jankratochvil added a comment.

In D92223#2422184 <https://reviews.llvm.org/D92223#2422184>, @labath wrote:

> I am not sure if this is the right way to implement this feature. Changing 
> ManualDWARFIndex to provide this additional information is easy enough, but 
> it means that the other index classes will never be able to support this 
> functionality

I wanted to post here rather my agreement with the patch. This looks to me as 
giving up on the high performance index benefits. (But maybe what you say is 
the right sweet spot compromise.)

I rather find missing a feature to cross-check ManualDWARFIndex vs. on-disk 
indexes (lld/LTO produced .debug_names, `gdb-add-index -dwarf-5` produced 
.debug_names and how Apple produces their index). Fedora is probably going to 
have two indexes (`.gdb_index` and `.debug_names`) for each LLVM-built binary 
as the format of GDB `.debug_names` is currently incompatible with LLVM/LLDB 
`.debug_names`. I also expect current `ManualDWARFIndex` is producing different 
index than lld.

> I am also worried about the increase to the manual index size, as this would 
> mean that every occurrence of the DW_TAG_member would be placed in the index, 
> whereas now we only place the one which has a location (which is normally 
> just in a single compile unit).

With `-flto` (even with `-O0`) there is only a single definition of each class.


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D92223/new/

https://reviews.llvm.org/D92223

_______________________________________________
lldb-commits mailing list
lldb-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/lldb-commits

Reply via email to