C++ Operators
Copyright 1990 by John M. Dlugosz
Typesetting notes:
@I(…) surrounds Italic text
@b(…) is for bold face.
Due to the heavy use of programming keywords
in the text, it would be impossible to read without some
representation for the alternate font. You can use a search and
replace to convert these to your preferred format (like ` ‘ or [ ]
for reading online) or the printer control codes you need or the
correct commands to import into your word processor.
Say you have defined a @I(complex) type to work with complex numbers.
In C, to add two complex numbers you would have to have a
function like @I(complex add(complex,complex)). An expression of any
size would end up looking more like LISP than C. In C++, you can
overload operators. In this case, I would much rather say @I(c=a+b;)
than @I(c=add(a,b)). Naturally, I can do exactly that.
The name of an operator is the reserved word @I(operator) followed by
the symbol of the operator being referred to. In this case,
@I(operator+) is what I am after. @I(operator+) can be treated just like
any other function name. That is how I define the function:
@I(complex operator+ (complex,complex);) It is exactly like @I(add())
except the name of the function is now @I(operator+). I can use it
just like I did above, as @I(c=operator+(a,b);) That is hardly an
improvement! But I can also use the operator in its natural
operator syntax, as @I(c=a+b;)
The operators are just like functions with a special name, but
some restrictions apply. You can only define those operators
that exist, with the correct number of parameters. The type of
the parameters are much more flexible, and is the whole point of
defining them. However, at least one parameter to an operator
function must be of a class type. So, you could not define @I(void
operator+ (char*, char*);) because all arguments are of built-in
types.
Operators can be member functions. In this case, the first
argument is the receiver and the function is declared with one
less argument than it actually has. So if I defined operator+ as
a member, I would have @I(complex complex::operator+ (complex);) It
can be used as @I(c=a.operator+(b);) as a member function with a
funny name, or as the normal infix operator. In the infix
syntax, the left argument is taken as the object being operated
on.
@SECTION(available operators)
These are the operators that can be overloaded:
new delete @I(storage allocation)
+ – * @I(both unary and binary forms
& @I(unary and binary, unary is special)
= @I(assignment operator is special)
/ % ^ | ~ ! < > << >> ==
!= += -= *= /= %= ^= &= |= ->*
&& || , @I(have left to right evaluation)
++ — @I(prefix and postfix, special way to distinguish
-> () [] @I(special in various ways)
In addition, @I(conversion operators) are also defined with the
operator keyword. Conversion operators are covered in a later
chapter.
Some operators can be used in more than one way. For example,
operator- can be used as unary (one argument) or binary (two
arguments). All operators keep the same order of precedence when
they are redefined.
@SECTION(defining an operator)
Back to the example of complex addition. Here is a @I(complex) class
that has operator+= and operator+ defined for complex numbers.
@BEGIN(LISTING)
//example using overloaded operators to define arithmetic
//on complex numbers.
class complex {
double x, y;
public:
complex (double xx, double yy) { x=xx; y=yy; }
complex& operator+= (const complex& b);
friend complex operator+ (const complex& a, const complex& b);
};
complex& complex::operator+= (const complex& b)
{
x += b.x;
y += b.y;
return *this;
}
complex operator+ (const complex& a, const complex& b)
{
complex temp= a;
return temp += b;
}
@END(LISTING)
The operator+= is defined as a member function. It takes one
argument in addition to the implicit @I(this). The line @I(return temp
+= b;) is equivalent to @I(return temp.operator+=(b);). Remember,
operators defined as members have one less argument defined than
they actually take.
@SECTION(standard input and output)
The @I(stream) library in C++ uses overloaded operators for standard
input and output. The operator<< is used as a “put to”, and
operator>> is used as a “get from”. There are classes @I(ostream)
and @I(istream) to handle output and input. Functions such as:
@BEGIN(LISTING)
ostream& operator<< (ostream&, int);
ostream& operator<< (ostream&, const char*);
istream& operator>> (istream&, int&);
istream& operator>> (istream&, char*&);
@END(LISTING)
Are used to handle input and output. Each operator returns the
first argument as the function result, so they can be chained
together. The standard input and output are available as
variables @I(cin) and @I(cout).
@BEGIN(LISTING)
cout << “hello world!n”;
cout << “the answer is:” << x;
cin >> x >> y; //read values into x and y;
@END(LISTING)
You can define your own output and input operators to operate on
your own types.
@BEGIN(LISTING)
ostream& operator<< (ostream& o, const complex& c)
{
//I’m a friend, and have access to c.x and c.y
o << ‘(‘ << c.x << ‘,’ << c.y << ‘)’;
return o;
}
istream& operator>> (istream& i, complex& c)
{
double x, y;
char c;
i >> c >> x >> c >> y >> c;
c= complex(x,y);
return o;
}
@END(LISTING)
The input operator is a little overkill for such a simple
structure– you could have made it a friend and just read in the
x and y components directly. But this illustrates a more
general technique. You read the data needed to create an object,
and then call a constructor to put it together.
In general, use @I(cin >> x;) to read a value, and @I(cout << x;) to
write a value. That is all you have to remember until you get to
the chapter on streams.
@SECTION(operator=)
Normally when you write an assignment such as @I(x=y;) the value of y
is moved to x simply by copying the bits. The assignment
operator lets you define how assignment will take place instead.
Consider a class that contains pointers. When you assign one
instance to another, what you really want is a “deep copy” where
the pointers are also duplicated to give the new copy its own.
@BEGIN(LISTING)
class C {
char* name;
int age;
public:
// other members omitted for example…
C& operator= (C&); //assignment operator
};
C& C::operator= (C& x)
{
age= x.age; //copy each element
free (name); //free `name’ before overwriting it!
name= strdup (x.name); //make unique copy
return *this;
}
@END(LISTING)
The assignment operator must be a member. It should be used to
provide assignment, and not for anything else. The parameter can
be any type though. In fact, you can have multiple assignment
operators defined for a class, so you can assign various things
to it.
@BEGIN(LISTING)
class String {
char* s;
int len;
public:
String& operator= (String&);
String& operator= (const char*);
String& operator= (char);
};
@END(LISTING)
In class C, notice that the operator= returns a @I(C&) and in fact
returns @I(*this). This is common practice, and allows chaining of
assignment and the use of the assignment in a larger expression
as with the built-in use. The built-in assignment operator
returns the first argument as its result, so you can do things
like @I(foo(a=b+1);) and @I(a=b=c;). It is common practice to continue
this tradition by returning @I(*this) from operator=.
Assignment in C does not return an lvalue, but my supplied
operator= does. In C++, this has been extended to the built-in
definitions as well. This means that you could say something
like @I(p= &(c=b);) which takes the address of the return value from
assignment, even for built-in types. You could not do that in C:
you would get the error @I(argument to `&’ must be an lvalue).
You can define operator= to return anything you like. Generally
it returns @I(*this) or has no return value (defined as void), but
there may be reasons to do otherwise.
The operator= is unique in that it is not inherited. Actually,
it is inherited in a special way. If you have a class that has
members or base classes that have an operator= defined, and do
not provide an operator= in this class, the compiler generates
one for you that calls operator= for those pieces of the class
that have an operator= defined, and copies the rest.
@BEGIN(LISTING)
// example of derived class needing operator=
class D {
int x;
C y; // C from above example
public:
D& operator= (D&);
};
D& D::operator= (D& second)
{
x= second.x; //normal copy
y= second.y; //calls C::operator=()
return *this;
}
@END(LISTING)
Since class D contained a member that should not be copied in the
plain way, it defined an operator= that copied it correctly by
calling the operator= on that member. However, if I had not
written @I(D::operator=()) then it would still have done the same
thing! This is sometimes called the Miranda rule: if you do not
write an operator=, one will be provided for you. Beware though that
if you do write an operator= for a class, make sure you indeed
copy everything in it.
@b(Version Notes)
In C++ versions prior to 2.0, there was no Miranda rule. The
operator= was not inherited at all. If you did not provide a new
operator= in a derived class, or did not write an operator= in a
class with members that need it, you get a bit-for-bit copy
anyway.
@SECTION(operator[])
The operator[] can be overloaded, and it is one of the special
ones. First of all, the syntax is different. Writing @I(x[y];) will
call @I(x.operator[](y);) Second, it must be a member function.
Here is an example of a vector class that uses this operator.
This implements an array that grows as needed to accommodate new
elements.
@BEGIN(LISTING)
#include <iostream.h>
typedef int eltype;
class vector {
eltype* elements;
int capacity;
public:
vector (int startsize= 10);
~vector() { delete[capacity] elements; }
int size() { return capacity; }
eltype& operator[] (int index);
};
vector::vector (int startsize)
{
capacity= startsize;
elements= new eltype[startsize];
}
eltype& vector::operator[] (int index)
{
if (index >= capacity) { //out of bounds
eltype* new_array= new eltype[index+1];
for (int loop= 0; loop < index; loop++)
new_array[loop]= elements[loop];
delete[capacity] elements;
elements= new_array;
capacity= index+1;
}
return elements[index];
}
main()
{
vector a;
for (;;) {
char c;
int index, value;
cout << “operation (r,w,q)? “;
cin >> c;
switch (c) {
case ‘q’: goto done;
case ‘r’: //read test
cout << ” index “;
cin >> index;
if (index < 0 || index >= a.size())
cout << “index out of range.”;
else cout << “contains ” << a[index];
break;
case ‘w’: //write test
cout << ” index and value “;
cin >> index >> value;
a[index]= value;
break;
case ‘s’: //size?
cout << “array holds ” << a.size() << ” elements.”;
break;
case ‘l’: //list them
for (index= 0; index < a.size(); index++)
cout << “n [” << index << “] ” << a[index];
break;
}
cout << ‘n’;
}
done:
cout << “program finished.n”;
}
@END(LISTING)
The operator[] returns an element of the array, and returns it by
reference. This allows it to be used on the left hand side of an
assignment, as seen in the write test in the main program.
This is a typical application– to access a collection of
elements. The parameter to operator[] can be of any type though.
For example, an associative array could associate strings and
numbers. Index it with a number and it returns the string, and
index it with a string and it returns the number! Interesting?
Here is the code:
@BEGIN(LISTING)
#include <iostream.h>
#include <string.h>
/* associative array of strings */
typedef char* eltype;
class asa { //associative string array
eltype* elements;
int capacity;
public:
asa (int startsize= 10);
~asa() { delete[capacity] elements; }
int size() { return capacity; }
eltype& operator[] (int index);
int operator[] (eltype s); //the ‘backwords’ version
};
asa::asa (int startsize)
{
capacity= startsize;
elements= new eltype[startsize];
}
int asa::operator[] (eltype s)
{
for (int loop= 0; loop < capacity; loop++)
if (!strcmp(s,elements[loop])) return loop; //found it
//did not find it– add it
(*this)[loop]= strdup(s);
return loop;
}
eltype& asa::operator[] (int index)
{
if (index >= capacity) { //out of bounds
eltype* new_array= new eltype[index+1];
for (int loop= 0; loop < index; loop++)
new_array[loop]= elements[loop];
delete[capacity] elements;
elements= new_array;
capacity= index+1;
}
return elements[index];
}
main()
{
asa a;
for (;;) {
char c;
int index;
char value[40];
cout << “operation (r,w,b,q)? “;
cin >> c;
switch (c) {
case ‘q’: goto done;
case ‘b’: // ‘backwards’ read test
cout << “enter string “;
cin >> value;
cout << “found at ” << a[value];
break;
case ‘r’: //read test
cout << ” index “;
cin >> index;
if (index < 0 || index >= a.size())
cout << “index out of range.”;
else cout << “contains ” << a[index];
break;
case ‘w’: //write test
cout << ” index and value “;
cin >> index >> value;
a[index]= strdup(value);
break;
case ‘s’: //size?
cout << “array holds ” << a.size() << ” elements.”;
break;
case ‘l’: //list them
for (index= 0; index < a.size(); index++)
cout << “n [” << index << “] ” << a[index];
break;
}
cout << ‘n’;
}
done:
cout << “program finished.n”;
}
@END(LISTING)
This program was based heavily on the previous. I started with
the vector class. A global search and replace changed the name
from @I(vector) to @I(asa), and simply changing the typedef of eltype
changed the type to operate on char*’s instead of ints. Then I
added the new function: @I(int asa::operator[] (eltype s)). It
simply searches the array for a matching string.
The test driver also required only minor changes. Changing the
definition of the variable @I(value) to the new type also changed the
behavior of all the input and output statements that use it. All
I had to do was add a new case to try the string look up.
Note that this program is not all that great. The simple stream
input does not let me enter strings with a space in them, the
array is not initialized so listing it can cause problems, and it
is up to the user of the class to free up pointers before they
are overwritten.
@SECTION(operator())
The operator() is another strange one. It is sometimes called
the @I(function-call operator). Like operator[], it must be a
member. The way to call it is easier shown than explained. Look
how it is used:
@BEGIN(LISTING)
class C {
//stuff…
public:
void operator() (int x);
};
C x;
//call operator()
x(5);
x.operator()(5); //same thing
@END(LISTING)
Here, @I(x) is a variable of type @I(C). Using the name as if it were a
function will call operator(). When defining operator(), you
give a parameter list just line any other operator, so you have
two sets of parenthesis in the definition. When calling it, you
just have the one set. The operator() can be defined to have any
number of arguments– it is the only operator that can do this.
@BEGIN(LISTING)
void C::operator()(); //no args
int C::operator() (char* s, int y); //2 args
@END(LISTING)
So what is the point in having such an operator? For one, it can
be used in a manner similar to the operator[]. Here is a vector
class that uses operator[] to access an element with range
checking, and operator() to access an element without range
checking.
@BEGIN(LISTING)
// example of using operator() on a vector similar to operator[]
typedef int eltype;
class vector {
eltype* contents;
int first, last; //bounds of the array
public:
vector (int first, int last);
~vector() { delete[last-first+1] contents; }
eltype& operator[] (int index); //access with range checking
eltype& operator() (int index) //access without range
checking
{ return contents[index-first]; }
};
vector::vector (int f, int l)
{
first= f;
last= l;
contents= new eltype[l-f+1];
}
eltype& vector::operator[] (int index)
{
if (index < first || index > last) {
//deal with the error somehow. In this case,
//I’ll return a special internal value that can
//be written to without trashing memory.
static eltype dump;
return dump;
}
return contents[index-first];
}
@END(LISTING)
If @I(a) is of type vector, you could access @I(a[5]) or @I(a(5)) which are
similar in meaning. Having both [] and () available gives two
different ways to subscript a vector.
Another thing you can do with the operator() is to take advantage
of its ability to have different numbers of arguments. The
operator[] can only take one argument, but you might have a
matrix class that takes two subscripts. You could use operator()
to subscript the class instead.
@BEGIN(LISTING)
// example of using operator() instead of operator[]
// so I can use two arguments.
#include <assert.h>
const int matsize= 3;
typedef double eltype;
class square {
eltype data[matsize][matsize];
public:
eltype& operator() (int x, int y);
};
eltype& square::operator() (int x, int y)
{
assert (x > 0 && y > 0 && x < matsize && y < matsize);
return data[x][y];
}
@END(LISTING)
Given a variable @I(M) of type @I(square), you could write @I(M(1,2)) to
access an element.
In the class definition, the eltype definition is used as usual.
In this example, the size of the matrix is specified with @I(matsize)
as well. To change the size, you only need to change this
definition. All other parts of the code reference the size by
this name. In C, you would have to use a #define for this
purpose. In C++, a const variable can be used in constant
expressions, such as the size of an array definition.
Another variation on this theme is a string class which uses
operator() to take a substring. Writing @I(s(3,7)) would return the
third through seventh characters in the string s.
Another time operator() is used in when a class only has one
method, or one very important method. Consider an iterator
class. It steps through a linked list, returning the next
element each time the @I(next()) method is called. Rather than
calling it @I(next()), this sometimes uses operator() for that
purpose.
@BEGIN(LISTING)
// example of using operator() in an iterator
class node {
friend class iterator; //grant class iterator access to my
private data
node* next;
public:
char* data;
void insert_after (node*); //insert a node after this node.
};
class iterator {
node* p; //keep track of my position in the list
public:
node* operator()(); //advance to the next position
iterator (node* n) { p= n; }
};
void node::insert_after (node* p)
{ //insert p after this node
p->next= next;
next= p;
}
node* iterator::operator()()
{ //return the node and advance to the next node
if (!p) return p; //end of the line, don’t advance
node* temp= p;
p= p->next;
return temp;
}
@END(LISTING)
@SECTION(operator->)
The operator-> is the strangest one of all. It must be a member.
How it is called, and what it does, takes some explaining.
The operator-> must be defined to return a pointer type or a
class type that itself has an operator-> defined. The call is
made with an object on the left and a member name on the right,
such as @I(x->a), but the @I(a) is not an argument to the function. The
operator-> is applied to @I(x), and then the result is used on the
left side of -> again. The @I(a) will be the name of a field, not an
argument of any kind.
The example @I(x->a) is equivalent to @I((x.operator->())->a). The
operator is “slipped in” to the member access.
Here is an example of using operator-> to implement a “smart
pointer”.
@BEGIN(LISTING)
//example of using operator-> to create a “smart pointer”
class C {
//members go here…
public:
int x; //a public data member
void dosomething(); //member function
};
class Cptr {
C* p;
public:
Cptr() { p=0; } //always initialized
Cptr& operator= (C* ptr) { p= ptr; return *this; };
C* operator->();
};
extern void error(); //report an error, somehow
C* Cptr::operator->()
{
if (!p) { //oops! NULL pointer
error();
}
else return p;
}
@END(LISTING)
The smart pointer class holds a pointer to a C, and has
operator-> defined on it so it will return that pointer. You
could have:
@BEGIN(LISTING)
C x;
Cptr p;
p= &x;
y= p->x; //refer to element in C
p->dosomething(); //even member functions
@END(LISTING)
The operator-> is a unary operator that returns a C*, and then
the -> operation is redone with that return value on the left.
So @I(p->x) is equivalent to @I((p.operator->())->x;). If that still
confuses you, remember that the operator is simply a member
function with a funny name. Calling it @I(fetch()) would let you say
@I((p.fetch())->x;) which is perhaps clearer.
@SECTION(operator&)
The unary form operator& is unusual because it is already defined
for all class type. It normally takes the address of the object.
You can redefine it to do anything you want. Normally it is used
to inform the system that an object is having its address taken.
It is sometimes used in debug code to report to the programmer
when objects have their address taken.
@BEGIN(LISTING)
someclass* someclass::operator& ()
{
report_on (this); //tell the programmer
return this; //and do what I came for.
}
@END(LISTING)
This operator can be defined as a member function as shown above,
or as a non-member function. You should watch out for pitfalls
when using an operator&. Namely, how do you take the address of
an object if operator& is defined? This suggests one use of
operator& is to make a class where you cannot take the address of
an object– operator& is private or causes a run-time error to be
printed.
In the example of operator-> a smart pointer was illustrated.
The Cptr type had an operator= that let you assign a C* to a
Cptr. Instead, you could use operator& to make you use smart
pointers exclusively: define a @I(Cptr C::operator&();) so that
taking the address of a C object gives a smart pointer directly.
@SECTION(operator,)
The comma operator is not all that unusual. It is rarely used
because the comma is used so much in C++ already. It is unusual
in the respect that the comma is already defined between class
types, so you have to be careful sometimes in knowing how
overloading with resolve.
The comma operator forces left-to-right evaluation. That can be
handy. The order of precedence is very low, so you will usually
need parenthesis around the comma expression.
@b(come up with an example)
@SECTION(operator++ and operator–)
These two are similar, and I’ll just talk about operator++. The
same comments apply to operator–.
There are two forms of ++ for built-in types. As a C programmer,
you know that @I(y= x++;) and @I(y= ++x); has different meanings. In C++
you can define both the prefix and postfix forms.
Defining the prefix form is exactly what you would expect, since
it is the same as any other unary operator. A member function
such as @I(C::operator()) or a nonmember such as @I(operator++(C&))
will do.
The postfix form is defined with an extra argument. This second
argument is an int, and is always passed a value of 0. Defining
@I(C::operator++(int)) or @I(operator++ (C&,int)) will define a postfix
operator.
Here is the smart pointer example again, showing prefix and
postfix operators added.
@BEGIN(LISTING)
//example of using operator++ with the smart pointers
class C {
//members go here…
public:
int x; //a public data member
void dosomething(); //member function
};
class Cptr {
C* p;
public:
Cptr() { p=0; } //always initialized
Cptr& operator= (C* ptr) { p= ptr; return *this; };
C* operator->(); //see example in operator-> section
Cptr& operator++(); //preincrement
Cptr operator++(int); //postincrement
};
Cptr& Cptr::operator++()
{ //preincrement
++p;
return *this;
}
Cptr operator++ (int)
{ //postincrement
Cptr temp= *this;
++p;
return temp; //return original unmodified copy
}
@END(LISTING)
@b(Version Note)
In C++ versions prior to 2.1, the postfix form was not available.
The only way to define an operator++ or operator– was with one
argument. It called this same function for either prefix or
postfix use. For compatibility, do not use such a function as a
postfix operator.