bug#6593: emacs-23.2 crashes in get_glyph_face_and_encoding with a null face returned by FACE_FROM_ID(0)

[Eli Zaretskii <eliz@gnu.org>]

> From: John Lumby <johnlumby@hotmail.com>
> Date: Fri, 9 Jul 2010 09:54:55 -0400
> Cc: 
> 
>  paste this text into any lisp-interaction buffer e.g. *scratch*,  position cursor at end, and press Ctl-X Ctl-E
> 
> 
> (progn (modify-frame-parameters (selected-frame) '((menu-bar-lines . 0)))
> (set-face-font 'modeline "-misc-fixed-medium-r-normal--13-120-75-75-c-70-iso8859-1")
>  (face-set-after-frame-default (selected-frame)) (set-face-background 'default (cdr (assoc 'background-color (frame-parameters (selected-frame))))) (set-face-background 'fringe (cdr (assoc 'background-color (frame-parameters (selected-frame))))) (set-face-foreground 'default (cdr (assoc 'foreground-color (frame-parameters (selected-frame))))) (server-start))
> 
> 
> 
> I get a crash in get_glyph_face_and_encoding. 

Thanks for your report.  It misses a few crucial details, though.

What system is that?  I couldn't reproduce the crash on my machine
(MS-Windows).

> I started emacs with something like
> emacs --name=sawlist --title=sawlist -ms "#14892b" -cr red -fn "-Misc-Fixed-bold-R-*--15-*-*-*-C-90-ISO8859-1" -bg "#E0EDED" -fg "#751503" -q -load sawlist.el
> 
> I can send my sawlist.el  if needed

Does the crash happen if you start Emacs with "emacs -Q"?  Without -Q,
your customizations on ~/.emacs also come into play.

> I obtained a core file after rebuilding to get the debugging information and here is the relevant piece
> (gdb) bt
> #0  0x00d57424 in __kernel_vsyscall ()
> #1  0x00876006 in kill () at ../sysdeps/unix/syscall-template.S:82
> #2  0x0810fe76 in fatal_error_signal (sig=11) at emacs.c:402
> #3  <signal handler called>
> #4  get_glyph_face_and_encoding (f=0x920c4f0, glyph=0x946b9a0, char2b=0xbfbeb890, two_byte_p=0xbfbeb85c) at xdisp.c:19511
> #5  0x0806bc58 in fill_glyph_string (s=0xbfbeb8b0, face_id=0, start=<value optimized out>, end=1, overlaps=0) at xdisp.c:19681
> #6  0x0806c68b in draw_glyphs (w=0x920c670, x=27, row=0x93fa530, area=TEXT_AREA, start=0, end=1, hl=DRAW_NORMAL_TEXT, overlaps=0)
>     at xdisp.c:20297

Please show the full backtrace, including the Lisp backtrace (shown if
you start GDB from the src directory).  The above only shows the top 6
stack frames.

Also, if you could rebuild without optimizations (-O0), it would help,
because the stack trace in an optimized build cannot be trusted and
too many variables have their values shown as ``optimized out'', which
doesn't really help debugging.

> This patch (below) fixes one obvious mistake (but not the cause) and also works around the problem.  This mistake is that the assert is located after the dereference of the face pointer  - should be before.   But that just changes the crash to an assert failure.   The workaround to is to try all other faces cached on the frame.    I don't know what the correct fix is for the default face being null.

Sorry, but I don't think this is the right fix.  The default face
should always be realized.  Looking up some face, any face, in the
face cache is not TRT.