Mockito: Why You Should Not Use InjectMocks Annotation to Autowire Fields

People like the way how Mockito is able to mock Spring’s auto-wired fields with the @InjectMocks annotation. When I read this post of Lubos Krnac last week, I thought I should explain why I think the use of InjectMocks is a bad signal and how you should avoid it. Hint: it’s about visibility.

Let’s say we have a PlannerServiceImpl which delegates to a PlannerClient. Uses Spring for auto-wiring all together; there’s no constructor, but Spring is able to use field injection.

@Service
public class PlannerServiceImpl implements PlannerService {
    private static final Logger LOG = LoggerFactory.getLogger(PlannerServiceImpl.class);

    @Autowired
    private PlannerClient plannerClient;

    @Override
    public Long createWeddingPlan() {
        try {
            CreateWeddingPlanResponse response = plannerClient.createWeddingPlan();
            return convert(response).getId();
        } catch (Exception e) {
            LOG.error("Unable to create wedding plan", e);
            return null;
        }
    }

An associated test could look like:

@RunWith(MockitoJUnitRunner.class)
public class PlannerServiceImplTest {

    @Mock
    private PlannerClient plannerClient;

    @InjectMocks
    private final PlannerServiceImpl plannerService = new PlannerServiceImpl();

    @Test
    public void testCreateWeddingPlanWhenClientReturnsUndefinedResponseThenNullIsReturned() throws Exception {
        when(plannerClient.createWeddingPlan()).thenReturn(null);

        final Long actual = plannerService.createWeddingPlan();

        assertThat(actual, is(nullValue()));
    }

The org.mockito.InjectMocks annotation can be seen as an equivalent of Spring’s own dependency injection. The Javadoc states:

Mockito will try to inject mocks only either by constructor injection, setter injection, or property injection in order and as described below. If any of the following strategy fail, then Mockito won’t report failure; i.e. you will have to provide dependencies yourself.

(Whoever would design this to fail silently at all?)

So what if someone decides to create a new dependency, say an AuditService and upgrades a bunch of services by adding it as an additional property, also marked as @Autowired?

@Service
public class PlannerServiceImpl implements PlannerService {
    private static final Logger LOG = LoggerFactory.getLogger(PlannerServiceImpl.class);

    @Autowired
    private PlannerClient plannerClient;

    @Autowired
    private AuditService auditService;

    @Override
    public Long createWeddingPlan() {
        try {
            CreateWeddingPlanResponse response = plannerClient.createWeddingPlan();
            auditService.addEntry("Wedding plan created.");
            return convert(response).getId();
        }

The test will fail, probably on a NullPointerException on a missing AuditService – and it is not visible why. InjectMocks will fail silently and there’s no indication the test needs this. Did I already ask whoever would design something like this to fail silently?

If you’re doing TDD or not (and we are able to change the test first) – clients of this code don’t know about an additional dependency, because it’s completely hidden. You shouldn’t use InjectMocks to deal with injecting private fields (err..or at all) , because this kind of Dependency Injection is evil – and signals you should change your design.

There, I said it.

Fix #1: Solve your design and make your dependencies visible.

Create a constructor. Pass along the PlannerClient.

@Service
public class PlannerServiceImpl implements PlannerService {
    private static final Logger LOG = LoggerFactory.getLogger(PlannerServiceImpl.class);

    private final PlannerClient plannerClient;

    @Autowired
    public PlannerServiceImpl(final PlannerClient plannerClient) {
        this.plannerClient = plannerClient;
    }

Now, when there are more dependencies needed, they’re clearly in sight because the constructor says so. So don’t go creating a bunch of setters now – they still don’t force you to pass along your required dependencies!

@Service
public class PlannerServiceImpl implements PlannerService {
    private static final Logger LOG = LoggerFactory.getLogger(PlannerServiceImpl.class);

    private final PlannerClient plannerClient;

    private final AuditService auditService;

    @Autowired
    PlannerServiceImpl(PlannerClient plannerClient, AuditService auditService) {
        this.plannerClient = plannerClient;
        this.auditService = auditService;
    }

The test itself won’t compile any more (luckily, because of the way we’ve been instantiating the field as plannerService = new PlannerServiceImpl()!) as soon as e.g. the AuditService is added to the constructor. So it’s time to..

Fix #2: Get rid of @InjectMocks

There’s no need to use @InjectMocks anymore. Instead instantiate the class-under-test properly in a @Before-annotated method – where it belongs, passing along all needed dependencies.

@RunWith(MockitoJUnitRunner.class)
public class PlannerServiceImplTest {

@Mock
private PlannerClient plannerClient;

@Mock
private AuditService auditService;

private PlannerServiceImpl plannerService;

@Before
void setUp() {
    plannerService = new PlannerServiceImpl(plannerClient, auditService);
}

Luckily, Lubos – which I mentioned earlier – completely independently came to the same conclusion in the mean time 🙂

Edit July 2020

Not long after this blog post, someone actually created an Mockito GitHub issue based on it, asking to have Mockito no longer silently fail. Apart from the issue itself, I value the response of one of the Mockito contributors explaining some of the history behind it.

“Nowhere in the javadoc, Spring, @Autowired, @Inject, etc is mentioned. And the shown behavior is documented. In that sense this blog post advocates to read the javadoc when one is using some product!

This feature was implemented when annotation based injection wasn’t even there in spring or in JEE.
And anyway the code being tested may even not live inside a Spring container, the code may even be legacy, i.e. without proper separation of concerns, the code may have various valid combination to initialize the bean (setters, fields), the code may have no tests at all. Still when one have to test these kind of code be it the full object or just a portion, it was kinda useful to have a framework that just did the boilerplate code.
Also note that so many Spring beans or maybe custom WhateverCompany beans have optional setters, i.e. they have default component.

So yes it fails silently, because Mockito is not able to confirm an object is correctly initialized or not when this object relies on fields/setters, it’s just impossible. And yes constructor injection is probably the best and the correct approach to dependency injection as the author even suggest (as a reminder @InjectMocks tries first to use constructor injection) ; I totally agree that the bean-y way (ie via fields/setters) to design object initialization is flawed.

The bottom line being @InjectMocks feature in mockito was never designed to be a full featured dependency injection mechanism and will probably never be. At best it’s shorthand way to inject mocks, but that’s all.”

Update: In the mean time I posted a sort of @InjectMocks follow-up.