TITLE: Pure virtuals, multiple inheritance, and dominance

PROBLEM: kmr@netcom.com (KMR)

I could not find an answer to my problem  of abstract base class
and pure virtual functions in ARM, any help will be appreciated.

Consider the following:

class A { 
  public:
    virtual void foo1() = 0;
};

class C : virtual public A {};

class B : virtual public A {
  public :
     virtual void foo1() {cout<<"I'm B::foo1\n";}
};

class D : public B, public C {};

When compiling the above on SUN 2.1 Cfront, the compiler fails 
informing me that D is an abstract class and thus cannot be "newed"


RESPONSE: carney@rocky.rational (Mike Carney)

I think that your real problem here is that class C does not have an
implementation of foo1(). If a class is an abstract class (i.e. any of
its virtual members is declared line foo1() and foo2() in class A
above), then all of its derived classes *must* have implementations of
those virtual member functions that were declared abstract. 


RESPONSE: maxtal@extro.ucc.su.OZ.AU (John MAX Skaller), 1 Oct 92

	No, only the most derived class D must have implementations.
The function B::foo1() dominates A::foo1(), therefore all virtual
calls

	d->foo1();
	c->foo1();
	b->foo1();
	a->foo1();

bind to B::foo1(). Give A::foo1() a dummy body instead of the pure
specifier to test this out. See: A::foo1() cannot
be virtually called.

	It follows D is concrete. It has no holes. Whether
A::foo1 is pure or not should make no difference (since
it cannot be called).

	There is a section in the ARM dealing with the sibling
call mechanism. It is the basis of c++ "mixin" style.
Its operation depends on the name dominance rule (also discussed),
which is claimed to be the basis of polymorphism in C++.
It is unfortunate the ARM example did not use pure virtuals
in the base class in the example.