https://gcc.gnu.org/bugzilla/show_bug.cgi?id=58909
--- Comment #8 from Jonathan Wakely <redi at gcc dot gnu.org> --- Well Roland's fix doesn't work, and his claims about libstdc++ problems are based on ancient history, so I don't find his arguments very persuasive. I'm happy to document the requirements in the libstdc++ manual, there's no downside to that. We can also avoid weak symbols for some functions that are only referenced from headers, e.g. pthread_call_once. We only need the weak symbols in definitions that end up in libstdc++.so and libstdc++.a, so that those libraries don't unconditionally depend on libphtread. But if user code instantiates the std::call_once function template then they should get a dependency on pthread_call_once, because they're actually using it. Going via the __gthread_call_once weak symbol has no benefit in that case. But the std::condition_variable definitions are in the library, and so we do need to use weak symbols for the cases involved in this PR.
