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

[Dmitry Gutov]

On 12/11/2023 04:36, João Távora wrote:
> > cl-lib is not that well-optimized either.
> That's certainly true.  I've been looking at the code and there
> are some serious optimization opportunities in cl-lib.el.  The
> one you found right away, destructive versions delegating to
> overconsing non-destructive ones, but also other opportunities.
>
> I attach a patch that I hope you can try with that bug report.
> Very lightly tested here, but seems to show a measurable
> improvement.

Looks like a solid improvement:

(setq cc (all-completions "" obarray))
;; (length cc) => 46002

(defvar list2 '("abc" "abc" "abc" "abc" "abc" "abc"))

(benchmark-run 10 (cl-set-difference cc list2 :test 'equal))
;; => 0.430

(benchmark-run 10 (setq cc (cl-nset-difference cc list2 :test 'equal)))
;; => 0.137 (!)

(benchmark-run 10 (my/set-difference cc list2))
;; => 0.144

(benchmark-run 10 (seq-difference cc list2))
;; => 0.814

(defun my/set-difference (list1 list2)
  (delq 'wrong
        (mapcar
         (lambda (c) (if (member c list2)
                    'wrong
                  c))
         list1)))

So it seems on par with the hand-written version (byte-compiled, too) 
that hardcodes the test function.

Curiously, though, if I make list2 longer (e.g. 12 elements rather than 
6), then my/set-difference starts to win noticeably (0.166 vs 0.200).

I'll keep the real-world case as-is (code in project--vc-list-files) 
both because the result with cl-nset-difference is still a little 
slower, and because it's slower in all released emacsen, but it 
definitely becomes an option.