views:

434

answers:

6

I know that discussions about coding styles tend to end in disaster and endless flame wars, but that’s not what I want to reach. During the last decade I mainly saw two different coding styles for dealloc methods in Objective-C. The first and most common one was to place dealloc at the bottom of the file. This is also the style Apple uses in the Xcode default templates. The logic behind this seems to be that dealloc is invoked when the end of the object approaches, so the end of the file seems to be a nice metaphor.

On the other hand a couple of people tend to put dealloc directly below the @synthesize directives. This has two major disadvantages in my opinion:

  1. The top of the file gets cluttered with boring code.
  2. It’s harder to find the essential parts in your class, you have to scroll down.

The huge advantage in my opinion is that you have a direct visual connection between properties and the corresponding release message.

The other thing is niling already released variables. While I don’t think that this is necessary, especially in object context where the whole variable gets desctructed after dealloc ends, I tend to also nil the variables. I’m used to do this for variables in function scope, so I’m just consistent with my coding style.

This is how most of my classes look like:

@implementation Bar

@synthesize foo;

- (void)dealloc
{
  [foo release], foo = nil;

  [super dealloc];
}

// Initializers and other methods…

I already mentioned a couple of pros and cons. What do you think about this topic? What is the coding style you use in dealloc and why? Are there other pros and cons I forgot to mention?

I don’t want to start a flame war here. I just want to know what style you use and if you have specific reasons for this or if this doesn’t matter for you in the end.

+3  A: 

I put my dealloc at the top, just under the @synthesize directives. It is a little clunky, and boring code, but oh-so-important code, so it gets top billing. Also, being able to compare between the properties and the -releases is vital.

Ben Gottlieb
+12  A: 

I like to put the dealloc implementation right below the initializers. That way, when I add a new instance variable, I remember to release it right after I init it.

Also, I find it really helpful to use the #pragma mark directive to make it easier to browse the file. So I "group" the init and dealloc methods together under a heading called "initializers". When browsing the file, having those headings makes it much easier to find what you're looking for without being distracted by the dealloc method.

It might be boring code, but man is it important.

Alex
A: 
- (id)init{
   self = [super init];
   if( self ) {
      someVar = [[NSMutableArray alloc] init];
      // something like the following shouldn't be released:
      someString = [NSString stringWithFormat:@"ANumber: %d",10];
   }
   return self;

   - (void)dealloc{
       [someVar release]; someVar = nil;
       [super dealloc];
   }

that's the way i do it :)

Antwan van Houdt
I'm pretty sure stringWithFormat returns an autoreleased object. You need to retain it in init and release in dealloc. Also, your braces don't balance...
walkytalky
read please, something like the following shouldn't be released as I state in my comment in the code, What i meant with that is that the object I am using there is autoreleased.
Antwan van Houdt
+7  A: 

Don't set your ivar to nil in dealloc if you don't have a specific reason to. It serves no purpose and at best masks programmer errors that you would do better finding out about than hiding.

+1: By not setting the instance variables to `nil` and using good ol' `NSZombieEnabled`, I've caught a few logical errors over the years I never would have found otherwise.
Alex
+2  A: 

I put it at the bottom. This allows me to simply hit end and go to it when I add something that needs deallocating. I also don't want it around my property synthesizers, because this is deceptive. Not everything I dealloc necessarily has a synthesized accessor attached to it. Heck, it isn't even necessarily all in the initializer. If I try to use a shortcut that way, I'm likely to mess it up.

Chuck
Agreed. I refer to all the member variables in the `@interface` when I'm writing `dealloc`.
Tom Dalling
+3  A: 

My order:

  1. Class methods (+load, +initialize, +sharedFoo, others)
  2. Initializers
  3. dealloc
  4. finalize
  5. Syntheses, @dynamic directives, and custom accessors
  6. Protocol conformance methods, grouped by protocol (usually with #pragma mark directives)
  7. Notification handler methods (usually declared in a class extension up at the top)
  8. Other methods (usually declared in a class extension up at the top)

Within the dealloc method:

  • Don't use accessor messages, implicit (property accesses) or explicit. Any impure custom accessor may not be safe to call on a partially-deallocated object. (The same goes for initializers.)
  • Don't set ivars to nil. The object is partially deallocated; why are you still sending it messages? (If you aren't, then nothing is looking at the values of the ivars.)
  • (If it were somehow appropriate to set ivars to nil) Don't abuse the comma operator. An expression like [foo release], foo = nil mixes the types (first void from the message expression, then id from the assignment expression). These are separate statements; write them as such.
  •  
    [super dealloc] is always last and always has an empty line above it, emphasizing its presence.

Of course, I also have “Treat Warnings as Errors” turned on, so if I forget [super dealloc], I break my build.

Peter Hosey
These are a lot of good tips for improving an Objective-C coding style. Thanks for sharing all of them.I know that it’s pointless to set the pointers to `nil` after releasing them in `dealloc`. I just do it, because I’m used to do it in a method context.I also use a blank line above `[super dealloc]` to emphasize the importance.An interesting thing is that you synthesize your properties in between of methods. Do you have a specific reason for this?
Rafael
It's a historical artifact: Before properties, all accessors were custom, and I didn't always declare them in the header. Thus, the definition would be the first the compiler heard of them, so I couldn't use them any earlier or the compiler would rightly complain about it. Thus, these undeclared accessors had to be defined before anything that might use them. Now I synthesize almost all of my accessors, but my choice of placement lives on.
Peter Hosey