[MERGE][1.2?][Bug #137823] Don't blindly raise UnavailableFeature during test setup

[Vincent Ladeuil]

>>>>> "martin" == Martin Pool <mbp at canonical.com> writes:

    martin> Martin Pool has voted approve.
    martin> Status is now: Resubmit
    martin> Comment:
    martin> The fix looks correct and I think this is reasonable for 1.2.

    martin> Robert, Vila did have a note about the change of logging.

    martin> I wonder if we should specially distinguish tests
    martin> that failed to setup,

I think they are, it's an *error* to fail to setup a test.

    martin> or even stop immediately if it happens.

selftest --one seems enough.

    martin> setup/teardown is such an awful api anyhow.

There are many cases where the tearDown must be called before
another test can be run, so we can't replace setUp by __init__
and wait for the test to be garbage collected. Plus, at least for
parametrization, we want to modify the test *before* setUp is
called.

        Vincent