bug#56393: Actually fix the long lines display bug

[Eli Zaretskii]

> Cc: gregory@heytings.org, larsi@gnus.org, 56393@debbugs.gnu.org
> Date: Thu, 07 Jul 2022 16:36:38 +0300
> From: Eli Zaretskii <eliz@gnu.org>
> 
> > modified   src/xdisp.c
> > @@ -18872,11 +18872,20 @@ set_vertical_scroll_bar (struct window *w)
> >       && NILP (echo_area_buffer[0])))
> >      {
> >        struct buffer *buf = XBUFFER (w->contents);
> > -      whole = BUF_ZV (buf) - BUF_BEGV (buf);
> > -      start = marker_position (w->start) - BUF_BEGV (buf);
> > -      /* I don't think this is guaranteed to be right.  For the
> > -    moment, we'll pretend it is.  */
> > -      end = BUF_Z (buf) - w->window_end_pos - BUF_BEGV (buf);
> > +      if (! BUFFER_AUTO_NARROWED_P (buf))
> > +   {
> > +     whole = BUF_ZV (buf) - BUF_BEGV (buf);
> > +     start = marker_position (w->start) - BUF_BEGV (buf);
> > +     /* I don't think this is guaranteed to be right.  For the
> > +        moment, we'll pretend it is.  */
> > +     end = BUF_Z (buf) - w->window_end_pos - BUF_BEGV (buf);
> > 
> > I can almost guarantee that it's not guaranteed that window_end_pos is
> > always right.  But I don't have an alternative, ATM.  Could you please
> > add a TODO or what's customary today in the comment, so it's easier to
> > find?
> 
> Yes, this should test window_end_valid before using window_end_pos.
> An alternative could be window-start point plus some estimation of the
> window's text, perhaps?

Actually, we could be more accurate: we could use move_it_to,
similarly to what pos_visible_p does when passed -1 as the position.