[plt-scheme] Multiple values

From: Bill Wood (william.wood3 at comcast.net)
Date: Mon Feb 20 19:01:30 EST 2006

On Tue, 2006-02-21 at 00:03 +0100, hufflen at lifc.univ-fcomte.fr wrote:
   . . .
>    Yes.  More generally, I often use multiple values because they allow
> multiple computation. For an "actual" example, let us consider a
> non-empty linear list. Let us assume that we have to get, on the one
> hand, the list but its last element, on the other hand, its last
> element. You can write two functions for these two purposes, but they
> would go along the list twice. Multiple values allow to get the two
> results in one pass:
> 
> (define (butlast-and-last l)
>   ;;  "l" is supposed to be non-empty, returns two values:  a linear list
> with
>   ;;  all the elements of "l" but the last one, and the last one.
>   (let ((first (car l))
>         (rest (cdr l)))
>     (if (null? rest)
>         (values '() first)
>         (call-with-values (lambda () (butlast-and-last rest))
>           (lambda (l0 last-0) (values (cons first l0) last-0))))))

I know I'm coming in *way* late to the discussion (particularly given
the presence of multiple values in Common Lisp), but while I find some
elegance in the multiple values mechanism I have to ask, is there a
killer argument for multiple values over an ordinary function returning
a (proper or improper) list?  Does the code above win big over, for
example,

(define (butlast-and-last l)
  (let ((first (car l))
        (rest (cdr l)))
    (if (null? rest)
        (list* '() first)
        (match (butlast-and-last rest)
          ([l0 . last0] (list* (cons first l0) last0))))))

or is it a matter of taste.  For example, can a compiler do something
about the non-tail recursion in M. Hufflen's example?

Not carping, just a question,

 -- Bill Wood




Posted on the users mailing list.