https://gcc.gnu.org/bugzilla/show_bug.cgi?id=66483
--- Comment #15 from Thomas Preud'homme <thopre01 at gcc dot gnu.org> --- (In reply to Thomas Preud'homme from comment #13) > Thanks doko. I meant to do it today but lacked the time. As to the variation > issue I've been experiencing, I'll create a new PR when I isolate the source > of variation. It was just a red herring. Having no compiler in the path lead to HAVE_COMDAT_GROUP not to be defined which change the result of can_alias_cdtor and thus the number of times a pass is called. Making HAVE_COMDAT_GROUP be true, I can see there is still a difference in code generation but I guess it's likely a similar thing. No issue here. Best regards.