Writing unit tests

Introduction

This chapter goes into details about writing and maintaining unit tests in the TYPO3 world. Core developers over the years gained quite some knowledge and experience on this topic, this section outlines some best practices and goes into details about some of the TYPO3 specific unit testing details that have been put on top of the native phpunit stack: At the time of this writing the TYPO3 Core contains about ten thousand unit tests - many of them are good, some are bad and we're constantly improving details. Unit testing is a great playground for interested contributors, and most extension developers probably learn something useful from reading this section, too.

Note this chapter is not a full "How to write unit tests" documentation: It contains some examples, but mostly goes into details of the additions typo3/testing-framework puts on top.

Furthermore, this documentation is a general guide. There can be reasons to violate them. These are no hard rules to always follow.

When to unit test

It depends on the code you're writing if unit testing that specific code is useful or not. There are certain areas that scream to be unit tested: You're writing a method that does some PHP array munging or sorting, juggling keys and values around? Unit test this! You're writing something that involves date calculations? No way to get that right without unit testing! You're throwing a regex at some string? The unit test data provider should already exist before you start with implementing the method!

In general, whenever a rather small piece of code does some dedicated munging on a rather small set of data, unit testing this isolated piece is helpful. It's a healthy developer attitude to assume any written code is broken. Isolating that code and throwing unit tests at it will proof its broken. Promised. Add edge cases to your unit test data provider, feed it with whatever you can think of and continue doing that until your code survives all that. Depending on your use case, develop test-driven: Test first, fail, fix, refactor, next iteration.

Good to-be-unit-tested code does usually not contain much state, sometimes it's static. Services or utilities are often good targets for unit testing, sometimes some detail method of a class that has not been extracted to an own class, too.

When not to unit test

Simply put: Do not unit test "glue code". There are persons proclaiming "100% unit test coverage". This does not make sense. As an extension developer working on top of framework functionality, it usually does not make sense to unit test glue code. What is glue code? Well, code that fetches things from one underlying part and feeds it to some other part: Code that "glues" framework functionality together.

Good examples are often Extbase MVC controller actions: A typical controller usually does not do much more than fetching some objects from a repository just to assign them to the view. There is no benefit in adding a unit test for this: A unit test can't do much more than verifying some specific framework methods are actually called. It thus needs to mock the object dependencies to only verify some method is hit with some argument. This is tiresome to set up and you're then testing a trivial part of your controller: Looking at the controller clearly shows the underlying method is called. Why bother?

Another example are Extbase models: Most Extbase model properties consist of a protected property, a getter and a setter method. This is near no-brainer code, and many developers auto-generate getters and setters by an IDE anyway. Unit testing this code leads to broken tests with each trivial change of the model class. That's tiresome and likely some waste of time. Concentrate unit testing efforts on stuff that does data munging magic as outlined above! One of your model getters initializes some object storage, then sorts and filters objects? That can be helpful if unit tested, your filter code is otherwise most likely broken. Add unit tests to prove it's not.

A much better way of testing glue code are functional tests: Set up a proper scenario in your database, then call your controller that will use your repository and models, then verify your view returns something useful. With adding a functional test for this you can kill many birds with one stone. This has many more benefits than trying to unit test glue code.

A good sign that your unit test would be more useful if it is turned into a functional test is if the unit tests needs lots of lines of code to mock dependencies, just to test something using ->shouldBeCalled() on some mock to verify on some dependency is actually called. Go ahead and read some unit tests provided by the Core: We're sure you'll find a bad unit test that could be improved by creating a functional test from it.

Unit test conventions

TYPO3 unit testing means using the phpunit testing framework. TYPO3 comes with as basic UnitTests.xml file that can be used by Core and extensions. This references a phpunit bootstrap file so phpunit does find our main classes. Apart from that, there are little conventions: Tests for some "system under test" class in the Classes/ folder should be located at the same position within the Test/Unit folder having the additional suffix Test.php to the system under test file name. The class of the test file should extend the basic unit test abstract TYPO3\TestingFramework\Core\Unit\UnitTestCase. Single tests should be named starting with the method that is tested plus some explaining suffix and should be annotated with @test.

Example for a system under test located at typo3/sysext/core/Utility/ArrayUtility.php (stripped):

typo3/sysext/core/Utility/ArrayUtility.php
<?php
namespace TYPO3\CMS\Core\Utility;
class ArrayUtility
{

    //...

    public static function filterByValueRecursive($needle = '', array $haystack = [])
    {
        // System under test code
    }
}
Copied!

The test file is located at typo3/sysext/core/Tests/Unit/Utility/ArrayUtilityTest.php (stripped):

typo3/sysext/core/Tests/Unit/Utility/ArrayUtilityTest.php
<?php
namespace TYPO3\CMS\Core\Tests\Unit\Utility;
use TYPO3\CMS\Core\Utility\ArrayUtility;
use TYPO3\TestingFramework\Core\Unit\UnitTestCase;
class ArrayUtilityTest extends UnitTestCase
{

    //...

    /**
     * @test
     * @dataProvider filterByValueRecursive
     */
    public function filterByValueRecursiveCorrectlyFiltersArray($needle, $haystack, $expectedResult)
    {
        // Unit test code
    }
}
Copied!

This way it is easy to find unit tests for any given file. Note PhpStorm understands this structure and can jump from a file to the according test file by hitting CTRL+Shift+T.

Keep it simple

This is an important rule in testing: Keep tests as simple as possible! Tests should be easy to write, understand, read and refactor. There is no point in complex and overly abstracted tests. Those are pain to work with. The basic guides are: No loops, no additional class inheritance, no additional helper methods if not really needed, no additional state. As simple example, there is often no point in creating an instance of the subject in setUp() just to park it as in property. It is easier to read to just have a $subject = new MyClass() call in each test at the appropriate place. Test classes are often much longer than the system under test. That is ok. It's better if a single test is very simple and to copy over lines from one to the other test over and over again than trying to abstract that away. Keep tests as simple as possible to read and don't use fancy abstraction features.

Extending UnitTestCase

Extending a unit test from class TYPO3\TestingFramework\Core\Unit\UnitTestCase of the typo3/testing-framework package instead of the native phpunit class PHPUnit\Framework\TestCase adds some functionality on top of phpunit:

  • Environment backup: If a unit test has to fiddle with the Environment class, setting property $backupEnvironment to true instructs the unit test to reset the state after each call.
  • If a system under test creates instances of classes implementing SingletonInterface, setting property $resetSingletonInstances to true instructs the unit test to reset internal GeneralUtility scope after each test. tearDown() will fail if there are dangling singletons, otherwise.
  • Adding files or directories to array property $testFilesToDelete instructs the test to delete certain files or entire directories that have been created by unit tests. This property is useful to keep the system clean.
  • A generic tearDown() method: That method is designed to test for TYPO3 specific global state changes and to let a unit test fail if it does not take care of these. For instance, if a unit tests add a singleton class to the system but does not declare that singletons should be flushed, the system will recognize this and let the according test fail. This is a great help for test developers to not run into side effects between unit tests. It is usually not needed to override this method, but if you do, call parent::tearDown() at the end of the inherited method to have the parent method kick in!
  • A getAccessibleMock() method: This method can be useful if a protected method of the system under test class needs to be accessed. It also allows to "mock-away" other methods, but keep the method that is tested. Note this method should not be used if just a full class dependency needs to be mocked. Use prophecy (see below) to do this instead. If you find yourself using that method, it's often a hint that something in the system under test is broken and should be modelled differently. So, don't use that blindly and consider extracting the system under test to a utility or a service. But yes, there are situations when getAccessibleMock() can be very helpful to get things done.

General hints

  • Creating an instance of the system under test should be done with new in the unit test and not using GeneralUtility::makeInstance().
  • Only use getAccessibleMock() if parts of the system under test class itself needs to be mocked. Never use it for an object that is created by the system under test itself.
  • Unit tests are by default configured to fail if a notice level PHP error is triggered. This has been used in the Core to slowly make the framework notice free. Extension authors may fall into a trap here: First, the unit test code itself, or the system under test may trigger notices. Developers should fix that. But it may happen a Core dependency triggers a notice that in turn lets the extensions unit test fail. At best, the extension developer pushes a patch to the Core to fix that notice. Another solution is to mock the dependency away, which may however not be desired or possible - especially with static dependencies.

A casual data provider

This is one of the most common use cases in unit testing: Some to-test method ("system under test") takes some argument and a unit tests feeds it with a series of input arguments to verify output is as expected. Data providers are used quite often for this and we encourage developers to do so, too. An example test from ArrayUtilityTest:

typo3/sysext/core/Tests/Unit/Utility/ArrayUtilityTest.php
/**
 * Data provider for removeByPathRemovesCorrectPath
 */
public function removeByPathRemovesCorrectPathDataProvider()
{
    return [
        'single value' => [
            [
                'foo' => [
                    'toRemove' => 42,
                    'keep' => 23
                ],
            ],
            'foo/toRemove',
            [
                'foo' => [
                    'keep' => 23,
                ],
            ],
        ],
        'whole array' => [
            [
                'foo' => [
                    'bar' => 42
                ],
            ],
            'foo',
            [],
        ],
        'sub array' => [
            [
                'foo' => [
                    'keep' => 23,
                    'toRemove' => [
                        'foo' => 'bar',
                    ],
                ],
            ],
            'foo/toRemove',
            [
                'foo' => [
                    'keep' => 23,
                ],
            ],
        ],
    ];
}

/**
 * @test
 * @dataProvider removeByPathRemovesCorrectPathDataProvider
 * @param array $array
 * @param string $path
 * @param array $expectedResult
 */
public function removeByPathRemovesCorrectPath(array $array, $path, $expectedResult)
{
    $this->assertEquals(
        $expectedResult,
        ArrayUtility::removeByPath($array, $path)
    );
}
Copied!

Some hints on this: Try to give the single data sets good names, here "single value", "whole array" and "sub array". This helps to find a broken data set in the code, it forces the test writer to think about what they are feeding to the test and it helps avoiding duplicate sets. Additionally, put the data provider directly before the according test and name it "test name" + "DataProvider". Data providers are often not used in multiple tests, so that should almost always work.

Mocking

Unit tests should test one thing at a time, often one method only. If the system under test has dependencies like additional objects, they should be usually "mocked away". A simple example is this, taken from TYPO3\CMS\Backend\Tests\Unit\Controller\FormInlineAjaxControllerTest:

typo3/sysext/backend/Tests/Unit/Controller/FormInlineAjaxControllerTest.php
/**
 * @test
 */
public function createActionThrowsExceptionIfContextIsEmpty(): void
{
    $requestProphecy = $this->prophesize(ServerRequestInterface::class);
    $requestProphecy->getParsedBody()->shouldBeCalled()->willReturn(
        [
            'ajax' => [
                'context' => '',
            ],
        ]
    );
    $this->expectException(\RuntimeException::class);
    $this->expectExceptionCode(1489751361);
    (new FormInlineAjaxController())->createAction($requestProphecy->reveal());
}
Copied!

Prophecy is a nice mocking framework bundled into phpunit by default. Many people prefer it nowadays over phpunit's own mock framework based on ->getMock() and we encourage to use prophecy: Prophecy code is often easier to read and the separation of the dummy object that is given to the system under test, the "revelation", and the object prophecy is quite handy. Prophecy is quite some fun to use, go ahead and play around with it.

The above case is pretty straight since the mocked dependency is hand over as argument to the system under test. If the system under test however creates an instance of the to-mock dependency on its own - typically using GeneralUtility::makeInstance(), the mock instance can be manually registered for makeInstance:

EXT:my_extension/Tests/Unit/SomeTest.php
use TYPO3\CMS\Core\Utility\GeneralUtility;
use TYPO3\CMS\Core\Imaging\IconFactory;

GeneralUtility::addInstance(IconFactory::class, $iconFactoryProphecy->reveal());
Copied!

This works well for prototypes. addInstance() adds objects to a LiFo, multiple instances of the same class can be stacked. The generic ->tearDown() later confirms the stack is empty to avoid side effects on other tests. Singleton instances can be registered in a similar way:

EXT:my_extension/Tests/Unit/SomeTest.php
use TYPO3\CMS\Core\Utility\GeneralUtility;
use TYPO3\CMS\Extbase\Service\EnvironmentService;

 GeneralUtility::setSingletonInstance(EnvironmentService::class, $environmentServiceMock);
Copied!

If adding singletons, make sure to set the property protected $resetSingletonInstances = true;, otherwise ->tearDown() will detect a dangling singleton and let's the unit test fail to avoid side effects on other tests.

Static dependencies

If a system under test has a dependency to a static method (typically from a utility class), then hopefully the static method is a "good" dependency that sticks to the general static method guide: A "good" static dependency has no state, triggers no further code that has state. If this is the case, think of this dependency code as being inlined within the system under test directly. Do not try to mock it away, just test it along with the system under test.

If however the static method that is called is a "bad" dependency that statically calls further magic by creating new objects, doing database calls and has own state, this is harder to come by. One solution is to extract the static method call to an own method, then use getAccessibleMock() to mock that method away. And yeah, that is ugly. Unit tests can quite quickly show which parts of the framework are not modelled in a good way. A typical case is TYPO3\CMS\Backend\Utility\BackendUtility - trying to unit test systems that have this class as dependency is often very painful. There is not much developers can do in this case. The Core tries to slowly improve these areas over time and indeed BackendUtility is shrinking each version.

Exception handling

Code should throw exceptions if something goes wrong. See working with exceptions for some general guides on proper exception handling. Exceptions are often very easy to unit test and testing them can be beneficial. Let's take a simple example, this is from TYPO3\CMS\Core\Tests\Unit\Cache\CacheManagerTest and tests both the exception class and the exception code:

typo3/sysext/core/Tests/Unit/Cache/CacheManagerTest.php
use TYPO3\CMS\Core\Cache\CacheManager;
use TYPO3\CMS\Core\Cache\Exception\NoSuchCacheGroupException;

/**
* @test
*/
public function flushCachesInGroupThrowsExceptionForNonExistingGroup()
{
  $this->expectException(NoSuchCacheGroupException::class);
  $this->expectExceptionCode(1390334120);
  $subject = new CacheManager();
  $subject->flushCachesInGroup('nonExistingGroup');
}
Copied!