whoopsie test failure since glib2.0 2.41.2-1 uploaded

Here we can see the test passing in a trusty-i386 chroot. (I've added some g_message calls to the test_monitor.c and monitor.c.)

/whoopsie/callback-triggered-once: ** Message: g_signal_connecting.
** Message: Callback registered.
** Message: crash file created: /tmp/whoopsie-test-7SOTNX/fake.crash
** Message: upload file work done: /tmp/whoopsie-test-7SOTNX/fake.upload
** Message: In changed_event.
** Message: In changed_event.
** Message: In changed_event.
** Message: In test_callback.
** Message: In changed_event.
** Message: In changed_event.
** Message: In test_callback.
** Message: Monitoring successful.
OK
(trusty-i386)root@impulse:/home/bdmurray/source-trees/whoopsie/upstream/src/tests#

And here we can see the test failing in a utopic-i386 chroot.

/whoopsie/callback-triggered-once: ** Message: g_signal_connecting.
** Message: Callback registered.
** Message: crash file created: /tmp/whoopsie-test-SLNTNX/fake.crash
** Message: upload file work done: /tmp/whoopsie-test-SLNTNX/fake.upload
** Message: Monitoring successful.

** (./test_monitor:9220): WARNING **: Expected file /tmp/whoopsie-test-SLNTNX/fake.uploaded doesn't exist.
Trace/breakpoint trap (core dumped)
(utopic-i386)root@impulse:/home/bdmurray/source-trees/whoopsie/upstream/src/tests#

Seb, can I ask that you (or someone on desktop) look at this, please? The whoopsie tests clearly need to be made autopkgtests, so that they're run at the right moment for catching glib2.0 regressions, but as that didn't happen, it's rather urgent now to figure out the cause of this problem.

Ryan said he's going to have a look, thanks!

Posting on behalf of Ryan, who cannot log in right now.

He found out that the problem is in the test case itself. It uses g_main_loop_quit() as a source callback from test_callback_never_triggered. Source callbacks are supposed to return bool for if they get issued again or not. Apparently on amd64 we randomly end up with 0 in the return value register so the source function gets unregistered. On x386 it's random junk, so the callback gets run again, with the already freed mainloop.

Also, g_timeout_add_seconds (0.5) is bogus - the first argument is a uint. Please change that to g_timeout_add(500).

This bug was fixed in the package whoopsie - 0.2.39

---------------
whoopsie (0.2.39) utopic; urgency=medium

  * Use G_SOURCE_CONTINUE and G_SOURCE_REMOVE for GSourceFunc return values,
    for clarity
  * tests/monitor: Call g_main_loop_quit from a callback function which
    removes the source instead of directly. GSourceFuncs need to return a
    boolean to say if they should be executed again or not. (LP: #1381804)
 -- Iain Lane <email address hidden> Thu, 16 Oct 2014 10:32:22 +0100