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

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

bb...@ma... wrote:

> On Tuesday, Jan 14, 2003, at 05:00 US/Eastern, Just van Rossum wrote:
> > Hm, I just discovered .release() returns None. If I do the
> > release() separately it works just fine. I still don't understand
> > why cls.alloc().init() would result in a retainCount of 2, though.
> > Every time I see a retain() or release() call in Python code I
> > can't help but think there's something terribly wrong :-(
> 
> It has to do with the way the retain/release mechanism works.   In
> any case, you have one -retain for alloc() and one for the assignment
> in python.

Why? The caller of alloc() becomes the owner, and I don't see why the
Python wrapper *wouldn't* become the owner.

> I agree that it would be very, very nice NOT to have to ever deal
> with retain/release/autorelease on the Python side of the PyObjC
> bridge. Unfortunately, it appears to be impossible [AFAICT] to
> automate the management of retain/release/autorelease with 100%
> reliability. Anything that makes assumptions regarding retain counts
> and adjusts the count downwards automatically will lead to situations
> where the app blows up.

I don't see why it's a bad "assumption" that the Python wrapper becomes
the owner of the reference, IMHO it's only natural. I've tried to read
the implementation (class-object.m) but I can't follow it at all because
of the complexity. I seem to see that it attempts to synchronize the
objc retainCount and the Python ob_refcnt. Why is that? I would think
the Python wrapper for an objc object only needs to own one reference
and vice versa for an objc wrapper for a Python object. I'm sure I'm
missing something subtle, but now is as good a time to learn about how
it works as any ;-)

When creating a Python wrapper, there should be a difference between
"wrap an existing ref" and "wrap a new ref".

> Ronald and I discussed this at length some time ago-- I believe it
> was before you had become active on the project.   There is actually
> code to maintain the release counts automatically, it is currently
> commented out.    Check the archives for more information....

I searched, to no avail. Any specific pointers or google keywords?

Just