This makes it slightly easier to debug test failures when one test
opens a repo, has a failure, and doesn't get a chance to close it for
the next test. Now, instead of getting no feedback, we at least see
test failure information.