https://gcc.gnu.org/bugzilla/show_bug.cgi?id=85553
--- Comment #4 from Paolo Carlini <paolo.carlini at oracle dot com> --- I see. Then, my humble opinion (we got Jason in CC, anyway) is that if we are still hoping to have this fixed in 8.1.0 we should just go ahead with something rather straightforward like your tweak, otherwise we should probably investigate the constexpr.c behavior a little more, because I don't really see why going from integer_zero_node to NULLPTR_TYPE_P is "better" than going from nullptr_node to NULLPTR_TYPE_P, if you see what I mean.
