Charlie's Corner

Pytest Pikachu

2019-10-01

I wanted to understand the whole process of how a Python package is distributed. Building a pytest plugin seemed like a way to build something small but interesting enough to publish.

You can find the source code on GitHub and the package on PyPI.

The most exciting thing for me is that anyone can install the plugin just by running

pip install pytest-pikachu

which shows how convenient Python package distribution is for people who use the packages.

After installing the plugin you can pass the --pikachu flag and, if all your tests pass, you won't be the only one looking surprised.

$ pytest --pikachu
============================= test session starts ==============================
platform linux -- Python 3.7.1, pytest-5.1.3, py-1.8.0, pluggy-0.13.0
rootdir: /home/.../pytest-pikachu, inifile: tox.ini, testpaths: tests
plugins: pikachu-0.1.0
collected 4 items

tests/test_plugin.py ....                                                [100%]

============================== 4 passed in 0.25s ===============================

⢀⣠⣾⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⠀⠀⠀⠀⣠⣤⣶⣶
⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⠀⠀⠀⢰⣿⣿⣿⣿
⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣧⣀⣀⣾⣿⣿⣿⣿
⣿⣿⣿⣿⣿⡏⠉⠛⢿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⡿⣿
⣿⣿⣿⣿⣿⣿⠀⠀⠀⠈⠛⢿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⠿⠛⠉⠁⠀⣿
⣿⣿⣿⣿⣿⣿⣧⡀⠀⠀⠀⠀⠙⠿⠿⠿⠻⠿⠿⠟⠿⠛⠉⠀⠀⠀⠀⠀⣸⣿
⣿⣿⣿⣿⣿⣿⣿⣷⣄⠀⡀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⢀⣴⣿⣿
⣿⣿⣿⣿⣿⣿⣿⣿⣿⠏⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠠⣴⣿⣿⣿⣿
⣿⣿⣿⣿⣿⣿⣿⣿⡟⠀⠀⢰⣹⡆⠀⠀⠀⠀⠀⠀⣭⣷⠀⠀⠀⠸⣿⣿⣿⣿
⣿⣿⣿⣿⣿⣿⣿⣿⠃⠀⠀⠈⠉⠀⠀⠤⠄⠀⠀⠀⠉⠁⠀⠀⠀⠀⢿⣿⣿⣿
⣿⣿⣿⣿⣿⣿⣿⣿⢾⣿⣷⠀⠀⠀⠀⡠⠤⢄⠀⠀⠀⠠⣿⣿⣷⠀⢸⣿⣿⣿
⣿⣿⣿⣿⣿⣿⣿⣿⡀⠉⠀⠀⠀⠀⠀⢄⠀⢀⠀⠀⠀⠀⠉⠉⠁⠀⠀⣿⣿⣿
⣿⣿⣿⣿⣿⣿⣿⣿⣧⠀⠀⠀⠀⠀⠀⠀⠈⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⢹⣿⣿
⣿⣿⣿⣿⣿⣿⣿⣿⣿⠃⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⢸⣿⣿

The actual logic isn't too complicated: it relies on existing hooks within the pytest framework which were well-documented. We wait until the end of the session and then print out an ASCII art Surprised Pikachu if the exit code shows the tests were successful.

The interesting bit was dipping my toes into Python packaging from the perspective of an author. I learned about how to write a valid setup.py file, complete with a pytest entrypoint, and use twine to upload the built distribution to PyPI.

It was also a good opportunity to use tox and GitHub Actions to run automated tests across a variety of Python language versions. At work I have only worked with a single version of Python at a time so this was new for me.

Overall I'm happy with how this turned out and that I learned some useful things about Python package distribution.