[Pan-devel] khaley testing branch: x-face segfaults from a4824a92 (update to using cairo)

[Duncan]

Recent khaley testing branch (thru current 463d68a0, at least)
segfaults on most x-faces I've seen, at least (I believe I've
seen at least one that didn't, however.  I bisected to:

commit a4824a92dc42a471768079f29b922aae387b4ac8
Author: K. Haley <address@hidden>
Date:   Wed Aug 18 15:49:51 2010 -0600

        Replace deprecated gdk_pixmap_create_from_data
        with cairo to render x-face.

The entire system recently recompiled with
gcc-4.5.1
glibc-2.12.1-r1 (-r1 indicates a gentoo revision)
gtk+-2.20.1-r1
cairo-1.10.0-r3
2.6.36-rc8-21-ga7011a3 (linus' git tree)

This is on amd64, dual older (3-digit) amd opteron 270s (dual-core),
radeon hd4650 graphics, native xorg ati freedomware driver.

xorg-server-1.9.0.902 (second rc of 1.9.1)
xf86-video-ati-6.13.2
mesa-7.9
libdrm-2.4.22

cairo-1.10.0-r3 was merged with these USE flags (generally, configure
options) ON:
X opengl qt4 svg

... but with these masked or OFF:
aqua (masked, this isn't Apple), debug, directfb, doc, drm (Gentoo masked
except on the live-git version, due to compile problems for the
maintainer), gallium (wouldn't be active without drm anyway), openvg
(masked, this isn't embedded), static-libs.

I suspect the issue may be a cairo issue, possibly related to immature
OpenGL support on this graphics hardware/driver combo, but I'm not sure. 
It could also be a glibc issue, since 2.12.1 is fairly new (Gentoo's
package introduced on August 5, according to the changelog), and that's
where the log shows the segfault at:

segfault at 7ffffc000000 ip 00007fa49823a530 sp 00007fffffb3ad20 error 4
in libc-2.12.1.so[7fa4981bd000+17a000]

Many test cases can be found on news.gmane.org, in the gmane.discuss
newsgroup.  In particular, posts by "Adam =?utf-8?Q?Sj=C3=B8gren?="
(that's the raw name in the from header), including this one, trigger the
segfault:

Subject: Re: Podcasts in gwene ?
Date: Tue, 19 Oct 2010 15:01:52 +0200
Message-ID: <address@hidden>

By restarting pan, selecting the message but NOT viewing it, and saving it
to disk, I was able to get the headers and verify that the cached version
pan's operating on is the same (and grab the headers for posting here). 
By then editing the cached version to remove the x-face header, reloading
the group in pan, and loading the message, I was able to verify that it
loaded fine without the x-face header, so that's what it's segfaulting on.

Here's the x-face header in question, tho I believe it's fine as Adam has
been using it for some time, and when I did the bisect and loaded the post
with a good pan version, I recognized the same x-face he normally uses.

X-Face: )qY&CseJ?.:=8F#^~GcSA?F=9eu'{KAFfL1C3/A&:nE?PW\i65"ba0NS)97,
        Q(address@hidden
        
rPuR#V8I(address@hidden($[ym:`K_+]*kjvW>xH5jbgLBVFGXY:(#4P>zVBklLbdL&XxL\M)%T}3S/IS9lMJ
        ^St'=VZBR<gm`!Dj`dIpp?+$"$l_'JKDN\w-jB;fo0Qy}Tbw

There's a second guy posting there whose posts trigger segfaults as well. 
Ted Ziatanov, IIRC, tho I'm composing this message ATM so can't verify,
lest I lose the message in the segfault. =:^(

-- 
Duncan - List replies preferred.   No HTML msgs.
"Every nonfree program has a lord, a master --
and if you use the program, he is your master."  Richard Stallman