https://gcc.gnu.org/bugzilla/show_bug.cgi?id=91707
--- Comment #6 from Martin Sebor <msebor at gcc dot gnu.org> --- The nowarning idea in comment #5 is close to what I've been moving toward with the new __builtin_warning intrinsic. Calls to it can be injected into the IL by any pass, even the front-ends or user-defined code, without worrying too much about false positives. Those not eliminated by DCE will be processed just before expansion and before issuing the warnings themselves. The processing I have implemented in the prototype so far only involves avoiding duplicate warnings, but it could include the checks for sanitization or anything else.