[clang] [Clang] disallow # operators in attribute argument lists (PR #147308)

[Aaron Ballman via cfe-commits]

AaronBallman wrote:

> @erichkeane @AaronBallman Thanks — it seems the changes will need to be revisited based on further guidance from WG21. Would it make sense to close this PR for now?

I could see WG21 going either way, but there's a case to be made that this is UB by omission and therefore we're fine to accept in both C and C++ (with a pedantic diagnostic about the extension). I think I'd prefer we accept with a diagnostic over rejecting outright, and if WG21 decides to make it ill-formed we can determine whether we want to turn it into an error at that time or whether we want to keep it as a conforming extension. It's not critical either way, we have no attributes currently that can make use of this, so it's not a strongly held opinion. WDYT?

https://github.com/llvm/llvm-project/pull/147308