[llvm] Aggregate errors from llvm-dwarfutil --verify (PR #79648)

[Kevin Frei via llvm-commits]

kevinfrei wrote:

> General idea I'm more or less OK with (though, like, clang/compilers in general don't have this sort of feature - so I'm not sure it's especially strongly motivated for DWARF either) - though it starts to look more like compiler diagnostics, with diagnostic groups & at that point I wonder whether it should use syntax that's more similar to the way compiler diagnostics and diagnostic groups are rendered - and maybe even share some implementation (well, layering might be difficult there, llvm can't depend on clang - but maybe some of the diagnostics stuff got refactored out to be used by flang and so might be in some usable place... )

Errors here are a little different than Clang/LLVM errors, IMO: There's no reasonable way to narrow down your output, and to make them useful, they contain a dramatic amount of detail that makes aggregating through simple scripts much harder. In general, the diagnostics output in the this tool also feels pretty inconsistent (I wanted to aggregate before changing output, just for cleanliness of the work itself).

I do feel like aggregation is quite useful here, and *much* easier to do properly than in Clang/LLVM. This is (mostly) a very "one-to-one" usage scenario: Something has produced DWARF data in some binary somewhere. Validate that it's okay. (The output currently still requires you to know which problems are catastrophic, which can be ignore safely, etc...)

https://github.com/llvm/llvm-project/pull/79648