lp.testing.layers.YUITestLayer tests fail in various ways without X

[Martin Packman]

Unlike lp.testing.layers.YUIAppServerLayer the layer setup doesn't fail, instead the tests themselves fail, then hang the test run. Failing during layer setup would be preferable. Not depending on X would be ideal.

First doctest fails with:

    Error in test lib/lp/app/javascript/inlinehelp/tests/test_inlinehelp.html
    Traceback (most recent call last):
      File "/home/ubuntu/launchpad/lp-branches/devel/lib/lp/testing/__init__.py", line 957, in setUp
        import html5browser
      ...
    RuntimeError: Gtk couldn't be initialized

Then following ones with:

    Error in test lib/lp/app/javascript/indicator/tests/test_indicator.html
    Traceback (most recent call last):
      File "/home/ubuntu/launchpad/lp-branches/devel/lib/lp/testing/__init__.py", line 962, in setUp
        incremental_timeout=self.incremental_timeout)
      File "/home/ubuntu/launchpad/lp-sourcedeps/eggs/html5browser-0.0.9-py2.7.egg/html5browser/__init__.py", line 103, in load_page timeout, initial_timeout, incremental_timeout)
      File "/home/ubuntu/launchpad/lp-sourcedeps/eggs/html5browser-0.0.9-py2.7.egg/html5browser/__init__.py", line 155, in _setup_listening_operation
        self._create_window()
      File "/home/ubuntu/launchpad/lp-sourcedeps/eggs/html5browser-0.0.9-py2.7.egg/html5browser/__init__.py", line 174, in _create_window
        self.browser_window = Gtk.Window()
      File "/usr/lib/python2.7/dist-packages/gi/overrides/Gtk.py", line 391, in __init__
        Gtk.Window.__init__(self, type=type, **kwargs)
    AttributeError: 'NoneType' object has no attribute 'Window'

Then the layer subprocess needs killing after it hangs:

     lib/lp/bugs/javascript/tests/test_buglisting_utils.html
    **********************************************************************
    Could not communicate with subprocess
    **********************************************************************

That's inside the call:

    write(2, "\n(test:2741): Gdk-CRITICAL **: g"..., 10

Perhaps just related to writing too much to a stderr that isn't being drained.