- Generated by
1.8.11
|
proxygen
|
You can find recipes for using Google Mock here. If you haven't yet, please read the ForDummies document first to make sure you understand the basics.
Note: Google Mock lives in the testing name space. For readability, it is recommended to write using ::testing::Foo; once in your file before using the name Foo defined by Google Mock. We omit such using statements in this page for brevity, but you should do it in your own code.
You must always put a mock method definition (MOCK_METHOD*) in a public: section of the mock class, regardless of the method being mocked being public, protected, or private in the base class. This allows ON_CALL and EXPECT_CALL to reference the mock function from outside of the mock class. (Yes, C++ allows a subclass to change the access level of a virtual function in the base class.) Example:
You can mock overloaded functions as usual. No special attention is required:
Note: if you don't mock all versions of the overloaded method, the compiler will give you a warning about some methods in the base class being hidden. To fix that, use using to bring them in scope:
To mock a class template, append _T to the MOCK_* macros:
Google Mock can mock non-virtual functions to be used in what we call hi-perf dependency injection.
In this case, instead of sharing a common base class with the real class, your mock class will be unrelated to the real class, but contain methods with the same signatures. The syntax for mocking non-virtual methods is the same as mocking virtual methods:
Note that the mock class doesn't define AppendPacket(), unlike the real class. That's fine as long as the test doesn't need to call it.
Next, you need a way to say that you want to use ConcretePacketStream in production code, and use MockPacketStream in tests. Since the functions are not virtual and the two classes are unrelated, you must specify your choice at compile time (as opposed to run time).
One way to do it is to templatize your code that needs to use a packet stream. More specifically, you will give your code a template type argument for the type of the packet stream. In production, you will instantiate your template with ConcretePacketStream as the type argument. In tests, you will instantiate the same template with MockPacketStream. For example, you may write:
Then you can use CreateConnection<ConcretePacketStream>() and PacketReader<ConcretePacketStream> in production code, and use CreateConnection<MockPacketStream>() and PacketReader<MockPacketStream> in tests.
It's possible to use Google Mock to mock a free function (i.e. a C-style function or a static method). You just need to rewrite your code to use an interface (abstract class).
Instead of calling a free function (say, OpenFile) directly, introduce an interface for it and have a concrete subclass that calls the free function:
Your code should talk to FileInterface to open a file. Now it's easy to mock out the function.
This may seem much hassle, but in practice you often have multiple related functions that you can put in the same interface, so the per-function syntactic overhead will be much lower.
If you are concerned about the performance overhead incurred by virtual functions, and profiling confirms your concern, you can combine this with the recipe for mocking non-virtual methods.
If a mock method has no EXPECT_CALL spec but is called, Google Mock will print a warning about the "uninteresting call". The rationale is:
EXPECT_CALL() to suppress the warning.However, sometimes you may want to suppress all "uninteresting call" warnings, while sometimes you may want the opposite, i.e. to treat all of them as errors. Google Mock lets you make the decision on a per-mock-object basis.
Suppose your test uses a mock class MockFoo:
If a method of mock_foo other than DoThis() is called, it will be reported by Google Mock as a warning. However, if you rewrite your test to use NiceMock<MockFoo> instead, the warning will be gone, resulting in a cleaner test output:
NiceMock<MockFoo> is a subclass of MockFoo, so it can be used wherever MockFoo is accepted.
It also works if MockFoo's constructor takes some arguments, as NiceMock<MockFoo> "inherits" MockFoo's constructors:
The usage of StrictMock is similar, except that it makes all uninteresting calls failures:
There are some caveats though (I don't like them just as much as the next guy, but sadly they are side effects of C++'s limitations):
NiceMock<MockFoo> and StrictMock<MockFoo> only work for mock methods defined using the MOCK_METHOD* family of macros directly in the MockFoo class. If a mock method is defined in a base class of MockFoo, the "nice" or "strict" modifier may not affect it, depending on the compiler. In particular, nesting NiceMock and StrictMock (e.g. NiceMock<StrictMock<MockFoo> >) is not supported.MockFoo) cannot have arguments passed by non-const reference, which happens to be banned by the Google C++ style guide.MockFoo, the mock object is not nice or strict. This may cause surprises if the constructor or destructor calls a mock method on this object. (This behavior, however, is consistent with C++'s general rule: if a constructor or destructor calls a virtual method of this object, that method is treated as non-virtual. In other words, to the base class's constructor or destructor, this object behaves like an instance of the base class, not the derived class. This rule is required for safety. Otherwise a base constructor may use members of a derived class before they are initialized, or a base destructor may use members of a derived class after they have been destroyed.)Finally, you should be very cautious about when to use naggy or strict mocks, as they tend to make tests more brittle and harder to maintain. When you refactor your code without changing its externally visible behavior, ideally you should't need to update any tests. If your code interacts with a naggy mock, however, you may start to get spammed with warnings as the result of your change. Worse, if your code interacts with a strict mock, your tests may start to fail and you'll be forced to fix them. Our general recommendation is to use nice mocks (not yet the default) most of the time, use naggy mocks (the current default) when developing or debugging tests, and use strict mocks only as the last resort.
Sometimes a method has a long list of arguments that is mostly uninteresting. For example,
This method's argument list is lengthy and hard to work with (let's say that the message argument is not even 0-terminated). If we mock it as is, using the mock will be awkward. If, however, we try to simplify this interface, we'll need to fix all clients depending on it, which is often infeasible.
The trick is to re-dispatch the method in the mock class:
By defining a new mock method with a trimmed argument list, we make the mock class much more user-friendly.
Often you may find yourself using classes that don't implement interfaces. In order to test your code that uses such a class (let's call it Concrete), you may be tempted to make the methods of Concrete virtual and then mock it.
Try not to do that.
Making a non-virtual function virtual is a big decision. It creates an extension point where subclasses can tweak your class' behavior. This weakens your control on the class because now it's harder to maintain the class' invariants. You should make a function virtual only when there is a valid reason for a subclass to override it.
Mocking concrete classes directly is problematic as it creates a tight coupling between the class and the tests - any small change in the class may invalidate your tests and make test maintenance a pain.
To avoid such problems, many programmers have been practicing "coding
to interfaces": instead of talking to the Concrete class, your code would define an interface and talk to it. Then you implement that interface as an adaptor on top of Concrete. In tests, you can easily mock that interface to observe how your code is doing.
This technique incurs some overhead:
However, it can also bring significant benefits in addition to better testability:
Concrete's API may not fit your problem domain very well, as you may not be the only client it tries to serve. By designing your own interface, you have a chance to tailor it to your need - you may add higher-level functionalities, rename stuff, etc instead of just trimming the class. This allows you to write your code (user of the interface) in a more natural way, which means it will be more readable, more maintainable, and you'll be more productive.Concrete's implementation ever has to change, you don't have to rewrite everywhere it is used. Instead, you can absorb the change in your implementation of the interface, and your other code and tests will be insulated from this change.Some people worry that if everyone is practicing this technique, they will end up writing lots of redundant code. This concern is totally understandable. However, there are two reasons why it may not be the case:
Concrete in different ways, so the best interfaces for them will be different. Therefore, each of them will have its own domain-specific interface on top of Concrete, and they will not be the same code.Concrete. You can check in the interface and the adaptor somewhere near Concrete (perhaps in a contrib sub-directory) and let many projects use it.You need to weigh the pros and cons carefully for your particular problem, but I'd like to assure you that the Java community has been practicing this for a long time and it's a proven effective technique applicable in a wide variety of situations. :-)
Some times you have a non-trivial fake implementation of an interface. For example:
Now you want to mock this interface such that you can set expectations on it. However, you also want to use FakeFoo for the default behavior, as duplicating it in the mock object is, well, a lot of work.
When you define the mock class using Google Mock, you can have it delegate its default action to a fake class you already have, using this pattern:
With that, you can use MockFoo in your tests as usual. Just remember that if you don't explicitly set an action in an ON_CALL() or EXPECT_CALL(), the fake will be called upon to do it:
Some tips:
ON_CALL() or using .WillOnce() / .WillRepeatedly() in EXPECT_CALL().DelegateToFake(), you only need to delegate the methods whose fake implementation you intend to use.ON_CALL()), see the "Selecting Between Overloaded Functions" section on this page; to disambiguate a fake function (the one you place inside Invoke()), use a static_cast to specify the function's type. For instance, if class Foo has methods char DoThis(int n) and bool DoThis(double x) const, and you want to invoke the latter, you need to write Invoke(&fake_, static_cast<bool (FakeFoo::*)(double) const>(&FakeFoo::DoThis)) instead of Invoke(&fake_, &FakeFoo::DoThis) (The strange-looking thing inside the angled brackets of static_cast is the type of a function pointer to the second DoThis() method.).Regarding the tip on mixing a mock and a fake, here's an example on why it may be a bad sign: Suppose you have a class System for low-level system operations. In particular, it does file and I/O operations. And suppose you want to test how your code uses System to do I/O, and you just want the file operations to work normally. If you mock out the entire System class, you'll have to provide a fake implementation for the file operation part, which suggests that System is taking on too many roles.
Instead, you can define a FileOps interface and an IOOps interface and split System's functionalities into the two. Then you can mock IOOps without mocking FileOps.
When using testing doubles (mocks, fakes, stubs, and etc), sometimes their behaviors will differ from those of the real objects. This difference could be either intentional (as in simulating an error such that you can test the error handling code) or unintentional. If your mocks have different behaviors than the real objects by mistake, you could end up with code that passes the tests but fails in production.
You can use the delegating-to-real technique to ensure that your mock has the same behavior as the real object while retaining the ability to validate calls. This technique is very similar to the delegating-to-fake technique, the difference being that we use a real object instead of a fake. Here's an example:
With this, Google Mock will verify that your code made the right calls (with the right arguments, in the right order, called the right number of times, etc), and a real object will answer the calls (so the behavior will be the same as in production). This gives you the best of both worlds.
Ideally, you should code to interfaces, whose methods are all pure virtual. In reality, sometimes you do need to mock a virtual method that is not pure (i.e, it already has an implementation). For example:
Sometimes you may want to call Foo::Concrete() instead of MockFoo::Concrete(). Perhaps you want to do it as part of a stub action, or perhaps your test doesn't need to mock Concrete() at all (but it would be oh-so painful to have to define a new mock class whenever you don't need to mock one of its methods).
The trick is to leave a back door in your mock class for accessing the real methods in the base class:
Now, you can call Foo::Concrete() inside an action by:
or tell the mock object that you don't want to mock Concrete():
(Why don't we just write Invoke(&foo, &Foo::Concrete)? If you do that, MockFoo::Concrete() will be called (and cause an infinite recursion) since Foo::Concrete() is virtual. That's just how C++ works.)
You can specify exactly which arguments a mock method is expecting:
You can use matchers to match arguments that have a certain property:
A frequently used matcher is _, which matches anything:
You can build complex matchers from existing ones using AllOf(), AnyOf(), and Not():
Google Mock matchers are statically typed, meaning that the compiler can catch your mistake if you use a matcher of the wrong type (for example, if you use Eq(5) to match a string argument). Good for you!
Sometimes, however, you know what you're doing and want the compiler to give you some slack. One example is that you have a matcher for long and the argument you want to match is int. While the two types aren't exactly the same, there is nothing really wrong with using a Matcher<long> to match an int - after all, we can first convert the int argument to a long before giving it to the matcher.
To support this need, Google Mock gives you the SafeMatcherCast<T>(m) function. It casts a matcher m to type Matcher<T>. To ensure safety, Google Mock checks that (let U be the type m accepts):
T can be implicitly cast to type U;T and U are built-in arithmetic types (bool, integers, and floating-point numbers), the conversion from T to U is not lossy (in other words, any value representable by T can also be represented by U); andU is a reference, T must also be a reference (as the underlying matcher may be interested in the address of the U value).The code won't compile if any of these conditions isn't met.
Here's one example:
If you find SafeMatcherCast<T>(m) too limiting, you can use a similar function MatcherCast<T>(m). The difference is that MatcherCast works as long as you can static_cast type T to type U.
MatcherCast essentially lets you bypass C++'s type system (static_cast isn't always safe as it could throw away information, for example), so be careful not to misuse/abuse it.
If you expect an overloaded function to be called, the compiler may need some help on which overloaded version it is.
To disambiguate functions overloaded on the const-ness of this object, use the Const() argument wrapper.
(Const() is defined by Google Mock and returns a const reference to its argument.)
To disambiguate overloaded functions with the same number of arguments but different argument types, you may need to specify the exact type of a matcher, either by wrapping your matcher in Matcher<type>(), or using a matcher whose type is fixed (TypedEq<type>, An<type>(), etc):
When a mock method is called, the last matching expectation that's still active will be selected (think "newer overrides older"). So, you can make a method do different things depending on its argument values like this:
Now, if foo.DoThis() is called with a value less than 5, 'a' will be returned; otherwise 'b' will be returned.
Sometimes it's not enough to match the arguments individually. For example, we may want to say that the first argument must be less than the second argument. The With() clause allows us to match all arguments of a mock function as a whole. For example,
says that the first argument of InRange() must not be 0, and must be less than the second argument.
The expression inside With() must be a matcher of type Matcher< ::testing::tuple<A1, ..., An> >, where A1, ..., An are the types of the function arguments.
You can also write AllArgs(m) instead of m inside .With(). The two forms are equivalent, but .With(AllArgs(Lt())) is more readable than .With(Lt()).
You can use Args<k1, ..., kn>(m) to match the n selected arguments (as a tuple) against m. For example,
says that Blah() will be called with arguments x, y, and z where x < y < z.
As a convenience and example, Google Mock provides some matchers for 2-tuples, including the Lt() matcher above. See the CheatSheet for the complete list.
Note that if you want to pass the arguments to a predicate of your own (e.g. .With(Args<0, 1>(Truly(&MyPredicate)))), that predicate MUST be written to take a ::testing::tuple as its argument; Google Mock will pass the n selected arguments as one single tuple to the predicate.
Have you noticed that a matcher is just a fancy predicate that also knows how to describe itself? Many existing algorithms take predicates as arguments (e.g. those defined in STL's <algorithm> header), and it would be a shame if Google Mock matchers are not allowed to participate.
Luckily, you can use a matcher where a unary predicate functor is expected by wrapping it inside the Matches() function. For example,
Since you can build complex matchers from simpler ones easily using Google Mock, this gives you a way to conveniently construct composite predicates (doing the same using STL's <functional> header is just painful). For example, here's a predicate that's satisfied by any number that is >= 0, <= 100, and != 50:
Since matchers are basically predicates that also know how to describe themselves, there is a way to take advantage of them in Google Test assertions. It's called ASSERT_THAT and EXPECT_THAT:
For example, in a Google Test test you can write:
which (as you can probably guess) executes Foo(), Bar(), and Baz(), and verifies that:
Foo() returns a string that starts with "Hello".Bar() returns a string that matches regular expression "Line \\\\d+".Baz() returns a number in the range [5, 10].The nice thing about these macros is that they read like English. They generate informative messages too. For example, if the first EXPECT_THAT() above fails, the message will be something like:
Credit: The idea of (ASSERT|EXPECT)_THAT was stolen from the Hamcrest project, which adds assertThat() to JUnit.
Google Mock provides a built-in set of matchers. In case you find them lacking, you can use an arbitray unary predicate function or functor as a matcher - as long as the predicate accepts a value of the type you want. You do this by wrapping the predicate inside the Truly() function, for example:
Note that the predicate function / functor doesn't have to return bool. It works as long as the return value can be used as the condition in statement if (condition) ....
When you do an EXPECT_CALL(mock_obj, Foo(bar)), Google Mock saves away a copy of bar. When Foo() is called later, Google Mock compares the argument to Foo() with the saved copy of bar. This way, you don't need to worry about bar being modified or destroyed after the EXPECT_CALL() is executed. The same is true when you use matchers like Eq(bar), Le(bar), and so on.
But what if bar cannot be copied (i.e. has no copy constructor)? You could define your own matcher function and use it with Truly(), as the previous couple of recipes have shown. Or, you may be able to get away from it if you can guarantee that bar won't be changed after the EXPECT_CALL() is executed. Just tell Google Mock that it should save a reference to bar, instead of a copy of it. Here's how:
Remember: if you do this, don't change bar after the EXPECT_CALL(), or the result is undefined.
Often a mock function takes a reference to object as an argument. When matching the argument, you may not want to compare the entire object against a fixed object, as that may be over-specification. Instead, you may need to validate a certain member variable or the result of a certain getter method of the object. You can do this with Field() and Property(). More specifically,
is a matcher that matches a Foo object whose bar member variable satisfies matcher m.
is a matcher that matches a Foo object whose baz() method returns a value that satisfies matcher m.
For example:
|
Field(&Foo::number, Ge(3))| Matchesxwherex.number >= 3. |
|:--------------------------—|:--------------------------------—|
|
Property(&Foo::name, StartsWith("John "))| Matchesxwherex.name()starts with"John ". |
Note that in Property(&Foo::baz, ...), method baz() must take no argument and be declared as const.
BTW, Field() and Property() can also match plain pointers to objects. For instance,
matches a plain pointer p where p->number >= 3. If p is NULL, the match will always fail regardless of the inner matcher.
What if you want to validate more than one members at the same time? Remember that there is AllOf().
C++ functions often take pointers as arguments. You can use matchers like IsNull(), NotNull(), and other comparison matchers to match a pointer, but what if you want to make sure the value pointed to by the pointer, instead of the pointer itself, has a certain property? Well, you can use the Pointee(m) matcher.
Pointee(m) matches a pointer iff m matches the value the pointer points to. For example:
expects foo.Bar() to be called with a pointer that points to a value greater than or equal to 3.
One nice thing about Pointee() is that it treats a NULL pointer as a match failure, so you can write Pointee(m) instead of
without worrying that a NULL pointer will crash your test.
Also, did we tell you that Pointee() works with both raw pointers and smart pointers (linked_ptr, shared_ptr, scoped_ptr, and etc)?
What if you have a pointer to pointer? You guessed it - you can use nested Pointee() to probe deeper inside the value. For example, Pointee(Pointee(Lt(3))) matches a pointer that points to a pointer that points to a number less than 3 (what a mouthful...).
Sometimes you want to specify that an object argument has a certain property, but there is no existing matcher that does this. If you want good error messages, you should define a matcher. If you want to do it quick and dirty, you could get away with writing an ordinary function.
Let's say you have a mock function that takes an object of type Foo, which has an int bar() method and an int baz() method, and you want to constrain that the argument's bar() value plus its baz() value is a given number. Here's how you can define a matcher to do it:
Sometimes an STL container (e.g. list, vector, map, ...) is passed to a mock function and you may want to validate it. Since most STL containers support the == operator, you can write Eq(expected_container) or simply expected_container to match a container exactly.
Sometimes, though, you may want to be more flexible (for example, the first element must be an exact match, but the second element can be any positive number, and so on). Also, containers used in tests often have a small number of elements, and having to define the expected container out-of-line is a bit of a hassle.
You can use the ElementsAre() or UnorderedElementsAre() matcher in such cases:
The above matcher says that the container must have 4 elements, which must be 1, greater than 0, anything, and 5 respectively.
If you instead write:
It means that the container must have 4 elements, which under some permutation must be 1, greater than 0, anything, and 5 respectively.
ElementsAre() and UnorderedElementsAre() are overloaded to take 0 to 10 arguments. If more are needed, you can place them in a C-style array and use ElementsAreArray() or UnorderedElementsAreArray() instead:
In case the array needs to be dynamically created (and therefore the array size cannot be inferred by the compiler), you can give ElementsAreArray() an additional argument to specify the array size:
Tips:
ElementsAre*() can be used to match any container that implements the STL iterator pattern (i.e. it has a const_iterator type and supports begin()/end()), not just the ones defined in STL. It will even work with container types yet to be written - as long as they follows the above pattern.ElementsAre*() to match nested (multi-dimensional) containers.Pointee(ElementsAre*(...)).ElementsAre*(). Therefore don't use it with containers whose element order is undefined (e.g. hash_map).Under the hood, a Google Mock matcher object consists of a pointer to a ref-counted implementation object. Copying matchers is allowed and very efficient, as only the pointer is copied. When the last matcher that references the implementation object dies, the implementation object will be deleted.
Therefore, if you have some complex matcher that you want to use again and again, there is no need to build it everytime. Just assign it to a matcher variable and use that variable repeatedly! For example,
ON_CALL is likely the single most under-utilized construct in Google Mock.
There are basically two constructs for defining the behavior of a mock object: ON_CALL and EXPECT_CALL. The difference? ON_CALL defines what happens when a mock method is called, but doesn't imply any expectation on the method being called. EXPECT_CALL not only defines the behavior, but also sets an expectation that the method will be called with the given arguments, for the given number of times (and in the given order when you specify the order too).
Since EXPECT_CALL does more, isn't it better than ON_CALL? Not really. Every EXPECT_CALL adds a constraint on the behavior of the code under test. Having more constraints than necessary is baaad - even worse than not having enough constraints.
This may be counter-intuitive. How could tests that verify more be worse than tests that verify less? Isn't verification the whole point of tests?
The answer, lies in what a test should verify. A good test verifies the contract of the code. If a test over-specifies, it doesn't leave enough freedom to the implementation. As a result, changing the implementation without breaking the contract (e.g. refactoring and optimization), which should be perfectly fine to do, can break such tests. Then you have to spend time fixing them, only to see them broken again the next time the implementation is changed.
Keep in mind that one doesn't have to verify more than one property in one test. In fact, it's a good style to verify only one thing in one test. If you do that, a bug will likely break only one or two tests instead of dozens (which case would you rather debug?). If you are also in the habit of giving tests descriptive names that tell what they verify, you can often easily guess what's wrong just from the test log itself.
So use ON_CALL by default, and only use EXPECT_CALL when you actually intend to verify that the call is made. For example, you may have a bunch of ON_CALLs in your test fixture to set the common mock behavior shared by all tests in the same group, and write (scarcely) different EXPECT_CALLs in different TEST_Fs to verify different aspects of the code's behavior. Compared with the style where each TEST has many EXPECT_CALLs, this leads to tests that are more resilient to implementational changes (and thus less likely to require maintenance) and makes the intent of the tests more obvious (so they are easier to maintain when you do need to maintain them).
If you are bothered by the "Uninteresting mock function call" message printed when a mock method without an EXPECT_CALL is called, you may use a NiceMock instead to suppress all such messages for the mock object, or suppress the message for specific methods by adding EXPECT_CALL(...).Times(AnyNumber()). DO NOT suppress it by blindly adding an EXPECT_CALL(...), or you'll have a test that's a pain to maintain.
If you are not interested in how a mock method is called, just don't say anything about it. In this case, if the method is ever called, Google Mock will perform its default action to allow the test program to continue. If you are not happy with the default action taken by Google Mock, you can override it using DefaultValue<T>::Set() (described later in this document) or ON_CALL().
Please note that once you expressed interest in a particular mock method (via EXPECT_CALL()), all invocations to it must match some expectation. If this function is called but the arguments don't match any EXPECT_CALL() statement, it will be an error.
If a mock method shouldn't be called at all, explicitly say so:
If some calls to the method are allowed, but the rest are not, just list all the expected calls:
A call to foo.Bar() that doesn't match any of the EXPECT_CALL() statements will be an error.
Uninteresting calls and unexpected calls are different concepts in Google Mock. Very different.
A call x.Y(...) is uninteresting if there's not even a single EXPECT_CALL(x, Y(...)) set. In other words, the test isn't interested in the x.Y() method at all, as evident in that the test doesn't care to say anything about it.
A call x.Y(...) is unexpected if there are some EXPECT_CALL(x, Y(...))s set, but none of them matches the call. Put another way, the test is interested in the x.Y() method (therefore it explicitly sets some EXPECT_CALL to verify how it's called); however, the verification fails as the test doesn't expect this particular call to happen.
An unexpected call is always an error, as the code under test doesn't behave the way the test expects it to behave.
By default, an uninteresting call is not an error, as it violates no constraint specified by the test. (Google Mock's philosophy is that saying nothing means there is no constraint.) However, it leads to a warning, as it might indicate a problem (e.g. the test author might have forgotten to specify a constraint).
In Google Mock, NiceMock and StrictMock can be used to make a mock class "nice" or "strict". How does this affect uninteresting calls and unexpected calls?
A nice mock suppresses uninteresting call warnings. It is less chatty than the default mock, but otherwise is the same. If a test fails with a default mock, it will also fail using a nice mock instead. And vice versa. Don't expect making a mock nice to change the test's result.
A strict mock turns uninteresting call warnings into errors. So making a mock strict may change the test's result.
Let's look at an example:
The sole EXPECT_CALL here says that all calls to GetDomainOwner() must have "google.com" as the argument. If GetDomainOwner("yahoo.com") is called, it will be an unexpected call, and thus an error. Having a nice mock doesn't change the severity of an unexpected call.
So how do we tell Google Mock that GetDomainOwner() can be called with some other arguments as well? The standard technique is to add a "catch all" EXPECT_CALL:
Remember that _ is the wildcard matcher that matches anything. With this, if GetDomainOwner("google.com") is called, it will do what the second EXPECT_CALL says; if it is called with a different argument, it will do what the first EXPECT_CALL says.
Note that the order of the two EXPECT_CALLs is important, as a newer EXPECT_CALL takes precedence over an older one.
For more on uninteresting calls, nice mocks, and strict mocks, read "The Nice, the Strict, and the Naggy".
Although an EXPECT_CALL() statement defined earlier takes precedence when Google Mock tries to match a function call with an expectation, by default calls don't have to happen in the order EXPECT_CALL() statements are written. For example, if the arguments match the matchers in the third EXPECT_CALL(), but not those in the first two, then the third expectation will be used.
If you would rather have all calls occur in the order of the expectations, put the EXPECT_CALL() statements in a block where you define a variable of type InSequence:
In this example, we expect a call to foo.DoThis(5), followed by two calls to bar.DoThat() where the argument can be anything, which are in turn followed by a call to foo.DoThis(6). If a call occurred out-of-order, Google Mock will report an error.
Sometimes requiring everything to occur in a predetermined order can lead to brittle tests. For example, we may care about A occurring before both B and C, but aren't interested in the relative order of B and C. In this case, the test should reflect our real intent, instead of being overly constraining.
Google Mock allows you to impose an arbitrary DAG (directed acyclic graph) on the calls. One way to express the DAG is to use the After clause of EXPECT_CALL.
Another way is via the InSequence() clause (not the same as the InSequence class), which we borrowed from jMock 2. It's less flexible than After(), but more convenient when you have long chains of sequential calls, as it doesn't require you to come up with different names for the expectations in the chains. Here's how it works:
If we view EXPECT_CALL() statements as nodes in a graph, and add an edge from node A to node B wherever A must occur before B, we can get a DAG. We use the term "sequence" to mean a directed path in this DAG. Now, if we decompose the DAG into sequences, we just need to know which sequences each EXPECT_CALL() belongs to in order to be able to reconstruct the orginal DAG.
So, to specify the partial order on the expectations we need to do two things: first to define some Sequence objects, and then for each EXPECT_CALL() say which Sequence objects it is part of. Expectations in the same sequence must occur in the order they are written. For example,
specifies the following DAG (where s1 is A -> B, and s2 is A -> C -> D):
This means that A must occur before B and C, and C must occur before D. There's no restriction about the order other than these.
When a mock method is called, Google Mock only consider expectations that are still active. An expectation is active when created, and becomes inactive (aka retires) when a call that has to occur later has occurred. For example, in
as soon as either #2 or #3 is matched, #1 will retire. If a warning "File too large." is logged after this, it will be an error.
Note that an expectation doesn't retire automatically when it's saturated. For example,
says that there will be exactly one warning with the message "File
too large.". If the second warning contains this message too, #2 will match again and result in an upper-bound-violated error.
If this is not what you want, you can ask an expectation to retire as soon as it becomes saturated:
Here #2 can be used only once, so if you have two warnings with the message "File too large.", the first will match #2 and the second will match #1 - there will be no error.
If a mock function's return type is a reference, you need to use ReturnRef() instead of Return() to return a result:
The Return(x) action saves a copy of x when the action is created, and always returns the same value whenever it's executed. Sometimes you may want to instead return the live value of x (i.e. its value at the time when the action is executed.).
If the mock function's return type is a reference, you can do it using ReturnRef(x), as shown in the previous recipe ("Returning References
from Mock Methods"). However, Google Mock doesn't let you use ReturnRef() in a mock function whose return type is not a reference, as doing that usually indicates a user error. So, what shall you do?
You may be tempted to try ByRef():
Unfortunately, it doesn't work here. The above code will fail with error:
The reason is that Return(value) converts value to the actual return type of the mock function at the time when the action is created, not when it is executed. (This behavior was chosen for the action to be safe when value is a proxy object that references some temporary objects.) As a result, ByRef(x) is converted to an int value (instead of a const int&) when the expectation is set, and Return(ByRef(x)) will always return 0.
ReturnPointee(pointer) was provided to solve this problem specifically. It returns the value pointed to by pointer at the time the action is executed:
Want to do more than one thing when a function is called? That's fine. DoAll() allow you to do sequence of actions every time. Only the return value of the last action in the sequence will be used.
Sometimes a method exhibits its effect not via returning a value but via side effects. For example, it may change some global state or modify an output argument. To mock side effects, in general you can define your own action by implementing testing::ActionInterface.
If all you need to do is to change an output argument, the built-in SetArgPointee() action is convenient:
In this example, when mutator.Mutate() is called, we will assign 5 to the int variable pointed to by argument #1 (0-based).
SetArgPointee() conveniently makes an internal copy of the value you pass to it, removing the need to keep the value in scope and alive. The implication however is that the value must have a copy constructor and assignment operator.
If the mock method also needs to return a value as well, you can chain SetArgPointee() with Return() using DoAll():
If the output argument is an array, use the SetArrayArgument<N>(first, last) action instead. It copies the elements in source range [first, last) to the array pointed to by the N-th (0-based) argument:
This also works when the argument is an output iterator:
If you expect a call to change the behavior of a mock object, you can use testing::InSequence to specify different behaviors before and after the call:
This makes my_mock.IsDirty() return true before my_mock.Flush() is called and return false afterwards.
If the behavior change is more complex, you can store the effects in a variable and make a mock method get its return value from that variable:
Here my_mock.GetPrevValue() will always return the argument of the last UpdateValue() call.
If a mock method's return type is a built-in C++ type or pointer, by default it will return 0 when invoked. Also, in C++ 11 and above, a mock method whose return type has a default constructor will return a default-constructed value by default. You only need to specify an action if this default value doesn't work for you.
Sometimes, you may want to change this default value, or you may want to specify a default value for types Google Mock doesn't know about. You can do this using the testing::DefaultValue class template:
Please note that changing the default value for a type can make you tests hard to understand. We recommend you to use this feature judiciously. For example, you may want to make sure the Set() and Clear() calls are right next to the code that uses your mock.
You've learned how to change the default value of a given type. However, this may be too coarse for your purpose: perhaps you have two mock methods with the same return type and you want them to have different behaviors. The ON_CALL() macro allows you to customize your mock's behavior at the method level:
As you may have guessed, when there are more than one ON_CALL() statements, the news order take precedence over the older ones. In other words, the last one that matches the function arguments will be used. This matching order allows you to set up the common behavior in a mock object's constructor or the test fixture's set-up phase and specialize the mock's behavior later.
If the built-in actions don't suit you, you can easily use an existing function, method, or functor as an action:
The only requirement is that the type of the function, etc must be compatible with the signature of the mock function, meaning that the latter's arguments can be implicitly converted to the corresponding arguments of the former, and the former's return type can be implicitly converted to that of the latter. So, you can invoke something whose type is not exactly the same as the mock function, as long as it's safe to do so - nice, huh?
Invoke() is very useful for doing actions that are more complex. It passes the mock function's arguments to the function or functor being invoked such that the callee has the full context of the call to work with. If the invoked function is not interested in some or all of the arguments, it can simply ignore them.
Yet, a common pattern is that a test author wants to invoke a function without the arguments of the mock function. Invoke() allows her to do that using a wrapper function that throws away the arguments before invoking an underlining nullary function. Needless to say, this can be tedious and obscures the intent of the test.
InvokeWithoutArgs() solves this problem. It's like Invoke() except that it doesn't pass the mock function's arguments to the callee. Here's an example:
Sometimes a mock function will receive a function pointer or a functor (in other words, a "callable") as an argument, e.g.
and you may want to invoke this callable argument:
Arghh, you need to refer to a mock function argument but C++ has no lambda (yet), so you have to define your own action. :-( Or do you really?
Well, Google Mock has an action to solve exactly this problem:
will invoke the N-th (0-based) argument the mock function receives, with arg_1, arg_2, ..., and arg_m. No matter if the argument is a function pointer or a functor, Google Mock handles them both.
With that, you could write:
What if the callable takes an argument by reference? No problem - just wrap it inside ByRef():
What if the callable takes an argument by reference and we do not wrap the argument in ByRef()? Then InvokeArgument() will make a copy of the argument, and pass a reference to the copy, instead of a reference to the original value, to the callable. This is especially handy when the argument is a temporary value:
Sometimes you have an action that returns something, but you need an action that returns void (perhaps you want to use it in a mock function that returns void, or perhaps it needs to be used in DoAll() and it's not the last in the list). IgnoreResult() lets you do that. For example:
Note that you cannot use IgnoreResult() on an action that already returns void. Doing so will lead to ugly compiler errors.
Say you have a mock function Foo() that takes seven arguments, and you have a custom action that you want to invoke when Foo() is called. Trouble is, the custom action only wants three arguments:
To please the compiler God, you can to define an "adaptor" that has the same signature as Foo() and calls the custom action with the right arguments:
But isn't this awkward?
Google Mock provides a generic action adaptor, so you can spend your time minding more important business than writing your own adaptors. Here's the syntax:
creates an action that passes the arguments of the mock function at the given indices (0-based) to the inner action and performs it. Using WithArgs, our original example can be written as:
For better readability, Google Mock also gives you:
WithoutArgs(action) when the inner action takes no argument, andWithArg<N>(action) (no s after Arg) when the inner action takes one argument.As you may have realized, InvokeWithoutArgs(...) is just syntactic sugar for WithoutArgs(Invoke(...)).
Here are more tips:
WithArgs and friends does not have to be Invoke() – it can be anything.WithArgs<2, 3, 3, 5>(...).WithArgs<3, 2, 1>(...).int and my_action takes a double, WithArg<4>(my_action) will work.The selecting-an-action's-arguments recipe showed us one way to make a mock function and an action with incompatible argument lists fit together. The downside is that wrapping the action in WithArgs<...>() can get tedious for people writing the tests.
If you are defining a function, method, or functor to be used with Invoke*(), and you are not interested in some of its arguments, an alternative to WithArgs is to declare the uninteresting arguments as Unused. This makes the definition less cluttered and less fragile in case the types of the uninteresting arguments change. It could also increase the chance the action function can be reused. For example, given
instead of
you could write
Just like matchers, a Google Mock action object consists of a pointer to a ref-counted implementation object. Therefore copying actions is also allowed and very efficient. When the last action that references the implementation object dies, the implementation object will be deleted.
If you have some complex action that you want to use again and again, you may not have to build it from scratch everytime. If the action doesn't have an internal state (i.e. if it always does the same thing no matter how many times it has been called), you can assign it to an action variable and use that variable repeatedly. For example:
However, if the action has its own state, you may be surprised if you share the action object. Suppose you have an action factory IncrementCounter(init) which creates an action that increments and returns a counter whose initial value is init, using two actions created from the same expression and using a shared action will exihibit different behaviors. Example:
versus
C++11 introduced move-only types. A move-only-typed value can be moved from one object to another, but cannot be copied. std::unique_ptr<T> is probably the most commonly used move-only type.
Mocking a method that takes and/or returns move-only types presents some challenges, but nothing insurmountable. This recipe shows you how you can do it.
Let’s say we are working on a fictional project that lets one post and share snippets called “buzzes”. Your code uses these types:
A Buzz object represents a snippet being posted. A class that implements the Buzzer interface is capable of creating and sharing Buzz. Methods in Buzzer may return a unique_ptr<Buzz> or take a unique_ptr<Buzz>. Now we need to mock Buzzer in our tests.
To mock a method that returns a move-only type, you just use the familiar MOCK_METHOD syntax as usual:
However, if you attempt to use the same MOCK_METHOD pattern to mock a method that takes a move-only parameter, you’ll get a compiler error currently:
While it’s highly desirable to make this syntax just work, it’s not trivial and the work hasn’t been done yet. Fortunately, there is a trick you can apply today to get something that works nearly as well as this.
The trick, is to delegate the ShareBuzz() method to a mock method (let’s call it DoShareBuzz()) that does not take move-only parameters:
Note that there's no need to define or declare DoShareBuzz() in a base class. You only need to define it as a MOCK_METHOD in the mock class.
Now that we have the mock class defined, we can use it in tests. In the following code examples, we assume that we have defined a MockBuzzer object named mock_buzzer_:
First let’s see how we can set expectations on the MakeBuzz() method, which returns a unique_ptr<Buzz>.
As usual, if you set an expectation without an action (i.e. the .WillOnce() or .WillRepeated() clause), when that expectation fires, the default action for that method will be taken. Since unique_ptr<> has a default constructor that returns a null unique_ptr, that’s what you’ll get if you don’t specify an action:
If you are not happy with the default action, you can tweak it. Depending on what you need, you may either tweak the default action for a specific (mock object, mock method) combination using ON_CALL(), or you may tweak the default action for all mock methods that return a specific type. The usage of ON_CALL() is similar to EXPECT_CALL(), so we’ll skip it and just explain how to do the latter (tweaking the default action for a specific return type). You do this via the DefaultValue<>::SetFactory() and DefaultValue<>::Clear() API:
What if you want the method to do something other than the default action? If you just need to return a pre-defined move-only value, you can use the Return(ByMove(...)) action:
Note that ByMove() is essential here - if you drop it, the code won’t compile.
Quiz time! What do you think will happen if a Return(ByMove(...)) action is performed more than once (e.g. you write ….WillRepeatedly(Return(ByMove(...)));)? Come think of it, after the first time the action runs, the source value will be consumed (since it’s a move-only value), so the next time around, there’s no value to move from – you’ll get a run-time error that Return(ByMove(...)) can only be run once.
If you need your mock method to do more than just moving a pre-defined value, remember that you can always use Invoke() to call a lambda or a callable object, which can do pretty much anything you want:
Every time this EXPECT_CALL fires, a new unique_ptr<Buzz> will be created and returned. You cannot do this with Return(ByMove(...)).
Now there’s one topic we haven’t covered: how do you set expectations on ShareBuzz(), which takes a move-only-typed parameter? The answer is you don’t. Instead, you set expectations on the DoShareBuzz() mock method (remember that we defined a MOCK_METHOD for DoShareBuzz(), not ShareBuzz()):
Some of you may have spotted one problem with this approach: the DoShareBuzz() mock method differs from the real ShareBuzz() method in that it cannot take ownership of the buzz parameter - ShareBuzz() will always delete buzz after DoShareBuzz() returns. What if you need to save the buzz object somewhere for later use when ShareBuzz() is called? Indeed, you'd be stuck.
Another problem with the DoShareBuzz() we had is that it can surprise people reading or maintaining the test, as one would expect that DoShareBuzz() has (logically) the same contract as ShareBuzz().
Fortunately, these problems can be fixed with a bit more code. Let's try to get it right this time:
Now, the mock DoShareBuzz() method is free to save the buzz argument for later use if this is what you want:
Using the tricks covered in this recipe, you are now able to mock methods that take and/or return move-only types. Put your newly-acquired power to good use - when you design a new API, you can now feel comfortable using unique_ptrs as appropriate, without fearing that doing so will compromise your tests.
Believe it or not, the vast majority of the time spent on compiling a mock class is in generating its constructor and destructor, as they perform non-trivial tasks (e.g. verification of the expectations). What's more, mock methods with different signatures have different types and thus their constructors/destructors need to be generated by the compiler separately. As a result, if you mock many different types of methods, compiling your mock class can get really slow.
If you are experiencing slow compilation, you can move the definition of your mock class' constructor and destructor out of the class body and into a .cpp file. This way, even if you #include your mock class in N files, the compiler only needs to generate its constructor and destructor once, resulting in a much faster compilation.
Let's illustrate the idea using an example. Here's the definition of a mock class before applying this recipe:
After the change, it would look like:
and
When it's being destoyed, your friendly mock object will automatically verify that all expectations on it have been satisfied, and will generate Google Test failures if not. This is convenient as it leaves you with one less thing to worry about. That is, unless you are not sure if your mock object will be destoyed.
How could it be that your mock object won't eventually be destroyed? Well, it might be created on the heap and owned by the code you are testing. Suppose there's a bug in that code and it doesn't delete the mock object properly - you could end up with a passing test when there's actually a bug.
Using a heap checker is a good idea and can alleviate the concern, but its implementation may not be 100% reliable. So, sometimes you do want to force Google Mock to verify a mock object before it is (hopefully) destructed. You can do this with Mock::VerifyAndClearExpectations(&mock_object):
Tip: The Mock::VerifyAndClearExpectations() function returns a bool to indicate whether the verification was successful (true for yes), so you can wrap that function call inside a ASSERT_TRUE() if there is no point going further when the verification has failed.
Sometimes you may want to "reset" a mock object at various check points in your test: at each check point, you verify that all existing expectations on the mock object have been satisfied, and then you set some new expectations on it as if it's newly created. This allows you to work with a mock object in "phases" whose sizes are each manageable.
One such scenario is that in your test's SetUp() function, you may want to put the object you are testing into a certain state, with the help from a mock object. Once in the desired state, you want to clear all expectations on the mock, such that in the TEST_F body you can set fresh expectations on it.
As you may have figured out, the Mock::VerifyAndClearExpectations() function we saw in the previous recipe can help you here. Or, if you are using ON_CALL() to set default actions on the mock object and want to clear the default actions as well, use Mock::VerifyAndClear(&mock_object) instead. This function does what Mock::VerifyAndClearExpectations(&mock_object) does and returns the same bool, plus it clears the ON_CALL() statements on mock_object too.
Another trick you can use to achieve the same effect is to put the expectations in sequences and insert calls to a dummy "check-point" function at specific places. Then you can verify that the mock function calls do happen at the right time. For example, if you are exercising code:
and want to verify that Foo(1) and Foo(3) both invoke mock.Bar("a"), but Foo(2) doesn't invoke anything. You can write:
The expectation spec says that the first Bar("a") must happen before check point "1", the second Bar("a") must happen after check point "2", and nothing should happen between the two check points. The explicit check points make it easy to tell which Bar("a") is called by which call to Foo().
Sometimes you want to make sure a mock object is destructed at the right time, e.g. after bar->A() is called but before bar->B() is called. We already know that you can specify constraints on the order of mock function calls, so all we need to do is to mock the destructor of the mock function.
This sounds simple, except for one problem: a destructor is a special function with special syntax and special semantics, and the MOCK_METHOD0 macro doesn't work for it:
The good news is that you can use a simple pattern to achieve the same effect. First, add a mock function Die() to your mock class and call it in the destructor, like this:
(If the name Die() clashes with an existing symbol, choose another name.) Now, we have translated the problem of testing when a MockFoo object dies to testing when its Die() method is called:
And that's that.
IMPORTANT NOTE: What we describe in this recipe is ONLY true on platforms where Google Mock is thread-safe. Currently these are only platforms that support the pthreads library (this includes Linux and Mac). To make it thread-safe on other platforms we only need to implement some synchronization operations in "gtest/internal/gtest-port.h".
In a unit test, it's best if you could isolate and test a piece of code in a single-threaded context. That avoids race conditions and dead locks, and makes debugging your test much easier.
Yet many programs are multi-threaded, and sometimes to test something we need to pound on it from more than one thread. Google Mock works for this purpose too.
Remember the steps for using a mock:
foo.ON_CALL() and EXPECT_CALL().foo.If you follow the following simple rules, your mocks and threads can live happily together:
foo. Obvious too, huh?If you violate the rules (for example, if you set expectations on a mock while another thread is calling its methods), you get undefined behavior. That's not fun, so don't do it.
Google Mock guarantees that the action for a mock function is done in the same thread that called the mock function. For example, in
if Foo(1) is called in thread 1 and Foo(2) is called in thread 2, Google Mock will execute action1 in thread 1 and action2 in thread 2.
Google Mock does not impose a sequence on actions performed in different threads (doing so may create deadlocks as the actions may need to cooperate). This means that the execution of action1 and action2 in the above example may interleave. If this is a problem, you should add proper synchronization logic to action1 and action2 to make the test thread-safe.
Also, remember that DefaultValue<T> is a global resource that potentially affects all living mock objects in your program. Naturally, you won't want to mess with it from multiple threads or when there still are mocks in action.
When Google Mock sees something that has the potential of being an error (e.g. a mock function with no expectation is called, a.k.a. an uninteresting call, which is allowed but perhaps you forgot to explicitly ban the call), it prints some warning messages, including the arguments of the function and the return value. Hopefully this will remind you to take a look and see if there is indeed a problem.
Sometimes you are confident that your tests are correct and may not appreciate such friendly messages. Some other times, you are debugging your tests or learning about the behavior of the code you are testing, and wish you could observe every mock call that happens (including argument values and the return value). Clearly, one size doesn't fit all.
You can control how much Google Mock tells you using the --gmock_verbose=LEVEL command-line flag, where LEVEL is a string with three possible values:
info: Google Mock will print all informational messages, warnings, and errors (most verbose). At this setting, Google Mock will also log any calls to the ON_CALL/EXPECT_CALL macros.warning: Google Mock will print both warnings and errors (less verbose). This is the default.error: Google Mock will print errors only (least verbose).Alternatively, you can adjust the value of that flag from within your tests like so:
Now, judiciously use the right flag to enable Google Mock serve you better!
You have a test using Google Mock. It fails: Google Mock tells you that some expectations aren't satisfied. However, you aren't sure why: Is there a typo somewhere in the matchers? Did you mess up the order of the EXPECT_CALLs? Or is the code under test doing something wrong? How can you find out the cause?
Won't it be nice if you have X-ray vision and can actually see the trace of all EXPECT_CALLs and mock method calls as they are made? For each call, would you like to see its actual argument values and which EXPECT_CALL Google Mock thinks it matches?
You can unlock this power by running your test with the --gmock_verbose=info flag. For example, given the test program:
if you run it with --gmock_verbose=info, you will see this output:
Suppose the bug is that the "c" in the third EXPECT_CALL is a typo and should actually be "a". With the above message, you should see that the actual F("a", "good") call is matched by the first EXPECT_CALL, not the third as you thought. From that it should be obvious that the third EXPECT_CALL is written wrong. Case solved.
If you build and run your tests in Emacs, the source file locations of Google Mock and Google Test errors will be highlighted. Just press <Enter> on one of them and you'll be taken to the offending line. Or, you can just type C-x` to jump to the next error.
To make it even easier, you can add the following lines to your ~/.emacs file:
Then you can type M-m to start a build, or M-up/M-down to move back and forth between errors.
Google Mock's implementation consists of dozens of files (excluding its own tests). Sometimes you may want them to be packaged up in fewer files instead, such that you can easily copy them to a new machine and start hacking there. For this we provide an experimental Python script fuse_gmock_files.py in the scripts/ directory (starting with release 1.2.0). Assuming you have Python 2.4 or above installed on your machine, just go to that directory and run
and you should see an OUTPUT_DIR directory being created with files gtest/gtest.h, gmock/gmock.h, and gmock-gtest-all.cc in it. These three files contain everything you need to use Google Mock (and Google Test). Just copy them to anywhere you want and you are ready to write tests and use mocks. You can use the scrpts/test/Makefile file as an example on how to compile your tests against them.
The MATCHER* family of macros can be used to define custom matchers easily. The syntax:
will define a matcher with the given name that executes the statements, which must return a bool to indicate if the match succeeds. Inside the statements, you can refer to the value being matched by arg, and refer to its type by arg_type.
The description string is a string-typed expression that documents what the matcher does, and is used to generate the failure message when the match fails. It can (and should) reference the special bool variable negation, and should evaluate to the description of the matcher when negation is false, or that of the matcher's negation when negation is true.
For convenience, we allow the description string to be empty (""), in which case Google Mock will use the sequence of words in the matcher name as the description.
For example:
allows you to write
or,
If the above assertions fail, they will print something like:
where the descriptions "is divisible by 7" and "not (is divisible
by 7)" are automatically calculated from the matcher name IsDivisibleBy7.
As you may have noticed, the auto-generated descriptions (especially those for the negation) may not be so great. You can always override them with a string expression of your own:
Optionally, you can stream additional information to a hidden argument named result_listener to explain the match result. For example, a better definition of IsDivisibleBy7 is:
With this definition, the above assertion will give a better message:
You should let MatchAndExplain() print any additional information that can help a user understand the match result. Note that it should explain why the match succeeds in case of a success (unless it's obvious) - this is useful when the matcher is used inside Not(). There is no need to print the argument value itself, as Google Mock already prints it for you.
Notes:
arg_type) is determined by the context in which you use the matcher and is supplied to you by the compiler, so you don't need to worry about declaring it (nor can you). This allows the matcher to be polymorphic. For example, IsDivisibleBy7() can be used to match any type where the value of (arg % 7) == 0 can be implicitly converted to a bool. In the Bar(IsDivisibleBy7()) example above, if method Bar() takes an int, arg_type will be int; if it takes an unsigned long, arg_type will be unsigned long; and so on.Sometimes you'll want to define a matcher that has parameters. For that you can use the macro:
where the description string can be either "" or a string expression that references negation and param_name.
For example:
will allow you to write:
which may lead to this message (assuming n is 10):
Note that both the matcher description and its parameter are printed, making the message human-friendly.
In the matcher definition body, you can write foo_type to reference the type of a parameter named foo. For example, in the body of MATCHER_P(HasAbsoluteValue, value) above, you can write value_type to refer to the type of value.
Google Mock also provides MATCHER_P2, MATCHER_P3, ..., up to MATCHER_P10 to support multi-parameter matchers:
Please note that the custom description string is for a particular instance of the matcher, where the parameters have been bound to actual values. Therefore usually you'll want the parameter values to be part of the description. Google Mock lets you do that by referencing the matcher parameters in the description string expression.
For example,
would generate a failure that contains the message:
If you specify "" as the description, the failure message will contain the sequence of words in the matcher name followed by the parameter values printed as a tuple. For example,
would generate a failure that contains the text:
For the purpose of typing, you can view
as shorthand for
When you write Foo(v1, ..., vk), the compiler infers the types of the parameters v1, ..., and vk for you. If you are not happy with the result of the type inference, you can specify the types by explicitly instantiating the template, as in Foo<long, bool>(5, false). As said earlier, you don't get to (or need to) specify arg_type as that's determined by the context in which the matcher is used.
You can assign the result of expression Foo(p1, ..., pk) to a variable of type FooMatcherPk<p1_type, ..., pk_type>. This can be useful when composing matchers. Matchers that don't have a parameter or have only one parameter have special types: you can assign Foo() to a FooMatcher-typed variable, and assign Foo(p) to a FooMatcherP<p_type>-typed variable.
While you can instantiate a matcher template with reference types, passing the parameters by pointer usually makes your code more readable. If, however, you still want to pass a parameter by reference, be aware that in the failure message generated by the matcher you will see the value of the referenced object but not its address.
You can overload matchers with different numbers of parameters:
While it's tempting to always use the MATCHER* macros when defining a new matcher, you should also consider implementing MatcherInterface or using MakePolymorphicMatcher() instead (see the recipes that follow), especially if you need to use the matcher a lot. While these approaches require more work, they give you more control on the types of the value being matched and the matcher parameters, which in general leads to better compiler error messages that pay off in the long run. They also allow overloading matchers based on parameter types (as opposed to just based on the number of parameters).
A matcher of argument type T implements testing::MatcherInterface<T> and does two things: it tests whether a value of type T matches the matcher, and can describe what kind of values it matches. The latter ability is used for generating readable error messages when expectations are violated.
The interface looks like this:
If you need a custom matcher but Truly() is not a good option (for example, you may not be happy with the way Truly(predicate) describes itself, or you may want your matcher to be polymorphic as Eq(value) is), you can define a matcher to do whatever you want in two steps: first implement the matcher interface, and then define a factory function to create a matcher instance. The second step is not strictly needed but it makes the syntax of using the matcher nicer.
For example, you can define a matcher to test whether an int is divisible by 7 and then use it like this:
You may improve the matcher message by streaming additional information to the listener argument in MatchAndExplain():
Then, EXPECT_THAT(x, DivisibleBy7()); may general a message like this:
You've learned how to write your own matchers in the previous recipe. Just one problem: a matcher created using MakeMatcher() only works for one particular type of arguments. If you want a polymorphic matcher that works with arguments of several types (for instance, Eq(x) can be used to match a value as long as value == x compiles – value and x don't have to share the same type), you can learn the trick from "gmock/gmock-matchers.h" but it's a bit involved.
Fortunately, most of the time you can define a polymorphic matcher easily with the help of MakePolymorphicMatcher(). Here's how you can define NotNull() as an example:
Note: Your polymorphic matcher class does not need to inherit from MatcherInterface or any other class, and its methods do not need to be virtual.
Like in a monomorphic matcher, you may explain the match result by streaming additional information to the listener argument in MatchAndExplain().
A cardinality is used in Times() to tell Google Mock how many times you expect a call to occur. It doesn't have to be exact. For example, you can say AtLeast(5) or Between(2, 4).
If the built-in set of cardinalities doesn't suit you, you are free to define your own by implementing the following interface (in namespace testing):
For example, to specify that a call must occur even number of times, you can write
If the built-in actions don't work for you, and you find it inconvenient to use Invoke(), you can use a macro from the ACTION* family to quickly define a new action that can be used in your code as if it's a built-in action.
By writing
in a namespace scope (i.e. not inside a class or function), you will define an action with the given name that executes the statements. The value returned by statements will be used as the return value of the action. Inside the statements, you can refer to the K-th (0-based) argument of the mock function as argK. For example:
allows you to write
Note that you don't need to specify the types of the mock function arguments. Rest assured that your code is type-safe though: you'll get a compiler error if *arg1 doesn't support the ++ operator, or if the type of ++(*arg1) isn't compatible with the mock function's return type.
Another example:
defines an action Foo() that invokes argument #2 (a function pointer) with 5, calls function Blah(), sets the value pointed to by argument #1 to 0, and returns argument #0.
For more convenience and flexibility, you can also use the following pre-defined symbols in the body of ACTION:
argK_type | The type of the K-th (0-based) argument of the mock function |
|---|---|
args | All arguments of the mock function as a tuple |
args_type | The type of all arguments of the mock function as a tuple |
return_type | The return type of the mock function |
function_type | The type of the mock function |
For example, when using an ACTION as a stub action for mock function:
we have:
| Pre-defined Symbol | Is Bound To |
|---|---|
arg0 | the value of flag |
arg0_type | the type bool |
arg1 | the value of ptr |
arg1_type | the type int* |
args | the tuple (flag, ptr) |
args_type | the type ::testing::tuple<bool, int*> |
return_type | the type int |
function_type | the type int(bool, int*) |
Sometimes you'll want to parameterize an action you define. For that we have another macro
For example,
will allow you to write
For convenience, we use the term arguments for the values used to invoke the mock function, and the term parameters for the values used to instantiate an action.
Note that you don't need to provide the type of the parameter either. Suppose the parameter is named param, you can also use the Google-Mock-defined symbol param_type to refer to the type of the parameter as inferred by the compiler. For example, in the body of ACTION_P(Add, n) above, you can write n_type for the type of n.
Google Mock also provides ACTION_P2, ACTION_P3, and etc to support multi-parameter actions. For example,
lets you write
You can view ACTION as a degenerated parameterized action where the number of parameters is 0.
You can also easily define actions overloaded on the number of parameters:
For maximum brevity and reusability, the ACTION* macros don't ask you to provide the types of the mock function arguments and the action parameters. Instead, we let the compiler infer the types for us.
Sometimes, however, we may want to be more explicit about the types. There are several tricks to do that. For example:
where StaticAssertTypeEq is a compile-time assertion in Google Test that verifies two types are the same.
Sometimes you want to give an action explicit template parameters that cannot be inferred from its value parameters. ACTION_TEMPLATE() supports that and can be viewed as an extension to ACTION() and ACTION_P*().
The syntax:
defines an action template that takes m explicit template parameters and n value parameters, where m is between 1 and 10, and n is between 0 and 10. name_i is the name of the i-th template parameter, and kind_i specifies whether it's a typename, an integral constant, or a template. p_i is the name of the i-th value parameter.
Example:
To create an instance of an action template, write:
where the ts are the template arguments and the vs are the value arguments. The value argument types are inferred by the compiler. For example:
If you want to explicitly specify the value argument types, you can provide additional template arguments:
where u_i is the desired type of v_i.
ACTION_TEMPLATE and ACTION/ACTION_P* can be overloaded on the number of value parameters, but not on the number of template parameters. Without the restriction, the meaning of the following is unclear:
Are we using a single-template-parameter action where bool refers to the type of x, or a two-template-parameter action where the compiler is asked to infer the type of x?
If you are writing a function that returns an ACTION object, you'll need to know its type. The type depends on the macro used to define the action and the parameter types. The rule is relatively simple:
| Given Definition | Expression | Has Type |
|---|---|---|
ACTION(Foo) | Foo() | FooAction |
ACTION_TEMPLATE(Foo, HAS_m_TEMPLATE_PARAMS(...), AND_0_VALUE_PARAMS()) | Foo<t1, ..., t_m>() | FooAction<t1, ..., t_m> |
ACTION_P(Bar, param) | Bar(int_value) | BarActionP<int> |
ACTION_TEMPLATE(Bar, HAS_m_TEMPLATE_PARAMS(...), AND_1_VALUE_PARAMS(p1)) | Bar<t1, ..., t_m>(int_value) | FooActionP<t1, ..., t_m, int> |
ACTION_P2(Baz, p1, p2) | Baz(bool_value, int_value) | BazActionP2<bool, int> |
ACTION_TEMPLATE(Baz, HAS_m_TEMPLATE_PARAMS(...), AND_2_VALUE_PARAMS(p1, p2)) | Baz<t1, ..., t_m>(bool_value, int_value) | FooActionP2<t1, ..., t_m, bool, int> |
| ... | ... | ... |
Note that we have to pick different suffixes (Action, ActionP, ActionP2, and etc) for actions with different numbers of value parameters, or the action definitions cannot be overloaded on the number of them.
While the ACTION* macros are very convenient, sometimes they are inappropriate. For example, despite the tricks shown in the previous recipes, they don't let you directly specify the types of the mock function arguments and the action parameters, which in general leads to unoptimized compiler error messages that can baffle unfamiliar users. They also don't allow overloading actions based on parameter types without jumping through some hoops.
An alternative to the ACTION* macros is to implement testing::ActionInterface<F>, where F is the type of the mock function in which the action will be used. For example:
The previous recipe showed you how to define your own action. This is all good, except that you need to know the type of the function in which the action will be used. Sometimes that can be a problem. For example, if you want to use the action in functions with different types (e.g. like Return() and SetArgPointee()).
If an action can be used in several types of mock functions, we say it's polymorphic. The MakePolymorphicAction() function template makes it easy to define such an action:
As an example, let's define an action that returns the second argument in the mock function's argument list. The first step is to define an implementation class:
This implementation class does not need to inherit from any particular class. What matters is that it must have a Perform() method template. This method template takes the mock function's arguments as a tuple in a single argument, and returns the result of the action. It can be either const or not, but must be invokable with exactly one template argument, which is the result type. In other words, you must be able to call Perform<R>(args) where R is the mock function's return type and args is its arguments in a tuple.
Next, we use MakePolymorphicAction() to turn an instance of the implementation class into the polymorphic action we need. It will be convenient to have a wrapper for this:
Now, you can use this polymorphic action the same way you use the built-in ones:
When an uninteresting or unexpected call occurs, Google Mock prints the argument values and the stack trace to help you debug. Assertion macros like EXPECT_THAT and EXPECT_EQ also print the values in question when the assertion fails. Google Mock and Google Test do this using Google Test's user-extensible value printer.
This printer knows how to print built-in C++ types, native arrays, STL containers, and any type that supports the << operator. For other types, it prints the raw bytes in the value and hopes that you the user can figure it out. Google Test's advanced guide explains how to extend the printer to do a better job at printing your particular type than to dump the bytes.
1.8.11