[llvm] [DebugInfo] getMergedLocation: match scopes based on their location (PR #132286)

[Stephen Tozer via llvm-commits]

SLTozer wrote:

> Let's consider two chains of Scopes:

Yes, I was thinking that this would be possible, and asymmetry isn't something that I believe can otherwise happen in this function (e.g. `getMergedLocation(L1, L2)` is always the same as `getMergedLocation(L2, L1)`). It _sounds_ like a bad property to have, but I don't know if it actually causes any problems anywhere - it's more "odd" than anything, but from my (C++) perspective the textual inclusion behaviour is odd as well. I _think_ you could create a stack of scopes like that in C/C++, but at the moment, I don't think either of the asymmetric results would be more incorrect than a line 0 (since either one is an accurate depiction of some layer of the textual-inclusion stack), or indeed any other source location, and I'm hard pressed to think of a practical scenario where this would cause more problems for debugging or profiling than omitting the information outright.

https://github.com/llvm/llvm-project/pull/132286