Arc Forum | This is a rare case where I think the tail-recursive solution is more readable: ...

Arc Forum

2 points by rkts 5905 days ago | link | parent

This is a rare case where I think the tail-recursive solution is more readable:

  (def partition (f xs)
    (if (alist xs)
      (let f (testify f)
        ((afn (xs a b)
           (if (no xs)
             (list (rev a) (rev b))
             (f (car xs))
               (self (cdr xs) (cons (car xs) a) b)
               (self (cdr xs) a (cons (car xs) b))))
         xs nil nil))
      (map [coerce _ 'string] (partition f (coerce xs 'cons)))))

Also, how about generalizing accum to multiple lists?

  (mac accums (accfns . body)
    (let gaccs (map [uniq] accfns)
      `(withs ,(mappend (fn (gacc accfn)
                          (list gacc 'nil accfn `[push _ ,gacc]))
                        gaccs accfns)
         ,@body
         (list ,@(map [list 'rev _] gaccs)))))

  (def partition (f xs)
    (if (alist xs)
      (let f (testify f)
        (accums (a b) (each x xs ((if (f x) a b) x))))
      (map [coerce _ 'string] (partition f (coerce xs 'cons)))))

3 points by almkglor 5905 days ago | link

True, although the final (rev a) (rev b) loses some of the speed benefit in the tail-recursion. If we didn't care about order (which happens quite often) we can drop the (rev ...) things. Otherwise, we can use a modulo-cons form, but this exceedingly complicated.

accums seems good... I suggest pushing it on nex-3's arc-wiki.

What I would like to see is a proper order-preserving accum. Here's a try at a "reasonably efficient" implementation:

  (mac accumf (accfn . body)
    " Collects or accumulates the values given to all calls to `accfn' within
      `body' and returns a list of those values.  The returned list
      is in the same order as calls to `accfn'.
      See also [[accum]] [[summing]] "
    (w/uniq (hd tl)
      `(let (,hd ,tl ,accfn) nil
            (= ,accfn
              (fn (s)
                (if ,hd
                    (do (= ,hd (cons s nil)) (= ,tl ,hd))
                    (do (= (cdr ,tl) (cons s nil)) (= ,tl (cdr ,tl))))))
            ,@body
            hd))))

-----

2 points by sacado 5905 days ago | link

Thanks for your answers ! I think that might be an interesting functionnality to add in the core functions. I used it a few times, in quite different contexts. Partitioning a list from a given criteria looks like a frequent action...

-----