[cfe-commits] r171899 - in /cfe/trunk: lib/Sema/SemaDeclAttr.cpp test/Sema/attr-availability.c

[Douglas Gregor]

On Jan 9, 2013, at 8:01 AM, Rafael Espíndola <rafael.espindola at gmail.com> wrote:

>>> +  NamedDecl *ND = dyn_cast<NamedDecl>(D);
>>> +  if (!ND) {
>>> +    S.Diag(Attr.getLoc(), diag::warn_attribute_ignored) << Attr.getName();
>>> +    return;
>>> +  }
>> 
>> Rafael,
>> 
>> I'm concerned by attribute abuse.  We just warn on too many weird
>> cases instead of throwing an error.  I can understand "legacy code"
>> reasons, but this particular case never compiled, so there's no reason
>> to be forgiving.
>> 
>> Maybe I am misunderstanding something...
> 
> I would be more than happy to make this an error, but
> 
> * We are very forgiving with the availability attribute for some
> reason. We warn if it is declared to be available after 10.5 and 10.7
> for example.
> * This is used mostly (only?) by apple, so I it is probably best if
> they decide whether this should be an error or warning.

Unfortunately, we're forced into making this a warning for historical reasons. On Darwin, there are a pile of availability macros (eg, in <Availability.h>) that currently expand to unavailable or deprecated attributes. Those macros should switch to use the availability attribute, but that means we can't be too strict or we break existing code.