https://gcc.gnu.org/bugzilla/show_bug.cgi?id=89571
--- Comment #4 from Paolo Carlini <paolo.carlini at oracle dot com> --- Seriously, I spent quite a bit of time today on this issue and, all in all, barring much more invasive changes, I think that not setting *spec_p to error_mark_node when maybe_instantiate_noexcept returns false in process_subob_fn makes for better error recovery. After all, in quite a few places we don't even check the return value of maybe_instantiate_noexcept... If nobody comes up with a better solution, in a few days I will send a message to the mailing list.
