[PATCH] D84115: [Debuginfo][Codegen] (2/7) Support for DW_OP_implicit_pointer for named and unnamed variables (second strategy).

[David Blaikie via Phabricator via llvm-commits]

dblaikie added a comment.

In D84115#2184545 <https://reviews.llvm.org/D84115#2184545>, @jmorse wrote:

> I sense this is the bulk of the work; and it looks good. I've posted on llvm-dev asking whether the creation of artificial variables could instead be deferred til DWARF emission. That'll avoid anything having to deal with a new form of DBG_VALUE, and us having to think about their lifetimes. Plus, there's greater scope for de-duplication, as I understand it this patch will generate a new artificial variable for each dbg.value, for each level of indirection.
>
> On that topic -- do we need to give the artificial variable a name? I'm no DWARF expert, but can we just not attach a DW_AT_name to the DIE?

I haven't looked closely at the code - but "do it later" sounds like the right theory to me. I believe the absence of a name is probably OK - there are unnamed types, for instance, so unnamed variables seem fine too. But some testing with real debuggers would be useful.


CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D84115/new/

https://reviews.llvm.org/D84115