terminate called (uncaught exception) from DefaultServerConfiguration::the_host_connection()

[errors.ubuntu.com bug bridge]

The Ubuntu Error Tracker has been receiving reports about a problem regarding unity8. This problem was most recently seen with version 8.00+14.10.20140811-0ubuntu1, the problem page at https://errors.ubuntu.com/problem/97fd02aaa9a970a142df70e81f399de409d23847 contains more details.

[kevin gunn (kgunn72)]

adding mir as the trace might indicate nested lifecycle

[Daniel van Vugt (vanvugt)]

It appears to be the system settings "wizard" nested server that's crashing. You can tell by "/run/user/32011/wizard_socket".

Not quite sure how the "wizard" is constructed in the grand scheme of things.

[Daniel van Vugt (vanvugt)]

The offending exception: /usr/include/c++/4.9/bits/shared_ptr_base.h:626
appears to be constructor/memory related, as if unity8 has exhausted available memory?

      template<typename _Tp, typename _Alloc, typename... _Args>
        __shared_count(_Sp_make_shared_tag, _Tp*, const _Alloc& __a,
                       _Args&&... __args)
        : _M_pi(0)
        {
          typedef _Sp_counted_ptr_inplace<_Tp, _Alloc, _Lp> _Sp_cp_type;
          typedef typename allocator_traits<_Alloc>::template
            rebind_traits<_Sp_cp_type> _Alloc_traits;
          typename _Alloc_traits::allocator_type __a2(__a);
          _Sp_cp_type* __mem = _Alloc_traits::allocate(__a2, 1);
          __try
            {
              _Alloc_traits::construct(__a2, __mem, std::move(__a),
                    std::forward<_Args>(__args)...);
              _M_pi = __mem;
            }
          __catch(...)
            {
              _Alloc_traits::deallocate(__a2, __mem, 1);
              __throw_exception_again;
            }
        }

Either that or the constructor is throwing, which could be:

mgn::MirClientHostConnection::MirClientHostConnection
 ...
    if (!mir_connection_is_valid(mir_connection))
    {
        std::string const msg =
            "Nested Mir Platform Connection Error: " +
            std::string(mir_connection_get_error_message(mir_connection));

        BOOST_THROW_EXCEPTION(std::runtime_error(msg)); <=== the only exception possible I can see
    }

either due to a startup race or due to unity-system-compositor having crashed...?

[Daniel van Vugt (vanvugt)]

It's possible that whatever is meant to provide:
    host_socket = "/run/user/32011/wizard_socket"
is not running yet, or has crashed.

What provides "wizard_socket"?

[Alexandros Frantzis (afrantzis)]

My guess is that host_socket = "/run/user/32011/wizard_socket" is wrong. Perhaps a stray "MIR_SOCKET" environment variable. Unity8 should only be connecting to USC.

[Daniel van Vugt (vanvugt)]

Certainly seems like a spurious value of $MIR_SOCKET. So Mir itself isn't doing anything wrong *

* Except that, I would argue, occasions like this call for error messages rather than exceptions leading to bug reports. It's not something that should have made it to the Mir bug tracker really. Although it's now useful that it did.

[Alexandros Frantzis (afrantzis)]

OK, so current investigation results indicate that this is probably an issue (race?) in our start-up scripts. The start-up wizard should be setting MIR_SOCKET=/run/user/32011/wizard_socket in the upstart environment [1], and restoring the previous value when shutting down [2].

However it seems that occasionally something goes wrong at the cleanup phase and the environment variable is not restored, or unity8 manages to start before the cleanup phase is performed.

[1] https://bazaar.launchpad.net/~system-settings-touch/ubuntu-system-settings/rtm-14.09/view/head:/wizard/ubuntu-system-settings-wizard.conf
[2] https://bazaar.launchpad.net/~system-settings-touch/ubuntu-system-settings/rtm-14.09/view/head:/wizard/ubuntu-system-settings-wizard-cleanup.conf

[Michael Terry (mterry)]

In such cases, is there a crash file for the wizard? If the wizard crashes, the cleanup script isn't run... And I bet unity8 would start with the bogus MIR_SOCKET value.

But cleaning it up in the post-stop phase of the wizard would be racy with the start of unity8. So we'd maybe want to add logic to unity8's startup to check for this case somehow...

[Launchpad Janitor (janitor)]

This bug was fixed in the package ubuntu-system-settings - 0.3+14.10.20140925-0ubuntu1

---------------
ubuntu-system-settings (0.3+14.10.20140925-0ubuntu1) utopic; urgency=low

  [ Michael Terry ]
  * Add a translator comment for what HERE means. (LP: #1368838) (LP:
    #1368838)
  * Prevent a crash in unity8 when the wizard also crashes or stops
    early. (LP: #1355892) (LP: #1355892)
 -- Ubuntu daily release <email address hidden> Thu, 25 Sep 2014 00:22:52 +0000