https://gcc.gnu.org/bugzilla/show_bug.cgi?id=123351
--- Comment #2 from maic <gnu.ojxq8 at dralias dot com> --- A fix is in the thread https://gcc.gnu.org/pipermail/gcc-patches/2026-January/704827.html So I presume the real regression was in commit 404c787e2bfe8cae666b075ed903990ea452220e, which is in GCC 12.1. Though, I am not sure if code exists to trigger non-determinism end-to-end without looking at internal intermediate state. So it may be fine to not backport this to GCC 12 and 13, but no strong opinion, just wanted to clarify that the regression may have happened earlier.
