[PATCH] D35923: Fix PR32332 - PCM nondeterminism with builtins caused by global module index

[Adrian Prantl via Phabricator via cfe-commits]

aprantl added a comment.

After thinking about this some more, I think it should be safe to use the global module index as a negative cache like in this example. The conditions under which a module is rebuilt should not affect the contents, and if the module were out-of-date it would have been rebuilt anyway.
I will see if I can track down why the builtin macros make it into the module's identifier table. We already know that they are being marked incorrectly as FromAST, but I do not yet understand why.


https://reviews.llvm.org/D35923