Using Accessors in Init and Dealloc
Introduction
There has been a change in the Cocoa community in the past few years, where the use of accessors in init/dealloc is frowned upon, recommend against, and outright considered to be wrong. It is much better, they say, to directly access instance variables instead. In other words, the recommendation is to write code like this:
- (id)initWithWhatever: (id)whatever
{
if((self = [self init]))
{
_whatever = [whatever retain];
}
return self;
}
- (void)dealloc
{
[_whatever release];
[super dealloc];
}
The alternative is to use accessors, like this:
- (id)initWithWhatever: (id)whatever
{
if((self = [self init]))
{
[self setWhatever: whatever];
}
return self;
}
- (void)dealloc
{
[self setWhatever: nil];
[super dealloc];
}
Pros of Accessors
The pros of using accessors in init/dealloc are pretty much the same as the pros of using them anywhere else. They decouple the code from your implementation, and in particular help you with memory management. How many times have you accidentally written code like this?
- (id)init
{
_ivar = [NSArray arrayWithObjects:...];
return self;
}
I expect the answer will vary quite a bit (I haven't made this particular error in quite a long time) but using an accessor will make sure you don't make this mistake:
- (id)init
{
[self setIvar: [NSArray arrayWithObjects:...]];
return self;
}
Furthermore, you might have ancillary state, like caches, summaries, etc. that need to be set up and torn down when the object value changes. Correct use of accessors can ensure that all of this happens as it needs to when the object is created and destroyed without duplicate code.
Cons of Accessors
The downside to using accessors in this way can be summed up in one short sentence: accessors can have side effects. Sometimes these side effects are undesirable forinit/dealloc.
When writing a setter, it needs to behave correctly if you're going to call it from init/dealloc. This means dealing with a partially constructed object.
Worse, if you ever override a setter in a subclass, you need to write it to handle the case where the superclass is using that setter to initialize or destroy its ivars. For example, this bit of innocuous-looking code is potentially dangerous:
- (void)setSomeObj: (id)obj
{
[anotherObj notifySomething];
[super setSomeObj: obj];
}
- (void)dealloc
{
[anotherObj release];
[super dealloc];
}
If the superclass uses the accessor to destroy someObj, then the override will execute after dealloc has already executed, causing the override to access a dangling reference toanotherObj, probably causing a nice crash.
It's not hard to fix this code to handle the situation gracefully. Simply assign anotherObj = nil after releasing it in dealloc, and everything works again. In general it's not difficult to make sure that your overrides behave properly, but if you're going to use accessors like this then you must remember to, and that is the difficult part.
Key-Value Observing
A topic which is frequently brought up in these discussions is key-value observing, because KVO is a common way for accessors to have side effects. Like other cases, if KVO side effects are triggered when the object is partially initialized or destroyed, and that code isn't written to tolerate such an object, bad things will occur. I personally think that it's mostly a red herring.
The reason it's mostly a red herring is because 99% of the time, KVO is not set up on an object until after it's already fully initialized, and is ceased before an object is destroyed. It isconceivable for KVO to be used such that it activates earlier or terminates later, but it's unlikely in practice.
It's not possible for outside code to do anything with your object until it's fully initialized, unless your initializer itself is passing pointers around to outside objects. And likewise, it's not possible for outside code to keep a KVO reference to your object as it's executing its dealloc method, because it's too late for it to remove it, unless your dealloc triggers something that allows it to do so. Superclass code does execute in these timeframes, but superclass code is extremely unlikely to do anything to cause external objects to observe properties that the superclass itself doesn't even have.
Conclusion
Now you know the pros and cons; should you use accessors in init and dealloc? In my opinion, it can go either way. The advantages aren't generally that great. The downsides are minor. I don't use accessors for the vast majority of my instance variables, but there are certain cases where the advantages become significant because I'm doing something special with an ivar, and in that case I don't hesitate to let the accessors save me some headache.
=================================================
you may invoke methods from inside your dealloc method, though you're wise to be cautious. Pretty much the only methods you should invoke should be "tear down" methods, or methods that help in cleaning up an object before its resources are reclaimed. Some of these cleanup methods include:
- unregistering for notifications via a notification center
- removing yourself as a key-value observer
- other general cleanup methods
Note, however, that in each of these methods, your object will be in an inconsistent state. It may be partially deallocated (some ivars may/will be invalid), and so you should never rely on a specific object state. These methods should only be used to continue deconstructing object state.
This is the fundamental reason why we're discouraged from using property setters (setFoo: methods) indealloc: another object may be registered as an observer, and using the property will trigger a KVO notification, and if the observer is expect the object to have a valid state, they could be out of luck and things can blow up very quickly.
=================================================
When you call [ivar release], nothing happens if the ivar is nil, but if it contains a pointer to an object, you would cause a memory leak by not releasing it.
It's preferable to not use self.ivar = nil in the dealloc method, as the setter method could contain logic that is unnecessary during deallocation, or could even cause tricky bugs.
Regarding @property declarations without an explicit ivar declaration, the "modern runtime" synthesizes the ivar automatically.
=================================================