Red Hat Bugzilla – Bug 227245
[a11y] openoffice.org segfaults on exit in at-spi atexit
Last modified: 2007-11-30 17:11:56 EST
Description of problem: Running in a terminal window one gets: /usr/lib64/openoffice.org/program/soffice: line 222: 23556 Segmentation fault which happens every time when program exits. I did not check what happens if exiting the program exits with anything but an empty writer window; the above was "open-and-exit". To see this with more detail one can try: $ sh -x /usr/lib64/openoffice.org/program/soffice -writer + '[' -z ']' + export SAL_USE_VCLPLUGIN=gtk + SAL_USE_VCLPLUGIN=gtk + export SAL_ENABLE_NATIVE_XFONTS=0 + SAL_ENABLE_NATIVE_XFONTS=0 + SAL_ENABLE_FILE_LOCKING=1 + export SAL_ENABLE_FILE_LOCKING ++ uname -s + sd_platform=Linux ++ uname -m + '[' Linux = Linux -a x86_64 = ppc ']' ++ pwd + sd_cwd=/root + '[' -h /usr/lib64/openoffice.org/program/soffice ']' ++ dirname /usr/lib64/openoffice.org/program/soffice + cd /usr/lib64/openoffice.org/program ++ pwd + sd_prog=/usr/lib64/openoffice.org/program + cd .. ++ basename /usr/lib64/openoffice.org/program/soffice + sd_binary=soffice.bin ++ pwd + sd_inst=/usr/lib64/openoffice.org + cd /root + case $sd_platform in + '[' ']' + LD_LIBRARY_PATH=/usr/lib64/openoffice.org/program + export LD_LIBRARY_PATH + for arg in '$@' + case "$arg" in + '[' -x /usr/lib64/openoffice.org/program/javaldx ']' ++ /usr/lib64/openoffice.org/program/javaldx + java_ld_library_path= + '[' '' '!=' '' ']' + OPENOFFICE_MOZILLA_FIVE_HOME=/usr/lib64/openoffice.org/program + export OPENOFFICE_MOZILLA_FIVE_HOME + unset XENVIRONMENT + for sd_arg in '${1+"$@"}' + case ${sd_arg} in + sd_binary=swriter.bin + break + '[' /usr/lib64/qt-3.3/bin:/usr/kerberos/sbin:/usr/kerberos/bin:/usr/local/sbin:/usr/local/bin:/sbin:/bin:/usr/sbin:/usr/bin:/usr/X11R6/bin:/root/bin ']' + PATH=/usr/lib64/openoffice.org/program:/usr/lib64/qt-3.3/bin:/usr/kerberos/sbin:/usr/kerberos/bin:/usr/local/sbin:/usr/local/bin:/sbin:/bin:/usr/sbin:/usr/bin:/usr/X11R6/bin:/root/bin + export PATH + /usr/lib64/openoffice.org/program/swriter.bin -writer + trap 'kill -9 $!' TERM + wait 23556 GTK Accessibility Module initialized libGL warning: 3D driver claims to not support visual 0x4b GTK Accessibility Module shutdown /usr/lib64/openoffice.org/program/soffice: line 222: 23556 Segmentation fault "$sd_prog/$sd_binary" "$@" + '[' 139 -eq 79 ']' + exit The above does not dump core even if 'ulimit' for the account sets core size to unlimited. With a script modified to run binaries under gdb this is not much more illuminating: Program received signal SIGSEGV, Segmentation fault. [Switching to Thread 46912496327472 (LWP 23841)] 0x00002aaaaaf0225e in main () from /usr/lib64/openoffice.org/program/libvclplug_gtk680lx.so (gdb) where #0 0x00002aaaaaf0225e in main () from /usr/lib64/openoffice.org/program/libvclplug_gtk680lx.so (gdb) c Continuing. Program terminated with signal SIGSEGV, Segmentation fault. The program no longer exists. (gdb) q I have no idea of a 32-bit version does the same or not. Version-Release number of selected component (if applicable): openoffice.org-2.2.0-5.1 How reproducible: always
This is triggered by an at-spi change where there are some glib thread stuff is now called in an atexit function. OOo has overridden the g_thread lock functions and OOo tried to shutdown a11y before shutting down it's gtk vcl plugin which destroys stuff which the g_thread_lock overrides requires but OOo can't cancel the execution of the registered at-spi atexit which will try and use the thread lockers. I can hack around this in the thread lockers to check to see if their requirements have disappeared when they get called after the fact in the at-spi shutdown but it's ugly. caolanm->mclasen: I suspect you'll begin seeing upstream bugreports about vanilla OOo crashing against the newer at-spis.
Created attachment 147348 [details] OOo bandaid
the atexit function in atk-bridge/bridge.c has the following comment: /* * FIXME: this may be incorrect for apps that do their own bonobo * shutdown, until we can explicitly shutdown to get the ordering * right. */ So at least there is some awareness of ordering issues. Did you file an upstream bug for this ?
logged as http://bugzilla.gnome.org/show_bug.cgi?id=404592
Should be a non-crasher in openoffice.org-2.2.0-5.2