Reset the socket default timeout in test setup harness

[Christian Reis]

We currently had PQM reject 3 days of submissions because a single test set the socket default timeout to 1 (and not resetting it to its original value). The reason this causes the tests to fail is because the Librarian daemon does a lot of connecting and querying, and with a client timeout of 1s it's simply not enough time for the Librarian to respond.

The test was doc/bugtracker.txt, and it issued socket.setdefaulttimeout(1)

That the librarian daemon takes too much time reconnecting is bug #140817. But the test suite should kick back the socket timeout to the default value.

[Francis J. Lacoste (flacoste)]

I don't think we need to add special support for reverting possible changes to that global.

Today, there are only two tests in RF that call setdefaulttimeout(): one in bzrlib and another is a unit test. This means that this new support wouldn't even be useful for these two tests. (bzrlib tests use their own infrastructure and unit tests don't use the layers system).

[Christian Reis (kiko)]

I proposed it to avoid the problem reoccuring, given the severity of the
outage and stress that ensued because of the modified value. I don't
care so much as long as it never happens again.

[Francis J. Lacoste (flacoste)]

I think I'll simply add an assertion in BaseLayer.testTearDown() that asserts
that the default timeout (None) is in effect. That way we will get a
complaint if a test forget to clean-up after itself.

[Christian Reis (kiko)]

That idea is much better than mine.

[Francis J. Lacoste (flacoste)]

Fixed in RF 4969.