[Bug 37078] New: crashes if connecting fails just before it succeeds

[bugzilla-daemon at freedesktop.org]

https://bugs.freedesktop.org/show_bug.cgi?id=37078

           Summary: crashes if connecting fails just before it succeeds
           Product: Telepathy
           Version: 0.13
          Platform: Other
        OS/Version: All
            Status: ASSIGNED
          Severity: normal
          Priority: medium
         Component: gabble
        AssignedTo: simon.mcvittie at collabora.co.uk
        ReportedBy: simon.mcvittie at collabora.co.uk
         QAContact: telepathy-bugs at lists.freedesktop.org


We've seen some Gabble crash-dumps with these criticals logged (in the
configuration where they were reported, they're non-fatal):

May  5 13:04:07 (none) telepathy-gabble[23476]: GLIB CRITICAL ** tp-glib -
tp_base_connection_add_interfaces: assertion `self->status !=
TP_CONNECTION_STATUS_DISCONNECTED' failed
May  5 13:04:07 (none) telepathy-gabble[23476]: GLIB CRITICAL ** tp-glib -
tp_base_connection_change_status: assertion `prev_status ==
TP_CONNECTION_STATUS_CONNECTING' failed

and then an abort caused by this assertion failure:

"(base->status == TP_CONNECTION_STATUS_DISCONNECTED) ||
(base->status == TP_INTERNAL_CONNECTION_STATUS_NEW)"

The reason that assertion should be OK is that GabbleConnectionManager holds a
ref to every GabbleConnection, from before the time it moves to CONNECTED until
after the time it moves to DISCONNECTED. The life-cycle is NEW -> CONNECTING ->
CONNECTED -> DISCONNECTED, except that you can also become DISCONNECTED from
any earlier state.

It looks as though the two criticals I quoted can only happen if:

* set_status_to_connected is called when we already DISCONNECTED
* which requires decrement_waiting_connected when we already DISCONNECTED

s_s_t_c should bail out early if we already DISCONNECTED; that's a Gabble bug.

Meanwhile, telepathy-glib's tp_base_connection_change_status() can critical()
but already have had its side-effect of changing the status to CONNECTED:
that's a telepathy-glib bug.

The backtrace contains GSimpleAsyncResult machinery, so I think what's happened
here is:

* bare_jid_disco_cb succeeds, we've disco'ed ourselves
* conn_presence_set_initial_presence_async succeeds, and schedules a call
  to its callback in an idle; we are now ready to be CONNECTED, but because
  of GAsyncResult calling conventions, we won't call the callback right now
* someone calls Disconnect(), and we do so, synchronously; we are now
  DISCONNECTED
* the idle goes off and connection_initial_presence_cb is called

-- 
Configure bugmail: https://bugs.freedesktop.org/userprefs.cgi?tab=email
------- You are receiving this mail because: -------
You are the QA contact for the bug.