PHPUnit @covers Tutorial

A small demonstration of using PHPUnit's @covers annotation to control code coverage.

PHPUnit utilizes the xdebug extension in order to analyze your code and determine which lines of your code are "covered" by your tests. By itself, this is useful for finding untested or poorly tested areas of your application. Issues can arise though if your app is structure in such a way where a test accidentally covers some of your code. Sometimes this is okay or even preferable when testing protected methods via their public interfaces, but this can lead to a false sense of security where atomic units of your code are only covered because something else that is tested happens to call them.

So PHPUnit's @covers annotation exists to provide a way to restrict which parts of your code a given unit test is meant to cover. Practically speaking this means that when PHPUnit generates coverage reports, it will use the @covers annotations to only count lines of code explicitly @covered by your test.

Setup

1. You should have a function installation of PHP v5.3+ on your system.

2. Install the xdebug extension.

3. Install PHPUnit 3.7+. (This repo also includes a composer file, so you should also be able to run composer install if you have it available.)

4. Download or clone this project: git clone https://github.com/beporter/phpunit-coverage-tutorial.git

5. All of the examples take place from the command line PHPUnit test runner, so open a terminal and navigate to the project folder.

Tutorial

Running existing tests and checking coverage

1. The first step is to get familiar with the code and the tests.

   • Open SampleClass.php. You'll see it is built to generate Fibonacci value arrays, and can format them into strings.
   • In SampleClassTest.php you'll find tests for all three methods (fib(), aryToStr() and printFibSequence()) but note that the first and last tests are set to be skipped via markTestSkipped() and markTestIncomplete(). (We'll come back to this later.)

2. Next we'll run the tests as-is and generate a code coverage report for the project.

   • In your terminal, run ./phpunit-runner.sh.
   • This will execute the tests in SampleClassTest.php and produce a code coverage report at ./coverage/index.html.
   • (You can examine the phpunit-runner.sh script to see how it is executing the tests and generating the HTML report if you want.)

3. Before moving on, let's take a quick look at the output of the command line test runner.

   • You'll see a bunch of S's (skipped tests), two .'s (passed tests) and a trailing I (incomplete test).
   • This is normal because as mentioned above the first and last tests has been intentionally disabled to start.

4. Open the coverage report index coverage/index.html in your browser. (On a Mac, type open coverage/index.html in your Terminal.)

   • In your browser, click on the SampleClass.php link to see detailed coverage for that class.
   • (You can keep this window open from now on, and just refresh the page after running ./phpunit-runner.sh.)
   • (Here is an example of the initial coverage report in case you are just poking around this project and not actually following around at home.)
   • The coverage seems pretty good in spite of those skipped tests! We're only missing one line of code in fib(). Except there's an issue here...
   • If you hover your mouse over any of the green "covered" lines of code, you'll see the names of the tests that executed that line.
   • In this case, all of our lines are covered by testPrintFibSequence(). Uh oh, that means our code coverage of fib() and aryToStr() is based on incidental calls from printFibSequence().
   • What's more, if you hover over the one line in aryToStr() you'll see it is covered by two tests: testAryToStr and testPrintFibSequence. This means your tests are executing your code redundantly (which is why tests doubles are usually preferable), but in our case it also means one of our tests is overreaching its intended scope and what we're concerned with at the moment is being able to analyze coverage accurately.

Fixing the coverage report

1. The first thing we want to do is isolate our active test so that it only "covers" the method we intend it to.

   • (Another [almost always better] way to accomplish this is by using test doubles but this example is about @covers, so bear with me.)
   • In SampleClassTest.php, find the comment TUTORIAL#1 which should be around L106.
   • Immediately below that line is a PHPUnit @covers annotation that has been disabled.
   • Remove the -disabled from the annotation (so the whole clause reads @covers SampleClass::printFibSequence) and save the file.

2. Now run the tests again (./phpunit-runner.sh) and refresh the code coverage report in your Terminal.

   • Back in your browser, the percentage of code covered should drop.
   • Now our testPrintFibSequence() test method is only covering the printFibSequence() method, and the incidental calls to the other class methods have been ignored.
   • This gives us a more accurate picture of what we're really testing.
   • Hovering your mouse over aryToStr()'s single line now reports only one test covering it.

Covering the remaining method

In this tutorial the hard work of writing the tests has been done for you, so all you need to do is enable them.

1. In SampleClassTest.php, find the line contain TUTORIAL#2 which should be around L78 at the top of testFib().

2. Delete this entire line (which will cause this test to no longer be skipped) and save the file.

3. Run your tests again (./phpunit-runner.sh), switch back to your browser and refresh the report.

   • On the command line, our tests are almost green now!
   • In our coverage report, we've covered fib() in its entirety thanks to testib().

BONUS: Data providers

• If you take a look at the now-active testFib() method, you'll notice that it has method arguments defined, and doesn't do any setup-- It only calls the assertEquals() assertion with the provided arguments.

• The arguments come from a data provider, which is another method in the test class that returns an array of "data sets" that should be fed to the test method.

• The testFib() method's doc block contains a @dataProvider provideTestFibArgs annotation that tells PHPUnit which method to get the data sets from.

• Each element from the array return by provideTestFibArgs() contains a set of values that will be provided to testFib() as a set of arguments.

• Data providers are an excellent way to control redundancy in your test methods and easily test highly algorithmic methods that vary only in input and output.

• It's also a lot easier to add new specific test samples as you encounter (and fix) problematic input/output pairs.

BONUS: Using a test double instead of @covers

• The last test method, SampleClassTest::testPrintFibSequenceTestDouble() (marked by a TUTORIAL#3 comment around L128) can be enabled by deleting the markTestIncomplete() line and is an alternate way of testing the printFibSequence().

• Once you've done this, it's safe to delete the original testPrintFibSequence() method entirely.

• This second version replaces the "real" copies of the fib() and aryToStr() methods with "test doubles" that do what we tell them to instead of what the actual code as-written does.

• When we call the (real) printFibSequence() method in this test double, our replaced versions of the "incidental" methods execute instead of the real ones, allowing us to never actually leave the printFibSequence() method during test execution.

• This approach is quite a bit more verbose, but doesn't require the @covers line in its doc block because it never actually calls the real fib() and aryToStr() methods.

• This also makes our test completely independent from changes made to fib() and aryToStr(), which is an extremely good thing.

• @covers shouldn't be used in place of using test doubles; only as an intelligent supplement to control the coverage reports.

Final thoughts

• The @covers annotation can be really excellent for limiting what your tests are intended to be "touching".

• They are especially handy when you can't easily control a method's use of external calls. (Static calls like CakePHP's Router::url() in particular are horrible for unit testing.)

• The resulting code coverage is much "truer" because you don't get any incidental method calls as an unwanted bonus in your coverage. This makes you work for the coverage more honestly.

• The downside is that the @covers annotations are really easy to miss, and when refactoring code or tests, you can end up with a test that @covers a method name that doesn't exist anymore (something I actually did accidentally while setting this tutorial up), which means that test contributes nothing to your coverage even if it is properly executing and verifying some code somewhere correctly.

Questions, Comments, Feedback, Contributions

Please post an issue. This was intended as a one-off lesson, so while I appreciate feedback and contributions, I don't intended to spend a lot of time maintaining or improving this project.

License

CC-BY-NC-SA-3.0

Copyright

© 2014 Brian Porter

Technical Details

• The composer.json file (in this case) is really only needed to pull in tools used for testing (phpunit and woodhouse.)
  • It's also used locally for "one-time" set up the rest of these configs, so you'll have to run composer install at least once to do the rest of this.
• You must have a script that will execute your full test suite, in our case that's `phpunit-runner.sh.
  • It's important that the phpunit script be set up to generate HTML coverage and a text coverage file for use by woodhouse later.
  • These generated files should be included in the .gitignore so they do not get accidentally committed to the repo.
• The .travis.yml config specifies the phpunit-runner.sh as the "script" to run during travis executions.
• The travis config specifies woodhouse.sh as the "after-run" script to publish code coverage back to the gh-pages branch of the Github repo.
• Woodhouse needs a Github auth token in order to do the publishing.
• The auth token can be created manually in your Github repo's settings, or using the woodhouse command line tool.
• That auth token needs to be encrypted as an environment variable that is available during travis test runs using travis encrypt ENV_VAR_NAME="env var value". (See also: installing the travis command line tool.)
• The encrypted value needs to be saved into your .travis.yml file under env: global: - secure:.
• The counterpart to this is that when you execute woodhouse (in this project, that happens in woodhouse.sh) you must tell woodhouse where to obtain the key using --auth-env-token ENV_VAR_NAME (where "ENV_VAR_NAME" is the same thing you encrypted earlier.
  • Woodhouse also depends on a text coverage file generated earlier by phpunit so it can grab the proper code coverage badge, which will also be published back to gh-pages.
• You should lint your .travis.yml file.
• You should lint your composer.json file.
• Finally, you must enable travis for your repo.

What this all accomplishes:

• When you push a commit, travis is notified and starts a test run.
• Travis uses a virtual machine to clone your code, init any git submodules, install any composer dependencies, and then executes a script which should run your entire test suite.
• The exit code of that script determines if the build passes (0) or fails (>0).
• Any other necessary steps are also run before/after (including woodhouse:)
  • woodhouse takes any local "artifacts" you designate and commits them back to the named repo into the gh-pages branch.
  • These artifacts become available at http://username.github.io/repository/, which you can link to in your README to display a build status or code coverage badge, for example.