https://gcc.gnu.org/bugzilla/show_bug.cgi?id=128428
--- Comment #2 from Alexandre Oliva <aoliva at gcc dot gnu.org> --- It would be easy enough to fix the testcase, but then we'd be accepting much lower-quality error messages for regular users, rather than the much saner and desirable type names we get when libstdc++ headers aren't considered system headers. That probably wouldn't be a regression, but the result wouldn't be the nice and informative error message that the patch that introduced these tests intended. We could hopefully do better.
