Before this, a failed test would keep failing because the output file
is not cleaned up and subsequent runs would refuse to overwrite it.