https://gcc.gnu.org/bugzilla/show_bug.cgi?id=121159
--- Comment #10 from Lukas Grätz <lukas.gra...@tu-darmstadt.de> --- (In reply to Andi Kleen from comment #8) > ><source>:3:30: error: cannot tail-call: volatile function type > > I wonder how that happened. It looks like a bug. volatile functions should > be rejected elsewhere and there is no volatile in the test case. > > Maybe some tree bit encoding gets confused? > > Should reopen the bug just for this. I guess there are two solutions: First, to allow no-return function calls with musttail attribute. Although it is currently not supported and tail-calls for no-return functions are not done in general because of debugging, it should be OK if the attribute is used explicitly. There was another issue referenced in 10837: https://gcc.gnu.org/pipermail/gcc-bugs/2000-October/040396.html I tested this (by explicitly given the exit() signature without noreturn attribute): extern void exit(int); void main() { [[gnu::musttail]] return exit(0); } gcc exitsibcall.c -o exitsibcall ./exitsibcall I checked it. At least on my x86-64, I could not observe the reported garbage exit status (it exits with status 0 or whatever we pass to exit()). I have not tested it on i686, but with -m32 I get an error "target is not able to optimize the call into a sibling call". The second solution is to keep it as-is and just fix the error message to something more meaningful like "cannot tail-call noreturn functions". But I see no reason in doing so, unless above issue manifests in another configuration. Have I missed something?
