[compiler-rt] [llvm] [TSan, SanitizerBinaryMetadata] Analyze the capture status for `alloca` rather than arbitrary `Addr` (PR #132756)

[Marco Elver via llvm-commits]

================
@@ -88,4 +88,35 @@ entry:
 ; CHECK: __tsan_write
 ; CHECK: ret void
 
----------------
melver wrote:

I understood what's happening. It's an unfortunate consequence of the poor Github PR experience that LLVM project enforces :-/ -- normally I'd prefer a PR with multiple commits and not multiple PRs, but the fact we're forced to "Squash and Merge" makes this hard. Not your fault. Once the first PR is committed, you can rebase this one and the problem should be "fixed".

With regards to the 2nd commit in this PR, it'd still be good to have an IR test in addition to the e2e test you already added.

https://github.com/llvm/llvm-project/pull/132756