[llvm] [llvm][DebugInfo] Encode DW_AT_object_pointer on method declarations with DW_FORM_implicit_const (PR #124790)

[Michael Buch via llvm-commits]

Michael137 wrote:

> re: producing the GCC-thing when gdb-tuning. I'd be hesitant to endorse that until someone has a specific gdb behavior/use case for this - it's possible GCC is producing something GDB doesn't need/use (there is precedent for this - view numbering, so far as I understand, is not implemented in gdb, for instance).
> 
> So, yeah, I'd be OK with the non-standard attr+form combo when tuning for lldb and not changing non-lldb tuning behavior.

> If we wanted to be more cautious, we could use an LLVM extension attribute for this - though I'd guess it'd produce the same-ish GDB behavior (it'd spit some warning about not knowing what the attribute is - but there'd be less risk that way, since GDB wouldn't be trying to interpret the index as an offset, which could collide, but is unlikely to - because it's always going to be 0)

Happy to go either way. Though it'd be nice not to have to maintain an LLVM extension which will be made obsolete by one of the ongoing DWARF proposals re. encoding `DW_AT_object_pointer` (be it using index or DIE relative offsets). We can also wait for the DWARF committee discussions to settle since there is no rush with this. It'll be useful for codebases adopting explicit object parameters, but LLDB's support is blocked on a couple of other things, not just this patch.

https://github.com/llvm/llvm-project/pull/124790