Excessive redraw of overlapping/overlapped rows?

Excessive redraw of overlapping/overlapped rows?

[YAMAMOTO Mitsuharu]

In window_update (dispnew.c), overlapping/overlapped rows are redrawn
if `changed_p' is non-zero.

	  if (changed_p && rif->fix_overlapping_area)
	    {
	      redraw_overlapped_rows (w, yb);
	      redraw_overlapping_rows (w, yb);
	    }

but this variable is set when either the mode line or the header line
is updated.

      if (mode_line_row->mode_line_p && mode_line_row->enabled_p)
	{
	  mode_line_row->y = yb;
	  update_window_line (w, MATRIX_ROW_VPOS (mode_line_row,
						  desired_matrix),
			      &mouse_face_overwritten_p);
	  changed_p = 1;
	}

So when a user moves the cursor vertically and the line number
displayed in the mode line is changed, the whole
overlapping/overlapped rows get redrawn.  It significantly degrades
performance of cursor movement for such an environment that text
drawing is not so fast.

Is it OK to not set `changed_p' for the case of mode/header line
updates?

				     YAMAMOTO Mitsuharu
				mituharu@math.s.chiba-u.ac.jp

Re: Excessive redraw of overlapping/overlapped rows?

[Kim F. Storm]

YAMAMOTO Mitsuharu <mituharu@math.s.chiba-u.ac.jp> writes:

> Is it OK to not set `changed_p' for the case of mode/header line
> updates?

I don't know for sure , but it seems ok.

-- 
Kim F. Storm <storm@cua.dk> http://www.cua.dk

Re: Excessive redraw of overlapping/overlapped rows?

[David Kastrup]

storm@cua.dk (Kim F. Storm) writes:

> YAMAMOTO Mitsuharu <mituharu@math.s.chiba-u.ac.jp> writes:
>
>> Is it OK to not set `changed_p' for the case of mode/header line
>> updates?
>
> I don't know for sure , but it seems ok.

If mode line updates change the height of the mode line (because of
descenders, or images or stuff like that), it is conceivable that
something happens with the display, though likely not on all lines at
once.  Even for line number updates, if you use line numbers with a
non-constant height (old-style numerals).

So it seems like it would at least appear prudent to check whether the
mode line dimensions changed or something like that, or check just
what lines the mode line might overlap.

-- 
David Kastrup, Kriemhildstr. 15, 44793 Bochum

Re: Excessive redraw of overlapping/overlapped rows?

[YAMAMOTO Mitsuharu]

>>>>> On Mon, 26 Sep 2005 16:26:47 +0200, storm@cua.dk (Kim F. Storm) said:

> YAMAMOTO Mitsuharu <mituharu@math.s.chiba-u.ac.jp> writes:
>> Is it OK to not set `changed_p' for the case of mode/header line
>> updates?

> I don't know for sure , but it seems ok.

Thanks.  I'll install the change if I (or someone who tries it) don't
find any problems for a few days.

				     YAMAMOTO Mitsuharu
				mituharu@math.s.chiba-u.ac.jp

Index: src/dispnew.c
===================================================================
RCS file: /cvsroot/emacs/emacs/src/dispnew.c,v
retrieving revision 1.356
diff -c -r1.356 dispnew.c
*** src/dispnew.c	1 Sep 2005 14:16:19 -0000	1.356
--- src/dispnew.c	26 Sep 2005 07:34:48 -0000
***************
*** 4143,4149 ****
  	  update_window_line (w, MATRIX_ROW_VPOS (mode_line_row,
  						  desired_matrix),
  			      &mouse_face_overwritten_p);
- 	  changed_p = 1;
  	}
  
        /* Find first enabled row.  Optimizations in redisplay_internal
--- 4143,4148 ----
***************
*** 4213,4219 ****
  	{
  	  header_line_row->y = 0;
  	  update_window_line (w, 0, &mouse_face_overwritten_p);
- 	  changed_p = 1;
  	}
  
        /* Fix the appearance of overlapping/overlapped rows.  */

Re: Excessive redraw of overlapping/overlapped rows?

[YAMAMOTO Mitsuharu]

>>>>> On Mon, 26 Sep 2005 21:19:32 +0200, David Kastrup <dak@gnu.org> said:

> If mode line updates change the height of the mode line (because of
> descenders, or images or stuff like that), it is conceivable that
> something happens with the display, though likely not on all lines
> at once.  Even for line number updates, if you use line numbers with
> a non-constant height (old-style numerals).

So the question would be: can't other parts that set `changed_p' in
update_window() detect such cases?  I'm not certain about this
question, but if not, then additional conditions would be needed as
you said.

				     YAMAMOTO Mitsuharu
				mituharu@math.s.chiba-u.ac.jp

Re: Excessive redraw of overlapping/overlapped rows?

[YAMAMOTO Mitsuharu]

>>>>> On Mon, 26 Sep 2005 21:19:32 +0200, David Kastrup <dak@gnu.org> said:

> storm@cua.dk (Kim F. Storm) writes:
>> YAMAMOTO Mitsuharu <mituharu@math.s.chiba-u.ac.jp> writes:
>> 
>>> Is it OK to not set `changed_p' for the case of mode/header line
>>> updates?
>> 
>> I don't know for sure , but it seems ok.

> If mode line updates change the height of the mode line (because of
> descenders, or images or stuff like that), it is conceivable that
> something happens with the display, though likely not on all lines
> at once.  Even for line number updates, if you use line numbers with
> a non-constant height (old-style numerals).

> So it seems like it would at least appear prudent to check whether
> the mode line dimensions changed or something like that, or check
> just what lines the mode line might overlap.

You mean the change like below?  But I'm still not certain whether
this kind of mode/header line changes may affect overlapping in other
part of the text area.

				     YAMAMOTO Mitsuharu
				mituharu@math.s.chiba-u.ac.jp

*** dispnew.c.~1.360.~	Thu Feb 23 18:08:35 2006
--- dispnew.c	Thu Feb 23 19:17:01 2006
***************
*** 4128,4135 ****
    if (force_p || !input_pending || !NILP (do_mouse_tracking))
      {
        struct glyph_row *row, *end;
!       struct glyph_row *mode_line_row;
!       struct glyph_row *header_line_row;
        int yb, changed_p = 0, mouse_face_overwritten_p = 0, n_updated;
  
        rif->update_window_begin_hook (w);
--- 4128,4135 ----
    if (force_p || !input_pending || !NILP (do_mouse_tracking))
      {
        struct glyph_row *row, *end;
!       struct glyph_row *mode_line_row, *current_mode_line_row;
!       struct glyph_row *header_line_row, *current_header_line_row;
        int yb, changed_p = 0, mouse_face_overwritten_p = 0, n_updated;
  
        rif->update_window_begin_hook (w);
***************
*** 4156,4162 ****
  	  update_window_line (w, MATRIX_ROW_VPOS (mode_line_row,
  						  desired_matrix),
  			      &mouse_face_overwritten_p);
! 	  changed_p = 1;
  	}
  
        /* Find first enabled row.  Optimizations in redisplay_internal
--- 4156,4166 ----
  	  update_window_line (w, MATRIX_ROW_VPOS (mode_line_row,
  						  desired_matrix),
  			      &mouse_face_overwritten_p);
! 	  current_mode_line_row = MATRIX_MODE_LINE_ROW (w->current_matrix);
! 	  if (!(current_mode_line_row->mode_line_p
! 		&& current_mode_line_row->enabled_p
! 		&& mode_line_row->height == current_mode_line_row->height))
! 	    changed_p = 1;
  	}
  
        /* Find first enabled row.  Optimizations in redisplay_internal
***************
*** 4226,4232 ****
  	{
  	  header_line_row->y = 0;
  	  update_window_line (w, 0, &mouse_face_overwritten_p);
! 	  changed_p = 1;
  	}
  
        /* Fix the appearance of overlapping/overlapped rows.  */
--- 4230,4240 ----
  	{
  	  header_line_row->y = 0;
  	  update_window_line (w, 0, &mouse_face_overwritten_p);
! 	  current_header_line_row = w->current_matrix->rows;
! 	  if (!(current_header_line_row->mode_line_p
! 		&& current_header_line_row->enabled_p
! 		&& header_line_row->height == current_header_line_row->height))
! 	    changed_p = 1;
  	}
  
        /* Fix the appearance of overlapping/overlapped rows.  */