Boost C++ Libraries: Ticket #4003: Error in destroy.hpp when methods "T()" and "operator~()" are defined.

attachment set

Sat, 13 Mar 2010 18:18:02 GMT

attachment → foo.h

minimized example header file to be wrapped

attachment set

Sat, 13 Mar 2010 18:19:48 GMT

attachment → wrap_foo.cpp

boost.python wrapping code for minimal example

attachment set

Sat, 13 Mar 2010 18:19:50 GMT

attachment → wrap_foo.2.cpp

boost.python wrapping code for minimal example

attachment set

Sat, 13 Mar 2010 18:20:42 GMT

attachment → inadequate_patch.patch

(imperfect) patch for destroy.hpp

Sat, 13 Mar 2010 18:22:29 GMT

Sorry about the two copies of wrap_foo.cpp. In my impatience I hit the "submit" button twice.

Steven Watanabe — Mon, 15 Mar 2010 21:57:39 GMT

It seems to work if I change the code to

p->~T()

However, this is slightly different, since it will invoke a virtual destructor, where the original will not. The only other alternative I can think of it to mangle the name so that there is no chance of anyone using the same name. (i.e. template<class T_boost_python_unique_id_82623817364646282>)

owner changed

Dave Abrahams — Tue, 16 Mar 2010 18:25:40 GMT

owner changed from Dave Abrahams to Steven Watanabe

Isn't using a virtual d'tor OK here?

What about a simple typedef either inside the function body or in the body of the class? Does VC still look up the wrong name?

owner changed

Steven Watanabe — Tue, 16 Mar 2010 19:35:15 GMT

owner changed from Steven Watanabe to Dave Abrahams

I would assume that a virtual destructor is okay, but I didn't write the code in question. A typedef doesn't seem to help. The first name after the -> always seems to be looked up in the context of T first. I even tried this:

template<class T>
struct Destroyer {
    typedef T Type;
    static void destroy(T* t) {
        t->Destroyer<Type>::Type::~Type();
    }
};
struct S {
    void Type();
    void Destroyer();
};
int main() {
    S t;
    Destroyer<S>::destroy(&t);
}

and it still fails.

anonymous — Wed, 17 Mar 2010 20:40:46 GMT

Replying to steven_watanabe:

I would assume that a virtual destructor is okay, but I didn't write the code in question.

Feel free to check in your fix, preferably after having run the tests once :-)

status changed; resolution set

Steven Watanabe — Fri, 26 Mar 2010 19:04:42 GMT

status new → closed
resolution → fixed

(In [60868]) Handle the destructor call in a way that keeps msvc happy. Fixes #4003