From: Charles Mills <cmills@...>
Date: 2004-08-06T23:08:47+09:00
Subject: Re: Sets are not extensional


On Aug 6, 2004, at 3:11 AM, Robert Klemme wrote:

>
> "Charles Mills" <cmills@freeshell.org> schrieb im Newsbeitrag
> news:2B60D928-E751-11D8-AF77-000A95A27A10@freeshell.org...
>>
>> On Aug 5, 2004, at 6:41 PM, George Ogata wrote:
>>
>>> Kristof Bastiaensen <kristof@vleeuwen.org> writes:
>>>
>>>> {}.hash == {}.hash
>>>> => false
>>>> Set[].hash == Set[].hash
>>>> => false
>>>> [].hash == [].hash
>>>> => true
>>>>
>>>> I have no idea for the reason why different hashes have different
>>>> hashvalues, while different arrays may get the same value (based on
>>>> the
>>>> contents of the array).
>>>
>>> This seems like a blatant bug to me.  `a.eql? b' must imply `a.hash 
>>> ==
>>> b.hash', yet:
>
> Yes, this issue has come up before IIRC.  Maybe Matz did something 
> about
> this but it's not included in a release version.
>
>> I agree:
>>> a = {} #=> {}
>>> a.hash #=> 1650134
>>> a.id #=> 1650134
>>> b = {} #=> {}
>>> b.hash #=> 1639814
>> # So hash on an empty hash returns the objects id....
>>> a = {"a"=>"b"} #=> {"a"=>"b"}
>>> a.hash #=> 1682094
>>> a.clear #=> {}
>>> a.hash #=> 1682094
>> # Looks like no two hashes hash to one another...
>> # Here is the reason
>> /* in object.c */
>>      rb_mKernel = rb_define_module("Kernel");
>>      rb_include_module(rb_cObject, rb_mKernel);
>> /* ..snip.. */
>>      rb_define_method(rb_mKernel, "object_id", rb_obj_id, 0);

/* I intended to show this line: */
     rb_define_method(rb_mKernel, "hash", rb_obj_id, 0);

Anyway, so the code above shows class Object including module Kernel.  
Module Kernel defines "hash" as rb_obj_id().  So Object gets 
rb_obj_id() as its "hash" method.

>>
>> /* in hash.c the "hash" method is never overridden.  So rb_obj_id() is
>> the "hash" method for hashes. */
>
> Where do you take that conclusion from?  From the source quotes you 
> name,
> there is no indication of a connection between #hash and #id.  And 
> you're
> showing code for Kernel which is unrelated to this thread.  The only
> conclusion so far is, that Hash#hash is Object#hash (because it's not
> overridden as you state) and thus can't take into account the contents 
> of
> the Hash.  Can you please clarify this?

Don't forget you can always look at the source yourself :)
-Charlie