[PATCH] IR: Move the complex address expression field out of DIVariable and into an extra argument of the dbg.declare/dbg.value intrinsics.

[David Blaikie]

>>! In D4919#6, @aprantl wrote:
> This updated version of the patch passes all tests.
> Still outstanding:
> - Auto-upgrading old-style intrinsics

Do we need this? We have the auto-upgrader stripping debug info already, don't we? Does it not strip the intrinsics? I assume we should just strip them if we aren't already.

> - DIExpression tagging (for pretty-dumping)
>   My plan is to add an offset to the ComplexAddrKind enum values (DW_TAG_user_low?) so DIExpressions can be recognized.

I'm still really on the fence about the whole annotated metadata support... it really is just getting lucky about certain integer prefixes. Is there something less likely to collide that we could/should use?

If we do continue to decide just to use integers, yeah, I suppose we could add a number to that enum. I'd probably generate a random number for the task (& write some clear comments about why that is). DW_TAG_user_low would just be confusing two mostly unrelated things for no particular benefit (I think it'd be confusing - it's like doing arithmetic with differently dimensioned quantities).

- David

http://reviews.llvm.org/D4919