原创 How to implement pure virtual functions

Last year, I wrote a number of articles on how virtual functions behave in C++ and how you can obtain similar behaviour in C. In short, you can emulate a polymorphic C++ class (a class with at least one virtual function) as a C structure that has an additional member commonly called a vptr (VEE-pointer).[1] The vptr points to a table of function pointers called a vtbl (VEE-table).

Last month, I introduced the concept of pure virtual functions in C++.[2] This month, I'll show how compilers typically implement pure virtual functions by showing how you can achieve similar behaviour in C.

As in my prior articles, my sample classes represent an assortment of two-dimensional geometric shapes such as circle, rectangle, and triangle, all derived from a common base class called shape. The C++ definition for the shape base class looks in part like:

class shape {
public:
shape(color o, color f); // constructor
virtual double area() const = 0;
virtual double perimeter() const = 0;
~~~
private:<
color outline, fill;
};

The characters = 0 appearing at the end of the declarations for area and perimeter indicates that they are pure virtual functions. They need not be defined. As I asked at the end of my previous column, if shape's area and perimeter functions are undefined, what happens when you try to call them?

Pure virtual function virtually impossible
C++ makes calling a pure virtual function nearly impossible. A class with at least one pure virtual function is called an abstract class. C++ doesn't let you create an abstract class object except as the base class subobject of some derived class object. For example, these declarations won't compile:

shape s; // compile error
shape x[10]; // compile error

You can declare a pointer to a shape, as in:

shape *ps; // OK

However, the compiler will reject a new-expression that tries to construct an abstract base class object, as in:

ps = new shape; // compiler error

Again, any class with at least one pure virtual function is an abstract class. If you can't create any such objects, then you don't have to worry that you might call a pure virtual function. Unfortunately, C++ compilers have trouble detecting all possible calls to pure virtual functions, such as when the constructor of an abstract class indirectly calls a pure virtual function of the same class. Thus, most compilers provide another line of defence. More on this shortly.

The definition for our circle class derived from shape looks like:

class circle: public shape {
public:
circle(double r, color o, color f); // constructor
virtual double area() const;
virtual double perimeter() const;
~~~
private:
double radius;
};

If you don't declare the area and perimeter functions in circle, then circle will inherit them as pure virtual functions, and circle will be an abstract class. You won't be able to create any circle objects. However, the class definition above does declare area and perimeter, so they're not pure virtual and circle isn't an abstract class. Now you're on the hook to define circle's area and perimeter functions.

Inasmuch as circle is not an abstract class, you can create circle objects, as in:

circle c (3);

or:

shape *ps = new circle (4);

If the compiler could detect and prevent all calls to pure virtual functions, implementing pure virtual functions would be trivial—simply initialise the vtbl entry for each pure virtual function as a null pointer. In C, the code to define the vtbl for the shape base class would look something like:

typedef struct shape_vtbl shape_vtbl;
struct shape_vtbl {
double (*area)(shape const *);
double (*perimeter)(shape const *);
};

shape_vtbl the_shape_vtbl = {
0, // 0 converted to null pointer
0 // ditto
};

A second line of defence
Unfortunately, compilers may fail to catch some pure virtual function calls. Such failures are rare, but nevertheless possible. As a second line of defence, if a pure virtual function has no definition, most compilers generate a definition for that function that makes a stink if executed. The C code to do this for the shape class might look like:

double shape_area(shape const *s) {
pure_virtual_alert();
return 0;
}

double shape_perimeter(shape const *s) {
pure_virtual_alert();
return 0;
}

~~~

shape_vtbl the_shape_vtbl = {
shape_area,
shape_perimeter
};

On a desktop system, the pure_virtual_alert function might simply write a message to the system's stderr stream or display a message box, and then terminate the program. For an embedded system, you would have to tailor the response to be something more appropriate.

Endnotes:
[1] Saks, Dan, "Implementing virtual functions in C," Eetindia.co.in, August 14, 2012. www.embedded.com/4391967.
[2] Saks, Dan, "The concept of pure virtual functions," Eetindia.co.in, December 20, 2012. www.embedded.com/4403313.