bug#23079: 25.0.92; Movement commands leave cursor in invisible line

[Michael Heerdegen <michael_heerdegen@web.de>]

Stefan Monnier <monnier@iro.umontreal.ca> writes:

> The default "stickiness" of overlay boundaries is the inverse of that if
> text-properties, for some reason.  So, this test above suffers from the
> bug#19200, but other than that, it seems to work correctly for me
> (i.e. it prefers putting point in front of "333" rather than in front
> of "555").
>
> And if I remove the "nil t t" args to make-overlay, then
> point-adjustment correctly tries to put point in front of "555" instead
> of putting it in front of "333".

Ok, we are coming closer...

They use the default stickiness for their overlays.  Nonetheless, the
behavior is the same as with "nil t t".

_But_ their local binding of post-command-hook looks like (t
magit-section-update-highlight).

When I set it to (t) only, the issue is fixed (but the highlighting
update doesn't work anymore, of course)!


magit-section-update-highlight is

#+begin_src emacs-lisp
(defun magit-section-update-highlight ()
  (let ((section (magit-current-section)))
    (unless (eq section magit-section-highlighted-section)
      (let ((inhibit-read-only t)
            (deactivate-mark nil)
            (selection (magit-region-sections)))
        (mapc #'delete-overlay magit-section-highlight-overlays)
        (setq magit-section-unhighlight-sections
              magit-section-highlighted-sections
              magit-section-highlighted-sections nil)
        (unless (eq section magit-root-section)
          (run-hook-with-args-until-success
           'magit-section-highlight-hook section selection))
        (--each magit-section-unhighlight-sections
          (run-hook-with-args-until-success
           'magit-section-unhighlight-hook it selection))
        (restore-buffer-modified-p nil)
        (unless (eq magit-section-highlighted-section section)
          (setq magit-section-highlighted-section
                (unless (magit-section-hidden section) section))))
      (setq deactivate-mark nil))))
#+end_src

This function doesn't move point, but it somehow interferes with point
adjustment in another way.


Michael