Intermittent acceptance test failure in ClientPidTestFixture.authorizer_may_prevent_connection_of_clients

[Alan Griffiths]

This is one failure mentioned in bug 1231341

$ MIR_SERVER_MSG_PROCESSOR_REPORT=log MIR_CLIENT_RPC_REPORT=log bin/acceptance-tests --gtest_filter=ClientPidTestFixture.authorizer_may_prevent_connection_of_clients --gtest_repeat=1000 --gtest_break_on_failure
...
Repeating all tests (iteration 54) . . .

Note: Google Test filter = ClientPidTestFixture.authorizer_may_prevent_connection_of_clients
[==========] Running 1 test from 1 test case.
[----------] Global test environment set-up.
[----------] 1 test from ClientPidTestFixture
[ RUN ] ClientPidTestFixture.authorizer_may_prevent_connection_of_clients
[DD, rpc] Invocation request: id: 0 method_name: connect
[EE, rpc] Invocation failed: id: 0 method_name: connect error: Broken pipe
pure virtual method called
terminate called without an active exception
/home/alan/display_server/development-branch/tests/mir_test_framework/testing_process_manager.cpp:127: Failure
Value of: result.succeeded()
  Actual: false
Expected: true
client terminate error=process::Result(child_terminated_by_signal, signal(6), )
Segmentation fault (core dumped)

It looks like the client isn't always handling the rejection well.

[Alan Griffiths (alan-griffiths)]

This is fun!

The client crashes;
The test fails;
And a server process is left running!

[Alan Griffiths (alan-griffiths)]

The server process left running is an artefact of "--gtest_break_on_failure" which exit the test process before sending SIGTERM to the server.

There are two failure modes:

1. Attempting to write to std::cerr after the stream objects are destroyed.
2. Attempting to access boost::system::error_code::message and calling a pure virtual function.

Both look like the process is closing down without first joining any active threads. (I'm still looking for the root cause.)