bug#77924: 31.0.50; [Feature branch] Change marker implementation

[Gerd Möllmann]

Visuwesh <visuweshm@gmail.com> writes:

> [வியாழன் ஏப்ரல் 24, 2025] Eli Zaretskii wrote:
>
>>> From: Stefan Monnier <monnier@iro.umontreal.ca>
>>> Cc: gerd.moellmann@gmail.com,  stefankangas@gmail.com,
>>>     77924@debbugs.gnu.org, Ihor Radchenko <yantar92@posteo.net>
>>> Date: Wed, 23 Apr 2025 11:41:43 -0400
>>> 
>>> > Btw, did I miss some benchmarks?  I presume that the main motivation
>>> > for this branch is to speed up markers and charpos-to-bytepos
>>> > conversions, is that right?  If so, the only benchmark I saw shows only
>>> > a very mild speed-up, but maybe I missed something?
>>> 
>>> AFAICT we have tuned the current charpos<->bytepos conversion code such
>>> that its worst case virtually never shows up in practice.
>>> 
>>> Supposedly Ihor has a use case where that conversion still can make
>>> a noticeable difference, but I don't know if he tried the `text-index`
>>> branch to confirm that it makes a measurable difference on his case.
>>> 
>>> Gerd and I played with some micro benchmarks to confirm that the
>>> pathological behaviors are indeed solved, but other than that, the focus
>>> was on making sure the branch does not make the normal case slower
>>> (contrary to my sorted-array-of-markers-with-gap).
>>
>> Still, for performance-oriented changes, it would be nice to see some
>> benchmarks.  Scrolling through some large file with many non-ASCII
>> characters comes to mind (the Unicode UCD has quite a few of such
>> files).
>
> Is there a standard benchmark code that I can try for this?  I have such
> a large HTML file with Tamil text, and I can scroll through the buffer
> produced by shr-render-buffer.

The elisp-benchmarks package has an elb-scroll.el, in its benchmarks
directory. AFAICT, that loads an xmenu.c from the resources
sub-directory of the benchmarks.

You could copy that, or modify it.

P.S.

One can set the benchmarks directory to somewhere else, and run
benchmarks like this:

  (setq elb-bench-directory "~/emacs/notes/code/benchmarks")
  (elisp-benchmarks-run ".*replace-region-contents.*" t 100)

The 100 is the number of runs, which should be chosen high enough that
the "err" column in the result buffer is reasonably low. I'd start with
1 to see how long that takes, and then increase it.