[llvm] [DebugInfo] Handle additional types of stores in assignment tracking (PR #129070)

[Orlando Cazalet-Hyams via llvm-commits]

https://github.com/OCHyams commented:

Tyvm for handling this.

Code changes SGTM, with one inline question.

More broadly, it's unfortunate to remove coverage for the entire variable. Part of me wonders whether it's better to go the other way, and state it's a def for the whole variable. This is obviously liable to introduce incorrect locations, but if we compare it to dbg.declare semantics (which this stuff sort-of replaces) I don't think it'd be a regression there. I'm not sure how convinced I am by my own argument; I'm not a massive fan of either option, but we do need to do something. cc @jmorse in case he wants to chime in.

https://github.com/llvm/llvm-project/pull/129070