Re: KVC weird

[Wolfgang Lux]

David Wetzel wrote:

> Hi,
>
> I am trying to find whats breaking GSWeb on current -base.
>
> The code in GSWApplication+Defaults.m does KVC on a class, not an  
> instance.
>
> [...]
>
> So my wild guess is that [sig numberOfArguments] is wrong.

0 points. The problem really is in -methodSignatureForSelector:,  
which returns nil rather than the class method's signature. This is  
due to a recent optimization by David C., which broke the  
implementation of this method for class methods.

The new fast path of the method uses class_respondsToSelector, but  
this method checks only for an *instance* method of the given class.  
So you get false negatives when asking a class object for one of its  
class methods, as happens in your case, except if there is an  
instance method with the same name. On the other hand, you get false  
positives when asking a class object for a class method which it does  
not implement and if an instance method with the same name exists.  
This will later lead to a crash in this method -- at least with the  
old runtime -- when dereferencing the result of GSGetMethod.

> Any ideas how to fix that? (other than changing the macros at  
> GSWApplication+Defaults.m)

Until somebody fixes this bug in svn, just revert the patch from  
r31231 in -base (svn merge -c-31231 should do the job).

Wolfgang