[cfe-commits] r72715 - /cfe/trunk/lib/AST/ASTContext.cpp

[Eli Friedman]

On Tue, Jun 2, 2009 at 9:25 AM, Fariborz Jahanian<fjahanian at apple.com> wrote:
> Thank you for cleaning this up. I have a question below.
>
> On Jun 1, 2009, at 10:28 PM, Eli Friedman wrote:
>
>>
>> +  // Strip off objc_gc attributes off the top level so they can be
>> merged.
>> +  // This is a complete mess, but the attribute itself doesn't make much
>> sense.
>
> These attributes provide necessary API to Next't runtime for GC. If you have
> questions
> please let us know. I'd be glad to provide the answers. It also helps us
> polishing up
> our documentation :).

I don't think I really have any questions, it's just that the way that
the qualifier is allowed at any layer is simply weird.  The one thing
I've considered that might make it a bit cleaner is always pulling the
attribute out to the outermost array/pointer type, and always pushing
it back in on a dereference, which would make the whole thing actually
work consistently.  Currently, there's an issue that for the two
lines:

__weak id* x;
id* __weak y;

The generated code for *x and *y is different.

-Eli