https://gcc.gnu.org/bugzilla/show_bug.cgi?id=89692
--- Comment #7 from Richard Biener <rguenth at gcc dot gnu.org> --- Hmm. I think we should prune the variant list down to what is actually used but to fix the GC issue I was raising make sure the pruned variants appear as main variants (thus NULL their TYPE_NEXT_VARIANT and set TYPE_MAIN_VARIANT to themselves).