Introduction to C++

and Object Oriented Programming

Wouter Verkerke (NIKHEF)

v39 – Edition for 2007 UvA APP Master course © 2006 Wouter Verkerke, NIKHEF
Introduction and Overview

0 Introduction
& Overview

© 2006 Wouter Verkerke, NIKHEF

Intended audience and scope of course

• This course is targeted to students with some

programming experience in procedural (i.e. non-OO)
programming languages like Fortran, C, Pascal
– No specific knowledge of C, C++ is assumed

• This course will cover

– Basic C/C++ syntax, language features
– Basics of object oriented programming

• This course has some extra focus on the application of

C++ in (High Energy) Physics
– Organized processing and analysis of data
– Focus mostly in exercises

© 2006 Wouter Verkerke, NIKHEF

Programming, design and complexity
• The goal of software – to solve a particular problem
– E.g. computation of numeric problems, maintaining an organized
database of information, finding the Higgs etc..

• Growing computational power in the last decades have

allowed us to tackle more and more complex problems

• As a consequence software has also grown more

powerful and complex
– For example Microsoft Windows XP OS, last generation video
games, often well over 1.000.000 lines of source code
– Growth also occurs in physics: e.g. collection of software
packages for reconstruction/analysis of the BaBar experiment is
~6.4M lines of C++

• How do we deal with such increasing complexity?

© 2006 Wouter Verkerke, NIKHEF

Programming philosophies
• Key to successfully coding complex systems is break
down code into smaller modules and minimize the
dependencies between these modules

• Traditional programming languages (C, Fortran, Pascal)

achieve this through procedure orientation
– Modularity and structure of software revolves around ‘functions’
encapsulate (sub) algorithms
– Functions are a major tool in software structuring but leave a few
major design headaches

• Object-oriented languages (C++, Java,…) take this

several steps further
– Grouping data and associated functions into objects
– Profound implications for modularity and dependency reduction

© 2006 Wouter Verkerke, NIKHEF

What are objects

• ‘Software objects’ are often found naturally in real-life

problems
• Object oriented programming  Finding these objects
and their role in your problem

Drop box object

Dialog box
object

Check box object Button

© 2006 Wouter object
Verkerke, NIKHEF
What are objects

• An object has
– Properties : position, shape, text label
– Behavior : if you click on the ‘Cancel button’ a defined action occurs

Drop box object

Dialog box
object

Check box object Button object

© 2006 Wouter Verkerke, NIKHEF

Relating objects

• Object-Oriented Analysis and Design seeks the relation

between objects
– ‘Is-A’ relationship (a PushButton Is-A ClickableObject)
– ‘Has-A’ relationship (a DialogBox Has-A CheckBox)

Drop box object

Dialog box
object

Check box object Button

© 2006 Wouter object
Verkerke, NIKHEF
Benefits of Object-Oriented programming

• Benefits of Object-oriented programming

– Reuse of existing code – objects can represent generic problems
– Improved maintainability – objects are more self contained that
‘subroutines’ so code is less entangled
– Often a ‘natural’ way to describe a system – see preceding
example of dialog box

• But…
– Object oriented modeling does not substitute for sound thinking
– OO programming does not guarantee high performance, but it
doesn’t stand in its way either

• Nevertheless
– OO programming is currently the best way we know
to describe complex systems

© 2006 Wouter Verkerke, NIKHEF

Basic concept of OOAD

• Object-oriented programming revolves around

abstraction of your problem.
– Separate what you do from how you do it

• Example – PushButton object

PushButton is a complicated
piece of software – Handling
of mouse input, drawing
of graphics etc..

Nevertheless you can use a

PushButton object and don’t
need to know anything about
that. Its public interface can
be very simple: My name is
‘cancel’ and I will call function
doTheCancel() when I get
clicked
© 2006 Wouter Verkerke, NIKHEF
Techniques to achieve abstraction

• Abstraction is achieved through

1. Modularity

2. Encapsulation

3. Inheritance

4. Polymorphism

© 2006 Wouter Verkerke, NIKHEF

Modularity
• Decompose your problem logically in independent units
– Minimize dependencies between units – Loose coupling
– Group things together that have logical connection – Strong cohesion

• Example
– Grouping actions and properties of a bank account together

long getBalance()
void print()
void calculateInterest()

char* ownersName
long accountNumber
long accountBalance

Account

© 2006 Wouter Verkerke, NIKHEF

Encapsulation

• Separate interface and implementation and shield

implementation from object ‘users’

long getBalance()
void print() interface
void calculateInterest()

char* ownersName
long accountNumber implementation
long accountBalance (not visible from outside)

Account

© 2006 Wouter Verkerke, NIKHEF

Inheritance

• Describe new objects in terms of existing objects

• Example of mortgage account

long getBalance()
void print() interface
void calculateInterest()

char* ownersName
long accountNumber implementation
long accountBalance (not visible from outside)

Account
char* collateralObject
long collateralValue
MortgageAccount

© 2006 Wouter Verkerke, NIKHEF

Polymorphism

• Polymorphism is the ability to treat objects of different

types the same way
– You don’t know exactly what object you’re dealing with but you
know that you can interact with it through a standardized
interface
– Requires some function call decisions to be taken at run time

• Example with trajectories

– Retrieve position at a flight length of 5 cm
– Same interface works for different objects with identical interface

Point p = Traj->getPos(5.0)

LineTrajectory HelixTrajectory
© 2006 Wouter Verkerke, NIKHEF
Introduction to C++
• Wide choice of OO-languages – why program in C++?
– It depends on what you need…

• Advantage of C++ – It is a compiled language

– When used right the fastest of all OO languages
– Because OO techniques in C++ are resolved and implemented at compile
time rather than runtime so
• Maximizes run-time performance
• You don’t pay for what you don’t use

• Disadvantage of C++ – syntax more complex

– Also, realizing performance advantage not always trivial

• C++ best used for large scale projects where performance

matters
– C++ rapidly becoming standard in High Energy Physics for mainstream data
processing, online data acquisition etc…
– Nevertheless, If your program code will be O(100) lines and performance is
not critical C, Python, Java may be more efficient

© 2006 Wouter Verkerke, NIKHEF

C++ and other programming languages
ALGOL60

Simula67
Pascal SmallTalk
C

Ada83 C with classes

Objective C

C++
Eiffel

ANSI/ISO C++ Java

• NB: Java very similar to C++, but simpler

– Simpler syntax as all OO support is implemented at run-time
– If you know C++ Java will be easy to learn

© 2006 Wouter Verkerke, NIKHEF

Outline of the course
1. Introduction and overview

2. Basics of C++

3. Modularity and Encapsulation – Files and Functions

4. Class Basics

5. Object Analysis and Design

6. The Standard Library I – Using IOstreams

7. Generic Programming – Templates

8. The Standard Library II – The template library

9. Object Orientation – Inheritance & Polymorphism

10.Robust programming – Exception handling

11.Where to go from here

© 2006 Wouter Verkerke, NIKHEF
The basics of C++

1 The basics
of C++

© 2006 Wouter Verkerke, NIKHEF

“Hello world” in C++

• Let start with a very simple C++ program

// my first program in C++

#include <iostream>

int main () {
std::cout << "Hello World!“ << std::endl;
return 0;
}

© 2006 Wouter Verkerke, NIKHEF

“Hello world” in C++

• Let start with a very simple C++ program

// my first program in C++

#include <iostream>

int main () {
std::cout << "Hello World!“ << std::endl;
return 0;
} Anything on line after // in C++ is
considered a comment

© 2006 Wouter Verkerke, NIKHEF

“Hello world” in C++

• Let start with a very simple C++ program

// my first program in C++

#include <iostream>

int main () { Lines starting with # are directives for

std::cout << "Hello World!“the<< preprocessor
std::endl;
return 0;
Here we include some standard function
} and type declarations of objects defined
by the ‘iostream’ library

• The preprocessor of a C(++) compiler processes the

source code before it is passed the compiler. It can
– Include other source files (using the #include directive)
– Define and substitute symbolic names (using the #define directive)
– Conditionally include source code (using the #ifdef, #else, #endif
directives) © 2006 Wouter Verkerke, NIKHEF
“Hello world” in C++

• Let start with a very simple C++ program

// my first program in C++

#include <iostream>
Beginning of the main()
function declaration.
int main () {
std::cout << "Hello World!“ << std::endl;
return 0;
}

• The main() function is the default function where all C++

program begin their execution.
– In this case the main function takes no input arguments and returns
an integer value
– You can also declare the main function to take arguments which will
be filled with the command line options given to the program
© 2006 Wouter Verkerke, NIKHEF
“Hello word” in C++

• Let start with a very simple C++ program

// my first program in C++

Use iostream library
#include <iostream> objects to print string to
standard output
int main () {
std::cout << "Hello World!“ << std::endl;
return 0;
}

• The names std::cout and std::endl are declared in the

‘header file’ included through the ‘#include <iostream>’
preprocessor directive.
• The std::endl directive represents the ‘carriage return / line
feed’ operation on the terminal
© 2006 Wouter Verkerke, NIKHEF
“Hello word” in C++

• Let start with a very simple C++ program

// my first program in C++

#include <iostream>

int main () {
std::cout << "Hello World!“ << std::endl;
return 0;
} The return statement
passed the return value
back to the calling function

• The return value of the main() function is passed back to

the operating system as the ‘process exit code’

© 2006 Wouter Verkerke, NIKHEF

Compiling and running ‘Hello World’

• Example using Linux, (t)csh and g++ compiler

unix> g++ -o hello hello.cc Convert c++ source code

into executable

unix> hello Run executable ‘hello’

Hello World!

unix> echo $status Print exit code of last

0 run process (=hello)

© 2006 Wouter Verkerke, NIKHEF

Outline of this section

• Jumping in: the ‘hello world’ application

• Review of the basics int main() {

– Built-in data types int a = 3 ;
float b = 5 ;

– Operators on built-in types float c = a * b + 5 ;

– Control flow constructs if ( c > 10) {

return 1 ;
}
– More on block {} structures
return 0 ;
– Dynamic Memory allocation }

© 2006 Wouter Verkerke, NIKHEF

Review of the basics – built-in data types
• C++ has only few built-in data types

type name type description

char ASCII character, 1 byte
int, Integer. Can be signed, unsigned, long or
signed int, unsigned int, short. Size varies and depends on CPU
short int, long int architecture (2,4,8 bytes)
float, double Floating point number, single and double
precision
bool Boolean, can be true or false (1 byte)
enum Integer with limited set of named states
enum fruit { apple,pear,citrus }, or
enum fruit { apple=0,pear=1,citrus}

• More complex types are available in the ‘Standard Library’

– A standard collection of tools that is available with every compiler
– But these types are not fundamental as they're implement using standard C++
– We will get to this soon © 2006 Wouter Verkerke, NIKHEF
Defining data objects – variables

• Defining a data object can be done in several ways

int main() {
int j ; // definition – initial value undefined
int k = 0 ; // definition with assignment initialization
int l(0) ; // definition with constructor initialization

int m = k + l // initializer can be any valid C++ expressions

int a,b=0,c(b+5) // multiple declaration – a,b,c all integers

}

• Data objects declared can also be declared constant

int main() {
const float pi = 3.14159268 ; // constant data object
pi = 2 ; // ERROR – doesn’t compile
}

© 2006 Wouter Verkerke, NIKHEF

Defining data objects – variables
• Const variables must be initialized

int main() {
const float pi ; // ERROR – forgot to initialize

const float e = 2.72; // OK

const float f = 5*e ;// OK – expression is constant
}

• Definition can occur at any place in code

int main() {
float pi = 3.14159268 ;
cout << “pi = “ << pi << endl ;

float result = 0; // ‘floating’ declaration OK

result = doCalculation() ;
}
– Style tip: always declare variables as close as possible to point of first use
© 2006 Wouter Verkerke, NIKHEF
Literal constants for built-in types

• Literal constants for integer types

int j = 16 // decimal
int j = 0xF // hexadecimal (leading 0x)
int j = 020 // octal (leading 0)

unsigned int k = 4294967280U // unsigned literal

– Hex, octal literals good for bit patterns

(hex digit = 4 bits, octal digit = 3 bits)
– Unsigned literals good for numbers that are
\n newline
too large for signed integers
\r carriage return
(e.g. between 2^32/2 and 2^32-1)
\t tabulation
\v vertical tabulation
\b backspace
• Literal constants for character types \f page feed
\a alert (beep)
char ch = ‘A’ // Use single quotes; \' single quotes (')
\" double quotes (")
– Escape sequences exist for special characters \? question (?)
© 2006 Wouter\\
Verkerke, NIKHEF
inverted slash (\)
Arrays

• C++ supports 1-dimensional and N-dimensional arrays

– Definition

Type name[size] ;
Type name[size1][size2]…[sizeN] ;

– Array dimensions in definition must be constants

float x[3] ; // OK

const int n=3 ;

float x[n] ; // OK

int k=5 ;
float x[k] ; // ERROR!

– First element is always 0

– Assignment initialization possible

float x[3] = { 0.0, 5.7 , 2.3 } ;

float y[2][2] = { 0.0, 1.0, 2.0, 3.0 } ;
float y[3] = { 1.0 } ; // Incomplete initialization
© 2006 Wouter Verkerke, NIKHEFOK
Declaration versus definition of data
• Important fine point: definition of a variable is two actions
1. Allocation of memory for object
2. Assigning a symbolic name to that memory space

C++ symbol name space Memory layout

int myArray[5]

float x

char name[256]

– C++ symbolic name is a way for programs to give understandable

names to segments of memory.
– But it is an artifact: no longer exists once©the
2006program is compiled
Wouter Verkerke, NIKHEF
References

• C++ allows to create ‘alias names’, a different symbolic

name references an already allocated data object
– Syntax: ‘Type& name = othername’
– References do not necessarily allocate memory

• Example

int x ; // Allocation of memory for in

// and declaration of name ‘x’
int& y = x ; // Declaration of alias name ‘y’
// for memory refenced by ‘x’

x = 3 ;
cout << x << endl ; // prints ‘3’
cout << y << endl ; // also prints ‘3’

– Concept of references will become more interesting when we’ll

talk about functions
© 2006 Wouter Verkerke, NIKHEF
References

• Illustration C++ of reference concept

– Reference is symbolic name that points to same memory as
initializer symbol

C++ symbol name space Memory layout

int myArray[5]

float x
float& y = x ;

char name[256]

© 2006 Wouter Verkerke, NIKHEF

Pointers

• Pointers is a variable that contains a memory address

– Somewhat similar to a reference in functionality, but fundamentally
different in nature: a pointer is always an object in memory itself
– Definition: ‘TYPE* name’ makes pointer to data of type TYPE

C++ symbol name space Memory layout

int myArray[5]

float x
float& y = x
char name[256]

float* y = &x

© 2006 Wouter Verkerke, NIKHEF

Pointers

• Working with pointers

– Operator & takes memory address of symbol object (=pointer value)
– Operator * turns memory address (=pointer value) into symbol object

• Creating and reading through pointers

int x = 3, y = 4 ;
int* px ; // allocate px of type ‘pointer to integer’
px = &x ; // assign ‘memory address of x’ to pointer px

cout << px << endl ; // Prints 0x3564353, memory address of x

cout << *px << e`ndl ;// Prints 3, value of x, object pointed to by px

• Modifying pointers and objects pointed to

*px = 5 ; // Change value of object pointed to by px (=x) ;
cout << x << endl // Prints 5 (since changed through px)
px = &y ; // Reseat pointer to point to symbol named ‘y’

cout << px << endl ; // Prints 0x4863813, memory address of y

cout << *px << endl ;// Prints 4, value of y, objected pointed to by px
© 2006 Wouter Verkerke, NIKHEF
Pointers continued

• Pointers are also fundamentally related to arrays

int a[3] = { 1,2,3} ; // Allocates array of 3 integers

int* pa = &a[0] ; // Pointer pa now points a[0] ;

cout << *pa << endl ; // Prints ‘1’

cout << *(pa+1) << endl ; // Prints ‘2’

• Pointer (pa+1) points to next element of an array

– This works regardless of the type in the array
– In fact a itself is a pointer of type int* pointing to a[0]

• The Basic Rule for arrays and pointers

– a[i] is equivalent to *(a+i)

© 2006 Wouter Verkerke, NIKHEF

Pointers and arrays of char – strings
• Some special facilities exists for arrays of char
– char[] holds strings and is therefore most commonly used array

• Initialization of character arrays:

– String literals in double quotes are of type ‘char *’, i.e.
const char* blah = “querty”
is equivalent to
const char tmp[7] = {‘q’,’w’,’e’,’r’,’t’,’y’,0}’ ;
const char* blah = tmp ;
– Recap: single quoted for a single char, double quotes for a const pointer to
an array of chars

• Termination of character arrays

– Character arrays are by convention ended with a null char (\0)
– Can detect end of string without access to original definition
• For example for strings returned by “a literation expression”

© 2006 Wouter Verkerke, NIKHEF

Strings and string manipulation
• Since char[] strings are such an common objects
– the ‘Standard Library’ provides some convenient manipulation functions

• Most popular char[] manipulation functions

// Length of string
int strlen(const char* str) ;

// Append str2 to str1 (make sure yourself str1 is large enough)

char* strcat(char* str1, const char* str2) ;

// Compares strings, returns 0 if strings are identical

strcmp(const char* str1, const char* str2) ;

• Tip: Standard Library also provides ‘class string’ with superior

handling
– We’ll cover class string later
– But still need ‘const char*’ to interact with operating system function calls
(open file, close file, etc)
© 2006 Wouter Verkerke, NIKHEF
Reading vs. Writing – LValues and RValues

• C++ has two important concepts to distinguish read-

only objects and writeable objects
– An LValue is writable and can appear on the left-hand side of an
assignment operation
– An RValue is read-only and may only appear on the right-hand
side of assignment operations

• Example

int i;
char buf[10] ;

i = 5 ; // OK, i is an lvalue
5 = i ; // ERROR, 5 is not an lvalue
// (it has no memory location)

buf[0] = ‘c’ ; // OK buf[0] is an lvalue

buf = “qwerty” ; // ERROR, buf is immutably tied to char[10]
© 2006 Wouter Verkerke, NIKHEF
Operators and expressions – arithmetic operators

• Arithmetic operators overview

Name Operator Name Operator
Unary minus -x Modulus x % y
Multiplication x * y Addition x + y
Division x / y Subtraction x - y

• Arithmetic operators are evaluated from left to right

– 40 / 4 * 5 = (40 / 4) * 5 = 50 (not 2)

• In case of mixed-type expressions compiler

automatically converts integers up to floats

int i = 3, j = 5 ;
float x = 1.5 ;

float y = i*x ; // = 4.5 ; int i promoted to float

float z = j/i ; // = 1.0 ; ‘/’ ©has
2006precedence over
Wouter Verkerke, ‘=‘
NIKHEF
Operators and expressions – increment/decrement operators

• In/Decrement operators
Name Operator
Prefix increment ++x
Postfix increment x++
Prefix decrement --x
Postfix decrement x--

• Note difference
– Prefix operators return value after operation
– Postfix operators return value before operation

• Examples
int x=0 ;
cout << x++ << endl ; // Prints 0
cout << x << endl ; // Prints 1

cout << ++x << endl ; // Prints 2

cout << x << endl ; // Prints 2
© 2006 Wouter Verkerke, NIKHEF
Operators and expressions – relational operators

• Relational operators
Name Operator
Less than x < y
Less than or equal to x <= y
Greater than or equal to x >= y
Greater than x > y
Equal to x == y
Not equal to x != y

• All relational operators yield bool results

• Operators ==, != have precedence over <,<=,>=,>

© 2006 Wouter Verkerke, NIKHEF

Operators and expressions – Logical operators

• Logical operators

Name Operator
Logical NOT !x
Logical AND x>3 && x<5 Do not confuse
with bit-wise AND (&)
Logical OR x==3 || x==5 and bit-wise OR (|)

• All logical operators take bool arguments and return bool

– If input is not bool it is converted to bool
– Zero of any type maps to false, anything else maps to true

• Logical operators are evaluated from left to right

– Evaluation is guaranteed to stop as soon as outcome is determined

float x, y ;
…
if (y!=0. && x/y < 5.2) // safe against divide
© 2006 Wouter
by zero
Verkerke, NIKHEF
Operators and expressions – Bitwise operators

• Bitwise operators

Name Operator Example

Bitwise complement ~x 0011000  1100111
Left shift x << 2 000001  000100
Right shift x >> 3 111111  000111
Bitwise AND x & y 1100 & 0101 = 0100
Bitwise OR x | y 1100 | 0101 = 1101
Bitwise XOR x ^ y 1100 ^ 0101 = 1001

• Remarks
– Bitwise operators cannot be applied to floating point types
– Mostly used in online, DAQ applications where memory is limited
and ‘bit packing is common’
– Do not confuse logical or, and (||,&&) with bitwise or, and (|,&)

© 2006 Wouter Verkerke, NIKHEF

Operators and expressions – Assignment operators

• Assignment operators

Name Operator
Assignment x = 5
Addition update x += 5
Subtraction update x -= 5
Multiplication update x *= 5
Division update x /= 5
Modulus update x %= 5
Left shift update x <<= 5
Right shift update x >>= 5
Bitwise AND update x &= 5
Bitwise OR update x |= 5
Bitwise XOR update x ^= 5

© 2006 Wouter Verkerke, NIKHEF

Operators and expressions – Assignment operators

• Important details on assignment operators

– Left-hand arguments must be lvalues (naturally)
– Assignment is evaluated right to left
– Assignment operator returns left-hand value of expression

• Return value property of assignment has important

consequences
– Chain assignment is possible!

x = y = z = 5 // OK! x = ( y = ( z = 5 ))
// x = ( y = 5)
// x = 5

– Inline assignment is possible

int x[5], i ;
x[i=2] = 3 ; // i is set to 2, x[2] is set to 3

© 2006 Wouter Verkerke, NIKHEF

Operators and expressions – Miscellaneous

• Inline conditional expression: the ternary ?: operator

– Executes inline if-then-else conditional expression

int x = 4 ;
cout << ( x==4 ? ”A” : ”B” ) << endl ; // prints “A” ;

• The comma operator (expr1, expr2, expr3)

– Evaluates expressions sequentially, returns rightmost expression

int i=0, j=1, k=2 ;

cout<< (i=5, j=5, k) <<endl ; // Prints ‘2’, but i,j set to 5

• The sizeof operator

– Returns size in bytes of operand, argument can be type or symbol

int size1 = sizeof(int) ; // =4 (on most 32-bit archs)

double x[10]
© 2006 Wouter Verkerke, NIKHEF
int size2 = sizeof(x) ; // =10*sizeof(double)
Conversion operators
• Automatic conversion
– All type conversion that can be done ‘legally’ and without loss of
information are done automatically
– Example: float to double conversion

float f = 5 ;
double d = f ; // Automatic conversion occurs here

• Non-trivial conversions are also possible, but not

automatic
– Example: float to int, signed int to unsigned int
– If conversion is non-trivial, conversion is not automatic  you
must it request it with a conversion operator

• C++ has a variety of ways to accomplish conversions

– C++ term for type conversion is ‘cast’
– Will focus on ‘modern’ methods and ignore ‘heritage’ methods

© 2006 Wouter Verkerke, NIKHEF

Conversion operators – Explicit casts

• For conversions that are ‘legal’ but may result in

truncation, loss of precision etc…: static_cast

float f = 3.1
int i = static_cast<int>(f) ; // OK, i=3 (loss of precision)
int* i = static_cast<int*>(f) ; // ERROR float != pointer

• For conversions from ‘const X’ to ‘X’, i.e. to override a

logical const declaration: const_cast

float f = 3.1 ;
const float& g = f ;
g = 5.3 ; // ERROR not allowed, g is const
float& h = const_cast<float&>(g) ; // OK g and h of same type
h = 5.3 ; // OK, h is not const

© 2006 Wouter Verkerke, NIKHEF

Conversion operators – Explicit casts

• Your last resort: reinterpret_cast

float* f ;
int* i = reintepret_cast<int*>(f) ; // OK, but you take
// responsibility for the ensuing mess…

• You may need more than one cast to do your

conversion

const float f = 3.1

int i = static_cast<int>(f) ; // ERROR static_cast cannot
// convert const into non-const

const float f = 3.1

int i = static_cast<int>( const_cast<float>(f) ) ; // OK

– It may look verbose it helps you to understand your code

as all aspects of the conversion are explicitly spelled out
© 2006 Wouter Verkerke, NIKHEF
Control flow constructs – if/else

• The if construct has three formats

– Parentheses around expression required
– Brackets optional if there is only one statement (but put them anyway)

if (expr) {
statements ; // evaluated if expr is true
}

if (expr) {
statements ; // evaluated if expr is true
} else {
statements ; // evaluated if expr is false
}

if (expr1) {
statements ; // evaluated if expr1 is true
} else if (expr2) {
statements ; // evaluated if expr2 is true
} else {
statements ; // evaluated if neither expr is true
} © 2006 Wouter Verkerke, NIKHEF
Intermezzo – coding style

• C++ is free-form so there are no rules

• But style matters for readability, some suggestions
– One statement per line
– Always put {} brackets even if statement is single line
– Common indentation styles for {} blocks

if (foo==bar) { if (foo==bar)
statements ; {
} else { statements ;
statements ; }
} else
{
statements ;
}

Try to learn yourself this style,

it is more compact and more
readable (especially when you’re
more experienced) © 2006 Wouter Verkerke, NIKHEF
Control flow constructs – while

• The while construct

while (expression) {
statements ;
}

– Statements will be executed if expression is true

– At end, expression is re-evaluated. If again true, statements are
again executed

• The do/while construct

do {
statements ;
} while (expression) ;

– Similar to while construct except that statements are always

executed once before expression is evaluated for the first time

© 2006 Wouter Verkerke, NIKHEF

Control flow constructs – for

• The for construct

for (expression1 ; expression2 ; expression3 ) {

statements ;
}

– is equivalent to

expression1 ;
while (expression2) {
statements ;
expression3 ;
}

• Most common looping construct

int i ;
for (i=0 ; i<5 ; i++) {
// Executes with i=0,1,2,3 and 4
}
© 2006 Wouter Verkerke, NIKHEF
Control flow constructs – for

• Expressions may be empty

for (;;) {
cout << “Forever more” << endl ;
}

• Comma operator can be useful to combine multiple

operations in expressions

int i,j
for (i=0,j=0 ; i<3 ; i++,j+=2) {
// execute with i=0,j=0, i=1,j=2, i=2,j=4
}

© 2006 Wouter Verkerke, NIKHEF

Control flow constructs – break and continue

• Sometimes you need to stop iterating a do, do/while or

for loop prematurely
– Use break and continue statements to modify control flow

• The break statement

– Terminate loop construct immediately

int i = 3 ;
while(true) { // no scheduled exit from loop
i -= 1 ;
if (i<0) break ; // exit loop
cout << i << endl ;
}

– Example prints ‘2’, ’1’ and ’0’. Print statement for i=-1 never
executed
© 2006 Wouter Verkerke, NIKHEF
Control flow constructs – break and continue
• The continue statement
– Continue stops execution of loops statements and returns to
evaluation of conditional expression

char buf[12] = “abc,def,ghi”

for (int i=0 ; i<12 ; i++) {
if (buf[i]==‘,’) continue ; // return to for()
// if ‘,’ is encountered
cout << buf[i] ;
}
cout << endl ;

– Output of example ‘abcdefghi’

– Do not confuse with FORTRAN ‘continue’ statement -- Very
different meaning!

• Both break and continue only affect the innermost

loop
– When you are using nested loops
© 2006 Wouter Verkerke, NIKHEF
Control flow constructs -- switch
• The switch construct
switch (expr) {
case constant1:
statements ; // Evaluated if expr==const1
break ;

case constant2:
case constant3:
statements ; // Evaluated if expr==const2 or const3
break ;

default:
statements ; // Evaluated expression matched none
break ;
}

– Most useful for decision tree algorithms

• If break is omitted execution continues with next case

evaluation
– Usually you don’t want this, so watch the©breaks
2006 Wouter Verkerke, NIKHEF
 Control flow constructs – switch (example)
enum color { red=1, green=2, blue=4 };
• switch works very color paint = getcolor() ;
elegantly with enum types switch (paint) {
case red:
– enum naturally has finite set case green:
of states case blue:
cout << “primary color” << endl ;
break ;
• case expressions must be
case red+green:
constant but can be any cout << “yellow” << end;
valid expression break ;
– Example: case red+blue:
cout << “magenta” << end;
break ;

case blue+green:
cout << “cyan” << end;
break ;

default:
cout << “white” << endl ;
break ;
© 2006 Wouter Verkerke, NIKHEF
}
Some details on the block {} statements

• Be sure to understand all consequences of a block {}

– The lifetime of automatic variables inside the block is limited to
the end of the block (i.e up to the point where the } is
encountered)

main() {
int i = 1 ;

if (x>0) { Memory for

int i = 0 ; ‘int i’ allocated
Memory for // code
‘int i’ released } else {
// code
}
}

– A block introduces a new scope : it is a separate name space in

which you can define new symbols, even if those names already
existed in the enclosing block

© 2006 Wouter Verkerke, NIKHEF

Scope – more symbol visibility in {} blocks

• Basic C++ scope rules for variable definitions

– In given location all variables defined in local scope are visible
– All variables defined in enclosing scopes are visible
– Global variables are always visible
– Example

int a ;
int main() {
int b=0 ;

a, b visible if (b==0) {
a, b, c visible
int c = 1;
}

© 2006 Wouter Verkerke, NIKHEF

Scoping rules – hiding

• What happens if two variables declared in different

scopes have the same name?
– Definition in inner scope hides definition in outer scope
– It is legal to have two variables with the same name defined in
different scopes

int a ;
int main() {
int b=0 ;
‘b’ declared in main() visible
if (b==0) {
LEGAL! int b = 1; ‘b’ declared in if() visible
} ‘b’ declared in main() hidden!

} int main() {
int b ;
– NB: It is not legal to have two definitions …
of the same name in the same scope, e.g. int b ; ERROR!
}
© 2006 Wouter Verkerke, NIKHEF
Scoping rules – The :: operator

• Global variables, even if hidden, can always be accessed

using the scope resolution operator ::

int a=1 ;

int main() {
int a=0 ; LEGAL, but hides global ‘a’

::a = 2 ; Changes global ‘a’

}

• No tools to resolve symbols from intermediate unnamed

scope
– Solution will be to use ‘named’ scopes: namespaces or classes
– More on classes later

© 2006 Wouter Verkerke, NIKHEF

More on memory use

• By default all objects defined outside {} blocks (global

objects) are allocated statically
– Memory allocated before execution of main() begins
– Memory released after main() terminates

• By default all defined objects defined inside {} blocks

are ‘automatic’ variables
– Memory allocated when definition occurs
– Memory released when closing bracket of scope is encountered
if (x>0) {
int i = 0 ; Memory for
// code ‘int i’ allocated
Memory for }
‘int i’ released
– You can override behavior of variables declared in {} blocks
to be statically allocated using the static keyword
© 2006 Wouter Verkerke, NIKHEF
More on memory allocation

• Example of static declaration

void func(int i_new) {

static int i = 0 ;
cout << “old value = “ << i << endl ;
i = i_new ;
cout << “new value = “ << i << endl ;
}

main() {
func(1) ;
func(2) ;
}

– Output of example

old value = 0 ;
new value = 1 ;
old value = 1 ; Value of static int i preserved between func() calls
new value = 2 ;
© 2006 Wouter Verkerke, NIKHEF
Dynamic memory allocation

• Allocating memory at run-time

– When you design programs you cannot always determine how
much memory you need
– You can allocate objects of size unknown at compile time using
the ‘free store’ of the C++ run time environment

• Basic syntax of runtime memory allocation

– Operator new allocates single object, returns pointer
– Operator new[] allocates array of objects, returns pointer

// Single objects
Type* ptr = new Type ;
Type* ptr = new Type(initValue) ;

// Arrays of objects
Type* ptr = new Type[size] ;
Type* ptr = new Type[size][size]…[sizeN] ;
© 2006 Wouter Verkerke, NIKHEF
Releasing dynamic memory allocation

• Operator delete releases dynamic memory allocated

with new
// Single objects
delete ptr ;

// Arrays of objects
delete[] ptr ;

– Be sure to use delete[] for allocated arrays. A mismatch will

result in an incomplete memory release

• How much memory is available in the free store?

– As much as the operating system lets you have
– If you ask for more than is available your program will terminate
in the new operator
– It is possible to intercept this condition and continue the program
using ‘exception handling’ (we’ll discuss this later)
© 2006 Wouter Verkerke, NIKHEF
Dynamic memory and leaks

• A common problem in programs are memory leaks

– Memory is allocated but never released even when it is not used
anymore
– Example of leaking code

void leakFunc() {
int* array = new int[1000] ;
// do stuff with array
} Leak happens right here
we loose the pointer array
here and with that our only
int main() { possibility to release memory
int i ; in future
for (i=0 ; i<1000 ; i++) {
leakFunc() ; // we leak 4K at every call
}
}

© 2006 Wouter Verkerke, NIKHEF

Dynamic memory and leaks

• Another scenario to leak memory

– Misunderstanding between two functions

int* allocFunc() {
int* array = new int[1000] ;
// do stuff with array
return array ; allocFunc() allocates memory
} but pointer as return value
memory is not leaked yet
int main() {
int i ;
for (i=0 ; i<1000 ; i++) {
allocFunc() ; Author of main() doesn’t know
} that it is supposed to delete
} array returned by allocFunc()

Leak occurs here, pointer to dynamically

allocated memory is lost before memory
is released
© 2006 Wouter Verkerke, NIKHEF
Dynamic memory and ownership

• Avoiding leaks is a matter of good bookkeeping

– All memory allocated should be released again

• Memory handling logistics usually described in terms of

ownership
– The ‘owner’ of dynamically allocated memory is responsible for
releasing the memory again
– Ownership is a ‘moral concept’, not a C++ syntax rule. Code
that never releases memory it allocated is legal, but may not work
well as program size will increase in an uncontrolled way over
time
– Document your memory management code in terms of ownership

© 2006 Wouter Verkerke, NIKHEF

Dynamic memory allocation

• Example of dynamic memory allocation with ownership

semantics
– Less confusion about division of responsabilities

int* makearray(int size) {

// NOTE: caller takes ownership of memory
int* array = new int(size) ;

int i ;
for (i=0 ; i<size ; i++) {
array[i] = 0 ;
}
}

int main() {
// Note: We own array ;
int* array = makearray(1000) ;

delete[] array ;
}
© 2006 Wouter Verkerke, NIKHEF
Files and Functions

2 Files and
Functions

© 2006 Wouter Verkerke, NIKHEF

Introduction – Files, functions and namespaces

• Contents of this chapter

– Encapsulating algorithms – functions

– Splitting your code into modules – working with multiple files

– Decoupling symbols in modules – namespaces

– Working with existing modules – The Standard Library

© 2006 Wouter Verkerke, NIKHEF

Structured programming – Functions
• Functions group statements into logical units
– Functions encapsulate algorithms

• Declaration

TYPE function_name(TYPE arg1, TYPE arg2, TYPE argN) ;

• Definition:

TYPE function_name(TYPE arg1, TYPE arg2, TYPE argN) {

// body
statements ;
return arg ;
}

• Ability to declare function separate from definition important

– Allows to separate implementation and interface
– But also solves certain otherwise intractable problems

© 2006 Wouter Verkerke, NIKHEF

Forward declaration of functions

• Example of trouble using function definitions only

int g() {
f() ; // g calls f – ERROR, f not known yet
}

int f() {
g() ; // f calls g – OK g is defined
}

– Reversing order of definition doesn’t solve problem,

– But forward declaration does solve it:

int f(int x) ;

int g() {
f(x*2) ; // g calls f – OK f declared now
}

int f(int x) {
g() ; // f calls g – OK g defined by now
}
© 2006 Wouter Verkerke, NIKHEF
Functions – recursion

• Recursive function calls are explicitly allowed in C++

– Example
int factorial(int number) {

if (number<=1) {
return number ;
}

return number*factorial(number-1) ;
}

– NB: Above example works only in pass-by-value implementation

• Attractive solution for inherently recursive algorithms

– Recursive (directory) tree searches, etc…

© 2006 Wouter Verkerke, NIKHEF

Function arguments
• Function input and return arguments are both optional
– Function with no input arguments: TYPE function() ;
– Function with no return argument: void function(TYPE arg,…) ;
– Function with neither: void function() ;

• Pseudo type void is used as place holder when no argument

is returned
• Returning a value
– If a function is declared to return a value, a value must be return using the
‘return <value>’ statement
– The return statement may occur anywhere in the function body,
but every execution path must end in a return statement

int func() { void func() {

if (special_condition) { if (special_condition) {
return -1 ; return ;
} }
return 0; return ; // optional
} }

– Void functions may terminate early using ‘return ;’

© 2006 Wouter Verkerke, NIKHEF
Function arguments – values

• By default all functions arguments are passed by value

– Function is passed copies of input arguments
void swap(int a, int b) ;

main() {
int a=3, b=5 ;
swap(a,b) ;
cout << “a=“ << a << “, b=“ << b << endl ;
}

void swap(int a, int b) {

int tmp ;
tmp = a ; a and b in swap() are copies of
a = b ; a and b in main()
b = tmp ;
}
// output = “a=3, b=5”

– Allows function to freely modify inputs without consequences

– Note: potentially expensive when passing©large objects (arrays)
2006 Wouter Verkerke, NIKHEF
by value is expensive!
Function arguments – references
• You can change this behavior by passing references as
input arguments

void swap(int& a, int& b) ;

main() {
int a=3, b=5 ;
swap(a,b) ;
cout << “a=“ << a << “, b=“ << b << endl ;
}

void swap(int& a, int& b) {

int tmp ;
tmp = a ; a and b in swap() are references to
a = b ; original a and b in main(). Any operation
b = tmp ; affects originals
}
// Output “a=5, b=3”

– Passing by reference is inexpensive, regardless of size of object

– But allows functions to modify input arguments which may have
potentially further consequences © 2006 Wouter Verkerke, NIKHEF
Function arguments – const references

• Functions with ‘const references’ take references but

promise not to change the object

void swap(const int& a, const int& b) {

int tmp ;
tmp = a ; // OK – does not modify a
a = b ; // COMPILER ERROR – Not allowed
b = tmp ; // COMPILER ERROR – Not allowed
}

• Use const references instead of ‘pass-by-value’ when

you are dealing with large objects that will not be
changed
– Low overhead (no copying of large overhead)
– Input value remains unchanged (thanks to const promise)

© 2006 Wouter Verkerke, NIKHEF

Function arguments – pointers

• You can of course also pass pointers as arguments

void swap(int* a, int* b) ;

main() {
int a=3, b=5 ;
swap(&a,&b) ;
cout << “a=“ << a << “, b=“ << b << endl ;
}

void swap(int* a, int* b) {

int tmp ;
tmp = *a ; a and b in swap() are pointers to
*a = *b ; original a and b in main(). Any operation
*b = tmp ; affects originals
}
// Output “a=5, b=3”

– Syntax more cumbersome, use references when you can, pointers

only when you have to
© 2006 Wouter Verkerke, NIKHEF
Function arguments – references to pointers

• Passing a pointer by reference allows function to modify

pointers
– Often used to pass multiple pointer arguments to calling function

bool allocate(int*& a, int*& b) ;

main() {
int* a(0), *b(0) ;
bool ok = allocate(a,b) ;
cout << a << “ “ << b << endl ;
// prints 0x4856735 0x4927847
}

bool allocate(int*& a, int*& b) {

a = new int[100] ;
b = new int[100*100] ;
return true ;
}

– NB: reverse is not allowed – you can’t make a pointer to a

reference as a reference is not (necessarily) an object in memory
© 2006 Wouter Verkerke, NIKHEF
Function arguments – arrays
• Example of passing a 1D arrays as argument

void square(int len, int array[]) ;

main() {
int array[3] = { 0, 1, 2 } ;
square( sizeof(array)/sizeof(int), array) ;
return 0 ;
}

void square (int len, int array[]) { Remark:

while(--len>=0) { Code exploits that len is a copy
array[len] *= array[len] ; Note prefix decrement use
Note use of *= operator
}
}

• Remember basic rule: array = pointer

– Writing ‘int* array’ is equivalent to ‘int array[]’
– Arrays always passed ‘by pointer’
– Need to pass length of array as separate argument
© 2006 Wouter Verkerke, NIKHEF
Function arguments – multi-dimensional arrays

• Multi-dimensional array arguments more complicated

– Reason: interpretation of array memory depends on dimension
sizes (unlike 1D array)

• Must specify all dimensions except 1st one in declaration

– Example
Logical layout
// arg of f declared to be N x 10 array
y
void f(int [][10]) ;

// Pass 5 x 10 array
int a[5][10] ;
x
f(a) ;

void f(int p[][10]) {

// inside f use 2-D array as usual
… p[i][j]… Memory layout
}

© 2006 Wouter Verkerke, NIKHEF

Function arguments – char[] (strings)

• Since char* strings are zero terminated by convention,

no separate length argument needs to be passed
– Example

void print(const char* str) ;

int main(){
const char* foo = “Hello World” ;
print(foo) ;
return 0 ;
}

void print (const char* str) {

const char* ptr = str ;
while (*ptr != 0) {
cout << *ptr << endl ;
ptr++ ;
}
}

© 2006 Wouter Verkerke, NIKHEF

Function arguments – main() and the command line

• If you want to access command line arguments you can

declare main() as follows
Array of (char*)

int main(int argc, char* argv[]) {

int i ;
for (i=0 ; i<argc ; i++) {
// argv[i] is ‘char *’
cout << “arg #” << i << “ = “ << argv[i] << endl ;
}
}
– Second argument is array of pointers

• Output of example program

unix> cc –o foo foo.cc
unix> foo Hello World
arg #0 = foo
arg #1 = Hello
arg #2 = World
© 2006 Wouter Verkerke, NIKHEF
Functions – default arguments
• Often algorithms have optional parameters with default values
– How to deal with these in your programs?

• Simple: in C++ functions, arguments can have default values

void f(double x = 5.0) ;
void g(double x, double y=3.0) ;
const int defval=3 ;
void h(int i=defval) ;

main() {
double x(0.) ;

f() ; // calls f(5.0) ;

g(x) ; // calls g(x,3.0) ;
g(x,5.0) ; // calls g(x,5.0) ;
h() ; // calls h(3) ;
}
• Rules for arguments with default values
– Default values can be literals, constants, enumerations or statics
– Positional rule: all arguments without default values must appear to the left
© 2006 Wouter Verkerke, NIKHEF
of all arguments with default values
Function overloading

• Often algorithms have different implementations with

the same functionality
int minimum3_int(int a, int b, int c) {
return (a < b ? ( a < c ? a : c ) : ( b < c ? b : c) ) ;
}

float minimum3_float(float a, float b, float c) {

return (a < b ? ( a < c ? a : c ) : ( b < c ? b : c) ) ;
}

int main() {
int a=3,b=5,c=1 ;
float x=4.5,y=1.2,z=-3 ;

int d = minimum3_int(a,b,c) ;
float w = minimum3_float(x,y,z) ;
}

– The minimum3 algorithm would be easier to use if both

implementations had the same name and the compiler would
automatically select the proper implementation with each use
© 2006 Wouter Verkerke, NIKHEF
Function overloading

• C++ function overloading does exactly that

– Reimplementation of example with function overloading

int minimum3(int a, int b, int c) {

return (a < b ? ( a < c ? a : c )
: ( b < c ? b : c) ) ; Overloaded
} functions have
same name,
float minimum3 (float a, float b, float c) { but different
return (a < b ? ( a < c ? a : c ) signature
: ( b < c ? b : c) ) ; (list of arguments)
}

int main() {
int a=3,b=5,c=1 ;
float x=4.5,y=1.2,z=-3 ;
Code calls same function name
int d = minimum3(a,b,c) ;
twice. Compiler selects appropriate
float w = minimum3(x,y,z) ;
overloaded function based on
}
argument list
© 2006 Wouter Verkerke, NIKHEF
Function overloading resolution

• How does the compiler match a list of arguments to an

implementation of an overloaded function? It tries
Rank Method Example
1 Exact Match
2 Trivial argument conversion int  int&
3 Argument Promotion float  double
4 Standard argument conversion intbool, doublefloat
5 User defined argument conversion (we’ll cover this later)

– Example

void func(int i) ;
void func(double d) ;

int main() {
int i ;
float f ;

func(i) ; // Exact Match

func(f) ; // Promotion to double
© 2006 Wouter Verkerke, NIKHEF
Function overloading – some fine points

• Functions can not be overloaded on return type

int function(int x) ;
float function(int x) ; // ERROR – only return type is different

• A call to an overloaded function is only legal if there is

exactly one way to match that call to an implementation
void func(bool i) ;
void func(double d) ;

int main() {
bool b ;
int i ;
float f ;

func(b) ; // Exact match

func(f) ; // Unique Promotion to double
func(i) ; // Ambiguous Std Conversion (intbool or intdouble)

– Its gets more complicated if you have >1 arguments

© 2006 Wouter Verkerke, NIKHEF
Pointers to functions

• You can create pointer to functions too!

– Declaration

Type (*pfname)(Type arg1, Type arg2,…) ;

– Example

double square(double x) {
return x*x ;
}

int main() {
double (*funcptr)(double i) ; // funcptr is function ptr
funcptr = &square ;

cout << square(5.0) << endl ; // Direct call

cout << (*funcptr)(5.0) << endl ; // Call via pointer
}

– Allows to pass function as function argument, e.g. to be used as

callback function © 2006 Wouter Verkerke, NIKHEF
Pointers to functions – example use

• Example of pointer to function – call back function

void sqrtArray(double x[], int len, void (*handler)(double x)) {

int i ;
for (i=0 ; i<len ; i++) {
if (x[i]<0) {
handler(x[i]) ; // call handler function if x<0
} else {
cout << "sqrt(" << x[i] << ") = " << sqrt(x[i]) << endl ;
}
}
}

void errorHandler(double x) {
cout << "something is wrong with input value " << x << endl ;
}

int main() {
double x[5] = { 0, 1, 2, -3, -4 } ;
sqrtArray(x , 5, &errorHandler) ;
}

© 2006 Wouter Verkerke, NIKHEF

Organizing your code into modules

• For all but the most trivial programs it is not convenient

to keep all C++ source code in a single file
– Split source code into multiple files

• Module: unit of source code offered to the compiler

– Usually module = file

• How to split your code into files and modules

1. Group functions with related functionality into a single file
• Follow guide line ‘strong cohesion’, ‘loose coupling’
• Example: a collection of char* string manipulation functions go together in a single
module

2. Separate declaration and definition in separate files

• Declaration part to be used by other modules that interact with given module
• Definition part only offered once to compiler for compilation

© 2006 Wouter Verkerke, NIKHEF

Typical layout of a module

• Declarations file

// capitalize.hh
void convertUpper(char* str) ;
void convertLower(char* str) ; Declarations

• Definitions file
// capitalize.cc
#include “capitalize.hh”
void convertUpper(char* ptr) { Definitions
while(*ptr) {
if (*ptr>=‘a’&&*ptr<=‘z’) *ptr -= ‘a’-’A’ ;
ptr++ ;
}
}
void convertLower(char* ptr) {
while(*ptr) {
if (*ptr>=‘A’&&*ptr<=‘Z’) *ptr += ‘a’-’A’ ;
ptr++ ;
} © 2006 Wouter Verkerke, NIKHEF
}
Using the preprocessor to include declarations

• The C++ preprocessor #include directive can be used

to include declarations from an external module

// demo.cc

#include “capitalize.hh”

int main(int argc, const char* argv[]) {

if (argc!=2) return 0 ;
convertUpper(argv[1]) ;
cout << argv[1] << endl ;
}

• But watch out for multiple inclusion of same source file

– Multiple inclusion can have unwanted effects or lead to errors
– Preferred solution: add safeguard in .hh file that gracefully
handles multiple inclusions
• rather than rely on cumbersome bookkeeping by module programming

© 2006 Wouter Verkerke, NIKHEF

Safeguarding against multiple inclusion

• Automatic safeguard against multiple inclusion

– Use preprocessor conditional inclusion feature

#ifdef NAME
(#else)
#endif
– NAME can be defined with #define

• Application in capitalize.hh example

– If already included, CAPITALIZE_HH is set and future inclusion will
be blank

// capitalize.hh
#ifndef CAPITALIZE_HH
#define CAPITALIZE_HH

void convertUpper(char* str) ;

void convertLower(char* str) ;

#endif © 2006 Wouter Verkerke, NIKHEF

Namespaces

• Single global namespace often bad idea

– Possibility for conflict: someone else (or even you inadvertently)
may have used the name want you use in your new piece of code
elsewhere  Linking and runtime errors may result
– Solution: make separate ‘namespaces’ for unrelated modules of
code

• The namespace feature in C++ allows you to explicitly

control the scope of your symbols
– Syntax: namespace name {
Code can access symbols
int global = 0 ;
inside same namespace
without further qualifications
void func() {
// code
cout << global << endl ;
}

}
© 2006 Wouter Verkerke, NIKHEF
Namespaces

• But code outside namespace must explicitly use scope

operator with namespace name to resolve symbol

namespace foo {

int global = 0 ;

void func() {
// code
cout << global << endl ;
}

void bar() {
cout << foo::global << endl ;

foo::func() ; Namespace applies to functions too!

}

© 2006 Wouter Verkerke, NIKHEF

Namespace rules

• Namespace declaration must occur at the global level

void function foo() {

namespace bar { ERROR!
statements
}
}

• Namespaces are extensible

namespace foo {
int bar = 0 ;
}

// other code

namespace foo { Legal

int foobar = 0 ;
}

© 2006 Wouter Verkerke, NIKHEF

Namespace rules

• Namespaces can nest

namespace foo {
int zap = 0 ;

namespace bar { Legal

int foobar = 0 ;
}

int main() {
cout << foo::zap << endl ;
cout << foo::bar::foobar << endl ;
}

Recursively use :: operator to resolve nested namespaces

© 2006 Wouter Verkerke, NIKHEF

Namespace rules

• Namespaces can be unnamed!

– Primary purpose: to avoid ‘leakage’ of private global symbols from
module of code

namespace {
int bar = 0 ;
}

void func() {
cout << bar << endl ;
}

Code in same module outside unnamed namespace

can access symbols inside unnamed namespace

© 2006 Wouter Verkerke, NIKHEF

Namespaces and the Standard Library

• All symbols in the Standard library are wrapped in the

namespace ‘std’
• The ‘Hello world’ program revisited:

// my first program in C++

#include <iostream>

int main () {
std::cout << "Hello World!“ << std::endl;
return 0;
}

© 2006 Wouter Verkerke, NIKHEF

Using namespaces conveniently

• It is possible to import symbols from a given

namespace into the current scope
– To avoid excessive typing and confusing due to repeated lengthy
notation

// my first program in C++

#include <iostream>
using std::cout ; Import selected symbols into global namespace
using std::endl ;

int main () {
cout << "Hello World!“ << endl;
return 0;
} Imported symbols can now be used
without qualification in this module

– Can also import symbols in a local scope. In that case import valid
only inside local scope
© 2006 Wouter Verkerke, NIKHEF
Using namespaces conveniently

• You can also import the symbol contents of an entire

namespace

// my first program in C++

#include <iostream>
using namespace std ;

int main () {
cout << "Hello World!“ << endl;
return 0;
}

• Style tip: If possible only import symbols you need

© 2006 Wouter Verkerke, NIKHEF

Modules and namespaces

• Namespace enhance encapsulation of modules

– Improved capitalize module
// capitalize.hh
#ifndef CAPITALIZE_HH
#define CAPITALIZE_HH
namespace capitalize {
void convertUpper(char* str) ;
void convertLower(char* str) ;
}
#endif // capitalize.cc
#include “capitalize.hh”
namespace capitalize {
void convertUpper(char* ptr) {
while(*ptr) {
if (*ptr>=‘a’&&*ptr<=‘z’) *ptr -= ‘a’-’A’ ;
ptr++ ;
}
}
void convertLower(char* ptr) {
while(*ptr) {
if (*ptr>=‘A’&&*ptr<=‘Z’) *ptr += ‘a’-’A’ ;
ptr++ ;
}
} © 2006 Wouter Verkerke, NIKHEF
}
The standard library as example

• Each C++ compiler comes with a standard suite of

libraries that provide additional functionality
– <math> -- Math routines sin(),cos(),exp(),pow(),…
– <stdlib> -- Standard utilities strlen(),strcat(),…
– <stdio> -- File manipulation utilities open(),write(),close(),…

• Nice example of modularity and use of namespaces

– All Standard Library routines are contained in namespace std

© 2006 Wouter Verkerke, NIKHEF

Compiling & linking code in multiple modules

• Compiling & linking code in a single module

– g++ -c demo.cc
• Converts demo.cc C++ code into demo.o (machine code)

– g++ -o demo demo.o

• Links demo.o with Standard Library code and makes standalone executable code

– Alternatively, ‘g++ -o demo demo.cc’ does all in one step

• Compiling & linking code in multiple modules

– g++ -c module1.cc
– g++ -c module2.cc
– g++ -c module3.cc
– g++ -o demo module1.o module2.o module3.o
• Link module1,2,3 to each other and the Standard Library code

© 2006 Wouter Verkerke, NIKHEF

Intermezzo – Concept of ‘object libraries’

• Operating systems usually also support concept ‘object

libraries’
– A mechanism to store multiple compile modules (.o files) into a
single library file
– Simplifies working with very large number of modules
– Example: all Standard Library modules are often grouped together
into a single ‘object library file’

• Creating an object library file from object modules

– Unix: ‘ar q libLibrary.a module1.o module2.o ….’

• Linking with a library of object modules

– Unix: ‘g++ -o demo demo.cc –lLibrary’
• Above syntax looks for library name libLibrary.a in current directory or in
‘standard locations’
• To add directory to library search path, specify –L<path> in g++ command line

© 2006 Wouter Verkerke, NIKHEF

Object-based programming – Classes

3 Class
Basics

© 2006 Wouter Verkerke, NIKHEF

Overview of this section

• Contents of this chapter

– structs and classes - Grouping data and functions together

– public vs private – Improving encapsulation through hiding of

internal details

– constructors and destructors – Improving encapsulation

through self-initialization and self-cleanup

– more on const – Improving modularity and encapsulation

through const declarations

© 2006 Wouter Verkerke, NIKHEF

Encapsulation

• OO languages like C++ enable you to create your own

data types. This is important because
– New data types make program easier to visualize and implement
new designs
– User-defined data types are reusable
– You may modify and enhance new data types as programs evolve
and specifications change
– New data types let you create objects with simple declarations

• Example

Window w ; // Window object

Database ood ; // Database object
Device d ; // Device object

© 2006 Wouter Verkerke, NIKHEF

Evolving code design through use of C++ classes

• Illustration of utility of C++ classes – Designing and

building a FIFO queue
– FIFO = ‘First In First Out’

• Graphical illustration of a FIFO queue

write read

‘S’ ‘A’ ‘Z’ ‘Q’ ‘W’ ‘L’

© 2006 Wouter Verkerke, NIKHEF

Evolving code design through use of C++ classes

• First step in design is to write down the interface

– How will ‘external’ code interact with our FIFO code?

write read
‘S’ ‘A’ ‘Z’ ‘Q’ ‘W’ ‘L’

• List the essential interface tasks

1. Create and initialize a FIFO
2. Write a character in a FIFO
3. Read a character from a FIFO
– Support tasks
1. How many characters are currently in the FIFO
2. Is a FIFO empty
3. Is a FIFO full
© 2006 Wouter Verkerke, NIKHEF
 Designing the C++ class FIFO – interface

• List of interface tasks

// Interface
1. Create and initialize a FIFO
void init() ;
2. Write a character in a FIFO void write(char c) ;
char read() ;
3. Read a character from a FIFO
int nitems() ;
• List desired support tasks bool full() ;
bool empty() ;
1. How many characters are
currently in the FIFO
2. Is a FIFO empty
3. Is a FIFO full

write read
‘S’ ‘A’ ‘Z’ ‘Q’ ‘W’ ‘L’

© 2006 Wouter Verkerke, NIKHEF

Designing the C++ struct FIFO – implementation

• Implement FIFO with array of elements

– Use index integers to keep track of front and rear, size of queue

// Implementation
char s[LEN] ;
‘A’
int rear ;
int front ;
‘Z’ int count ;

‘Q’

‘W’

© 2006 Wouter Verkerke, NIKHEF

Designing the C++ struct FIFO – implementation

• Implement FIFO with array of elements

– Use index integers to keep track of front and rear, size of queue
– Indices revolve: if they reach end of array, they go back to 0

// Implementation
void init() { front = rear = count = 0 ; }

void write(char c) { count++ ;

‘A’ if(rear==LEN) rear=0 ;
s[rear++] = c ; }
‘Z’
char read() { count-- ;
‘Q’ if (front==LEN) front=0 ;
return s[front++] ; }
‘W’

int nitems() { return count ; }

bool full() { return (count==LEN) ; }
bool empty() { return (count==0) ; }

© 2006 Wouter Verkerke, NIKHEF

Designing the C++ struct FIFO – implementation

• Animation of FIFO write operation

void write(char c) { count++ ;
if(rear==LEN) rear=0 ;
s[rear++] = c ; }

count=4 count=5 count=5

‘X’ ‘X’ rear=5

‘A’ rear=4 ‘A’ rear=4 ‘A’

‘Z’ ‘Z’ ‘Z’

‘Q’ ‘Q’ ‘Q’

‘W’ front=1 ‘W’ front=1 ‘W’ front=1

© 2006 Wouter Verkerke, NIKHEF

Designing the C++ struct FIFO – implementation

• Animation of FIFO read operation

char read() { count-- ;
if (front==LEN) front=0 ;
return s[front++] ; }

count=5 count=4 count=4

‘X’ rear=5 ‘X’ rear=5 ‘X’ rear=5

‘A’ ‘A’ ‘A’

‘Z’ ‘Z’ ‘Z’

‘Q’ ‘Q’ ‘Q’ front=2

‘W’ front=1 ‘W’ front=1 ‘W’

© 2006 Wouter Verkerke, NIKHEF

Putting the FIFO together – the struct concept

• The finishing touch: putting it all together in a struct

const int LEN = 80 ; // default fifo length

struct Fifo {
// Implementation
char s[LEN] ;
int front ;
int rear ;
int count ;

// Interface
void init() { front = rear = count = 0 ; }
int nitems() { return count ; }
bool full() { return (count==LEN) ; }
bool empty() { return (count==0) ; }
void write(char c) { count++ ;
if(rear==LEN) rear=0 ;
s[rear++] = c ; }
char read() { count-- ;
if (front==LEN) front=0 ;
return s[front++] ; }
} ; © 2006 Wouter Verkerke, NIKHEF
Characteristics of the ‘struct’ construct

• Grouping of data members facilitates storage allocation

– Single statement allocates all data members

// Allocate struct data type ‘Fifo’

Fifo f ;

// Access function through name ‘f’

f.init() ;

// Access data member through name ‘f’

cout << f.count << endl ;

• A struct organizes access to data members and

functions through a common symbolic name

© 2006 Wouter Verkerke, NIKHEF

Characteristics of the ‘struct’ construct

• Concept of ‘member functions’ automatically ties

manipulator functions to their data
– No need to pass data member operated on to interface function

// Solution without // Solution with

// member functions // member functions

struct fifo { struct fifo {

int front, rear, count ; int front, rear, count ;
} ; char read() {
count-- ;
char read_fifo(fifo& f) { …
f.count-- ; }
… } ;
}
fifo f ;
fifo f ; f.read() ;
read_fifo(f) ;

© 2006 Wouter Verkerke, NIKHEF

Using the FIFO example code

• Example code using the FIFO struct

const char* data = “data bytes” ;
int i, nc = strlen(data) ;

Fifo f ;
f.init() ; // initialize FIFO

// Write chars into fifo

const char* p = data ;
for (i=0 ; i<nc && !f.full() ; i++) {
f.write(*p++) ;
}

// Count chars in fifo

cout << f.nitems() << “ characters in fifo” << endl ;

// Read chars back from fifo

for (i=0 ; i<nc && !f.empty() ; i++) {
cout << f.read() << endl ; Program Output
} 10 characters in fifo
data ©bytes
2006 Wouter Verkerke, NIKHEF
Characteristics of the FIFO code
• Grouping data, function members into a struct
promotes encapsulation
– All data members needed for fifo operation allocated in a single
statement
– All data objects, functions needed for fifo operation have
implementation contained within the namespace of the FIFO
object
– Interface functions associated with struct allow implementation
of a controlled interface functionality of FIFO
• For example can check in read(), write() if FIFO is full or empty and
take appropriate action depending on status

• Problems with current implementation

– User needs to explicitly initialize fifo prior to use
– User needs to check explicitly if fifo is not full/empty when
writing/reading
– Data objects used in implementation are visible to user and
subject to external modification/corruption

© 2006 Wouter Verkerke, NIKHEF

Controlled interface

• Improving encapsulation
– We improve encapsulation of the FIFO implementation by
restricting access to the member functions and data members
that are needed for the implementation

• Objective – a controlled interface

– With a controlled interface, i.e. designated member functions that
perform operations on the FIFO, we can catch error conditions on
the fly and validate offered input before processing it
– With a controlled interface there is no ‘back door’ to the data
members that implement the fifo thus guaranteeing that no
corruption through external sources can take place
• NB: This also improves performance you can afford to be less paranoid.

© 2006 Wouter Verkerke, NIKHEF

Private and public
• C++ access control keyword: ‘public’ and ‘private’
struct Name {
private:

… members … // Implementation

public:

… members … // Interface

} ;
• Public data
– Access is unrestricted. Situation identical to no access control declaration

• Private data
– Data objects and member functions in the private section can only accessed
by member functions of the struct (which themselves can be either private
or public)

© 2006 Wouter Verkerke, NIKHEF

Redesign of fifo class with access restrictions

const int LEN = 80 ; // default fifo length

struct Fifo {
private: // Implementation
char s[LEN] ;
int front ;
int rear ;
int count ;

public: // Interface
void init() { front = rear = count = 0 ; }
int nitems() { return count ; }
bool full() { return (count==LEN) ; }
bool empty() { return (count==0) ; }
void write(char c) { count++ ;
if(rear==LEN) rear=0 ;
s[rear++] = c ; }
char read() { count-- ;
if (front==LEN) front=0 ;
return s[front++] ; }
} ;
© 2006 Wouter Verkerke, NIKHEF
Using the redesigned FIFO struct

• Effects of access control in improved fifo struct

Fifo f ;
f.init() ; // initialize FIFO

f.front = 5 ; // COMPILER ERROR – not allowed

cout << f.count << endl ; // COMPILER ERROR – not allowed

cout << f.nitems() << endl ; // OK – through

// designated interface

front is an implementation details that’s not part of the

abstract FIFO concept. Hiding this detail promotes encapsulation
as we are now able to change the implementation later
with the certainty that we will not break existing code

write read
‘S’ ‘A’ ‘Z’ ‘Q’ ‘W’ ‘L’

© 2006 Wouter Verkerke, NIKHEF

Class – a better struct

• In addition to ‘struct’ C++ also defines ‘class’ as a

method to group data and functions
– In structs members are by default public,
In classes member functions are by default private
– Classes have several additional features that we’ll cover shortly

Equivalent
struct Name { class Name {
private:
… members …
… members …
public:
public:
… members …
… members …
} ; } ;

© 2006 Wouter Verkerke, NIKHEF

Classes and namespaces

• Classes (and structs) also define their own namespace

– Allows to separate interface and implementation even further by
separating declaration and definition of member functions

Declaration and definition Declaration only

class Fifo { class Fifo {
public: // Interface public: // Interface
char read() { char read() ;
count-- ; } ;
if (front==len) front=0 ;
return s[front++] ;
}
} ; Definition
#include “fifo.hh”
char Fifo::read() {
count-- ;
Use of scope operator ::
if (front==len) front=0 ;
to specify read() function
return s[front++] ;
of Fifo class when outside
}
class definition
© 2006 Wouter Verkerke, NIKHEF
Classes and namespaces

• Scope resolution operator can also be used in class

member function to resolve ambiguities

class Fifo {
public: // Interface
char read() {
…
std::read() ;
…
}
} ; Use scope operator to specify that you want
to call the read() function in the std namespace
rather than yourself

© 2006 Wouter Verkerke, NIKHEF

Classes and files

• Class declarations and definitions have a natural

separation into separate files
– A header file with the class declaration
To be included by everybody that uses the class
– A definition file with definition fifo.hh
that is only offered once
to the compiler #ifndef FIFO_HH
#define FIFO_HH
– Advantage: You do not need to class Fifo {
recompile code using public: // Interface
class fifo if only implementation char read() ;
(file fifo.cc) changes } ;
#endif

fifo.cc
#include “fifo.hh”
char Fifo::read() {
count-- ;
if (front==len) front=0 ;
return s[front++] ;
}
© 2006 Wouter Verkerke, NIKHEF
Constructors

• Abstraction of FIFO data typed can be further enhanced

by letting it take care of its own initialization
– User should not need to know if and how initialization should
occur
– Self-initialization makes objects easier to use and gives less
chances for user mistakes

• C++ approach to self-initialization – the Constructor

member function
– Syntax: member function with function name identical to class
name

class ClassName {
…
ClassName() ;
…
} ;

© 2006 Wouter Verkerke, NIKHEF

Adding a Constructor to the FIFO example

• Improved FIFO example

class Fifo { class Fifo {

public: public:
void init() ; Fifo() { init() ; }
…
private:
void init() ;
…

• Simplified use of FIFO

Fifo f ; // creates raw FIFO

f.init() ; // initialize FIFO

Fifo f ; // creates initialized FIFO

© 2006 Wouter Verkerke, NIKHEF

Default constructors vs general constructors

• The FIFO code is an example of a default constructor

– A default constructor by definition takes no arguments

• Sometimes an objects requires user input to properly

initialize itself
– Example: A class that represents an open file – Needs file name
– Use ‘regular constructor’ syntax

class ClassName {
…
ClassName(argument1,argument2,…argumentN) ;
…
} ;

– Supply constructor arguments at construction

ClassName obj(arg1,…,argN) ;
ClassName* ptr = new ClassName(Arg1,…ArgN) ;
© 2006 Wouter Verkerke, NIKHEF
Constructor example – a File class
class File {

private:
int fh ;

public:
File(const char* name) {
fh = open(name) ;
}

void read(char* p, int n) { ::read(fh,p,n) ; }

void write(char* p, int n) { ::write(fh,p,n) ; }
void close() { ::close(fh) ; }
} ;

File* f1 = new File(“dbase”) ;

File f2(“records”) ; Supply constructor arguments here

© 2006 Wouter Verkerke, NIKHEF

Multiple constructors

• You can define multiple constructors with different

signatures
– C++ function overloading concept applies to class member
functions as well, including the constructor function

class File {

private:
Int fh ;

public:
File() {
fh = open(“Default.txt”) ;
}
File(const char* name) {
fh = open(name) ;
}

read(char* p, int n) { ::read(p,n) ; }

write(char* p, int n) { ::write(p,n) ; }
close() { ::close(fh) ; }
} ; © 2006 Wouter Verkerke, NIKHEF
Default constructor and default arguments

• Default values for function arguments can be applied to

all class member functions, including the constructor
– If any constructor can be invoked with no arguments (i.e. it has
default values for all arguments) it is also the default constructor

class File {

private:
Int fh ;

public:
File(const char* name=“Default.txt”) {
fh = open(name) ;
}

read(char* p, int n) { ::read(p,n) ; }

write(char* p, int n) { ::write(p,n) ; }
close() { ::close(fh) ; }
} ;

© 2006 Wouter Verkerke, NIKHEF

Default constructors and arrays
• Array allocation of objects does not allow for
specification of constructor arguments

Fifo* fifoArray = new Fifo[100] ;

• You can only define arrays of classes that have a

default constructor
– Be sure to define one if it is logically allowed
– Workaround for arrays of objects that need constructor
arguments: allocate array of pointers ;

Fifo** fifoPtrArray = new (Fifo*)[100] ;

int i ;
for (i=0 ; i<100 ; i++) {
fifoPtrArray[i] = new Fifo(arguments…) ;
}

– Don’t forget to delete elements in addition to array afterwards!

© 2006 Wouter Verkerke, NIKHEF
Classes contained in classes – member initialization

• If classes have other classes w/o default constructor as

data member you need to initialize ‘inner class’ in
constructor of ‘outer class’

class File {
public:
File(const char* name) ;
…
} ;

class Database {
public:
Database(const char* fileName) ;

private:
File f ;
} ;

Database::Database(const char* fileName) : f(fileName) {

// Database constructor
}
© 2006 Wouter Verkerke, NIKHEF
Class member initialization

• General constructor syntax with member initialization

ClassName::ClassName(args) :
member1(args),
member2(args), …
memberN(args) {
// constructor body
}

– Note that insofar order matters, data members are initialized in

the order they are declared in the class, not in the order they
are listed in the initialization list in the constructor
– Also for basic types (and any class with default ctor) the member
initialization form can be used
Initialization through assignment Initialization through constructor
File(const char* name) { File(const char* name) :
fh = open(name) ; fh(open(name)) {
} }
– Performance tip: for classes constructor initialization tends to be
faster than assignment initialization (more on this later)
© 2006 Wouter Verkerke, NIKHEF
Common initialization in multiple constructors

• Overlapping functionality is a common design issue with

multiple constructors
– How to avoid unnecessary code duplication (i.e member initialization)

• Common mistake – attempts to make one constructor

function call another one
class Array {
public:
Array(int size) {
_size = size ;
_x = new double[size] ;
} Not Allowed!!!
(Compiler Error)
Array(const double* input, int size) : Array(size) {
int i ;
for (i=0 ; i<size ; i++) _x[i] = input[i] ;
}

private
int _size ;
double* _x ; © 2006 Wouter Verkerke, NIKHEF
}
Common initialization in multiple constructors
• Another clever but wrong solution
– Idea: Call Array(size) as if it were a regular member function, which
will then perform the necessary initialization steps

Array(const double* input, int size) {

Array(size) ; // This doesn’t work either!

int i ;
for (i=0 ; i<size ; i++) _x[i] = input[i] ;
}

– Problem: It is legal C++ (it compiles fine) but it doesn’t do what you
think it does!
– Calling a constructor like this creates a temporary object that is
initialized with size and immediately destroyed again. It does not
initialize the instance of array you are constructing with the
Array(double*,int) constructor

© 2006 Wouter Verkerke, NIKHEF

Common initialization in multiple constructors

• The correct solution is to make a private initializer

function that is called from all relevant constructors
class Array {
public:
Array(int size) {
initialize(size) ;
}

Array(const double* input, int size) {

initialize(size) ;
int i ;
for (i=0 ; i<size ; i++) _x[i] = input[i] ;
}

private:
void initialize(int size) {
_size = size ;
_x = new double[size] ;
}
int _size ;
double* _x ;
© 2006 Wouter Verkerke, NIKHEF
}
Destructors

• Classes that define constructors often allocate dynamic

memory or acquire resources
– Example: File class acquires open file handles, any other class
that allocates dynamic memory as working space

• C++ defines Destructor function for each class to be

called at end of lifetime of object
– Can be used to release memory, resources before death

• Class destructor syntax:

class ClassName {
…
~ClassName() ;
…
} ;

© 2006 Wouter Verkerke, NIKHEF

Example of destructor in File class

class File {

private:
int fh ;
close() { ::close(fh) ; }

public:
File(const char* name) { fh = open(name) ; }
~File() { close() ; }
…
} ;
File is automatically closed
when object is deleted

void readFromFile() {
File *f = new File(“theFile.txt”) ; Opens file automatically
// read something from file
delete f ; Closes file automatically
}

© 2006 Wouter Verkerke, NIKHEF

Automatic resource control

• Destructor calls can take care of automatic resource

control
– Example with dynamically allocated File object

void readFromFile() {
File *f = new File(“theFile.txt”) Opens
; file automatically
// read something from file
delete f ; Closes file automatically
}

– Example with automatic File object

void readFromFile() {
File f(“theFile.txt”) ; Opens file automatically
// read something from file
} Deletion of automatic
variable f calls destructor
– Great example of abstraction of & closes file automatically
file concept and of encapsulation
of resource control © 2006 Wouter Verkerke, NIKHEF
Classes vs Instances – an important concept

• There is an important distinction between classes and

instances of classes (objects)
– A class is a unit of code
– An instance is an object in memory that is managed by the class
code

Array a ; // creates an array object

Class Instance

• A class can have more than one instance

Array a1 ; // first instance

Array a2 ; // second instance

© 2006 Wouter Verkerke, NIKHEF

Classes vs Instances – an important concept
• The concept that a single unit of code can work with multiple
objects in memory has profound consequences
– Start with program that makes two arrays like this

Array a1 ; // first instance

Array a2 ; // second instance

– Now what happens inside the array’s initialize() code

void Array::initialize(int size) {

_size = size ;
_x = new double[size] ;
}
– Q: To which memory object does data member _size belong, a1 or a2?
– A: It depends on who calls initialize()!

If you call a1.initialize() data member _size automatically refers to

a1._size, if you call a2.initialize() it refers to a2._size etc…
– Concept is called ‘automatic binding’

© 2006 Wouter Verkerke, NIKHEF

Intermezzo – Referring to yourself – this

• Can you figure which instance you are representing in a

member function? A: Yes, using the special object this
– The ‘this’ keyword return a pointer to yourself inside a member
function

void Array::initialize() {
cout << “I am a array object, my pointer is “ << this << endl ;
}

• How does it work?

– In case you called a1.initialize() from the main program,
this=&a1
– In case
you called a2.initialize() then this=&a2 etc…

© 2006 Wouter Verkerke, NIKHEF

Intermezzo – Referring to yourself – this
• You don’t need this very often.
– If you think you do, think hard if you can avoid it, you usually can

• Most common cases where you really need this are

– Identifying yourself to an outside function (see below)
– In assignment operations, to check that you’re not copying onto yourself
(e.g. a1=a1). We’ll come back to this later

• How to identify yourself to the outside world?

– Example: Member function of classA needs to call external function
externalFunc() that takes reference to classA

void externalFunction(ClassA& obj) {

…
}

void classA::memberFunc() {
if (certain_condition) {
externFunction(*this) ;
}
} © 2006 Wouter Verkerke, NIKHEF
Copy constructor – a special constructor
• The copy constructor is the constructor with the
signature

ClassA::ClassA(const ClassA&) ;

• It is used to make a clone of your object

ClassA a ;
ClassA aclone(a) ; // aclone is an identical copy of a

• It exists for all objects because the C++ compiler

provides a default implementation if you don’t supply
one
– The default copy constructor calls the copy constructor for all data
members. Basic type data members are simply copied
– The default implementation is not always right for your class, we’ll
return to this shortly

© 2006 Wouter Verkerke, NIKHEF

Taking good care of your property

• Use ‘ownership’ semantics in classes as well

– Keep track of who is responsible for resources allocated by your
object
– The constructor and destructor of a class allow you to
automatically manage your initialization/cleanup
– All private resources are always owned by the class so make sure
that the destructor always releases those

• Be careful what happens to ‘owned’ objects when you

make a copy of an object
– Remember: default copy constructor calls copy ctor on all class
data member and copies values of all basic types
– Pointers are basic types
– If an ‘owned’ pointer is copied by the copy constructor it is no
longer clear which instance owns the object  danger ahead!

© 2006 Wouter Verkerke, NIKHEF

Taking good care of your property

• Example of default copy constructor wreaking havoc

class Array {
public:
Array(int size) {
initialize(size) ;
}
~Array() {
delete[] _x ;
}

private:
void initialize(int size) {
_size = size ;
_x = new double[size] ;
}
int _size ;
double* _x ; Watch out! Pointer data member
}

© 2006 Wouter Verkerke, NIKHEF

Taking good care of your property

• Add illustration
void example { Array a
double[]
Array a(10) ; _x
// ‘a’ Constructor allocates _x ;

if (some_condition)
Array b(a) ; Array b
// ‘b’ Copy Constructor does
// b._x = a._x ; _x

// b appears to be copy of a
}
// ‘b’ Destructor does Array a


// delete[] _b.x
_x Problem is here:
// BUT _b.x == _a.x  Memory b._x points to
// allocated by ‘Array a’ has same array
// been released by ~b() ; as a._x!
<Do something with Array>
// You are dead!
} © 2006 Wouter Verkerke, NIKHEF
Taking good care of your property

• Example of default copy constructor wreaking havoc

class Array { void example {
public:
Array(int size) { Array a(10) ;
initialize(size) ; // ‘a’ Constructor allocates _x ;
}
~Array() { if (some_condition)
delete[] _x ; Array b(a) ;
} // ‘b’ Copy Constructor does
Whenever your class owns dynamically allocated
// b._x = a._x ;
private: memory or similar resources you need to implement
your{ own copy
void initialize(int size) // bconstructor!
appears to be copy of a
_size = size ; }
_x = new double[size] ; // ‘b’ Destructor does
} // delete[] _b.x
int _size ;
double* _x ; // BUT _b.x == _a.x  Memory
} // allocated by ‘Array a’ has
// been released by ~b() ;

<Do something with Array>

// You are dead!
} © 2006 Wouter Verkerke, NIKHEF
Example of a custom copy constructor
class Array {
public:
Array(int size) {
initialize(size) ;
}

Array(const double* input, int size) {

initialize(size) ;
int i ;
for (i=0 ; i<size ; i++) _x[i] = input[i] ;
}
Classes vs Instances
Array(const Array& other) {
initialize(other._size) ; Here we are dealing
int i ; explicitly with one
for (i=0 ; i<_size ; i++) _x[i] = other._x[i] ; class and two instances
}
Symbol other._x
private:
void initialize(int size) {
refers to data
Symbol _x refers
_size = size ; member of other
to data member
_x = new double[size] ; instance
of this instance
}
int _size ;
double* _x ;
} © 2006 Wouter Verkerke, NIKHEF
Another solution to copy constructor problems

• You can disallow objects being copied by declaring their

copy constructor as ‘private’
– Use for classes that should not copied because they own non-
clonable resources or have a unique role
– Example: class File – logistically and resource-wise tied to a
single file so a clone of a File instances tied to the same file
make no sense

class File {

private:
Int fh ;
close() { ::close(fh) ; }
File(const File&) ; // disallow copying

public:
File(const char* name) { fh = open(name) ; }
~File() { close() ; }
…
} ;
© 2006 Wouter Verkerke, NIKHEF
Ownership and defensive programming
• Coding mistakes happen, but by programming
defensively you will spot them easier
– Always initialize owned pointers to zero if you do not allocate your
resources immediately
– Always set pointers to zero after you delete the object they point
to

• By following these rules you ensure that you never have

‘dangling pointers’
– Dangling pointers = Pointers pointing to a piece memory that is
no longer allocated which may return random values
– Result – more predictable behavior
– Dereferencing a dangling pointer may
• Work just fine in case the already released memory has not been overwritten yet
• Return random results
• Cause your program to crash

– Dereferencing a zero pointer will always terminate your program

immediately in a clean and understandable way

© 2006 Wouter Verkerke, NIKHEF

Const and Objects
• ‘const’ is an important part of C++ interfaces.
– It promotes better modularity by enhancing ‘loose coupling’

• Reminder: const and function arguments

void print(int value) // pass-by-value, value is copied

void print(int& value) // pass-by-reference,

print may change value
void print(const int& value) // pass-by-const-reference,
print may not change value

• Const rules simple to enforce for basic types: ‘=‘ changes

contents
– Compile can look for assignments to const reference and issue error
– What about classes? Member functions may change contents, difficult
to tell?
– How do we know? We tell the compiler which member functions
change the object! © 2006 Wouter Verkerke, NIKHEF
Const member functions

• By default all member functions of an object are

presumed to change an object
– Example

class Fifo {
…
void print() ;
…
}

main() {
Fifo fifo ;
showTheFifo(fifo) ;
}

void showTheFifo(const Fifo& theFifo)

{
theFifo.print() ; // ERROR – print() is allowed
// to change the object
}

© 2006 Wouter Verkerke, NIKHEF

Const member functions

• Solution: declare print() to be a member function that

does not change the object

class Fifo { A member function is declared

… const by putting ‘const’ behind
void print() const ; the function declaration
…
}

main() {
Fifo fifo ;
showTheFifo(fifo) ;
}

void showTheFifo(const Fifo& theFifo)

{
theFifo.print() ; // OK print() does not change object
}

© 2006 Wouter Verkerke, NIKHEF

Const member function – the flip side

• The compiler will enforce that no statement inside a

const member function modifies the object

class Fifo {
…
void print() const ;
…
int size ;
}

void Fifo::print() const {

cout << size << endl ; // OK
size = 0 ; // ERROR const function is not
allows to modify data member
}

© 2006 Wouter Verkerke, NIKHEF

Const member functions – indecent exposure

• Const member functions are also enforced not to ‘leak’

non-const references or pointers that allows users to
change its content

class Fifo {
…
char buf[80] ;
…
char* buffer() const {
return buf ; // ERROR – Const function exposing
non-const pointer to data member
}
}

© 2006 Wouter Verkerke, NIKHEF

 Const return values

• Lesson: Const member functions can only return const

references to data members
– Fix for example of preceding page

This const says that this

class Fifo { member function will not
… change the Fifo object
char buf[80] ;
…
const char* buffer() const {
return buf ; // OK
}
}

This const says the returned

pointer cannot be used to
modify what it points to

© 2006 Wouter Verkerke, NIKHEF

Why const is good
• Getting all your const declarations in your class correct
involves work! – Is it work the trouble?
• Yes! – Const is an important tool to promote encapsulation
– Classes that are ‘const-correct’ can be passed through const references to
functions and other objects and retain their full ‘read-only’ functionality
– Example

main() {
Fifo fifo ;
showTheFifo(fifo) ;
}

void showTheFifo(const Fifo& theFifo)

{
theFifo.print() ;
}

– Const correctness of class Fifo loosens coupling between main() and

showTheFifo() since main()’s author does not need to closely follow if
future version of showTheFifo() may have undesirable side effects on the
object
© 2006 Wouter Verkerke, NIKHEF
Mutable data members

• Occasionally it can be useful to be able to modify

selected data members in a const object
– Most frequent application: a cached value for a time-consuming
operation
– Your way out: declare that data member ‘mutable’. In that case it
can be modified even if the object itself is const

class FunctionCalculation {
…
mutable float cachedResult ;
…
float calculate() const {
// do calculation
cachedResult = <newValue> ; // OK because cachedResult
// is declared mutable
return cachedResult ;
}
}
– Use sparingly!
© 2006 Wouter Verkerke, NIKHEF
Static data members

• OO programming minimizes use of global variables

because they are problematic
– Global variable cannot be encapsulated by nature
– Changes in global variables can have hard to understand side
effects
– Maintenance of programs with many global variables hard

• C++ preferred alternative: static variables

– A static data member encapsulates a variable inside a class
• Optional ‘private’ declaration prevents non-class members to access variable

– A static data member is shared by all instances of a class

– Syntax
class ClassName {
… Declaration
static Type Name ;
…
}
Definition and initialization
© 2006
Type ClassName::Name = value ; Wouter Verkerke, NIKHEF
Static data members

• Don’t forget definition in addition to declaration!

– Declaration in class (in .hh) file. Definition in .cc file

• Example use case:

– class that keeps track of number of instances that exist of it

class Counter { int main() {

public: Counter c1 ;
Counter() { count++ ; } c1.Print() ;
~Counter() { count-- ; }
if (true) {
void print() { Counter c2,c3,c4 ;
cout << “there are “ c1.Print() ;
<< count }
<< “instances of count c1.Print() ;
<< endl ;
return 0 ;
}
}
private:
static int count ;
} ; there are 1 instances of count
there are 4 instances of count
int Counter::count = 0 ; there© are 1 instances
2006 Wouter of count
Verkerke, NIKHEF
Static function members

• Similar to static data member, static member functions

can be defined
– Syntax like regular function, with static keyword prefixed in
declaration only

class ClassName {
…
static Type Name(Type arg,…) ;
…
}

type ClassName::Name(Type arg,…) {

// body goes here
}

– Static function can access static data members only since

function is not associated with particular instance of class
– Can call function without class instance
ClassName::Name(arg,…) ;

© 2006 Wouter Verkerke, NIKHEF

Static member functions

• Example use case – modification of preceding example

class Counter { int main() {
public: Counter::print() ;
Counter() { count++ ; }
~Counter() { count-- ; } Counter c1 ;
static void print() { Counter::print() ;
cout << “there are “
<< count if (true) {
<< “instances of count Counter c2,c3,c4 ;
<< endl ; Counter::print() ;
} }
private: Counter::print() ;
static int count ; return 0 ;
} ; }

Counter::count = 0 ; there are 0 instances of count

there are 1 instances of count
there are 4 instances of count
there are 1 instances of count

© 2006 Wouter Verkerke, NIKHEF

Class Analysis and Design

4 Class Analysis
& Design

© 2006 Wouter Verkerke, NIKHEF

Overview of this section

• Contents of this chapter

– Object Oriented Analysis and Design – A first shot at

decomposing your problem into classes

– Designing the class interface – Style guide and common issues

– Operator overloading – Making your class behave more like

built-in types

– Friends – Breaking access patterns to enhance encapsulation

© 2006 Wouter Verkerke, NIKHEF

Class Analysis and Design
• We now understand the basics of writing classes
– Now it’s time to think about how to decompose your problem into
classes

• Writing good OO software involves 3 separate steps

1.Analysis
2.Design
3.Programming
– You can do them formally or informally, well or poorly, but you
can’t avoid them

• Analysis
– How to divide up your problem in classes
– What should be the functionality of each class

• Design
– What should the interface of your class look like?
© 2006 Wouter Verkerke, NIKHEF
Analysis – Find the class

• OO Analysis subject of many text books, many different

approaches
– Here some basic guidelines

1. Try to describe briefly in plain English (or Dutch) what you intend
your software to do
• Rationale – This naturally makes you think about your software in a high abstraction
level

2. Associate software objects with natural objects (‘objects in the

application domain’)
• Actions translate to member functions
• Attributes translate to data members

3. Make hierarchical ranking of objects using ‘has-a’ relationships

• Example: a ‘BankAccount’ has-a ‘Client’
• Has-a relationships translate into data members that are objects

4. Iterate! Nobody gets it right the first time

© 2006 Wouter Verkerke, NIKHEF
Analysis – A textbook example

• Example of telephone hardware represented as class

hierarchy using ‘has-a’ relationships
– Programs describing or simulating hardware usually have an
intuitive decomposition and hierarchy

Each line represents

Telephone
a ‘has-a’ relationship

Cable Housing Dialer Handset

Earpiece Mouthpiece Cable

© 2006 Wouter Verkerke, NIKHEF

Analysis – Example from High Energy Physics

• Real life often not so clean cut

• Example problem from High Energy physics

– We have a file with experimental data from a calorimeter.
– A calorimeter is a HEP detector that detects energy through
absorption. A calorimeter consists of a grid of detector modules
(cells) that each individually measure deposited energy

Calorimeter

Cells with energy deposit

Cell

Incoming particle © 2006 Wouter Verkerke, NIKHEF

 Analysis – Example from High Energy Physics

• First attempt to identify objects in data processing

model and their containment hierarchy
– Calorimeter global position and cell coordinates are not physical
objects but separate logical entities so we make separate classes
for those too

Calorimeter
Calorimeter
has-a has-a
Coordinate

Position CaloCell

has-a

CaloCell
Coordinate

Position

© 2006 Wouter Verkerke, NIKHEF

Analysis – Example from High Energy Physics

• Key Analysis sanity check – Can we describe what each

object is, in addition to what it does?
– Answer yes

Calorimeter
has-a has-a

Position CaloCell

has-a

Coordinate

© 2006 Wouter Verkerke, NIKHEF

 Analysis – Example from High Energy Physics

• Iterating the design – are there other/better solutions?

– Remember ‘strong cohesion’ and ‘loose coupling’
– Try different class decomposition, moving functionality from one
class to another

• Example of alternative solution

– We can store the CaloCells in an intelligent container class
CellGrid that mimics a 2D array and keeps track of coordinates

Calorimeter Calorimeter

CellGrid has-a has-a

Position CellGrid

has-a

CaloCell
CaloCell
Position © 2006 Wouter Verkerke, NIKHEF
Analysis – Example from High Energy Physics
• Which solution is better?
– Source of ambiguity: cell coordinate not really intrinsic property of
calorimeter cell
– Path to solution: what are cell coordinates used for? Import for insight in
best solution. Real-life answer: to find adjacent (surrounding cells)
– Solution: Adjacency algorithms really couple strongly to layout of cells, not
to property of individual cells  design with layout in separate class
probably better

Calorimeter Calorimeter
has-a has-a has-a has-a

Position CaloCell Position CellGrid

has-a has-a

Coordinate CaloCell

© 2006 Wouter Verkerke, NIKHEF

Extending the example – Has-A vs Uses-A
• Next step in analysis of calorimeter data is to reconstruct
properties of incoming particles
– Reconstruct blobs of energy deposited into multiple cells
– Output stored in new class CaloCluster, which stores properties of cluster
and refers back to cells that form the cluster

Calorimeter
has-a has-a

Position CellGrid CaloCluster

has-a has-a?, uses-a!

CaloCell

– Now we run into some problems with ‘has-a’ semantics: All CaloCells in
Calorimeter are owned by Calorimeter, so CaloCluster doesn’t really
‘have’ them. Solution: ‘Uses-A’ semantic.
– A ‘Uses-A’ relation translates into a pointer or ©
reference to an object
2006 Wouter Verkerke, NIKHEF
Summary on OO analysis
• Choosing classes: You should be able to say what a class is
– A ‘Has-A relation’ translates into data members, a ‘Uses-A’ relation into a
pointer
– Functionality of your natural objects translates in member functions

• Be wary of complexity
– Signs of complexity: repeated identical code, too many function arguments,
too many member functions, functions with functionality that cannot be
succinctly described
– A complex class is difficult to maintain  Redesign into smaller units

• There may not be a unique or ‘single best’ decomposition of

your class analysis
– Such is life. Iterate your design, adapt to new developments

• We’ll revisit OOAD again in a while when we will discuss

polymorphism and inheritance which open up many new
possibility (and pitfalls)
© 2006 Wouter Verkerke, NIKHEF
The art of proper class design
• Class Analysis tells you what functionality your class should have
• Class Design now focuses on how to package that best

• Focus: Make classes easy to use

– Robust design: copying objects, assigning them (even to themselves) should
not lead to corruption, memory leaks etc
– Aim for intuitive behavior: mimic interface of built-in types where possible
– Proper functionality for ‘const objects’

• Reward: better reusability of code, easier maintenance, shorter

documentation

• And remember: Write the interface first, then the implementation

– While writing the interface you might still find flaws or room for improvements
in the design. It is less effort to iterate if there is no implementation to data

© 2006 Wouter Verkerke, NIKHEF

The art of proper class design

• Focus on following issues next

– Boilerplate class design

– Accessors & Modifiers – Proper interface for const objects

– Operator overloading

– Assignment – Why you need it

– Overloading arithmetic, and subscript operators

– Overloading conversion operators, use of explicit

– Spilling your guts – friends

© 2006 Wouter Verkerke, NIKHEF

Check list for class interface

• A boilerplate class design

• When writing a class it helps to group member functions
into the following categories

– Initialization – Constructors and helper functions

– Assignment
– Cleanup – Destructors and helper functions

– Accessors – Function providing read-only access to data members

– Modifiers – Functions that allow to modify data members

– Algorithmic functions
– I/O functions
– Error processing functions

© 2006 Wouter Verkerke, NIKHEF

Accessor / modifier pattern
• For each data member that is made publicly available
implement an accessor and a modifier
• Pattern 1 – Encapsulate read & write access in separate functions
– Complete control over input and output. Modifier can be protected for better
access control and modifier can validate input before accepting it
– Note that returning large data types by value is inefficient. Consider to return a
const reference instead

class Demo { const here is important

private: otherwise this will fail
float _val ;
public: const Demo demo ;
// accessor demo.getVal() ;
float getVal() const ; {
return _val ;
}
// modifier
void setVal(float newVal) {
// Optional validity checking goes here
_val = newVal ;
} © 2006 Wouter Verkerke, NIKHEF
}
Accessor / modifier pattern

• Pattern 2 – Return reference to internal data member

– Must implement both const reference and regular reference!
– Note that no validation is possible on assignment. Best for built-in
types with no range restrictions or data members that are classes
themselves with built-in error checking and validation in their
modifier function

const version here is essential,

class Demo { otherwise code below will fail
private:
float _val ; const Demo demo ;
float demoVal = demo.val() ;
public:
float& val() { return _val ; }
const float& val() const { return _val ; }

© 2006 Wouter Verkerke, NIKHEF

Making classes behave like built-in objects

• Suppose we have written a ‘class complex’ that

represents complex numbers
– Execution of familiar math through add(),multiply() etc member
functions easily obfuscates user code

complex a(3,4), b(5,1) ;

b.multiply(complex(0,1)) ;
a.add(b) ;
a.multiply(b) ;
b.subtract(a) ;

– Want to redefine meaning of C++ operators +,* etc to perform

familiar function on newly defined classes, i.e. we want compiler
to automatically translate:

c = a * b c.assign(a.multiply(b)) ;

• Solution: C++ operator overloading

© 2006 Wouter Verkerke, NIKHEF
Operator overloading
• In C++ operations are functions too, i.e.
What you write What the compiler does

complex c = a + b c.operator=(operator+(a,b))

• Operators can be both regular functions as well as class

member functions
– In example above operator=() is implemented as member
function of class complex, operator+() is implemented as global
function
– You have free choice here, operator+() can also be implemented
as member function in which case the code would be come

c.operator=(a.operator+(b))

– Design consideration: member functions (including operators) can

access ‘private’ parts, so operators that need this are easier to
implement as member functions
• More on this in a while…
© 2006 Wouter Verkerke, NIKHEF
An assignment operator – declaration

• Lets first have a look at implementing the assignment

operator for our fictitious class complex
• Declared as member operator of class complex:
– Allows to modify left-hand side of assignment
– Gives access to private section of right-hand side of assignment

class complex {
public:
complex(double r, double i) : _r(r) _i(i) {} ;
complex& operator=(const complex& other) ;

private:
double _r, _i ;
} ;

© 2006 Wouter Verkerke, NIKHEF

An assignment operator – implementation

Copy content of other object Handle self-assignment explicitly

It is the same class, so you It happens, really!
have
access to its private members

complex& complex::operator=(const complex& other) {

// handle self-assignment
if (&other == this) return *this ;

// copy content of other

_r = other._r ;
_i = other._i ;

// return reference to self

return *this ;
}

Return reference to self

Takes care
© of chain
2006 assignments
Wouter Verkerke, NIKHEF
An assignment operator – implementation

Copy content of other object Handle self-assignment explicitly

It is the same class, so you It happens, really!
have
access to its private members

complex& complex::operator=(const complex& other) {

// handle self-assignment
if (&other == this) return *this ;

// copy content
Why ignoring self-assignment of bad
can be other
Image you store_r = other._r
information ;
in a dynamically allocated array
that needs to be_i = other._i
reallocated ;
on assignment…

A& A::operator=(const A& other) { Oops if (other==*this)

// return
delete _array ; reference to selfyou just deleted your own
return *this ;
_array = new int[other._len] ; array!
}
// Refill array here
return *this ;
} © 2006 Wouter Verkerke, NIKHEF
An assignment operator – implementation
Why you should return a reference to yourself
Returning a reference to yourself allows chain assignment

complex a,b,c ; complex a,b,c ;

a = b =c ; a.operator=(b.operator=(c)) ;

Returns reference to b

complex&
Not mandatory, complex::operator=(const
but essential complex&
if you want to mimic behavior of built-inother)
types {

// handle self-assignment
if (&other == this) return *this ;

// copy content of other

_r = other._r ;
_i = other._i ;

// return reference to self

return *this ;
}

Return reference to self

Takes care
© of chain
2006 assignments
Wouter Verkerke, NIKHEF
The default assignment operator
• The assignment operator is like the copy constructor:
it has a default implementation
– Default implementation calls assignment operator for each data member

• If you have data member that are pointers to ‘owned’ objects

this will create problems
– Just like in the copy constructor

• Rule: If your class owns dynamically allocated memory or

similar resources you should implement your own assignment
operator

• You can disallow objects being assigned by declaring their

assignment operator as ‘private’
– Use for classes that should not copied because they own non-assignable
resources or have a unique role (e.g. an object representing a file)

© 2006 Wouter Verkerke, NIKHEF

Example of assignment operator for owned data members
class A {
private:
float* _arr ;
int _len ;
public:
operator=(const A& other) ;
}
C++ default operator=() Custom operator=()
A& operator=(const A& other) { A& operator=(const A& other) {
if (&other==this) return *this; if (&other==this) return *this;
_arr = other._arr ; _len = other._len ;
_len = other._len ; delete[] _arr ;
return *this ; _arr = new int[_len] ;
} int i ;
for (i=0 i<len ; i++) {
YOU DIE.
If other is deleted before us, _arr will point _arr[i] = other._arr[i] ;
to garbage. Any subsequent use of self has }
undefined results return *this ;
}
If we are deleted before other, we will delete
_arr=other._arr, which is not owned by us:
other._arr will point to garbage and will
attempt to delete array again © 2006 Wouter Verkerke, NIKHEF
Overloading other operators
• Overloading of operator=() mandatory if object owns
other objects
• Overloading of other operators voluntary
– Can simplify use of your classes (example: class complex)
– But don’t go overboard – Implementation should be congruent
with meaning of operator symbol
• E.g. don’t redefine operator^() to implement exponentiation

– Comparison operators (<,>,==,!=) useful to be able to put class

in sortable container
– Addition/subtraction operator useful in many contexts: math
objects, container class (add new content/ remove content)
– Subscript operator[] potentially useful in container classes
– Streaming operators <<() and operator>>() useful for printing in
many objects

• Next: Case study of operator overloading with a custom

string class

© 2006 Wouter Verkerke, NIKHEF

The custom string class

• Example string class for illustration of operator overloading

class String {
private:
char* _s ;
int _len ; Data members, array & length

void insert(const char* str) { // private helper function

_len = strlen(str) ;
if (_s) delete[] _s ;
_s = new char[_len+1] ;
strcpy(_s,str) ;
}

public:
String(const char* str= “”) : _s(0) { insert(str) ; }
String(const String& a) : _s(0) { insert(a._s) ; }
~String() { delete[] _s ; }

int length() const { return _len ; }

const char* data() const { return _s ; }
String& operator=(const String& a) {
if (this != &a) insert(a._s) ;
return *this ;
}
© 2006 Wouter Verkerke, NIKHEF
} ;
The custom string class

• Example string class for illustration of operator overloading

class String {
private:
char* _s ;
int _len ;

void insert(const char* str) { // private helper function

_len = strlen(str) ;
if (_s) delete[] _s ; Delete old buffer,
_s = new char[_len+1] ; allocate new buffer,
strcpy(_s,str) ; copy argument into new buffer
}

public:
String(const char* str= “”) : _s(0) { insert(str) ; }
String(const String& a) : _s(0) { insert(a._s) ; }
~String() { delete[] _s ; }

int length() const { return _len ; }

const char* data() const { return _s ; }
String& operator=(const String& a) {
if (this != &a) insert(a._s) ;
return *this ;
}
© 2006 Wouter Verkerke, NIKHEF
} ;
The custom string class

• Example string class for illustration of operator overloading

class String {
private:
char* _s ;
int _len ;

void insert(const char* str) { // private helper function

_len = strlen(str) ;
if (_s) delete[] _s ;
_s = new char[_len+1] ;
strcpy(_s,str) ;
}

public:
String(const char* str= “”) : _s(0) { insert(str) ; } Ctor
String(const String& a) : _s(0) { insert(a._s) ; }
Dtor
~String() { delete[] _s ; }

int length() const { return _len ; }

const char* data() const { return _s ; }
String& operator=(const String& a) {
if (this != &a) insert(a._s) ;
return *this ;
}
© 2006 Wouter Verkerke, NIKHEF
} ;
The custom string class

• Example string class for illustration of operator overloading

class String {
private:
char* _s ;
int _len ;

void insert(const char* str) { // private helper function

_len = strlen(str) ;
if (_s) delete[] _s ;
_s = new char[_len+1] ;
strcpy(_s,str) ;
}

public:
String(const char* str= “”) : _s(0) { insert(str) ; }
String(const String& a) : _s(0) { insert(a._s) ; }
~String() { delete[] _s ; }

int length() const { return _len ; }

const char* data() const { return _s ; }
String& operator=(const String& a) { Overloaded
if (this != &a) insert(a._s) ;
assignment
return *this ;
} operator
© 2006 Wouter Verkerke, NIKHEF
} ;
Overloading operator+(), operator+=()

• Strings have a natural equivalent of addition

– “A” + “B” = “AB”
– Makes sense to implement operator+

• Coding guideline: if you implement +, also implement +=

– In C++ they are separate operators.
– Implementing + will not automatically make += work.
– Implementing both fulfills aim to mimic behavior of built-in types

• Practical tip: Do operator+=() first.

– It is easier
– Operator+ can trivially be implemented in terms of operator+=
(code reuse)

© 2006 Wouter Verkerke, NIKHEF

Overloading operator+(), operator+=()

• Example implementation for String

– Argument is const (it is not modified after all)
– Return is reference to self, which allows chain assignment

class String {
public:
String& operator+=(const String& other) {
int newlen = _len + other._len ; // calc new length
char* newstr = new char[newlen+1] ; // alloc new buffer

strcpy(newstr,_s) ; // copy own contents

strcpy(newstr+_len,other._s) ; // append new contents

delete[] _s ; // release orig memory

_s = newstr ; // install new buffer

return *this ;
}
} ;

© 2006 Wouter Verkerke, NIKHEF

Overloading operator+(), operator+=()

• Now implement operator+() using operator +=()

– Operator is a global function rather than a member function – no
privileged access is needed to String class content
– Both arguments are const as neither contents is changed
– Result string is passed by value

String operator+(const String& s1, const String& s2) {

String result(s1) ; // clone s1 using copy ctor
result += s2 ; // append s2
return result ; // return new result
}

© 2006 Wouter Verkerke, NIKHEF

Overloading operator+() with different types

• You can also add heterogeneous types with operator+()

– Example: String(“A”) + “b”

• Implementation of heterogeneous operator+ similar

– Illustration only, we’ll see later why we don’t need it in this particular
case

String operator+(const String& s1, const char* s2) {

String result(s1) ; // clone s1 using copy ctor
result += String(s2) ; // append String converted s2
return result ; // return new result
}

• NB: Arguments of operator+() do not commute

operator+(const& A, const& B)!=operator+(const& B, const& A)

© 2006 Wouter Verkerke, NIKHEF

– If you need both, implement both
Overloading comparison operators ==,!=,<,>

• Comparison operators make sense for strings

– “A” != “B”, “Foo” == “Foo”, “ABC” < “XYZ”
– Comparison operators are essential interface to OO sorting

• Example implementation
– Standard Library function strcmp return 0 if strings are identical,
less than 0 if s1<s2, and greater than 0 if s1>s2
– Input arguments are const again
– Output type is bool
– Operators <,>,<=,>= similar

bool operator==(const String& s1, const String& s2) {

return (strcmp(s1.data(),s2.data())==0) ;
}

bool operator!=(const String& s1, const String& s2) {

return (strcmp(s1.data(),s2.data())!=0) ;
} © 2006 Wouter Verkerke, NIKHEF
Overloading subscript operators

• Subscript operators make sense for indexed collections

such as strings
– String(“ABCD”)[2] = ‘C’

• Example implementation for String

– Non-const version allows string[n] to be use as lvalue
– Const version allows access for const objects

char& String::operator[](int i) {
// Don’t forget range check here
return _s[i] ;
}

const char& String::operator[](int i) const {

// Don’t forget range check here
return _s[i] ;
}

© 2006 Wouter Verkerke, NIKHEF

Overloading subscript operators
• Note 1: Any argument type is allowed in []
– Example class PhoneBook {
public:
int PhoneBook::operator[](const char* name) ;
} ;

void example() {
PhoneBook pbook ;
pbook[“Bjarne Stroustrup”] = 0264524 ;
int number = phoneBook[“Brian Kernigan”] ;
}
– Powerful tool for indexed container objects
– More on this later in the Standard Template Library section

• Note 2: C++ does not have multi-dimensional array operator

like array[5,3]
– Instead it has array[5][3] ;
– If you design a container with multi-dimensional indexing consider
overloading the () operator, which works exactly like the [] operator,
except that it allows multiple arguments
© 2006 Wouter Verkerke, NIKHEF
Overloading conversion operators

• Conversions (such as int to float) are operators too!

• Sometimes it makes sense to define custom
conversions for your class
– Example: String  const char*, const char*  String

• General syntax for conversions for classA to classB

ClassA {
operator ClassB() const ; // conversion creates copy
// so operation is const
}

• Example implementation for class String

String::operator const char*() const {

return _s ;
}

© 2006 Wouter Verkerke, NIKHEF

Using conversion operators

• Conversion operators allow the compiler to convert

types automatically for you.
– Example

int strlen(const char* str) ; // Standard Library function

String foo(“Hello World”) ;
int len = strlen(foo) ;

int strlen(const char* str) ; // Standard Library function

String foo(“Hello World”) ;
int len = strlen(foo.operator const char*()) ;

• Constructors aid the automatic conversion process for

reverse conversion from (from another type to yourself)
– Example: allows automatic conversion from ‘const char*’ to String
class String {
String(const char* str) ;
} © 2006 Wouter Verkerke, NIKHEF
How conversion operators save you work

• Remember that we defined operator+(const& String,

const char*)
– It turns out we don’t need it if String to ‘const char*’ conversion is
defined
– Compiler automatically fills in the necessary conversions for you

String s(“Hello”) ;
String s2 = s + “ World” ;

String s(“Hello”) ;
String s2 = s + String(“ World”) ;

– No need for our operator+(const String&, const char*).

– Of course we can define a dedicated operator that is
computationally more efficient we should still implement it.
The compiler will use the dedicated operator instead
© 2006 Wouter Verkerke, NIKHEF
Curbing an overly enthusiastic compiler

• Suppose you want define the constructor

class String {
String(const char*) ;
}

but you do not want to compiler to use it for automatic

conversions
• Solution: make the constructor explicit

class String {
explicit String(const char*) ;
}

– Useful in certain cases

© 2006 Wouter Verkerke, NIKHEF

Recap on operator definition
• Operators can be implemented as
– Global functions
– Member functions

• For binary operators a member function implementation

always binds to the left argument
– I.e. ‘a + b’  a.operator+(b)

• Rule of thumb:
– Operators that modify an object should be member functions of
that object
– Operators that don’t modify an object can be either a member
function or a global function

• But what about operators that modify the rightmost

argument?
– Example cin >> phoneBook  operator>>(cin,phoneBook)
© 2006 Wouter Verkerke, NIKHEF
What friends are for

• But what about operators that modify the rightmost

argument?
– Example cin >> phoneBook  operator>>(cin,phoneBook)

– Sometimes you can use public interface to modify object (e.g. see
string example)
– Sometimes this is not desirable (e.g. interface to reconstitute
object from stream is considered private) – what do you do?

• Solution: make friends

– A friend declaration allows a specified class or function to the
private parts of a function
– A global function declared as friend does NOT become a member
function it is only given the same access privileges
class String {
public:
String(const char*==“”) ;
private:
friend istream& operator>>(istream&, String&)
© 2006 Wouter Verkerke, ;
NIKHEF
}
Friend and encapsulation

• Worked out string example

class String {
public:
String(const char*==“”) ;
private:
char* _buf ;
int _len ;
friend istream& operator>>(istream&, String&) ;
} ;

istream& operator>>(istream& is, String& s) {

const int bufmax = 256 ;
static char buf[256] ;
is >> buf ;
delete[] s._buf ; // Directly
s._len = strlen(buf) ; // manipulate
s._buf = new char[s._len] ; // private members
strcpy(s._buf,buf) ; // of String s
return is ;
}
© 2006 Wouter Verkerke, NIKHEF
Friends and encapsulation

• Friends technically break encapsulation, but when

properly used they enhance encapsulation
– Example: class String and global
operator>>(istream&,String&) are really a single module
(strong cohesion)
– Friend allow parts of single logical module to communicate with
each other without exposing private interface to the outer world

• Friend declarations are allowed for functions, operators

and classes
– Following declaration makes all member functions of class
StringManipulator friend of class String

class String {
public:
String(const char*=“”) ;
private:
friend class StringManipulator ;
} © 2006 Wouter Verkerke, NIKHEF
Class string

• The C++ Standard Library provides a class string

very similar to the example class String that we have
used in this chapter
– Nearly complete set of operators defined, internal buffer memory
expanded as necessary on the fly
– Declaration in <string>
– Example
string dirname(“/usr/include”) ;
string filename ;

cout << “Give first name:”

// filename buffer will expand as necessary

cin >> filename ;

// Append char arrays and string intuitively

string pathname = dirname+ ”/” + filename ;

// But conversion string  char* must be done explicitly

ifstream infile(pathname.c_str()) ©; 2006 Wouter Verkerke, NIKHEF
The Standard Library – Using I/O streams

5 Standard Library –
Using I/O streams

© 2006 Wouter Verkerke, NIKHEF

Introduction
• The Standard Library organizes all kinds of I/O operations
through a standard class ‘iostream’
• We’ve already used class iostream several types through the
objects cin and cout

#include <iostream>
using namespace std ;

int main() {
double x;

// Read x from standard input

cin >> x ;

// Write x to standard output

cout << “x = “ << x << endl ;

return 0 ;
}

• We will now take a better look at how streams work

© 2006 Wouter Verkerke, NIKHEF
A look behind the scenes

• I/O in C++ involves three distinct steps

int, float, char[], Employee

Writing
Conversion from/to byte stream

Buffering of byte stream

Writing to/reading from I/O channel

Reading
Physical or Logical Device
© 2006 Wouter Verkerke, NIKHEF
 I/O classes and operators in C++
• Operators<<(), >>() do step 1, classes istream, ostream do step 2

int, float, char[], Employee

operator<<() Conversion from/to byte stream operator>>()

Buffering of byte stream

class ostream class istream

Writing to/reading from I/O channel

Physical or Logical Device

© 2006 Wouter Verkerke, NIKHEF
 Stream object in the Standard Library
• Stream classes in Standard Library
Logical or Direction of byte stream
Include file physical device
Input Output Both

<iostream> Generic
istream ostream iostream
(e.g.terminal)
<fstream> File ifstream ofstream fstream

<sstream> std::string istringstream ostringstream stringstream

• Standard Library stream classes also implement all operators

to convert built-in types to byte streams
– Implemented as member operators of stream class
– Example: ostream::operator<<(int) ;

• Standard Library also provides three global stream objects for

‘standard I/O’
– istream object cin for ‘standard input’
– © 2006 Woutererror’
ostream objects cout,cerr for ‘standard output’,’standard Verkerke, NIKHEF
Using streams without operators >>(),<<()

• Streams provide several basic functions to read and

write bytes
– Block operations

char buf[100] ;
int count(99) ;

// read ‘count’ bytes from input stream

cin.read(buf, count) ;

// write ‘count’ bytes to output stream

cout.write(buf,count) ;

© 2006 Wouter Verkerke, NIKHEF

Using streams without operators >>(),<<()

• Streams provide several basic functions to read and

write bytes
– Line oriented operations

// read line from stdin up to and including the newline char

cin.get(buf,100) ;

// read line from std up to newline char

cin.getline(buf,100) ;

// read line up to and including ‘:’

cin.get(buf,100,’:’) ;

// read single character

cin.get(c ) ;

// write buffer up to terminating null byte

cout.write(buf,strlen(buf)) ;

// write single character

cout.put(c ) ;
© 2006 Wouter Verkerke, NIKHEF
How is the stream doing?

• Member functions give insight into the state of the stream

Function Meaning
bool good() Next operation might succeed
bool eof() End of input seen
bool fail() Next operation will fail
bool bad() Stream is corrupted

• Example – reading lines from a file till the end of the file
ifstream ifs(“file.txt”) ;
char buf[100] ;

// Loop as long as stream is OK

while(!ifs.fail()) {
ifs.getline(buf,100) ;

// Stop here if we have reached end of file

if (ifs.eof()) break ;

cout << “just read ‘” << buf << “’” << Wouter
© 2006 endl Verkerke,
; NIKHEF
}
Some handy abbreviations

• Streams overload operator void*() to return !fail()

– Can shorten preceding example to

while(ifs) { // expanded to while(ifs.operator void*())

ifs.getline(buf,100) ;
if (ifs.eof()) break ;
cout << “just read ‘” << buf << “’” << endl ;
}

• Also return value of getline() provides similar

information
– Returns true if stream is good() and stream is not at eof() after
operation

while(ifs.getline(buf,100)) {
cout << “just read ‘” << buf << “’” << endl ;
}

© 2006 Wouter Verkerke, NIKHEF

Using stream operators

• The next step is to use the streaming operators instead

of the ‘raw’ IO routines
– Encapsulation, abstraction  let objects deal with their own
streaming

• Solution: use operator>>() instead of getline()

shoesize.txt
Bjarne 42
Leif 47
Thor 52

ifstream ifs(“shoesize.txt”) ;
string name ;
int size ;

while(ifs >> name >> size) {

cout << name << “ has shoe size “ << size << endl ;
}

© 2006 Wouter Verkerke, NIKHEF

Using stream operators

• Remember: syntax of stream operators is like that of

any other operator
string name ;
int size
cin >> name >> size
;

string name ;
int size
cin.operator>>( cin.operator>>(name), size) ;

• For all built-in types

– operator<<(ostream,TYPE) and operator>>(istream,TYPE) are
implemented as member functions of the streams
– Special case: operator<<(const char*) and operator>>(char*)
read and write char[] strings

© 2006 Wouter Verkerke, NIKHEF

Parsing input – some fine points
• Delimiters
– How does the text line

Bjarne Stroustrup 42

map on to the statement

cin >> firstName >> lastName >> shoeSize

– Because each operator()>> stops reading when it encounter ‘white

space’
– White space is ‘space’,’tab’,’vertical tab’,’form feed’ and ‘newline’
– White space between tokens is automatically ‘eaten’ by the stream

• Reading string tokens

– Be careful using char[] to read in strings: operator>>(const
char*) does not know your buffer size and it can overrun!
– Better to use class string

© 2006 Wouter Verkerke, NIKHEF

Formatting output of built-in types

• For built-in types streams have several functions that

control formatting
– Example: manipulating the base of integer output

cout.setf(ios_base::oct,ios_base::basefield) ; // set octal

cout << 1234 << endl ; // shows ‘2322’

cout.setf(ios_base::hex,ios_base::basefield) ; // set hex

cout << 1234 << endl ; // shows ‘4d2’

– But it is often inconvenient to use this as calling formatting

function interrupt chained output commands

• To accomplish formatting more conveniently streams

have ‘manipulators’
– Manipulators are ‘pseudo-objects’ that change the state of the
stream on the fly:
cout << oct << 1234 << endl << hex << 1234 << endl ;
// shows ‘2322’ ‘4d2’
© 2006 Wouter Verkerke, NIKHEF
Overview of manipulators

• So manipulators are the easiest way to modify the

formatting of built-in types
• What manipulators exist?
– Integer formatting
Manipulator Stream type Description
dec iostream decimal base for integer
hex iostream hexadecimal base for integer
oct iostream octal base for integer
[no]showpos iostream show ‘+’ for positive integers
setbase(int n) iostream base n for integer

– Floating point formatting

Manipulator Stream type Description
setprecision(int n) iostream show n places after decimal point
[no]showpoint iostream [don’t ]show trailing decimal point
scientific iostream scientific format x.xxexx
uppercase iostream print 0XFF, nnExx
© 2006 Wouter Verkerke, NIKHEF
fixed iostream format xxxx.xx
Manipulators – continued
– Alignment & general formatting

Manipulator Stream type Description

left iostream align left
right iostream align right
internal iostream use internal alignment for each type
setw(int n) iostream next field width is n positions
setfill(char c) iostream set field fill character to c

– Miscellaneous

Manipulator Stream type Description

endl ostream put ‘\n’ and flush
ends ostream put ‘\0’ and flush
flush ostream flush stream buffers
ws istream eat white space
setfill(char c) iostream set field fill character to c

– Include <iomanip> for most manipulator definitions

© 2006 Wouter Verkerke, NIKHEF
Formatting output with manipulators
• Very clever, but how do manipulators work?
– A manipulator is a ‘pseudo-object’ that modifies the state of the
stream
– More precisely: a manipulator is a static member function of the
stream that takes a stream as argument, for example
class ostream {
static ostream& oct(ostream& os) {
os.setf(ios::oct,ios::basefield) ;
}
}
– The manipulator applies its namesake modification to the stream
argument
– You put manipulators in your print statement because class ostream
also defines
operator<<(ostream&(*f)(ostream&)) {
return f(*this) ;
}
This operator processes any function that takes a single ostream& as
argument and returns an ostream. The operator calls the function
with itself as argument, which then causes the wanted operation to be
© 2006 Wouter Verkerke, NIKHEF
executed on itself
Random access streams

• Streams tied to files and to strings also allow random

access
– Can move ‘current’ position for reading and writing to arbitrary
location in file or string

member function stream Description

streampos tellg() input return current location of ‘get()’ position
seekg(streampos) input set location of ‘get()’ position
streampos tellp() output return current location of ‘put()’ position
seekp(streampos) output set location of ‘put()’ position

– Streams open for both input and output (fstream,stringstream)

have all four methods, where put() and get() pointer can be in
different positions

© 2006 Wouter Verkerke, NIKHEF

Random access streams
File Layout
• Example use of tell(),seek() get() put()
position position
#include <fstream>

FileHeader
// Open file for reading and writing
fstream iofile(“file.dat”,ios::in|ios::out) ;

FileDataObj
© 2006 Wouter Verkerke, NIKHEF
Random access streams
File Layout
• Example use of tell(),seek() get() put()
position position
#include <fstream>

FileHeader
// Open file for reading and writing
fstream iofile(“file.dat”,ios::in|ios::out) ;

// Read in (fictitious) file header

FileHeader hdr ;
iofile >> hdr ;

FileDataObj
© 2006 Wouter Verkerke, NIKHEF
Random access streams
File Layout
• Example use of tell(),seek() get() put()
position position
#include <fstream>

FileHeader
// Open file for reading and writing
fstream iofile(“file.dat”,ios::in|ios::out) ;

// Read in (fictitious) file header

FileHeader hdr ;
iofile >> hdr ;

// Store current location of stream ‘get()’ pointer

streampos marker = ifs.tellg() ;

FileDataObj
© 2006 Wouter Verkerke, NIKHEF
Random access streams
File Layout
• Example use of tell(),seek() get() put()
position position
#include <fstream>

FileHeader
// Open file for reading and writing
fstream iofile(“file.dat”,ios::in|ios::out) ;

// Read in (fictitious) file header

FileHeader hdr ;
iofile >> hdr ;

// Store current location of stream ‘get()’ pointer

streampos marker = ifs.tellg() ;

FileDataObj
// Read (fictitious) file data object
FileDataObj fdo ;
iofile >> fdo ;

© 2006 Wouter Verkerke, NIKHEF

Random access streams
File Layout
• Example use of tell(),seek() get() put()
position position
#include <fstream>

FileHeader
// Open file for reading and writing
fstream iofile(“file.dat”,ios::in|ios::out) ;

// Read in (fictitious) file header

FileHeader hdr ;
iofile >> hdr ;

// Store current location of stream ‘get()’ pointer

streampos marker = ifs.tellg() ;

FileDataObj
// Read (fictitious) file data object
FileDataObj fdo ;
iofile >> fdo ;

// modify file data object

// Move current’ location of stream

// ‘put ‘()’ pointer to marked position
ifs.tellp(marker) ;

© 2006 Wouter Verkerke, NIKHEF

Random access streams
File Layout
• Example use of tell(),seek() get() put()
position position
#include <fstream>

FileHeader
// Open file for reading and writing
fstream iofile(“file.dat”,ios::in|ios::out) ;

// Read in (fictitious) file header

FileHeader hdr ;
iofile >> hdr ;

// Store current location of stream ‘get()’ pointer

streampos marker = ifs.tellg() ;

FileDataObj
// Read (fictitious) file data object
FileDataObj fdo ;
iofile >> fdo ;

// modify file data object

// Move current’ location of stream

// ‘put ‘()’ pointer to marked position
ifs.tellp(marker) ;

// Write modified object over old location in file

iofile << fdo ; © 2006 Wouter Verkerke, NIKHEF
Streaming custom classes
• You can stream custom classes by defining your
matching operator<<(),operator>>() for those classes
– Standard Library stream classes implement operators <<, >> as
member functions for streaming of all basic types basic types
– This is not an option for you as you can’t modify the Standard
Library classes

– But in general, binary operators can be

1. member of class ostream(cout),
2. member of your class, or
3. be a global function.

– Option 1) already ruled out

– Option 2) doesn’t work because class being read/written needs to
be rightmost argument of operator, while as a member function it
is by construction the left argument of the operator
– Option 3) works: implement operator<< as global operator

© 2006 Wouter Verkerke, NIKHEF

Streaming custom classes

• For types that can be printed on a single line,

overloading the operator<<, operator>> is sensible
– Class string obvious example

String s(“Hello”) ;
cout << string << “ World” ;

String s(“Hello”) ;
cout.operator<<(operator<<(cout,string),”World”) ;

• For classes that read/write multi-line output, consider a

separate function
– operator>>,<< syntax for such cases potentially confusing:
processing white space etc traditionally handled by stream not by
operator
– Example names: readFromStream(),writeToStream()
© 2006 Wouter Verkerke, NIKHEF
Implementing your own <<,>> operators

• Important: operators <<,>> need to return a reference

to the input ostream, istream respectively
– Essential for ability to chain << operations
cin >> a >> b ;
cout << a << b << c ;

• Example implementation for class string

ostream& operator<<(ostream& os, const String& s) {
os << s._s ;
return os ;
}

istream& operator>>(istream& is, String& s) {

const int bufmax = 256 ;
static char buf[256] ;
is >> buf ; Note: no const here
s = buf ; as String is modified
return is ;
}
© 2006 Wouter Verkerke, NIKHEF
Generic programming – Templates

6 Generic
Programming
– Templates

© 2006 Wouter Verkerke, NIKHEF

Introduction to generic programming

• So far concentrated on definitions of objects as means

of abstraction
• Next: Abstracting algorithms to be independent of the
type of data they work with
• Naive – max()
– Integer implementation

// Maximum of two values

int max(int a, int b) {
return (a>b) ? a : b ;
}

– (Naïve) real-life use

int m = 43, n = 56 ;
cout << max(m,n) << endl ; // displays 56 (CORRECT)

double x(4.3), y(5.6) ;

cout << max(x,y) << endl ; //©displays
2006 Wouter 5 (INCORRECT)
Verkerke, NIKHEF
Generic algorithms – the max() example

• First order solution – function overloading

– Integer and float implementations

// Maximum of two values

int max(int a, int b) {
return (a>b) ? a : b ;
}

// Maximum of two values

float max(float a, float b) {
return (a>b) ? a : b ;
}

– (Naïve) real-life use

int m = 43, n = 56 ;
cout << max(m,n) << endl ; // displays 56 (CORRECT)

double x(4.3), y(5.6) ;

cout << max(x,y) << endl ; // displays 5.6 (CORRECT)
© 2006 Wouter Verkerke, NIKHEF
Generic algorithms – the template solution

• Overloading solution works but not elegant

– Duplicated code (always a sign of trouble)
– We need to anticipate all use cases in advance

• C++ solution – a template function

template<class TYPE>
TYPE max(const TYPE& a, const TYPE& b) {
return (a>b) ? a : b ;
}

int m = 43, n = 56 ;
cout << max(m,n) << endl ; // displays 56 (CORRECT)

double x(4.3), y(5.6) ;

cout << max(x,y) << endl ; // displays 5.6 (CORRECT)

© 2006 Wouter Verkerke, NIKHEF

Basics of templates

• A template function is function or algorithm for a

generic TYPE
– Whenever the compiler encounter use of a template function with
a given TYPE that hasn’t been used before the compiler will
instantiate the function for that type

template<class TYPE>
TYPE max(const TYPE& a, const TYPE& b) {
return (a>b) ? a : b ;
}

int m = 43, n = 56 ;
// compiler automatically instantiates max(int&, int&)
cout << max(m,n) << endl ; // displays 56 (CORRECT)

double x(4.3), y(5.6) ;

// compiler automaticallu instantiates max(float&, float&)
cout << max(x,y) << endl ; // displays 5.6 (CORRECT)

© 2006 Wouter Verkerke, NIKHEF

Basics to templates – assumptions on TYPE

• A template function encodes a generic algorithm but not

a universal algorithm
– TYPE still has to meet certain criteria to result in proper code
– For example:

template<class TYPE>
TYPE max(const TYPE& a, const TYPE& b) {
return (a>b) ? a : b ;
}

assumes that TYPE.operator>(TYPE&) is defined

• Style tip: When you write a template spell out in the

documentation what assumption you make (if any)

© 2006 Wouter Verkerke, NIKHEF

Basic templates – another example

• Here is another template function example

template <class TYPE>

void swap(TYPE& a, TYPE& b) {
TYPE tmp = a ; // declare generic temporary
a = b ;
b = tmp ;
}

– Allocation of generic storage space

– Only assumption of this swap function: TYPE::operator=()
defined
– Since operator=() has a default implementation for all types this
swap function truly universal
• Unless of course a class declares operator=() to be private in which case no
copies can be made at all

© 2006 Wouter Verkerke, NIKHEF

Template formalities

• Formal syntax of template function

– Template declaration

template <class TYPE>

TYPE function_name(TYPE& t) ;

– Template definition
template <class TYPE>
TYPE function_name(TYPE& t){
// body
}

• What’s OK
– Multiple template classes allowed
template <class TYPE1, class TYPE2,…class TYPEN>
TYPE1 function_name(TYPE1&, TYPE2&,… TYPEN&) ;

– Non-template function arguments allowed

template <class TYPE>
TYPE function_name(TYPE& t, int x, ©float y) ;Verkerke, NIKHEF
2006 Wouter
Template formalities

• And what’s not

– Template definitions must be in the global scope

int myFunction() {
template <class T> // ERROR – not allowed
void myTemplFunc(T& t) ;
}

– Template TYPE must appear in signature

template <class TYPE>

TYPE function_name(int i) // ERROR cannot overload
// on return type

– Reason: function overloading by return type is not allowed

© 2006 Wouter Verkerke, NIKHEF

Templates, files and export
• Like for regular functions, template functions can be declared
and defined in separates files
– Template declaration file server to ‘consumer code’
– Template definition only offered once to compiler

• But one extra detail – template definitions in a separate file

require the ‘export’ keyword

// swap.hh -- Declaration // swap.cc -- Definition

template <class TYPE> export template <class TYPE>
void swap(TYPE& a, TYPE& b) ; void swap(TYPE& a, TYPE& b) {
TYPE tmp = a ;
a = b ;
b = tmp ;
}

– Reason: when templates definitions are in separate file logistics of

instantiating all templates exactly once in a program consisting of many
modules requires extra logistic support from compiler and linker. Export
keyword tells compiler to make necessary preparations for these logistics
© 2006 Wouter Verkerke, NIKHEF
NB: Keyword export is not supported by all compilers yet
Template specialization
• Sometimes you have a template function that is almost
generic because
– It doesn’t work (right) with certain types.
For example max(const char* a, const char* b)

template<class TYPE>
TYPE max(const TYPE& a, const TYPE& b) {
return (a>b) ? a : b ; // comparing pointer not sensible
}

– Or for certain types there is a more efficient implementation of the

algorithm

• Solution: provide a template specialization

– Can only be done in definition, not in declaration
– Tells compiler that specialized version of function for given template should
be used when appropriate

template<>
const char* max(const char* a, const char* b) {
return strcmp(a,b)>0 ? a : b ; // Use string comparison instead
© 2006 Wouter Verkerke, NIKHEF
}
Template classes

• Concept of templates also extends to classes

– Can define a template class just like a template function

template<class T>
class Triplet {
public:
Triplet(T& t1, T& t2, T& t3) () ;
private:
T _array[3] ;
}

• Class template mechanism allows to create generic

classes
– A generic class provide the same set of behaviors for all types
– Eliminates code duplication
– Simplifies library design
– Use case per excellence: container classes (arrays, stacks etc…)

© 2006 Wouter Verkerke, NIKHEF

Generic container class example

• A generic stack example

template<class TYPE>
class Stack {
public:
Stack(int size) : _len(size), _top(0) { // constructor
_v = new TYPE[_len] ;
}
Stack(const Stack<TYPE>& other) ; // copy constructor
~Stack() { delete[] _v ; }

void push(const TYPE& d) { _v[_top++] = d ; }

TYPE pop() { return _v[--_top] ; }

Stack<TYPE>& operator=(const Stack<TYPE>& s) ; // assignment

private:
TYPE* _v ; Assumptions on TYPE
int _len ; -Default constructor
int _top ; -Assignment defined

} © 2006 Wouter Verkerke, NIKHEF

Using the generic container class

• Example using Stack

void example() {

Stack<int> s(10) ; // stack of 10 integers

Stack<String> t(20) ; // stack of 20 Strings

s.push(1) ;
s.push(2) ;
cout << s.pop() << endl ;

// OUTPUT ‘2’

t.push(“Hello”) ; // Exploit automatic

t.push(“World”) ; // const char*  String conversion

cout << t.pop() << “ “ << t.pop() << endl ;

// OUTPUT ‘World Hello’

}
© 2006 Wouter Verkerke, NIKHEF
Non-class template parameters

• You can also add non-class parameters to a template

declaration, e.g.
– parameter must be constant expression (template is instantiated
at compile time!)

template<class T, int DIM>

class Vector {
public:
Vector() ;
Vector(T[] input) ;
private:
T _array[DIM] ;
}

– Advantage: avoid dynamic memory allocation (runtime

performance penalty)

void example() {
Vector<double,3> ThreeVec ;
Vector<double,4> FourVec ;
}
© 2006 Wouter Verkerke, NIKHEF
Template default parameters

• Default values for template parameters are also allowed

template<class T=double, int DIM=3>

class Vector {
public:
Vector(T[] input) ;
private:
T _array[DIM] ;
}

void example() {
Vector ThreeVecDouble ;
Vector<int> ThreeVecInt ;
Vector<float,4> FourVecFloat ;
}

© 2006 Wouter Verkerke, NIKHEF

Containment, composition

• No real limit on complexity of template constructions

– Containment

template<class TYPE>
class A {
private:
B<TYPE> member1 ; // OK – generic template member
C<int> member2 ; // OK – specific template member
D member3 ; // OK – non-template member
public:
A(args) : B(args),C(args),D(args) {} // initialization
} ;

– Composition
template<class TYPE> class Vec { … } ; // Vector container
template<class TYPE> class Arr { … } ; // Array container
template<class TYPE> class Sta { … } ; // Stack container

Vec<String> c1 ; // Vector of Strings

Vec<Arr<String> > c2 ; // Vector of Arrays of Strings
Vec<Arr<Sta<String> > > c3 ; // Vector of Arrays of Stacks of Strings
Vec<Vec<Vec<String> > > c4 ; // Vector of Vectors of Vectors of Strings
© 2006 Wouter Verkerke, NIKHEF
Note extra space to distguish from operator>>
The Standard Template Library

7 Standard Library II
the Template Library

© 2006 Wouter Verkerke, NIKHEF

Introduction to STL

• STL = The Standard Template Library

– A collection of template classes and functions for general use
– Started out as experimental project by Hewlett-Packard
– Now integral part of ANSI C++ definition of ‘Standard Library’
– Excellent design!

• Core functionality – Collection & Organization

– Containers (such as lists)
– Iterators (abstract methods to iterate of containers)
– Algorithms (such as sorting container elements)

• Some other general-purpose classes

– classes string, complex, bits

© 2006 Wouter Verkerke, NIKHEF

Overview of STL components

• Containers
– Storage facility of objects

Container

Object Object Object Object Object Object

• Iterators
– Abstract access mechanism to collection contents
– “Pointer to container element” with functionality to move pointer

• Algorithms
– Operations (modifications) of container organization of contents
– Example: Sort contents, apply operation to each of elements

© 2006 Wouter Verkerke, NIKHEF

STL Advantages
• STL containers are generic
– Templates let you use the same container class with any class or built-in
type

• STL is efficient
– The various containers provide different data structures.
– No inheritance nor virtual functions are used (we’ll cover this shortly).
– You can choose the container that is most efficient for the type of
operations you expect

• STL has a consistent interface

– Many containers have the same interface, making the learning curve easier

• Algorithms are generic

– Template functions allow the same algorithm to be applied to different
containers.

• Iterators let you access elements consistently

– Algorithms work with iterators
– Iterators work like C++ pointers

• Many aspects can be customized easily © 2006 Wouter Verkerke, NIKHEF

Overview of STL containers classes

• Sequential containers (with a defined order)

– vector
Fundamental container
– list implementations
with different performance tradeoffs
– deque (double-ended queue)

– stack
Adapters of fundamental
– queue containers
that provide a modified functionality
– priority_queue

• Associative containers (no defined order, access by key)

– set
– multiset
– map
– multimap

© 2006 Wouter Verkerke, NIKHEF

Common container facilities

• Common operations on fundamental containers

– insert – Insert element at defined location
– erase – Remove element at defined location

– push_back – Append element at end

– pop_back – Remove & return element at end

– push_front – Append element at front

– pop_front – Remove & return element at front

– at – Return element at defined location (with range checking)

– operator[] – Return element at defined location (no range checking)

– Not all operations exist at all containers (e.g. push_back is undefined on

a set as there is no ‘begin’ or ‘end’ in an associative container)

© 2006 Wouter Verkerke, NIKHEF

Vector <vector>

• Vector is similar to an array

0 1 2

– Manages its own memory allocation

– Initial length at construction, but can be extended later
– Elements initialized with default constructor
– Offers fast random access to elements
– Example
#include <vector>
vector<int> v(10) ;

v[0] = 80 ;
v.push_back(70) ; // creates v[10] and sets to 11

vector<double> v2(5,3.14) ; // initialize 5 elements to 3.14

© 2006 Wouter Verkerke, NIKHEF

 Template<class T>
List <list> Struct ListElem {
T elem ;
ListElem* prev ;
• Implemented as doubly linked list ListElem* next ;
}

front end

– Fast insert/remove of in the middle of the collection

front end

– No random access
– Example iterator ‘pointer’ in collection

#include <list>
list<double> l ;
l.push_front(30.5) ; // append element in front
l.insert(somewhere,47.5) ; // insert in middle

© 2006 Wouter Verkerke, NIKHEF

Double ended queue <deque>

• Deque is sequence optimized for insertion at both ends

– Inserting at ends is as efficient as list
– Random access of elements efficient like vector

rear front

• Example of deque’s in real life

– String of beads
– Deck of cards
– Train

“Deque, it rhymes with ‘check’ (B. Stroustrup)”

© 2006 Wouter Verkerke, NIKHEF

Stack <stack>

• A stack is an adapter of deque top

– It provides a restricted view of a deque
push() pop()
– Can only insert/remove elements elements
at end (‘top’ in stack view’)
– No random access

• Example
bottom
void sender() {
stack<string> s ;
s.push(“Aap”) ;
s.push(“Noot”) ;
s.push(“Mies”) ;
receiver(s) ;
}
void receiver(stack<int>& s) {
while(!s.empty()) cout << s.pop() << “ “ ;
}

// output “Mies Noot Aap”

© 2006 Wouter Verkerke, NIKHEF
Queue <queue>

• A queue is another adapter of deque

– It provides a different restricted view of a deque
– Can only insert elements at back, remove elements from front
– No random access

push() rear front pop()

• Plenty of real-life applications

– Ever need anything at city hall? Take a number!
– Example implementation
void sender() {
queue<string> s ;
s.push(“Aap”) ; s.push(“Noot”) ;s.push(“Mies”) ;
receiver(s) ;
}
void receiver(stack<int>& s) {
while(!s.empty()) cout << s.pop() << “ “ ;
}
// output “Aap Noot Mies” (reverse ©compared
2006 Wouter Verkerke, NIKHEF
to stack)
Priority_queue <queue>

• Like queue with priority control

– In priority queue pop() returns element with highest rank,
rank determined with operator<()
– Also a container adapter (by default of vector)

33 17
push() pop()
returns and removes
35 40 42 element with
highest priority

void sender() {
priority_queue<int> s ;
s.push(10) ; s.push(30) ;s.push(20) ;
receiver(s) ;
}
void receiver(stack<int>& s) {
while(!s.empty()) cout << s.pop() << “ “ ;
}
// output “30 20 10” © 2006 Wouter Verkerke, NIKHEF
Sequential container performance comparison

• Performance/capability comparison of sequential

containers

Insert/remove element
index [] In Middle At Front At Back
vector const O(n) + const +
list const const const
deque const O(n) const const
stack const +
queue const const +
prio_que O(logN) O(logN)

– const : constant CPU cost

– O(n) : CPU cost proportional to Nelem
– O(logN) : CPU cost proportional to log(Nelem)

© 2006 Wouter Verkerke, NIKHEF

Sequential versus associative containers
• So far looked at several forms of sequential containers
– Defining property: storage organization revolves around ordering: all
elements are stored in a user defined order
– Access to elements is always done by relative or absolute position in container
– Example:
vector<int> v ;
v[3] = 3rd element of vector v

list<int> l ;
double tmp = *(l.begin()) ; // 1st element of list

• For many types of problems access by key is much more natural

– Example: Phone book. You want to know the phone number (=value) for a
name (e.g. ‘B. Stroustrup’ = key)
– You don’t care in which order collection is stored as you never retrieve the
information by positional reference (i.e. you never ask: give me the 103102nd
entry in the phone book)
– Rather you want to access information with a ‘key’ associated with each value

• Solution: the associative container

© 2006 Wouter Verkerke, NIKHEF
Sequential versus associative containers
Sequential

front end

Give me 3rd element

Associative

key: value: key: value:

Ivar 49 Brian 52
key: value:
Bjarne 43
key: value:
key: value: Leif 47
Thor 39

Give me value of element

© 2006 Wouter Verkerke, NIKHEF
with key “Leif”
Pair <utility>

• Utility for associative containers – stores a key-value pair

template<type T1, type T2>
struct pair {
T1 first ;
T2 second ;
pair(const T1&, const T2&) ;
} ;

template<type T1, type T2>

pair<T1,T2> make_pair(T1,T2) ; // exists for convenience

– Main use of pair is as input or return value

pair<int,float> calculation() {
return make_pair(42,3.14159) ;
}
int main() {
pair<int,float> result = calculation() ;
cout << “result = “ << pair.first
<< “ “ << pair.second << endl ;
}
© 2006 Wouter Verkerke, NIKHEF
Map <map>

• Map is an associative container

– It stores pairs of const keys and values
– Elements stored in ranking by keys (using key::operator<())
– Provides direct access by key
– Multiple entries with same key prohibited

map<T1,T2>

pair<const T1,T2>
Bjarne 33

Gunnar 42

Leif 47

Thor 52

© 2006 Wouter Verkerke, NIKHEF

Map <map>

• Map example

map<string,int> shoeSize ;

shoeSize.insert(pair<string,int>(“Leif”,47)) ;
showSize.insert(make_pair(“Leif”,47)) ;

shoeSize[“Bjarne”] = 43 ;
shoeSize[“Thor”] = 52 ;

int theSize = shoeSize[“Bjarne”] ; // theSize = 43

int another = shoeSize[“Stroustrup”] ; // theSize = 0

– If element is not found, new entry is added using default

constructors

© 2006 Wouter Verkerke, NIKHEF

Set <set>

• A set is a map without values set<T1>

– I.e. it is just a collection of keys
const T1
– No random access through []
– Storage in ranked by key (<) Bjarne

– No duplicates allowed
Gunnar

set<string> people ; Leif

people.insert(“Leif”) ;
people.insert(“Bjarne”) ; Thor
people.insert(“Stroustrup”) ;

© 2006 Wouter Verkerke, NIKHEF

Multiset <set>, Multimap <map>

• Multiset
– Identical to set
– Except that multiple keys of the same value are allowed

• Multimap
– Identical to map
– Except that multiple keys of the same value are allowed

© 2006 Wouter Verkerke, NIKHEF

Taking a more abstract view of containers
• So far have dealt directly with container object to insert
and retrieve elements
– Drawback: Client code must know exactly what kind of container
it is accessing
– Better solution: provide an abstract interface to the container.
– Advantage: the containers will provide the same interface (as far
as possible within the constraints of its functionality)
– Enhanced encapsulation – You can change the type of container
class you use later without invasive changes to your client code

• STL abstraction mechanism for container access:

the iterator
– An iterator is a pointer to an element in a container
– So how is an iterator difference from a regular C++ pointer? – An
iterator is aware of the collection it is bound to.
– How to you get an iterator: A member function of the collection
will give it to you

© 2006 Wouter Verkerke, NIKHEF

Taking a more abstract view of containers

• Illustration of iterators vs C++ pointers

double v[10] ; vector<double> v[10] ;

int i = 0 ;
double* ptr ; vector<double>::iterator iter ;

ptr = &array[0] ; iter = v.begin() ;

while(i<10) { while(iter!=v.end()) {

cout << *ptr << endl ; cout << *iter << endl ;

++ptr ; ++iter ;
++i ; }
}

Allocate C++ array of 10 elements Allocate STL vector of 10 elements

© 2006 Wouter Verkerke, NIKHEF

Taking a more abstract view of containers

• Illustration of iterators vs C++ pointers

double v[10] ; vector<double> v[10] ;

int i = 0 ;
double* ptr ; vector<double>::iterator iter ;

ptr = &array[0] ; iter = v.begin() ;

while(i<10) { while(iter!=v.end()) {

cout << *ptr << endl ; cout << *iter << endl ;

++ptr ; ++iter ;
++i ; }
}

Allocate a pointer. Allocate an STL iterator to a vector

Also allocate an integer to keep
track of when you’re at the end
of the array © 2006 Wouter Verkerke, NIKHEF
Taking a more abstract view of containers

• Illustration of iterators vs C++ pointers

double v[10] ; vector<double> v[10] ;

int i = 0 ;
double* ptr ; vector<double>::iterator iter ;

ptr = &array[0] ; iter = v.begin() ;

while(i<10) { while(iter!=v.end()) {

cout << *ptr << endl ; cout << *iter << endl ;

++ptr ; ++iter ;
++i ; }
}

Make the pointer point to Make the iterator point to

the first element of the the first element of the vector
array
© 2006 Wouter Verkerke, NIKHEF
Taking a more abstract view of containers

• Illustration of iterators vs C++ pointers

double v[10] ; vector<double> v[10] ;

int i = 0 ;
double* ptr ; vector<double>::iterator iter ;

ptr = &array[0] ; iter = v.begin() ;

while(i<10) { while(iter!=v.end()) {

cout << *ptr << endl ; cout << *iter << endl ;

++ptr ; ++iter ;
++i ; }
}

Check if you’re at the end Check if you’re at the end of

of your array your vector

© 2006 Wouter Verkerke, NIKHEF

Taking a more abstract view of containers

• Illustration of iterators vs C++ pointers

double v[10] ; vector<double> v[10] ;

int i = 0 ;
double* ptr ; vector<double>::iterator iter ;

ptr = &array[0] ; iter = v.begin() ;

while(i<10) { while(iter!=v.end()) {

cout << *ptr << endl ; cout << *iter << endl ;

++ptr ; ++iter ;
++i ; }
}

Access the element the pointer Access the element the iterator
is currently pointing to is currently pointing to

© 2006 Wouter Verkerke, NIKHEF

Taking a more abstract view of containers

• Illustration of iterators vs C++ pointers

double v[10] ; vector<double> v[10] ;

int i = 0 ;
double* ptr ; vector<double>::iterator iter ;

ptr = &array[0] ; iter = v.begin() ;

while(i<10) { while(iter!=v.end()) {

cout << *ptr << endl ; cout << *iter << endl ;

++ptr ; ++iter ;
++i ;
} }

Modifiy the pointer to point Modify the iterator to point to

to the next element in the the next element in the array
array
© 2006 Wouter Verkerke, NIKHEF
Containers Iterators – A closer look at the formalism

begin() returns iterator

Declaration pointing to start of container
container<T>::iterator

// Iterator loop
vector<int> v(10) ;
vector<int>::iterator iter ;
for (iter=v.begin() ;
operator++
iter!=v.end() ;
moves iterator one
++iter) {
element ahead
*iter = 0 ;
}
dereference
operator*() returns
T& to pointed-to
element end() returns iterator pointing
to start of container
assignment between
iterators transfer
‘current position’
© 2006 Wouter Verkerke, NIKHEF
Why iterators are a better interface

• Iterators hide the structure of the container

– Iterating over a vector, list or map works the same. You client
code needs to make no (unnecessary assumptions) on the type of
collection you’re dealing with

// Iterator loop over a vector

vector<int> v(10) ;
int sum = calcSum(v.begin()) ;

int calcSum(vector<int>::iterator iter) {

int sum(0) ;
while(iter) { // Iterator loop over a list
sum += *iter ; list<int> l ;
++iter ; int sum = calcSum(l.begin()) ;
}
return sum ; int calcSum(list<int>::iterator iter) {
} int sum(0) ;
while(iter) {
sum += *iter ;
++iter ;
}
return sum ;
© 2006 Wouter Verkerke, NIKHEF
}
Iterators and ranges
• A pair of iterators is also an efficient way to define subsets of
containers
– Example with multimap-based phone book

#include <map>

multimap<string,int> pbook ;
pbook.insert(pair<string,int>(“Bjarne”,00205726666)) ; // office phone
pbook.insert(pair<string,int>(“Bjarne”,00774557612)) ; // home phone
pbook.insert(pair<string,int>(“Bjarne”,0655765432)) ; // cell phone
pbook.insert(pair<string,int>(“Fred”,0215727576)) ; // office phone

multimap<string,int>::iterator iter
begin=pbook.lower_bound(“Bjarne”),
end=pbook.upper_bound(“Bjarne”) ;

for(iter=begin ; iter!=end ; ++iter) {

cout << iter->first << “ “ << iter->second << endl ;
}

– Running iterator in standard way between bounds supplied by

lower_bound() and upper_bound() accesses all elements with key
“Bjarne”
– No special mechanism needed to indicate subsets
© 2006 Wouter Verkerke, NIKHEF
Iterators and container capabilities

• Not all containers have the same capabilities

– For example list does not allow random access (i.e. you cannot
say: give me the 3rd element. You can say: give the next element
with respect to the current iterator position, or the preceding
element)

• Solution: STL provides multiple types of iterators with

different capabilities
– All iterators look and feel the same so you don’t notice they're
different, except that if a container can’t do a certain thing the
corresponding iterator won’t let you do it either
– Within a given set of functionality (e.g. only sequential access but
no random access) all STL containers that provide that interface
look and feel the same when access through iterators

• What classes of iterators exist:

– Input, Ouput, Forward, Bidirectional, RandomAccess

© 2006 Wouter Verkerke, NIKHEF

Types of iterators

• Input iterator
– Special iterator for input, for example from keyboard
– Iterator allows to read input and must be incremented before next
input is allowed
var = *iter++ ;

• Output iterator
– Special iterator for output sequence, for example to standard
output
– Iterator allows to write and must be incremented before next
output is allowed
*iter++ = var ;

• Forward iterator
– Input and output are both allowed
– Iteration must occur in positive increments of one
*iter = var ;
var = *iter ;
© 2006 Wouter Verkerke, NIKHEF
++iter ;
Types of iterators

• Bidirectional iterator
– Can move forward and backward in steps of one

*iter = var ;
var = *iter ;
++iter ;
--iter ;

• Random access iterator

– Can move with arbitrary step size, or move to absolute locations

*iter = var ;
var = *iter ;
++iter ;
--iter ;
iter[3] = var ;
iter += 5 ;
iter -= 3 ;

© 2006 Wouter Verkerke, NIKHEF

Containers and iterators

• Iterator functionality overview

Input Iterator
Forward Iterator Bidirectional Iterator Random Access Iterator
Output Iterator

Less functionality More functionality

• Iterators provided by STL containers

Container Provided iterator

vector random access Obtaining iterators
deque random access
// Forward iterators
list bidirectional C.begin()
(multi)set bidirectional C.end()
(multi)map bidirectional
// Reverse iterators
stack none C.rbegin()
(priority_)queue none C.rend()

© 2006 Wouter Verkerke, NIKHEF

Iterators as arguments to generic algorithms

• Iterators allow generic algorithms to be applied to

contains
– Iterators hide structure of container  Access through iterator
allows single algorithm implementation to operate on multiple
container types
– Examples
vector vec<int> vec(10) ;

// Shuffle vector elements in random order

random_shuffle(vec.begin(),vec.end()) ;

// Sort vector elements according to operator< ranking

sort(vec.begin(),vec.end()) ;

list l<string> l ;
// Sort list elements according to operator< ranking
sort(l.begin(),l.end()) ;

© 2006 Wouter Verkerke, NIKHEF

STL algorithms – sort

• Sorts elements of a sequential container

– Uses operator<() for ranking
– Needs RandomAccess iterators

• Example
vector<int> grades(100) ;
sort(grades.begin(),grades.end()) ; // sort all elements

vector<int>::iterator halfway = grades.begin()+50 ;

sort(grades.begin(),halfway) ; // sort elements [0,49]

• Notes
– Pair of iterators is used to indicate range to be sorted
– Range does not include element pointed to by upper bound iterator
• Function end() returns a ‘past the end’ iterator so that the last element of
the container is included in the range when end() is used as endpoint

© 2006 Wouter Verkerke, NIKHEF

STL algorithms – find

• Finds element of a certain value in a sequential

container
– Uses operator==() to establish matches
– Expects ForwardIterator
– Return value: iterator pointing to element, end() otherwise

• Example list<int> l(10) ;

int value = 3 ;

// Find first occurrence

list<int>::iterator iter ;
iter = find(l.begin(),l.end(),value) ;
if (iter != l.end()) {
// element found
}

// Find remaining occurrences

iter = find(iter,l.end(),value)
while (iter != l.end()) {
// element found
iter = find(++iter,l.end(),value))
} © 2006 Wouter Verkerke, NIKHEF
STL algorithm – for_each

• Calls function for each element in sequential container

– Pass call back function to algorithm
– Call back function should take single argument of type matching
container, returning void
– Expects ForwardIterator

• Example
// the call back function
void printRoot(float number) {
cout << sqrt(number) << endl ;
}

// Execute the algorithm

vector<float> v(10) ;
for_each(v.begin(),v.end(),printRoot) ;

// Calls printRoot for each element of v,

// passing its value
© 2006 Wouter Verkerke, NIKHEF
STL algorithm – copy

• Copies (part of) sequential container to another

sequential container
– Takes two input iterators and one output iterator

• Example

list<int> l(10) ;
vector<int> v(10) ;

// copy from list to vector

copy(l.begin(), l.end(), v.begin()) ;

// copy from list to standard output

copy(l.begin(), l.end(), ostream_iterator<int>(cout,”\n”)) ;

• Note on ostream_iterator
– Construct output iterator tied to given ostream.
– Optional second argument is printed after each object is printed
© 2006 Wouter Verkerke, NIKHEF
 STL algorithm overview

• There are many algorithms!

accumulate adjacent_difference binary_search copy copy_backward

count count_if equal equal_range fill

fill_n find find_if for_each generate

generate_n includes inner_product inplace_merge iota

iter_swap lexographical_compare compare lower_bound make_heap

max max_element merge min_element mismatch

next_permutation partial_sort_copy partial_sum partition pop_heap

prev_permutation push_heap random_shuffle remove remove_copy

remove_copy_if remove_if replace replace_copy replace_copy_if

replace_if reverse reverse_copy rotate rotate_copy

search set_difference set_intersection set_symmetric_difference set_union

sort sort_heap stable_partition stable_sort swap

swap_ranges transform unique nthe_element partial_sort

unique_copy upper_bound

© 2006 Wouter Verkerke, NIKHEF

Modifying default behavior – Function objects
• STL container and algorithm behavior can easily be
adapted using optional call back functions or function
objects
// the call back function
void printRoot(float number) {
– Example from for_each cout << sqrt(number) << endl ;
}

// Execute the algorithm

vector<float> v(10) ;
for_each(v.begin(),v.end(),printRoot) ;

• Mechanism generalized in several ways in STL

– Customization argument not necessarily a function, but can also
be a ‘function object’, i.e. anything that can be evaluated with
operator().
– STL provides a set of standard function objects to apply common
behavior modifications to algorithms and containers

© 2006 Wouter Verkerke, NIKHEF

Function objects

• Illustration of function object concept

– A class with operator()(int) has the same call signature as
function(int)

// the call back function // the function object

class printRoot {
void printRoot(float number) { public:
cout << sqrt(number) << endl ; operator()(float number) {
} cout << sqrt(number) << endl ;
}
}

// Execute the algorithm // Execute the algorithm

vector<float> v(10) ; vector<float> v(10) ;
for_each(v.begin(),v.end(), for_each(v.begin(),v.end(),
printRoot) ; printRoot()) ;

© 2006 Wouter Verkerke, NIKHEF

Template function objects

• Function objects can also be templated

– Introduces generic function objects
– printRoot is example of unary function object (takes 1 argument)
– binary function objects are also common (2 arguments, returns bool)
• To implement comparisons, ranking etc

// the template function object

template <class T>
class printRoot {
public:
void operator()(T number) {
cout << sqrt(T) << endl ;
}
}

// Execute the algorithm on vector of float

vector<float> v(10) ;
for_each(v.begin(),v.end(),printRoot<float>()) ;

// Execute the algorithm on vector of int

vector<int> v(10) ;
for_each(v.begin(),v.end(),printRoot<int>()) ©;2006 Wouter Verkerke, NIKHEF
STL Function objects <functional>
• STL already defines several template function objects
– Arithmetic: plus, minus, divides,…
– Comparison: less, greater, equal_to
– Logical: logical_and, logical_or,…

• Easy to use standard function objects to tailor

STL algorithms
– Example

vector<int> v(10) ;

// Default sort using int::operator<()

sort(v.begin(),v.end()) ;

// Customized sort in reversed order

sort(v.begin(),v.end(),greater<int>()) ;

– Also used in some container constructors

bool NoCase(const string& s1, const string& s2) ;

© 2006 Wouter Verkerke, NIKHEF
map<string,int,NoCase> phoneBook ;
Numeric support in STL
• STL has some support for numeric algorithms as well
– Won’t mention most of them here except for one:

• Class complex
– STL implements complex numbers as template

complex<double,double> a(5,3), b(1,7) ;

complex<double,double> c = a * b ;

– Performance optimized template specializations exist for

<double,double>, <float,float>, <long double, long
double>
– Default template argument is float

complex a(5,3), b(1,7) ;

complex c = a * b ;

– Nearly all operators are implemented

• complex * double, int * complex etc etc… © 2006 Wouter Verkerke, NIKHEF
© 2006 Wouter Verkerke, NIKHEF
© 2006 Wouter Verkerke, NIKHEF
OO programming – Inheritance & Polymorphism

8 Inheritance &
Polymorphism

© 2006 Wouter Verkerke, NIKHEF

Inheritance – Introduction
• Inheritance is
– a technique to build a new class based on an old class

• Example
– Class employee holds employee personnel record

class Employee {
public:
Employee(const char* name, double salary) ;
const char* name() const ;
double salary() const ;
private:
string _name ;
double _salary ;
} ;
– Company also employs managers, which in addition to being
employees themselves supervise other personnel
• Manager class needs to contain additional information: list of subordinates

– Solution: make Manager class that inherits from Employee

© 2006 Wouter Verkerke, NIKHEF

Inheritance – Syntax

• Example of Manager class constructed through

inheritance

Declaration of public
inheritance

class Manager : public Employee {

public:
Manager(const char* name, double salary,
vector<Employee*> subordinates) ;
list<Employee*> subs() const ;
private:
list<Employee*> _subs ;
} ;

Additional data members

in Manager class

© 2006 Wouter Verkerke, NIKHEF

Inheritance and OOAD

• Inheritance means: Manager Is-An Employee

– Object of class Manager can be used in exactly the same way as
you would use an object of class Employee because
– class Manager also has all data members and member functions of
class Employee
– Detail: examples shows ‘public inheritance’ – Derived class
inherits public interface of Base class

• Inheritance offers new possibilities in OO Analysis and

Design
– But added complexity is major source for conceptual problems
– We’ll look at that in a second, let’s first have a better look at
examples

© 2006 Wouter Verkerke, NIKHEF

Inheritance – Example in pictures

• Schematic view of Manager class Terminology

class Manager ‘Derived class’

public:
list<Employee*> subs() const ;
private:
list<Employee*> _subs ;

class Employee ‘Base class’

public:
const char* name() const ;
double salary() const ;
private:
string _name ;
double _salary ;

© 2006 Wouter Verkerke, NIKHEF

Inheritance – Example in pictures

• Inheritance can be used recursively Terminology

class Director ‘Derived class’

public:
int numShares() const ;
private:
int _numShares() ;

class Manager ‘Derived class’

public: & ‘Base class’
list<Employee*> subs() const ;
private:
list<Employee*> _subs ;

class Employee ‘Base class’

public:
const char* name() const ;
double salary() const ;
private:
string _name ;
double _salary ;
© 2006 Wouter Verkerke, NIKHEF
Inheritance – Using it

• Demonstration of Manager-IS-Employee concept

// Create employee, manager record

Employee* emp = new Employee(“Wouter”,10000) ;

list<Employee*>subs(1) ;
subs.push_back(wouter) ;

Manager* mgr = new Manager(“Stan”,20000,subs) ;

// Print names and salaries using

// Employee::salary() and Employee::name()
cout << emp->name() << endl ; // prints Wouter
cout << emp->salary() << endl ; // prints 10000

cout << mgr->name() << endl ; // prints Stan

cout << mgr->salary() << endl ; // prints 20000

© 2006 Wouter Verkerke, NIKHEF

Inheritance – Using it

• Demonstration of Manager-IS-Employee concept

– A Pointer to a derived class is a pointer to the base class

// Pointer-to-derived IS Pointer-to-base
void processEmployee(Employee& emp) {
cout << emp.name() << “ : “ << emp.salary() << endl ;
}

processEmployee(*emp) ;
processEmployee(*mgr) ; // OK Manager IS Employee

– But there reverse is not true!

// Manager details are not visible through Employee* ptr

Employee* emp2 = mgr ; // OK Manager IS Employee
emp2->subs() ; // ERROR – Employee is not manager

© 2006 Wouter Verkerke, NIKHEF

OO Analysis and Design – ‘Is-A’ versus ‘Has-A’

• How is an ‘Is-A’ relationship different from a ‘Has-A’

relationship
– An Is-A relationship expresses inheritance (A is B)
– An Has-A relationship expresses composition (A is a component of B)

a Calorimeter HAS-A Position An Manager IS-An Employee

class Calorimeter { class Manager :

public: public Employee {
Position& p() { return p ; } public:
private:
Position p ; private:
} ; } ;

Calorimeter calo ; Manager mgr ;

// access position part // Use employee aspect of mgr
© 2006 Wouter Verkerke,;NIKHEF
mgr.salary()
calo.p() ;
Inheritance – constructors, initialization order

• Construction of derived class involves construction of base

object and derived object
– Derived class constructor must call base class constructor
– The base class constructor is executed before the derived class ctor
– Applies to all constructors, including the copy constructor

Manager::Manager(const char* name, double salary,

list<Employee*>& l) :
Employee(name,salary),
_subs(l) {
cout << name() << endl ; // OK - Employee part of object
} // is fully constructed at this
// point so call to base class
// function is well defined

Manager::Manager(const Manager& other) :

Employee(other), // OK Manager IS Employee
_subs(other._subs) {
// body of Manager copy constructor
} © 2006 Wouter Verkerke, NIKHEF
Inheritance – Assignment

• If you define your own assignment operator for an

inherited class (e.g. because you allocate memory) you
need to handle the base class assignment as well
– Virtual function call mechanism invokes call to derived class
assignment operator only.
– You should call the base class assignment operator in the derived
class assignment operator

Manager::operator=(const Manager& other) {

// Handle self assignment

if (&other != this) return *this ;

// Handle base class assignment

Employee::operator=(other) ;

// Derived class assignment happens here

return *this ;
}
© 2006 Wouter Verkerke, NIKHEF
Inheritance – Destructors, call sequence

• For destructors the reverse sequences is followed

– First the destructor of the derived class is executed
– Then the destructor of the base class is executed

• Constructor/Destructor sequence example

class A {
A() { cout << “A constructor” << endl ;}
~A() { cout << “A destructor” << endl ; }
}

class B : public A {
B() { cout << “B constructor” << endl ;}
~B() { cout << “B destructor” << endl ; }
}
Output
int main() {
A constructor
B b ;
B constructor
cout << endl ;
}
B destructor
A destructor © 2006 Wouter Verkerke, NIKHEF
Sharing information – protected access

• Inheritance preserves existing encapsulation

– Private part of base class Employee is not accessible by derived
class Manager
Manager::giveMyselfRaise() {
_salary += 1000 ; // NOT ALLOWED: private in base class
}

• Sometimes useful if derived class can access part of

private data of base class
– Solution: ‘protected’ -- accessible by derived class, but not by
public class Derived : public Base {
class Base { void foo() {
public: a = 3 ; // OK public
int a ; b = 3 ; // OK protected
protected: }
int b ; }
private:
int c ; Base base ;
} base.a = 3 ; // OK public
base.b = 3© ; //Wouter
2006 ERROR protected
Verkerke, NIKHEF
Better example of protected interface
class Employee {
public:
Employee(const char* name, double salary) ;
annualRaise() { setSalary(_salary*1.03) ; }
double salary() const { return _salary ; }

protected:
void setSalary(double newSalary) {
if (newSalary<_salary) }
cout << “ERROR: salary must always increase” << endl ;
} else {
_salary = newSalary ;
}
}
The setSalary() function is
private: protected:
string _name ;
double _salary ; Public cannot change salary
} except in controlled way
through public
annualRaise() method

© 2006 Wouter Verkerke, NIKHEF

Better example of protected interface
class Employee { Managers can also get
public: additional raise through
Employee(const char* name, double salary) ;giveBonus()
annualRaise() { setSalary(_salary*1.03) ; }
double salary() const { return _salary ;Access
} to protected
setSalary() method allows
protected: giveBonus() to modify salary
void setSalary(double newSalary) {
if (newSalary<_salary) }
cout << “ERROR: salary must always increase” << endl ;
} else {
_salary = newSalary ;
}
} class Manager : public Employee {
public:
private: Manager(const char* name, double salary,
string _name ; list<Employee*> subs) ;
double _salary ;
} giveBonus(double amount) {
setSalary(salary()+amount) ;
}
private:
list<Employee*> _subs ;
© 2006 Wouter Verkerke, NIKHEF
}
 Better example of protected interface
class Employee {
public:
Employee(const char* name, double salary) ;
annualRaise() { setSalary(_salary*1.03) ; }
double salary() const { return _salary ; }

protected:
void setSalary(double newSalary) {
if (newSalary<_salary) }
cout << “ERROR: salary must always increase” << endl ;
} else {
_salary = newSalary ;
}
} class Manager : public Employee {
public:
Note how accessor/modifier
private: Manager(const char* name, double salary,
stringpattern
_name ; list<Employee*> subs) ;
salary()/setSalary() is also
double _salary ;
useful for protected access giveBonus(double amount) {
}
setSalary(salary()+amount) ;
Manager is only allowed to
change salary through }
controlled method: negative private:
bonuses are not allowed… list<Employee*> _subs ;
© 2006 Wouter Verkerke, NIKHEF
}
Object Oriented Analysis & Design with Inheritance

• Principal OOAD rule for inheritance: an Is-A relation is an

extension of an object, not a restriction
– manager Is-An employee is good example of a valid Is-A relation:

A manager conceptually is an employee in all respects, but with

some extra capabilities
– Many cases are not that simple however

• Some other cases to consider

– A cat is a carnivore that knows how to meow (maybe)
– A square is a rectangle with equal sides (no!)
• ‘Is-A except‘ is a restriction, not an extension
– A rectangle is a square with method to change side lengths (no!)
• Code in square can make legitimate assumptions that both sides
are of equal length

© 2006 Wouter Verkerke, NIKHEF

Object Oriented Analysis & Design with Inheritance

• Remarkably easy to get confused

– Particularly if somebody else inherits from your class later (and you might
not even know about that)

• The Iron-Clad rule: The Liskov Subtitution Principle

– Original version

‘If for each object o1 of type S there is an object o2 of type T

such that for all programs P defined in terms of T, the behavior
of P is unchanged when o1 is substituted for o2, then S a subtype of T’

– In plain English:

‘An object of a subclass must behave indistinguishably from an

object of the superclass when referenced as an object of the superclass’

– Keep this in mind when you design class hierarchies

© 2006using Is-A
Wouter relationships
Verkerke, NIKHEF
Object Oriented Analysis & Design with Inheritance

• Extension through inheritance can be quite difficult

– ‘Family trees’ seen in text books very hard to do in real designs

• Inheritance for “extension” is non-intuitive, but for

“restriction” is wrong

• Inheritance is hard to get right in advance

– Few things are straightforward extensions
– Often behavior needs to be overridden rather than extended
– Design should consider entire hierarchy

• But do not despair:

– Polymorphism offers several new features that will make OO
design with inheritance easier
© 2006 Wouter Verkerke, NIKHEF
Advanced features of inheritance
• Multiple inheritance is also allowed
– A class with multiple base classes

class Manager : public Employee, public ShareHolder {

…
} ;
– Useful in certain circumstances, but things become complicated very quickly

• Private, protected inheritance

– Derived class does not inherit public interface of base class
– Example declaration

class Stack : private List {

…
} ;

– Private inheritance does not describe a ‘Is-A’ relationship but rather a

‘Implemented-by-means-of’ relationship
– Rarely useful
– Rule of thumb: Code reuse through inheritance©is2006
a bad idea Verkerke, NIKHEF
Wouter
Polymorphism
• Polymorphism is the ability of an object to retain its true
identity even when access through a base pointer
– This is perhaps easiest understood by looking at an example
without polymorphism

• Example without polymorphism

– Goal: have name() append “(Manager”) to name tag for manager
– Solution: implement Manager::name() to do exactly that

class Manager : public Employee {

public:
Manager(const char* name, double salary,
vector<Employee*> subordinates) ;

const char* name() const {

cout << _name << “ (Manager)” << endl ;
}

list<Employee*> subs() const ;

private:
list<Employee*> _subs ;
© 2006 Wouter Verkerke, NIKHEF
} ;
Example without polymorphism
• Using the improved manager class

Employee emp(“Wouter”,10000) ;
Manager mgr(“Stan”,20000,&emp) ;

cout << emp.name() << endl ; // Prints “Wouter”

cout << mgr.name() << endl ; // Prints “Stan (manager)”

• But it doesn’t work in all circumstances…

void print(Employee& emp) {

cout << emp.name() << endl ;
}
print(emp) ; // Prints “Wouter”
print(mgr) ; // Prints “Stan” – NOT WHAT WE WANTED!

– Why does this happen?

– Function print() sees mgr as employee, thus the compiler calls
Employee:name() rather than Manager::name() ;
– Problem profound: name() function call selected at compile time. No way
© 2006 Wouter Verkerke, NIKHEF
for compiler to know that emp really is a Manager!
Polymorphism
• Polymorphism is the ability of an object to retain its true
identity even when access through a base pointer
– I.e. we want this:

Employee emp(“Wouter”,10000) ;
Manager mgr(“Stan”,20000,&emp) ;

void print(Employee& emp) {

cout << emp.name() << endl ;
}
print(emp) ; // Prints “Wouter”
print(mgr) ; // Prints “Stan (Manager)”

• In other words: Polymorphism is the ability to treat

objects of different types the same way
– To accomplish that we will need to tell C++ compiler to look at
run-time what emp really points to.
– In compiler terminology this is called ‘dynamic binding’ and
involves the compiler doing some extra work prior executing the
emp->name() call © 2006 Wouter Verkerke, NIKHEF
Dynamic binding in C++ – keyword virtual

• The keyword virtual in a function declaration activates

dynamic binding for that function
– The example class Employee revisited

class Employee {
public:
Employee(const char* name, double salary) ;
virtual const char* name() const ;
double salary() const ;
private:
…
} ;

– No further changes to class Manager needed

… And the broken printing example now works

void print(Employee& emp) {

cout << emp.name() << endl ;
}
print(emp) ; // Prints “Wouter”
print(mgr) ; // Prints “Stan (Manager)”
© 2006 Wouter EUREKA
Verkerke, NIKHEF
Keyword virtual – some more details

• Declaration ‘virtual’ needs only to be done in the base

class
– Repetition in derived classes is OK but not necessary

• Any member function can be virtual

– Specified on a member-by-member basis

class Employee {
public:
Employee(const char* name, double salary) ;
~Employee() ;

virtual const char* name() const ; // VIRTUAL

double salary() const ; // NON-VIRTUAL

private:
…
} ;

© 2006 Wouter Verkerke, NIKHEF

Virtual functions and overloading

• For overloaded virtual functions either all or none of the

functions variants should be redefined
OK – all redefined
class A {
virtual void func(int) ;
virtual void func(float) ; NOT OK – partially redefined
}
class A {
virtual void func(int) ;
class B : public A {
virtual void func(float) ;
void func(int) ;
}
void func(float) ;
}
class B : public A {
void func(float) ;
OK – none redefined
}
class A {
virtual void func(int) ;
virtual void func(float) ;
}

class B : public A {
} © 2006 Wouter Verkerke, NIKHEF
Virtual functions – Watch the destructor
• Watch the destructor declaration if you define virtual functions
– Example

Employee* emp = new Employee(“Wouter”,1000) ;

Manager* mgr = new Manager(“Stan”,20000,&emp) ;

void killTheEmployee(Employee* emp) {

delete emp ;
}

killTheEmployee(emp) ; // OK
killTheEmployee(mgr) ; // LEGAL but WRONG!
// calls ~Employee() only, not ~Manager()

– Any resources allocated in Manager constructor will not be released as

Manager destructor is not called (just Employee destructor)
– Solution: make the destructor virtual as well

• Lesson: if you ever delete a derived class through a base

pointer your class should have a virtual destructor
– In practice: Whenever you have any virtual function, make the destructor
virtual
© 2006 Wouter Verkerke, NIKHEF
Abstract base classes – concept

• Virtual functions offer an important tool to OOAD – the

Abstract Base Class
– An Abstract Base Class is an interface only. It describes how an
object can be used but does not offer a (full) implementation

class Trajectory
public: Interface
virtual Point x(float& t)=0; only

class LineTrajectory class HelixTrajectory

public: public:
Point x(float &t) Point x(float&t) ;
Imple-
private: private: mentation
Vector _orig ; Vector _orig ;
Vector _dir ; double _rho, _phi, _d,
_kappa, _lambda ;
© 2006 Wouter Verkerke, NIKHEF
Abstract base classes – pure virtual functions

• A class becomes an abstract base class when it has one

or more pure virtual functions
– A pure virtual function is a declaration without an implementation
– Example
class Trajectory {
public:
Trajectory() ;
virtual ~Trajectory() ;
virtual Point x(float& t) const = 0 ;
} ;

– It is not possible to create an instance of an abstract base

class, only of implementations of it

Trajectory* t1 = new Trajectory(…) ; // ERROR abstract class

Trajectory* t2 = new LineTrajectory(…); // OK
Trajectory* t3 = new HelixTrajectory(…);// OK

© 2006 Wouter Verkerke, NIKHEF

Abstract base classes and design

• Abstract base classes are a way to express common

properties and behavior without implementation
– Especially useful if there are multiple implementation of a
commons interface possible
– Example: a straight line ‘is a’ trajectory,
but a helix also ‘is a’ trajectory

• Enables you to write code at a higher level abstraction

– For example, you don’t need to know how trajectory is
parameterized, just how to get its position at a give flight time.
– Powered by polymorphism

• Simplifies extended/augmenting existing code

– Example: can write new class SegmentedTrajectory. Existing
code dealing with trajectories can use new class without
modifications (or even recompilation!)
© 2006 Wouter Verkerke, NIKHEF
Abstract Base classes – Example

• Example on how to use abstract base classes

void processTrack(Trajectory& track) ;

int main() {
// Allocate array of trajectory pointers
Trajectory* tracks[3] ;

// Fill array of trajectory pointers

tracks[0] = new LineTrajectory(…) ;
tracks[1] = new HelixTrajectory(…) ;
tracks[2] = new HelixTrajectory(…) ;

for (int i=0 ; i<3 ; i++) {

processTrack(*tracks[i]) ;
}
Use Trajectory
} interface to
manipulate track
void processTrack(Trajectory& track) { without knowing
cout << “position at flight length 0 is “ the exact class
<< track.pos(0) << endl ; you’re dealing with
} (HelixTrajectory
© 2006 Wouter or
Verkerke, NIKHEF
LineTrajectory)
The power of abstract base classes
• You can even reuse existing compiled code with new
implementations of abstract base classes
• Example of reusing compiled code with a new class
– First iteration – no magnetic field
1. Write abstract class Trajectory
2. Write implementation LineTrajectory
3. Write algorithm class TrackPointPOCA to find closest point of approach
between given cluster position and trajectory using Trajectory
interface
– Second iteration – extend functionality to curved tracks in
magnetic field
1. Write implementation HelixTrajectory, compile HelixTrajectory
code
2. Link HelixTrajectory code with existing compiled code into new
executable
3. Your executable can use the newly defined HelixTrajectory objects
without further modification

• Higher level code TrackPointPOCA transparent to

future code changes!
© 2006 Wouter Verkerke, NIKHEF
Object Oriented Analysis and Design and Polymorphism

• Design of class hierarchies can be much simplified if

only abstract base classes are used
– In plain inheritance derived class forcibly inherits full specifications
of base type
– Two classes that inherit from a common abstract base class can
share any subset of their common functionality

Abstract
Base Common
Interface

Concrete Concrete
Derived Implementation Implementation
I II

© 2006 Wouter Verkerke, NIKHEF

Polymorphic objects and storage
• Polymorphic inheritance simplifies many aspects of object use
and design – but there are still some areas where you still
need to pay attention:

• Storage of polymorphic object collections

– Reason: when you start allocating memory the true identity of the matters.
you need to know exactly how large it is after all…
– Storage constructions that assume uniform size of objects also no longer
work – Use of arrays, STL container classes not possible

• Cloning of polymorphic object collections

– Reason: you want to clone the implementation class not the interface class
so you must know the true type
– Ordinarily virtual functions solves such problems, however there is no such
thing as a virtual copy constructor…

• Will look into this in a bit more detail in the next sides…
© 2006 Wouter Verkerke, NIKHEF
Collections of Polymorphic objects – storage

• Dealing with storage

– Naïve attempt to make STL list of trajectories

LineTrajectory track1(…) ;
HelixTrajectory track2(…) ;

list<Trajectory> trackList ; // ERROR

– Why Error: list<X> calls default constructor for X, but can not
instantiate X if X is an abstract classes such as Trajectory
– Solution: make a collection of pointers

Trajectory* track1 = new LineTrajectory(…) ;

Trajectory* track2 = new HelixTrajectory (…) ;

list<Trajectory*> trackList ; // OK
trackList.push_back(&track1) ;
trackList.push_back(&track2) ;

© 2006 Wouter Verkerke, NIKHEF

Collections of Polymorphic objects – storage

• But remember ownership semantics

– STL container will delete pointers to objects, but not objects
themselves
– In other words: deleting trackList does NOT delete the
tracks!

• Technical Solution
– Write a new container class, or inherit it from a STL container
class that takes ownership of objects pointed to.
– NB: This is not so easy – think about what happens if replace
element in container: does removed element automatically get
deleted on the spot?

• Bookkeeping Solution
– Document clearly in function that creates trackList that contents
of tracklist is owned by caller in addition to list itself
– More prone to mistakes © 2006 Wouter Verkerke, NIKHEF
Collections of polymorphic objects – copying

• Copying a polymorphic collection also has its issues

list<Trajectory*> trackList ;
list<Trajectory*> clonedTrackList ;

list<Trajectory*>::iterator iter ;
for(iter=trackList.begin() ; iter!=trackList.emd() ; ++iter) {
Trajectory* track = *iter ;

newTrack = new Trajectory(*track) // NOPE – attempt to

// instantiate abstract class
cloneTrackList.push_back(newTrack) ;
}

• Solution: make your own ‘virtual copy constructor’

– Add a pure virtual clone() function to your abstract base class
class Trajectory {
public:
Trajectory() ;
virtual ~Trajectory() ;
virtual Trajectory* clone() const = 0 ;
virtual Point x(float& t) const ©= 2006
0 ;Wouter Verkerke, NIKHEF
} ;
The virtual copy constructor
• Implementing the clone() function

class LineTrajectory : public Trajectory {

LineTrajectory(…) ;
LineTrajectory(const LineTrajectory& other) ;
virtual ~LineTrajectory() ;

// ‘virtual copy constructor’

virtual Trajectory* clone() const {
return new LineTrajectory(*this) ; calls copy ctor
}
}

• Revisiting the collection copy example

list<Trajectory*>::iterator iter ;
for(iter=tl.begin() ; iter!=tl.end() ; ++iter) {
Trajectory* track = *iter ;
newTrack = track->clone() ;
clonedTrackList.push_back(newTrack) ;
}
– clone() returns a Trajectory* pointer to a LineTrajectory for track1
– clone() returns a Trajectory* pointer to a HelixTrajectory for track
© 2006 Wouter Verkerke, 2
NIKHEF
Run-time type identification

• Sometimes you need to cheat…

– Example: The preceding example of cloning a list of tracks
– Proper solution: add virtual clone() function
– But what if (for whatever reason) we cannot touch the base class?
• For example: it is designed by somebody else that doesn’t want you change it,
or it is part of a commercial library for which you don’t have the source code

– Can you still tell what the true type is given a base class pointer?

• Solution: the dynamic_cast<> operator

– Returns valid pointer if you guessed right, null otherwise
Trajectory* track ;
LineTrajectory* lineTrack =
dynamic_cast<LineTrajectory*>(track) ;

if (lineTrack != 0) {
cout << “track was a LineTrajectory” << endl ;
} else {
cout << “track was something else” << endl ;
} © 2006 Wouter Verkerke, NIKHEF
Run time type identification

• Solution to trackList clone problem

list<Trajectory*>::iterator iter ;
for(iter=tl.begin() ; iter!=tl.end() ; ++iter) {
Trajectory* track = *iter ;

LineTrajectory* line = dynamic_cast<LineTrajectory*> track ;

if (line) {
newTrack = new LineTrajectory(*line) ;
continue ;
}

HelixTrajectory* helix = dynamic_cast<HelixTrajectory*> track;

if (helix) {
newTrack = new HelixTrajectory(*helix) ;
continue ;
}

cout << “ERROR: track is neither helix nor line”

}

– Obviously ugly, maintenance prone, incomplete

– Use dynamic_cast<> as last resort only!
© 2006 Wouter Verkerke, NIKHEF
C++ competes with your government
• Flip side of polymorphic inheritance – performance

• Inheritance can be taxed!

– In C++ you incur a performance overhead if you use virtual
functions instead of regular (statically bound) functions
– Reason: every time you call a virtual function the C++ compiler
inserts code that identifies the true identity of the object and
decided based on that information what function to call
– Overhead only applies to virtual functions. Regular function in a
class with other virtual functions do not incur the overhead

• Use virtual functions judiciously

– Don’t make every function in your class virtual
– Overhead is not always waste of time. If alternative is figuring out
the true identity of the object yourself the lookup step is intrinsic
to your algorithms.

© 2006 Wouter Verkerke, NIKHEF

Robust programming – Exception handling

9 Exception
handling

© 2006 Wouter Verkerke, NIKHEF

Introduction to Exception handling

• Exception handling = handling of anomalous events

outside regular chain of execution
– Purpose: Handle error conditions that cannot be dealt with locally

• Your traditional options for dealing with an error

1. Terminate the program
– Not acceptable for embedded program, e.g high voltage controller

2. Return a value representing an error

– Often return type has no ‘reserved values’ that can be used as error flag

3. Return a legal value but set a global error flag

– You make the tacit assumption that somebody will check the flag

4. Call a user supplied error function

– Just passing the buck – user function also has no good option to deal with
problem
© 2006 Wouter Verkerke, NIKHEF
Throwing and catching

• Simple example of C++ exception handling

– Exception is ‘thrown’ in case of run-time problem
– If exception is not caught (as in example below) program
terminates

int main() {
double x(-3) ;
double y = calc(x) ;
}

double calc(double x) {
if (x<0) {
// run-time error condition
throw x ;
}
return sqrt(x) ;
}

© 2006 Wouter Verkerke, NIKHEF

Exceptions and automatic variables

• How is throwing an exception better than calling abort()?

– If exception is thrown destructor is called for all automatic variables
up to point where exception is handled (in this case up to the end of
main())
– In example below, buffers of output file ofs flushed, and file closes
properly prior to program termination

int main() {
SomeClass obj ;
double x(-3) ;
double y = calc(x) ;
}
time line
1) throw f
double calc(double x) { 2) ofs destructor
ofstream ofs ; (closes file)
if (x<0) { 3) obj destructor
// run-time error condition 4) exit
throw x ;
}
return sqrt(x) ;
© 2006 Wouter Verkerke, NIKHEF
}
Catching exceptions

• You can also deal explicitly with exception ‘catching’ the

exception
– You can pass information to the handler via the thrown object
int main() {
double x(-3) ;
try { Exceptions are caught
double y = calc(x) ; in this {} block
}
catch (double x) {
cout << “oops, sqrt of “
<< ”negative number:” Exceptions handled
<< x << endl ; in this block
}
}
Display details on error
double calc(double x) {
using information passed
if (x<0) {
by exception
// run-time error condition (value of negative number
throw x ; in this case)
}
return sqrt(x) ; © 2006 Wouter Verkerke, NIKHEF
}
Catching deep exceptions

• Exceptions are also caught by try{},catch{} if they

occur deeply inside nested function calls

int main() {
double x(-3) ;
try {
double y = wrapper(x) ;
}
catch (float x) {
cout << “oops, sqrt of “
<< ”negative number”
double wrapper(double x) {
<< endl ;
// do some other stuff
}
return calc(x) – 5 ;
}
}
double calc(double x) {
if (x<0) {
// run-time error condition
throw f ;
}
return sqrt(x) ;
} © 2006 Wouter Verkerke, NIKHEF
 Solving the problem

• You can try to solve the problem in the catch block

– If you fail, call throw inside the catch block to indicate that you
give up
int main() {
double x(-3) ;
try {
int* array = allocate(10000) ;
}
catch (int size) {
cout << “error allocating array of size “
<< size << endl ;

// do some housekeeping
If allocate throws
an exception again
if (problem_solved) {
it will not be caught array = allocate(size) ;
by surrounding catch } else {
block throw ; // give up handling error here
}
}
} © 2006 Wouter Verkerke, NIKHEF
Solving the problem in steps

• A chain of error handlers int main() {

double x(-3) ;
– A rethrow allows a higher try {
level error handler to deal double y = wrapper(x) ;
}
with the problem if you can’t
catch (float x) {
// second level error handling
}
}

double wrapper(double x) {
// do some other stuff
try {
calc(x) ;
}
catch (float x) {
// first resort error handling
if (!problem_solved) {
throw ; // forward
}
}
}

double calc(double x) {
throw©f2006
; Wouter Verkerke, NIKHEF
}
There are exceptions and exceptions

• Usually more than one kind of error can occur

– Give each error its own type of exception. Each type that is
thrown can be assigned a separate catch handler:

int main() {
double x(-3) ;
try {
double y = calc(x) ;
}
catch (int x) {
cout << “oops, sqrt of “
<< ”negative integer number”
<< endl ;
}
catch (float x) {
cout << “oops, sqrt of “
<< ”negative floating point number”
<< endl ;
}
}
© 2006 Wouter Verkerke, NIKHEF
The catchall

• When you specify multiple exception handler they are

tried in order
– The catch(…) handler catches any exception. Useful to put last in
chain of handlers

int main() {
double x(-3) ;
try {
double y = calc(x) ;
}
catch (int x) {
// deal with int exception
}
catch (float x) {
// deal with float exception
}
catch (…) {
// deal with any other exception
}
}

© 2006 Wouter Verkerke, NIKHEF

Exceptions and objects – hierarchy

• So far example have thrown floats and ints as

exceptions
• What else can we throw?
– Actually, anything, including objects!

• Throwing objects is particularly nice for several reason

1. Error often have a hierarchy, just like objects can have
2. Objects can carry more information, error status, context etc …
– Example of exception hierarchy

FloatingPointError

Underflow Overflow DivideByZero

© 2006 Wouter Verkerke, NIKHEF
Throwing objects

• Example of hierarchical error handling with classes

– The hierarchy of throwable classes

class FloatingPointError {
FloatingPointError(float val) : value(val) {}
public:
float value ;
}

class UnderflowError : public FloatingPointError {

UnderflowError(float val) : FloatingPointError(val) {}
}

class OverflowError : public FloatingPointError {

OverflowError(float val) : FloatingPointError(val) {}
}

class DivZeroError : public FloatingPointError {

DivZeroError(float val) : FloatingPointError(val) {}
}

© 2006 Wouter Verkerke, NIKHEF

Catching objects

• Hierarchical handling of floating point errors

– Note that if we omit the DivZeroError handler that exception is
still caught (but then by the FloatingPointError handler)

void mathRoutine() {
try {
doTheMath() ;
}

// Catches divide by zero errors specifically

catch(DivZeroError zde) {
cout << “You divided “ << zde.value
<< “ by zero!” << endl ;
}

// Catches all other types of floating point errors

catch(FloatingPointError fpe) {
cout << “A generic floating point error occurred,”
<< ” value = “ << fpe.value << endl ;
}
}
© 2006 Wouter Verkerke, NIKHEF
Throwing polymorphic objects

• We can simplify error further handling by throwing

polymorphic objects
class FloatingPointError {
FloatingPointError(float val) : value(val) {}
virtual const char* what() { return “FloatingPointError” ; }
public:
float value ;
}

class UnderflowError : public FloatingPointError {

UnderflowError(float val) : FloatingPointError(val) {}
const char* what() { return “UnderflowError” ; }
}

class OverflowError : public FloatingPointError {

OverflowError(float val) : FloatingPointError(val) {}
const char* what() { return “OverflowError” ; }
}

class DivZeroError : public FloatingPointError {

DivZeroError(float val) : FloatingPointError(val) {}
const char* what() { return “DivZeroError” ; }
} © 2006 Wouter Verkerke, NIKHEF
Catching polymorphic objects

• Handling of all FloatingPointErrors in a single handler

– Of course only works if action for all type of errors is the same
(or implemented as virtual function in exception class)
– If structurally different error handling is needed for one particular
type of FloatingPointError you can still insert separate handler

void mathRoutine() {
try {
doTheMath() ;
}

// Catches all types of floating point errors

catch(FloatingPointError fpe) {
cout << fpe.what() << endl ;
}
}

© 2006 Wouter Verkerke, NIKHEF

 Standard Library exception classes

• The Standard Library includes a hierarchy of objects to

be thrown as exception objects
– Base class ‘exception’ in header file <exception>

• Hierarchy of standard exceptions

– Member function what() returns error message

exception

logic_error runtime_error

length_error range_error

domain_error bad_alloc bad_cast overflow_error

out_of_range bad_exception bad_typeid underflow_error

invalid_argument ios_base::failure

– Note that catching class exception does not guarantee catching

all exceptions, people are free to start their own hierarchy
© 2006 Wouter Verkerke, NIKHEF
Where to go from here

10 Where to go
from here

© 2006 Wouter Verkerke, NIKHEF

Recommended reading

• The C++ syntax in alls its detail

– The C++ programming Language , 3rd edition (Bjarne Stroustrup)

• How to program good C++

– Catching those hidden performance issues
– Good coding practices etc etc
– ‘Effective C++’ & ‘More Effective C++’ (Scott Meyers)

• A book with a lot of practical examples

– ‘Accelerated C++’ (Andrew Koenig & Barbara Moe)

• Software design
– ‘Design Patterns’ (Gamma et al.) [ not for beginners ]

© 2006 Wouter Verkerke, NIKHEF

Web resources

• Frequently Asked Questions on C++ (and answers)

– http://www.parashift.com/c++-faq-lite/

• A reference guide to the Standard Template Library

– http://www.sgi.com/tech/stl/

© 2006 Wouter Verkerke, NIKHEF