[PATCH] D82838: Parse section ranges when verifying DWARF so we can exclude addresses that should have been stripped from DWARF.

[David Blaikie via Phabricator via llvm-commits]

dblaikie added a comment.

I think maybe this is sort of orthogonal to 46453... maybe not, but kind of.

Seems like we should filter out known-tombstoned ranges (the only ones we can know for sure are the new -1/-2 tombstones - all the others have ambiguities). Then we should maybe flag maybe-tombstones with a little "eh, maybe?". Then we should warn for anything left that's even partially outside the .text range (this patch), then we should warn for overlaps/etc on the remaining ones?

But as @jhenderson said, maybe those first ones come later & we use the .text range to determine which things to look at for overlap first, then add new verifier checks for "things outside .text that aren't clearly tombstoned" knowing that some of those are expected limitations of (at least gold's) previous tombstoning strategies.

(I'd sort of like to avoid actually looking at the object's executable sections - but I can't really fault the strategy & even if we added all the other verifier checks/warnings/etc, it'd still be super reasonable to warn about ranges that are otherwise totally valid, but extend beyond/are entirely outside the actual executable .text)


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D82838/new/

https://reviews.llvm.org/D82838