bug#72960: 31.0.50; PGTK Wayland exhibits more lag than X11 version

[Eli Zaretskii <eliz@gnu.org>]

> Date: Tue, 03 Sep 2024 16:36:44 +0100
> From: "Stephane Travostino" <sph@combo.cc>
> Cc: 72960@debbugs.gnu.org
> 
> > Please try profiling the lagging cases with "M-x profiler", and post
> > the profile here.
> 
> I don't know how to make a consistent test case. I have tried here to profile opening Emacs (same commit with and without PGTK) on the same 547-line Elixir file, and holding the Down key until it reaches the bottom and then back to the top of the buffer. I have (setopt scroll-conservatively 101) so after the first page the contents are continuously redrawn for every new line. 
> 
> The PGTK version feels like it's skipping frames while it's relatively smooth on X11:
> 
> X11:        
>         8795  86% + redisplay_internal (C function)
>         1141  11% + command-execute
>           54   0% + direnv--maybe-update-environment
>           49   0% + gcmh-register-idle-gc
>           42   0% + winner-save-old-configurations
>           20   0% + timer-event-handler
>           18   0% + ...
>           18   0% + jit-lock--antiblink-post-command
> 
> 
> PGTK:
>         9387  91% + redisplay_internal (C function)
>          698   6% + command-execute
>           19   0% + ...
>           19   0% + timer-event-handler
>           12   0% + direnv--maybe-update-environment
>           11   0% + winner-save-old-configurations
>  
> I have run this a few times and in Wayland `redisplay_internal` takes always a few percent more time than on X11, though I am not sure these numbers can prove anything as they are quite close.

Thanks.  Maybe Po Lu will have some ideas.

> Is there some kind of consistent UI benchmark I can run? The frame skipping reminds me of missed vsync deadlines one might experience in games.

Try this:

  (defun scroll-up-benchmark ()
    (interactive)
    (let ((oldgc gcs-done)
	  (oldtime (float-time)))
      (condition-case nil (while t (scroll-up) (redisplay))
	(error (message "GCs: %d Elapsed time: %f seconds"
			(- gcs-done oldgc) (- (float-time) oldtime))))))

Evaluate this function, then visit a large file, like src/xdisp.c from
the Emacs sources, and invoke "M-x scroll-up-benchmark RET".  It will
show the time it took at the end.  Record the results and compare with
the other configuration.