[r6rs-discuss] [Formal] Add (sort) and (vector-sort!) procedures

[Eli Barzilay]

I think that the `sort' issue is hopeless, since there are many
features you need to make everyone happy.  But:


On Oct  2, John Cowan wrote:
> Marcin 'Qrczak' Kowalczyk scripsit:
> > If any sorting is adopted, I call for making Schwartzian transform
> > available.

Please don't use that name.  It's obnoxious to put a label that
credits a Perl hacker for rediscovering a popular Lisp feature, and
making the result sound like some method in calculus.


> > People must not be advised to pass
> >    (lambda (x y) (< (car x) (car y)))
> > as the comparison. They should pass
> >    car <
> > as two separate arguments (or just car with < being the default).
> 
> If this is a convenience feature, I'm against it.  If it supposedly
> provides improved performance or function of some sort, I'd like to
> see how.

Marcin gave an example that doesn't show the saving.  Compare

  (sort file-names
        (lambda (f1 f2) (< (file-date f1) (file-date f2))))

and

  (map cdr
       (sort (map (lambda (f) (cons (file-date f) f)) file-names)
             (lambda (p1) (< (car p1) (car p2)))))

The scond version is doing half the number of FS accesses.  Some Lisp
implementations will do something similar when receiving a `:key'
argument.  (And a Perl hacker took it, and it turned to the above
"Transform", as wikipedia will happily tell you.)  I personally think
that since

  (lambda (x y) (compare (access x) (access y)))

is a very common pattern, then an additional key argument is a good
thing to have for convenience.

-- 
          ((lambda (x) (x x)) (lambda (x) (x x)))          Eli Barzilay:
                  http://www.barzilay.org/                 Maze is Life!