Gevorg Voskanyan:

Hi folks,

I've came across a problem with addressof() when the argument is of a class type with a template conversion operator. A minimal test case reproducing the issue is attached. Using boost 1.38, the example fails to compile under gcc 4.2.2, with the following error:

../../../../vendor/boost/boost_1_38_0/boost/utility/addressof.hpp: In function 'T* boost::addressof(T&) [with T = convertible]': addressof_conversion.cpp:17: instantiated from here boost/boost_1_38_0/boost/utility/addressof.hpp:42: error: ISO C++ says that these are ambiguous, even though the worst conversion for the first is better than the worst conversion for the second: ../../../../vendor/boost/boost_1_38_0/boost/utility/addressof.hpp:26: note: candidate 1: static T* boost::detail::addressof_impl<T>::f(T&, long int) [with T = convertible] ../../../../vendor/boost/boost_1_38_0/boost/utility/addressof.hpp:32: note: candidate 2: static T* boost::detail::addressof_impl<T>::f(T*, int) [with T = convertible]

AFAICT gcc is right complaining here, according to 13.3.3

The same example compiles OK with boost 1.33.1, and the changeset that causes the difference seems to be this one: https://svn.boost.org/trac/boost/changeset/44705

Can this be considered a regression?

Yes, this is a regression. Unfortunately, my attempts so far to fix it have been alternately thwarted by MSVC's ability to convert function pointers to void* and by the inclination of your convertible class to convert to anything. I'll keep trying.