The tests compare the command output (including of error cases) with the
expected output and exit code.
Not all tests are executed, since some expect to have a known good bc and
dc binary installed and compare results of large amounts of generated data
being processed by both versions to test for regressions.
