jimingham wrote: It's been a long time since I've looked at this code, but the change to return a 0 value rather than an error seems consistent at least. Was there no way to test this? https://github.com/llvm/llvm-project/pull/93548