Re: 23.0.60; SEGV in font_score()

[Kenichi Handa <handa@m17n.org>]

In article <m3k5h1cc6m.fsf@lugabout.jhcloos.org>, James Cloos <cloos@jhcloos.com> writes:

> Compiled again, this time w/o -O2.
> With -O2 entity got optimized out, so the crash was reported on line
> 1990:

>         if (xstrcasecmp (SDATA (spec_str), SDATA (entity_str)))

> but w/o -O2 it occurs a couple of lines up at:

>         Lisp_Object entity_str = SYMBOL_NAME (AREF (entity, i));

Ah.

> Since I can only trigger this randomly while reading mail with gnus, and
> since nnimap takes about an hour to start up, this is very fustrating to
> debug.

> I was able to convince the xpr macro in gdb to tell me that the first
> element of entity is "xft",

Hmmm, then it seems that at least entity is a correct
font-entity object.  So, perhaps the 2nd element (you wrote
that `i' was 2) of it was set to something other than symbol.

> but I couldn't figure out how to get to the
> next element.

I've just installed a new gdb function xfont of .gdbinit.
With that, you can do something like this:

(gdb) p entity
$1 = 147592732
(gdb) xfont
$2 = (struct font_entity *) 0x8cc1618
(gdb) p $2->props[FONT_FAMILY_INDEX]
$3 = 148609265
(gdb) xtype
Lisp_Symbol

> Handa-san: I can provide a -ggdb core dump off list.  It looks to have
> much more value than the dump I mentioned in the initial bug report.
> (It is quite large, though.  More than 64 Megs uncompressed.)

I think that kind of dump doesn't help that much because the
problem is in Lisp object.  If you still keep that gdb
session, please try above (after you get the new .gdbinit
from the CVS repository, and do "(gdb) source .gdbinit").
 
And also this will help.

(gdb) up 2
#2  0x081e64d7 in font_find_for_lface (f=0x84e5300, attrs=0x89ea6f8, spec=147460412, c=-1) at font.c:2893
(gdb) pr spec
(gdb) pr prefer

---
Kenichi Handa
handa@ni.aist.go.jp