[Boost-users] Re: Serialization, multiple inheritance, bad cast at smart_cast.hpp:202

From: Robert Ramey (ramey_at_[hidden])
Date: 2005-03-10 21:19:15

Zoltan Szatmary wrote:
> Dear all
>
> I use the serialization library of boost, and I am very satisfied with
> it except one annoying issue. It turns out, that in many cases (not
> always!!!) when there is a multiple inheritance structure in the class
> hierarchy I am trying to serialize I get a std::bad_cast exception at
> line 202 in smart_cast.hpp. The corresponding boost code is:
>
>             template<class U>
>             static T cast(U * u){
>                 T tmp = dynamic_cast<T>(u);
>                 #ifndef NDEBUG
>                     if ( tmp == 0 ) throw std::bad_cast();
>                 #endif
>                 return tmp;
>             }
>
> I have debugged it, and I could not really find out what the problem
> was. For example I have a class hierarchy similar to this:
>
> class A
> class B : public A
> class C : public virtual B
> class D : public virtual B
> class E : public C, public D
>
> so, a classic diamond inheritance which the serialization lib is
> supposed to be able to tackle with. Now, if I have a piece of code
> like this
>
> boost::serialization::xml_oarchive oa(somestream);
> A *a = new E();
> oa << BOOST_SERIALIZATION_NVP(a);
>
> it throws exception at the aforementioned line. Examining the template
> substitutions, template class T is substituted with const B* and
> template class U is substituted with const A. So, the program should
> do a downcast on a const A* to const B* which is ought to be possible.

Your example does indicate whether A has any pure virtual functions or not.
The dynamic down cast will only function if A is polymorphic - that is has
at least one virtual function.  Soooo, I would expect this work if A has at
least one virtual function and I would expect it to fail if has no virtual
function.  Double check this and get back to us.

> The funny thing is, that the error does not always appear! I could put
> together a simple example code demonstrating this effect, but I am not
> sure that attachments to this mail get through the list server, so
> anybody kind enough to take a look at this problem can ask me to send
> him/her the example code. I have tested the example code both with
> Intel C++ compiler v8.1 and G++, so I concluded this may not be a
> compiler
> error but rather some boost flaw.
>
> Any help is warmly welcome and thanked in advance!
>
> Cheers
>
> Zoltan