https://gcc.gnu.org/bugzilla/show_bug.cgi?id=85553
--- Comment #3 from Jakub Jelinek <jakub at gcc dot gnu.org> --- (In reply to Paolo Carlini from comment #2) > Thanks Jakub for looking into this. For sure, the whole fold and convert > machinery should not be necessary for something this simple, but, to be > honest, isn't immediately obvious to me why it does the wrong thing in this > case. Because it isn't folded into INTEGER_CST, but NOP_EXPR with the decltype(nullptr) type wrapping zero INTEGER_CST and constexpr.c doesn't consider that a valid reduced constant expression.
