[PATCH] D11969: Avoid using of DataLayout::getPointerSize() without address space argument in DWARF info generation

[Mehdi AMINI via llvm-commits]

joker.eph added a comment.

Oh I thought you were talking about defining the result of getPointerSize() as being a "base pointer size" or something like that. Got what you meant now.

I discussed with a few people around and my impression so far is that

- The DataLayout does not defined anything about which address space the code resides in
- The code for a single module can only be generated in one address space
- Everything is fine as long as the DataLayout is specified using address space 0 for the code
- MCAsmInfo::PointerSize is intended to describe the size of a pointer to the memory space where the code resides.

If this is correct, it seems to me that on the short term we should to solve your problem:

- rename MCAsmInfo::PointerSize to something more explicit like MCAsmInfo::CodePointerSize
- remove the hook AsmPrinter::getPointerSize() and redirect client to use explicitly MCAsmInfo::getCodePointerSize or DataLayout::getPointerSize

On a longer term, I wonder if this information shouldn't be part of the DataLayout as well. I'm not sure what the implication on the IR level optimizations can be at this point.

I hope this makes sense?


Repository:
  rL LLVM

http://reviews.llvm.org/D11969