https://gcc.gnu.org/bugzilla/show_bug.cgi?id=123668
--- Comment #3 from Patrick Palka <ppalka at gcc dot gnu.org> ---
AFAICT before r14-9938 we always rejected this testcase, so this is hardly a
regression IMHO.
Here's a simpler testcase that we always ICEd on (since r9-4045 which
implemented unevaluated lambdas):
struct bad {
template <auto L = [] {}> static void g();
};
int main() {
bad::g();
[] {};
bad::g<[] {}>();
}
