https://gcc.gnu.org/bugzilla/show_bug.cgi?id=79937
--- Comment #15 from Jakub Jelinek <jakub at gcc dot gnu.org> --- Created attachment 43578 --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=43578&action=edit gcc8-pr79937.patch Actually, seems TREE_HAS_CONSTRUCTOR is set on a CONSTRUCTOR only by finish_compound_literal. So if those are something we don't want to walk into and everything else we do, then this patch might be it (passed make check-c++-all, but otherwise untested).