Code, Angular, iOS and more by Aviv Ben-Yosef

Testing the Untested

| Comments

I’m quite of a testing nut. TDD gave me a big leap forward in my career. I attribute to it a major part of my ability to tackle big tasks.

I won’t go into it again, you can read some of my thoughts about it here. As the years passed I must admit that I’ve become less of a zealot about this. Yes, I write some code without tests (gasp!). Especially when I crank out prototypes for clients to get fast feedback (after all, a good feedback loop is what TDD is all about).

But, even when I don’t write tests I still do it responsibly. Once the code reaches some stability (and this usually happens after a week or two of development, not months) I add tests as I go along.

This is incredibly important. How many times did you let a spike turn into a big ball of mud that is super hard to test? Having the discipline to act on time is a big part of being a professional.

How should you go about doing this? First, start covering the parts of the code that break the most or that you find hard/scary to change. This means we first add the tests with the biggest bang for the buck.

I tend to add test cases almost as if I were writing the whole thing from scratch in TDD. I’ll write many small test cases and they give me the confidence that everything is in order. I might delete some of these tests later, but this isn’t waste! Once you’re skilled in testing, writing a few more tests is cheap and worth it – even if you delete them two hours later.

An important thing here is to see your damn tests fail. The TDD cycle is red-green-refactor. If you haven’t seen a test fail, how can you tell it’s even working? You need to test your tests.

Yes, you already wrote the code, and so it makes sense that the test is passing. But the solution is stupidly simple: break the code. Find the conditional or whatever that this specific test is checking, change it so it’s invalid and see the test fail with your own eyes. That’s the only way to make sure that it works, checks what you meant it to check, and that the error it produces is decipherable.

Don’t test hard, test well :)