Re: [Pyobjc-dev] Headache over hacking an XML editor

[Just v. R. <ju...@le...>]

Ronald Oussoren wrote:

> On Monday, Jan 13, 2003, at 21:27 Europe/Amsterdam, Just van Rossum 
> wrote:
>
> > Btw. there's in interesting thing going on in this app: if the
> > PythonItem class is a subclass of NSObject everything works fine,
> > but if it isn't (a pure Python class), the app crashes as soon as
> > you click inside the tree widget. Maybe there's some refcount
> > problem for Python objects passed to objc?
> 
> This is a feature(*) of NSOutlineView. The outline view does not call
> retain on the objects you return from methods like 
> outlineView:child:ofItem:, but it does keep around references to
> those objects. This is documented in 'Using an Outline View Data
> Source' in Apple's developer documents.

Aha, I'll try to find that document.

> The reason your code is crashing when PythonItem is not subclassed
> from NSObject and runs correctly when it is twofold.
> 
> First of all 'normal' python objects are brigded to Objective-C using
> autoreleased proxy objects. Because the outline-view doesn't call
> retain these will be released 'soon'. And because the outline-view
> does actually use the pointer later your program crashes...
> 
> If you subclass from NSObject your object is bridged as itself (more
> or less). And if you keep around a reference to the objects you
> return these objects won't go away before the outline-view uses the
> pointer it stashed away. Which brings me to your second problem: You
> code is leaking memory.
> 
> PythonItem.__new__ should return cls.alloc().init().release().

Ha, when I make that change the app crashes right away ;-) But the
reason is now more clear: I didn't keep any child references around at
all, so the instances NSObject subclass would be alive just long enough
(due to the autorelease pool magic I assume) and the non-NSObject
subclass instances weren't. Or they indeed leaked.

I changed the code to keep references to the children and everything
works just fine as a non-NSObject subclass. But now it *does* crash when
subclassing NSObject if I add .release() to the __new__ method, so
something isn't right, most likely in my understanding of things...

> (*) Actually I think this is not a feature but a bug. I've filed a
> bug for this at Apple, but did not receive a response for this yet. I
> hope Apple changes this, especially because adding the proper calls
> to retain and release should be backward compatible (although I have
> no idea what effect this would have on the performance of
> NSOutlineView)

It's at least a lousy feature :-(

Just