[Lldb-commits] [PATCH] D94077: Support unscoped enumeration members in the expression evaluator.

[Pavel Labath via Phabricator via lldb-commits]

labath added a comment.

In D94077#2489064 <https://reviews.llvm.org/D94077#2489064>, @werat wrote:

> Thanks for the explanation, this makes sense. I've checked the mailing list archives and it seems there was already a discussion about the enumerators in the `.debug_names` index back in 2018 -- http://lists.dwarfstd.org/pipermail/dwarf-discuss-dwarfstd.org/2018-April/004443.html. You were the one to bring it up and the consensus was that the enumerators should go into the index too.
>
> But it seems this was never actually implemented, the latest version of the toolchain doesn't add `DW_TAG_enumerator` entries to the `.debug_names` index. Is there a reason for that or it just slipped through the cracks? Should I bring it up again on the mailing list or we can assume the consensus is still the same and this should be just implemented?

As I recall, I myself started having doubts about the correctness of that approach, as I started discovering other interesting named objects that were not in the index:

- using bar = foo; // another name for foo
- using foo = bar::foo; // same name but in a different namespace
- using namespace foo; // another name for _everything_ in foo

I started being unsure where should the line be drawn, and (coupled with insufficient time to investigate this in depth) decided to stick to what the spec says. I am also unsure whether that discussion would be sufficiently authoritative to decide on the matter, since this is pretty clearly in contradiction with the wording of the spec.  And there's still the question of how much extra space will this feature consume (there's no way to "home" these declarations, so they index entries would have to be emitted for each compile unit that uses them)...

> Regarding the changes in the manual index. Do you think there's a better way to implement enumerators lookup?

Maybe? It depends on what kinds of contexts we'd want these enumerators to be recognised. If we're fine with seeing just the enumerators visible in the current scope, then we could parse the current compile unit, looking for enum declarations -- I believe "using namespace" declarations are handled in this way. If we'd want them to be visible globally, then things get trickier. Though there's always the possibility to say that the feature is not worth the cost it brings...

> Assuming `.debug_names` and `apple_names` are updated to include the enumerators, `manual_index` should be updated as well to match the behaviour.

Yep, that much we can agree on. :)


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D94077/new/

https://reviews.llvm.org/D94077