> -----Original Message-----
> From: Mathieu Bouchard [mailto:matju / cam.org]
[...]
> ListMixin #(common to Array and String)
> ArrayMixin #(includes ArrayInterface, ListMixin)
> ArrayInterface #(the basic operations that are left to implement)
> ArrayInterfaceSafe #(assertion-checked version of the preceding)
> CommonAssert #(used by ArrayInterfaceSafe, StringInterfaceSafe)
> ArrayUsingArray #(dummy impl of ArrayInterfaceSafe, includes ArrayMixin)
> 
> As you see, methods are not divided according to individual purposes but
> according to a larger scheme of small interface vs large interface /
> interface vs implementation vs verification / shared with string vs
> specific to array.
> 
> The "set operations" in Array are somewhat more restrictive than set
> operations. See [ruby-talk:6756], [ruby-talk:6723]. Order of elements is
> important (repeating elements do collapse, however).

I remember this discussion. The problem with set operations is 
that they really are somewhat arbitrary (ultimately indicator
Hash functions are a better way of implementing sets IMHO) due to 
the fact that there is no natural way to order elements/objects 
(besides ``elements.id'' but this depends on the context) and the
built-in set operations are modeled around the set theoretic 
interpretation - a multi set interpretation is just as sensible. 
Also self-referential Arrays - try to run  
 ruby -e 'x = [1]; x << y = ["a"]; y << [x]; x & x' 
- are the enemies of the set theoretic interpretation. Because of
this a ArrayMixin writer might not want to provide any set-theoretic
facilities at all (see below).
  
> I don't see the usefulness of putting lists/queues/stacks/sets
> functionalities in different modules (that i'll have to include into
> ArrayMixin to respect the builtin Array protocol) if there is nowhere they
> can be reused.
>
> The "set operations" could be somewhat reusable in another context (though
> they could be optimized out of their restrictions in those contexts, and 
> #uniq would become irrelevant, etc).
> 
> May I know what you have in mind in a detailed fashion?

Actually the main motivation behind organizing lists/queues/stacks/sets
functionality into modules would be that you could choose NOT to include
all of these functionalities  - i.e. it would be nice if you could
write  ListMixin/QuesesMixin ... 's which would be functional on their 
own but provide the full Array facilities when included into a single 
Module.



[...] 
> The builtin semantics follow my implementation.
> 
> Or rather:
> 
> My implementation follows the builtin semantics.
> 
> It's just that when you try to #flatten! an array that immediately
> contains self-references, it skips them as a special case (i don't know
> why Array#flatten! does this). However when you do #flatten, a dup is

The built-in #flatten! does not throw exceptions on more 
complicated examples like 
$ ruby -e 'x = [1]; x << y = ["a"]; y << [x]; x.flatten!; p x'
[1, "a"]
either, it simply  throws out a self-referential entry if it 
encounters it a second time - kind of fun if you #flatten! 
in a threading environment. 
   
> performed first, then #flatten! is called on the copy, and so the
> base-level self-references are not base-level anymore, and #flatten!
> explodes.
> 

Christoph