https://gcc.gnu.org/bugzilla/show_bug.cgi?id=122185
Jakub Jelinek <jakub at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
Status|NEW |ASSIGNED
Assignee|unassigned at gcc dot gnu.org |jakub at gcc dot gnu.org
CC| |jakub at gcc dot gnu.org
--- Comment #25 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
Created attachment 62873
--> https://gcc.gnu.org/bugzilla/attachment.cgi?id=62873&action=edit
gcc16-pr122185.patch
You can't provide a valid testcase for this, because it can only ICE on invalid
ones. In particular the -ffold-simple-inlines code was just assuming standard
conforming std::move, std::forward, std::forward_like, std::addressof,
std::as_const, std::to_underlying and std::__addressof with same behavior as
std::addressof.
The INDIRECT_TYPE_P && INDIRECT_TYPE_P check was actually correct before for
preventing folding (and possibly ICEs or weird stuff) on invalid definitions of
these. Now, std::to_underlying doesn't satisfy that and guess that is why it
was removed, in that case it wants to convert ENUMERAL_TYPE to INTEGRAL_TYPE_P.
So, this patch checks that too.
