https://gcc.gnu.org/bugzilla/show_bug.cgi?id=10837
--- Comment #25 from Josh Haberman <jhaberman at gmail dot com> --- (In reply to Lukas Grätz from comment #24) > (In reply to Josh Haberman from comment #23) > > (In reply to Lukas Grätz from comment #22) > > > I think the status should be changed into RESOLVED FIXED now. After PR > > > 119483 and PR 121159 with GCC 16+ and 15.2+, tail call optimization is > > > finally possible for noreturn functions. Just use the musttail attribute > > > introduced by PR 83324. > > > > That works as long as the function signature of the noreturn function > > precisely matches the function signature of the caller. In many cases, > > you would need to add a wrapper around the noreturn function to make > > them match. > > Sorry, but if I understand you correctly, this has nothing to do with > noreturn; it is a limitation of tail call optimization in general. > > So this does not affect the present issue. If I am wrong, please provide an > example to explain. I was mistaken; I thought that GCC implemented the same rules as Clang for musttail. In Clang, musttail is rejected unless the parameters for the caller and callee match, even if a tail call would technically be possible. But after experimenting on Godbolt, I see that GCC does not have these requirements. Note how this example compiles on GCC but is rejected for Clang: https://godbolt.org/z/Yc589hMM3 As a result, I now agree with you that `musttail` addresses the complaint in this bug. While it is odd and unexpected that `noreturn` will inhibit tail calls, I agree that `musttail` appears to override this behavior, providing an escape hatch for users who want tail calls on `noreturn` functions. It is somewhat unfortunate that this requires the attribute to be at every call site, instead of on the `noreturn` function itself, but at least it is possible.
