bug#5858: 24.0.50; bidi aborts: "can't happen" happens

bug#5858: 24.0.50; bidi aborts: "can't happen" happens

[Ari Roponen]

Hi,

the following code crashes emacs by making code in bidi.c call
abort():

  (let ((bidi-display-reordering t))
    (insert #x650 #x650)		; ARABIC KASRA
    (redisplay))


The place that calls abort() is in bidi_resolve_neutral():

#0  abort () at emacs.c:430
#1  0x0000000000490abe in bidi_resolve_neutral (bidi_it=0x7fffffff9000)
    at bidi.c:1577
#2  0x000000000049119a in bidi_level_of_next_char (bidi_it=0x7fffffff9000)
    at bidi.c:1713
...

(gdb) up
#1  0x0000000000490abe in bidi_resolve_neutral (bidi_it=0x7fffffff9000)
    at bidi.c:1577
1577			  abort ();		/* can't happen: BNs are skipped */
(gdb) l
1572			next_type = STRONG_R;
1573			saved_it.next_for_neutral.type = STRONG_R;
1574			break;
1575		      case WEAK_BN:
1576			if (!bidi_explicit_dir_char (bidi_it->ch))
1577			  abort ();		/* can't happen: BNs are skipped */
1578			/* FALLTHROUGH */
1579		      case NEUTRAL_B:
1580			/* Marched all the way to the end of this level run.
1581			   We need to use the eor type, whose information is


In GNU Emacs 24.0.50.3 (x86_64-unknown-linux-gnu, GTK+ Version 2.19.7)
 of 2010-04-07 on arirop
Windowing system distributor `Fedora Project', version 11.0.10799901

-- 
Ari Roponen

bug#5858: 24.0.50; bidi aborts: "can't happen" happens

[Eli Zaretskii]

> From: Ari Roponen <ari.roponen@gmail.com>
> Date: Thu, 08 Apr 2010 12:03:55 +0300
> Cc: 
> 
> the following code crashes emacs by making code in bidi.c call
> abort():
> 
>   (let ((bidi-display-reordering t))
>     (insert #x650 #x650)		; ARABIC KASRA
>     (redisplay))

Heh, a brain-fart from 9 years ago that came back to haunt me...

> The place that calls abort() is in bidi_resolve_neutral():
> 
> #0  abort () at emacs.c:430
> #1  0x0000000000490abe in bidi_resolve_neutral (bidi_it=0x7fffffff9000)
>     at bidi.c:1577
> #2  0x000000000049119a in bidi_level_of_next_char (bidi_it=0x7fffffff9000)
>     at bidi.c:1713
> ...
> 
> (gdb) up
> #1  0x0000000000490abe in bidi_resolve_neutral (bidi_it=0x7fffffff9000)
>     at bidi.c:1577
> 1577			  abort ();		/* can't happen: BNs are skipped */
> (gdb) l
> 1572			next_type = STRONG_R;
> 1573			saved_it.next_for_neutral.type = STRONG_R;
> 1574			break;
> 1575		      case WEAK_BN:
> 1576			if (!bidi_explicit_dir_char (bidi_it->ch))
> 1577			  abort ();		/* can't happen: BNs are skipped */
> 1578			/* FALLTHROUGH */
> 1579		      case NEUTRAL_B:
> 1580			/* Marched all the way to the end of this level run.
> 1581			   We need to use the eor type, whose information is

Yours is an optimized build, right?  Because, in an unoptimized build,
it's a different call to `abort' that is triggered:

  #1  0x011a3931 in bidi_resolve_neutral (bidi_it=0x82d2f8) at bidi.c:1600
  1600                    abort ();
  (gdb) l
  1595                        bidi_copy_it (bidi_it, &saved_it);
  1596                        return bidi_it->type;
  1597                      }
  1598                    break;
  1599                  default:
  1600                    abort ();  <<<<<<<<<<<<<<<<<
  1601                }
  1602              type = bidi_resolve_neutral_1 (saved_it.prev_for_neutral.type,

  1603                                             next_type, current_level);
  1604              saved_it.type = type;

Anyway, I think I fixed it (revno 99851); please try again after
updating your tree.  If it still doesn't work, please re-open this
bug.

Thanks for reporting this.