https://gcc.gnu.org/bugzilla/show_bug.cgi?id=103755
--- Comment #7 from Jonathan Wakely <redi at gcc dot gnu.org> --- (In reply to Dmitry Prokoptsev from comment #6) > That would also work, I suppose (it even outperforms my original approach by > a tiny bit -- 33 ns for v2 vs 36 for my original implementation). > > There are a few build errors in the alternative implementations, however: > is_convertible<locale::facet*, _Facet*>::value is false, as it tries to > downcast the facet -- did you mean is_base_of<>, which would also better > align with the static assert message? Yes, I've already changed it to use that in my local branch. > Also, despite the always_inline attribute of __try_use_facet, my attempts to > build the library with -O0 -ggdb and link my benchmark yielded a bunch of > errors like > > ld: > /home/av/prgs/gcc/debug/x86_64-pc-linux-gnu/libstdc++-v3/src/.libs/libstdc++. > so: undefined reference to `std::num_get<char, > std::istreambuf_iterator<char, std::char_traits<char> > > const* > std::__try_use_facet<std::num_get<char, std::istreambuf_iterator<char, > std::char_traits<char> > > >(std::locale const&)' Yes, that's what happens for the debug libs built with --enable-libstdcxx-debug > Suggest explicitly instantiating __try_use_facet where has_facet() and > use_facet() is instantiated. Yes, I wanted to avoid that :-( > See attached fix for build problems I discovered. Thanks, I'll compare it with the fixes I've already got locally and finish retesting.
