https://gcc.gnu.org/bugzilla/show_bug.cgi?id=118874
--- Comment #19 from Jason Merrill <jason at gcc dot gnu.org> --- (In reply to Jason Merrill from comment #17) > I wonder about moving much of the return_temp handling to > gimplify_modify_expr... This runs into ordering trouble with cp_gimplify_init_expr, which elides the INIT_EXPR before gimplify_modify_expr gets to see it.