bug#46350: 28.0.50; touchpad-scrolling-eats-lots-of-cpu-samples

[Eli Zaretskii <eliz@gnu.org>]

> From: Andrey Orst <andreyorst@gmail.com>
> Date: Sat, 6 Feb 2021 21:34:29 +0300
> Cc: 46350@debbugs.gnu.org
> 
> > Please load mwheel.el (NOT the .elc file!), and then profile the slow
> > scrolling again and show a fully-expanded profile.  That might help us
> > understand what part of mwheel-scroll takes the lion's share of CPU
> > cycles.
> 
> Here's whole report expanded (btw can't find if there's expand-all feature?):

Yes, "C-u RET".

>        11593  91% - command-execute
>        11549  91%  - funcall-interactively
>        11542  91%   - mwheel-scroll
>        11536  91%    - let*
>        11420  90%     - condition-case
>        11381  90%      - unwind-protect
>        11377  90%       - let
>        11373  90%        - cond
>        11372  90%         - condition-case
>        10868  86%          - funcall
>          324   2%           - scroll-down
>          263   2%            - jit-lock-function
>          255   2%             - jit-lock-fontify-now
>          235   1%              - jit-lock--run-functions

I cannot make sense out of this: this profile says that funcall takes
most of the time, and the function it calls, scroll-down, takes almost
no time?  Does anyone have any idea what could be going on here?

If this is just some artifact of "M-x profile", and most of the time
is spent in scroll-down, then I'm not sure what can be done, nor why
you see something I don't.  Do you use a large frame and/or a small
font?  And how many mouse-wheel events Emacs receives in your
scenario, anyway?

Oh, and what happens if you raise gc-cons-threshold to a large value?