Boost C++ Libraries: Ticket #11323: is_virtual_base_of fails to build when virtual bases and their ancestors define identical virtual functions

John Maddock — Wed, 20 May 2015 17:17:56 GMT

That's nasty, as I assume this could be a common occurrence, but off the top of my head I don't see a way to fix this :(

Will have to think on this.

leandro.gracia.gil@… — Wed, 20 May 2015 17:46:45 GMT

I don't see any either... I'll keep thinking too.

BTW, in case somebody seeing this is wondering why we don't simply make foo a pure virtual in A and avoid the issue altogether, let me share a more meaningful example where this becomes a problem.

#include <iostream>
#include <boost/type_traits.hpp>
// Let's say enabling RTTI is not an option in our platform / project.
#define ASSIGN_CLASS_ID() virtual int GetClassId() { return __COUNTER__; }
struct A { ASSIGN_CLASS_ID(); };
struct B : public virtual A { ASSIGN_CLASS_ID(); };
struct C : public B { ASSIGN_CLASS_ID(); };
int main() {
  std::cout << boost::is_virtual_base_of<B, C>::value << std::endl;
  return 0;
}

douwegelling@… — Wed, 26 Oct 2016 12:39:07 GMT

Using apple LLVM 8.0.0, this fails with pure virtual base class as well:

#include <iostream>
#include <boost/type_traits.hpp>
struct A { virtual void foo() = 0; };
struct B : public virtual A { void foo() override {} };
struct C : public B {};
int main() {
  std::cout << boost::is_virtual_base_of<B, C>::value << std::endl;
  return 0;
}

fails to compile on my machine. This makes boost::serialize completely unusable in combination with virtual inheritance

status changed; resolution set

John Maddock — Sat, 12 May 2018 07:21:51 GMT

status new → closed
resolution → fixed

Well it's only taken 3 years, but thanks to the "detection idiom" this should now be fixed in develop for all current (post C++11) compilers.