Re: sending notifications in -dealloc

[Richard Frith-Macdonald]

On Friday, May 31, 2002, at 04:21 PM, Helge Hess wrote:

> Richard Frith-Macdonald wrote:
> > Perhaps NSDecrementExtraRefCountWasZero() and 
> > NSIncrementExtraRefCount() should simply not
> > operate once the extra ref count goes to (unsigned)-1  (with the 
> > former returning NO) ?
>
> Hm, I'm unsure about that. I'll check MacOSX. Even though I don't 
> remember that I have written this somewhere, I could imagine code which 
> essentially does that:
>
> - (void)dealloc {
>   if ([self shouldCache]) {
>     [MyCache addObject:self];
>     return;
>   }
>   [super dealloc];
> }
>
> That is, the reference count really is raised up again and the 
> deallocation is canceled. I'm aware that this is very dangerous 
> (especially with subclassing) ...
>
> I'll check MacOSX ...

I've just done that ... it seems that the behavior is simply that 
NSDecrementExtraRefCountWasZero() does *NOT*
decrement the extra ref count if it was zero.

I've changed the base library to do the same, and documented the 
function and the -release method to describe
how they *actually* behave.

This allows the behavior you describe above, since it is now impossible 
for a retainCount to become zero.