[clang] [Modules] Don't fail when an unused textual header is missing. (PR #138227)

[Ilya Biryukov via cfe-commits]

ilya-biryukov wrote:

> In general, though, I share the same concern as @vsapsai about leaving this patch reverted for too long.

I have a concrete proposal to move forward,  #141792. I have checked that the failure we've seen in our infrastructure goes away with both changes applied. Given that multiple modules owning the same header seem to be a rare case outside of Google, I hope this also won't cause any breakages and the change acceptable (it's only changing the logic that kicks in when we do have multiple modules for the same header).

I really appreciate the steps forward here and would be happy to hop on a video call if necessary. I would still suggest to discuss #141792 first and if that stalls, we can schedule a call. Does that work?
 
Unfortunately, I will be OOO until next Monday and won't be able to respond here.
If folks feel there is urgency in relanding this or are happy with merging #141792 without discussion, feel free to do so. (We can definitely live with a local revert for a week). The only major objection that I would raise would be to a path forward that breaks the test case I've added above without a migration path that we can employ.

https://github.com/llvm/llvm-project/pull/138227