From: Ryan Hinton <iobass@...>
Date: 2007-05-25T00:55:06+09:00
Subject: Re: Enumerator for Hash

On May 23, 12:18 pm, Robert Klemme <shortcut...@googlemail.com> wrote:
> You could wrap your keys in a custom class.  For added convenience you
> could also wrap Hash with another class which wraps keys on insertion
> and unwraps on reading.
> ...
>
> In that case I'd probably rather do this:
>
> my_array.inject({}) {|hs, obj| hs[obj.key] ||= obj; hs }.values

Now I remember why I wanted my earlier approach.  My object has
several Array member variables.  By default, my #hash and #eql?
methods use all of them.  In one case I need to ignore 2 of the 4
arrays, e.g. for a uniqueness hash.  I can use your second method
thus.

  my_array.inject({}) {|hs, obj| hs[obj.arr1 + obj.arr2] ||= obj;
hs }.values

But won't this create a bevy of new Array objects with the "+" method
(two for each item in the original array)?  I could instead pre-hash
the values something like the following.

  my_array.inject({}) {|hs, obj| hs[obj.arr1.hash ^ obj.arr2.hash] ||=
obj; hs }.values

This avoids the new Array spew, but I will lose unique objects if my
fabricated hash keys clash (i.e. same hash key for unique arrays).  If
I can specify different methods to use in place of #hash and #eql?, I
can get the same behavior without the performance penalty.  (My code
is already slow.)

I think I'll use your wrapping Hash idea.  Thanks!