How to use shouldNotAllowAnyObjectWithMixedVarargs method of org.mockito.internal.invocation.InvocationsFinderTest class

You may use the following:

model = Mockito.mock(Model.class);
final Map<Foo, List<Bar>> value = new HashMap<Foo, List<Bar>>();

Mockito.when(model.getValue()).thenAnswer(new Answer<Map<Foo, List<Bar>>> () {
  public Map<Foo, List<Bar>> answer(InvocationOnMock invocation) throws Throwable {
    return value;
  }
});

Above can be shortened using lambda as:

Mockito.when(model.getValue()).thenAnswer(invocationOnMock -> value)

TL-DR

• write plain unit tests for components that you can straightly test without loading a Spring container (run them in local and in CI build).

• write partial integration tests/slicing unit test for components that you cannot straightly test without loading a Spring container such as components related to JPA, controllers, REST clients, JDBC ... (run them in local and in CI build)

• write some full integration tests (end-to-end tests) for some high-level components where it brings values (run them in CI build).

---

3 main ways to test a component

• plain unit test (doesn't load a Spring container)
• full integration test (load a Spring container with all configuration and beans)
• partial integration test/ test slicing (load a Spring container with very restricted configurations and beans)

Can all components be tested in these 3 ways ?

In a general way with Spring any component can be tested in integration tests and only some kinds of components are suitable to be tested unitary(without container).
But note that with or without spring, unitary and integration tests are not opposed but complementary.

How to determine if a component can be plain tested (without spring) or only tested with Spring?

You recognize a code to test that doesn't have any dependencies from a Spring container as the component/method doesn't use Spring feature to perform its logical.
Take that FooService class :

@Service
public class FooService{

   private FooRepository fooRepository;
   
   public FooService(FooRepository fooRepository){
       this.fooRepository = fooRepository;
   }

   public long compute(...){
      List<Foo> foos = fooRepository.findAll(...);
       // core logic
      long result = 
           foos.stream()
               .map(Foo::getValue)
               .filter(v->...)
               .count();
       return result;
   }
}

FooService performs some computations and logic that don't need Spring to be executed.
Indeed with or without container the compute() method contains the core logic we want to assert.
Reversely you will have difficulties to test FooRepository without Spring as Spring Boot configures for you the datasource, the JPA context, and instrument your FooRepository interface to provide to it a default implementation and multiple other things.
Same thing for testing a controller (rest or MVC).
How could a controller be bound to an endpoint without Spring? How could the controller parse the HTTP request and generate an HTTP response without Spring? It simply cannot be done.

1)Writing a plain unit test

Using Spring Boot in your application doesn't mean that you need to load the Spring container for any test class you run.
As you write a test that doesn't need any dependencies from the Spring container, you don't have to use/load Spring in the test class.
Instead of using Spring you will instantiate yourself the class to test and if needed use a mock library to isolate the instance under test from its dependencies.
That is the way to follow because it is fast and favors the isolation of the tested component.
Here how to unit-test the FooService class presented above.
You just need to mock FooRepository to be able to test the logic of FooService.
With JUnit 5 and Mockito the test class could look like :

import org.mockito.junit.jupiter.MockitoExtension;
import org.mockito.Mock;
import org.mockito.Mockito;
import org.junit.jupiter.api.extension.ExtendWith;
import org.junit.jupiter.api.Assertions;
import org.junit.jupiter.api.BeforeEach;


@ExtendWith(MockitoExtension.class)
class FooServiceTest{

    FooService fooService;  

    @Mock
    FooRepository fooRepository;

    @BeforeEach 
    void init{
        fooService = new FooService(fooRepository);
    }

    @Test
    void compute(){
        List<Foo> fooData = ...;
        Mockito.when(fooRepository.findAll(...))
               .thenReturn(fooData);
        long actualResult = fooService.compute(...);
        long expectedResult = ...;
        Assertions.assertEquals(expectedResult, actualResult);
    }

}

2)Writing a full integration test

Writing an end-to-end test requires to load a container with the whole configuration and beans of the application.
To achieve that @SpringBootTest is the way :

The annotation works by creating the ApplicationContext used in your tests through SpringApplication

You can use it in this way to test it without any mock :

import org.springframework.beans.factory.annotation.Autowired;
import org.springframework.boot.test.context.SpringBootTest;
import org.junit.jupiter.api.Test;

@SpringBootTest
public class FooTest {

   @Autowired
   Foo foo;

   @Test
   public void doThat(){
      FooBar fooBar = foo.doThat(...);
      // assertion...
   }    
   
}

But you can also mock some beans of the container if it makes sense :

import org.springframework.beans.factory.annotation.Autowired;
import org.springframework.boot.test.mock.mockito.MockBean;
import org.springframework.boot.test.context.SpringBootTest;
import org.junit.jupiter.api.Test;
import org.mockito.Mockito;

@SpringBootTest
public class FooTest {

   @Autowired
   Foo foo;

   @MockBean
   private Bar barDep;

   @Test
   public void doThat(){
      Mockito.when(barDep.doThis()).thenReturn(...);
      FooBar fooBar = foo.doThat(...);
      // assertion...
   }    
   
}

Note the difference for mocking as you want to mock a plain instance of a Bar class (org.mockito.Mock annotation)and that you want to mock a Bar bean of the Spring context (org.springframework.boot.test.mock.mockito.MockBean annotation).

Full integration tests have to be executed by the CI builds

Loading a full spring context takes time. So you should be cautious with @SpringBootTest as this may make unit tests execution to be very long and generally you don't want to strongly slow down the local build on the developer's machine and the test feedback that matters to make the test writing pleasant and efficient for developers.
That's why "slow" tests are generally not executed on the developer's machines.
So you should make them integration tests (IT suffix instead of Test suffix in the naming of the test class) and make sure that these are executed only in the continuous integration builds.
But as Spring Boot acts on many things in your application (rest controllers, MVC controllers, JSON serialization/deserialization, persistence, and so for...) you could write many unit tests that are only executed on the CI builds and that is not fine either.
Having end-to-end tests executed only on the CI builds is ok but having also persistence, controllers or JSON tests executed only on the CI builds is not ok at all.
Indeed, the developer build will be fast but as drawback the tests execution in local will detect only a small part of the possible regressions...
To prevent this caveat, Spring Boot provides an intermediary way : partial integration test or the slice testing (as they call it) : the next point.

3)Writing a partial integration test focusing on a specific layer or concern thanks to slice testing

As explained in the point "Recognizing a test that can be plain tested (without spring))", some components can be tested only with a running container.
But why using @SpringBootTest that loads all beans and configurations of your application while you would need to load only a few specific configuration classes and beans to test these components?
For example why loading a full Spring JPA context (beans, configurations, in memory database, and so forth) to test the controller part?
And reversely why loading all configurations and beans associated to Spring controllers to test the JPA repository part?
Spring Boot addresses this point with the slice testing feature.
These are not as much as fast than plain unit tests (that is without container) but these are really much faster than loading a whole spring context. So executing them on the local machine is generally very acceptable.
Each slice testing flavor loads a very restricted set of auto-configuration classes that you can modify if needed according to your requirements.

Some common slice testing features :

• Auto-configured JSON Tests : @JsonTest

To test that object JSON serialization and deserialization is working as expected, you can use the @JsonTest annotation.

• Auto-configured Spring MVC Tests : @WebMvcTest

To test whether Spring MVC controllers are working as expected, use the @WebMvcTest annotation.

• Auto-configured Spring WebFlux Tests : @WebFluxTest

To test that Spring WebFlux controllers are working as expected, you can use the @WebFluxTest annotation.

• Auto-configured Data JPA Tests : @DataJpaTest

You can use the @DataJpaTest annotation to test JPA applications.

And you have still many other slice flavors that Spring Boot provides to you.
See the testing part of the documentation to get more details.
Note that if you need to define a specific set of beans to load that the built-in test slice annotations don't address, you can also create your own test slice annotation(https://spring.io/blog/2016/08/30/custom-test-slice-with-spring-boot-1-4).

4)Writing a partial integration test focusing on specific beans thanks to lazy bean initialization

Some days ago, I have encountered a case where I would test in partial integration a service bean that depends on several beans that themselves also depend on other beans. My problem was that two deep dependency beans have to be mocked for usual reasons (http requests and a query with large data in database).
Loading all the Spring Boot context looked an overhead, so I tried to load only specific beans. To achieve that, I annotation the test class with @SpringBootTest and I specified the classes attribute to define the configuration/beans classes to load.
After many tries I have gotten something that seemed working but I had to define an important list of beans/configurations to include.
That was really not neat nor maintainable.
So as clearer alternative, I chose to use the lazy bean initialization feature provided by Spring Boot 2.2 :

@SpringBootTest(properties="spring.main.lazy-initialization=true")
public class MyServiceTest { ...}

That has the advantage to load only beans used at runtime.
I don't think at all that using that property has to be the norm in test classes but in some specific test cases, that appears the right way.

There are several disadvantages to the when/thenReturn, when/thenThrow and when/then syntaxes. For example,

• In the case of when/thenReturn, if the return type is a generic with a wildcard, and you wish to return a mock of the same type, you will be unable to avoid a compile warning.
• You can't use when/thenThrow and when/then for a void method.
• You can't use these syntaxes on Mockito spies.
• You can only call when once for each combination of mock object, method and arguments, unless you call reset on the mock.
• Calling when multiple times for one combination of mock object and method, when you are using argument matchers, can lead to problems.

I find these cases difficult to remember. So instead of trying to keep track of when the when/thenReturn, when/thenThrow and when/then syntaxes will and won't work, I prefer to avoid them completely, in favour of the doReturn/when, doThrow/when and doAnswer/when alternatives. That is to say, since you'll occasionally need doReturn/when, doThrow/when and doAnswer/when, and you can ALWAYS use these methods, there is no point in learning how to use when/thenReturn, when/thenThrow and when/then.

Note that doReturn, doThrow and doAnswer can be chained together in the same way as thenReturn, thenThrow and then. What they don't have is an option for returning several values (or throwing several exceptions, or running several answers) within a single call to doReturn, doThrow and doAnswer. But I find that I need to do this so seldom, that it doesn't really matter.

There's one more disadvantage to doReturn, which I consider insignificant. You don't get compile time checking of the type of its argument, like you do with when/thenReturn. So if you get the argument type wrong, you won't find out until you run your test. Frankly, I don't care.

In summary then, I have been using Mockito for more than two years, and I consider the consistent use of doReturn, doThrow and doAnswer to be a Mockito best practice. Other Mockito users disagree.

when(mock.someMethod(any()).thenReturn(yourValue);

The any() matcher basically says you can have any value or a null. Check out the documentation at mockito, especially the section on Argument Matchers.

Use a Hamcrest matcher, as shown in the documentation:

when(carFinderMock.find(argThat(isRed()))).thenReturn(car1);

where isRed() is defined as

private Matcher<MappingFilter> isRed() {
    return new BaseMatcher<MappingFilter>() {
        // TODO implement abstract methods. matches() should check that the filter is RED.
    }
}

Try the anyVararg() matcher. This was introduced in 1.8.1.

I find the most robust way to deal with time-critical code is to wrap up all of your time-critical functions in their own class. I usually call it TimeHelper. So this class might look like the following.

import java.util.Date;
public class TimeHelper{
    public long currentTimeMillis(){
        return System.currentTimeMillis();
    }
    public Date makeDate(){
        return new Date();
    }
}

and it might have more methods of the same type. Now, any class that uses such functions should have (at least) two constructors - the normal one that you'll use in your application, plus a package-private one in which a TimeHelper is a parameter. This TimeHelper needs to be stored away for later use.

public class ClassThatDoesStuff {
    private ThingConnector connector;
    private TimeHelper timeHelper;

    public ClassThatDoesStuff(ThingConnector connector) {
        this(connector, new TimeHelper());
    }

    ClassThatDoesStuff(ThingConnector connector, TimeHelper timeHelper) {
        this.connector = connector;
        this.timeHelper = timeHelper;
    } 
}

Now, within your class, instead of writing System.currentTimeMillis(), write timeHelper.currentTimeMillis(). This will, of course, have exactly the same effect; except now, your class has magically become much more testable.

When you test your class, make a mock of TimeHelper. Configure this mock (using Mockito's when and thenReturn, or alternatively doReturn) to return any time values you like - whatever you need for your test. You can even return multiple values here, if you're going to have multiple calls to currentTimeMillis() in the course of the test.

Now use the second constructor to make the object that you're going to test, and pass in the mock. This gives you perfect control of what time values will be used in the test; and you can make your assertions or verifications assert that precisely the right value has been used.

public class ClassThatDoesStuffTest{
    @Mock private TimeHelper mockTime;
    @Mock private ThingConnector mockConnector;
    private ClassThatDoesStuff toTest;

    @Test
    public void doesSomething(){
        // Arrange
        initMocks(this);
        when(mockTime.currentTimeMillis()).thenReturn(1000L, 2000L, 5000L);
        toTest = new ClassThatDoesStuff(mockConnector, mockTime);

        // Act
        toTest.doSomething();

        // Assert
        // ... ???
    }
}        

If you do this, you know that your test will always work, and never be dependent on the time slicing policies of your operating system. You also have the power to verify the exact values of your timestamps, rather than asserting that they fall within some approximate interval.

Make sure you add @RunWith(PowerMockRunner.class) to the top of your class as well.

::edit:: two years later...

Don't ever use PowerMockito, you shouldn't need to.

If you do need to, you have most likely broken the SOLID principles and your design is wrong.

Fix your design instead.

You cant use when on getMethod, because getMethod is not a mock. It is still real object created by your class.

ArgumentCaptor has quite different purpose. Check section 15 here.

You could make your code more testable. Generally, classes that are creating new instances of other classes are difficult to test. Put some factory to this class for creating get/post methods, then in test mock this factory, and make it mock get/post methods.

public class YourClass {
  MethodFactory mf;

  public YourClass(MethodFactory mf) {
    this.mf = mf;
  }

  public void handleHttpClient(HttpClient httpClient) {
    httpClient.executeMethod(mf.createMethod());
    //your code here
  }
}

Then in test you can do:

HttpClient mockHttpClient = mock(HttpClient.class);
when(mockHttpClient.executeMethod(any(GetMethod.class)).thenReturn(HttpStatus.SC_OK);

MethodFactory factory = mock(MethodFactory.class);
GetMethod get = mock(GetMethod.class);
when(factory.createMethod()).thenReturn(get);
when(get.getResponseBodyAsStream()).thenReturn(new FileInputStream(source));

UPDATE

You can also try some nasty hack, and Answer and accessing GetMethod's private parts ;) by reflection. (This is really nasty hack)

when(mockHttpClient.executeMethod(any(GetMethod.class))).thenAnswer(new Answer() {
  Object answer(InvocationOnMock invocation) {
    GetMethod getMethod = (GetMethod) invocation.getArguments()[0];

    Field respStream = HttpMethodBase.class.getDeclaredField("responseStream");
    respStream.setAccessible(true);
    respStream.set(getMethod, new FileInputStream(source));

    return HttpStatus.SC_OK;
  }
});