Python Unit Tests

Yesterday I added unit tests and continuous integration to Bear-as-a-Service. This post briefly describes the first half of this (unit testing). Tomorrow will describe the continuous integration piece.

Unit tests

Commit #d393031 adds unit tests for (some of) the functionality in the mqtt_json subdirectory.

This uses the pytest framework. Write a file that ends in, add functions that end in _test, and use assert statements in these functions. Then run pytest from the command line to run the tests.

Note that this commit does three things: it adds the test file, adds the test runner to requirements.txt, and it adds the instructions for how to run the tests to the README. The code, the requirements, and the README (and other docs) should all be kept in sync.

The setup instructions lead to a running system”, and “the docs tell me how to use and develop the system”. Most changes to the code don’t require touching the other two; adding a package or workflow (this change does both) are exceptions.

Also note that pytest is a different test framework from the unittest module from the Python standard library, and that you may have used in the SoftDes unittest toolbox. I prefer pytest because it’s less verbose. It also automatically extracts the values to print alongside assertion failures, which is handy.


In unit testing, it’s useful to test one class or function without bringing in object from another class. unittest.mock is useful for this. It allows you to create a stub: a function call that just returns, or a class all of whose methods are stubs as well; and to temporarily replace (“patch”) the classes and functions in a module by these stubs. Python stubs are also mocks, that can return a value or have other attached behavior, and spies, that record whether a function was called so that you can write an assert against it later. In unittest.mock, these three roles are bundles together and called “mocks”; in other frameworks, you may find them provided separately.

test_config is an example of patching a mock. test_create_subscription_queue is an atypically complicated example; it’s what I needed in order to test this function, but I don’t recommend it as an example. (I may go back and refactor or comment it more later.)

Test Watchers

Most languages’ test runners have a “watch” option. This re-runs the test when a file changes. Watch mode lets you leave the test running in one pane, and see when your edits break the tests (turn the “red”) or fix them (turn them “green”).

Pytest doesn’t have a watch option. Commit #d393031 therefore also includes pytest-watch among the new package dependencies, and describes how to use in the README.