[PATCH] D71839: [Dsymutil][Debuginfo][NFC] Reland: Refactor dsymutil to separate DWARF optimizing part. #2.

David Blaikie via Phabricator via llvm-commits llvm-commits at lists.llvm.org
Thu Jan 2 11:24:37 PST 2020


dblaikie added a comment.

In D71839#1801654 <https://reviews.llvm.org/D71839#1801654>, @JDevlieghere wrote:

> In D71839#1799149 <https://reviews.llvm.org/D71839#1799149>, @dblaikie wrote:
>
> > In D71839#1795483 <https://reviews.llvm.org/D71839#1795483>, @avl wrote:
> >
> > > > Not sure if this is better/worse off as a subdirectory of lib/CodeGen, rather than as a top level library on its own (lib/DWARFOptimizer). Figured I'd mention it, but mostly leave it up to the other dsymutil folks to make the choices here.
> > >
> > > My idea was that DWARFOptimizer does generation task. F.e. AsmPrinter - generates asm and it is under CodeGen.  DWARFOptimizer generates optimized DWARF. That is why I put it under CodeGen.
> > >  Though if it is better to put it on top level - lib/DWARFOptimizer - I am OK to do this.
> >
> >
> > @aprantl @JDevlieghere @probinson any preferences here? If I've got to make the call I'll probably pick lib/DWARFOptimizer or lib/DWARFLinker (probably the latter)
>
>
> My personal preference goes to `lib/DWARFLinker`.


Let's go with that then - thanks for weighing in!

>> (totally random aside: I Imagine we could have a mode where LLVM's debug info is aware it will be linked with a DWARF aware linker, so it doesn't need to use relocations in some cases (even with DWARFv5, the string relocations (in str_offsets) are still pretty significant & would be nice to skip (as split DWARF manages to skip))
> 
> That sounds like an interesting idea indeed!

(realizing I'm not sure who might implement this - it's not a problem for the Apple/MachO debug info distribution model, which is always DWARF aware and I think already avoids/skips using relocations - and it's not going to be a super high priority for Google, I think, since we use Split DWARF a lot (though switching to DWARFv5 accelerator tables which separate strings from the tables (unlike (gnu_)pubnames/pubtypes) it might be valuable to be able to skip those relocations))


CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D71839/new/

https://reviews.llvm.org/D71839





More information about the llvm-commits mailing list