bug#11367: 24.0.95.1 Crash: Windows 7 using egg

[Eli Zaretskii <eliz@gnu.org>]

> Date: Sat, 28 Apr 2012 23:44:15 +0300
> From: Eli Zaretskii <eliz@gnu.org>
> Cc: 11367@debbugs.gnu.org
> 
> > From: Michael Kleehammer <michael@kleehammer.com>
> > Date: Sat, 28 Apr 2012 14:27:00 -0500
> > 
> > > I see I will need to install msysgit and debug this locally.  Last
> > > request:
> > >
> > >  (gdb) p (g-2)->object
> > 
> > $13 = 60228101
> > 
> > >  (gdb) xtype
> > 
> > Lisp_Vectorlike
> > PVEC_BUFFER
> > 
> > >  (gdb) p it3.object
> > 
> > $14 = 84618657
> > 
> > >  (gdb) xtype
> > 
> > Lisp_String
> 
> Thanks.  I've reproduced the problem, and will debug it (probably
> tomorrow; gray hair says it's unwise to try fixing redisplay after
> 11PM ;-).

I think I fixed this (revision 107922 on the emacs-24 branch).  The
diffs are below, in case you want (and are able) to try them now.

Thanks.

=== modified file 'src/ChangeLog'
--- src/ChangeLog	2012-04-24 02:58:26 +0000
+++ src/ChangeLog	2012-04-29 17:19:08 +0000
@@ -1,3 +1,10 @@
+2012-04-29  Eli Zaretskii  <eliz@gnu.org>
+
+	* xdisp.c (pos_visible_p): If already at a newline from the
+	display string before the 'while' loop, don't walk back the glyphs
+	from it3.glyph_row.  Solves assertion violation when the display
+	string begins with a newline (egg.el).  (Bug#11367)
+
 2012-04-24  Chong Yidong  <cyd@gnu.org>
 
 	* xselect.c (x_convert_selection): Initialize a pointer (Bug#11315).

=== modified file 'src/xdisp.c'
--- src/xdisp.c	2012-04-23 16:22:23 +0000
+++ src/xdisp.c	2012-04-29 17:19:08 +0000
@@ -1375,6 +1375,7 @@ pos_visible_p (struct window *w, EMACS_I
 		  Lisp_Object startpos, endpos;
 		  EMACS_INT start, end;
 		  struct it it3;
+		  int it3_moved;
 
 		  /* Find the first and the last buffer positions
 		     covered by the display string.  */
@@ -1431,6 +1432,15 @@ pos_visible_p (struct window *w, EMACS_I
 		     begins.  */
 		  start_display (&it3, w, top);
 		  move_it_to (&it3, -1, 0, top_y, -1, MOVE_TO_X | MOVE_TO_Y);
+		  /* If it3_moved stays zero after the 'while' loop
+		     below, that means we already were at a newline
+		     before the loop (e.g., the display string begins
+		     with a newline), so we don't need to (and cannot)
+		     inspect the glyphs of it3.glyph_row, because
+		     PRODUCE_GLYPHS will not produce anything for a
+		     newline, and thus it3.glyph_row stays at its
+		     stale content it got at top of the window.  */
+		  it3_moved = 0;
 		  /* Finally, advance the iterator until we hit the
 		     first display element whose character position is
 		     CHARPOS, or until the first newline from the
@@ -1442,6 +1452,7 @@ pos_visible_p (struct window *w, EMACS_I
 		      if (IT_CHARPOS (it3) == charpos
 			  || ITERATOR_AT_END_OF_LINE_P (&it3))
 			break;
+		      it3_moved = 1;
 		      set_iterator_to_next (&it3, 0);
 		    }
 		  top_x = it3.current_x - it3.pixel_width;
@@ -1452,7 +1463,8 @@ pos_visible_p (struct window *w, EMACS_I
 		     display string, move back over the glyphs
 		     produced from the string, until we find the
 		     rightmost glyph not from the string.  */
-		  if (IT_CHARPOS (it3) != charpos && EQ (it3.object, string))
+		  if (it3_moved
+		      && IT_CHARPOS (it3) != charpos && EQ (it3.object, string))
 		    {
 		      struct glyph *g = it3.glyph_row->glyphs[TEXT_AREA]
 					+ it3.glyph_row->used[TEXT_AREA];