Functional Test Service or Unit Test Class

[saamorim]

Hi Guys.

I’m creating a new project, this time in Symfony, and an old question keeps poping up and I would like to hear your thoughts.

When creating tests, should I focus on:

1. Testing the service container definition
2. Unit testing the implementation of that service
3. Both
4. A mix

As an example, lets assume that I’m defined in the services.yml the following:

foobar_service:
   class: Foo\Bar
   arguments: [@"dep1", @"dep2"]

and it’s implementation is:

namespace Foo;
public class Bar implements IBar {
   /** private dependencies accessors */
   public __constructor($dep1, $dep2) { ... }

   //IBar implementation
   public doSomething($arg1) { return $arg1 * 2; }
}

The tests for scenario 1, where I want to test the provided service, could be:

public function testDoSomethingSuccess() {
   $kernel = static::createKernel();
   $kernel->boot();
   $foobar_service = $kernel->getContainer()->get('foobar_service');
   $this->assertEquals(2, $foobar_service->doSomething(1));
}

The tests for scenario 2, where I want to test the implementation. could be:

public function testDoSomethingSuccess() {
   $stubDep1 = $this->mockDep1();
   $stubDep2 = $this->mockDep2();
   
   $foobar = new \Foo\Bar($stubDep1, $stubDep2);
   $this->assertEquals(2, $foobar->doSomething(1));
}

The first one I consider a functional test and “assures” that the interface is tested whatever the implementation would be, and I could switch implementation and “assure” that it works. But it’s a pain to test as it doesn’t allow me to control the dependencies, the input data is harder to control, and I can’t assure what I can’t see (the interactions with dep1 and dep2) But, as a user of service foobar_service this is what I want to rely on.

The second one I consider a unit test and assures that a specific implementation is tested. I can control the dependencies and the interaction with them, but I need to stub each dependency and method used by \Foo\Bar in order to successfully make the test pass, and because of it, I could make incorrect assumptions which could give me false sense of correctness. Multiple implementations of the same service could assume different . But, this is what I want as a developer of foobar_service.

Creating both seems like a waste of effort… or doesn’t it?

Thanks