Re: NSTextView memory leak

[Richard Frith-Macdonald]

On Sunday, October 20, 2002, at 06:11  pm, Alexander Malmberg wrote:

> Caba Conti wrote:
> > The following code is from NSTextView's dealloc method:
> > (CVS version from october 19th)
> >
> > -------------------
> >           /* Balance the RELEASE we sent to us to break the retain 
> > cycle
> >              in initWithFrame: or initWithCoder: (otherwise releasing 
> > the
> >              _textStorage will make our retain count go below zero 
> > ;-) */
> >           RETAIN (self);
> >
> >           /* This releases all the text objects (us included) in
> >            * fall.  */
> >           DESTROY (_textStorage);
> >
> >           /* When the rest of the text network is released, we'll be
> >            * released again and be deallocated for real.  That will
> >            * likely happen during the DESTROY of the _textStorage, or
> >            * later if parts of the text network are maybe retained
> >            * elsewhere (in autorelease pools etc) for slightly longer.
> >            */
> >           return;
> > -------------------
>
> I had a quick look at it, and it seems like the semantic changes in
> NSDecrementExtraRefCountWasZero() (and thus -release/-dealloc;
> 2002-06-01) broke this. If the new behavior is correct, the
> RETAIN(self); can simply be removed.

Well, we went into it in enough detail at the time that I think we
can be pretty certain that the new behavior is 'correct'.

Both the new and old behavior are 'correct' to the letter of the
OpenStaep/MacOS-X documentation, but the new behavior is consistent
with the MacOS-X implementation, and seems to me to be more in line
with what I think they *meant* to say when they wrote the spec.

So yes, I think the RETAIN can be removed.