https://gcc.gnu.org/bugzilla/show_bug.cgi?id=86661
--- Comment #3 from Paolo Carlini <paolo.carlini at oracle dot com> --- Note, I'm simply fixing the regression, we will still reject the testcase without ICEing, as we always did. I'm keeping the bug open, in case somebody wants to argue for accepting it even without -fpermissive, I'm not going to do the required analysis any time soon, however.