bug#69555: 30.0.50; shr - Preserve indentation when shr-fill-text is nil

["Kévin Le Gouguec" <kevin.legouguec@gmail.com>]

[-- Attachment #1: Type: text/plain, Size: 2090 bytes --]

Eli Zaretskii <eliz@gnu.org> writes:

>> From: Kévin Le Gouguec <kevin.legouguec@gmail.com>
>> Cc: 69555@debbugs.gnu.org,  rahguzar@zohomail.eu
>> Date: Tue, 05 Mar 2024 20:22:52 +0100
>> 
>> Eli Zaretskii <eliz@gnu.org> writes:
>> 
>> > Also, please time the code on some substantially large body of text,
>> > with and without shr-fill-text, and compare that with the current
>> > version.  I think performance is an important aspect of any change in
>> > this area.
>> 
>> Can do; would "(elisp) Profiling" be the starting point?
>
> I think benchmark-run is a better starting point.
>
>> (Also wondering if we have any "standard" or preferred HTML documents or
>> websites to throw at shr.el for benchmarking purposes; if not, I guess
>> I'll peruse <https://en.wikipedia.org/wiki/Special:LongPages> 🤔)
>
> Actually, something with a lot of text in large paragraphs, sometimes
> indented, would be better.  Those Wikipedia pages are basically long
> lists, but there's not much opportunity there to perform filling and
> indentation of large amounts of text, which is what is sought here.
> Here's one possible candidate:
>
>   https://debbugs.gnu.org/Developer.html

Thanks for the pointers.  Attaching the over-engineered scripts I built
from that, which with 1000 REPETITIONS yield:

2024-03-05; 33976ecf244; 30.0.50; master          shr-fill-text=nil  ( 3.013 23  0.313)
2024-03-04; b06916cb218; 30.0.50; shr-blockquote  shr-fill-text=nil  ( 3.121 24  0.328)

2024-03-05; 33976ecf244; 30.0.50; master          shr-fill-text=t    (32.331 65  0.904)
2024-03-04; b06916cb218; 30.0.50; shr-blockquote  shr-fill-text=t    (32.045 65  0.902)

I can bump up REPETITIONS if that would help; sending the scripts &
results as-is before hitting the hay since I figure they might have some
glaring methodology issues, or there is more information I might not
have thought of reporting.

If not, the tentative conclusion would be "shr-fill-text nil gets 4%
slower; shr-fill-text t is none the worse for wear; nil still runs
circles around t"?


[-- Warning: decoded text below may be mangled, UTF-8 assumed --]
[-- Attachment #2: bench.el --]
[-- Type: text/x-emacs-lisp, Size: 1027 bytes --]

;; -*- lexical-binding: t; -*-

(require 'shr)

(defun bench/git (&rest args)
  (car (apply 'process-lines "git" "-C" source-directory args)))

(defun bench/describe-emacs ()
  (let* ((fork-point
          (bench/git
           "merge-base" "--fork-point" "origin/master"
           emacs-repository-version))
         (git-desc
          (bench/git
           "show" "--date=short" "--format=format:%cd; %h" fork-point)))
    (format "%s; %-7s; %s" git-desc emacs-version emacs-repository-branch)))

(let ((dom (with-temp-buffer
             (insert-file-contents "test.html")
             (libxml-parse-html-region)))
      (emacs-desc (bench/describe-emacs)))
  (dolist (val '(nil t))
    (setopt shr-fill-text val)
    (pcase-let ((`(,total-time ,gc-count ,gc-time)
                 (benchmark-run 1000
                   (with-temp-buffer (shr-insert-document dom)))))
      (message
       "%s\tshr-fill-text=%s\t(%6.3f %d %6.3f)"
       (bench/describe-emacs)
       shr-fill-text
       total-time gc-count gc-time))))

[-- Attachment #3: bench.sh --]
[-- Type: application/x-shellscript, Size: 210 bytes --]