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

[David Blaikie]

On Fri, Jun 21, 2013 at 10:33 AM, Renato Golin <renato.golin at linaro.org> wrote:
> On 21 June 2013 18:18, David Blaikie <dblaikie at gmail.com> wrote:
>>
>> !llvm.hardref.foo.h.myClass = ...;
>> !llvm.hardref.bar.h.myClass = ...;
>
>
> I like this idea! It's much easier to read than metadata fields, but will it
> need to have mangled names that today, are quoted?

Mangled in some form - using the actual language mangling's probably a
good idea, since that's the uniqueness we require anyway (this would
help/be necessary for local types in inline functions, etc - chances
are we would've had to do that even if we were using the previously
discussed string/mdnode pair scheme)

> Do we still have the notion that named metadata cannot be deleted?

That's the invariant, yes. Named metadata are the roots of the
glorious metadata tree from which the debug info fruit grows. Or
something.

>
> cheers,
> --renato