`depth_first_visit_impl`: `finish_edge` not invoked when it should on MSVC

[jrsala]

Lines 94-104 in depth_first_search.hpp:

    template < typename E, typename G, typename Vis >
    void call_finish_edge(Vis& vis, E e, const G& g)
    { // Only call if method exists
#if ((defined(__GNUC__) && (__GNUC__ > 4)               \
         || ((__GNUC__ == 4) && (__GNUC_MINOR__ >= 9))) \
    || defined(__clang__)                               \
    || (defined(__INTEL_COMPILER) && (__INTEL_COMPILER >= 1200)))
        do_call_finish_edge< has_member_function_finish_edge< Vis, void,
            boost::mpl::vector< E, const G& > >::value >::call_finish_edge(vis,
            e, g);
#else
        do_call_finish_edge< has_member_function_finish_edge< Vis,
            void >::value >::call_finish_edge(vis, e, g);
#endif
    }

On MSVC, the section after the #else will not call Vis::finish_edge even when it is available and has the correct signature because has_member_function_finish_edge<Vis, void>::value is false. I think this is because instead of checking that Vis::finish_edge exists and has return type void as was seemingly intended, it in facts checks that it has the exact signature void(). So on MSVC, even when providing a visitor type with a correct void finish_edge(E, const G&), it is never called.

I don't know how to fix this because I don't know why it was determined that some compilers should be treated differently here.

I circumvented the bug in my code by adding a dummy void finish_edge() {} member function in my visitor type next to the actual void finish_edge(E, const G&) function just so the member function detection would yield true.

Alternatively this could be a bug in boost/tti/has_member_function.hpp but I guess that's less likely.

[marvin-the-mathematician]

I am having this problem with Visual Studio 2022 and Boost 1.82.0, but worse :-( Since I am using a EventVisitorList (https://www.boost.org/doc/libs/1_82_0/libs/graph/doc/EventVisitorList.html) with make_dfs_visitor I have no alternative, but to ditch that, write my own visitor and use the workaround outlined above, which I don't really want to do. A rather annoying bit of meta-programming that is getting in the way of making a simple call! Why not just call the finish_edge member function and let the compiler decide whether it actually exists or not?

[jeremy-murphy]

There is a lot of very old code in this library. We're not supporting 'old' compilers any more, so probably the best thing to do is remove the #if-else altogether, or at least try to push it up into a helper utility outside the library. If you make a PR to fix it then I will see to fast-track approving it.