bug#12745: crash in bidi_pop_it during (idle) redisplay

[Eli Zaretskii <eliz@gnu.org>]

> Date: Wed, 31 Oct 2012 17:46:48 +0200
> From: Eli Zaretskii <eliz@gnu.org>
> Cc: alptekin.aker@gmail.com, 12745@debbugs.gnu.org
> 
> I will meanwhile fix init_from_display_pos, because it's certainly
> looks broken, and could well be the root cause for this.

Now done in revision 110767 on the emacs-24 branch.  Will appear on
the trunk on the next merge.  The patch is below, FYI.

=== modified file 'src/xdisp.c'
--- src/xdisp.c	2012-10-20 21:30:51 +0000
+++ src/xdisp.c	2012-11-03 09:25:34 +0000
@@ -928,6 +928,7 @@ static enum move_it_result
        move_it_in_display_line_to (struct it *, ptrdiff_t, int,
 				   enum move_operation_enum);
 void move_it_vertically_backward (struct it *, int);
+static void get_visually_first_element (struct it *);
 static void init_to_row_start (struct it *, struct window *,
                                struct glyph_row *);
 static int init_to_row_end (struct it *, struct window *,
@@ -3113,6 +3114,40 @@ init_from_display_pos (struct it *it, st
       eassert (STRINGP (it->string));
       it->current.string_pos = pos->string_pos;
       it->method = GET_FROM_STRING;
+      it->end_charpos = SCHARS (it->string);
+      /* Set up the bidi iterator for this overlay string.  */
+      if (it->bidi_p)
+	{
+	  it->bidi_it.string.lstring = it->string;
+	  it->bidi_it.string.s = NULL;
+	  it->bidi_it.string.schars = SCHARS (it->string);
+	  it->bidi_it.string.bufpos = it->overlay_strings_charpos;
+	  it->bidi_it.string.from_disp_str = it->string_from_display_prop_p;
+	  it->bidi_it.string.unibyte = !it->multibyte_p;
+	  bidi_init_it (IT_STRING_CHARPOS (*it), IT_STRING_BYTEPOS (*it),
+			FRAME_WINDOW_P (it->f), &it->bidi_it);
+
+	  /* Synchronize the state of the bidi iterator with
+	     pos->string_pos.  For any string position other than
+	     zero, this will be done automagically when we resume
+	     iteration over the string and get_visually_first_element
+	     is called.  But if string_pos is zero, and the string is
+	     to be reordered for display, we need to resync manually,
+	     since it could be that the iteration state recorded in
+	     pos ended at string_pos of 0 moving backwards in string.  */
+	  if (CHARPOS (pos->string_pos) == 0)
+	    {
+	      get_visually_first_element (it);
+	      if (IT_STRING_CHARPOS (*it) != 0)
+		do {
+		  /* Paranoia.  */
+		  eassert (it->bidi_it.charpos < it->bidi_it.string.schars);
+		  bidi_move_to_visually_next (&it->bidi_it);
+		} while (it->bidi_it.charpos != 0);
+	    }
+	  eassert (IT_STRING_CHARPOS (*it) == it->bidi_it.charpos
+		   && IT_STRING_BYTEPOS (*it) == it->bidi_it.bytepos);
+	}
     }
 
   if (CHARPOS (pos->string_pos) >= 0)
@@ -3122,6 +3157,9 @@ init_from_display_pos (struct it *it, st
 	 IT should already be filled with that string.  */
       it->current.string_pos = pos->string_pos;
       eassert (STRINGP (it->string));
+      if (it->bidi_p)
+	bidi_init_it (IT_STRING_CHARPOS (*it), IT_STRING_BYTEPOS (*it),
+		      FRAME_WINDOW_P (it->f), &it->bidi_it);
     }
 
   /* Restore position in display vector translations, control