[llvm] [RFC] Emit dwarf data for signature-changed or new functions (PR #157349)

[Vladislav Dzhidzhoev via llvm-commits]

dzhidzhoev wrote:

> Tbh, I'm not sure why upstream insists on this being a separate pass. From ad-hoc patterns in e.g. `getArg()` it appears that original idea to handle signature changes in each individual optimization is simpler and more reliable way handle this.

Still, it would probably be nice to have a separate pass (or something in DwarfDebug) to find and somehow mark DISubprograms that do not correspond to IR function signatures anymore. But updating DISubprogram signatures at the place where Functions themselves are modified sounds more reasonable, indeed. A pass that modifies Function signature should know better how to indicate this change in debug info metadata.

https://github.com/llvm/llvm-project/pull/157349