GNU bug report logs - #21556
25.0.50; Memory leak in emacs -Q with lucid (font cache cleared prematurely)

Previous Next

Full log

Message #14 received at 21556 <at> debbugs.gnu.org (full text, mbox):

From: Dima Kogan <dima <at> secretsauce.net>
To: Eli Zaretskii <eliz <at> gnu.org>
Cc: 21556 <at> debbugs.gnu.org
Subject: Re: bug#21556: 25.0.50; Memory leak in emacs -Q with lucid (font)
Date: Fri, 25 Sep 2015 01:13:15 -0700

Eli Zaretskii <eliz <at> gnu.org> writes:

>> From: Dima Kogan <dima <at> secretsauce.net>
>> Date: Thu, 24 Sep 2015 17:05:12 -0700
>> 
>> Main question of this bug report so far:
>> 
>>   Is there a way to put a watchpoint on a lisp object?
>
> Theoretically, yes.

Hi. I just came back to this, figured out how to (semi-manually) put in
a watchpoint, and this watchpoint told me what's resetting the font
cache: the garbage collector. The backtrace in question:

    0x00000000005d2172 in compact_font_cache_entry (entry=24121763) at alloc.c:5313
    #0  0x00000000005d2172 in compact_font_cache_entry (entry=24121763) at alloc.c:5313
    #1  0x00000000005d221b in compact_font_caches () at alloc.c:5339
    #2  0x00000000005d2742 in garbage_collect_1 (end=0x7ffcdb166830) at alloc.c:5515
    #3  0x00000000005d2e1d in Fgarbage_collect () at alloc.c:5720
    #4  0x000000000054eb21 in maybe_gc () at lisp.h:4515
    #5  0x00000000005f638c in Ffuncall (nargs=3, args=0x7ffcdb166988) at eval.c:2584

    Lisp Backtrace:
    "Automatic GC" (0x0)
    "map-keymap" (0xdb166990)
    "keymap-canonicalize" (0xdb166f38)
    "x-create-frame" (0xdb1678a0)
    "x-create-frame-with-faces" (0xdb167dd8)
    0x12b9d80 PVEC_COMPILED
    "apply" (0xdb168450)
    "frame-creation-function" (0xdb1689f0)
    "make-frame" (0xdb168f40)
    "make-frame-on-display" (0xdb1694a8)
    "server-create-window-system-frame" (0xdb169a78)
    "server-process-filter" (0xdb169ff8)
    Hardware watchpoint 24: ((struct Lisp_Cons *) 0x1701190)->u.cdr


As a test I asked gdb to never drop fonts in the gc by acting as if
drop==0 in compact_font_cache_entry(), and that made the leak in this
bug go away. Clearly that's not a fix, but it's evidence that the
understanding of the problem is correct, so progress is being made.

Previous Next

GNU bug tracking system
Copyright (C) 1999 Darren O. Benham, 1997,2003 nCipher Corporation Ltd, 1994-97 Ian Jackson.