[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:

> I wouldn't say I am opposed to that (I don't consider myself a decision maker here), but emitting a (non-standard!) form that gdb doesn't understand doesn't sound that much better than not emitting anything. It might be different if we went to gdb and asked them to implement it, but I don't think anyone's planning to do that.

Yea fair point. I'll have a look at what happens with GDB and report back. Just to clarify though, the form itself is standard, just the use of it for `DW_AT_object_pointer` isn't standard

> I guess my preferred solution would be to emit the standard form for gdb tuning, and use the non-standard thing for lldb (assuming it doesn't break gdb catastrophically), and also start using lldb tuning for Google. I don't know how hard the last part would be, but I can try to find out.

Ah that seems like a good middle-ground! And could be done in separate steps (i.e., not emit anything under gdb tuning until google moves adopts `-glldb` in order to mitigate the size regression).


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