https://gcc.gnu.org/bugzilla/show_bug.cgi?id=113835
--- Comment #10 from Jason Merrill <jason at gcc dot gnu.org> --- (In reply to Jason Merrill from comment #9) > Removing the maybe_constant_init from expand_default_init speeds up the > testcase by more than 10x, but also regresses a bunch of testcases, so it's > not that simple. Yeah, the speedup was from never trying to do constant evaluation at all; properly doing it in store_init_value instead provides no improvement.