Test-Driven Development

Every Developer has to test his software - for developers a not much liked theme, but one you cannot walk by. How work tests in the classic software development? Programmers in the first way construct a test case by putting certain data into the database, write little programs to test or manipulate the url parameter in the browser. A big number of little functions are often implemented at one time, before they are tested (look at figure 2-7). After that, these steps are repeated in periodic steps, as often until the function you want to have is implemented. After this the next function is in line. But this way leads to problem: the tests are not made systematic, but selective. By following this way you are implementing failures into some routines by accident. Furthermore by a test is not only a little code fragment tested, but often a more complex routine.


Figure 2-7: in the classic software development exist a strict isolation between the Development- and the Test- phase.

By using Test-Driven Development (TDD) these problems should be solved. Tests could be fast completed and reproducible. These increases the developers motivation to start the tests constantly and by this way to get faster a callback, if a failure is implemented into the existing functions.


Field report

When Robert Lemke and other Core Developers suggested to make the development for FLOW test driven, I was skeptic. Test-Driven Development sounded like a nice concept, but I did not know how to test a framework this size reasonable. Also in the internet there were often only very simple academic examples to find. Until this time I had only a theoretical overview over TDD. Even when I started to test, when the Fluid development started. The first test were not Unit- but Integration tests. This means they tested Fluid in the view of a user: There were not parsed little Template-Snippets and compared with the expectations. The first tests took there time - it felt strange to test things, which were not implemented at this time. But after the first test was written and this test run through successfully, I was able to make the following development cycles extremely fast. Because of Test-Driven Development I was able during a train ride to totally reconstruct the core of Fluid. Without tests, it seriously would have took me days until all would have worked at the end. Especially the feedback I got at once, I really appreciate. You click on a button and after a few seconds you got your feedback. After this I am infected, learned about Mock- and Stub- objects, and today I do not want to miss it. (In this chapter you will get an introduction into these concepts.) If you want to learn TDD, you will jump in at the deep end, and to try it at the next project. Until the first Unit test is finished, it will take a while and after this it will go really faster. – Sebastian Kurfürst

First test, then implementing

The goal of Test-Driven Development is to make test explicit and automatic repeatable. The workflow is different of the classic programming, separated into very small iterations.

Using Test-Driven Development you write your Unit tests for your features before you write your features themselves. Unit tests are automatically repeatable tests of methods, classes and little program parts. Even during the writing of the tests, you should seriously think about the desired functionality. When you are running your tests, they naturally will fail, because the tested functionality is not implemented yet. In the next step you will have to write the code which is minimal needed to run the test successfully (have a look at figure 2-8). A new started test will finish without a problem. In the next step you should begin to think about which functionality is missing. The reason is that you have only implemented the minimal necessary code. When you know, what you want to implement you will write a new Unit test, with which you can test this new functionality. In this way the process of testing an implementing starts over.


Figure 2-8: with Test-Driven Development testing and development are alternating

The fortune of this method is obviously.

Because you are forced by writing of the tests to separate big features into smaller pieces, you have during the implementation the possibility to fully concentrate on the functionality you have actually to implement. Furthermore these change of the point of view between Developer and User of a class is the reason for better code which does what it should. TDD is especially useful for designing of APIs: Because of the testing you as developer change your perspective into the one of the user of the API and so you are able to identify inconsistencies and missing functionality much faster. In addition Unit tests are a kind of safety net against unrequested behavior. In the case of destroying a functionality when you are programming, you are getting through the Unit tests feedback directly and are able to correct the bug at once. This is the reason why the chance of regressions (producing bugs same time you are correcting another) is furthermore decreasing. According to some studies Developers who are using TDD are fast at the same rate or even faster than developers who are coding the conventional style, although the first mentioned have not only to write the Implementation but also the according tests. Furthermore is the quality of the code in the Test-Driven Development clearly higher than in the conventional Programming.


Let us assume we want to model an online shop for a customer. These owns a name, which is transferred in the constructor. Through these tests we want to make sure that the name is correctly saved into the object. At first you have to install the phpunit-Extension from the TYPO3 Extension Repository (TER), because we will run the tests with this. The next step is to go to our own extension and create a folder "Tests/Unit/" in the main folder of the extension, if it not exists. This will contain later all our unit tests. Our customer object which we have to create, will be situated in Classes/Domain/Model/Customer.php because it is part of our domain model in our extension.

Similarly we create the test class in the file Tests/Unit/Domain/Model/CustomerTest.php. Now we create a minimal test case with which we get used with PHPUnit.


All our test cases are named after the same name scheme like normal classes and they must be extended with TYPO3CMSCoreTestsBaseTestCase. One test class can contain many test methods. These have to be public and have to contain the annotation @test in their PHPDOC-Block, so they can be performed. Please keep in mind that the name of the test method should make clear which expectations the test should fulfill. Now we can run the test for the first time. Therefore go to the TYPO3-Backend to the module PHPUnit which is to find under the Admin Tools. Then you can choose your extension and click on Run all tests. Now you should, like it is shown in the figure 2-9, see a (yellow) bar and the Error message Not yet implemented. Because you will work much with the PHPUnit Environment, you should familiarize yourself with this. Try to run the test for extbase and fluid and also try the different Display options. For example you can let show you all tests or only the failed tests.

Now we know that our test case is running, we can write our first useful test case. This should test, if a name which is specified in the constructor, can be accessed again.


// Code


Figure 2-9: With the test runner you are able to run easily the Unit tests in the TYPO3-Backend.

When we run the test case, we will be displayed a fatal error from PHP, because the class we want to test does not exist already. Now we are changing roles: We are not the user of the class anymore, but now we are the developer, who should implement the class. At first we create in the file Classes/Domain/Model/Customer.php an empty class with the needed methods to get rid of the fatal error:


When we now let run the test suite again there should not be a fatal error anymore but instead our Unit-Tests will fail because the getName() method returns the false value. Now we are able, motivated by getting the red bar fast as possible into green, to start with implementing:


Now the Unit-Tests are running without failure, the expected value is given out. At this time we are not satisfied at all - finally, now is always 'Sebastian Elector', as name returned. The next step is a refactoring phase: We clean up the code and always make sure that the unit tests continue successfully passed. After several iterations we arrive at the following code:


h1. Page 48

The unit tests always run through yet, and we have the desired functionality reached. Now we can once again slip into the role of the developer from the role of the user of the class and specify with other test cases additional functionality.

Test individual objects

Our first example about Unit-Tests was very simple. In this section we show you how to test classes that depend on other classes. Suppose we have a program which is writing log messages and they should be send per mail. For those there is a class EmailLogger that send the log data via e-mail. These class implements the potential complex goal of the e-mail sending on is own, but is using another class which is called EmailService. EmailService uses, depending on the used configuration a SMTP-Server or the mail() function of PHP. This is shown in the figure 2-10: The email logger class has a reference on the email service.


Figure 2-10: The EmailLogger uses for sending of the emails the EmailService.

We now want to test the class EmailLogger without using the EmailService. We do not want to send real emails with every test run. To reach that goal we need two sub elements. Dependency Injection and the use of Mock objects. Both concepts we will introduce below.

Dependency Injection

You often see classes that are constructed according to the following structure:


The EmailLogger requires the EmailService to function correctly, and instantiated this in the constructor. However, this strongly coupled to these two classes together: When you create to test a new instance of the class EmailLogger, you automatically get an instance of an EmailService and this would implicitly be tested. In addition, it is not possible to exchange the EmailService at run time, without changing the source code. A solution to this dilemma is to use Dependency Injection:

This instantiates a class does not itself have dependencies but she gets from the outside passed. The EmailLogger gets a new method injectsEmailService, the EmailService in the class sets. This looks e.g. like this:


Extbase offers currently not a framework support for Dependency Injection. Therefore, we recommend that the instantiation of classes and their Dependency Injection in respective factories to outsource. A possible Factory looks as follows from:


We can now operate in a test case from the outside, which the EmailService the EmailLogger gets. We could write a TestEmailService, for example, which simple does nothing (to avoid a fatal error), or we use the Mock objects that are shown below.


Through the usage of Dependency Injection we are able to instantiate EmailLogger without its dependencies. Because the EmailLogger needs the EmailService to work, we must provide these in the tests.

But more than that: We also want to ensure that the EmailLogger really calls the method for sending the e-mail! Therefore we can use Mocks. Mocks are more or less "dummies" for real objects that emulate the behavior of objects. They are also help to ensure specific calls or parameters. A test that tests the EmailLogger that could be as follows:

// code

The procedure in detail: In line 6, the variable $message with our dummy message filled, we want to log. This message we need several more times, so it makes sense to store them in a variable. In lines 7 through 9, we instantiate the EmailLogger and initiate him a mock object of the EmailService.

In line 10 passes through the truly fascinating: We expect that in the EmailService an unique method call is sent, with the parameters 'logging@domain.local','Message Log', $message. Once we have specified our expectations, we can at line 11 let the EmailLogger log the message. At the end of the test cases our expectations are automatically controlled.

If the the method send was called exactly once or with the false parameter values, the test will fail with a detailed error message. What have we achieved? We have tested the EmailLogger without the use of Email-Service and still ensure that the Email-Service is called with the correct parameters. Also, we had no separate "placeholder" class to write for Email-Service, because we used the mock-functionality from PHPUnit.


You have to get used to the writing style for Mock objects; But it will go on with the time in flesh and blood.