https://gcc.gnu.org/bugzilla/show_bug.cgi?id=94734
--- Comment #15 from Richard Biener <rguenth at gcc dot gnu.org> --- (In reply to Jakub Jelinek from comment #13) > Even better. Note none of the committed testcases would be handled with this. There's also the issue of store data races (not sure if the notrap handling is correct there, it looks like not) - for auto variables that would not be an issue (but would need to be checked extra). A dominating store to the same location isn't enough to prove validity either since that might be inside a different protected region.
