[PATCH] D14687: Macro support in LLVM IR

Amjad Aboud via llvm-commits llvm-commits at lists.llvm.org
Wed Dec 9 06:42:07 PST 2015


aaboud added a comment.

Thanks Paul for the comment.
This is what I was trying to say, I understand the concerns regarding DWARF5, but I am sure that once DWARF5 is official this design will easily be upgraded to support it.

Do you still think I need to change anything with this patch?
Otherwise, can you give me approval for committing?

Thanks,
Amjad

In http://reviews.llvm.org/D14687#305607, @probinson wrote:

> In http://reviews.llvm.org/D14687#304960, @aaboud wrote:
>
> > In http://reviews.llvm.org/D14687#304938, @aprantl wrote:
> >
> > > The only concern I've got with this is whether we shouldn't future-proof this by using the DWARF5-style DW_MACRO_.* constants internally, since the old ones aren't even defined in the DWARF5 standard document any more and thus might be harder to find documentation for.
> >
>
>
> Older versions of the DWARF spec will remain available; dwarfstd.org has them back to 1.1.  I'm not concerned about the constant names becoming unavailable.
>  On the contrary, the DWARF 5 constants aren't officially defined/published yet, and using them is speculative.  I'd prefer to see DW_MACINFO for now, and add DW_MACRO later.
>
> > > Context: In DWARF5 ( http://dwarfstd.org/ShowIssue.php?issue=110722.1 ) the DW_MACINFO_.* constants will be renamed to DW_MACRO_*. This will be fun to implement in dwarfdump because the same numeric value has different names depending on the DWARF version.
>
> > 
>
>
> Yeah... reading the data from the object file will be unambiguous (different section name) but pretty printing it might get kind of tedious.
>
> > Thanks Adrian for the comment.
>
> >  I understand your concern, but I think it is not related to this patch of macro support in LLVM IR.
>
> >  I intentionally, split code into several parts, so the concern can be raised only on the problematic ones.
>
> > 
>
> > Do you think that this patch might not fit DWARF5?
>
>
> Conveying the macro data from the frontend through the IR should work the same way for either, I'd guess. DWARF 5 will have differences in detail of emitting the data into the object file and that will probably have to be done separately anyway.





Repository:
  rL LLVM

http://reviews.llvm.org/D14687





More information about the llvm-commits mailing list