Exception "The mock object was garbage collected."

[basdebakker]

We received the following exception while running tests using Mockito, either version 3.0.0 or 3.1.0, and I'm filing this bug as requested:

Exception in thread "main" java.lang.IllegalStateException: The mock object was garbage collected. This should not happen in normal circumstances when using public API. Typically, the test class keeps strong reference to the mock object and it prevents getting the mock collected. Mockito internally needs to keep weak references to mock objects to avoid memory leaks for certain types of MockMaker implementations. If you see this exception using Mockito public API, please file a bug. For more information see issue #1313.
	at org.mockito.internal.invocation.mockref.MockWeakReference.get(MockWeakReference.java:32)
	at org.mockito.internal.invocation.InterceptedInvocation.getMock(InterceptedInvocation.java:106)
	at org.mockito.internal.stubbing.InvocationContainerImpl.invokedMock(InvocationContainerImpl.java:157)
	at org.mockito.internal.stubbing.OngoingStubbingImpl.<init>(OngoingStubbingImpl.java:22)
	at org.mockito.internal.handler.MockHandlerImpl.handle(MockHandlerImpl.java:83)
	at org.mockito.internal.handler.NullResultGuardian.handle(NullResultGuardian.java:29)
	at org.mockito.internal.handler.InvocationNotifierHandler.handle(InvocationNotifierHandler.java:35)
	at org.mockito.internal.creation.bytebuddy.MockMethodInterceptor.doIntercept(MockMethodInterceptor.java:61)
	at org.mockito.internal.creation.bytebuddy.MockMethodInterceptor.doIntercept(MockMethodInterceptor.java:49)
	at org.mockito.internal.creation.bytebuddy.MockMethodInterceptor$DispatcherDefaultingToRealMethod.interceptAbstract(MockMethodInterceptor.java:126)
	at org.mockito.codegen.Runnable$MockitoMock$1840149894.run(Unknown Source)
	at MockitoTest.runTest(MockitoTest.java:13)
	at MockitoTest.main(MockitoTest.java:6)

I've reduced the problem to the following test program:

import org.mockito.*;

public class MockitoTest {

  public static void main(String[] args) {
    while (true) runTest();
  }

  private static void runTest() {
    Node list = createList();
    while (list != null) {
      Node next = list.next;
      list.object.run();
      list = next;
    }
  }

  private static Node createList() {
    Node node = null;
    for (int i = 0; i < 1000; ++i) {
      Node next = new Node();
      next.next = node;
      node = next;
    }
    return node;
  }

  private static class Node {
    final Runnable object = Mockito.mock(Runnable.class);
    Node next;
  }
}

Whether the problem occurs depends on what exactly the JVM does with the code. It looks like the mock object can be garbage collected between the call to the mocked method and Mockito trying to use the weak reference to it. I'm using AdoptOpenJDK 11.0.3.7-hotspot on Windows. My test program more often than not reproduces the exception within a few seconds, but occasionally it can run forever without failing.

[tokuhirom]

I tried bisecting. This issue happens after this PR.
#1544

[raphw]

That makes sence, if the BaseStubbing instance wraps the mocks and the latter instance is referenced via a weak reference then that latter instance can be GC-ed even if the actual mock instance is still in use.

Thanks for the test, since we both have a test for the "too weak" and the "not sufficiently weak" mock applications, I hope that I can come up with a scenario that satisfies both conditions.

[ghackett]

This issue just came up as we tried to upgrade from 3.3.3 to either 3.4.6 or 3.5.9. A large chunk of our unit tests became flaky as a result. Holding onto a strong reference of MockitoAnnotations.openMocks() does not seem to have any impact. The only thing that seemed to rectify the issue was creating and retaining a MockitoSession like via MockitoJUnit.rule() (Using the MockspressoJUnitRunner shows the same flakiness as no session is initialized in it).

Considering that Mockito.mockitoSession() is labeled as optional and @Incubating, and considering that the junit rule & TestRunner have always been optional historically, this seems like a legitimate bug that will block mockito upgrades in many large codebases (especially those relying on mockito-kotlin's inline mock() method).

[raphw]

Yes, this needs to be addressed. The alternatives right now are unfortunately a leak or premature discarding. I think a leak is the better alternative, though since you can explicitly clear Mockito's mock store.

The updates within 3 should not have an impact but can have one as a result of changed GC cycles.

I will try to solve this one the next time I find some free time to work on Mockito.

[ghackett]

Awesome! Thanks so much for addressing this!