[LLVMdev] Proposal: type uniquing of debug info for LTO

[Eric Christopher]

On Fri, Jun 21, 2013 at 11:58 AM, Robinson, Paul
<Paul_Robinson at playstation.sony.com> wrote:
>> From: llvmdev-bounces at cs.uiuc.edu [mailto:llvmdev-bounces at cs.uiuc.edu]
>> On Thu, Jun 20, 2013 at 10:52 PM, Manman Ren <mren at apple.com> wrote:
>> > ---------------------
>> > Option b) I am going to expand David's description with more details,
>> correct me if I am wrong.
>>
>> I'll let David continue this part of the thread with you.
>>
>> >
>> > I personally prefer option b over a since it is much cleaner.
>> >
>>
>> Option b is more general and should probably work. I suggested a
>> originally because I know it would work and the impact is limited to
>> debug info. Also it doesn't involve touching the generic metadata
>> interface or code and could be done quickly and incrementally. b may
>> also require further changes to the debug info infrastructure to deal
>> with lookup, etc so may end up being as much churn as a. b is more
>> useful if we think other users of metadata may want/need similar
>> functionality in the future.
>
> I can't say I've followed all the details but would this be useful
> in moving toward support of .debug_types at all?  There's a
> superficial resemblance in conjuring up hashes of types referenced
> from elsewhere, and I can't help wondering if the resemblance is
> any deeper than that.  (I'm not insisting that it has to be, at
> worst it shouldn't get in the way if we can anticipate what would
> be required for .debug_types to work.)
>

Nope. Completely orthogonal.

I've almost got dwarf4 type hashing ready for commit anyhow.

-eric