Alternative to https://reviews.llvm.org/D56587
The advantages of this approach as I see it would be
- does not require the infrastructure update to allow a debug expression reference a base_type DIE (especially the emit part of this seems hairy)
- even if we have full DW5 ops there is still the problem of when we need to deref. This pseudo-op has that baked in (which is perhaps not such a good idea after all?). Actually I think that when and where to DW_OP_deref (and DW_OP_deref_size as is required for big-endian) is the hardest part. I am trying to raise that issue on llvm-dev https://lists.llvm.org/pipermail/llvm-dev/2019-January/129522.html and hopefully it will result in some useful insight.