"Richard A. O'Keefe" wrote:
> 
> "Lex Spoon" <lex at cc.gatech.edu> wrote:
>         > (There are other orders which are also compatible with =, such as
>         >  - first compare the lengths,
>         >  - then compare the elements ONLY if the lengths are the same.
>         >  This is almost always faster, and has some nice properties.
> 
>         Gee, yet another way to compare collections.  Note, though, that this is
>         not a total ordering, either -- arrays of different lengths will be ~=
>         but they will be neither #< nor #> .
> 
> No, it's a total order.
>         #(4) < #(1 2)           (length is less)
>         #(1 2) < #(4 2)         (length =, first element <)
>         #(1 2) < #(1 4)         (length =, last element <)
> 
> If two arrays have different lengths, then one of them is shorter,
> and that one is less.  If they have the same length, then either
> all elements are equal, so the arrays count as equal,
> or one position has unequal elements, and the first such position
> determines the order.  (Of course, taking the last position would
> also work.)
> 
> This order was used on strings in PL/CV,
> and it's _almost_ the order Prolog uses.
> (Prolog actually compares function symbols first, then arity,
>  and then elements.  So if you use f, f(X1), f(X1,X2), .. and so
>  on for sequences, you get this order.)
> 
> I'm not advocating that ordering; lexicographic has always suited me.
> I brought it up because that's the reason why I didn't just post an
> [ENH].

Adding #<, #>, etc. methods to SequenceableCollection seems like a reasonable idea in general... certainly no worse than #+, #sum, etc.  And one of its subclasses (String) already implements them, so moving the *same* behavior up to SequenceableCollection makes some sense.

Using an unusual ordering like the one described above (length-first, elements-second) seems like a bad idea to me, though, even if it has some performance benefits.  I would go with the most obvious ordering, which would be whatever String uses (lexicographic), which I guess is what you originally proposed.

You mentioned that it already exists in Python... does Python use lexicographic ordering?  (Also, do any other Smalltalks implement these methods?)

- Doug Way
  dway at riskmetrics.com