bug#28710: 27.0.50; eassert failure in maybe_produce_line_number

[Eli Zaretskii <eliz@gnu.org>]

> From: Alex <agrambot@gmail.com>
> Cc: 28710@debbugs.gnu.org
> Date: Sun, 08 Oct 2017 15:19:53 -0600
> 
> Eli Zaretskii <eliz@gnu.org> writes:
> 
> > OK, so we need to understand the path the code takes after try_window
> > returns the value 1.  This means, after typing "finish" 3 times, type
> > "next", then continue pressing RET until redisplay_window returns.  I
> > need to see the path through the code until we exit redisplay_window
> > to understand where to put the missing call to clear_glyph_matrix.
> 
> Thread 1 "emacs" hit Hardware watchpoint 2: -location $1->desired_matrix->rows->enabled_p

Thanks.  I installed a change that should fix the problem, please try
the latest emacs-26 branch.

As to why this doesn't happen to me: for some reason, on your system,
when all the frame's windows have been redisplayed, their frame's
'redisplay' flag is set, and that causes redisplay_internal to
immediately redisplay all the windows again, see this part of your
transcript:

> 14320			  if (!f_redisplay_flag && f->redisplay)
> (gdb) 
> 14321	                    goto retry_frame;

On my system, the 'redisplay' flag stays reset, so this goto is
bypassed, and the problem doesn't happen.  If you can afford one last
effort, please re-run the recipe with a watchpoint set on the frame's
'redisplay' flag, and show the backtraces from every one of the
watchpoint's hits, then perhaps I will know next time what else to try
to reproduce such cases.

Specifically, after invoking redraw-display, which causes GDB to kick
in, do this:

  Thread 1 hit Breakpoint 3, Fredraw_display () at dispnew.c:3032
  3032      FOR_EACH_FRAME (tail, frame)
  (gdb) n
  3033        if (FRAME_VISIBLE_P (XFRAME (frame)))
  (gdb) p XFRAME(frame)
  $1 = (struct frame *) 0x1b5e380 <dumped_data+4020672>
  (gdb) p $1->redisplay
  $2 = true
  (gdb) watch -l $1->redisplay
  Hardware watchpoint 4: -location $1->redisplay
  (gdb) commands
     > bt
     > continue
     > end
  (gdb) continue

and then continue with the recipe, and show all the backtraces you get.