nico wrote: That would effectively get us back to the behavior before this PR (i.e. users never see deprecation warnings for headers from libc++ in practice), yes? If so, yes, that sounds like a good temporary workaround. https://github.com/llvm/llvm-project/pull/163356