POLYMORPHISM with C++ - OBJECT ORIENTED PROGRAMMING

REQUIREMENTS:

-> Should know Inheritance in C++

In the end I explain how the compiler implements polymorphism and so knowledge of Assembly Language will be a great help in understanding Polymorphism. Nevertheless, I will still try to make the assembly explanation as simple as possible. So even if you do not know Assembly, try understanding it.

I. WHAT IS POLYMORPHISM?

Polymorphism is in short the ability to call different functions by just using one type of function call. It is a lot useful since it can group classes and their functions together. Polymorphism is the most important part of Object-Oriented Programming. Some people feel that if they have an idea of what classes are they have stepped in the object-oriented world. But this is not true. Polymorphism is the core of object-oriented programming and if anybody stops here he's missing out the best part of Object Oriented Programming (OOP). All this may seem a lot hard to understand but read on... you'll understand better as you keep reading.

Let us now try to understand polymorphism with the help of an example. Suppose we want to draw a picture consisting of circles, squares, lines and triangles. So we can make a class Shape and create an instance of it like this:

      Shape *s[100];

Now all the addresses of the objects of the other classes (line, circle, etc.) are stored in the Shape Array. And then to draw the Picture all we have to do is this:

   for(int i=0;i<=100;i++)
       s[i]->draw();

Now as the loop runs different draw functions of each class is called. This is great because:

1. Functions from different classes are executed through the same function call.
2. The Array s[] has been defined to contain shape pointers and not square or triangle pointers.

This can be done by using Virtual Functions.

II. VIRTUAL FUNCTIONS ???

The Literal Meaning of Virtual means to appear like something while in reality it is something else ie. when virtual functions are used, a program appears to call a function of one class but actually it may be calling a function from another class. In the previous example draw() is a virtual function since it calls different draw functions from different classes by using the same function call draw();

Now how do we know which version of draw() would be called during execution? Which draw() function would get used depends on the contents of s[i].But for this polymorphic approach to work we must satisfy the following conditions:

• The Base class must contain a draw() function which is declared virtual.
• All other classes (line, circle, etc.) should be derived from the base class.

Well, all this may be hard to understand in just one go so we'll start using programs that'll help us understand better. Here's the First One.

#include 

class base                          //Base Class
{
 public:
   void func()
   {
     cout<<"In base::func()\n";
   }
};

class d1:public base                       // Derived Class 1
{
 public:
   void func()
   {
    cout<<"In d1::func()\n";
   }
};

class d2:public base     // Derived Class 2
{
 public:
   void func()
   {
    cout<<"In d2::func()\n";
   }
};

void main()
{
  d1 d;
  base *b=&d;
  b->func();
  d2 e;
  b=&e;
  b->func();
}

Run this program and you would see that the output would be:

In base::func()
In base::func()

Shouldn't this statement give an error? (b=&e;) No. Since the compiler allows a pointer of a base class to accept addresses of derived class objects. This is known as UPCASTING. Here the Compiler looks at the type of pointer b and since it belongs to the base class it calls the base class function.

But now, let's make a slight modification in our program. Precede the declaration of func() in the base class with the keyword virtual so that it looks like this:

virtual void func()
{
   cout<<"In base::func()\n";
}

Now Compile and Run the Program. Now the Output is:

In d1::func()
In d2::func()

This time the Compiler looks at the contents of the pointer instead of it's type. Hence since addresses of objects of d1 and d2 classes are stored in *b the respective func() is called. But this way how does the compiler know which function to compile when it doesn't know which object's address 'b' might contain? Which version does the compiler call?

Actually even the compiler does not know which function to call at compile-time. Hence it decides which function to call at run-time with the help of a table called VTABLE. Using this table the compiler finds what object is pointed by the pointer b and then calls the appropriate function. VTABLE is explained later.

The method by which the compiler decides which function to call at run-time is known as late-binding or dynamic-binding. It slows down the program but makes it a lot more flexible.

III. PURE VIRTUAL FUNCTIONS

Now we realise that since the base class virtual function never gets called anyway we'd better keep it's body blank. But there's a better way to do this. We can change the virtual function func() in the base class to the following:

virtual void func()=0;

The =0 is not an assignment operator here but it is just a way of telling the compiler that the function has no body. But there is another side of this. An object of a class which contains a pure virtual function cannot be created. It seems logical enough ie. If you have classes triangle, square, circle derived from shape class we wouldn't want to make an object of the shape class. Hence the shape class should be provided with a pure virtual function. If you even try to create an object of a class containing a pure virtual function the compiler would report an error even pointing out which pure virtual function prevents you from creating an object.

IV. HOW VIRTUAL FUNCTIONS WORK

Using Virtual Functions is just one part of polymorphism and knowing how they work completes the other half. When the keyword 'virtual' is inserted in the declaration of the function the compiler inserts all mechanisms in the program to use Virtual Functions. Each Class has a VTABLE that stores the functions that it can access and each class contains a VPTR that can access the VTABLE. Look at this program and the table below it and you will understand the VTABLE and the VPTR.

#include 

class item
{
  public:
     virtual void price()
     {
       printf("In item::price()\n");
     }
     virtual void type()
     {
       printf("In item::type()\n");
     }
     void display();
};
void item::display(){printf("In item::display()\n");}

class microwave:public item
{
 public:
    void price()
    {
      printf("Microwave::Price()\n");
    }

    void type()
    {
      printf("Microwave::type()\n");
    }
};

class computer:public item
{
  public:
     void price()
     {
        printf("Compuer::Price()\n");
     }
};

class radio:public item
{
 public:
    void type()
    {
      printf("radio::type()\n");
    }
};

void main()
{
  microwave m1;
  computer c1;
  radio r1;

  item *i=&m1;
  i->price();
  i->type();
  printf("\n");

  i=&c1;
  i->price();
  i->type();
  i->display();
  printf("\n");

  i=&r1;
  i->price();
  i->type();
  printf("\n");

  microwave m2;
  i=&m2;
  i->price();
  i->type();
  i->display();
  printf("\n");
}

The Output of this Program would be:

Microwave::Price()
Microwave::type()

Compuer::Price()
In item::type()
In item::display()

In item::price()
radio::type()

Microwave::Price()
Microwave::type()
In item::display()

Now here is how the VTABLE of each class Looks Like:

-------------------------------------------------------------------------------|
ITEM POINTERS   |     OBJECTS         |                 VTABLES                |
-------------------------------------------------------------------------------|
i-------------> |    item{VPTR}---->  |        &item::price()                  |
                |                     |        &item::type()                   |
-------------------------------------------------------------------------------|
                |                     |                                        |
i-------------> |microwave m1,m2{VPTR}|        µwave::price()             |
                |                     |        µwave::type()              |
-------------------------------------------------------------------------------|
                |                     |                                        |
i-------------> | computer c1{VPTR}-> |        &computer::price()              |
                |                     |        &item::type()                   |
-------------------------------------------------------------------------------|
                |                     |                                        |
i-------------->| radio r1{VPTR}----> |        &item::price()                  |
                |                     |        &radio::type()                  |
-------------------------------------------------------------------------------|

First the VPTR Pointer is initialised to it's proper VTABLE by the constructor which is automatically done by the compiler. When a Virtual Function is being called the VPTR looks up the VTABLE and calls the virtual function. If the function is not present in the VTABLE [like here display()] then the function of the base class is called. So everywhere where the display function is called item::display() is called every time. No matter how many objects of a class are created they all point to the same VTABLE of the class.

V. ARE VIRTUAL FUNCTIONS OPTIONAL?

Normal Function calls are called by the Assembly instruction 'call' while virtual functions require complex instructions. This takes up code space as well as execution time.

Virtual Functions reduce the code's speed. Some languages like SmallTalk perform Late Binding every time a function is called and hence SmallTalk Programs aren't fast enough. But C++ is a Superset of C where Efficiency is important and hence C++ allows both static binding as well as late binding. The default convention used is static binding so that there is no loss in speed.

Don't stop using Virtual Functions in your classes just because they reduce execution speed. In fact, it makes it easier to manage and code and its advantages are more than its disadvantages. So wherever possible use Virtual Functions in your classes.

VI. MISCELLANEOUS

1. If a Virtual Function is called within a derived classes constructor or destructor then the derived function is always called.
2. If b is a base class pointer and d is a derived class pointer then b=d will copy only the base class contents and remove the derived class contents. This is known as Object Slicing and should be avoided.
3. For a Virtual Function to work the function must be present in that base class even though it is declared virtual in the derived class.
4. Virtual Destructors can also be used and they allow execution of the derived destructor first and then it calls itself.

VII. VIRTUAL BASE CLASSES

Consider a situation when a 'base' class has two classes derived from it. For example derived1 and derived2. Suppose we create another class which derives itself from both the derived classes ie. derived3. Now suppose a member function of derived3 wants to access data or functions in a base class. Since derived1 and derived2 are derived from base each inherits a copy of base. This copy is referred to as a sub object. Now when derived3 refers to the data in the base class, which of the two copies should it access? The compiler notices this ambiguous situation and reports an error. To get rid of this we should make derived1 and derived2 as virtual base classes. This is shown in the following program.

#include 

class base
{
   protected:
   int data;
   
   public:
   base()
   {
    data=10;
   }
};

class derived1 : virtual public base
{};

class derived2 : virtual public base
{};

class derived3 : public derived1,public derived2
{
   public:
   int getdata()
   {
      return data;
   }
};

void main()
{
  derived3 d3;
  int val=d3.getdata();
  cout<<val<<endl;
}

Using the keyword virtual in the two classes derived1 and derived2 makes them share a single subobject of the base class hence eliminating all ambiguity since there is only one subobject for derived3 to access. Hence derived1 and derived2 are known as virtual base classes.

VIII. VTABLE COMPILATION PROOF

THIS PROGRAM IS COMPILED BY BORLAND C++ 5.02

Lot of people know that the compiler uses a VTABLE to implement Virtual Functions but few get to see the actual code that the compiler generates. To save space I shall compile the same program that uses the item, microwave, computer and radio class and explain how the compiler implements Virtual Functions. I have used IDA Pro to disassemble this program. Here is the disassembled listing followed by the explanation. I have included line numbers so that I can just refer to the line numbers while explaining how the program works.

;                           THE FUNCTION  main()
_main		proc near

m2_VPTR		= dword	ptr -10h
r1_VPTR		= dword	ptr -0Ch
c1_VPTR		= dword	ptr -8
m1_VPTR		= dword	ptr -4
argc		= dword	ptr  8
argv		= dword	ptr  0Ch
envp		= dword	ptr  10h

1       	push	ebp
2		mov	ebp, esp
3		add	esp, 0FFFFFFF0h	; 16 bytes allocated on	the stack
4		push	ebx		; Register Saved
5		mov	[ebp+m1_VPTR], offset item_VTABLE
6		mov	[ebp+m1_VPTR], offset microwve_VTABLE
7		mov	[ebp+c1_VPTR], offset item_VTABLE
8		mov	[ebp+c1_VPTR], offset computer_VTABLE
9		mov	[ebp+r1_VPTR], offset item_VTABLE
10		mov	[ebp+r1_VPTR], offset radio_VTABLE
11		lea	ebx, [ebp+m1_VPTR]
12		push	ebx		; this*	pushed
13		mov	eax, [ebx]
14		call	dword ptr [eax]	; Calls	microwve_price
15		pop	ecx		; 4 bytes freed	used up	by this*
16		push	ebx		; this*	pushed
17		mov	edx, [ebx]
18		call	dword ptr [edx+4] ; Calls microwve_type
19		pop	ecx		; 4 bytes freed	used up	by this*
20		push	offset newline	; __va_args
21		call	_printf
22		pop	ecx		; 4 bytes cleared used up by argument
23		lea	ebx, [ebp+c1_VPTR]
24		push	ebx		; this*	pushed
25		mov	eax, [ebx]
26		call	dword ptr [eax]
27		pop	ecx
28		push	ebx
29		mov	edx, [ebx]
30		call	dword ptr [edx+4]
31		pop	ecx
32		push	ebx
33		call	item_display
34		pop	ecx
35		push	offset newline1	; __va_args
36		call	_printf
37		pop	ecx
38		lea	ebx, [ebp+r1_VPTR]
39		push	ebx
40		mov	eax, [ebx]
41		call	dword ptr [eax]
42		pop	ecx
43		push	ebx
44		mov	edx, [ebx]
45		call	dword ptr [edx+4]
46		pop	ecx
47		push	offset newline2	; __va_args
48		call	_printf
49		pop	ecx
50		mov	[ebp+m2_VPTR], offset item_VTABLE
51		mov	[ebp+m2_VPTR], offset microwve_VTABLE
52		lea	ebx, [ebp+m2_VPTR]
53		push	ebx
54		mov	eax, [ebx]
55		call	dword ptr [eax]
56		pop	ecx
57		push	ebx
58		mov	edx, [ebx]
59		call	dword ptr [edx+4]
60		pop	ecx
61		push	ebx
62		call	item_display
63		pop	ecx
64		push	offset newline3	; __va_args
65		call	_printf
66		pop	ecx
67		pop	ebx
68		mov	esp, ebp
69		pop	ebp
70		retn
_main		endp

item_price	proc near              ; Function item::price()
1		push	ebp
2		mov	ebp, esp
3		push	offset aInItemPrice ;__va_args
4		call	_printf
5		pop	ecx
6		pop	ebp
7		retn
item_price	endp


radio_type	proc near               ;   Function radio::type()
1		push	ebp
2		mov	ebp, esp
3		push	offset aRadioType ; __va_args
4		call	_printf
5		pop	ecx
6		pop	ebp
7		retn
radio_type	endp


computer_price	proc near                ;  Function computer::price()
1		push	ebp
2		mov	ebp, esp
3		push	offset aCompuerPrice ; __va_args
4		call	_printf
5		pop	ecx
6		pop	ebp
7		retn
computer_price	endp


item_type	proc near                ; Function item::type()

1		push	ebp
2		mov	ebp, esp
3		push	offset aInItemType ; __va_args
4		call	_printf
5		pop	ecx
6		pop	ebp
7		retn
item_type	endp


microwve_price	proc near                 ;Function microwave::price()
1		push	ebp
2		mov	ebp, esp
3		push	offset aMicrowavePrice ; __va_args
4		call	_printf
5		pop	ecx
6		pop	ebp
7		retn
microwve_price	endp


microwve_type	proc near                  ;Function microwave::type()
1		push	ebp
2		mov	ebp, esp
3		push	offset aMicrowaveType ;	__va_args
4		call	_printf
5		pop	ecx
6		pop	ebp
7		retn
microwve_type	endp


;                             THE  DATA   SECTION


aInItemDisplay	db 'In item::display()',0Ah,0
newline		db 0Ah,0
newline1	db 0Ah,0
newline2	db 0Ah,0
newline3	db 0Ah,0

aInItemPrice	db 'In item::price()',0Ah,0
aRadioType	db 'radio::type()',0Ah,0
aCompuerPrice	db 'Compuer::Price()',0Ah,0
aInItemType	db 'In item::type()',0Ah,0
aMicrowavePrice	db 'Microwave::Price()',0Ah,0
aMicrowaveType	db 'Microwave::type()',0Ah,0

;                            THE   VTABLES

radio_VTABLE	dd offset item_price
		dd offset radio_type

computer_VTABLE	dd offset computer_price
		dd offset item_type

item_VTABLE	dd offset item_price
		dd offset item_type

microwve_VTABLE	dd offset microwve_price
		dd offset microwve_type
		
		

----------------------------EXPLANATION-----------------------------------------

For those who know nothing about Assembly but still dared to read till here should remember that in Assembly code anything after the semicolon ';' is a Remark. So between code I have inserted remarks so that it is easier to read.

Since we did not include the constructor in our code the compiler automatically inserts it between the main() section. Hence we can see in LINE 5. that m1_VPTR is first initialised to the VTABLE of item. Then in the next LINE m1_VPTR points to the address of the VTABLE of Microwave class. The same process continues till LINE 9.

Now in LINE 13. EAX contains the address pointed to by EBX ie. microwave_VTABLE. In LINE 14. it calls the function which is located at the address of eax ie. microwave::price(). This is the call of a virtual function. Look through the data Section and see how the VTABLES are set.

In LINE 33. there is a straight-forward call to item::display() and this is how a non-virtual function is called. Now compare the calling process of a non-virtual function to that of a virtual function. You will notice that the calling of a virtual function includes the following process:

1. Initialise the VPTR to that of the Base class VTABLE
2. Set the VPTR to the Derived class VTABLE
3. Load the address of the VTABLE in a Register
4. Call the Function located at the address pointed to by the VTABLE

But you require just one step to call a non-virtual function ie. the call instruction followed by the function name.

You should notice that the other microwave object also points to the same VTABLE but has a different VPTR. Each Class has it's own VTABLE and is shared among all VPTR's of the Same class.

In LINE 30. the instruction call dword ptr [edx+4] calls the function located at the (address+4 bytes) pointed by the EDX Register. This means that the 2nd Function in the VTABLE is called.

Notice that in the VTABLE of the radio class the address of item::price() is present. I hope you understand and appreciate the implementation of virtual functions by the compiler.

This is the end of our tutorial. I hope you use Virtual Functions in your programs since it is easier to code your programs using these functions.

Author Information:

Sanchit Karve

http://www.freewebs.com/born2c0de/

born2c0de@hotmail.com

Comments: