zygoloid wrote: > I have a question about the "sub-object" part. Do we attempt to discover the sub-object past any casting? Good question. I'm going to merge this PR, so I've filed #96737 for this ongoing discussion. https://github.com/llvm/llvm-project/pull/96573