What is virtual inheritance?

Answer 1

Virtual inheritance is used in multiple inheritance hierarchies whenever two or more classes inherit from the same base class. For example, consider the following minimal example:

struct A {int x;};

struct B : A {};

struct C : A {};

struct D : B, C {};

Here, A is a common base class of both B and C, thus B::A and C::A are independent instances of A. However, D inherits from both B and C thus indirectly inherits both B::A and C::A. This introduces an ambiguity when implicitly referring to A from D, as in the following example:

D d; d.x = 42; // error: ambiguous

When the compiler sees this it won't know whether we are referring to B::A.x or C::A.x, thus we must be explicit when referring to A via D:

d.B::x = 42; // ok

The problem with this is that we still have two separate instances of x so unless we are extremely careful about which x we are referring to we can easily end up with two different values for x with no way of knowing which was the correct value with respect to D.

What we really need is for B and C to share the same instance of A and we achieve that through virtual inheritance:

struct A {int x;};

struct B : virtual A {};

struct C : virtual A {};

struct D : B, C {};

D d;

d.x = 42; // ok

Note that d.x can be referred to as d::A.x, d::B::A.x or d::C::A.x because they all refer to the same instance of x, thus the ambiguity is eliminated. In effect, the most-derived class in the hierarchy (D) is now a direct descendant of A rather than an indirect descendant of both B::A and C::A.

Note also that the virtual keyword has no effect on independent instances of either B or C because they then become the most-derived classes within their own hierarchy and therefore behave exactly as they would had the virtual keyword been omitted. The virtual keyword only comes into play when we derive from a class with a virtual base.

It is also possible to have a base class that is both virtual and non-virtual in the same hierarchy:

struct A {int x;};

struct B : virtual A {};

struct C : virtual A {};

struct D: A {}

struct E : B, C, D {};

Here, A is a virtual base of both B and C, as before, but not of D. Thus E::A implicitly refers to the virtual A while E::D::A must be explicitly referred to as E::D::A because it is a separate instance. Such classes are best avoided, but if we don't have access to the source code we sometimes don't have a choice in the matter.

However, if a base class holds no data it really doesn't matter whether it is declared virtual or not because a class that has no data consumes just one byte regardless of how many instances there are.