Re: building dbuskit fails with llvm/clang

[Nicola Pero]

> I think the correct solution is to allow something like this:
> 
> @class<NSObject> GSStackTrace;
> 
> This would forward-declare the class, and also advertise that
> it conforms to the NSObject protocol.

It's an interesting suggestion, and has good merit. :-)

I would personally not like it that much because:

 * well, I like a simple language and I'm worried about feature-creep. :-)

 * @class only exists to allow you to deal with circular references
   when declaring classes.  So, really, every @class should later
   be followed by a corresponding @interface.

 * I would agree that compilers very lax behaviour in that respect (where
   '@class A' makes anything of class A practically equivalent to 'id') is
   counter-productive as it doesn't enforce or recommend that at all.
   If you message an object of a class which is only declared using @class,
   you should get a warning.  Both GCC and clang don't warn about this.
   GCC also allows class methods to the class again with no warnings;
   clang doesn't (in fact it produces a hard error, which is the "problem"
   reported above).  I like the clang behaviour a lot, and I submitted a bug
   report to GCC requesting the warning (maybe not an error) for GCC too.
   I'm not entirely sure why clang makes a difference between class and
   instance methods.  I would like the warning on instance methods too! :-)

Having said that, the idea is creative and nice.

Thanks