Best practices for overriding isEqual: and hash

Dave_Dribin_26825 0
Viewed 94221

How do you properly override isEqual: in Objective-C? The "catch" seems to be that if two objects are equal (as determined by the isEqual: method), they must have the same hash value.

The Introspection section of the Cocoa Fundamentals Guide does have an example on how to override isEqual:, copied as follows, for a class named MyWidget:

- (BOOL)isEqual:(id)other {
    if (other == self)
        return YES;
    if (!other || ![other isKindOfClass:[self class]])
        return NO;
    return [self isEqualToWidget:other];
}

- (BOOL)isEqualToWidget:(MyWidget *)aWidget {
    if (self == aWidget)
        return YES;
    if (![(id)[self name] isEqual:[aWidget name]])
        return NO;
    if (![[self data] isEqualToData:[aWidget data]])
        return NO;
    return YES;
}

It checks pointer equality, then class equality, and finally compares the objects using isEqualToWidget:, which only checks the name and data properties. What the example doesn't show is how to override hash.

Let's assume there are other properties that do not affect equality, say age. Shouldn't the hash method be overridden such that only name and data affect the hash? And if so, how would you do that? Just add the hashes of name and data? For example:

- (NSUInteger)hash {
    NSUInteger hash = 0;
    hash += [[self name] hash];
    hash += [[self data] hash];
    return hash;
}

Is that sufficient? Is there a better technique? What if you have primitives, like int? Convert them to NSNumber to get their hash? Or structs like NSRect?

(Brain fart: Originally wrote "bitwise OR" them together with |=. Meant add.)

objective-c equality
16 Answers

Start with

 NSUInteger prime = 31;
 NSUInteger result = 1;

Then for every primitive you do

 result = prime * result + var

For objects you use 0 for nil and otherwise their hashcode.

 result = prime * result + [var hash];

For booleans you use two different values

 result = prime * result + ((var)?1231:1237);

Explanation and Attribution

This is not tcurdt's work, and comments were asking for more explanation, so I believe an edit for attribution is fair.

This algorithm was popularized in the book "Effective Java", and the relevant chapter can currently be found online here. That book popularized the algorithm, which is now a default in a number of Java applications (including Eclipse). It derived, however, from an even older implementation which is variously attributed to Dan Bernstein or Chris Torek. That older algorithm originally floated around on Usenet, and certain attribution is difficult. For example, there is some interesting commentary in this Apache code (search for their names) that references the original source.

Bottom line is, this is a very old, simple hashing algorithm. It is not the most performant, and it is not even proven mathematically to be a "good" algorithm. But it is simple, and a lot of people have used it for a long time with good results, so it has a lot of historical support.

user_331650

I'm just picking up Objective-C myself, so I can't speak for that language specifically, but in the other languages I use if two instances are "Equal" they must return the same hash - otherwise you are going to have all sorts of problems when trying to use them as keys in a hashtable (or any dictionary-type collections).

On the other hand, if 2 instances are not equal, they may or may not have the same hash - it is best if they don't. This is the difference between an O(1) search on a hash table and an O(N) search - if all your hashes collide you may find that searching your table is no better than searching a list.

In terms of best practices your hash should return a random distribution of values for its input. This means that, for example, if you have a double, but the majority of your values tend to cluster between 0 and 100, you need to make sure that the hashes returned by those values are evenly distributed across the entire range of possible hash values. This will significantly improve your performance.

There are a number of hashing algorithms out there, including several listed here. I try to avoid creating new hash algorithms as it can have large performance implications, so using the existing hash methods and doing a bitwise combination of some sort as you do in your example is a good way to avoid it.

Brian_B_218170

A simple XOR over the hash values of critical properties is sufficient 99% of the time.

For example:

- (NSUInteger)hash
{
    return [self.name hash] ^ [self.data hash];
}

Solution found at http://nshipster.com/equality/ by Mattt Thompson (which also referred to this question in his post :~)

user_12206420

The easy but inefficient way is to return the same -hash value for every instance. Otherwise, yes, you must implement hash based only on objects which affect equality. This is tricky if you use lax comparisons in -isEqual: (e.g. case-insensitive string comparisons). For ints, you can generally use the int itself, unless you’ll be comparing with NSNumbers.

Don’t use |=, though, it will saturate. Use ^= instead.

Random fun fact: [[NSNumber numberWithInt:0] isEqual:[NSNumber numberWithBool:NO]], but [[NSNumber numberWithInt:0] hash] != [[NSNumber numberWithBool:NO] hash]. (rdar://4538282, open since 05-May-2006)

Ahruman_64430

I've found this page to be a helpful guide in override equals- and hash-type methods. It includes a decent algorithm for calculating hash codes. The page is geared towards Java, but it's pretty easy to adapt it to Objective-C/Cocoa.

user_288040

This doesn't directly answer your question (at all) but I've used MurmurHash before to generate hashes: murmurhash

Guess I should explain why: murmurhash is bloody fast...

schwa_231130

I'm an Objective C newbie too, but I found an excellent article about identity vs. equality in Objective C here. From my reading it looks like you might be able to just keep the default hash function (which should provide a unique identity) and implement the isEqual method so that it compares data values.

ceperry_704100

Note that if you're creating a object that can be mutated after creation, the hash value must not change if the object is inserted into a collection. Practically speaking, this means that the hash value must be fixed from the point of the initial object creation. See Apple's documentation on the NSObject protocol's -hash method for more information:

If a mutable object is added to a collection that uses hash values to determine the object’s position in the collection, the value returned by the hash method of the object must not change while the object is in the collection. Therefore, either the hash method must not rely on any of the object’s internal state information or you must make sure the object’s internal state information does not change while the object is in the collection. Thus, for example, a mutable dictionary can be put in a hash table but you must not change it while it is in there. (Note that it can be difficult to know whether or not a given object is in a collection.)

This sounds like complete whackery to me since it potentially effectively renders hash lookups far less efficient, but I suppose it's better to err on the side of caution and follow what the documentation says.

user_103450
Related