RexTechnology1 wrote: Closing this. #202006 landed and fixes the same assertion from the other direction, by keeping the invalid primary in the instantiation instead of guarding the lookup, so the lookup is never empty. I checked on current main and my reproducer from #202956 no longer crashes, and neither do the variants I could think of that might still leave the lookup empty, including an invalid template parameter in the primary, an out-of-line partial specialization, and the errorful-PCH path. So there is nothing left here to fix. Thanks for the pushback, the questions about why serialization mattered were what made me go and actually measure it, and the answer turned out to be that it did not.
https://github.com/llvm/llvm-project/pull/202958 _______________________________________________ cfe-commits mailing list [email protected] https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
