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

[Eli Zaretskii <eliz@gnu.org>]

> Date: Sun, 28 Oct 2012 12:00:49 -0700
> From: Ami Fischman <ami@fischman.org>
> Cc: 12745@debbugs.gnu.org
> 
> > Thanks.  But I'd like to see the report from your normal session,
> > invoked just as you invoke those that crash.
> >
> 
> Of course now I don't have such a session b/c I'm running w/ git HEAD :)

If it loads the same .emacs, that is good enough.

> So this seems to say that there's at least one overlay string at
> > buffer position 1295.  Is that reasonable?  What was the current
> > buffer when this crashed?  You can find that out by typing this at GDB
> > prompt:
> >   (gdb) pp current_buffer->name_
> >
> 
> (gdb) pp current_buffer->name_
> Cannot access memory at address 0x8b6a00

How about this:

  (gdb) p current_buffer->name_
  (gdb) xtype

(Note: "p", not "pp".)

If the last command says it's a Lisp string, display the contents of
'struct Lisp_String' whose address it shows.

> >   (gdb) p current_buffer->text->beg[1200]@100
> >
> 
>  (gdb) p current_buffer->text->beg[1200]@100
> $1 = "num to avoid later static_cast in\n// PluginInstance.\nenum
> MediaKeyError {\n  kUnknownError = 1,\n  kCl"
> which tells me the current buffer was an edited version of
> http://src.chromium.org/viewvc/chrome/trunk/src/webkit/media/crypto/ppapi/cdm_wrapper.cc?view=markup(which

Did that buffer have any minor mode or some other optional feature
turned on, in addition to C++ Mode?

> FWIW, there's nothing non-7-bit-ascii in this file, and nothing that
> should have triggered any bidi-specific logic.  It's just a cc-mode
> C++ file.

This crash is not about bidi.  It is about handling overlays.  That it
barfs inside a function whose name begins with "bidi" is just sheer
luck--or lack thereof.

> Possibly interestingly, if I print p current_buffer->text->beg[0]@100000 to
> emit the entire buffer, I see this text starting at char 1675:
> http://go", '\000' <repeats 2000 times>, "/b
> Those 2000 NULs are definitely out of place (the URL should have started
> with http://go/b) but I don't know if that's a debugging artifact, or what.

This could be the gap, you should see its position and size like this:

  (gdb) p current_buffer->text->gpt
  (gdb) p current_buffer->text->gap_size

> If I load the modified buffer into my HEAD session (overlays-at 1295)
> returns nil.

That's what I suspected, this is the reason for the abort: Emacs
thinks there's an overlay string at that position, when there is none,
actually.

> > (gdb) frame 6
> 
> > #6  0x0000000000447aa1 in pop_it (it=0x7fff2251f1e0) at xdisp.c:5769
> 
> > 5769          bidi_pop_it (&it->bidi_it);
> 
> > (gdb) pgrowx it->glyph_row
> You can't do that without a process to debug.

So are you debugging a core dump?