https://gcc.gnu.org/bugzilla/show_bug.cgi?id=69699
--- Comment #4 from Jonathan Wakely <redi at gcc dot gnu.org> --- (In reply to bastian.beischer from comment #3) > Jonathan, maybe so but it should be mentioned on the page nevertheless, > right? Yes, although I plan to deprecate that macro. > My issue is that we are using (as an example): > > #if (defined(__GLIBCXX__) && __GLIBCXX__ > 20150626) || > (defined(_LIBCPP_VERSION) && _LIBCPP_VERSION >= 1101) > static_assert(std::is_trivially_constructible<MyClass>::value, "MyClass > should be trivially constructible."); > #endif > > since "std_is_trivially_constructible" exists only in libstdc++ shipped with > GCC 5.1.0+, but not in libstdc++ shipped with GCC 4.9.3, for example. > > If i understand your comment correctly you are saying that, if GCC 4.9.4 > were released, it would include a version of libstdc++ with __GLIBCXX__ > > today, so it would pass the above check, but still there would't be any > "std_is_trivially_constructible". Exactly. GCC 4.9.4 definitely will be released, and when it is your check will not work. Someone could even use a version of GCC 4.9.3 built from the gcc-4_9-branch in subversion, and your check would fail today. > But how else can I do the check then?! Using __GNUC__ / __GNUC_MINOR__ is > not going to work because we have people using other compilers (clang, Intel > compiler) which use libstdc++ shipped with GCC but do not define __GNUC__ / > __GNUC_MINOR__. We need a macro that actually reflects the version of the library, not one that refers to the almost meaningless release date.
