16 October 2012

We do not trust our unit tests

There is a simple rule, why we write unit test or better why we should. Unit tests give us the required trust to refactor our code. Changing code is the reason, why we test it. But many developers do not trust their unit tests. They expect a wrong kind of trust, that unit test can not supply. Time to change how we think about quality and changing code.

The missing trust in units

The trust developers want is a large one. They want to trust their application as a whole. They want to know changing a line of code breaks the application. There is a good reason for this. If you write a method in a class that is used in many places everywhere in your code and you change a line in this method, you can break many parts in your software.

The solution is to do tests on the highest reasonable level. In the best case this level is the UI of your software. But UI tests are also the slowest solution. They require a lot of things. You should run them on all supported platforms or browsers. You should run them with a real database and all these other services, that will be involved in your production environment. All these things must be done for just changing one line of code. You only trust in this probably simple change, when the whole application is set up and its UI still does what it should.

As you might guess there is a better solution. Unit tests should run very fast. You should run them as part of your development process many times in a hour. But they simply test a simple method or class. How can we trust them? How can we prove the application will not be broken by changing code?

Write testable code

Many people try to tell us TDD results in testable code. But this is not true. You first need to understand what testable code really is. Testable code is written in a way, that allows you to trust your unit tests. This includes testable code is easy to test, but it means a lot more. Testable code is build of small units with a well defined API. The API is defined by the unit tests and the tests prove, the API works as expected.

A simple example. You have got a method, that is used all over your code. It is invoked in hundreds of places. This might be good sign. Your code is highly reusable. But having such a method means, its API is used very frequently and changing the API has a big potential to break your application. If you need to change this method, try to keep its API stable. If you have good unit tests, you can easily prove the stability.

A simple step to get code more testable is writing small classes doing just a few things. Small classes are easier to test, but it is also easier to keep there APIs stable. If classes have few dependencies and small classes tend to have few dependencies, there might be less potential side effects from other objects at runtime. This will help you to make classes' APIs more stable.

If you write testable code, your code will become a big bunch of small pieces each with a very stable API. If you make a change to your code, it might affect just few places and hopefully it does not break your internal APIs.

What about the need of API breaks?

Stable APIs are a good idea and you should pursue them. But sometimes breaks are unavoidable. How can we deal with them?

A breaking API affects the users of the API. If your API breaks and thus your existing tests have been changed, you need to work on the users of the class or method too. Check their unit tests and write new tests, if required to prove the API changes in the changed class do not break their API. If their API also has been broken, go on and work on their users too. Do this until you reach the point, where an API keeps stable.

Testable code helps to deal with this. Breaking a classes API in a testable code base does not affect the whole code. Instead it affects just few classes or methods. If an API break also breaks the APIs of the code's users, the code base is not testable.

Improve quality

We have seen, what kind of trust unit tests can supply to us. Although this trust does not work very well, when your code is not testable improving your code's quality writing testable code is a much better way than doing all tests through the UI and let your code rot.