Hello,
Andy Wingo <wi...@pobox.com> writes:
> On Tue 20 Oct 2009 10:38, l...@gnu.org (Ludovic Courtès) writes:
>
>> Andy Wingo <wi...@pobox.com> writes:
>>
>>> (sort '((a . 1) (b . -2) (c . 3)) < cdr)
>>> => ((b . -2) (a . 1) (c . 3))
>>
>> FWIW I find:
>>
>> (sort '((a . 1) (b . -2) (c . 3))
>> (lambda (x y)
>> (< (cdr x) (cdr y))))
>>
>> more in the spirit of not “piling feature on top of feature”.
>
> (sort l < #:key cdr) is a bit contrived. But if your key takes time to
> compute, you need to use the decorate-sort-undecorate idiom, otherwise
> you compute the sort key O(n log n) times instead of O(n) times. It's
> part of our old Lisp tradition :)
If you insist:
(sort '((a . 1) (b . -2) (c . 3))
(memoized-lambda
(lambda (x y)
(< (cdr x) (cdr y)))))
(With ‘memoized-lambda’ trivially implemented using a weak hash table,
etc.)
:-)
Also, it would be a departure from R5RS (and R6RS’ ‘list-sort’, FWIW).
Thanks,
Ludo’.
