Re: What's missing in ELisp that makes people want to use cl-lib?

[Augusto Stoffel]

On Wed, 15 Nov 2023 at 14:51, João Távora wrote:

> That and the fact they aren't aware they're anywhere from 5x to
> 10x slower, sometimes more (I measured 18x for decently realistic),
> than what they could be.

seq can surely be optimized, and I'm skeptical the issue lies in the
generic method dispatch (which would be a fundamental problem).  It
seems more likely due to the fact that the default implementations wrap
predicate functions and macro bodies into lambdas.

I tried one of the your benchmarks, with results consistent with yours:

  (setq cc (all-completions "" obarray))
  (setq list2 (make-list 12 "shooveedoowaa"))

  (when nil
     (benchmark-run 100 (cl-set-difference cc list2 :test #'equal))
     ;(11.517403284999999 45 4.082223424999995)

     (benchmark-run 100 (seq-difference cc list2))
     ;(18.507667340999998 119 10.999239876000019)
     )

Then I tried the first optimization that comes to mind, which is to
redefine seq-doseq as

  (defmacro seq-doseq (spec &rest body)
    "Loop over a SEQUENCE, evaluating BODY with VAR bound to each of its 
elements.

  Similar to `dolist' but can be applied to lists, strings, and vectors.

  \(fn (VAR SEQUENCE) BODY...)"
    (declare (indent 1) (debug ((symbolp form &optional form) body)))
    `(let ((var1234 ,(cadr spec)))
       (cond
        ((listp var1234)
         (dolist ,spec ,@body))
        (t
         (seq-do (lambda (,(car spec))
                   ,@body)
                 var1234)))))

And now I get:

  (when nil
     (benchmark-run 100 (cl-set-difference cc list2 :test #'equal))
     ;(11.517403284999999 45 4.082223424999995)

     (benchmark-run 100 (seq-difference cc list2))
     ;(7.308574982 29 2.6960850209999876)
     )

> Also the fact they don't know cl-seq.el provides all that with
> a much more versatile interface, and frequently much better
> performance.

I for one like the simple aesthetics of seq.