PEP 218 Re: ANN: set-0.1 module available

[Gonçalo Rodrigues]

On 21 May 2002 11:07:35 +0200, Bernhard Herzog <bh at intevation.de> wrote:

>Gonçalo Rodrigues <op73418 at mail.telepac.pt> writes:
>
>> >If an object is stored in a dict as a key, the only things that matter
>> >are that
>> >
>> >1. its hash value doesn't change as long as the object is used as a key.
>> 
>> Doesn't this mean, in practice, that it is immutable?
>
>Not necessarily. If the hash value can change if the state of the object
>changes, you have to make sure that you don't change the object in such
>a way as long as it's used as a key in a dict.
>
>The easiest way to achieve this is of course to have the hash value
>constant for the life-time of the object.
>
>> To me, this requirement is just shifting responsabilities from one
>> place to the other and not really solving the problem.
>
>It may, yes. It's certainly not the best solution for objects in a set.
>
>> >2. equal objects have equal hash values
>> >
>> >There may be some more subtle requirements. I'm not sure what happens
>> >when two objects that have the same hash value but do not compare eqal
>> >at first later become equal, for instance.
>> 
>> I believe 2. is more stringent:
>> 
>> equal objects iff equal hash values
>
>That can't work. Hash values are ints which today means that it's a C
>long and therefore either 32bit or 64 bit on most machines. Obviously
>there are far fewer possible hash values than strings for instance, so
>there must be at least two strings which are unequal but have the same
>hash value.
>

Ah, thanks for the clarification. Since nowadays I only program in
Python, when I see int i see integer-arbitrarily-big.

>An easy example for hashes of ints:
>
>Python 2.2 (#1, Jan 17 2002, 21:04:07) 
>[GCC 2.95.4 20011006 (Debian prerelease)] on linux2
>Type "help", "copyright", "credits" or "license" for more information.
>>>> -1 == -2
>0
>>>> hash(-1) == hash(-2)
>1
>>>> 
>
>Most of the time, i == hash(i) for any python int except -1, because -1
>is indicates an error in the C-level equivalent of __hash__.
>
>   Bernhard

Best regards,
Gonçalo Rodrigues