On 24 May 2018 at 02:44, Brian Groose wrote: > I'm moving from g++ 5.3.0 to 8.1.0 at the moment, and bumped into an > unexpected linker failure on AIX. I don't currently have a small > piece of code that reproduces the issue, but I'm working on narrowing > it down. Hopefully someone can make a suggestion without code, > though. I've build gcc 8.1.0 from source with no special options > except --prefix and --enable-languages=c,c++ > > These are the AIX linker errors using g++ 8.1.0: > ld: 0711-317 ERROR: Undefined symbol: typeinfo for std::thread::_State > ld: 0711-317 ERROR: Undefined symbol: .std::thread::_State::~_State() > ld: 0711-317 ERROR: Undefined symbol: > .std::thread::_M_start_thread(std::unique_ptr<std::thread::_State, > std::default_delete<std::thread::_State> >, void (*)()) > ld: 0711-317 ERROR: Undefined symbol: vtable for std::thread::_State > > This worked fine, the same code, on g++ 5.3.0. > > This looks similar to > https://gcc.gnu.org/bugzilla/show_bug.cgi?id=52887 except for > std::thread. It seems to happen when std::thread is used to call a > member function, though a simple test case worked fine. That bug was fixed by adding explicit instantiations to the library, but the undefined symbols you're seeing are not templates, and are defined in libstdc++.so The only way those symbols would not be in libstdc++.so is if this condition is false when building libstdc++.so but true when compiling your own code: #if defined(_GLIBCXX_HAS_GTHREADS) && defined(_GLIBCXX_USE_C99_STDINT_TR1) Alternatively, you're simply not linking to libstdc++ correctly (or at all). What is your linker command that fails?
