Re: Observing with GC

FROM : Chris Suter
DATE : Tue Feb 26 23:15:47 2008

On 27/02/2008, at 4:59 AM, Quincey Morris wrote:

>
> On Feb 26, 2008, at 02:42, Chris Suter wrote:
>

>> The observation info uses a weak reference and it will have been 
>> zeroed before finalize is called so neither of the above are a 
>> problem. Even so, I'm not sure it's a good idea to call 
>> removeObserver: from the finalize method.

>
> A weak reference to what? If you mean to the observed object, that 
> brings us full circle. My original question was whether the 
> references to observed objects are weak, and the answer I got was no.


The reference to the observer is a weak reference. The reference to 
the observed object isn't and that's where the leak comes from.

>> Knowing this, you might wonder if you have to call removeObserver 
>> at all in the GC case. I just ran some tests and it looks like you 
>> do otherwise you'll get a memory leak which is a shame because it 
>> would be easy to clean up automatically.

>
> Do you know what's leaking? Are we talking about an observation info 
> object that's sitting there with a zeroed reference to the observee? 
> If so, I agree it's a shame, but I'm not sure how easy it would be 
> for the observation code to detect without a lot of scanning.


Yes, it's the observation info that gets leaked. All observation info 
appears to get stored in a global dictionary: 
NSKeyValueObservationInfoPerObject and it's an entry in that that gets 
leaked. It should be easy to make it automatically clear up; use a 
NSMapTable rather than a dictionary.

- Chris