[PATCH] D71732: Move the sysroot attribute from DIModule to DICompileUnit

Adrian Prantl via Phabricator via llvm-commits llvm-commits at lists.llvm.org
Tue Jan 14 12:30:53 PST 2020


aprantl added a comment.

In D71732#1820401 <https://reviews.llvm.org/D71732#1820401>, @dblaikie wrote:

> In D71732#1813229 <https://reviews.llvm.org/D71732#1813229>, @aprantl wrote:
>
> > > Some mysterious CFI failures I haven't looked into yet, but were pretty clearly blamed on this change & went away with a revert.
> >
> > Because the flag was previously not (de)serialized and now shows up in LTO builds where it was dropped before?
>
>
> Maybe, but don't see how that'd turn up anything interesting in CFI behavior - again, no one's turning this flag on...
>
> > Well I guess I can serialize false for now without preventing this from getting fixed.
> > 
> > Any opinions on this patch otherwise?
>
> Seems alright - but I'd like to understand it a bit better. Looks like this'll add the attribute to every (.o - so skeleton with Split DWARF, or the normal/unit with non-split) CU, where previously it was only in DIModules which were only emitted with -gmodules? What's it used for? Could it be put in the full unit, rather than the skeleton unit? Could it be only emitted when tuning for lldb?


Yes and yes. I'll update the patch. The attribute helps LLDB to construct a Clang compiler invocation that can be used to import Clang modules from source so you have access to the entire module (not just what was used and thus put into the debug info) in the expression evaluator.


CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D71732/new/

https://reviews.llvm.org/D71732





More information about the llvm-commits mailing list