https://gcc.gnu.org/bugzilla/show_bug.cgi?id=84590
--- Comment #5 from Jason Merrill <jason at gcc dot gnu.org> --- (In reply to Jakub Jelinek from comment #3) > assumes that at least for non-vector cp_fully_fold of a CONSTRUCTOR will > yield again a CONSTRUCTOR as well, but that is generally not the case. > It could (at least in theory) result in error_mark_node, or VIEW_CONVERT_EXPR > around the CONSTRUCTOR: > or a TARGET_EXPR around the CONSTRUCTOR: Hmm, I think cp_fully_fold should probably strip those.