion effectively requires practice and more practice. For many students,
mastering the concept takes years. But because recursion will turn out to be one of the
most powerful techniques in your programming repertoire, that time will be well spent.

Review questions
1. Define the terms recursive and iterative. Is it possible for a function to employ both
strategies?

2. What is the fundamental difference between recursion and stepwise refinement?

3. In the pseudocode for the CollectContributions function, the if statement looks

like this:

if (n <= 100)

Why is it important to use the <= operator instead of simply checking whether n is
exactly equal to 100?

4. What is the standard recursive paradigm?

5. What two properties must a problem have for recursion to make sense as a solution
strategy?

6. Why is the term divide and conquer appropriate to recursive techniques?

Introduction to Recursion – 196 –

7. What is meant by the recursive leap of faith? Why is this concept important to you
as a programmer?

8. In the section entitled “Tracing the recursive process,” the text goes through a long
analysis of what happens internally when Fact(4) is called. Using this section as a
model, trace through the execution of Fib(4) , sketching out each stack frame
created in the process.

9. Modify Fibonacci’s rabbit problem by introducing the additional rule that rabbit
pairs stop reproducing after giving birth to three litters. How does this assumption
change the recurrence relation? What changes do you need to make in the simple
cases?

10. How many times is Fib(1) called when calculating Fib(n) using the recursive
implementation given in Figure 5-1?

11. What would happen if you eliminated the if (n == 1) check from the function
AdditiveSequence, so that the implementation looked like this:

int AdditiveSequence(int n, int t0, int t1) {

if (n == 0) return t0;
return AdditiveSequence(n - 1, t1, t0 + t1);
}

Would the function still work? Why or why not?

12. What is a wrapper function? Why are they often useful in writing recursive
functions?

13. Why is it important that the implementation of IsPalindrome in Figure 5-3 check
for the empty string as well as the single character string? What would happen if the
function didn’t check for the single character case and instead checked only whether
the length is 0? Would the function still work correctly?

14. Explain the effect of the function call

CheckPalindrome(str, firstPos + 1, lastPost - 1)

in the IsPalindrome implementation given in Figure 5-4.

15. What is mutual recursion?

16. What would happen if you defined IsEven and IsOdd as follows:

bool IsEven(unsigned int n) {

return !IsOdd(n);
}

bool IsOdd(unsigned int n) {

return !IsEven(n);
}

Which of the errors explained in the section “Avoiding the common pitfalls” is
illustrated in this example?
Introduction to Recursion – 197 –

17. The following definitions of IsEven and IsOdd are also incorrect:
bool IsEven(unsigned int n) {
if (n == 0) {
return true;
} else {
return IsOdd(n - 1);
}
}

bool IsOdd(unsigned int n) {

if (n == 1) {
return true;
} else {
return IsEven(n - 1);
}
}
Give an example that shows how this implementation can fail. What common pitfall
is illustrated here?

Programming exercises
1. Spherical objects, such as cannonballs, can be stacked to form a pyramid with one
cannonball at the top, sitting on top of a square composed of four cannonballs, sitting
on top of a square composed of nine cannonballs, and so forth. Write a recursive
function Cannonball that takes as its argument the height of the pyramid and returns
the number of cannonballs it contains. Your function must operate recursively and
must not use any iterative constructs, such as while or for.
2. Unlike many programming languages, C++ does not include a predefined operator
that raises a number to a power. As a partial remedy for this deficiency, write a
recursive implementation of a function
int RaiseIntToPower(int n, int k)

that calculates n k. The recursive insight that you need to solve this problem is the
mathematical property that
 1 if k = 0

nk = 
 n x n k–1 otherwise

3. The greatest common divisor (g.c.d.) of two nonnegative integers is the largest
integer that divides evenly into both. In the third century B. C ., the Greek
mathematician Euclid discovered that the greatest common divisor of x and y can
always be computed as follows:
• If x is evenly divisible by y, then y is the greatest common divisor.
• Otherwise, the greatest common divisor of x and y is always equal to the greatest
common divisor of y and the remainder of x divided by y.
Use Euclid’s insight to write a recursive function GCD(x, y) that computes the
greatest common divisor of x and y.
4. Write an iterative implementation of the function Fib(n).
Introduction to Recursion – 198 –

5. For each of the two recursive implementations of the function Fib(n) presented in
this chapter, write a recursive function (you can call these CountFib1 and
CountFib2 for the two algorithms) that counts the number of function calls made
during the evaluation of the corresponding Fibonacci calculation. Write a main
program that uses these functions to display a table showing the number of calls
made by each algorithm for various values of n, as shown in the following sample
run:

FibCount
This program compares the performance of two
algorithms to compute the Fibonacci sequence.

Number of calls:
N Fib1 Fib2
-- ---- ----
0 1 2
1 1 3
2 3 4
3 5 5
4 9 6
5 15 7
6 25 8
7 41 9
8 67 10
9 109 11
10 177 12
11 287 13
12 465 14

6. Write a recursive function DigitSum(n) that takes a nonnegative integer and returns
the sum of its digits. For example, calling DigitSum(1729) should return
1 + 7 + 2 + 9, which is 19.
The recursive implementation of DigitSum depends on the fact that it is very easy
to break an integer down into two components using division by 10. For example,
given the integer 1729, you can divide it into two pieces as follows:

1729

172 9
n / 10 n % 10

Each of the resulting integers is strictly smaller than the original and thus represents
a simpler case.

7. The digital root of an integer n is defined as the result of summing the digits
repeatedly until only a single digit remains. For example, the digital root of 1729
can be calculated using the following steps:

Step 1: 1 + 7 + 2 + 9 → 19
Step 2: 1 + 9 → 10
Step 3: 1 + 0 → 1
Introduction to Recursion – 199 –

Because the total at the end of step 3 is the single digit 1, that value is the digital
root.
Write a function DigitalRoot(n) that returns the digital root of its argument.
Although it is easy to implement DigitalRoot using the DigitSum function from
exercise 6 and a while loop, part of the challenge of this problem is to write the
function recursively without using any explicit loop constructs.
8. The mathematical combinations function C(n, k) is usually defined in terms of
factorials, as follows:
n!
C(n, k) =
k! x (n–k)!
The values of C(n, k) can also be arranged geometrically to form a triangle in which
n increases as you move down the triangle and k increases as you move from left to
right. The resulting structure,, which is called Pascal’s Triangle after the French
mathematician Blaise Pascal, is arranged like this:
C(0, 0)
C(1, 0) C(1, 1)
C(2, 0) C(2, 1) C(2, 2)
C(3, 0) C(3, 1) C(3, 2) C(3, 3)
C(4, 0) C(4, 1) C(4, 2) C(4, 3) C(4, 4)
Pascal’s Triangle has the interesting property that every entry is the sum of the two
entries above it, except along the left and right edges, where the values are always 1.
Consider, for example, the circled entry in the following display of Pascal’s
Triangle:

1
1 1
1 2 1
1 3 3 1
1 4 6 4 1
1 5 10 10 5 1
1 6 15 20 15 6 1

This entry, which corresponds to C(6, 2), is the sum of the two entries—5 and 10—
that appear above it to either side. Use this relationship between entries in Pascal’s
Triangle to write a recursive implementation of the Combinations function that uses
no loops, no multiplication, and no calls to Fact.
9. Write a recursive function that takes a string as argument and returns the reverse of
that string. The prototype for this function should be
string Reverse(string str);
and the statement
cout << Reverse("program") << endl;
Introduction to Recursion – 200 –

should display
ReverseString
margorp

Your solution should be entirely recursive and should not use any iterative constructs
such as while or for.

10. The strutils.h library contains a function IntegerToString,. You might have
wondered how the computer actually goes about the process of converting an integer
into its string representation. As it turns out, the easiest way to implement this
function is to use the recursive decomposition of an integer outlined in exercise 6.
Rewrite the IntegerToString implementation so that it operates recursively without
using use any of the iterative constructs such as while and for.
Chapter 6
Recursive Procedures

Nor would I consider the magnitude and complexity of my

plan as any argument of its impracticability.
— Mary Shelley, Frankenstein, 1818
Recursive Procedures – 202 –

When a recursive decomposition follows directly from a mathematical definition, as it

does in the case of the Fact and Fib functions in Chapter 5, applying recursion is not
particularly hard. In most cases, you can translate the mathematical definition directly
into a recursive implementation by plugging the appropriate expressions into the standard
recursive paradigm. The situation changes, however, as you begin to solve more
complex problems.
This chapter introduces several programming problems that seem—at least on the
surface—much more difficult than those in Chapter 5. In fact, if you try to solve these
problems without using recursion, relying instead on more familiar iterative techniques,
you will find them quite difficult to solve. Even so, each of the problems has a recursive
solution that is surprisingly short. If you exploit the power of recursion, a few lines of
code are sufficient for each task.
The brevity of these solutions endows them with a deceptive aura of simplicity. The
hard part of solving these problems does not lie in the intricacy or length of the code.
What makes these programs difficult is identifying the appropriate recursive
decomposition in the first place. Doing so occasionally requires some cleverness, but
what you need even more is confidence. You have to accept the recursive leap of faith.
As you develop your solution, you must strive to come to a point at which you are faced
with a problem that is identical to the original in form but simpler in scale. When you do,
you have to be willing to stop and declare the problem solved, without trying to trace the
program further.

6.1 The Tower of Hanoi

The first example in this chapter is a simple puzzle that has come to be known as the
Tower of Hanoi. Invented by French mathematician Edouard Lucas in the 1880s, the
Tower of Hanoi puzzle quickly became popular in Europe. Its success was due in part to
the legend that grew up around the puzzle, which was described as follows in La Nature
by the French mathematician Henri De Parville (as translated by the mathematical
historian W. W. R. Ball):

In the great temple at Benares beneath the dome which marks the
center of the world, rests a brass plate in which are fixed three
diamond needles, each a cubit high and as thick as the body of a bee.
On one of these needles, at the creation, God placed sixty-four disks
of pure gold, the largest disk resting on the brass plate and the others
getting smaller and smaller up to the top one. This is the Tower of
Brahma. Day and night unceasingly, the priests transfer the disks
from one diamond needle to another according to the fixed and
immutable laws of Brahma, which require that the priest on duty
must not move more than one disk at a time and that he must place
this disk on a needle so that there is no smaller disk below it. When
all the sixty-four disks shall have been thus transferred from the
needle on which at the creation God placed them to one of the other
needles, tower, temple and Brahmins alike will crumble into dust,
and with a thunderclap the world will vanish.

Over the years, the setting has shifted from India to Vietnam, but the puzzle and its
legend remain the same.
Recursive Procedures – 203 –

As far as I know, the Tower of Hanoi puzzle has no practical use except one: teaching
recursion to computer science students. In that domain, its value is unquestioned because
the solution involves almost nothing besides recursion. In contrast to most practical
examples of recursion, the Tower of Hanoi problem has no extraneous complications that
might interfere with your understanding and keep you from seeing how the recursive
solution works. Because it works so well as an example, the Tower of Hanoi is included
in most textbooks that treat recursion and has become part of the cultural heritage that
computer programmers share.
In most commercial versions of the puzzle, the 64 golden disks of legend are replaced
with eight wooden or plastic ones, which makes the puzzle considerably easier to solve,
not to mention cheaper. The initial state of the puzzle looks like this:

A B C

At the beginning, all eight disks are on spire A. Your goal is to move these eight disks
from spire A to spire B, but you must adhere to the following rules:

• You can only move one disk at a time.

• You are not allowed to move a larger disk on top of a smaller disk.

Framing the problem

In order to apply recursion to the Tower of Hanoi problem, you must first frame the
problem in more general terms. Although the ultimate goal is moving eight disks from A
to B, the recursive decomposition of the problem will involve moving smaller subtowers
from spire to spire in various configurations. In the more general case, the problem you
need to solve is moving a tower of a given height from one spire to another, using the
third spire as a temporary repository. To ensure that all subproblems fit the original form,
your recursive procedure must therefore take the following arguments :

1. The number of disks to move

2. The name of the spire where the disks start out
3. The name of the spire where the disks should finish
4. The name of the spire used for temporary storage

The number of disks to move is clearly an integer, and the fact that the spires are labeled
with the letters A, B, and C suggests the use of type char to indicate which spire is
involved. Knowing the types allows you to write a prototype for the operation that
moves a tower, as follows:

void MoveTower(int n, char start, char finish, char temp);

To move the eight disks in the example, the initial call is

Recursive Procedures – 204 –

MoveTower(8, 'A', 'B', 'C');

This function call corresponds to the English command “Move a tower of size 8 from
spire A to spire B using spire C as a temporary.” As the recursive decomposition
proceeds, MoveTower will be called with different arguments that move smaller towers in
various configurations.

Finding a recursive strategy

Now that you have a more general definition of the problem, you can return to the
problem of finding a strategy for moving a large tower. To apply recursion, you must
first make sure that the problem meets the following conditions:

1. There must be a simple case. In this problem, there is an obvious simple case.
Because the rules of the puzzle require you to move only one disk at a time, any
tower with more than one disk must be moved in pieces. If, however, the tower only
contains one disk, you can go ahead and move it, as long as you obey the other rules
of the game. Thus, the simple case occurs when n is equal to 1.
2. There must be a way to break the problem down into a simpler problem in such a way
that solving the smaller problem contributes to solving the original one. This part of
the problem is harder and will require closer examination.

To see how solving a simpler subproblem helps solve a larger problem, it helps to go
back and consider the original example with eight disks.

A B C

The goal here is to move eight disks from spire A to spire B. You need to ask yourself
how it would help if you could solve the same problem for a smaller number of disks. In
particular, you should think about how being able to move a stack of seven disks would
help you to solve the eight-disk case.
If you think about the problem for a few moments, it becomes clear that you can solve
the problem by dividing it into these three steps:

1. Move the entire stack consisting of the top seven disks from spire A to spire C.
2. Move the bottom disk from spire A to spire B.
3. Move the stack of seven disks from spire C to spire B.

Executing the first step takes you to the following position:

Recursive Procedures – 205 –

A B C

Once you have gotten rid of the seven disks on top of the largest disk, the second step is
simply to move that disk from spire A to spire B, which results in the following
configuration:

A B C

All that remains is to move the tower of seven disks back from spire C to spire B, which
is again a smaller problem of the same form. This operation is the third step in the
recursive strategy, and leaves the puzzle in the desired final configuration:

A B C

That’s it! You’re finished. You’ve reduced the problem of moving a tower of size
eight to one of moving a tower of size seven. More importantly, this recursive strategy
generalizes to towers of size N, as follows:

1. Move the top N–1 disks from the start spire to the temporary spire.
2. Move a single disk from the start spire to the finish spire.
3. Move the stack of N–1 disks from the temporary spire back to the finish spire.

At this point, it is hard to avoid saying to yourself, “Okay, I can reduce the problem to
moving a tower of size N–1, but how do I accomplish that?” The answer, of course, is
that you move a tower of size N–1 in precisely the same way. You break that problem
down into one that requires moving a tower of size N–2, which further breaks down into
Recursive Procedures – 206 –

the problem of moving a tower of size N–3, and so forth, until there is only a single disk
to move. Psychologically, however, the important thing is to avoid asking that question
altogether. The recursive leap of faith should be sufficient. You’ve reduced the scale of
the problem without changing its form. That’s the hard work. All the rest is
bookkeeping, and it’s best to let the computer take care of that.
Once you have identified the simple cases and the recursive decomposition, all you
need to do is plug them into the standard recursive paradigm, which results in the
following pseudocode procedure:
void MoveTower(int n, char start, char finish, char temp) {
if (n == 1) {
Move a single disk from start to finish.
} else {
Move a tower of size n - 1 from start to temp.
Move a single disk from start to finish.
Move a tower of size n - 1 from temp to finish.
}
}

Validating the strategy

Although the pseudocode strategy is in fact correct, the derivation up to this point has
been a little careless. Whenever you use recursion to decompose a problem, you must
make sure that the new problems are identical in form to the original. The task of moving
N–1 disks from one spire to another certainly sounds like an instance of the same problem
and fits the MoveTower prototype. Even so, there is a subtle but important difference. In
the original problem, the destination and temporary spires are empty. When you move a
tower of size N–1 to the temporary spire as part of the recursive strategy, you’ve left a
disk behind on the starting spire. Does the presence of that disk change the nature of the
problem and thus invalidate the recursive solution?
To answer this question, you need to think about the subproblem in light of the rules of
the game. If the recursive decomposition doesn’t end up violating the rules, everything
should be okay. The first rule—that only one disk can be moved at a time—is not an
issue. If there is more than a single disk, the recursive decomposition breaks the problem
down to generate a simpler case. The steps in the pseudocode that actually transfer disks
move only one disk at a time. The second rule—that you are not allowed to place a larger
disk on top of a smaller one—is the critical one. You need to convince yourself that you
will not violate this rule in the recursive decomposition.
The important observation to make is that, as you move a subtower from one spire to
the other, the disk you leave behind on the original spire—and indeed any disk left
behind at any previous stage in the operation—must be larger than anything in the current
subtower. Thus, as you move those disks among the spires, the only disks below them
will be larger in size, which is consistent with the rules.

Coding the solution

To complete the Tower of Hanoi solution, the only remaining step is to substitute
function calls for the remaining pseudocode. The task of moving a complete tower
requires a recursive call to the MoveTower function. The only other operation is moving a
single disk from one spire to another. For the purposes of writing a test program that
displays the steps in the solution, all you need is a function that records its operation on
the console. For example, you can implement the function MoveSingleDisk as follows:
Recursive Procedures – 207 –

void MoveSingleDisk(char start, char finish) {

cout << start << " -> " << finish << endl;
}

The MoveTower code itself looks like this:

void MoveTower(int n, char start, char finish, char temp) {

if (n == 1) {
MoveSingleDisk(start, finish);
} else {
MoveTower(n - 1, start, temp, finish);
MoveSingleDisk(start, finish);
MoveTower(n - 1, temp, finish, start);
}
}

Tracing the recursive process

The only problem with this implementation of MoveTower is that it seems like magic. If
you’re like most students learning about recursion for the first time, the solution seems so
short that there must be something missing. Where is the strategy? How can the
computer know which disk to move first and where it should go?
The answer is that the recursive process—breaking a problem down into smaller
subproblems of the same form and then providing solutions for the simple cases—is all
you need to solve the problem. If you trust the recursive leap of faith, you’re done. You
can skip this section of the book and go on to the next. If you’re still suspicious, it may
be necessary for you to go through the steps in the complete process and watch what
happens.
To make the problem more manageable, consider what happens if there are only three
disks in the original tower. The main program call is therefore
MoveTower(3, 'A', 'B', 'C');

To trace how this call computes the steps necessary to transfer a tower of size 3, all you
need to do is keep track of the operation of the program, using precisely the same strategy
as in the factorial example from Chapter 5. For each new function call, you introduce a
stack frame that shows the values of the parameters for that call. The initial call to
MoveTower, for example, creates the following stack frame:

MoveTower
n start finish temp
3 A B C

→ if (n == 1) {
MoveSingleDisk(start, finish);
} else {
MoveTower(n - 1, start, temp, finish);
MoveSingleDisk(start, finish);
MoveTower(n - 1, temp, finish, start);
}
Recursive Procedures – 208 –

As the arrow in the code indicates, the function has just been called, so execution begins
with the first statement in the function body . The current value of n is not equal to 1, so
the program skips ahead to the else clause and executes the statement

MoveTower(n-1, start, temp, finish);

As with any function call, the first step is to evaluate the arguments. To do so, you
need to determine the values of the variables n, start, temp, and finish. Whenever you
need to find the value of a variable, you use the value as it is defined in the current stack
frame. Thus, the MoveTower call is equivalent to

MoveTower(2, 'A', 'C', 'B');

This operation, however, indicates another function call, which means that the current
operation is suspended until the new function call is complete. To trace the operation of
the new function call, you need to generate a new stack frame and repeat the process. As
always, the parameters in the new stack frame are initialized by copying the calling
arguments in the order in which they appear. Thus, the new stack frame looks like this:

MoveTower
MoveTower
n start finish temp
2 A C B

→ if (n == 1) {
MoveSingleDisk(start, finish);
} else {
MoveTower(n - 1, start, temp, finish);
MoveSingleDisk(start, finish);
MoveTower(n - 1, temp, finish, start);
}

As the diagram illustrates, the new stack frame has its own set of variables, which
temporarily supersede the variables in frames that are further down on the stack. Thus, as
long as the program is executing in this stack frame, n will have the value 2, start will
be 'A', finish will be 'C', and temp will be 'B'. The old values in the previous frame
will not reappear until the subtask represented by this call is created and the function
returns.
The evaluation of the recursive call to MoveTower proceeds exactly like the original
one. Once again, n is not 1, which requires another call of the form

MoveTower(n-1, start, temp, finish);

Because this call comes from a different stack frame, however, the value of the individual
variables are different from those in the original call. If you evaluate the arguments in the
context of the current stack frame, you discover that this function call is equivalent to

MoveTower(1, 'A', 'B', 'C');

The effect of making this call is to introduce yet another stack frame for the MoveTower
function, as follows:
Recursive Procedures – 209 –

MoveTower
MoveTower
MoveTower
n start finish temp
1 A B C

→ if (n == 1) {
MoveSingleDisk(start, finish);
} else {
MoveTower(n - 1, start, temp, finish);
MoveSingleDisk(start, finish);
MoveTower(n - 1, temp, finish, start);
}

This call to MoveTower, however, does represent the simple case. Since n is 1, the
program calls the MoveSingleDisk function to move a disk from A to B, leaving the
puzzle in the following configuration:

A B C

At this point, the most recent call to MoveTower is complete and the function returns.
In the process, its stack frame is discarded, which brings the execution back to the
previous stack frame, having just completed the first statement in the else clause:

MoveTower
MoveTower
n start finish temp
2 A C B

if (n == 1) {
MoveSingleDisk(start, finish);
} else {
MoveTower(n - 1, start, temp, finish);
→ MoveSingleDisk(start, finish);
MoveTower(n - 1, temp, finish, start);
}

The call to MoveSingleDisk again represents a simple operation, which leaves the
puzzle in the following state:
Recursive Procedures – 210 –

A B C

With the MoveSingleDisk operation completed, the only remaining step required to
finish the current call to MoveTower is the last statement in the function:

MoveTower(n-1, temp, finish, start);

Evaluating these arguments in the context of the current frame reveals that this call is
equivalent to

MoveTower(1, 'B', 'C', 'A');

Once again, this call requires the creation of a new stack frame. By this point in the
process, however, you should be able to see that the effect of this call is simply to move a
tower of size 1 from B to C, using A as a temporary repository. Internally, the function
determines that n is 1 and then calls M o v e S i n g l e D i s k to reach the following
configuration:

A B C

This operation again completes a call to MoveTower, allowing it to return to its caller
having completed the subtask of moving a tower of size 2 from A to C. Discarding the
stack frame from the just-completed subtask reveals the stack frame for the original call
to MoveTower, which is now in the following state:

MoveTower
n start finish temp
3 A B C

if (n == 1) {
MoveSingleDisk(start, finish);
} else {
MoveTower(n - 1, start, temp, finish);
→ MoveSingleDisk(start, finish);
MoveTower(n - 1, temp, finish, start);
}

The next step is to call MoveSingleDisk to move the largest disk from A to B, which
results in the following position:
Recursive Procedures – 211 –

A B C

The only operation that remains is to call

MoveTower(n-1, temp, finish, start);

with the arguments from the current stack frame, which are
MoveTower(2, 'C', 'B', 'A');

If you’re still suspicious of the recursive process, you can draw the stack frame created
by this function call and continue tracing the process to its ultimate conclusion. At some
point, however, it is essential that you trust the recursive process enough to see that
function call as a single operation having the effect of the following command in English:

Move a tower of size 2 from C to B, using A as a temporary repository.

If you think about the process in this holistic form, you can immediately see that
completion of this step will move the tower of two disks back from C to B, leaving the
desired final configuration:

A B C

6.2 Generating permutations

Many word games such as Scrabble® require the ability to rearrange a set of letters to
form a word. Thus, if you wanted to write a Scrabble program, it would be useful to have
a facility for generating all possible arrangements of a particular set of tiles. In word
games, such arrangements are generally called anagrams. In mathematics, they are
known as permutations.
Let’s suppose you want to write a function ListPermutations(s) that displays all
permutations of the string s. For example, if you call

ListPermutations("ABC");

your program should display the six arrangements of "ABC", as follows:

ListPermutations
ABC
ACB
BAC
BCA
CBA
CAB
Recursive Procedures – 212 –

The order of the output is unimportant, but each of the possible arrangements should
appear exactly once.
How would you go about implementing the ListPermutations function? If you are
limited to iterative control structures, finding a general solution that works for strings of
any length is difficult. Thinking about the problem recursively, on the other hand, leads
to a relatively straightforward solution.
As is usually the case with recursive programs, the hard part of the solution process is
figuring out how to divide the original problem into simpler instances of the same
problem. In this case, to generate all permutations of a string, you need to discover how
being able to generate all permutations of a shorter string might contribute to the solution.
Stop and think about this problem for a few minutes. When you are first learning
about recursion, it is easy to look at a recursive solution and believe that you could have
generated it on your own. Without trying it first, however, it is hard to know whether you
would have come up with the same insight.
To give yourself more of a feel for the problem, you need to consider a concrete case.
Suppose you want to generate all permutations of a five-character string, such as
"ABCDE" . In your solution, you can apply the recursive leap of faith to generate all
permutations of any shorter string. Just assume that the recursive calls work and be done
with it. Once again, the critical question is how being able to permute shorter strings
helps you solve the problem of permuting the original five-character string.
The recursive insight
The key to solving the permutation problem is recognizing that the permutations of the
five-character string "ABCDE" consist of the following strings:
• The character 'A' followed by every possible permutation of "BCDE"
• The character 'B' followed by every possible permutation of "ACDE"
• The character 'C' followed by every possible permutation of "ABDE"
• The character 'D' followed by every possible permutation of "ABCE"
• The character 'E' followed by every possible permutation of "ABCD"
More generally, to display all permutations of a string of length n, you can take each of
the n characters in turn and display that character followed by every possible permutation
of the remaining n – 1 characters.
The only difficulty with this solution strategy is that the recursive subproblem does not
have exactly the same form as the original. The original problem requires you to display
all permutations of a string. The subproblem requires you to display a character from a
string followed by all permutations of the remaining letters. As the recursion proceeds,
the character in front will become two characters, then three, and so forth. The general
subproblem, therefore, is to generate all permutations of a string, with some characters at
the beginning of the string already fixed in their positions.
As discussed in Chapter 5, the easiest way to solve the problem of asymmetry between
the original problem and its recursive subproblems is to define ListPermutations as a
simple wrapper function that calls a subsidiary function to solve the more general case.
In this example, the general problem can be solved by a new procedure
RecursivePermute, which generates all permutations of the remaining characters in a
string having already chosen some characters as the prefix. The prefix starts empty and
Recursive Procedures – 213 –

all the original letters still remain to be examined, which gives you the original problem.
As the prefix grows and there are fewer characters remaining, the problem becomes
simpler. When there are no characters remaining to be permuted, all characters have been
placed in the prefix, and it can be displayed exactly as it appears. The definition of
ListPermutations itself looks like this:

void ListPermutations(string str) {

RecursivePermute("", str);
}

The RecursivePermute procedure follows the outline of the recursive permutation

algorithm and has the following pseudocode form:
void RecursivePermute(string prefix, string rest) {
if (rest is empty) {
Display the prefix string.
} else {
For each character in rest {
Add the character to the end of prefix.
Remove character from rest.
Use recursion to generate permutations with the updated values for prefix and rest.
}
}
}

Translating this function from pseudocode to C++ is reasonably simple. The full
definition of RecursivePermute looks like this:
void RecursivePermute(string prefix, string rest) {
if (rest == "") {
cout << prefix << endl;
} else {
for (int i = 0; i < rest.length(); i++) {
string newPrefix = prefix + rest[i];
string newRest = rest.substr(0, i) + rest.substr(i+1);
RecursivePermute(newPrefix, newRest);
}
}
}

6.3 Graphical applications of recursion

Back in the early 1990s, I developed a simple graphics library that makes it possible to
construct simple drawings on a computer display using line segments and circular arcs.1
Using the graphics library makes programming more fun, but it also has proven to be
useful in illustrating the concepts of recursion. Many of the figures you can create with
the graphics library are fundamentally recursive in nature and provide excellent examples
of recursive programming.
The remainder of this chapter includes an overview of the graphics library, followed
by two illustrations of how recursion can be applied in the graphics domain. This
material is not essential to learning about recursion, and you can skip it if you don’t have
ready access to the graphics library. On the other hand, working through these examples
might make recursion seem a lot more powerful, not to mention more fun.

1 Note that the C-based graphics library uses a very different drawing model from the acm.graphics
package available in Java. We’re working on a new graphics library for C++ that looks like the Java one.
Recursive Procedures – 214 –

The graphics library

The graphics.h interface provides you with access to a collection of functions that
enable you to create simple line drawings on the computer screen. When you initialize
the graphics package, a new rectangular window called the graphics window is created
on the screen and used as the drawing surface. Whenever you call procedures and
functions in the graphics library, the results are displayed in the graphics window.
To specify points within the graphics window, the graphics library uses an approach
that should be familiar from high-school geometry or algebra. All drawing in the
graphics window takes place on a conceptual grid, as illustrated in Figure 6-1. As in
traditional geometry, points are identified by specifying their position relative to the
origin, which is the point at the lower left corner of the graphics window. The horizontal
and vertical lines that emanate from the origin along the edges of the graphics window
are called the axes; the x-axis runs along the bottom of the window, and the y-axis runs
up the left side. Every point in the graphics window is identified by a pair of values,
usually written as (x, y), that specifies the position of that point along the x and y axes.
These values are called the coordinates of the point. Coordinates are measured in inches
relative to the origin, which is the point (0, 0). From there, x values increase as you move
to the right, and y values increase as you move up.
Coordinates in the graphics library come in two forms:

• Absolute coordinates specify a point in the window by giving its coordinates with
respect to the origin. For example, the solid dot in Figure 6-1 is at absolute
coordinates (2.0, 1.5).
• Relative coordinates specify a position in the window by indicating how far away that
point is along each axis from the last position specified. For example, the open dot in
Figure 6-1 has absolute coordinates (2.5, 1.5). If, however, you express its coordinates
in relation to the solid dot, this point is shifted by the relative coordinates (0.5, 0.0). If
you want to connect these dots with a line, the standard approach is to specify the first
point in absolute coordinates and the endpoint of the line in relative coordinates.
Figure 6-1 Coordinates in the graphics library
3

0
0 1 2 3
Recursive Procedures – 215 –

The best mental model to use for the drawing process is to imagine that there is a pen
positioned over a piece of transparent graph paper covering the screen. You can move
the pen to any location on the screen by specifying the absolute coordinates. You then
draw a straight line by moving the pen to a new point specified using relative coordinates,
making sure that the pen continuously touches the graph paper as you draw the line.
From there, you can start another line beginning where the last one ended.
The functions exported by the graphics.h interface are shown in Table 6-1. Graphics
applications begin by calling InitGraphics, after which the graphical image itself is
created by calls to MovePen , DrawLine , and DrawArc . The remaining functions—
GetWindowWidth , GetWindowHeight , GetCurrentX , and GetCurrentY —make it
possible to retrieve information about the dimensions and state of the graphics window.
These functions come up less frequently, but are nonetheless useful enough that it makes
sense to include them in the interface.
To get a better sense of how the graphics library works, consider the following
program, which draws a simple archway:

Table 6-1 Functions exported by graphics.h

Function call Operation
InitGraphics() This procedure creates the graphics window on the screen.
The call to InitGraphics must precede any output
operations of any kind and is usually the first statement in the
function main.
MovePen(x, y) This procedure picks up the pen and moves it—without
drawing any lines—to the position (x, y), which is specified
in absolute coordinates.
DrawLine(dx, dy) This procedure draws a line extending from the current point
by moving the pen dx inches in the x direction and dy inches
in the y direction. The final position becomes the new current
point.
DrawArc(r, start, sweep) This procedure draws a circular arc, which always begins at
the current point. The arc itself has radius r, and starts at the
angle specified by the parameter start, relative to the center of
the circle. This angle is measured in degrees
counterclockwise from the 3 o’clock position along the x-
axis, as in traditional mathematics. For example, if start is 0,
the arc begins at the 3 o’clock position; if start is 90, the arc
begins at the 12 o’clock position; and so on. The fraction of
the circle drawn is specified by the parameter sweep, which is
also measured in degrees. If sweep is 360, DrawArc draws a
complete circle; if sweep is 90, it draws a quarter of a circle.
If the value of s w e e p is positive, the arc is drawn
counterclockwise from the current point; if sweep is negative,
the arc is drawn clockwise. The current point at the end of
the DrawArc operation is the final position of the pen along
the arc.
GetWindowWidth() These functions return the width and height of the graphics
GetWindowHeight() window, respectively.
GetCurrentX() These functions return the absolute coordinates of the current
GetCurrentY() point.
Recursive Procedures – 216 –

int main() {
InitGraphics();
MovePen(2.0, 0.5);
DrawLine(1.0, 0.0);
DrawLine(0.0, 1.0);
DrawArc(0.5, 0, 180);
DrawLine(0.0, -1.0);
return 0;
}

The program begins, like all graphics programs, with a call to InitGraphics , which
creates an empty graphics window. The next two statements then move the pen to the
point (2.0, 0.5) and draw a line with the relative coordinates (1.0, 0.0). The effect of
these statements is to draw a 1-inch horizontal line near the bottom of the window. The
next call to DrawLine adds a vertical line that begins where the first line ended. Thus, at
this point, the graphics window contains two lines in the following configuration:

The next statement

DrawArc(0.5, 0, 180);

draws a circular arc with a radius of 0.5 inches. Because the second argument is 0, the
arc begins at the 0 degree mark, which corresponds to the 3 o’clock position. From there,
the third argument indicates that the arc runs in the positive direction (counterclockwise)
for a total of 180 degrees, or halfway around the circle. Adding this semicircle to the line
segments generated earlier makes the graphics window look like this:

The last line in the program draws a 1-inch vertical line in the downward direction,
which completes the archway, as shown:
Recursive Procedures – 217 –

An example from computer art

In the early part of the twentieth century, a controversial artistic movement arose in Paris,
largely under the influence of Pablo Picasso and Georges Braque. The Cubists—as they
were called by their critics—rejected classical artistic notions of perspective and
representationalism and instead produced highly fragmented works based on primitive
geometrical forms. Strongly influenced by Cubism, the Dutch painter Piet Mondrian
(1872–1944) produced a series of compositions based on horizontal and vertical lines,
such as the one shown in Figure 6-2.
Suppose that you want to generate compositions such as the following, which—like
much of Mondrian’s work—consists only of horizontal and vertical lines:

How would you go about designing a general strategy to create such a figure using the
graphics library?

Figure 6-2 Grid pattern from Piet Mondrian, “Composition with Grid 6,” 1919
Recursive Procedures – 218 –

To understand how a program might produce such a figure, it helps to think about the
process as one of successive decomposition. At the beginning, the canvas was simply an
empty rectangle that looked like this:

If you want to subdivide the canvas using a series of horizontal and vertical lines, the
easiest way to start is by drawing a single line that divides the rectangle in two:

If you’re thinking recursively, the thing to notice at this point is that you now have two
empty rectangular canvases, each of which is smaller in size. The task of subdividing
these rectangles is the same as before, so you can perform it by using a recursive
implementation of the same procedure. Since the new rectangles are taller than they are
wide, you might choose to use a horizontal dividing line, but the basic process remains
the same.
At this point, the only thing needed for a complete recursive strategy is a simple case.
The process of dividing up rectangles can’t go on indefinitely. As the rectangles get
smaller and smaller, at some point the process has to stop. One approach is to look at the
area of each rectangle before you start. Once the area of a rectangle falls below some
threshold, you needn’t bother to subdivide it any further.
The mondrian.cpp program shown in Figure 6-3 implements the recursive algorithm,
using the entire graphics window as the initial canvas.
Recursive Procedures – 219 –

Figure 6-3 Program to generate a Mondrian-style drawing

/*
* File: mondrian.cpp
* ------------------
* This program creates a random line drawing in a style reminiscent
* of the Dutch painter Piet Mondrian. The picture is generated by
* recursively subdividing the canvas into successively smaller
* rectangles with randomly chosen horizontal and vertical lines.
*/

#include "genlib.h"
#include "graphics.h"
#include "random.h"

/*
* Constants
* ---------
* MIN_AREA -- Smallest square that will be split
* MIN_EDGE -- Minimum fraction on each side of dividing line
*/

const double MIN_AREA = 0.5;

const double MIN_EDGE = 0.15;
/* Private function prototypes */

void SubdivideCanvas(double x, double y, double width, double height);

/* Main program */

int main() {
InitGraphics();
Randomize();
SubdivideCanvas(0, 0, GetWindowWidth(), GetWindowHeight());
return 0;
}

/*
* Function: SubdivideCanvas
* Usage: SubdivideCanvas(x, y, width, height);
* --------------------------------------------
* This function decomposes a canvas by recursive subdivision. The
* lower left corner of the canvas is the point (x, y), and the
* dimensions are given by the width and height parameters. The
* function first checks for the simple case, which is obtained
* when the size of the rectangular canvas is too small to subdivide
* (area < MIN_AREA). In the simple case, the function does nothing.
* If the area is larger than the minimum, the function first
* decides whether to split the canvas horizontally or vertically,
* choosing the larger dimension. The function then chooses a
* random dividing line, making sure to leave at least MIN_EDGE on
* each side. The program then uses a divide-and-conquer strategy
* to subdivide the two new rectangles.
*/
Recursive Procedures – 220 –

void SubdivideCanvas(double x, double y, double width, double height) {

if (width * height >= MIN_AREA) {
if (width > height) {
double mid = width * RandomReal(MIN_EDGE, 1 - MIN_EDGE);
MovePen(x + mid, y);
DrawLine(0, height);
SubdivideCanvas(x, y, mid, height);
SubdivideCanvas(x + mid, y, width - mid, height);
} else {
double mid = height * RandomReal(MIN_EDGE, 1 - MIN_EDGE);
MovePen(x, y + mid);
DrawLine(width, 0);
SubdivideCanvas(x, y, width, mid);
SubdivideCanvas(x, y + mid, width, height - mid);
}
}
}

In mondrian.cpp , the recursive function SubdivideCanvas does all the work. The
arguments give the position and dimensions of the current rectangle on the canvas. At
each step in the decomposition, the function simply checks to see whether the rectangle is
large enough to split. If it is, the function checks to see which dimension—width or
height—is larger and accordingly divides the rectangle with a vertical or horizontal line.
In each case, the function draws only a single line; all remaining lines in the figure are
drawn by subsequent recursive calls.

Fractals
In the late 1970s, a researcher at IBM named Benoit Mandelbrot generated a great deal of
excitement by publishing a book on fractals, which are geometrical structures in which
the same pattern is repeated at many different scales. Although mathematicians have
known about fractals for a long time, there was a resurgence of interest in the subject
during the 1980s, partly because the development of computers made it possible to do so
much more with fractals than had ever been possible before.
One of the earliest examples of fractal figures is called the Koch snowflake after its
inventor, Helge von Koch. The Koch snowflake begins with an equilateral triangle like
this:

This triangle, in which the sides are straight lines, is called the Koch fractal of order 0.
The figure is then revised in stages to generate fractals of successively higher orders. At
each stage, every straight-line segment in the figure is replaced by one in which the
middle third consists of a triangular bump protruding outward from the figure. Thus, the
first step is to replace each line segment in the triangle with a line that looks like this:
Recursive Procedures – 221 –

Applying this transformation to each of the three sides of the original triangle generates
the Koch fractal of order 1, as follows:

If you then replace each line segment in this figure with a new line that again includes
a triangular wedge, you create the following order-2 Koch fractal:

Replacing each of these line segments gives the order-3 fractal shown in the following
diagram, which has started to resemble a snowflake:

Because figures like the Koch fractal are much easier to draw by computer than by
hand, it makes sense to write a program that uses the graphics library to generate this
design. Before designing the program itself, however, it helps to introduce a new
procedure that will prove useful in a variety of graphical applications. The DrawLine
primitive in the graphics library requires you to specify the relative coordinates of the
new endpoint as a pair of values, dx and dy. In many graphical applications, it is much
easier to think of lines as having a length and a direction. For example, the solid line in
the following diagram can be identified by its length (r) and its angle from the x-axis (θ):
Recursive Procedures – 222 –

θ
x-axis
In mathematics, the parameters r and θ are called the polar coordinates of the line.
Converting from polar coordinates to the more traditional Cartesian coordinates used in
the graphics library requires a little trigonometry, as shown in the following
implementation of the procedure DrawPolarLine, which draws a line of length r in the
direction theta, measured in degrees counterclockwise from the x-axis:
const double PI = 3.1415926535;
void DrawPolarLine(double r, double theta) {
double radians = theta / 180 * PI;
DrawLine(r * cos(radians), r * sin(radians));
}

If you don’t understand trigonometry, don’t worry. You don’t need to understand the
implementation of DrawPolarLine to use it. If you have had a trigonometry course, most
of this implementation should be straightforward; the only complexity comes from the
fact that the library functions sin and cos are defined to take their arguments in radian
measure, which means that the implementation must convert the theta parameter from
degrees to radians prior to calling the trigonometric functions.
Given DrawPolarLine, it is very easy to draw the equilateral triangle that represents
the Koch snowflake of order 0. If size is the length of one side of the triangle, all you
need to do is position the pen at the lower left corner of the figure and make the following
calls:

DrawPolarLine(size, 0);
DrawPolarLine(size, 120);
DrawPolarLine(size, 240);

But how would you go about drawing a more complicated Koch fractal, that is, one of
a higher order? The first step is simply to replace each of the calls to DrawPolarLine
with a call to a new procedure that draws a fractal line of a specified order. Thus, the
three calls in the main program look like this:

DrawFractalLine(size, 0, order);
DrawFractalLine(size, 120, order);
DrawFractalLine(size, 240, order);

The next task is to implement DrawFractalLine, which is easy if you think about it
recursively. The simple case for DrawFractalLine occurs when order is 0, in which
case the function simply draws a straight line with the specified length and direction. If
order is greater than 0, the fractal line is broken down into four components, each of
which is itself a fractal line of the next lower order. Thus, the implementation of
DrawFractalLine looks like this:
Recursive Procedures – 223 –

void DrawFractalLine(double len, double theta, int order) {

if (order == 0) {
DrawPolarLine(len, theta);
} else {
DrawFractalLine(len/3, theta, order - 1);
DrawFractalLine(len/3, theta - 60, order - 1);
DrawFractalLine(len/3, theta + 60, order - 1);
DrawFractalLine(len/3, theta, order - 1);
}
}

The complete implementation of the koch.cpp program is shown in Figure 6-4.

Figure 6-4 Program to draw a Koch fractal snowflake

/*
* File: koch.cpp
* ---------------
* This program draws a Koch fractal.
*/

#include <iostream>
#include <cmath>
#include "simpio.h"
#include "graphics.h"
#include "genlib.h"

/* Constants */

const double PI = 3.1415926535;

/* Private function prototypes */

void KochFractal(double size, int order);

void DrawFractalLine(double len, double theta, int order);
void DrawPolarLine(double r, double theta);

/* Main program */

int main() {
InitGraphics();
cout << "Program to draw Koch fractals" << endl;
cout << "Enter edge length in inches: ";
double size = GetReal();
cout << "Enter order of fractal: ";
int order = GetInteger();
KochFractal(size, order);
return 0;
}

/*
* Function: KochFractal
* Usage: KochFractal(size, order);
* --------------------------------
* This function draws a Koch fractal snowflake centered in
* the graphics window of the indicated size and order.
*/
Recursive Procedures – 224 –

void KochFractal(double size, int order) {

double x0 = GetWindowWidth() / 2 - size / 2;
double y0 = GetWindowHeight() / 2 - sqrt(3) * size / 6;
MovePen(x0, y0);
DrawFractalLine(size, 0, order);
DrawFractalLine(size, 120, order);
DrawFractalLine(size, 240, order);
}

/*
* Function: DrawFractalLine
* Usage: DrawFractalLine(len, theta, order);
* ------------------------------------------
* This function draws a fractal line of the given length, starting
* from the current point and moving in direction theta. If order
* is 0, the fractal line is a straight line. If order is greater
* than zero, the line is divided into four line segments, each of
* which is a fractal line of the next lower order. These segments
* connect the same endpoints as the straight line, but include
* a triangular wedge replacing the center third of the segment.
*/
void DrawFractalLine(double len, double theta, int order) {
if (order == 0) {
DrawPolarLine(len, theta);
} else {
DrawFractalLine(len/3, theta, order - 1);
DrawFractalLine(len/3, theta - 60, order - 1);
DrawFractalLine(len/3, theta + 60, order - 1);
DrawFractalLine(len/3, theta, order - 1);
}
}

/*
* Function: DrawPolarLine
* Usage: DrawPolarLine(r, theta);
* -------------------------------
* This function draws a line of length r in the direction
* specified by the angle theta, measured in degrees.
*/
void DrawPolarLine(double r, double theta) {
double radians = theta / 180 * PI;
DrawLine(r * cos(radians), r * sin(radians));
}

Summary
Except for the discussion of the graphics library in the section entitled “Graphical
applications of recursion” earlier in this chapter, relatively few new concepts have been
introduced in Chapter 6. The fundamental precepts of recursion were introduced in
Chapter 5. The point of Chapter 6 is to raise the sophistication level of the recursive
examples to the point at which the problems become difficult to solve in any other way.
Because of this increase in sophistication, beginning students often find these problems
much harder to comprehend than those in the preceding chapter. Indeed, they are harder,
but recursion is a tool for solving hard problems. To master it, you need to practice with
problems at this level of complexity.
Recursive Procedures – 225 –

The important points in this chapter include:

• Whenever you want to apply recursion to a programming problem, you have to devise
a strategy that transforms the problem into simpler instances of the same problem.
Until you find the correct insight that leads to the recursive strategy, there is no way to
apply recursive techniques.
• Once you identify a recursive approach, it is important for you to check your strategy
to ensure that it does not violate any conditions imposed by the problem.
• When the problems you are trying to solve increase in complexity, the importance of
accepting the recursive leap of faith increases.
• Recursion is not magical. If you need to do so, you can simulate the operation of the
computer yourself by drawing the stack frames for every procedure that is called in the
course of the solution. On the other hand, it is critical to get beyond the skepticism
that forces you to look at all the underlying details.
• Wrapper functions are useful in complex recursive procedures, just as they are for the
simple recursive functions presented in Chapter 5.
• The graphics.h library makes it possible to display simple graphical drawings, many
of which have an interesting recursive structure.

Review questions
1. In your own words, describe the recursive insight necessary to solve the Tower of
Hanoi puzzle.
2. What is wrong with the following strategy for solving the recursive case of the
Tower of Hanoi puzzle:
a. Move the top disk from the start spire to the temporary spire.
b. Move a stack of N–1 disks from the start spire to the finish spire.
c. Move the top disk now on the temporary spire back to the finish spire.
3. If you call
MoveTower(16, 'A', 'B', 'C')
what line is displayed by MoveSingleDisk as the first step in the solution? What is
the last step in the solution?
4. What is a permutation?
5. In your own words, explain the recursive insight necessary to enumerate the
permutations of the characters in a string.
6. How many permutations are there of the string "WXYZ"?
7. Why is it necessary to define both ListPermutations and RecursivePermute in
the permutation problem?
8. Where is the origin located in the graphics window?
9. What is the difference between absolute and relative coordinates?
10. What are the eight functions exported by the graphics.h interface?
11. What simple case is used to terminate the recursion in mondrian.cpp?
Recursive Procedures – 226 –

12. Draw a picture of the order-1 Koch fractal.

13. How many line segments appear in the order-2 Koch fractal?
14. From the caller’s point of view, describe the effect of the function DrawPolarLine.

Programming exercises
1. Following the logic of the MoveTower function, write a recursive function
NHanoiMoves(n) that calculates the number of individual moves required to solve
the Tower of Hanoi puzzle for n disks.
2. To make the operation of the program somewhat easier to explain, the
implementation of MoveTower in this chapter uses
if (n == 1)
as its simple case test. Whenever you see a recursive program use 1 as its simple
case, it pays to be a little skeptical; in most applications, 0 is a more appropriate
choice. Rewrite the Tower of Hanoi program so that the MoveTower function checks
whether n is 0 instead. What happens to the length of the M o v e T o w e r
implementation?
3. Rewrite the Tower of Hanoi program so that it uses an explicit stack of pending tasks
instead of recursion. In this context, a task can be represented most easily as a
structure containing the number of disks to move and the names of the spires used
for the start, finish, and temporary repositories. At the beginning of the process, you
push onto your stack a single task that describes the process of moving the entire
tower. The program then repeatedly pops the stack and executes the task found there
until no tasks are left. Except for the simple cases, the process of executing a task
results in the creation of more tasks that get pushed onto the stack for later execution.
4. As presented in the text, the function RecursivePermute takes two strings, which
indicate how far the permutation process has progressed. You could also design the
program so that RecursivePermute takes a string and an integer. where the string is
the concatenation of the fixed prefix with a suffix whose characters can still be
permuted. The integer indicates the number of characters in the prefix and thus the
index at which the remaining characters begin within the string. For example, if you
call the redesigned RecursivePermute function on the arguments "ABCD" and 2, the
output should be
LPEX1
ABCD
ABDC

which is all strings beginning with "AB" followed by some permutation of "CD".
Rewrite the permutation program so that it uses this new design.
5. Update the permutation algorithm from the text to generate the correct list of
permutations even if the string contains repeated letters. For example, if you call
ListPermutations on the string "AABB" , your program should not generate as
many permutations as it does for the string "ABCD" because some of the strings
Recursive Procedures – 227 –

generated by the standard algorithm would be indistinguishable from others. Your

program should instead generate the following six:
LPEX2
AABB
ABAB
ABBA
BAAB
BABA
BBAA

Write a new implementation of ListPermutations that works correctly even if the

string contains duplicated letters. In writing this implementation, you should not
merely keep a list of the permutations that have already been encountered and avoid
generating duplicates. Instead, you should think carefully about the recursive
structure of the problem and find a way to avoid generating the extra permutations in
the first place.
6. On a telephone keypad, the digits are mapped onto the alphabet as shown in the
diagram below:

ABC DEF
1 2 3

GHI JKL MNO

4 5 6

PQRS TUV WXYZ

7 8 9

* 0 #

In order to make their phone numbers more memorable, service providers like to find
numbers that spell out some word (called a mnemonic) appropriate to their business
that makes that phone number easier to remember. For example, the phone number
for a recorded time-of-day message in some localities is 637-8687 (NERVOUS).
Imagine that you have just been hired by a local telephone company to write a
function ListMnemonics that will generate all possible letter combinations that
correspond to a given number, represented as a string of digits. For example, the call
ListMnemonics("723")
should generate the following 36 possible letter combinations that correspond to that
prefix:
PAD PBD PCD QAD QBD QCD RAD RBD RCD SAD SBD SCD
PAE PBE PCE QAE QBE QCE RAE RBE RCE SAE SBE SCE
PAF PBF PCF QAF QBF QCF RAF RBF RCF SAF SBF SCF

7. Rewrite the program from exercise 6 so that it uses the Lexicon class and the
EnglishWords.dat file from Chapter 4 so that the program only lists mnemonics
that are valid English words.
Recursive Procedures – 228 –

8. Using the ListPermutations example as a starting point, write a function

ListSubsets that generates all possible subsets of a given set, where the set is
represented by a string of letters. For example, if you call the function

ListSubsets("ABC");

your function should produce the following output:

Subsets
This program lists all subsets of a set.
Enter a string representing a set: ABC
{ABC}
{AB}
{AC}
{A}
{BC}
{B}
{C}
{}

Like permutations, the subset problem has a recursive formulation. If you

represent a set of characters using a string that either contains or does not contain a
given letter, you can calculate all possible subsets by (1) including the first character
in the subset and concatenating it onto the front of all subsets of the remaining N–1
characters and then (2) displaying the subsets of the remaining N–1 without this
character.

9. Inside a computer system, integers are represented as a sequence of bits, each of

which is a single digit in the binary number system and can therefore have only the
value 0 or 1. With N bits, you can represent 2N distinct integers. For example, three
bits are sufficient to represent the eight (23) integers between 0 and 7, as follows:

000 → 0
001 → 1
010 → 2
011 → 3
100 → 4
101 → 5
110 → 6
111 → 7

Each entry in the left side of the table is written in its standard binary representation,
in which each bit position counts for twice as much as the position to its right. For
instance, you can demonstrate that the binary value 110 represents the decimal
number 6 by following the logic shown in the following diagram:

place value 4 2 1
x x x
binary digits 1 1 0
=

4 + 2 + 0 = 6
Recursive Procedures – 229 –

Write a recursive function GenerateBinaryCode(nBits) that generates the bit

patterns for the standard binary representation of all integers that can be represented
using the specified number of bits. For example, calling GenerateBinaryCode(3)
should produce the following output:
GenerateBinaryCode
000
001
010
011
100
101
110
111

10. Although the binary coding used in exercise 7 is ideal for most applications, it has
certain drawbacks. As you count in standard binary notation, there are some points
in the sequence at which several bits change at the same time. For example, in the
three-bit binary code, the value of every bit changes as you move from 3 (011) to 4
(100).
In some applications, this instability in the bit patterns used to represent adjacent
numbers can lead to problems. Imagine for the moment that you are using some
hardware measurement device that produces a three-bit value from some real-world
phenomenon that happens to be varying between 3 and 4. Sometimes, the device
will register 011 to indicate the value 3; at other times, it will register 100 to indicate
4. For this device to work correctly, the transitions for each of the individual bits
must occur simultaneously. If the first bit changes more quickly than the others, for
example, there may be an intermediate state in which the device reads 111, which
would be a highly inaccurate reading.
It is interesting to discover that you can avoid this problem simply by changing
the numbering system. If instead of using binary representation in the traditional
way, you can assign three-bit values to each of the numbers 0 through 7 with the
highly useful property that only one bit changes in the representation between every
pair of adjacent integers. Such an encoding is called a Gray code (after its inventor,
the mathematician Frank Gray) and looks like this:
000 → 0
001 → 1
011 → 2
010 → 3
110 → 4
111 → 5
101 → 6
100 → 7
Note that, in the Gray code representation, the bit patterns for 3 and 4 differ only in
their leftmost bit. If the hardware measurement device used Gray codes, a value
oscillating between 3 and 4 would simply turn that bit on and off, eliminating any
problems with synchronization.
The recursive insight that you need to create a Gray code of N bits is summarized
in the following informal procedure:
Recursive Procedures – 230 –

1. Write down the Gray code for N – 1 bits.

2. Copy that same list in reverse order below the original one.
3. Add a 0 bit in front of the codings in the original half of the list and a 1 bit in
front of those in the reversed copy.
This procedure is illustrated in the following derivation of the Gray code for three
bits:
3-bit Gray code 2-bit Gray code 1-bit Gray code

0 0 0
0 0 1
0 1 1 0 0
0 1 0 0 1 0
1 1 0 1 1 reversed
1
1 1 1 reversed 1 0
1 0 1
1 0 0
Write a recursive function GenerateGrayCode(nBits) that generates the Gray
code patterns for the specified number of bits.
11. Given a set of numbers, the partition problem is to find a subset of the numbers that
add up to a specific target number. For example, there are two ways to partition the
set {1, 3, 4, 5} so that the remaining elements add up to 5:
• Select the 1 and the 4
• Select just the 5
By contrast, there is no way to partition the set {1, 3, 4, 5} to get 11.
Write a function NumberOfPartitions that takes an array of integers, the length
of that array, and a target number, and returns the number of partitions of that set of
integers which add up to the target. For example, suppose that the array sampleSet
has been initialized as follows:

int sampleSet[] = {1, 3, 4, 5};

Given this definition of sampleSet, calling

NumberOfPartitions(sampleSet, 4, 5);

should return 2 (there are two ways to make 5), and calling

NumberOfPartitions(sampleSet, 4, 11)

should return 0 (there are no ways to make 11).

The prototype for NumberOfPartitions is
Recursive Procedures – 231 –

int NumberOfPartitions(int array[], int length, int target);

In order to see the recursive nature of this problem, think about any specific element
in the set, such as the first element. If you think about all the partitions of a
particular target number, some of them will include the first element and some
won’t. If you count those that do include the first element and then add that total to
the number of those which leave out that element, you get the total number of
partitions. Each of these two computations, however, can be expressed as a problem
in the same form as the outer partition problem and can therefore be solved
recursively.

12. I am the only child of parents who weighed, measured, and priced
everything; for whom what could not be weighed, measured, and
priced had no existence.
—Charles Dickens, Little Dorrit, 1857

In Dickens’s time, merchants measured many commodities using weights and a two-
pan balance—a practice that continues in many parts of the world today. If you are
using a limited set of weights, however, you can only measure certain quantities
accurately.
For example, suppose that you have only two weights: a 1-ounce weight and a 3-
ounce weight. With these you can easily measure out 4 ounces, as shown:

1 3

It is somewhat more interesting to discover that you can also measure out 2 ounces
by shifting the 1-ounce weight to the other side, as follows:

1 3

Write a recursive function

bool IsMeasurable(int target, int weights[], int nWeights)

that determines whether it is possible to measure out the desired target amount with a
given set of weights. The available weights are stored in the array weights, which
has nWeights as its effective size. For instance, the sample set of two weights
illustrated above could be represented using the following pair of variables:

int sampleWeights[] = { 1, 3 };
int nSampleWeights = 2;

Given these values, the function call

Recursive Procedures – 232 –

IsMeasurable(2, sampleWeights, nSampleWeights)

should return true because it is possible to measure out 2 ounces using the sample
weight set as illustrated in the preceding diagram. On the other hand, calling
IsMeasurable(5, sampleWeights, nSampleWeights)
should return false because it is impossible to use the 1- and 3-ounce weights to add
up to 5 ounces.
The fundamental observation you need to make for this problem is that each
weight in the array can be either
1. Put on the opposite side of the balance from the sample
2. Put on the same side of the balance as the sample
3. Left off the balance entirely
If you consider one of the weights in the array and determine how choosing one of
these three options affects the rest of the problem, you should be able to come up
with the recursive insight you need to solve the problem.
13. In countries like the United States that still use the traditional English system of
measurement, each inch on a ruler is marked off into fractions using tick marks that
look like this:

The longest tick mark falls at the half-inch position, two smaller tick marks indicate
the quarter inches, and even smaller ones are used to mark the eighths and sixteenths.
Write a recursive program that draws a 1-inch line at the center of the graphics
window and then draws the tick marks shown in the diagram. Assume that the
length of the tick mark indicating the half-inch position is given by the constant
definition

const double HALF_INCH_TICK = 0.2;

and that each smaller tick mark is half the size of the next larger one.

14. One of the reasons that fractals have generated so much interest is that they turn out
to be useful in some surprising practical contexts. For example, the most successful
techniques for drawing computer images of mountains and certain other landscape
features involve using fractal geometry.
As a simple example of where this issue comes up, consider the problem of
connecting two points A and B with a fractal that looks like a coastline on a map.
The simplest possible strategy would be to draw a straight line between the two
points:

A B

This is the order 0 coastline and represents the base case of the recursion.
Of course, a real coastline will have small peninsulas or inlets somewhere along
its length, so you would expect a realistic drawing of a coastline to jut in or out
occasionally like a real one. As a first approximation, you could replace the straight
Recursive Procedures – 233 –

line with precisely the same fractal line used to create the Koch snowflake in the
program described in the section on “Fractals” earlier in the chapter, as follows:

A B
This process gives the order 1 coastline. However, in order to give the feeling of a
traditional coastline, it is important for the triangular wedge in this line sometimes to
point up and sometimes down, with equal probability.
If you then replace each of the straight line segments in the order 1 fractal with a
fractal line in a random direction, you get the order 2 coastline, which might look
like this:

A B

Continuing this process eventually results in a drawing that conveys a remarkably

realistic sense, as in this order 5 coastline:

Write a function DrawCoastline that fits the following interface description:

/*
* Function: DrawCoastline
* Usage: DrawCoastline(length, theta, order);
* -------------------------------------------
* The DrawCoastline function starts at the current (x, y)
* position and draws a fractal coastline of the specified
* length moving in the direction given by the angle theta
* (as defined in the definition of DrawPolarLine in the
* preceding problem); order gives the number of recursive
* subdivisions into which each segment will be divided.
*/

void DrawCoastline(double length, double theta, int order);

15. Recursive decomposition can also be used to draw a stylized representation of a tree.
The tree begins as a simple trunk indicated by a straight vertical line, as follows:
Recursive Procedures – 234 –

The trunk may branch at the top to form two lines that veer off at an angle, as shown:

These branches may themselves split to form new branches, which split to form new
ones, and so on. If the decision to branch is made randomly at each step of the
process, the tree will eventually become unsymmetrical and will end up looking a
little more like trees in nature, as illustrated by the following diagram:

If you think about this process recursively, however, you can see that all trees
constructed in this way consist of a trunk, optionally topped by two trees that veer
off at an angle. If the probability of branching is a function of the length of the
current branch, the process will eventually terminate as the branches get
progressively shorter.
Write a program drawtree.cpp that uses this recursive strategy and the graphics
library to draw a stylized line drawing of a tree.
Chapter 7
Backtracking Algorithms

Truth is not discovered by proofs but by exploration. It is

always experimental.
— Simone Weil, The New York Notebook, 1942
Backtracking Algorithms – 236 –

For many real-world problems, the solution process consists of working your way
through a sequence of decision points in which each choice leads you further along some
path. If you make the correct set of choices, you end up at the solution. On the other
hand, if you reach a dead end or otherwise discover that you have made an incorrect
choice somewhere along the way, you have to backtrack to a previous decision point and
try a different path. Algorithms that use this approach are called backtracking
algorithms.
If you think about a backtracking algorithm as the process of repeatedly exploring
paths until you encounter the solution, the process appears to have an iterative character.
As it happens, however, most problems of this form are easier to solve recursively. The
fundamental recursive insight is simply this: a backtracking problem has a solution if and
only if at least one of the smaller backtracking problems that results from making each
possible initial choice has a solution. The examples in this chapter are designed to
illustrate this process and demonstrate the power of recursion in this domain.

7.1 Solving a maze by recursive backtracking

Once upon a time, in the days of Greek mythology, the Mediterranean island of Crete was
ruled by a tyrannical king named Minos. From time to time, Minos demanded tribute
from the city of Athens in the form of young men and women, whom he would sacrifice
to the Minotaur—a fearsome beast with the head of a bull and the body of a man. To
house this deadly creature, Minos forced his servant Daedelus (the engineering genius
who later escaped the island by constructing a set of wings) to build a vast underground
labyrinth at Knossos. The young sacrifices from Athens would be led into the labyrinth,
where they would be eaten by the Minotaur before they could find their way out. This
tragedy continued until young Theseus of Athens volunteered to be one of the sacrifices.
Following the advice of Minos’s daughter Ariadne, Theseus entered the labyrinth with a
sword and a ball of string. After slaying the monster, Theseus was able to find his way
back to the exit by unwinding the string as he went along.

The right-hand rule

Theseus’s strategy represents an algorithm for escaping from a maze, but not everyone in
such a predicament is lucky enough to have a ball of string or an accomplice clever
enough to suggest such an effective approach. Fortunately, there are other strategies for
escaping from a maze. Of these strategies, the best known is called the right-hand rule,
which can be expressed in the following pseudocode form:

Put your right hand against a wall.

while (you have not yet escaped from the maze) {
Walk forward keeping your right hand on a wall.
}

As you walk, the requirement that you keep your right hand touching the wall may force
you to turn corners and occasionally retrace your steps. Even so, following the right-
hand rule guarantees that you will always be able to find an opening to the outside of any
maze.
To visualize the operation of the right-hand rule, imagine that Theseus has successfully
dispatched the Minotaur and is now standing in the position marked by the first character
in Theseus’s name, the Greek letter theta (Θ):
Backtracking Algorithms – 237 –

If Theseus puts his right hand on the wall and then follows the right-hand rule from there,
he will trace out the path shown by the dashed line in this diagram:

Θ
Finding a recursive approach
As the while loop in its pseudocode form makes clear, the right-hand rule is an iterative
strategy. You can, however, also think about the process of solving a maze from a
recursive perspective. To do so, you must adopt a different mindset. You can no longer
think about the problem in terms of finding a complete path. Instead, your goal is to find
a recursive insight that simplifies the problem, one step at a time. Once you have made
the simplification, you use the same process to solve each of the resulting subproblems.
Let’s go back to the initial configuration of the maze shown in the illustration of the
right-hand rule. Put yourself in Theseus’s position. From the initial configuration, you
have three choices, as indicated by the arrows in the following diagram:

The exit, if any, must lie along one of those paths. Moreover, if you choose the correct
direction, you will be one step closer to the solution. The maze has therefore become
Backtracking Algorithms – 238 –

simpler along that path, which is the key to a recursive solution. This observation
suggests the necessary recursive insight. The original maze has a solution if and only if it
is possible to solve at least one of the new mazes shown in Figure 7-1. The × in each
diagram marks the original starting square and is off-limits for any of the recursive
solutions because the optimal solution will never have to backtrack through this square.
By looking at the mazes in Figure 7-1, it is easy to see—at least from your global
vantage point—that the submazes labeled (a) and (c) represent dead-end paths and that
the only solution begins in the direction shown in the submaze (b). If you are thinking
recursively, however, you don’t need to carry on the analysis all the way to the solution.
You have already decomposed the problem into simpler instances. All you need to do is
rely on the power of recursion to solve the individual subproblems, and you’re home free.
You still have to identify a set of simple cases so that the recursion can terminate, but the
hard work has been done.

Identifying the simple cases

What constitutes the simple case for a maze? One possibility is that you might already be
standing outside the maze. If so, you’re finished. Clearly, this situation represents one
simple case. There is, however, another possibility. You might also reach a blind alley
where you’ve run out of places to move. For example, if you try to solve the sample
maze by moving north and then continue to make recursive calls along that path, you will
eventually be in the position of trying to solve the following maze:

× ×
Θ ×
×

At this point, you’ve run out of room to maneuver. Every path from the new position is
either marked or blocked by a wall, which makes it clear that the maze has no solution
from this point. Thus, the maze problem has a second simple case in which every
direction from the current square is blocked, either by a wall or a marked square.

Figure 7-1 Recursive decomposition of a maze

Θ
× × Θ ×
Θ

(a) (b) (c)

Backtracking Algorithms – 239 –

It is easier to code the recursive algorithm if, instead of checking for marked squares as
you consider the possible directions of motion, you go ahead and make the recursive calls
on those squares. If you check at the beginning of the procedure to see whether the
current square is marked, you can terminate the recursion at that point. After all, if you
find yourself positioned on a marked square, you must be retracing your path, which
means that the solution must lie in some other direction.
Thus, the two simple cases for this problem are as follows:
1. If the current square is outside the maze, the maze is solved.
2. If the current square is marked, the maze is unsolvable.

Coding the maze solution algorithm

Although the recursive insight and the simple cases are all you need to solve the problem
on a conceptual level, writing a complete program to navigate a maze requires you to
consider a number of implementation details as well. For example, you need to decide on
a representation for the maze itself that allows you, for example, to figure out where the
walls are, keep track of the current position, indicate that a particular square is marked,
and determine whether you have escaped from the maze. While designing an appropriate
data structure for the maze is an interesting programming challenge in its own right, it has
very little to do with understanding the recursive algorithm, which is the focus of this
discussion. If anything, the details of the data structure are likely to get in the way and
make it more difficult for you to understand the algorithmic strategy as a whole.
Fortunately, it is possible to put such details aside by introducing a new abstraction
layer. The purpose of the abstraction is to provide the main program with access to the
information it needs to solve a maze, even though the details are hidden. An interface that
provides the necessary functionality is the mazelib.h interface shown in Figure 7-2.
Once you have access to the mazelib.h interface, writing a program to solve a maze
becomes much simpler. The essence of the problem is to write a function SolveMaze that
uses recursive backtracking to solve a maze whose specific characteristics are maintained
by the mazelib module. The argument to SolveMaze is the starting position, which
changes for each of the recursive subproblems. To ensure that the recursion can
terminate when a solution is found, the SolveMaze function must also return some
indication of whether it has succeeded. The easiest way to convey this information is to
define SolveMaze as a predicate function that returns true if a solution has been found,
and false otherwise. Thus, the prototype for SolveMaze looks like this:
bool SolveMaze(pointT pt);

Given this definition, the main program is simply

int main() {
ReadMazeMap(MazeFile);
if (SolveMaze(GetStartPosition())) {
cout << "The marked squares show a solution path." << endl;
} else {
cout << "No solution exists." << endl;
}
return 0;
}
Backtracking Algorithms – 240 –

Figure 7-2 The mazelib.h interface

/*
* File: mazelib.h
* ---------------
* This interface provides a library of primitive operations
* to simplify the solution to the maze problem.
*/

#ifndef _mazelib_h
#define _mazelib_h

#include "genlib.h"

/*
* Type: directionT
* ----------------
* This type is used to represent the four compass directions.
*/

enum directionT { North, East, South, West };

/*
* Type: pointT
* ------------
* The type pointT is used to encapsulate a pair of integer
* coordinates into a single value with x and y components.
*/

struct pointT {
int x, y;
};

/*
* Function: ReadMazeMap
* Usage: ReadMazeMap(filename);
* -----------------------------
* This function reads in a map of the maze from the specified
* file and stores it in private data structures maintained by
* this module. In the data file, the characters '+', '-', and
* '|' represent corners, horizontal walls, and vertical walls,
* respectively; spaces represent open passageway squares. The
* starting position is indicated by the character 'S'. For
* example, the following data file defines a simple maze:
*
* +-+-+-+-+-+
* | |
* + +-+ + +-+
* |S | |
* +-+-+-+-+-+
*
* Coordinates in the maze are numbered starting at (0,0) in
* the lower left corner. The goal is to find a path from
* the (0,0) square to the exit east of the (4,1) square.
*/

void ReadMazeMap(string filename);

Backtracking Algorithms – 241 –

/*
* Function: GetStartPosition
* Usage: pt = GetStartPosition();
* -------------------------------
* This function returns a pointT indicating the coordinates of
* the start square.
*/

pointT GetStartPosition();

/*
* Function: OutsideMaze
* Usage: if (OutsideMaze(pt)) . . .
* ---------------------------------
* This function returns true if the specified point is outside
* the boundary of the maze.
*/

bool OutsideMaze(pointT pt);

/*
* Function: WallExists
* Usage: if (WallExists(pt, dir)) . . .
* -------------------------------------
* This function returns true if there is a wall in the indicated
* direction from the square at position pt.
*/

bool WallExists(pointT pt, directionT dir);

/*
* Functions: MarkSquare, UnmarkSquare, IsMarked
* Usage: MarkSquare(pt);
* UnmarkSquare(pt);
* if (IsMarked(pt)) . . .
* ------------------------------
* These functions mark, unmark, and test the status of the
* square specified by the coordinates pt.
*/

void MarkSquare(pointT pt);

void UnmarkSquare(pointT pt);
bool IsMarked(pointT pt);

#endif
Backtracking Algorithms – 242 –

The code for the SolveMaze function itself turns out to be extremely short and is
shown in Figure 7-3. The entire algorithm fits into approximately 10 lines of code with
the following pseudocode structure:
If the current square is outside the maze, return true to indicate that a solution has been found.
If the current square is marked, return false to indicate that this path has already been tried.
Mark the current square.
for (each of the four compass directions) {
if (this direction is not blocked by a wall) {
Move one step in the indicated direction from the current square.
Try to solve the maze from there by making a recursive call.
If this call shows the maze to be solvable, return true to indicate that fact.
}
}
Unmark the current square.
Return false to indicate that none of the four directions led to a solution.

The only function called by SolveMaze that is not exported by the mazelib.h
interface is the function AdjacentPoint(pt, dir), which returns the coordinates of the
square that is one step away from pt in the direction dir . The following is a simple
implementation of AdjacentPoint that copies the original point and then adjusts the
appropriate coordinate value:

Figure 7-3 The SolveMaze function

/*
* Function: SolveMaze
* Usage: if (SolveMaze(pt)) . . .
* -------------------------------
* This function attempts to generate a solution to the current
* maze from point pt. SolveMaze returns true if the maze has
* a solution and false otherwise. The implementation uses
* recursion to solve the submazes that result from marking the
* current square and moving one step along each open passage.
*/

bool SolveMaze(pointT pt) {

if (OutsideMaze(pt)) return true;
if (IsMarked(pt)) return false;
MarkSquare(pt);
for (int i = 0; i < 4; i++) {
directionT dir = directionT(i);
if (!WallExists(pt, dir)) {
if (SolveMaze(AdjacentPoint(pt, dir))) {
return true;
}
}
}
UnmarkSquare(pt);
return false;
}
Backtracking Algorithms – 243 –

pointT AdjacentPoint(pointT pt, directionT dir) {

pointT newpt = pt;
switch (dir) {
case North: newpt.y++; break;
case East: newpt.x++; break;
case South: newpt.y--; break;
case West: newpt.x--; break;;
}
return newpt;
}

The code to unmark the current square at the end of the for loop is not strictly
necessary in this implementation and in fact can reduce the performance of the algorithm
if there are loops in the maze (see exercise 3). The principal advantage of including it is
that doing so means that the solution path ends up being recorded by a chain of marked
squares from the original starting position to the exit. If you are using a graphical
implementation of this algorithm, erasing the marks as you retreat down a path makes it
much easier to see the current path.

Convincing yourself that the solution works

In order to use recursion effectively, at some point you must be able to look at a recursive
function like the SolveMaze example in Figure 7-3 and say to yourself something like
this: “I understand how this works. The problem is getting simpler because more squares
are marked each time. The simple cases are clearly correct. This code must do the job.”
For most of you, however, that confidence in the power of recursion will not come easily.
Your natural skepticism makes you want to see the steps in the solution. The problem is
that, even for a maze as simple as the one shown earlier in this chapter, the complete
history of the steps involved in the solution is far too large to think about comfortably.
Solving that maze, for example, requires 66 calls to SolveMaze that are nested 27 levels
deep when the solution is finally discovered. If you attempt to trace the code in detail,
you will inevitably get lost.
If you are not yet ready to accept the recursive leap of faith, the best you can do is
track the operation of the code in a more general sense. You know that the code first tries
to solve the maze by moving one square to the north, because the for loop goes through
the directions in the order defined by the directionT enumeration. Thus, the first step in
the solution process is to make a recursive call that starts in the following position:

Θ
×

At this point, the same process occurs again. The program again tries to move north
and makes a new recursive call in the following position:
Backtracking Algorithms – 244 –

Θ
×
×

At this level of the recursion, moving north is no longer possible, so the for loop cycles
through the other directions. After a brief excursion southward, upon which the program
encounters a marked square, the program finds the opening to the west and proceeds to
generate a new recursive call. The same process occurs in this new square, which in turn
leads to the following configuration:

× ×
Θ ×
×

In this position, none of the directions in the for loop do any good; every square is
either blocked by a wall or already marked. Thus, when the for loop at this level exits at
the bottom, it unmarks the current square and returns to the previous level. It turns out
that all the paths have also been explored in this position, so the program once again
unmarks the square and returns to the next higher level in the recursion. Eventually, the
program backtracks all the way to the initial call, having completely exhausted the
possibilities that begin by moving north. The for loop then tries the eastward direction,
finds it blocked, and continues on to explore the southern corridor, beginning with a
recursive call in the following configuration:

×
Θ

From here on, the same process ensues. The recursion systematically explores every
corridor along this path, backing up through the stack of recursive calls whenever it
reaches a dead end. The only difference along this route is that eventually—after
Backtracking Algorithms – 245 –

descending through an additional recursive level for every step on the path—the program
makes a recursive call in the following position:

× × × ×
× × × ×
× × ×
× × ×
× × × × ×
× × × ×
× × ×
Θ

At this point, Theseus is outside the maze. The simple case kicks in and returns true to
its caller. This value is then propagated back through all 27 levels of the recursion, at
which point the original call to SolveMaze returns to the main program.

7.2 Backtracking and games

Although backtracking is easiest to illustrate in the context of a maze, the strategy is
considerably more general. For example, you can apply backtracking to most two-player
strategy games. The first player has several choices for an initial move. Depending on
which move is chosen, the second player then has a particular set of responses. Each of
these responses leads in turn to new options for the first player, and this process continues
until the end of the game. The different possible positions at each turn in the game form
a branching structure in which each option opens up more and more possibilities.
If you want to program a computer to take one side of a two-player game, one
approach is simply to have the computer follow all the branches in the list of possibilities.
Before making its first move, the computer would try every possible choice. For each of
these choices, it would then try to determine what its opponent’s response would be. To
do so, it would follow the same logic: try every possibility and evaluate the possible
counterplays. If the computer can look far enough ahead to discover that some move
would leave its opponent in a hopeless position, it should make that move.
In theory, this strategy can be applied to any two-player strategy game. In practice, the
process of looking at all the possible moves, potential responses, responses to those
responses, and so on requires too much time and memory, even for modern computers.
There are, however, several interesting games that are simple enough to solve by looking
at all the possibilities, yet complex enough so that the solution is not immediately obvious
to the human player.

The game of nim

To see how recursive backtracking applies to two-player games, it helps to consider a
simple example such as the game of nim. The word nim actually applies to a large class
of games in which players take turns removing objects from some initial configuration.
In this particular version, the game begins with a pile of 13 coins in the center of a table.
On each turn, players take either one, two, or three coins from the pile and put them
aside. The object of the game is to avoid being forced to take the last coin. Figure 7-4
shows a sample game between the computer and a human player.
Backtracking Algorithms – 246 –

Figure 7-4 Sample game of nim

Nim
Hello. Welcome to the game of nim.
In this game, we will start with a pile of
13 coins on the table. On each turn, you
and I will alternately take between 1 and
3 coins from the table. The player who
takes the last coin loses.
There are 13 coins in the pile.
How many would you like? 2
There are 11 coins in the pile.
I'll take 2.
There are 9 coins in the pile.
How many would you like? 3
There are 6 coins in the pile.
I'll take 1.
There are 5 coins in the pile.
How many would you like? 1
There are 4 coins in the pile.
I'll take 3.
There is only one coin left.
I win.

How would you go about writing a program to play a winning game of nim? The
mechanical aspects of the game—keeping track of the number of coins, asking the player
for a legal move, determining the end of the game, and so forth—are a straightforward
programming task. The interesting part of the program consists of figuring out how to
give the computer a strategy for playing the best possible game.
Finding a successful strategy for nim is not particularly hard, particularly if you work
backward from the end of the game. The rules of nim state that the loser is the player
who takes the last coin. Thus, if you ever find yourself with just one coin on the table,
you’re in a bad position. You have to take that coin and lose. On the other hand, things
look good if you find yourself with two, three, or four coins. In any of these cases, you
can always take all but one of the remaining coins, leaving your opponent in the
unenviable position of being stuck with just one coin. But what if there are five coins on
the table? What can you do then? After a bit of thought, it’s easy to see that you’re also
doomed if you’re left with five coins. No matter what you do, you have to leave your
opponent with two, three, or four coins—situations that you’ve just discovered represent
good positions from your opponent’s perspective. If your opponent is playing
intelligently, you will surely be left with a single coin on your next turn. Since you have
no good moves, being left with five coins is clearly a bad position.
This informal analysis reveals an important insight about the game of nim. On each
turn, you are looking for a good move. A move is good if it leaves your opponent in a
bad position. But what is a bad position? A bad position is one in which there is no good
move. Although these definitions of good move and bad position are circular, they
nonetheless constitute a complete strategy for playing a perfect game of nim. All you
have to do is rely on the power of recursion. If you have a function FindGoodMove that
takes the number of coins as its argument, all it has to do is try every possibility, looking
for one that leaves a bad position for the opponent. You can then assign the job of
determining whether a particular position is bad to the predicate function
IsBadPosition, which calls FindGoodMove to see if there is one. The two functions call
each other back and forth, evaluating all possible branches as the game proceeds.
The FindGoodMove function has the following pseudocode formulation:
Backtracking Algorithms – 247 –

int FindGoodMove(int nCoins) {

for (each possible move) {
Evaluate the position that results from making that move.
If the resulting position is bad, return that move.
}
Return a sentinel value indicating that no good move exists.
}

The legal values returned by FindGoodMove are 1, 2, and 3. The sentinel indicating that
no good move exists can be any integer value outside that range. For example, you can
define the constant NO_GOOD_MOVE as follows:

const int NO_GOOD_MOVE = -1;

The code for FindGoodMove then looks like this:

int FindGoodMove(int nCoins) {

for (int nTaken = 1; nTaken <= MAX_MOVE; nTaken++) {
if (IsBadPosition(nCoins - nTaken)) return nTaken;
}
return NO_GOOD_MOVE;
}

The code for the IsBadPosition function is even easier. After checking for the
simple case that occurs when there is only a single coin to take, the function simply calls
FindGoodMove to see if a good move exists. The code for IsBadPosition is therefore
simply

bool IsBadPosition(int nCoins) {

if (nCoins == 1) return true;
return FindGoodMove(nCoins) == NO_GOOD_MOVE;
}

This function encapsulates the following ideas:

• Being left with a single coin indicates a bad position.

• A position is bad if there are no good moves.

The functions FindGoodMove and IsBadPosition provide all the strategy that the nim
program needs to play a perfect game. The rest of the program just takes care of the
mechanics of playing nim with a human player, as shown in Figure 7-5.

A generalized program for two-player games

The code for nim shown in Figure 7-5 is highly specific to that game. The
FindGoodMove function, for example, is written so that it incorporates directly into the
structure of the code the knowledge that the computer may take one, two, or three coins.
The basic idea, however, is far more general. Many two-player games can be solved
using the same overall strategy, even though different games clearly require different
codings of the details.
Backtracking Algorithms – 248 –

Figure 7-5 Program to play the game of nim

/*
* File: nim.cpp
* -------------
* This program simulates a simple variant of the game of nim.
* In this version, the game starts with a pile of 13 coins
* on a table. Players then take turns removing 1, 2, or 3
* coins from the pile. The player who takes the last coin
* loses. This simulation allows a human player to compete
* against the computer.
*/

#include "genlib.h"
#include "simpio.h"
#include <iostream>

/*
* Constants
* ---------
* N_COINS -- Initial number of coins
* MAX_MOVE -- The maximum number of coins a player may take
* NO_GOOD_MOVE -- Sentinel indicating no good move is available
*/

const int N_COINS = 13;

const int MAX_MOVE = 3;
const int NO_GOOD_MOVE = -1;

/*
* Type: playerT
* -------------
* This enumeration type distinguishes the turns for the human
* player from those for the computer.
*/

enum playerT { Human, Computer };

/* Private function prototypes */

void GiveInstructions();
void AnnounceWinner(int nCoins, playerT whoseTurn);
int GetUserMove(int nCoins);
bool MoveIsLegal(int nTaken, int nCoins);
int ChooseComputerMove(int nCoins);
int FindGoodMove(int nCoins);
bool IsBadPosition(int nCoins);
Backtracking Algorithms – 249 –

/*
* Main program
* ------------
* This program plays the game of nim. In this implementation,
* the human player always goes first.
*/

int main() {
int nCoins, nTaken;
playerT whoseTurn;

GiveInstructions();
nCoins = N_COINS;
whoseTurn = Human;
while (nCoins > 1) {
cout << "There are " << nCoins << " coins in the pile."<<endl;
switch (whoseTurn) {
case Human:
nTaken = GetUserMove(nCoins);
whoseTurn = Computer;
break;
case Computer:
nTaken = ChooseComputerMove(nCoins);
cout << "I'll take " << nTaken << "." << endl;
whoseTurn = Human;
break;
}
nCoins -= nTaken;
}
AnnounceWinner(nCoins, whoseTurn);
return 0;
}

/*
* Function: GiveInstructions
* Usage: GiveInstructions();
* --------------------------
* This function explains the rules of the game to the user.
*/

void GiveInstructions() {
cout << "Hello. Welcome to the game of nim." << endl;
cout << "In this game, we will start with a pile of" << endl;
cout << N_COINS << " coins on the table. " << endl;
cout << "On each turn, you" << endl;
cout << "and I will alternately take between 1 and" << endl;
cout << MAX_MOVE << " coins from the table." << endl;
cout << "The player who" << endl;
cout << "takes the last coin loses." << endl;
cout << endl;
}
Backtracking Algorithms – 250 –

/*
* Function: AnnounceWinner
* Usage: AnnounceWinner(nCoins, whoseTurn);
* -----------------------------------------
* This function announces the final result of the game.
*/

void AnnounceWinner(int nCoins, playerT whoseTurn) {

if (nCoins == 0) {
cout << "You took the last coin. You lose." << endl;
} else {
cout << "There is only one coin left." << endl;
switch (whoseTurn) {
case Human: cout << "I win." << endl; break;
case Computer: cout << "I lose." << endl; break;
}
}
}

/*
* Function: GetUserMove
* Usage: nTaken = GetUserMove(nCoins);
* ------------------------------------
* This function is responsible for the human player's turn.
* It takes the number of coins left in the pile as an argument,
* and returns the number of coins that the player removes
* from the pile. The function checks the move for legality
* and gives the player repeated chances to enter a legal move.
*/

int GetUserMove(int nCoins) {

int nTaken, limit;
while (true) {
cout << "How many would you like? ";
nTaken = GetInteger();
if (MoveIsLegal(nTaken, nCoins)) break;
limit = (nCoins < MAX_MOVE) ? nCoins : MAX_MOVE;
cout << "That's cheating! Please choose a number";
cout << " between 1 and " << limit << endl;
cout << "There are " << nCoins << " coins in the pile."<<endl;
}
return nTaken;
}

/*
* Function: MoveIsLegal
* Usage: if (MoveIsLegal(nTaken, nCoins)) . . .
* ---------------------------------------------
* This predicate function returns true if it is legal to take
* nTaken coins from a pile of nCoins.
*/

bool MoveIsLegal(int nTaken, int nCoins) {

return (nTaken > 0 && nTaken <= MAX_MOVE && nTaken <= nCoins);
}
Backtracking Algorithms – 251 –

/*
* Function: ChooseComputerMove
* Usage: nTaken = ChooseComputerMove(nCoins);
* -------------------------------------------
* This function figures out what move is best for the computer
* player and returns the number of coins taken. The function
* first calls FindGoodMove to see if a winning move exists.
* If none does, the program takes only one coin to give the
* human player more chances to make a mistake.
*/

int ChooseComputerMove(int nCoins) {

int nTaken = FindGoodMove(nCoins);
if (nTaken == NO_GOOD_MOVE) nTaken = 1;
return nTaken;
}

/*
* Function: FindGoodMove
* Usage: nTaken = FindGoodMove(nCoins);
* -------------------------------------
* This function looks for a winning move, given the specified
* number of coins. If there is a winning move in that
* position, the function returns that value; if not, the
* function returns the constant NoWinningMove. This function
* depends on the recursive insight that a good move is one
* that leaves your opponent in a bad position and a bad
* position is one that offers no good moves.
*/

int FindGoodMove(int nCoins) {

for (int nTaken = 1; nTaken <= MAX_MOVE; nTaken++) {
if (IsBadPosition(nCoins - nTaken)) return nTaken;
}
return NO_GOOD_MOVE;
}

/*
* Function: IsBadPosition
* Usage: if (IsBadPosition(nCoins)) . . .
* ---------------------------------------
* This function returns true if nCoins is a bad position.
* A bad position is one in which there is no good move.
* Being left with a single coin is clearly a bad position
* and represents the simple case of the recursion.
*/

bool IsBadPosition(int nCoins) {

if (nCoins == 1) return true;
return FindGoodMove(nCoins) == NO_GOOD_MOVE;
}
Backtracking Algorithms – 252 –

One of the principal ideas in this text is the notion of abstraction, which is the process
of separating out the general aspects of a problem so that they are no longer obscured by
the details of a specific domain. You may not be terribly interested in a program that
plays nim; after all, nim is rather boring once you figure it out. What you would probably
enjoy more is a program that is general enough to be adapted to play nim, or tic-tac-toe,
or any other two-player strategy game you choose.
The first step in creating such a generalization lies in recognizing that there are several
concepts that are common to all games. The most important such concept is state. For
any game, there is some collection of data that defines exactly what is happening at any
point in time. In the nim game, for example, the state consists of the number of coins on
the table and whose turn it is to move. For a game like chess, the state would instead
include what pieces were currently on which squares. Whatever the game, however, it
should be possible to combine all the relevant data together into a single record structure
and then refer to it using a single variable. Another common concept is that of a move.
In nim, a move consists of an integer representing the number of coins taken away. In
chess, a move might consist of a pair indicating the starting and ending coordinates of the
piece that is moving, although this approach is in fact complicated by the need to
represent various esoteric moves like castling or the promotion of a pawn. In any case, a
move can also be represented by a single structure that includes whatever information is
appropriate to that particular game. The process of abstraction consists partly of defining
these concepts as general types, with names like stateT and moveT, that transcend the
details of any specific game. The internal structure of these types will be different for
different games, but the abstract algorithm can refer to these concepts in a generic form.
Consider, for example, the following main program, which comes from the tic-tac-toe
example introduced in Figure 7-6 at the end of this chapter:
int main() {
GiveInstructions();
stateT state = NewGame();
moveT move;
while (!GameIsOver(state)) {
DisplayGame(state);
switch (WhoseTurn(state)) {
case Human:
move = GetUserMove(state);
break;
case Computer:
move = ChooseComputerMove(state);
DisplayMove(move);
break;
}
MakeMove(state, move);
}
AnnounceResult(state);
return 0;
}

At this level, the program is easy to read. It begins by giving instructions and then calls
NewGame to initialize a new game, storing the result in the variable state. It then goes
into a loop, taking turns for each side until the game is over. On the human player’s
turns, it calls a function GetUserMove to read in the appropriate move from the user. On
its own turns, the program calls ChooseComputerMove, which has the task of finding the
best move in a particular state. Once the move has been determined by one of these two
functions, the main program then calls MakeMove, which updates the state of the game to
Backtracking Algorithms – 253 –

show that the indicated move has been made and that it is now the other player’s turn. At
the end, the program displays the result of the game and exits.
It is important to notice that the main program gives no indication whatsoever about
what the actual game is. It could just as easily be nim or chess as tic-tac-toe. Each game
requires its own definitions for stateT , moveT , and the various functions like
GiveInstructions, MakeMove, and GameIsOver . Even so, the implementation of the
main program as it appears here is general enough to work for many different games.
The minimax strategy
The main program, however, is hardly the most interesting part of a game. The real
challenge consists of providing the computer with an effective strategy. In the general
program for two-player games, the heart of the computer’s strategy is the function
FindBestMove , which is called by the function ChooseComputerMove in the main
program. Given a particular state of the game, the role of FindBestMove is to return the
optimal move in that position.
From the discussion of nim earlier in this chapter, you should already have some sense
of what constitutes an optimal move. The best move in any position is simply the one
that leaves your opponent in the worst position. The worst position is likewise the one
that offers the weakest best move. This idea—finding the position that leaves your
opponent with the worst possible best move—is called the minimax strategy because the
goal is to find the move that minimizes your opponent’s maximum opportunity.
The best way to visualize the operation of the minimax strategy is to think about the
possible future moves in a game as forming a branching diagram that expands on each
turn. Because of their branching character, such diagrams are called game trees. The
current state is represented by a dot at the top of the game tree. If there are, for example,
three possible moves from this position, there will be three lines emanating down from
the current state to three new states that represent the results of these moves, as shown in
the following diagram:

For each of these new positions, your opponent will also have options. If there are again
three options from each of these positions, the next generation of the game tree looks like
this:

Which move do you choose in the initial position? Clearly, your goal is to achieve the
best outcome. Unfortunately, you only get to control half of the game. If you were able
to select your opponent’s move as well as your own, you could select the path to the state
two turns away that left you in the best position. Given the fact that your opponent is also
Backtracking Algorithms – 254 –

trying to win, the best thing you can do is choose the initial move that leaves your
opponent with as few winning chances as possible.
In order to get a sense of how you should proceed, it helps to add some quantitative
data to the analysis. Determine whether a particular move is better than some alternative
is much easier if it is possible to assign a numeric score to each possible move. The
higher the numeric score, the better the move. Thus, a move that had a score of +7, for
example, is better than a move with a rating of –4. In addition to rating each possible
move, it makes sense to assign a similar numeric rating to each position in the game.
Thus, one position might have a rating of +9 and would therefore be better than a position
with a score of only +2.
Both positions and moves are rated from the perspective of the player having the
move. Moreover, the rating system is designed to be symmetric around 0, in the sense
that a position that has a score of +9 for the player to move would have a score of –9
from the opponent’s point of view. This interpretation of rating numbers captures the
idea that a position that is good for one player is therefore a bad position for the
opponent, as you saw in the discussion of the Nim game earlier in this chapter. More
importantly, defining the rating system in this way makes it easy to express the
relationship between the scores for moves and positions. The score for any move is
simply the negative of the score for the resulting position when rated by your opponent.
Similarly, the rating of any position can be defined as the rating of its best move.
To make this discussion more concrete, let’s consider a simple example. Suppose that
you have looked two steps ahead in the game, covering one move by you and the possible
responses from your opponent. In computer science, a single move for a single player is
called a ply to avoid the ambiguity associated with the words move and turn, which
sometimes imply that both players have a chance to play. If you rate the positions at the
conclusion of the two-ply analysis, the game tree might look like this:

+7 +6 –9 –5 +9 –4 –1 +1 –2
Because the positions at the bottom of this tree are again positions in which—as at the top
of the tree—you have to move, the rating numbers in those positions are assigned from
your perspective. Given these ratings of the potential positions, what move should you
make from the original configuration? At first glance, you might be attracted by the fact
that the leftmost branch has the most positive total score or that the center one contains a
path that leads to a +9, which is an excellent outcome for you. Unfortunately, none of
these considerations matter much if your opponent is playing rationally. If you choose
the leftmost branch, your opponent will surely take the rightmost option from there,
which leaves you with a –9 position. The same thing happens if you choose the center
branch; your opponent finds the worst possible position, which has a rating of –5. The
best you can do is choose the rightmost branch, which only allows your opponent to end
up with a –2 rating. While this position is hardly ideal, it is better for you than the other
outcomes.
Backtracking Algorithms – 255 –

The situation becomes easier to follow if you add the ratings for each of your
opponent’s responses at the second level of the tree. The rating for a move—from the
perspective of the player making it—is the negative of the resulting position. Thus, the
move ratings from your opponent’s point of view look like this:

–7 +9 +5 +4 +1 +2
–6 –9 –1

In these positions, your opponent will seek to play the move with the best score. By
choosing the rightmost path, you minimize the maximum score available to your
opponent, which is the essence of the minimax strategy.

Implementing the minimax algorithm

The minimax algorithm is quite general and can be implemented in a way that does not
depend on the specific characteristics of the game. In many respects, the implementation
of the minimax algorithm is similar to the strategy section in nim.cpp because it consists
of two mutually recursive functions, one that finds the best move and another than
evaluates the quality of a position. If you want to make your algorithm as general as
possible, however, you must modify the approach used in the nim program to
accommodate the following extensions:

• It must be possible to limit the depth of the recursive search. For games that involve
any significant level of complexity, it is impossible to search the entire game tree in a
reasonable amount of time. If you try to apply this approach to chess, for example, a
program running on the fastest computers available would require many times the
lifetime of the universe to make the first move. As a result, a practical implementation
of the minimax algorithm must include a provision for cutting off the search at a
certain point. One possible approach is to limit the depth of the recursion to a certain
number of moves. You could, for example, allow the recursion to proceed until each
player had made three moves and then evaluate the position at that point using some
nonrecursive approach.
• It must be possible to assign ratings to moves and positions. Every position in nim is
either good or bad; there are no other options. In a more complex game, it is
necessary—particularly if you can’t perform a complete analysis—to assign ratings to
positions and moves so that the algorithm has a standard for comparing them against
other possibilities. The rating scheme used in this implementation assigns integers to
positions and moves. Those integers extend in both the positive and the negative
direction and are centered on zero, which means that a rating of –5 for one player is
equivalent to a rating of +5 from the opponent’s point of view. Zero is therefore the
neutral rating and is given the name NeutralPosition. The maximum positive rating
is the constant WinningPosition , which indicates a position in which the player
whose turn it is to move will invariably win; the corresponding extreme in the negative
direction is LosingPosition, which indicates that the player will always lose.
Taking these general considerations into account requires some changes in the design
of the mutually recursive functions that implement the minimax algorithm, which are
called FindBestMove and EvaluatePosition. Both functions take the state of the game
Backtracking Algorithms – 256 –

as an argument, but each also requires the current depth of the recursion so that the
recursive search can be restricted if necessary. Moreover, in order to avoid a
considerable amount of redundant calculation, it is extremely useful if FindBestMove can
return a rating along with the best move, so it uses a reference parameter along with the
return value to return the two pieces of information.. Given these design decisions, the
prototypes for FindBestMove and EvaluatePosition look like this:

moveT FindBestMove(stateT state, int depth, int & rating);

int EvaluatePosition(stateT state, int depth);

The strategy for FindBestMove can be expressed using the following pseudocode:

moveT FindBestMove(stateT state, int depth, int & rating) {

for (each possible move or until you find a forced win) {
Make the move.
Evaluate the resulting position, adding one to the depth indicator.
Keep track of the minimum rating so far, along with the corresponding move.
Retract the move to restore the original state.
}
Store the move rating into the reference parameter.
Return the best move.
}

The corresponding implementation, which follows this pseudocode outline, looks like
this:

moveT FindBestMove(stateT state, int depth, int & rating) {

Vector<moveT> moveList;
GenerateMoveList(state, moveList);
int nMoves = moveList.size();
if (nMoves == 0) Error("No moves available");
moveT bestMove;
int minRating = WINNING_POSITION + 1;
for (int i = 0; i < nMoves && minRating != LOSING_POSITION; i++) {
moveT move = moveList[i];
MakeMove(state, move);
int curRating = EvaluatePosition(state, depth + 1);
if (curRating < minRating) {
bestMove = move;
minRating = curRating;
}
RetractMove(state, move);
}
rating = -minRating;
return bestMove;
}

The function GenerateMoveList(state, moveArray) is implemented separately for

each game and has the effect of filling the elements in moveArray with a list of the legal
moves in the current position; the result of the function is the number of available moves.
The only other parts of this function that require some comment are the line

minRating = WinningPosition + 1;

which initializes the value of minRating to a number large enough to guarantee that this
value will be replaced on the first cycle through the for loop, and the line
Backtracking Algorithms – 257 –

rating = -minRating;

which stores the rating of the best move in the reference parameter. The negative sign is
included because the perspective has shifted: the positions were evaluated from the point-
of-view of your opponent, whereas the ratings express the value of a move from your
own point of view. A move that leaves your opponent with a negative position is good
for you and therefore has a positive value.
The EvaluatePosition function is considerably simpler. The simple cases that allow
the recursion to terminate occur when the game is over or when the maximum allowed
recursive depth has been achieved. In these cases, the program must evaluate the current
state as it exists without recourse to further recursion. This evaluation is performed by
the function EvaluateStaticPosition, which is coded separately for each game. In the
general case, however, the rating of a position is simply the rating of the best move
available, given the current state. Thus, the following code is sufficient to the task:
int EvaluatePosition(stateT state, int depth) {
int rating;
if (GameIsOver(state) || depth >= MAX_DEPTH) {
return EvaluateStaticPosition(state);
}
FindBestMove(state, depth, rating);
return rating;
}

Using the general strategy to solve a specific game

The minimax strategy embodied in the functions FindBestMove and EvaluatePosition
takes care of the conceptually complicated work of finding the best move from a given
position. Moreover, because it is written in an abstract way, the code does not depend on
the details of a particular game. Once you have these functions, the task of coding a new
two-player strategy game is reduced to the problem of designing moveT and stateT
structures that are appropriate to the game and then writing code for the functions that
must be supplied independently for each game.
The long program in Figure 7-6 illustrates how to use the general minimax facility to
construct a program that plays tic-tac-toe. The implementations of the main program and
the functions FindBestMove and EvaluatePosition are completely independent of the
details of tic-tac-toe, even though they are responsible for calculating most of the
strategy. What makes this program play tic-tac-toe is the definition of the types and
functions that surround the basic framework.
The types moveT and stateT are defined so that they are appropriate for tic-tac-toe. If
you number the squares in the board like this:

1 2 3
4 5 6
7 8 9

a move can be represented as a single integer, which means that the appropriate definition
for the type moveT is simply
typedef int moveT;
Backtracking Algorithms – 258 –

Figure 7-6 Program to play the game of tic-tac-toe

/*
* File: tictactoe.cpp
* -------------------
* This program plays a game of tic-tac-toe with the user. The
* code is designed to make it easy to adapt the general structure
* to other games.
*/

#include "genlib.h"
#include "simpio.h"
#include "vector.h"
#include "grid.h"
#include <iostream>

/*
* Constants: WINNING_POSITION, NEUTRAL_POSITION, LOSING_POSITION
* --------------------------------------------------------------
* These constants define a rating system for game positions. A
* rating is an integer centered at 0 as the neutral score: ratings
* greater than 0 are good for the current player, ratings less than
* 0 are good for the opponent. The constants WINNING_POSITION and
* LOSING_POSITION are opposite in value and indicate a position that
* is a forced win or loss, respectively. In a game in which the
* analysis is complete, no intermediate values ever arise. If the
* full tree is too large to analyze, the EvaluatePosition function
* returns integers that fall between the two extremes.
*/
const int WINNING_POSITION = 1000;
const int NEUTRAL_POSITION = 0;
const int LOSING_POSITION = -WINNING_POSITION;

/*
* Type: playerT
* -------------
* This type is used to distinguish the human and computer
* players and keep track of who has the current turn.
*/
enum playerT { Human, Computer };

/*
* Type: moveT
* -----------
* For any particular game, the moveT type must keep track of the
* information necessary to describe a move. For tic-tac-toe,
* a moveT is simply an integer identifying the number of one of
* the nine squares, as follows:
*
* 1 | 2 | 3
* ---+---+---
* 4 | 5 | 6
* ---+---+---
* 7 | 8 | 9
*/
typedef int moveT;
Backtracking Algorithms – 259 –

/*
* Type: stateT
* ------------
* For any game, the stateT structure records the current state of
* the game. As in Chapter 4, the tic-tac-toe board is represented
* using a Grid<char>; the elements must be either 'X', 'O', or ' '.
* In addition to the board array, the code stores a playerT value
* to indicate whose turn it is. In this game, the stateT structure
* also contains the total number of moves so that functions can
* find this value without counting the number of occupied squares.
*/
struct stateT {
Grid<char> board;
playerT whoseTurn;
int turnsTaken;
};

/*
* Constant: MAX_DEPTH
* -------------------
* This constant indicates the maximum depth to which the recursive
* search for the best move is allowed to proceed.
*/
const int MAX_DEPTH = 10000;

/*
* Constant: FIRST_PLAYER
* ----------------------
* This constant indicates which player goes first.
*/
const playerT FIRST_PLAYER = Computer;

/* Private function prototypes */

void GiveInstructions();
moveT FindBestMove(stateT state, int depth, int & pRating);
int EvaluatePosition(stateT state, int depth);
stateT NewGame();
void DisplayGame(stateT state);
void DisplayMove(moveT move);
char PlayerMark(playerT player);
moveT GetUserMove(stateT state);
moveT ChooseComputerMove(stateT state);
void GenerateMoveList(stateT state, Vector<moveT> & moveList);
bool MoveIsLegal(moveT move, stateT state);
void MakeMove(stateT & state, moveT move);
void RetractMove(stateT & state, moveT move);
bool GameIsOver(stateT state);
void AnnounceResult(stateT state);
playerT WhoseTurn(stateT state);
playerT Opponent(playerT player);
int EvaluateStaticPosition(stateT state);
bool CheckForWin(stateT state, playerT player);
bool CheckForWin(Grid<char> & board, char mark);
bool CheckLine(Grid<char> & board, char mark,
int row, int col, int dRow, int dCol);
Backtracking Algorithms – 260 –

/*
* Main program
* ------------
* The main program, along with the functions FindBestMove and
* EvaluatePosition, are general in their design and can be
* used with most two-player games. The specific details of
* tic-tac-toe do not appear in these functions and are instead
* encapsulated in the stateT and moveT data structures and a
* a variety of subsidiary functions.
*/

int main() {
GiveInstructions();
stateT state = NewGame();
moveT move;
while (!GameIsOver(state)) {
DisplayGame(state);
switch (WhoseTurn(state)) {
case Human:
move = GetUserMove(state);
break;
case Computer:
move = ChooseComputerMove(state);
DisplayMove(move);
break;
}
MakeMove(state, move);
}
AnnounceResult(state);
return 0;
}

/*
* Function: GiveInstructions
* Usage: GiveInstructions();
* --------------------------
* This function gives the player instructions about how to
* play the game.
*/

void GiveInstructions() {
cout << "Welcome to tic-tac-toe. The object of the game" << endl;
cout << "is to line up three symbols in a row," << endl;
cout << "vertically, horizontally, or diagonally." << endl;
cout << "You'll be " << PlayerMark(Human) << " and I'll be "
<< PlayerMark(Computer) << "." << endl;
}
Backtracking Algorithms – 261 –

/*
* Function: NewGame
* Usage: state = NewGame();
* -------------------------
* This function starts a new game and returns a stateT that
* has been initialized to the defined starting configuration.
*/

stateT NewGame() {
stateT state;
state.board.resize(3, 3);
for (int i = 0; i < 3; i++) {
for (int j = 0; j < 3; j++) {
state.board[i][j] = ' ';
}
}
state.whoseTurn = FIRST_PLAYER;
state.turnsTaken = 0;
return state;
}

/*
* Function: DisplayGame
* Usage: DisplayGame(state);
* --------------------------
* This function displays the current state of the game.
*/

void DisplayGame(stateT state) {

if (GameIsOver(state)) {
cout << "The final position looks like this:" << endl << endl;
} else {
cout << "The game now looks like this:" << endl << endl;
}
for (int i = 0; i < 3; i++) {
if (i != 0) cout << "---+---+---" << endl;
for (int j = 0; j < 3; j++) {
if (j != 0) cout << "|";
cout << " " << state.board[i][j] << " ";
}
cout << endl;
}
cout << endl;
}

/*
* Function: DisplayMove
* Usage: DisplayMove(move);
* -------------------------
* This function displays the computer's move.
*/

void DisplayMove(moveT move) {

cout << "I'll move to " << move << endl;
}
Backtracking Algorithms – 262 –

/*
* Function: FindBestMove
* Usage: move = FindBestMove(state, depth, pRating);
* --------------------------------------------------
* This function finds the best move for the current player, given
* the specified state of the game. The depth parameter and the
* constant MAX_DEPTH are used to limit the depth of the search
* for games that are too difficult to analyze in full detail.
* The function returns the best move by storing an integer in
* the variable to which pRating points.
*/

moveT FindBestMove(stateT state, int depth, int & rating) {

Vector<moveT> moveList;
GenerateMoveList(state, moveList);
int nMoves = moveList.size();
if (nMoves == 0) Error("No moves available");
moveT bestMove;
int minRating = WINNING_POSITION + 1;
for (int i = 0; i < nMoves && minRating != LOSING_POSITION; i++) {
moveT move = moveList[i];
MakeMove(state, move);
int curRating = EvaluatePosition(state, depth + 1);
if (curRating < minRating) {
bestMove = move;
minRating = curRating;
}
RetractMove(state, move);
}
rating = -minRating;
return bestMove;
}

/*
* Function: EvaluatePosition
* Usage: rating = EvaluatePosition(state, depth);
* -----------------------------------------------
* This function evaluates a position by finding the rating of
* the best move in that position. The depth parameter and the
* constant MAX_DEPTH are used to limit the depth of the search.
*/

int EvaluatePosition(stateT state, int depth) {

int rating;
if (GameIsOver(state) || depth >= MAX_DEPTH) {
return EvaluateStaticPosition(state);
}
FindBestMove(state, depth, rating);
return rating;
}
Backtracking Algorithms – 263 –

/*
* Function: NewGame
* Usage: state = NewGame();
* -------------------------
* This function starts a new game and returns a stateT that
* has been initialized to the defined starting configuration.
*/

stateT NewGame() {
stateT state;

for (int i = 1; i <= 9; i++) {

state.board[i] = ' ';
}
state.whoseTurn = FIRST_PLAYER;
state.turnsTaken = 0;
return state;
}

/*
* Function: DisplayGame
* Usage: DisplayGame(state);
* --------------------------
* This function displays the current state of the game.
*/

void DisplayGame(stateT state) {

if (GameIsOver(state)) {
cout << "The final position looks like this:" << endl << endl;
} else {
cout << "The game now looks like this:" << endl << endl;
}
for (int row = 0; row < 3; row++) {
if (row != 0) cout << "---+---+---" << endl;
for (int col = 0; col < 3; col++) {
if (col != 0) cout << "|";
cout << " " << state.board[row * 3 + col + 1] << " ";
}
cout << endl;
}
cout << endl;
}

/*
* Function: DisplayMove
* Usage: DisplayMove(move);
* -------------------------
* This function displays the computer's move.
*/

void DisplayMove(moveT move) {

cout << "I'll move to " << move << endl;
}
Backtracking Algorithms – 264 –

/*
* Function: PlayerMark
* Usage: mark = PlayerMark(player);
* ---------------------------------
* This function returns the mark used on the board to indicate
* the specified player. By convention, the first player is
* always X, so the mark used for each player depends on who
* goes first.
*/

char PlayerMark(playerT player) {

if (player == FIRST_PLAYER) {
return 'X';
} else {
return 'O';
}
}

/*
* Function: GetUserMove
* Usage: move = GetUserMove(state);
* ---------------------------------
* This function allows the user to enter a move and returns the
* number of the chosen square. If the user specifies an illegal
* move, this function gives the user the opportunity to enter
* a legal one.
*/

moveT GetUserMove(stateT state) {

cout << "Your move." << endl;
moveT move;
while (true) {
cout << "What square? ";
move = GetInteger();
if (MoveIsLegal(move, state)) break;
cout << "That move is illegal. Try again." << endl;
}
return move;
}

/*
* Function: ChooseComputerMove
* Usage: move = ChooseComputerMove(state);
* ----------------------------------------
* This function chooses the computer's move and is primarily
* a wrapper for FindBestMove. This function also makes it
* possible to display any game-specific messages that need
* to appear at the beginning of the computer's turn. The
* rating value returned by FindBestMove is simply discarded.
*/

moveT ChooseComputerMove(stateT state) {

int rating;
cout << "My move." << endl;
return FindBestMove(state, 0, rating);
}
Backtracking Algorithms – 265 –

/*
* Function: GenerateMoveList
* Usage: GenerateMoveList(state, moveList);
* -----------------------------------------
* This function generates a list of the legal moves available in
* the specified state. The list of moves is returned in the
* vector moveList, which must be allocated by the client.
*/

void GenerateMoveList(stateT state, Vector<moveT> & moveList) {

for (int i = 1; i <= 9; i++) {
moveT move = moveT(i);
if (MoveIsLegal(move, state)) {
moveList.add(moveT(i));
}
}
}

/*
* Function: MoveIsLegal
* Usage: if (MoveIsLegal(move, state)) . . .
* ------------------------------------------
* This function returns true if the specified move is legal.
*/

bool MoveIsLegal(moveT move, stateT state) {

if (move < 1 || move > 9) return false;
int row = (move - 1) / 3;
int col = (move - 1) % 3;
return state.board[row][col] == ' ';
}

/*
* Function: MakeMove
* Usage: MakeMove(state, move);
* -----------------------------
* This function changes the state by making the indicated move.
*/

void MakeMove(stateT & state, moveT move) {

int row = (move - 1) / 3;
int col = (move - 1) % 3;
state.board[row][col] = PlayerMark(state.whoseTurn);
state.whoseTurn = Opponent(state.whoseTurn);
state.turnsTaken++;
}
Backtracking Algorithms – 266 –

/*
* Function: RetractMove
* Usage: RetractMove(state, move);
* --------------------------------
* This function changes the state by "unmaking" the indicated move.
*/

void RetractMove(stateT & state, moveT move) {

int row = (move - 1) / 3;
int col = (move - 1) % 3;
state.board[row][col] = ' ';
state.whoseTurn = Opponent(state.whoseTurn);
state.turnsTaken--;
}

/*
* Function: GameIsOver
* Usage: if (GameIsOver(state)) . . .
* -----------------------------------
* This function returns true if the game is complete.
*/

bool GameIsOver(stateT state) {

return (state.turnsTaken == 9
|| CheckForWin(state, state.whoseTurn)
|| CheckForWin(state, Opponent(state.whoseTurn)));
}

/*
* Function: AnnounceResult
* Usage: AnnounceResult(state);
* -----------------------------
* This function announces the result of the game.
*/

void AnnounceResult(stateT state) {

DisplayGame(state);
if (CheckForWin(state, Human)) {
cout << "You win." << endl;
} else if (CheckForWin(state, Computer)) {
cout << "I win." << endl;
} else {
cout << "Cat's game." << endl;
}
}
Backtracking Algorithms – 267 –

/*
* Function: WhoseTurn
* Usage: player = WhoseTurn(state);
* ---------------------------------
* This function returns whose turn it is, given the current
* state of the game. The reason for making this a separate
* function is to ensure that the common parts of the code do
* not need to refer to the internal components of the state.
*/

playerT WhoseTurn(stateT state) {

return state.whoseTurn;
}

/*
* Function: Opponent
* Usage: opp = Opponent(player);
* ------------------------------
* This function returns the playerT value corresponding to the
* opponent of the specified player.
*/

playerT Opponent(playerT player) {

return (player == Human) ? Computer : Human;
}

/*
* Function: EvaluateStaticPosition
* Usage: rating = EvaluateStaticPosition(state);
* ----------------------------------------------
* This function gives the rating of a position without looking
* ahead any further in the game tree. Although this function
* duplicates much of the computation of GameIsOver and therefore
* introduces some runtime inefficiency, it makes the algorithm
* somewhat easier to follow.
*/

int EvaluateStaticPosition(stateT state) {

if (CheckForWin(state, state.whoseTurn)) {
return WINNING_POSITION;
}
if (CheckForWin(state, Opponent(state.whoseTurn))) {
return LOSING_POSITION;
}
return NEUTRAL_POSITION;
}
Backtracking Algorithms – 268 –

/*
* Function: CheckForWin
* Usage: if (CheckForWin(state, player)) . . .
* --------------------------------------------
* This function returns true if the specified player has won
* the game. The check on turnsTaken increases efficiency,
* because neither player can win the game until the fifth move.
*/

bool CheckForWin(stateT state, playerT player) {

if (state.turnsTaken < 5) return false;
return CheckForWin(state.board, PlayerMark(player));
}

/*
* Checks to see whether the specified player identified by mark
* ('X' or 'O') has won the game. To reduce the number of special
* cases, this implementation uses the helper function CheckLine
* so that the row, column, and diagonal checks are the same.
*/

bool CheckForWin(Grid<char> & board, char mark) {

for (int i = 0; i < 3; i++) {
if (CheckLine(board, mark, i, 0, 0, 1)) return true;
if (CheckLine(board, mark, 0, i, 1, 0)) return true;
}
if (CheckLine(board, mark, 0, 0, 1, 1)) return true;
return CheckLine(board, mark, 2, 0, -1, 1);
}

/*
* Checks a line extending across the board in some direction.
* The starting coordinates are given by the row and col
* parameters. The direction of motion is specified by dRow
* and dCol, which show how to adjust the row and col values
* on each cycle. For rows, dRow is always 0; for columns,
* dCol is 0. For diagonals, the dRow and dCol values will
* be +1 or -1 depending on the direction of the diagonal.
*/

bool CheckLine(Grid<char> & board, char mark, int row, int col,
int dRow, int dCol) {
for (int i = 0; i < 3; i++) {
if (board[row][col] != mark) return false;
row += dRow;
col += dCol;
}
return true;
}
Backtracking Algorithms – 269 –

The state of the tic-tac-toe game is determined by the symbols on the nine squares of
the board and whose turn it is to move. Thus, you can represent the essential components
of the state as follows:
struct stateT {
char board[(3 * 3) + 1];
playerT whoseTurn;
};

The board is represented as a single-dimensional array to correspond with the numbering

of the squares from 1 to 9; the fact that position 0 is unused makes it necessary to add an
extra element to the array. As the complete program shows, adding a turnsTaken
component to the state makes it much easier to determine whether the game is complete.
The rules of tic-tac-toe are expressed in the implementation of functions like
GenerateMoveList, MoveIsLegal , and EvaluateStaticPosition , each of which is
coded specifically for this game. Even though these functions require some thought, they
tend to be less conceptually complex than those that implement the minimax strategy. By
coding the difficult functions once and then reusing them in various different game
programs, you can avoid having to go through the same logic again and again.

Summary
In this chapter, you have learned to solve problems that require making a sequence of
choices as you search for a goal, as illustrated by finding a path through a maze or a
winning strategy in a two-player game. The basic strategy is to write programs that can
backtrack to previous decision points if those choices lead to dead ends. By exploiting
the power of recursion, however, you can avoid coding the details of the backtracking
process explicitly and develop general solution strategies that apply to a wide variety of
problem domains.
Important points in this chapter include:
• You can solve most problems that require backtracking by adopting the following
recursive approach:
If you are already at a solution, report success.
for (every possible choice in the current position ) {
Make that choice and take one step along the path.
Use recursion to solve the problem from the new position.
If the recursive call succeeds, report the success to the next higher level.
Back out of the current choice to restore the state at the beginning of the loop.
}
Report failure.
• The complete history of recursive calls in a backtracking problem—even for relatively
simple applications—is usually too complex to understand in detail. For problems that
involve any significant amount of backtracking, it is essential to accept the recursive
leap of faith.
• You can often find a winning strategy for two-player games by adopting a recursive-
backtracking approach. Because the goal in such games involves minimizing the
winning chances for your opponent, the conventional strategic approach is called the
minimax algorithm.
• It is possible to code the minimax algorithm in a general way that keeps the details of
any specific game separate from the implementation of the minimax strategy itself.
This approach makes it easier to adapt an existing program to new games.
Backtracking Algorithms – 270 –

Review questions
1. What is the principal characteristic of a backtracking algorithm?
2. Using your own words, state the right-hand rule for escaping from a maze. Would a
left-hand rule work equally well?
3. What is the insight that makes it possible to solve a maze by recursive backtracking?
4. What are the simple cases that apply in the recursive implementation of SolveMaze?
5. Why is important to mark squares as you proceed through the maze? What would
happen in the SolveMaze function if you never marked any squares?
6. What is the purpose of the UnmarkSquare call at the end of the for loop in the
SolveMaze implementation? Is this statement essential to the algorithm?

7. What is the purpose of the Boolean result returned by SolveMaze?

8. In your own words, explain how the backtracking process actually takes place in the
recursive implementation of SolveMaze.
9. In the simple nim game, the human player plays first and begins with a pile of 13
coins. Is this a good or a bad position? Why?
10. Write a simple C++ expression based on the value of nCoins that has the value true
if the position is good for the current player and false otherwise. (Hint: Use the %
operator.)
11. What is the minimax algorithm? What does its name signify?
12. Suppose you are in a position in which the analysis for the next two moves shows the
following rated outcomes from your opponent’s point-of-view:

–4 –9 + 4 –3 –2 –4 –3 + 2 –5 +1 0 –5
If you adopt the minimax strategy, what is the best move to make in this position?
What is the rating of that move from your perspective?
13. Why is it useful to develop an abstract implementation of the minimax algorithm?
14. What two data structures are used to make the minimax implementation independent
of the specific characteristics of a particular game?
15. What is the role of each of the three arguments to the FindBestMove function?
16. Explain the role of the EvaluateStaticPosition function in the minimax
implementation.
Backtracking Algorithms – 271 –

Programming exercises
1. The SolveMaze function shown in Figure 7-3 implements a recursive algorithm for
solving a maze but is not complete as it stands. The solution depends on several
functions exported by the mazelib.h interface, which is specified in Figure 7-2 but
never actually implemented.
Write a file mazelib.cpp that implements this interface. Your implementation
should store the data representing the maze as part of the private state of the module.
This design requires you to declare static global variables within the implementation
that are appropriate to the data you need to implement the operations. If you design
the data structure well, most of the individual function definitions in the interface are
quite simple. The hard parts of this exercise are designing an appropriate internal
representation and implementing the function ReadMapFile, which initializes the
internal structures from a data file formatted as described in the interface
documentation.
2. In many mazes, there are multiple paths. For example, the diagrams below show
three solutions for the same maze:

Θ × × × Θ Θ
× × × × ×
× × × × × × × ×
× × × × × × ×
× × × × ×
× × × × × × × × × ×
length = 13 length = 15 length = 13
None of these solutions, however, is optimal. The shortest path through the maze
has a path length of 11:

Θ
×
× ×
× ×
×
× × × ×

Write a function
int ShortestPathLength(pointT pt);
that returns the length of the shortest path in the maze from the specified position to
any exit. If there is no solution to the maze, ShortestPathLength should return the
constant NO_SOLUTION, which is defined to have a value larger than the maximum
permissible path length, as follows:
const int NO_SOLUTION = 10000;
Backtracking Algorithms – 272 –

3. As implemented in Figure 7-3, the SolveMaze function unmarks each square as it

discovers there are no solutions from that point. Although this design strategy has
the advantage that the final configuration of the maze shows the solution path as a
series of marked squares, the decision to unmark squares as you backtrack has a cost
in terms of the overall efficiency of the algorithm. If you’ve marked a square and
then backtracked through it, you’ve already explored the possibilities leading from
that square. If you come back to it by some other path, you might as well rely on
your earlier analysis instead of exploring the same options again.
To give yourself a sense of how much these unmarking operations cost in terms of
efficiency, extend the SolveMaze program so that it records the number of recursive
calls as it proceeds. Use this program to calculate how many recursive calls are
required to solve the following maze if the call to UnmarkSquare remains part of the
program:

Run your program again, this time without the call to UnmarkSquare. What happens
to the number of recursive calls?
4. As the result of the preceding exercise makes clear, the idea of keeping track of the
path through a maze by using the MarkSquare facility in mazelib.h has a substantial
cost. A more practical approach is to change the definition of the recursive function
so that it keeps track of the current path as it goes. Following the logic of
SolveMaze, write a function
bool FindPath(pointT pt, Vector<pointT> path);
that takes, in addition to the coordinates of the starting position, a vector of pointT
values called path. Like SolveMaze, FindPath returns a Boolean value indicating
whether the maze is solvable. In addition, FindPath initializes the elements of the
path vector to a sequence of coordinates beginning with the starting position and
ending with the coordinates of the first square that lies outside the maze. For
example, if you use FindPath with the following main program
int main() {
ReadMazeMap(MazeFile);
Vector<pointT> path;
if (FindPath(GetStartPosition(), path)) {
cout << "The following path is a solution:" << endl;
for (int i = 0; i < path.size(); i++) {
cout << "(" << path[i].x << ", "
<< path[i].y << ")" << endl;
}
} else {
cout << "No solution exists." << endl;
}
return 0;
}
it will display the coordinates of the points in the solution path on the screen.
Backtracking Algorithms – 273 –

For this exercise, it is sufficient for FindPath to find any solution path. It need
not find the shortest one.
5. If you have access to the graphics library described in Chapter 6, extend the
maze.cpp program so that, in addition to keeping track of the internal data structures
for the maze, it also displays a diagram of the maze on the screen and shows the final
solution path.
6. Most drawing programs for personal computers make it possible to fill an enclosed
region on the screen with a solid color. Typically, you invoke this operation by
selecting a paint-bucket tool and then clicking the mouse, with the cursor somewhere
in your drawing. When yo