ansaurus

Question

why dict objects are unhashable in python?

Answer 1

+11 A:

Short answer: because they are mutable containers.

If a dict was hashed, its hash would change as you changed its contents.

Ben James 2009-12-24 08:36:25

I don't even want to begin to imagine the nightmarish logic it would take to use an object as the key.

David 2009-12-24 08:40:24

Actually Python makes hashing objects easy, it gives each one a unique and constant identity which can be hashed.

Ben James 2009-12-24 08:42:34

yes, but by default instances of user defined objects always compare unequal.

hop 2009-12-24 08:47:52

but I wonder what's the reason behind it - to keem dict. unhashable?

Tumbleweed 2009-12-24 08:51:30

make a hashable representation of the dict if you must -- for example `frozenset(D.items())` (for `D` dict). A `frozenset` is a hashable and immutable set -- a `set` is unhashable for the same reason as `dict`.

kaizer.se 2009-12-24 10:37:30

@shahjapan: the reason is that it prevents hard-to-find bugs, especially for people that don't understand hashing all that well. As Dave Kirby shows below, there is a simple workaround, but that you have to use a workaround is a simple warning that you need to understand what you are doing, lest you 'lose' key-value pairs.

Confusion 2009-12-24 13:07:04

@Confusion yes thanks I understood now, but I was Wondering because if you consider java or c# Hashtables are hashable by default :-)

Tumbleweed 2009-12-24 14:10:32

@kaizer.se: for frozenset to freeze a dictionary, all the dictionary keys and values need to be hashable too. try `frozenset({1:2, 3:[]})`.

ΤΖΩΤΖΙΟΥ 2009-12-25 09:37:56

Answer 2

+1 A:

None of the mutable container types in Python are hashable, because they are mutable and thus their hash value can change over their lifetime.

hop 2009-12-24 08:38:41

tuples and strings are containers, yet hashable and immutable.

Ignacio Vazquez-Abrams 2009-12-24 08:40:18

a stupid typo that you could easily have corrected

hop 2009-12-24 08:57:56

@Ignacio Vazquez-Abrams: strings are sequences but not containers. Also, a tuple containing a non-hashable item is also non-hashable; try using `([1], {2:3})` as a dict key.

ΤΖΩΤΖΙΟΥ 2009-12-25 09:41:28

Answer 3

+2 A:

As others have said, the hash value of a dict changes as the contents change.

However if you really need to use dicts as keys, you can subclass dict to make a hashable version.

>>> class hashabledict(dict):
...    def __hash__(self):
...        return id(self)
... 
>>> hd = hashabledict()
>>> d = dict()
>>> d[hd] = "foo"
>>> d
{{}: 'foo'}

>>> hd["hello"] = "world"
>>> d
{{'hello': 'world'}: 'foo'}

This replaces the hash value used for the dict with the object's address in memory.

Dave Kirby 2009-12-24 13:00:31

and then can I replace normal dict byedict = hashabledict

Tumbleweed 2009-12-24 14:09:01

But this is useless: if I store a value under `{}`, I can't look it up with `{}`, because the two empty hashabledicts have different ids, and different hashes. The important thing about a hash function is that it must return the same hash for two "equal" values.

Ned Batchelder 2009-12-24 15:04:51

@Ned - Doh! You are right. What is really needed is a frozendict that acts the same way as a frozenset. You could subclass dict to define one, as in this recipe on ASPN: http://code.activestate.com/recipes/414283/

Dave Kirby 2009-12-24 15:52:33

ansaurus

tags:

views:

answers:

why dict objects are unhashable in python?

related questions