11:21 p.m.
This is one of the oldest crashes on my to do list.
loading the following in a vanilla Mule XEmacs will trigger a crash.
(defun delete-and-split ()
(interactive)
(delete-other-windows)
(split-window))
(split-window-vertically)
(insert "$B$,(B")
(backward-char)
(redraw-frame)
(delete-and-split)
Fatal error: assertion failed, file /deb-huge/home/gunnar/src/XEmacs/xemacs/src/objects.h,
line 143, RECORD_TYPEP (obj, &lrecord_font_instance)
#3 0x808eb81 in assert_failed (file=0x8218c60
"/deb-huge/home/gunnar/src/XEmacs/xemacs/src/objects.h", line=143,
expr=0x8218d40 "RECORD_TYPEP (obj, &lrecord_font_instance)")
at /deb-huge/home/gunnar/src/XEmacs/xemacs/src/emacs.c:2825
#4 0x80c9db4 in error_check_font_instance (obj=136776524)
at /deb-huge/home/gunnar/src/XEmacs/xemacs/src/objects.h:143
#5 0x81c730f in x_output_string (w=0x85a8c10, dl=0x8609808, buf=0x85e7640, xpos=4,
xoffset=0, clip_start=4,
width=16, findex=0, cursor=1, cursor_start=4, cursor_width=16, cursor_height=13)
at /deb-huge/home/gunnar/src/XEmacs/xemacs/src/redisplay-x.c:883
#6 0x81c60d5 in x_output_display_block (w=0x85a8c10, dl=0x8609808, block=0, start=0,
end=1, start_pixpos=4,
cursor_start=4, cursor_width=16, cursor_height=13)
at /deb-huge/home/gunnar/src/XEmacs/xemacs/src/redisplay-x.c:390
#7 0x81799c2 in redisplay_output_display_block (w=0x85a8c10, dl=0x8609808, block=0,
start=0, end=1,
start_pixpos=4, cursor_start=4, cursor_width=16, cursor_height=13)
at /deb-huge/home/gunnar/src/XEmacs/xemacs/src/redisplay-output.c:1088
#8 0x8178a28 in output_display_line (w=0x85a8c10, cdla=0x0, ddla=0x85e2668, line=1,
force_start=4, force_end=20)
at /deb-huge/home/gunnar/src/XEmacs/xemacs/src/redisplay-output.c:751
#9 0x8179839 in redraw_cursor_in_window (w=0x85a8c10, run_end_begin_meths=1)
at /deb-huge/home/gunnar/src/XEmacs/xemacs/src/redisplay-output.c:1029
#10 0x81798be in redisplay_redraw_cursor (f=0x8577f00, run_end_begin_meths=1)
at /deb-huge/home/gunnar/src/XEmacs/xemacs/src/redisplay-output.c:1054
#11 0x81760e9 in Fforce_cursor_redisplay (frame=139951872)
at /deb-huge/home/gunnar/src/XEmacs/xemacs/src/redisplay.c:8938
#12 0x8176761 in text_cursor_visible_p_changed (specifier=137474936, w=0x85e8148,
oldval=136988916)
at /deb-huge/home/gunnar/src/XEmacs/xemacs/src/redisplay.c:9005
#13 0x81999b1 in recompute_one_cached_specifier_in_window (specifier=137474936,
w=0x85e8148)
at /deb-huge/home/gunnar/src/XEmacs/xemacs/src/specifier.c:2812
#14 0x8199b67 in recompute_all_cached_specifiers_in_window (w=0x85e8148)
at /deb-huge/home/gunnar/src/XEmacs/xemacs/src/specifier.c:2854
#15 0x81d5e1e in Fset_window_buffer (window=140411208, buffer=137541312)
at /deb-huge/home/gunnar/src/XEmacs/xemacs/src/window.c:3198
#16 0x81d6b4e in Fsplit_window (window=140151824, chsize=136988916, horflag=136988916)
at /deb-huge/home/gunnar/src/XEmacs/xemacs/src/window.c:3503
#17 0x8094ed9 in Feval (form=140010872) at
/deb-huge/home/gunnar/src/XEmacs/xemacs/src/eval.c:2975
#18 0x8090870 in Fprogn (args=140010812) at
/deb-huge/home/gunnar/src/XEmacs/xemacs/src/eval.c:767
#19 0x8096629 in funcall_lambda (fun=140010896, nargs=0, args=0xbffff048)
at /deb-huge/home/gunnar/src/XEmacs/xemacs/src/eval.c:3497
#20 0x8095522 in Feval (form=140007804) at
/deb-huge/home/gunnar/src/XEmacs/xemacs/src/eval.c:3081
#21 0x8146b51 in readevalloop (readcharfun=140386240, sourcename=139813548,
evalfun=0x80947b0 <Feval>, printflag=0)
at /deb-huge/home/gunnar/src/XEmacs/xemacs/src/lread.c:1448
[...]
#43 0x808d520 in xemacs_21_2_b23_i586_pc_linux (argc=4, argv=0xbffffc34, envp=0xbffffc48,
restart=0)
at /deb-huge/home/gunnar/src/XEmacs/xemacs/src/emacs.c:1898
#44 0x808e0ab in main (argc=4, argv=0xbffffc34, envp=0xbffffc48)
at /deb-huge/home/gunnar/src/XEmacs/xemacs/src/emacs.c:2323
The problem is that the face cache doesn't contain the font instance
for the relevant charset or to be precise that the cache has been
cleared but we do not regenerate the necessary entries.
A quick fix for the crash would be adding a call to
ensure_face_cachel_contains_charset() at the relevant place in
x_output_string like this:
Index: redisplay-x.c
===================================================================
RCS file: /usr/CVSroot/XEmacs/xemacs/src/redisplay-x.c,v
retrieving revision 1.23.2.17
diff -u -r1.23.2.17 redisplay-x.c
--- redisplay-x.c 1999/10/15 10:11:51 1.23.2.17
+++ redisplay-x.c 1999/12/13 22:58:35
＠＠ -879,10 +879,14 ＠＠
for (i = 0; i < nruns; i++)
{
- Lisp_Object font = FACE_CACHEL_FONT (cachel, runs[i].charset);
- struct Lisp_Font_Instance *fi = XFONT_INSTANCE (font);
+ Lisp_Object font;
+ struct Lisp_Font_Instance *fi;
int this_width;
int need_clipping;
+
+ ensure_face_cachel_contains_charset(cachel, window, runs[i].charset);
+ font = FACE_CACHEL_FONT (cachel, runs[i].charset);
+ fi = XFONT_INSTANCE (font);
if (EQ (font, Vthe_null_font_instance))
continue;
This works but is kindof expensive to do, as x_output_string is called
very very often :-)
Maybe our Mule and redisplay experts can try to come up with something
more clever...
Gunnar
11:52 a.m.
Gunnar Evermann <ge204(a)eng.cam.ac.uk> wrote on 13 Dec 1999:
This is one of the oldest crashes on my to do list.
loading the following in a vanilla Mule XEmacs will trigger a crash.
(defun delete-and-split ()
(interactive)
(delete-other-windows)
(split-window))
(split-window-vertically)
(insert "が")
(backward-char)
(redraw-frame)
(delete-and-split)
Fatal error: assertion failed, file
/deb-huge/home/gunnar/src/XEmacs/xemacs/src/objects.h, line 143, RECORD_TYPEP (obj,
&lrecord_font_instance)
[... rest of backtrace see Gunnar's original message,
Message-ID: <874sdmtok3.fsf(a)eng.cam.ac.uk> ...]
Gunnar offered the following patch, which I am using all the time
since then and apply to every new release of XEmacs (21.1 and 21.2):
The problem is that the face cache doesn't contain the font
instance
for the relevant charset or to be precise that the cache has been
cleared but we do not regenerate the necessary entries.
A quick fix for the crash would be adding a call to
ensure_face_cachel_contains_charset() at the relevant place in
x_output_string like this:
Index: redisplay-x.c
===================================================================
RCS file: /usr/CVSroot/XEmacs/xemacs/src/redisplay-x.c,v
retrieving revision 1.23.2.17
diff -u -r1.23.2.17 redisplay-x.c
--- redisplay-x.c 1999/10/15 10:11:51 1.23.2.17
+++ redisplay-x.c 1999/12/13 22:58:35
＠＠ -879,10 +879,14 ＠＠
for (i = 0; i < nruns; i++)
{
- Lisp_Object font = FACE_CACHEL_FONT (cachel, runs[i].charset);
- struct Lisp_Font_Instance *fi = XFONT_INSTANCE (font);
+ Lisp_Object font;
+ struct Lisp_Font_Instance *fi;
int this_width;
int need_clipping;
+
+ ensure_face_cachel_contains_charset(cachel, window, runs[i].charset);
+ font = FACE_CACHEL_FONT (cachel, runs[i].charset);
+ fi = XFONT_INSTANCE (font);
if (EQ (font, Vthe_null_font_instance))
continue;
This works but is kindof expensive to do, as x_output_string is called
very very often :-)
I don't notice any slowdown or any other negative effects.
Maybe our Mule and redisplay experts can try to come up with
something
more clever...
If nothing better can be found, then maybe Gunnars patch could be
applied to XEmacs?
For my usage pattern, Gunnars patch is essential, without it XEmacs
crashes extremely often while using the 'lookup.el' interface to
electronic dictionaries.
I can reproduce a crash using lookup.el easily by doing the following:
~$ xemacs
C-x C-y t e s t (C-x C-y runs `lookup-pattern')
C-x 1 (C-x 1 runs `delete-other-windows')
C-x C-y t e s t (C-x C-y runs `lookup-pattern')
M-x p (M-p runs `lookup-history-previous')
=> crash.
The dictionary I use to reproduce this crash with lookup.el is
"Iwanami Shoten Koujien Daiyonban"[1], a Japanese dictionary. To
reproduce the crash, a Japanese dictionary is certainly necessary.
Gunnar's elisp example code above, which reproduces the crash, uses
(insert "が"), which inserts a Japanese character.
When using 'lookup.el' wich a Japanese dictionary, similar things like
in Gunnar's example code happen all the time. When looking up a few
random words in the dictionary, splitting windows a few times, going
back and forth in the history of the looked up words, XEmacs will
certainly crash soon. That means XEmacs won't usually survive longer
than a few minutes if one is using 'lookup.el' with a Japanese
dictionary.
What do you think? Doesn't it make sense to apply Gunnar's patch? It
solves the above problems and doesn't seem to do any harm. Or does
anybody know a better solution?
Mike
Footnotes:
[1] 岩波書店 広辞苑（第四版）
--
Mike Fabian <mike.fabian(a)gmx.de>
3:23 a.m.
Mike Fabian <mike.fabian(a)gmx.de> writes:
> (defun delete-and-split ()
> (interactive)
> (delete-other-windows)
> (split-window))
> (split-window-vertically)
> (insert "が")
> (backward-char)
> (redraw-frame)
> (delete-and-split)
> Fatal error: assertion failed, file
/deb-huge/home/gunnar/src/XEmacs/xemacs/src/objects.h, line 143, RECORD_TYPEP (obj,
&lrecord_font_instance)
>
> [... rest of backtrace see Gunnar's original message,
> Message-ID: <874sdmtok3.fsf(a)eng.cam.ac.uk> ...]
Gunnar offered the following patch, which I am using all the time
since then and apply to every new release of XEmacs (21.1 and 21.2):
Sorry for late answer but my time and expertise are limited.
I think this is a right patch but I'm not 100% sure because
I don't understand redisplay.
The problem is, Fsplit_window incorrectly invalidates face
cache when splitted window has no parents. This doesn't
make sense because Fsplit_window only adds new modeline
between windows and scrolls window to make point visible.
They don't change any faces. Function and variable names
are very cryptic in these functions. I'll probably refactor
it when I have some more time.
2000-12-04 Yoshiki Hayashi <yoshiki(a)xemacs.org>
* window.c (Fsplit_window): Don't invalidate face cache.
Index: window.c
===================================================================
RCS file: /usr/CVSroot/XEmacs/xemacs/src/window.c,v
retrieving revision 1.41.2.66
diff -u -r1.41.2.66 window.c
--- window.c 2000/11/07 23:24:35 1.41.2.66
+++ window.c 2000/12/04 02:31:50
＠＠ -3649,7 +3649,12 ＠＠
|| NILP (XWINDOW (o->parent)->vchild))
{
make_dummy_parent (window);
+#if 0
+ /* #### I can't understand why you have to reset face
+ cachels here. This can cause crash so let's disable it
+ and see the difference. See redisplay-tests.el --yh */
reset_face_cachels (XWINDOW (window));
+#endif
new = o->parent;
XWINDOW (new)->vchild = window;
XFRAME (o->frame)->mirror_dirty = 1;
＠＠ -3666,7 +3671,10 ＠＠
|| NILP (XWINDOW (o->parent)->hchild))
{
make_dummy_parent (window);
+#if 0
+ /* #### See above. */
reset_face_cachels (XWINDOW (window));
+#endif
new = o->parent;
XWINDOW (new)->hchild = window;
XFRAME (o->frame)->mirror_dirty = 1;
--
Yoshiki Hayashi
8538
days inactive
8895
days old
2 comments
3 participants
participants (3)
-
Gunnar Evermann -
Mike Fabian -
Yoshiki Hayashi