Interaction-based testing is a design and testing technique that emerged in the Extreme Programming (XP) community in the early 2000’s. Focusing on the behavior of objects rather than their state, it explores how the object(s) under specification interact, by way of method calls, with their collaborators.

For example, suppose we have a Publisher that sends messages to its `Subscriber`s:

class Publisher {
  List<Subscriber> subscribers = []
  int messageCount = 0
  void send(String message){
    subscribers*.receive(message)
    messageCount++
  }
}

interface Subscriber {
  void receive(String message)
}

class PublisherSpec extends Specification {
  Publisher publisher = new Publisher()
}

How are we going to test Publisher? With state-based testing, we can verify that the publisher keeps track of its subscribers. The more interesting question, though, is whether a message sent by the publisher is received by the subscribers. To answer this question, we need a special implementation of Subscriber that listens in on the conversation between the publisher and its subscribers. Such an implementation is called a mock object.

While we could certainly create a mock implementation of Subscriber by hand, writing and maintaining this code can get unpleasant as the number of methods and complexity of interactions increases. This is where mocking frameworks come in: They provide a way to describe the expected interactions between an object under specification and its collaborators, and can generate mock implementations of collaborators that verify these expectations.

How Are Mock Implementations Generated?

Like most Java mocking frameworks, Spock uses JDK dynamic proxies (when mocking interfaces) and Byte Buddy or CGLIB proxies (when mocking classes) to generate mock implementations at runtime. Compared to implementations based on Groovy meta-programming, this has the advantage that it also works for testing Java code.

The Java world has no shortage of popular and mature mocking frameworks: JMock, EasyMock, Mockito, to name just a few. Although each of these tools can be used together with Spock, we decided to roll our own mocking framework, tightly integrated with Spock’s specification language. This decision was driven by the desire to leverage all of Groovy’s capabilities to make interaction-based tests easier to write, more readable, and ultimately more fun. We hope that by the end of this chapter, you will agree that we have achieved these goals.

Except where indicated, all features of Spock’s mocking framework work both for testing Java and Groovy code.

Creating Mock Objects

Mock objects are created with the MockingApi.Mock() method.[1] Let’s create two mock subscribers:

def subscriber = Mock(Subscriber)
def subscriber2 = Mock(Subscriber)

Alternatively, the following Java-like syntax is supported, which may give better IDE support:

Subscriber subscriber = Mock()
Subscriber subscriber2 = Mock()

Here, the mock’s type is inferred from the variable type on the left-hand side of the assignment.

Note
If the mock’s type is given on the left-hand side of the assignment, it’s permissible (though not required) to omit it on the right-hand side.

Mock objects literally implement (or, in the case of a class, extend) the type they stand in for. In other words, in our example subscriber is-a Subscriber. Hence it can be passed to statically typed (Java) code that expects this type.

Default Behavior of Mock Objects

Lenient vs. Strict Mocking Frameworks

Like Mockito, we firmly believe that a mocking framework should be lenient by default. This means that unexpected method calls on mock objects (or, in other words, interactions that aren’t relevant for the test at hand) are allowed and answered with a default response. Conversely, mocking frameworks like EasyMock and JMock are strict by default, and throw an exception for every unexpected method call. While strictness enforces rigor, it can also lead to over-specification, resulting in brittle tests that fail with every other internal code change. Spock’s mocking framework makes it easy to describe only what’s relevant about an interaction, avoiding the over-specification trap.

Initially, mock objects have no behavior. Calling methods on them is allowed but has no effect other than returning the default value for the method’s return type (false, 0, or null). An exception are the Object.equals, Object.hashCode, and Object.toString methods, which have the following default behavior: A mock object is only equal to itself, has a unique hash code, and a string representation that includes the name of the type it represents. This default behavior is overridable by stubbing the methods, which we will learn about in the Stubbing section.

Injecting Mock Objects into Code Under Specification

After creating the publisher and its subscribers, we need to make the latter known to the former:

class PublisherSpec extends Specification {
  Publisher publisher = new Publisher()
  Subscriber subscriber = Mock()
  Subscriber subscriber2 = Mock()

  def setup() {
    publisher.subscribers << subscriber // << is a Groovy shorthand for List.add()
    publisher.subscribers << subscriber2
  }

We are now ready to describe the expected interactions between the two parties.

Mocking

Mocking is the act of describing (mandatory) interactions between the object under specification and its collaborators. Here is an example:

def "should send messages to all subscribers"() {
  when:
  publisher.send("hello")

  then:
  1 * subscriber.receive("hello")
  1 * subscriber2.receive("hello")
}

Read out aloud: "When the publisher sends a 'hello' message, then both subscribers should receive that message exactly once."

When this feature method gets run, all invocations on mock objects that occur while executing the when block will be matched against the interactions described in the then: block. If one of the interactions isn’t satisfied, a (subclass of) InteractionNotSatisfiedError will be thrown. This verification happens automatically and does not require any additional code.

Interactions

Is an Interaction Just a Regular Method Invocation?

Not quite. While an interaction looks similar to a regular method invocation, it is simply a way to express which method invocations are expected to occur. A good way to think of an interaction is as a regular expression that all incoming invocations on mock objects are matched against. Depending on the circumstances, the interaction may match zero, one, or multiple invocations.

Let’s take a closer look at the then: block. It contains two interactions, each of which has four distinct parts: a cardinality, a target constraint, a method constraint, and an argument constraint:

1 * subscriber.receive("hello")
|   |          |       |
|   |          |       argument constraint
|   |          method constraint
|   target constraint
cardinality

Cardinality

The cardinality of an interaction describes how often a method call is expected. It can either be a fixed number or a range:

1 * subscriber.receive("hello")      // exactly one call
0 * subscriber.receive("hello")      // zero calls
(1..3) * subscriber.receive("hello") // between one and three calls (inclusive)
(1.._) * subscriber.receive("hello") // at least one call
(_..3) * subscriber.receive("hello") // at most three calls
_ * subscriber.receive("hello")      // any number of calls, including zero
                                     // (rarely needed; see 'Strict Mocking')

Target Constraint

The target constraint of an interaction describes which mock object is expected to receive the method call:

1 * subscriber.receive("hello") // a call to 'subscriber'
1 * _.receive("hello")          // a call to any mock object

Method Constraint

The method constraint of an interaction describes which method is expected to be called:

1 * subscriber.receive("hello") // a method named 'receive'
1 * subscriber./r.*e/("hello")  // a method whose name matches the given regular expression
                                // (here: method name starts with 'r' and ends in 'e')

When expecting a call to a getter method, Groovy property syntax can be used instead of method syntax:

1 * subscriber.status // same as: 1 * subscriber.getStatus()

When expecting a call to a setter method, only method syntax can be used:

1 * subscriber.setStatus("ok") // NOT: 1 * subscriber.status = "ok"

Argument Constraints

The argument constraints of an interaction describe which method arguments are expected:

1 * subscriber.receive("hello")        // an argument that is equal to the String "hello"
1 * subscriber.receive(!"hello")       // an argument that is unequal to the String "hello"
1 * subscriber.receive()               // the empty argument list (would never match in our example)
1 * subscriber.receive(_)              // any single argument (including null)
1 * subscriber.receive(*_)             // any argument list (including the empty argument list)
1 * subscriber.receive(!null)          // any non-null argument
1 * subscriber.receive(_ as String)    // any non-null argument that is-a String
1 * subscriber.receive(endsWith("lo")) // an argument matching the given Hamcrest matcher
                                       // a String argument ending with "lo" in this case
1 * subscriber.receive({ it.size() > 3 && it.contains('a') })
// an argument that satisfies the given predicate, meaning that
// code argument constraints need to return true of false
// depending on whether they match or not
// (here: message length is greater than 3 and contains the character a)

Argument constraints work as expected for methods with multiple arguments:

1 * process.invoke("ls", "-a", _, !null, { ["abcdefghiklmnopqrstuwx1"].contains(it) })

When dealing with vararg methods, vararg syntax can also be used in the corresponding interactions:

interface VarArgSubscriber {
    void receive(String... messages)
}

...

subscriber.receive("hello", "goodbye")
Spock Deep Dive: Groovy Varargs

Groovy allows any method whose last parameter has an array type to be called in vararg style. Consequently, vararg syntax can also be used in interactions matching such methods.

Equality Constraint

The equality constraint uses groovy equality to check the argument, i.e, argument == constraint. You can use

  • any literal 1 * check('string') / 1 * check(1) / 1 * check(null),

  • a variable 1 * check(var),

  • a list or map literal 1 * check([1]) / 1 * check([foo: 'bar']),

  • an object 1 * check(new Person('sam')),

  • or the result of a method call 1 * check(person())

as an equality constraint.

Hamcrest Constraint

A variation of the equality constraint, if the constraint object is a Hamcrest matcher, then it will use that matcher to check the argument.

Wildcard Constraint

The wildcard constraint will match any argument null or otherwise. It is the , i.e. 1 * subscriber.receive(). There is also the spread wildcard constraint *_ which matches any number of arguments 1 * subscriber.receive(*_) including none.

Code Constraint

The code constraint is the most versatile of all. It is a groovy closure that gets the argument as its parameter. The closure is treated as an condition block, so it behaves like a then block, i.e., every line is treated as an implicit assertion. It can emulate all but the spread wildcard constraint, however it is suggested to use the simpler constraints where possible. You can do multiple assertions, call methods for assertions, or use with/verifyAll.

1 * list.add({
  verifyAll(it, Person) {
    firstname == 'William'
    lastname == 'Kirk'
    age == 45
  }
})

Negating Constraint

The negating constraint ! is a compound constraint, i.e. it needs to be combined with another constraint to work. It inverts the result of the nested constraint, e.g, 1 * subscriber.receive(!null) is the combination of an equality constraint checking for null and then the negating constraint inverting the result, turning it into not null.

Although it can be combined with any other constraint it does not always make sense, e.g., 1 * subscriber.receive(!_) will match nothing. Also keep in mind that the diagnostics for a non matching negating constraint will just be that the inner constraint did match, without any more information.

Type Constraint

The type constraint checks for the type/class of the argument, like the negating constraint it is also a compound constraint. It usually written as _ as Type, which is a combination of the wildcard constraint and the type constraint. You can combined it with other constraints as well, 1 * subscriber.receive({ it.contains('foo')} as String) will assert that it is a String before executing the code constraint to check if it contains foo.

Matching Any Method Call

Sometimes it can be useful to match "anything", in some sense of the word:

1 * subscriber._(*_)     // any method on subscriber, with any argument list
1 * subscriber._         // shortcut for and preferred over the above

1 * _._                  // any method call on any mock object
1 * _                    // shortcut for and preferred over the above
Note
Although (_.._) * _._(*_) >> _ is a valid interaction declaration, it is neither good style nor particularly useful.

Strict Mocking

Now, when would matching any method call be useful? A good example is strict mocking, a style of mocking where no interactions other than those explicitly declared are allowed:

when:
publisher.publish("hello")

then:
1 * subscriber.receive("hello") // demand one 'receive' call on 'subscriber'
_ * auditing._                  // allow any interaction with 'auditing'
0 * _                           // don't allow any other interaction

0 * only makes sense as the last interaction of a then: block or method. Note the use of _ * (any number of calls), which allows any interaction with the auditing component.

Note
_ * is only meaningful in the context of strict mocking. In particular, it is never necessary when Stubbing an invocation. For example, _ * auditing.record() >> "ok" can (and should!) be simplified to auditing.record() >> "ok".

Where to Declare Interactions

So far, we declared all our interactions in a then: block. This often results in a spec that reads naturally. However, it is also permissible to put interactions anywhere before the when: block that is supposed to satisfy them. In particular, this means that interactions can be declared in a setup method. Interactions can also be declared in any "helper" instance method of the same specification class.

When an invocation on a mock object occurs, it is matched against interactions in the interactions' declared order. If an invocation matches multiple interactions, the earliest declared interaction that hasn’t reached its upper invocation limit will win. There is one exception to this rule: Interactions declared in a then: block are matched against before any other interactions. This allows to override interactions declared in, say, a setup method with interactions declared in a then: block.

Spock Deep Dive: How Are Interactions Recognized?

In other words, what makes an expression an interaction declaration, rather than, say, a regular method call? Spock uses a simple syntactic rule to recognize interactions: If an expression is in statement position and is either a multiplication (*) or a right-shift (>>, >>>) operation, then it is considered an interaction and will be parsed accordingly. Such an expression would have little to no value in statement position, so changing its meaning works out fine. Note how the operations correspond to the syntax for declaring a cardinality (when mocking) or a response generator (when stubbing). Either of them must always be present; foo.bar() alone will never be considered an interaction.

Declaring Interactions at Mock Creation Time

If a mock has a set of "base" interactions that don’t vary, they can be declared right at mock creation time:

Subscriber subscriber = Mock {
   1 * receive("hello")
   1 * receive("goodbye")
}

This feature is particularly attractive for Stubbing and with dedicated Stubs. Note that the interactions don’t (and cannot [2]) have a target constraint; it’s clear from the context which mock object they belong to.

Interactions can also be declared when initializing an instance field with a mock:

class MySpec extends Specification {
    Subscriber subscriber = Mock {
        1 * receive("hello")
        1 * receive("goodbye")
    }
}

Grouping Interactions with Same Target

Interactions sharing the same target can be grouped in a Specification.with block. Similar to Declaring Interactions at Mock Creation Time, this makes it unnecessary to repeat the target constraint:

with(subscriber) {
    1 * receive("hello")
    1 * receive("goodbye")
}

A with block can also be used for grouping conditions with the same target.

Mixing Interactions and Conditions

A then: block may contain both interactions and conditions. Although not strictly required, it is customary to declare interactions before conditions:

when:
publisher.send("hello")

then:
1 * subscriber.receive("hello")
publisher.messageCount == 1

Read out aloud: "When the publisher sends a 'hello' message, then the subscriber should receive the message exactly once, and the publisher’s message count should be one."

Explicit Interaction Blocks

Internally, Spock must have full information about expected interactions before they take place. So how is it possible for interactions to be declared in a then: block? The answer is that under the hood, Spock moves interactions declared in a then: block to immediately before the preceding when: block. In most cases this works out just fine, but sometimes it can lead to problems:

when:
publisher.send("hello")

then:
def message = "hello"
1 * subscriber.receive(message)

Here we have introduced a variable for the expected argument. (Likewise, we could have introduced a variable for the cardinality.) However, Spock isn’t smart enough (huh?) to tell that the interaction is intrinsically linked to the variable declaration. Hence it will just move the interaction, which will cause a MissingPropertyException at runtime.

One way to solve this problem is to move (at least) the variable declaration to before the when: block. (Fans of Data Driven Testing might move the variable into a where: block.) In our example, this would have the added benefit that we could use the same variable for sending the message.

Another solution is to be explicit about the fact that variable declaration and interaction belong together:

when:
publisher.send("hello")

then:
interaction {
  def message = "hello"
  1 * subscriber.receive(message)
}

Since an MockingApi.interaction block is always moved in its entirety, the code now works as intended.

Scope of Interactions

Interactions declared in a then: block are scoped to the preceding when: block:

when:
publisher.send("message1")

then:
1 * subscriber.receive("message1")

when:
publisher.send("message2")

then:
1 * subscriber.receive("message2")

This makes sure that subscriber receives "message1" during execution of the first when: block, and "message2" during execution of the second when: block.

Interactions declared outside a then: block are active from their declaration until the end of the containing feature method.

Interactions are always scoped to a particular feature method. Hence they cannot be declared in a static method, setupSpec method, or cleanupSpec method. Likewise, mock objects should not be stored in static or @Shared fields.

Verification of Interactions

There are two main ways in which a mock-based test can fail: An interaction can match more invocations than allowed, or it can match fewer invocations than required. The former case is detected right when the invocation happens, and causes a TooManyInvocationsError:

Too many invocations for:

2 * subscriber.receive(_) (3 invocations)

To make it easier to diagnose why too many invocations matched, Spock will show all invocations matching the interaction in question:

Matching invocations (ordered by last occurrence):

2 * subscriber.receive("hello")   <-- this triggered the error
1 * subscriber.receive("goodbye")

According to this output, one of the receive("hello") calls triggered the TooManyInvocationsError. Note that because indistinguishable calls like the two invocations of subscriber.receive("hello") are aggregated into a single line of output, the first receive("hello") may well have occurred before the receive("goodbye").

The second case (fewer invocations than required) can only be detected once execution of the when block has completed. (Until then, further invocations may still occur.) It causes a TooFewInvocationsError:

Too few invocations for:

1 * subscriber.receive("hello") (0 invocations)

Note that it doesn’t matter whether the method was not called at all, the same method was called with different arguments, the same method was called on a different mock object, or a different method was called "instead" of this one; in either case, a TooFewInvocationsError error will occur.

To make it easier to diagnose what happened "instead" of a missing invocation, Spock will show all invocations that didn’t match any interaction, ordered by their similarity with the interaction in question. In particular, invocations that match everything but the interaction’s arguments will be shown first:

Unmatched invocations (ordered by similarity):

1 * subscriber.receive("goodbye")
1 * subscriber2.receive("hello")

Invocation Order

Often, the exact method invocation order isn’t relevant and may change over time. To avoid over-specification, Spock defaults to allowing any invocation order, provided that the specified interactions are eventually satisfied:

then:
2 * subscriber.receive("hello")
1 * subscriber.receive("goodbye")

Here, any of the invocation sequences "hello" "hello" "goodbye", "hello" "goodbye" "hello", and "goodbye" "hello" "hello" will satisfy the specified interactions.

In those cases where invocation order matters, you can impose an order by splitting up interactions into multiple then: blocks:

then:
2 * subscriber.receive("hello")

then:
1 * subscriber.receive("goodbye")

Now Spock will verify that both "hello"'s are received before the "goodbye". In other words, invocation order is enforced between but not within then: blocks.

Note
Splitting up a then: block with and: does not impose any ordering, as and: is only meant for documentation purposes and doesn’t carry any semantics.

Mocking Classes

Besides interfaces, Spock also supports mocking of classes. Mocking classes works just like mocking interfaces; the only additional requirement is to put byte-buddy 1.9+ or cglib-nodep 3.2.0+ on the class path.

When using for example

  • normal Mocks or Stubs or

  • Spys that are configured with useObjenesis: true or

  • Spys that spy on a concrete instance like Spy(myInstance)

it is also necessary to put objenesis 3.0+ on the class path, except for classes with accessible no-arg constructor or configured constructorArgs unless the constructor call should not be done, for example to avoid unwanted side effects.

If either of these libraries is missing from the class path, Spock will gently let you know.

Stubbing

Stubbing is the act of making collaborators respond to method calls in a certain way. When stubbing a method, you don’t care if and how many times the method is going to be called; you just want it to return some value, or perform some side effect, whenever it gets called.

For the sake of the following examples, let’s modify the Subscriber's receive method to return a status code that tells if the subscriber was able to process a message:

interface Subscriber {
    String receive(String message)
}

Now, let’s make the receive method return "ok" on every invocation:

subscriber.receive(_) >> "ok"

Read out aloud: "Whenever the subscriber receives a message, make it respond with 'ok'."

Compared to a mocked interaction, a stubbed interaction has no cardinality on the left end, but adds a response generator on the right end:

subscriber.receive(_) >> "ok"
|          |       |     |
|          |       |     response generator
|          |       argument constraint
|          method constraint
target constraint

A stubbed interaction can be declared in the usual places: either inside a then: block, or anywhere before a when: block. (See Where to Declare Interactions for the details.) If a mock object is only used for stubbing, it’s common to declare interactions at mock creation time or in a given: block.

Returning Fixed Values

We have already seen the use of the right-shift (>>) operator to return a fixed value:

subscriber.receive(_) >> "ok"

To return different values for different invocations, use multiple interactions:

subscriber.receive("message1") >> "ok"
subscriber.receive("message2") >> "fail"

This will return "ok" whenever "message1" is received, and "fail" whenever "message2" is received. There is no limit as to which values can be returned, provided they are compatible with the method’s declared return type.

Returning Sequences of Values

To return different values on successive invocations, use the triple-right-shift (>>>) operator:

subscriber.receive(_) >>> ["ok", "error", "error", "ok"]

This will return "ok" for the first invocation, "error" for the second and third invocation, and "ok" for all remaining invocations. The right-hand side must be a value that Groovy knows how to iterate over; in this example, we’ve used a plain list.

Computing Return Values

To compute a return value based on the method’s argument, use the the right-shift (>>) operator together with a closure. If the closure declares a single untyped parameter, it gets passed the method’s argument list:

subscriber.receive(_) >> { args -> args[0].size() > 3 ? "ok" : "fail" }

Here "ok" gets returned if the message is more than three characters long, and "fail" otherwise.

In most cases it would be more convenient to have direct access to the method’s arguments. If the closure declares more than one parameter or a single typed parameter, method arguments will be mapped one-by-one to closure parameters:[3]

subscriber.receive(_) >> { String message -> message.size() > 3 ? "ok" : "fail" }

This response generator behaves the same as the previous one, but is arguably more readable.

If you find yourself in need of more information about a method invocation than its arguments, have a look at org.spockframework.mock.IMockInvocation. All methods declared in this interface are available inside the closure, without a need to prefix them. (In Groovy terminology, the closure delegates to an instance of IMockInvocation.)

Performing Side Effects

Sometimes you may want to do more than just computing a return value. A typical example is throwing an exception. Again, closures come to the rescue:

subscriber.receive(_) >> { throw new InternalError("ouch") }

Of course, the closure can contain more code, for example a println statement. It will get executed every time an incoming invocation matches the interaction.

Chaining Method Responses

Method responses can be chained:

subscriber.receive(_) >>> ["ok", "fail", "ok"] >> { throw new InternalError() } >> "ok"

This will return "ok", "fail", "ok" for the first three invocations, throw InternalError for the fourth invocations, and return ok for any further invocation.

Returning a default response

If you don’t really care what you return, but you must return a non-null value, you can use _. This will use the same logic to compute a response as Stub (see Stubs), so it is only really useful for Mock and Spy instances.

subscriber.receive(_) >> _

You can of course use this with chaining as well. Here it might be useful for Stub instances as well.

subscriber.receive(_) >>> ["ok", "fail"] >> _ >> "ok"

Combining Mocking and Stubbing

Mocking and stubbing go hand-in-hand:

1 * subscriber.receive("message1") >> "ok"
1 * subscriber.receive("message2") >> "fail"

When mocking and stubbing the same method call, they have to happen in the same interaction. In particular, the following Mockito-style splitting of stubbing and mocking into two separate statements will not work:

given:
subscriber.receive("message1") >> "ok"

when:
publisher.send("message1")

then:
1 * subscriber.receive("message1")

As explained in Where to Declare Interactions, the receive call will first get matched against the interaction in the then: block. Since that interaction doesn’t specify a response, the default value for the method’s return type (null in this case) will be returned. (This is just another facet of Spock’s lenient approach to mocking.). Hence, the interaction in the given: block will never get a chance to match.

Note
Mocking and stubbing of the same method call has to happen in the same interaction.

Other Kinds of Mock Objects

So far, we have created mock objects with the MockingApi.Mock method. Aside from this method, the MockingApi class provides a couple of other factory methods for creating more specialized kinds of mock objects.

Stubs

A stub is created with the MockingApi.Stub factory method:

Subscriber subscriber = Stub()

Whereas a mock can be used both for stubbing and mocking, a stub can only be used for stubbing. Limiting a collaborator to a stub communicates its role to the readers of the specification.

Note
If a stub invocation matches a mandatory interaction (like 1 * foo.bar()), an InvalidSpecException is thrown.

Like a mock, a stub allows unexpected invocations. However, the values returned by a stub in such cases are more ambitious:

  • For primitive types, the primitive type’s default value is returned.

  • For non-primitive numerical values (such as BigDecimal), zero is returned.

  • If the value is assignable from the stub instance, then the instance is returned (e.g. builder pattern)

  • For non-numerical values, an "empty" or "dummy" object is returned. This could mean an empty String, an empty collection, an object constructed from its default constructor, or another stub returning default values. See class org.spockframework.mock.EmptyOrDummyResponse for the details.

Note
If the response type of the method is a final class or if it requires a class-mocking library and cglib or ByteBuddy are not available, then the "dummy" object creation will fail with a CannotCreateMockException.

A stub often has a fixed set of interactions, which makes declaring interactions at mock creation time particularly attractive:

Subscriber subscriber = Stub {
    receive("message1") >> "ok"
    receive("message2") >> "fail"
}

Spies

(Think twice before using this feature. It might be better to change the design of the code under specification.)

A spy is created with the MockingApi.Spy factory method:

SubscriberImpl subscriber = Spy(constructorArgs: ["Fred"])

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 no-arg constructor will be used.

If the given constructor arguments lead to an ambiguity, you can cast the constructor arguments as usual using as or Java-style cast. If the testee for example has one constructor with a String parameter and one with a Pattern parameter and you want null as constructorArg:

SubscriberImpl subscriber = Spy(constructorArgs: [null as String])
SubscriberImpl subscriber2 = Spy(constructorArgs: [(Pattern) null])

You may also create a spy from an instantiated object. This may be useful in cases where you do not have full control over the instantiation of types you are interested in spying. (For example when testing within a Dependency Injection framework such as Spring or Guice.)

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.

After creating a spy, you can listen in on the conversation between the caller and the real object underlying the spy:

1 * subscriber.receive(_)

Apart from making sure that receive gets called exactly once, the conversation between the publisher and the SubscriberImpl instance underlying the spy remains unaltered.

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".

Sometimes, it is desirable to both execute some code and delegate to the real method:

subscriber.receive(_) >> { String message -> callRealMethod(); message.size() > 3 ? "ok" : "fail" }

Here we use callRealMethod() to delegate the method invocation to the real object. Note that we don’t have to pass the message argument along; this is taken care of automatically. callRealMethod() returns the real invocation’s result, but in this example we opted to return our own result instead. If we had wanted to pass a different message to the real method, we could have used callRealMethodWithArgs("changed message").

Please note that while semantically both callRealMethod() and callRealMethodWithArgs(…​) only make sense with spies, technically you can also call these methods on mock or stub objects, kind of turning them into (pseudo) spy objects "through the backdoor". The only precondition is that the mocked/stubbed object actually has a real method implementation, i.e. for interface mocks there must be a default method, for class mocks there must be a (non-abstract) original method.

Partial Mocks

(Think twice before using this feature. It might be better to change the design of the code under specification.)

Spies can also be used as partial mocks:

// this is now the object under specification, not a collaborator
MessagePersister persister = Spy {
  // 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")

Groovy Mocks

So far, all the mocking features we have seen work the same no matter if the calling code is written in Java or Groovy. By leveraging Groovy’s dynamic capabilities, Groovy mocks offer some additional features specifically for testing Groovy code. They are created with the MockingApi.GroovyMock(), MockingApi.GroovyStub(), and MockingApi.GroovySpy() factory methods.

Tip
When Should Groovy Mocks be Favored over Regular Mocks? Groovy mocks should be used when the code under specification is written in Groovy and some of the unique Groovy mock features are needed. When called from Java code, Groovy mocks will behave like regular mocks. Note that it isn’t necessary to use a Groovy mock merely because the code under specification and/or mocked type is written in Groovy. Unless you have a concrete reason to use a Groovy mock, prefer a regular mock.

Mocking Dynamic Methods

All Groovy mocks implement the GroovyObject interface. They support the mocking and stubbing of dynamic methods as if they were physically declared methods:

Subscriber subscriber = GroovyMock()

1 * subscriber.someDynamicMethod("hello")

Mocking All Instances of a Type

(Think twice before using this feature. It might be better to change the design of the code under specification.)

Usually, Groovy mocks need to be injected into the code under specification just like regular mocks. However, when a Groovy mock is created as global, it automagically replaces all real instances of the mocked type for the duration of the feature method:[4]

def publisher = new Publisher()
publisher << new RealSubscriber() << new RealSubscriber()

RealSubscriber anySubscriber = GroovyMock(global: true)

when:
publisher.publish("message")

then:
2 * anySubscriber.receive("message")

Here, we set up the publisher with two instances of a real subscriber implementation. Then we create a global mock of the same type. This reroutes all method calls on the real subscribers to the mock object. The mock object’s instance isn’t ever passed to the publisher; it is only used to describe the interaction.

Note
A global mock can only be created for a class type. It effectively replaces all instances of that type for the duration of the feature method.

Since global mocks have a somewhat, well, global effect, it’s often convenient to use them together with GroovySpy. This leads to the real code getting executed unless an interaction matches, allowing you to selectively listen in on objects and change their behavior just where needed.

How Are Global Groovy Mocks Implemented?

Global Groovy mocks get their super powers from Groovy meta-programming. To be more precise, every globally mocked type is assigned a custom meta class for the duration of the feature method. Since a global Groovy mock is still based on a CGLIB proxy, it will retain its general mocking capabilities (but not its super powers) when called from Java code.

Mocking Constructors

(Think twice before using this feature. It might be better to change the design of the code under specification.)

Global mocks support mocking of constructors:

RealSubscriber anySubscriber = GroovySpy(global: true)

1 * new RealSubscriber("Fred")

Since we are using a spy, the object returned from the constructor call remains unchanged. To change which object gets constructed, we can stub the constructor:

new RealSubscriber("Fred") >> new RealSubscriber("Barney")

Now, whenever some code tries to construct a subscriber named Fred, we’ll construct a subscriber named Barney instead.

Mocking Static Methods

(Think twice before using this feature. It might be better to change the design of the code under specification.)

Global mocks support mocking and stubbing of static methods:

RealSubscriber anySubscriber = GroovySpy(global: true)

1 * RealSubscriber.someStaticMethod("hello") >> 42

The same works for dynamic static methods.

When a global mock is used solely for mocking constructors and static methods, the mock’s instance isn’t really needed. In such a case one can just write:

GroovySpy(RealSubscriber, global: true)

Advanced Features

Most of the time you shouldn’t need these features. But if you do, you’ll be glad to have them.

A la Carte Mocks

At the end of the day, the Mock(), Stub(), and Spy() factory methods are just canned ways to create mock objects with a certain configuration. If you want more fine-grained control over a mock’s configuration, have a look at the org.spockframework.mock.IMockConfiguration interface. All properties of this interface [5] can be passed as named arguments to the Mock() method. For example:

def person = Mock(name: "Fred", type: Person, defaultResponse: ZeroOrNullResponse.INSTANCE, verified: false)

Here, we create a mock whose default return values match those of a Mock(), but whose invocations aren’t verified (as for a Stub()). Instead of passing ZeroOrNullResponse, we could have supplied our own custom org.spockframework.mock.IDefaultResponse for responding to unexpected method invocations.

Detecting Mock Objects

To find out whether a particular object is a Spock mock object, use a org.spockframework.mock.MockUtil:

MockUtil mockUtil = new MockUtil()
List list1 = []
List list2 = Mock()

expect:
!mockUtil.isMock(list1)
mockUtil.isMock(list2)

An util can also be used to get more information about a mock object:

IMockObject mock = mockUtil.asMock(list2)

expect:
mock.name == "list2"
mock.type == List
mock.nature == MockNature.MOCK

Further Reading

If you would like to dive deeper into interaction-based testing, we recommend the following resources:

Endo-Testing: Unit Testing with Mock Objects

Paper from the XP2000 conference that introduces the concept of mock objects.

Mock Roles, not Objects

Paper from the OOPSLA2004 conference that explains how to do mocking right.

Mocks Aren’t Stubs

Martin Fowler’s take on mocking.

Growing Object-Oriented Software Guided by Tests

TDD pioneers Steve Freeman and Nat Pryce explain in detail how test-driven development and mocking work in the real world.


1. For additional ways to create mock objects, see Other Kinds of Mock Objects and A la Carte Mocks.
2. The subscriber variable cannot be referenced from the closure because it is being declared as part of the same statement.
3. The destructuring semantics for closure arguments come straight from Groovy.
4. You may know this behavior from Groovy’s MockFor and StubFor facilities.
5. Because mock configurations are immutable, the interface contains just the properties' getters.