nikic wrote: Yes, reverting the change did recover the regression. Did you test exactly the change that was landed? It's also possible that it interacted with something that landed in the meantime... https://github.com/llvm/llvm-project/pull/126088