nikic wrote: For some reason, this seems to have made the "can't start new thread" problem on i686 much worse for us, going from a few dozen Clang tests to 2-3 thousand LLVM tests. https://github.com/llvm/llvm-project/pull/186712