Test Driven Development was mind-bending when I first grappled with it. Last summer I was building a web app using Django and began to break things unintentionally when adding new features. This soon led to lots of clicking around different pages to test if stuff was still working each time I made an update.
This soon led to me thinking there must be a better way. Which brought me to Test Driven Development (TDD).
It should have just led me to writing tests, which it did. But googling whatever I googled got me down the TDD rabbit hole rather than just the “write some tests” rabbit hole. Write tests for your code before you write the code. Write tests for bugs you’ve fixed to check they stay fixed. Write tests as a kind of documentation to show what stuff is supposed to be doing. Errr…
Django was a big enough pile of abstractions as it was. Views, ORMs, mixins, serializers… Trying to add factories and fixtures into that took some getting used to. But eventually I made some progress, and now I quite enjoy running coverage reports to keep coverage close to 100%1.
Some of the main things I’ve learnt about writing tests:
- Use PyTest as much as possible rather than other testing libraries - its
assertstatements are more intuitive than Django’s own testing framework, and you can use it in any Python codebase, not just Django. It has lots of extensions and seems good at getting the job done fairly easily.
- Write tests as you go. I haven’t (yet) reached the elevated level of writing tests before I write the code to be tested, though I see why that would sometimes be useful. I do think writing tests sooner rather than later is best though, ideally as soon as you’ve got a basic version of your feature working.
Coverageto show you which code is covered by your tests, and which branches or edge cases are not. But be warned, it doesn’t tell you if the test is useful or not, only that it passes and which methods or functions it uses.
- Fixtures are great for keeping tests fairly DRY.
Freezegunis great for testing anything to do with dates and times.
- Static type checkers, like
Mypy, get more attractive in proportion to codebase complexity and size.
- Which is fun and all, but testing for the sake of it doesn’t necessarily stop bad things happening. Its very possible to write a test that covers the code you’ve just written without ensuring that only the intended behaviours happen. ↩