Hash of None varies per-machine

Dave Angel davea at ieee.org
Fri Apr 3 21:09:26 CEST 2009

ben.taylor at email.com wrote:
> Found this while trying to do something unrelated and was curious...
> If you hash an integer (eg. hash(3)) you get the same integer out. If
> you hash a string you also get an integer. If you hash None you get an
> integer again, but the integer you get varies depending on which
> machine you're running python on (which isn't true for numbers and
> strings).
> This raises the following questions:
> 1. Is it correct that if you hash two things that are not equal they
> might give you the same hash value? Like, for instance, None and the
> number 261862182320 (which is what my machine gives me if I hash
> None). Note this is just an example, I'm aware hashing integers is
> probably daft. I'm guessing that's fine, since you can't hash
> something to a number without colliding with that number (or at least
> without hashing the number to something else, like hashing every
> number to itself * 2, which would then mean you couldn't hash very
> large numbers)
> 2. Should the hash of None vary per-machine? I can't think why you'd
> write code that would rely on the value of the hash of None, but you
> might I guess.
> 3. Given that presumably not all things can be hashed (since the
> documentation description of hash() says it gives you the hash of the
> object "if it can be hashed"), should None be hashable?
> Bit esoteric perhaps, but like I said, I'm curious. ;-)
> Ben
1. Most definitely.  Every definition of hash (except for "perfect 
hash") makes it a many-to-one mapping.  Its only intent is to reduce the 
likelihood of collision between dissimilar objects.  And Python's spec 
that says that integers, longs and floats that are equal are guaranteed 
the same hash value is a new one for me.  Thanks for making me look it up.

2. Nothing guarantees that the Python hash() will return the same value 
for the same object between implementations, or even between multiple 
runs with the same version on the same machine.  In fact, the default 
hash for user-defined classes is the id() of the object, which will 
definitely vary between program runs.  Currently, id() is implemented to 
just return the address of the object.

3. Normally, it's just mutable objects that are unhashable.  Since None 
is definitely immutable, it should have a hash.  Besides, if it weren't 
hashable, it couldn't be usable as a key in a dictionary.

All my opinions, of course.

More information about the Python-list mailing list