Bug#881301: gnome-shell: intermittent "invalid unclassed pointer in cast to 'ClutterText'" + segfault

[Simon McVittie]

Control: retitle 881301 gnome-shell: intermittent "invalid unclassed pointer in cast to 'ClutterText'" + segfault
Control: tags 881301 + moreinfo

On Thu, 09 Nov 2017 at 18:27:59 -0300, Lucas Dyer wrote:
> Sometimes gnome-shell crashes, randomly while using the computer or not.
> Also happens when the screen is locked, even without any programs running.

I'm retitling this bug report to make it more distinctive, since a segfault
in libgobject is very non-specific (passing any invalid pointer where any
GObject is expected would have the same effect).

Are you using any GNOME Shell extensions? If so, which ones?

Approximately how often does this crash occur?

We are very unlikely to be able to solve this without a backtrace. The bug
appears to be that an invalid pointer is used where a ClutterText object
is expected (most likely a use-after-free or a null pointer somewhere),
but that isn't something that can be fixed without knowing where/how it
happens.

One very convenient way to get a backtrace from infrequent crashes is to
install systemd-coredump and gdb. After a crash occurs, run "coredumpctl"
to list recent crashes. One of them should be gnome-shell. Get the process
ID from the most recent gnome-shell crash (a number up to 5 digits long)
and use a command like "coredumpctl gdb 12345" to inspect the crash dump.
At the (gdb) prompt, use the "bt" command to get a backtrace.

To make the backtrace most useful, you will need detached debug symbols for
gnome-shell, libglib-2.0, mutter and perhaps other libraries: see
<https://wiki.debian.org/HowToGetABacktrace#Installing_the_debugging_symbols>.
There is more information about gdb on the same page.

>From your syslog attachments, it looks as though a lot of other programs
exit unexpectedly as a side-effect of the GNOME Shell crash, and some of
those other programs (like dleyna-renderer) are in fact also crashing.
This is a side-effect of this bug, rather than part of its cause. In
particular, details of the dleyna-renderer crash (which you'd probably
also see listed by coredumpctl) are not useful for solving this bug,
although you could report them as a dleyna bug if you want.

Regards,
    smcv