Re: Core Data and retain count

FROM : Bill
DATE : Mon Mar 03 04:50:05 2008

On Mar 2, 2008, at 5:35 PM, Ben Trumbull wrote:
>> My question is, why would changing a property value cause another
>> property to have its retain count increase?

>
> No idea.  Why don't you run it in gdb and break on the -retain 
> method and get some stack traces ?
> This works best if the class you're debugging (in this case the 
> value window controller) has a custom retain/release method that you 
> can break on instead of NSObject's etc.  If it doesn't, you can 
> trivially make one by adding a category on it and putting -retain/-
> release/-autorelease methods on it that simply call super.


Good idea.  Here's the stack trace:
#0    0x000029aa in -[NSWindowController(MainWindow) retain] at 
MainWindowController.m:35
#1    0x92dfa378 in -[NSManagedObject(_NSInternalMethods) 
_newPropertiesForRetainedTypes:andCopiedTypes:preserveFaults:]
#2    0x92dfa250 in -[NSManagedObject(_NSInternalMethods) _allProperties]
#3    0x92dfa185 in -[NSManagedObjectContext(_NSInternalChangeProcessing) 
_establishEventSnapshotsForObject:]
#4    0x92dfd516 in _PFFastMOCObjectWillChange
#5    0x92dfb4e3 in -[NSManagedObject willChangeValueForKey:]
#6    0x00021a6c in -[Window setOpenAtLaunch:] at Window.m:117

Not sure what to make of it, nor why setting the property openAtLaunch 
increments the retain count for the window controller property.


>> So by the time the user wants to delete a managed object from the 
>> list, the window controller could have a high
>> value and is never released!

>
> This sounds like a hypothetical concern as opposed to an empirically 
> observed leak.  Each retainer is obliged to balance itself out 
> eventually.  That's not your problem.  Generally speaking, the 
> retain count is an implementation dependent value and not really 
> meaningful API.


I made the statement because I have an NSLog statement in the window 
controller dealloc method.  It never gets executed unless the retain 
count is 0.

Thanks,
Bill