https://gcc.gnu.org/bugzilla/show_bug.cgi?id=109177
--- Comment #2 from Alex Coplan <acoplan at gcc dot gnu.org> --- Part of the problem here seems to be that dg-error doesn't care how many times the same diagnostic is emitted, so it doesn't seem able to detect this failure mode. Otherwise, I expect this would have shown up in failures of the existing g++.dg/ext/attr-unavailable*.C tests.