From: "Mauricio Fernández" <mfp@...>
Date: 2005-09-14T22:09:07+09:00
Subject: Re: Sets, uniqueness not unique.

On Wed, Sep 14, 2005 at 09:14:10PM +0900, Hugh Sasse wrote:
> On Wed, 14 Sep 2005, David A. Black wrote:
> >On Wed, 14 Sep 2005, Hugh Sasse wrote:
> >>This being a Set I don't really need the call to include? now, but
> >>it's there (from when I was using a hash for this).
> >>
> >>I find two things that seem odd to me:
> >>1. eql? is never getting called, despite include?.
> >>2. I end up with duplicate students.
> >>
> >>Sets *can't* hold duplicates, and include depends on eql? for Sets.
> >
> >Are you sure about that latter point?  In set.rb:
> >
> > def include?(o)
> >   @hash.include?(o)
> > end
> >
> >and in hash.c:
> >
> >   if (st_lookup(RHASH(hash)->tbl, key, 0)) {
> >       return Qtrue;
> >   ... }
> >
> >I haven't followed the trail beyond that... but I think any two
> >student objects will count as different hash keys, even if they have
> >similar string data.

object.c:

VALUE
rb_obj_id(VALUE obj)
{   
    if (SPECIAL_CONST_P(obj)) {
        return LONG2NUM((long)obj);
    }
    return (VALUE)((long)obj|FIXNUM_FLAG);
}   
[...]
rb_define_method(rb_mKernel, "hash", rb_obj_id, 0);

[...]
> What i don't really know is what the sufficient conditions are for
> this?  Is it *necessary* to change hash and eql together? What are the
> defaults for Set?

The defaults are actually those of Hash. You can follow the call chain
starting from 

static struct st_hash_type objhash = {
    rb_any_cmp,
    rb_any_hash,
};

in hash.c. For user-defined classes, it will end up using #hash and #eql?
defined in Kernel. [rb_any_cmp and rb_any_hash have some extra logic for
Symbol, Fixnum and String values, and some core classes redefine the
associated methods].

Given the above definition of Kernel#hash, if you redefine it, you'll
most probably want to change #eql? too (see below). As far as Hash
objects (and hence Sets) are concerned, modifying #eql? while keeping
#hash unchanged would be effectless (unless you restrict it further so
that obj.eql?(obj) is false, which doesn't seem quite right).


static VALUE
rb_obj_equal(VALUE obj1, VALUE obj2)
{   
    if (obj1 == obj2) return Qtrue;
    return Qfalse;
}   

[...]
 rb_define_method(rb_mKernel, "eql?", rb_obj_equal, 1);

-- 
Mauricio Fernandez