[llvm] [DebugInfo] Fix incorrect debug attribution for inlined allocas (PR #144345)

[David Blaikie via llvm-commits]

dwblaikie wrote:

Mostly happy to defer this to @jmorse and @SLTozer - but I am a bit confused.

On the one hand, I'm not sure any allocas should have source locations? They all effectively get lumped into the start of the function and merged anyway, and it's in the prelude, and it's not interesting (unlikely to fail - ultimately just adding some offset to the stack/frame pointer)?

On the other hand - it is accurate to attribute the alloca to the inlined function - what's going wrong if we do that? Should that thing be going wrong? Should the stack protection code not have a location rather than inheriting one from an arbitrary alloca? 

https://github.com/llvm/llvm-project/pull/144345