TDD: Designing our code, test-first

Standard

Hi, dear readers! Welcome to my blog. On this post, we will talk about TDD, a methodology that preaches about focusing first on tests, instead of our production code. I have already talked about test technologies on my post about mockito, so in this post we will focus more on the theoretical aspects of this subject. So, without further delay, let’s begin talking about TDD!

Test-driven development

TDD, also known as Test-driven development, is a development technique created by Kent Beck, which also created JUnit, a highly known Java test library.

TDD, as the name implies, dictates that the development must be guided by tests, that must be written before even the “real” code that implements the requirements is written! Let’s see in more detail how the steps of TDD work on the next section.

TDD steps

As we can see on the picture above, the following steps must be met to proceed in using the TDD paradigm:

  1. Write a test (that fails): Represented by the red circle, it means that before we implement the code – a method on a class, for example – we implement a test case, like a JUnit method for instance, and invoke our code, in this case a simple empty method. Of course, on this scenario the test will fail, since there is no implementation, which leads to our next step;
  2. Write code just enough to pass: Represented by the green circle, it means that we write code on our method just enough to make our test case pass with success. The principle behind this step is that we write code with simplicity in mind, in other words, keeping it simple, like the famous kiss principle;
  3. Refactor the code, improving his quality: Represented by the blue circle, it means that, after we make our code to pass the test, we analyse our code, looking for chances to improve his quality. Is there duplicate code? Remove the duplications! Is there hard coded constants? Consider changing to a enum or a properties table, for instance.

A important thing to notice is that the focus of the third step and the previous one is to not implement more functionality then the test code is testing. The reason for this is pretty obvious: since our focus is to implement the tests for our scenarios first, any functionality that we implement without the test reflecting the scenario is automatically a code without test coverage;

After we conclude the third step, the construction returns to the first step, reinitiating the cycle. On the next iteration, we implement another test case for our method, representing another scenario. We then code the minimum code to implement the new scenario and conclude the process by refactoring the code. This keeps going until all the scenarios are met, leaving at the end not only our final code, but all the test cases necessary to effective test our method, with all the functionalities he is intended to implement.

The reader may be asking: “but couldn’t we get the same by coding the tests after we write our production code?”. The answer is yes, we could get the same results by adding our tests at the end of the process, but then we wouldn’t have something called feedback from our code.

Feedback

When we talk about feedback, we mean the perception we have about how our code will perform. If we think about, the test code is the first client of our code: he will prepare the data necessary for the invocation, invoke our code and then collect the results. By implementing the tests first, we receive more rapidly this feedbacks from our implementation, not only on the sphere of correctness, but also on design levels.

For instance, imagine that we are building a test case and realize that the test is getting difficult to implement because of the complex class structure we have to populate just to make the invocation of our test’s target. This could mean that our class model is too complex and a refactoring is necessary.

By getting this feedback when we are just beginning the development – remember that we always code just enough to the test to pass! – it makes a lot more cheap and less complex to refactor the model then if we receive this feedback just at the end of the construction, when a lot of time and effort was already made! By this point of view, we can easily see the benefits of the TDD approach.

Types of tests

When we talked about tests on the previous sections, we talked about a type of test called unit test. Unit tests are tests focused on testing a single program unit, like a class, not focusing on testing the access to external resources, for example. Of course, there are other types of tests we can implement, as follows:

Unit tests: Unit tests have their focus to test individually each program unit that composes the system, without external interferences.

Integration tests: Integration tests also have the focus to test program units, but in this case to test the integration of the system with external resources, like a database, for example. A good example of a integration test is test cases for a DAO class, testing the code that implement insertions, updates, etc.

System tests: System tests, as the name implies, are tests that focus on testing the whole system, across all his layers. In a web application, for example, that means a automated test that turns on a server, execute some HTTP requests to the web application and analyse the responses. A good example of technology that could be used to test the web tier is a tool called Selenium.

Acceptance tests: Acceptance tests, commonly, are tests made by the end user of the system. The focus of this kind of test is to measure the quality of the implementation of requirements specified by the user. Other requirements such as usability are also evaluated by this kind of test.

A important thing to notice is that this kind of test is also referred as a automated system test, with the objective of testing the requirement as it is, for example:

  • Requirement: the invoice must be inserted on the system by the web interface;
  • Test: create a system test that executes the insertion by the web interface and verifies if the data is correctly inserted;

This technique, called ATDD (Acceptance Test-Driven Development) preaches that first a acceptance test must be created, and then the TDD technique is applied, until the acceptance test is satisfied. The diagram bellow shows this technique in practice:

Mock objects and unit tests

When we talk about unit tests, as said before, it is important to isolate the program unit we are testing, avoiding the interference from other tiers and dependencies that the program unit uses. On the Java world, a good framework we can use to create mocks is Mockito, which we talked about on my previous post.

With mocks, following the principles of TDD, we can, for example, create just the interfaces our code depends on and mock that interfaces, this way already establishing the communication channel that will be created, without leaving our focus from the program unit we are creating.

Another benefit of this approach is on the creation of the interfaces themselves, since our focus is always to implement the minimum necessary for the tests to pass, the resulting interfaces will be simple and concise, improving their quality.

Considerations

When do I use mocks?

A important note about mocks is that not always is good to use them. Taking for example a DAO class, that essentially is just a class that implement code necessary to interact with a database, for instance, the use of mocks won’t bring any value, since the code of the class itself is too simple to benefit from a unit test. On this cases, typically just a integration test is enough, using for example in-memory databases such as HSQLDB to act as the database.

Should I code more then one test case at a time?

In general, is considered a bad practice to write more then one test case at once before running and getting the fail results. The reason for this is that the point of the technique is to test one functionality at a time, which of course is “ruined” by the practice of coding more then one test at once.

How much of a functionality do I code on the test?

On the TDD terminology, we can also call the “amounts” of functionality we code at each iteration as baby steps. There is no universal convention of how much must be implement in each of this baby steps, but it is a good advice to follow common sense. If the developer is experienced and the functionality is simple, it could be even possible to implement almost the whole code on just one iteration.

If the developer is less experienced and/or the functionality is more complex, it makes more sense to spend more time with more baby steps, since it will create more feedbacks for the developer, making it easier to implement the complex requirements.

Should I test private code?

Private code, as the name implies, are code – like a method, for instance – that are accessible only inside the same program unit, like a class, for example. Since the test cases are normally implemented on another program unit (class), the test code can’t access this code, which in turn begs the question: should I make any code to test that private code?

Generally speaking, a common consensus is that private code is intended to implement little things, like a portion of code that is duplicated across multiple methods, for example. In that scenario, if you have for instance a private method on a Java class that it is enormous with lots of functionality, then maybe it means that this method should be made public, maybe even moved to a different class, like a utility class.

If that is not the case, then it is really necessary to design test cases to efficiently test the method, by invoking him indirectly by his public consumer. Talking specifically on the Java World, one way to test the code without the “barrier” of the public consumer is by using reflection.

My target code has too much test cases, is this normal?

When implementing the test cases for a target production code – a method, for example – the developer could end up on a scenario that lots of test cases are created just to test the whole functionality that single code composes. When this kind of situation happens, it is a good practice that the developer analyse if the code doesn’t have too much functionality implemented on itself, which leads to what in OO we call as low cohesion.

To avoid this situation, a refactoring from the code is needed, maybe splitting the code on more methods, or even classes on Java’s case.

Conclusion

And this concludes our post about TDD. By shifting the focus from implementing the code first to implementing the tests first, we can easily make our code more robust, simple and testable.

In a time that software is more important then ever – been on places like airplanes, cars and even inside human beans -, testing efficiently the code is really important, since the consequences of a bad code are getting more and more devastating. Thank you for following me on another post, until next time.

Advertisements

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s