How is the return value of __hash__ used?

Question

Suppose I write a class, but don\'t define a __hash__ for it. Then __hash__(self) defaults to id(self) (self\'s memory addres

Answer 1

Of course logically (from the view of code that uses the hash table) the object itself is the key. If you search for key "foo" in the hash table, no matter what other objects in the hash table have the same hash value as "foo", the corresponding value will only be returned if one of the key-value pairs stored in the hash table has key equal to "foo".

I don't know exactly what Python does, but a hash table implementation has to account for hash collisions. If the hash table array has N slots, then if you insert N + 1 values (and the table is not resized first), there must be a collision. Also, as in the case you mentioned where __hash__ always returns 1, or just as a quirk of the hash function implementation, it is possible to have two objects with exactly the same hash code.

There are two major strategies used to deal with hash collisions in a hash table for a single machine in memory (different techniques used for distributed hash tables, etc.):

1. Each slot in the array is a list (typically a linked list), and all values that hash to k modulo N are placed into the list at slot k. So if hash values collide, that isn't a problem because both objects with the same hash value end up in the same list.
2. Some kind of probing scheme. Basically, if the object you're inserting has hash value equal to k modulo N, you look at slot k. If it's full, you apply some formula to the current location (maybe just add 1), and look at the next slot. You follow a regular pattern to choose the next slot, given the original hash value and the number of probes so far, and keep probing until you find an open slot. This is less used, since if you aren't careful about your implementation you can run into clustering problems i.e. have to probe many many times before finding the object.

Wikipedia talks a lot more about hash table implementations here.