2026年1月18日 16:45:44 (+09:00) で、Takashi Yano さんが書きました:
> Thanks for pointing this out. > > The comment in the emutls.c explains that. > https://github.com/llvm/llvm-project/blob/main/compiler-rt/lib/builtins/emutls.c#L378-L389 > > Do you think this is a defect we should accept? > The restriction might be acceptable if there is a great gain with using compiler-rt, but I feel linking to cyggcc_s.dll is not so bad. If you are planning to provide a package built with compiler-rt, especially one that contains a DLL, I would say, "Please don't do that." It's very different if someone decides to build their own application with compiler-rt and takes responsibility for doing so. Before providing such a package, the implementation of native TLS in cygwin1.dll is needed, as well as an update to the toolchains to use it (and, I guess, it is impossible to switch with keeping ABI). That changes the fundamental design of the distribution. BTW, the profiler that provided by compiler-rt (unfortunately, it isn't included in your package; available without -rtlib=compiler-rt, it works fine even with libgcc_s) is definitely great; clang built with profile guided optimization reduces 20% of walltime to build LLVM in GitHub Actions. -- Tomohiro Kashiwada (@kikairoya) -- Problem reports: https://cygwin.com/problems.html FAQ: https://cygwin.com/faq/ Documentation: https://cygwin.com/docs.html Unsubscribe info: https://cygwin.com/ml/#unsubscribe-simple
