I am trying to mock a call to a protected method of one of my classes:
import com.couchbase.client.java.view.Stale; // an enum
import com.google.common.base.Optional;
public class MyClass {
public List<String> myList(Optional<Integer> arg1, Optional<Stale> arg2) {
...
}
}
The mock shall be accomplished in the following way:
// Providing types for any() does not change anything
Mockito.when(myClass.myList(Mockito.any(), Mockito.any()).thenReturn(new ArrayList());
Whenever the previous line is executed the actual myList() method is called with null values for arg1 and arg2. Why is the method called, at all? After all, I am trying to avoid any executing thereof...
As Brice mentioned, if your myList method is final, then Java skips virtual method dispatch and will call the original object (not your mock).
If you are spying on an actual class, it is expected behavior that when will call the actual object as part of the stub: after all, in the expression when(foo.bar()) Java doesn't know anything special about when and assumes that it cares about the return value) of foo.bar(), not the call itself. (I walk through the stubbing process in the "Implementation details" section of my answer here.)
This syntax is better for spies:
doReturn(new ArrayList()).when(myClass).myList(any(), any());
Because this different when method receives an object, Mockito can prepare the object to do nothing during the stubbing, which avoids any spurious calls to your myList method.
Although Jeff's answer did not show a workaround for my problem it pointed me into the right direction.
After changing the mocking behaviour to doReturn... I suddenly got an error message. This message told me that myClass is not a mock which makes sense since you can only mock (or stub?) methods of mocked or spied objects. So as Jeff's answer indicates and is explained in the documentation of mockito I created a partial mock of MyClass with
MyClass myClass = Mockito.spy(new MyClass());
With this partial mock Jeff's approach to method mocking suddenly worked (mine still does not and should therefore be avoided).
So: Thank you, Jeff!
Related
We are trying to define a UnitTest where we mock an object which I here called x for simplicity:
...
import static org.mockito.ArgumentMatchers.any;
import static org.mockito.Mockito.doNothing;
import org.kubek2k.springockito.annotations.SpringockitoContextLoader;
import org.kubek2k.springockito.annotations.WrapWithSpy;
...
#ContextConfiguration(
loader = SpringockitoContextLoader.class,
inheritLocations = true)
public class SyncServiceIntegrationTest extends AbstractIntegrationTest {
#Autowired
#WrapWithSpy
private EventDrivenIssueDeliveryConfirmer x;
...
#Before
public void setUp() {
...
doNothing().when(x).foobar(any(Event.class));
}
...
i.e. we want our UT (not shown here) to later NOT call the method foobar on that object x.
Strange enough we get an NPE during initialization of this UT-class. The NPE is thrown by method foobar(), when the passed argument is null.
As turned out this call with argument null happens in the line doNothing()... in the setup-method which in our understanding is supposed to just define the mock-object's stubbing. But instead it evaluates the any(Event.class)-expression which apparently yields null and with that result it then calls the foobar(...)-method on x which causes the NPE.
Besides the NullPointerException we also get an error message from Mockito:
java.lang.NullPointerException: null
... <stack trace omitted for brevity>
org.mockito.exceptions.misusing.InvalidUseOfMatchersException:
Misplaced or misused argument matcher detected here:
-> at ch.sst.integration.SyncServiceIntegrationTest .setUp(SyncServiceIntegrationTest.java:69)
You cannot use argument matchers outside of verification or stubbing.
Examples of correct usage of argument matchers:
... <examples omitted for brevity>
org.mockito.exceptions.misusing.UnfinishedStubbingException:
Unfinished stubbing detected here:
-> at ch.sst.integration.SyncServiceIntegrationTest .setUp(SyncServiceIntegrationTest.java:69)
...
Why is that so??? Why is our stubbing considered "unfinished"? What are we missing here?
Later addition:
The issue seems to have to do with the fact that class
EventDrivenIssueDeliveryConfirmer is marked with #Transactional. Removing/commenting that lets the UT succeed. But of course that's no workaround - we need that annotation.
At least this provides a hint in which direction to search. The wrapping caused by #Transactional and the wrapping done by Mockito seem to step on each other's foot here.
I have the same issue but with a totally different setup: kotlin, mockito and, of course, mockito-kotlin.
I comment on this issue because maybe somebody will come to this question with the kotlin mockito problem in the future? I sure did. Anyhow.
When not declaring a method as open in kotlin it's compiled as a final method which can't be mocked by mockito-kotlin. As a result the method gets executed which to me is kind of weird but that's what it does. It's mentioned in the mockito-kotlin github issues under https://github.com/mockito/mockito-kotlin/issues/314
From the documentation:
A spy is always based on a real object. Hence you must provide a class type rather than an interface type, along with any constructor arguments for the type. If no constructor arguments are provided, the type’s default constructor will be used.
Method calls on a spy are automatically delegated to the real object. Likewise, values returned from the real object’s methods are passed back to the caller via the spy.
Also:
When stubbing a method on a spy, the real method no longer gets called:
subscriber.receive(_) >> "ok"
Instead of calling SubscriberImpl.receive, the receive method will now simply return "ok".
If a spy is just an interface layer between a real object and the caller, why not just use the real object? What does using a spy offer that using the real object or a Mock do not?
It seems to be in this void between a Mock and a real object to me.
Spies can be used in different scenarios. However, it is good if you can implement your tests without resorting to spies.
(Think twice before using this feature. It might be better to change the design of the code under specification.)
They can be used to verify that a method was called without mocking the method itself
You can stub out calls that you don't want to happen
You can use partial mocks to test the object itself
// this is now the object under specification, not a collaborator
def persister = Spy(MessagePersister) {
// stub a call on the same object
isPersistable(_) >> true
}
when:
persister.receive("msg")
then:
// demand a call on the same object
1 * persister.persist("msg")
Example and quote are from the docs # http://spockframework.org/spock/docs/1.1/all_in_one.html#Spies
In my practice I prefer to use a real objects as much as possible. In case when only one method is to be mocked I still use a real object but with overridden needed method:
MyDomainClass myRealObjectWithMockedMethod = new MyDomainClass() {
#Override
Object doSomething() {
return "hard coded or mocked result";
}
}
// test what you need
myRealObjectWithMockedMethod.action();
Note, this way works only of overridden method is not final. Otherwise Spy will help to define a behavior of this method.
A spy offers the possibility to use the original object but also mock out one method. For example you have a class where you want to test the implementation of the toString() method. But this calls an long running method which needs some external access like a database. In this case you use a spy and let your long running method return some test string and then use the toString from the original object.
Or like the spock example the method subscriber.receive maybe needs a server which sends out asynchronous messages. To write an test for subscriber not relying on the server or to handle asynchronous complexity you let the spy return ok and can easily test your methods which will rely on a server ok.
I have a db service save() method which allows method chaining:
#Service
public class Service {
...
public Service save(...) {
...
return this;
}
and this works just great as:
service.save(this).save(that).save(other);
When I come to mock it with Mockito though it breaks unless I use
Service serviceMock = mock(Service.class, RETURNS_DEEP_STUBS);
IIUC though, the use of RETURNS_DEEP_STUBS is considered bad. Is there a better way to mock a class with method call chaining?
Your pattern for save is very similar to a Builder pattern, which makes the question similar to "How to mock a builder with mockito" elsewhere on SO.
Per David Wallace's answer there, you can write an Answer that detects whether the mock is an instance of the return type of the method, and return the mock in only that case. This functionality was also built into the Mockito library as RETURNS_SELF in Mockito 2.0. As with any Answer, you can use this on any specific method call with thenAnswer or as the second parameter of mock to make it the default answer, but bear in mind the Mockito documentation warning that methods with generous return types (e.g. Object) will return the mock whether or not that was intended.
I would like to know is the thing in the description possible and how to do it.
I know you can call original method and then do the Answer like this:
when(presenter, "myMethod").doAnswer(<CUSTOMANSWER>)
but I would like to order them differently, first do CUSTOMANSWER and then call the original method.
For followers, it's actually possible to doAnswer and callRealMethod at the same time...
doAnswer(new Answer<Object>() {
public Object answer(InvocationOnMock invocationOnMock) throws Throwable {
invocationOnMock.callRealMethod(); // this one
return null;
}
}).when(subject).method(...);
You won't ever see when(...).doAnswer() in Mockito. Instead, you'll see either of the following, which includes the "call real method" behavior you're describing. As usual with Mockito stubbing, Mockito will select the most recent chain of calls that matches the method call and argument values in the invocation, and do each action in the chain once until the final action (which it will do for all calls afterwards.
// Normal Mockito syntax assuming "myMethod" is accessible. See caveat below.
when(presenter.myMethod()).thenAnswer(customAnswer).thenCallRealMethod();
// ...or...
doAnswer(customAnswer).doCallRealMethod().when(presenter).myMethod();
That said, there's a deficiency in the PowerMockito API that makes this difficult, because after the first doAnswer call all subsequent calls you get a normal Mockito Stubber instance rather than a PowerMockitoStubber instance. The bug 599 was misinterpreted, so for the time being you'll still have to make the cast yourself.
((PowerMockitoStubber) doAnswer(customAnswer).doCallRealMethod())
.when(presenter, "myMethod");
I need to run some code whenever a property value is retrieved, so naturally it made sense to define the getProperty method in my class. This method will get automatically called whenever a property value is retrieved. Here's roughly what I have in my class:
class MyClass
{
def getProperty(String name)
{
// Run some code ...
return this.#"${name}"
}
}
The problem with the above method occurs when someone tries to make the following call somewhere:
MyClass.class
This call ends up in the getProperty method looking for a property named "class", however, there is not actual property named "class" so we get a MissingFieldException.
What would be the correct way to implement running code whenever a property value is retrieved and deal with these kind of situtations.
Best is not to have a getProperty method if not needed. If you need one and you want to fall back on standard Groovy logic, then you can use return getMetaClass().getProperty(this, property), as can be found in GroovyObjectSupport. This will cover more than just fields.
This seems to be a common problem with this method. Map has the same issue. The developers of groovy got around the problem with Map by saying you need to use getClass() directly.