The docs say that I should not release any modeled property in -dealloc. For me, that feels like violating the big memory management rules. I see a big retain in the header and no -release, because Core Data seems to do it at any other time.
Is it because Core Data may drop the value of a property dynamically, at any time when needed? And what's Core Data doing when dropping an managed object? If there's no -dealloc, then how and when are the properties getting freed up?
Core Data has NSManagedObjects as their base object, they are managed by the system and you do not have to do any memory managment on them...well unless u declare your own properties in a class that are not defined in the object model...
The tricky bit of managed object memory comes when the objects are turned into faults. A fault is a ghost of the object. It exists and responds to messages but it doesn't have it's attributes populated. For example, you can count the number of objects in a to-many relationship even if the objects themselves are faults.
Faults are used to maintain the object graph in memory without the overhead of the actual data. The object is still technically alive so its dealloc has never been called.
When the context turns an object into a fault, the object releases and nils it's attributes in memory. It does this in willTurnIntoFault and didTurnIntoFault. If you need to do some special releasing you should override those methods (usually the latter.)
Faulting makes managed objects rely on different methods than other classes for intialization and clean up. You intialize in awakeFromInsert or awakeFromFetch and you dealloc in willTurnIntoFault and didTurnIntoFault.
It's important to follow the rules for managed objects because the context might keep thousands of faulted objects in memory. If you create a custom attribute but do not release it when the object becomes a fault, then you could easily tie up a large amount of memory accidentally.