Sunday, August 30, 2015

TDD - It'll only hurt for a little while ;)

TDD is a very useful tool. Done well, it offers assurance that your code does what it is supposed to do, provides a set of baked in regression tests, allows you to refactor code without worrying that you will break something in the process and acts as a sort of living design document.

It is not, however, a silver bullet. If done poorly, it can cause code to become rigid. This generally happens when people test the implementation details rather than the behavior. In a language like Python, where there are no truly private methods, it can be even easier to test implementation rather than behavior if you aren't cautious. I'll also say that it can be really difficult to effectively test Android code because TDD isn't a first class citizen in that ecosystem.

In my opinion, the possible downsides of TDD are generally far outweighed by the bonuses – especially since the downsides are usually able to be mitigated by exercising some self-discipline.

That's not to say that every piece of code you ever write has to be done with TDD. There are some small, usually one-liner style programs, that I write which I don't test drive. If I'm going to be using something once and tossing it, provided that there isn't really a danger of permanent damage if the code doesn't do what I want the first time, I may decide that it isn't worth the effort to set up a testing harness.

Now that we've got my opinions on the topic out of the way, I have to say that I don't get the hate that some developers have for TDD. I can understand a non-technical manager not understanding how writing tests (more code) isn't a drawback, but people who write software for a living are another story.

If I had to guess, I'd say it was probably some combination of fear of the unknown, a previous bad experience (whether because of lackluster tools, lack of guidance, or inexperience), or posturing on the part of the developer (after all, we are often not known for a lack of ego).

There are occasionally reasons to not test something (few and far between), but as a general rule, having the tests will only help you now as well as helping future you when you have to dig back into the code in 6 months. Not wanting all of the tools that could help you is just crazy.

This whole post was inspired by a tweet that a friend of mine re-tweeted (and its associated tweets), so let's take a little time to critique the conversation.

Yes. Because it works. I've already covered that above, so I'm not going to repeat myself.

If you are working in a shop that doesn't do test driving (I'd be tempted to ask why not, but just for the sake of argument), then yes, a person who doesn't do TDD will be fine. If you are in a shop that does TDD and the person is willing and able to learn TDD, then consider helping them get up to speed. If they refuse, then they're probably not a good fit for your team/company. Sorry, but that's the way it is.

If your team all does TDD, but one person refuses to do so, you lose a lot of the benefits. The baked in set of regression tests is no longer complete. The living documentation is no longer complete. Refactoring becomes trickier. Bugs can become harder to track down. These are problems.

Then you're probably doing TDD wrong. In all likelihood, you are testing implementation instead of behavior. This is actually a pretty common mistake for people who aren't used to TDD and it takes some experience to get over. Heck, even people who have been doing it for a while can slip up sometimes. None of us are perfect. That doesn't change the fact that you're probably using the tool wrong.

You don't TDD with Integration Tests, or at least not only with integration tests. You use Unit Tests to test functions (given input, check output for example). Integration Tests are used to test a series of Units (kind of like checking to see that a bunch of legos you put together actually build a wall).

Then you're looking at it wrong. Each unit test should be a condition that your function should fulfil. You only test the criteria that your code is supposed to fulfill. That's not to say that you can't spike out something you're not sure of (especially if you're working with a new language), but before that code actually makes it into your codebase, it needs to be under test, and that test needs to fail first.

I can sort of get behind this for one very simple reason – some things are just ridiculously difficult (if not damned near impossible) to test. These cases don't come up often, but they do occasionally come up. Other than that, test your code. For everyone's sake.

I disagree. If the entire team (or company) does TDD, then yes, it is a valid requirement for the job. If you refuse to use TDD, then you don't have a place on the team for reasons that I've already outlined. In fact, I'm glad more places are making it a requirement (provided that their people write good tests).

The sad thing is that I hear these sorts of arguments a lot. Thankfully, however, I seem to be hearing them less lately. I honestly welcome constructive criticism of practices because it helps people grow. You need to be prepared to have others respond to that criticism though.

Just complaining about a methodology you don't like without valid reasons doesn't help anyone (and that's exactly what that last tweet was).

Current mood: Calm
Current music: INXS – The Stairs