https://gcc.gnu.org/bugzilla/show_bug.cgi?id=89585
--- Comment #12 from Andrew Pinski <pinskia at gcc dot gnu.org> --- (In reply to Harald van Dijk from comment #11) > > That doesn't mean it's acceptable to change how that's handled in a minor > release. The whole concept of a release series doesn't make sense if you > take that route. Take a look at how signed integer overflow was handled, for > instance. We all know that's invalid and has always been invalid, but the > change to enable -fstrict-overflow at -O2 was saved for the next major > release and clearly documented in the GCC 4.2 Changes page > <https://gcc.gnu.org/gcc-4.2/changes.html>. > > A release series should mean that it is safe for distros to upgrade within > the release series. Right now, it's not. This is wrong too since some of what -fstrict-overflow did was enabled at -O1 (and -O2) before hand; just the option was added for GCC 4.2.0 and only enabled at -O2 and above.