Re: Speed of all-completions

Re: Speed of all-completions

[Stefan Monnier]

Barely 6 years ago, Jesper Harder <harder@ifa.au.dk> wrote:
> I use icomplete-mode and I've noticed that it feels more sluggish in
> cvs Emacs than in Emacs 21.3.
> Profiling shows that the bulk of the time is spent in
> `all-completions', so I tried to benchmark it with this example:
> (defmacro time (form)
>   `(let ((t1 (float-time)))
>     ,form
>     (- (float-time) t1)))
> (let ((oba (make-vector 255 0)))
>   (dotimes (i 10000)
>     (intern (format "f%i" i) oba))
>   (time
>    (all-completions "f" oba)))
> The time used was:
>   cvs Emacs:   0.16448211669921875
>   Emacs 21.3:  0.01507115364074707
> i.e. `all-completions' is roughly an order of magnitude slower in cvs
> Emacs.

I cannot reproduce it now.  I tried it with `emacs21', `emacs22', and
`emacs23' (under Debian testing, and adding a 0 to the number of
iterations of your test), and I do see that Emacs is getting slower, the
difference is not nearly as large, changing from 0.05s to
0.06s, basically.


        Stefan "you can always count on a prompt answer"

Re: Speed of all-completions

[Lawrence Mitchell]

Stefan Monnier wrote:
> Barely 6 years ago, Jesper Harder <harder@ifa.au.dk> wrote:

[...]

> I cannot reproduce it now.  I tried it with `emacs21', `emacs22', and
> `emacs23' (under Debian testing, and adding a 0 to the number of
> iterations of your test), and I do see that Emacs is getting slower, the
> difference is not nearly as large, changing from 0.05s to
> 0.06s, basically.

FWIW, I see the same behaviour here 2 times out of 3.  On the
third occasion, Emacs takes about half a second in GC.  With the
following test case:

(defvar foo-ob nil)
(defun foo (n)
  (when (null foo-ob)
    (setq foo-ob (make-vector 255 0))
    (dotimes (i n)
      (intern (format "f%i" i) foo-ob)))
  (time (all-completions "f" foo-ob)))

(byte-compile 'foo)
(setq garbage-collection-messages t)
(setq gc-cons-threshold 400000)
(foo 100000)


With (setq gc-cons-threshold 4000000) all-completions GCs one
time in five.

Cheers,
Lawrence
-- 
wence@gmx.li