LaunchpadTestRequest should not always set features to NullFeatureController

[Gavin Panella]

In a test, the following code did not work:

    def test_a_view(self):
        set_derived_series_ui_feature_flag(self)
        with person_logged_in(self.simple_user):
            view = create_initialized_view(...)
            html = view()
        self.assertThat(...)

Because LaunchpadTestRequest overrides self.features to
NullFeatureController, effectively ignoring any flags that have been
set up in the test, and preventing more flags from being configured.

In this instance the code worked when modified:

    from lp.services.features import get_relevant_feature_controller

    def test_a_view(self):
        set_derived_series_ui_feature_flag(self)
        with person_logged_in(self.simple_user):
            view = create_initialized_view(...)
            view.request.features = get_relevant_feature_controller()
            html = view()
        self.assertThat(...)

In this instance it was the template that queried request.features, so
it was possible to poke a sensible feature controller into place
before it was needed. However, this would not have worked if the view
initialization code had looked at request.features.

I'm not aware of the rationale behind clobbering LTR.features in this
way, especially with a not very useful feature controller. We *need*
to be able to write tests for code that is activated by feature flags.

This might be a duplicate of bug 631884 depending on your point of
view.

I am marking this as High importance because trying to diagnose this
problem consumed well over 2h of developer time.

[Martin Pool (mbp)]

I think NullFeatureController was quite possibly a mistake on my part
(specifically leftover tdd scaffolding), and if someone tackles this
please think about killing it entirely.

Martin

[Launchpad QA Bot (lpqabot)]

Fixed in stable r12824 <http://bazaar.launchpad.net/~launchpad-pqm/launchpad/stable/revision/12824>.