bug#40323: 28.0.50; error in process filter: Invalid search bound (wrong side of point)

[Noam Postavsky <npostavs@gmail.com>]

> The patch you provided does indeed seem to fix the issue, or at least
> it hasn't happened for the time I tested it (I can verify various
> processes were running and outputting things to the async buffer).
>
> I do not have any idea as to why did those markers get swapped
> around. Is there any specific place that deals with them that I could
> check? Could this be a bug with comint mode, maybe clashing with
> another package?

Try reproducing the issue after evaluating the code below, perhaps
*trace-output* will have some useful clues.

    (load-library "comint.el") ;; Can only trace set-marker from Lisp source.

    (defun bug-40323-get-comint-output-marker ()
      (list :comint-pmark
            (and (markerp comint-last-output-start)
                 (eq (marker-buffer comint-last-output-start)
                     (current-buffer))
                 (process-mark (get-buffer-process (current-buffer))))))

    (dolist (fun '(set-marker
                   comint-send-input
                   comint-output-filter
                   comint-adjust-window-point
                   comint-adjust-point
                   ansi-color-process-output))
      (trace-function fun nil #'bug-40323-get-comint-output-marker))