Some tests failed with label `failed', which is incorrect because the
completions just weren't installed. Completions weren't installed
because the commands to complete weren't installed.
Fixed so that if commands aren't available, the tests fail with label
`unsupported'.