https://gcc.gnu.org/bugzilla/show_bug.cgi?id=66578
--- Comment #15 from Thomas Koenig <tkoenig at gcc dot gnu.org> --- Is this something which could also have caused wrong-code issues in other circumstances, which we just didn't happen to find until this was exposed by Andre's patch and the test case? In other words, should we backport Mikael's patch?
