Published on May 05, 2015

Imagine that there is a power in the universe that said to Apple: “Hey, convince developers to do a silly thing and you’ll get $1 for the every demonstration of silliness**”.

I believe it was a hard task, but Apple managed to find that thing:

self = [super init];
if (self) { // <- this is silly
}
return self;

**Don’t want to offend you, I did this silly thing for years

TL;DR;

Check the value of self in constructor only if you use ivars.

Why?

Documentation of NSObject’s -init method says:

Return Value

An initialized object, or nil if an object could not be created for some reason that would not result in an exception.

Which is completely OK, but why should we check if self is nil or not?

To get the answer let’s take a look at a few possible variations of a constructor:

• super.init returns nil, programmer uses properties (send messages)

self = [super init];
self.foo = @"Foo";
return self;

• super.init returns nil, programmer uses ivars

self = [super init];
_foo = @"Foo";
return self;

I skipped options when super.init returns a valid object, because they’re considered correct and valid in any situation

Nil object and messaging

In a first case an assignment expression just converts into a function call:

// note: 'self' is nil
self.foo = @"Foo";
// converts into
objc_msgSend(self, @selector(setFoo:), @"Foo");

What happens when you send a message to a Nil object?

Well, nothing happens.

Nil object and pointers

The second constructor also has a different final form:

// note: 'self' is nil
_foo = @"Foo";
// actually means
self->_foo = @"Foo";

OK, another question: what happens when you dereference null pointer?

Behaviour in this case is undefined, but for iOS/OS X platforms it’s usually a crash.

Conclusion

Based on the statements above we can conclude that we don’t need to check self in every constructor, but only when ivars are used.

As far as I see some developers do this check, but they don’t know why.

The reason, imho, is obvious: ObjC is an old language and tons of documentation/examples were written when properties were not introduced yet and programmers had to use ivars.

After doing this for a years it’s just turned into a habit.

Nowadays we have properties, that allow us to write clean and robust code, so let’s do it and, please, check the value of self in constructor only if you use ivars.

Happy and sane hacking!