It looks like #198 introduced a regression that should not have made it into master/release. Better testing would have caught this.
Use pytest and add a minimum coverage requirement. Also add unit tests for tricky argument parsing, such as what was brought up in #272.
