Doubles is used by creating stubs and mocks with the allow and expect functions. Each of these functions takes a “target” object as an argument. The target is the object whose methods will be allowed or expected to be called. For example, if you wanted to expect a call to something like User.find_by_id, then User would be the target. Using a real object from your system as the target creates a so-called “partial double.”
There are also three constructors, InstanceDouble, ClassDouble, and ObjectDouble, which can be used to create so-called “pure double” targets, meaning they are unique objects which don’t modify any existing object in the system.
The details of allow, expect, and the three pure double constructors follow.
Stubs and allowances¶
Stubs are doubles which have a predetermined result when called. To stub out a method on an object, use the allow function:
from doubles import allow from myapp import User def test_allows_get_name(): user = User('Carl') allow(user).get_name assert user.get_name() is None
On the first line of the test, we create a user object from a theoretical class called User in the application we’re testing. The second line declares an allowance, after which user.get_name will use the stub method rather than the real implementation when called. The default return value of a stub is None, which the third line asserts.
To instruct the stub to return a predetermined value, use the and_return method:
from doubles import allow from myapp import User def test_allows_get_name(): user = User('Carl') allow(user).get_name.and_return('Henry') assert user.get_name() == 'Henry'
By default, once a method call has been allowed, it can be made any number of times and it will always return the value specified.
The examples shown so far will allow the stubbed method to be called with any arguments that match its signature. To specify that a method call is allowed only with specific arguments, use with_args:
from doubles import allow from myapp import User def test_allows_set_name_with_args(): user = User('Carl') allow(user).set_name.with_args('Henry') user.set_name('Henry') # Returns None user.set_name('Teddy') # Raises an UnallowedMethodCallError
You do not need to specifically call with_args, calling the allowance directly is the same as calling with_args. The following example is identical to the code above:
from doubles import allow from myapp import User def test_allows_set_name_with_args(): user = User('Carl') allow(user).set_name('Henry') user.set_name('Henry') # Returns None user.set_name('Teddy') # Raises an UnallowedMethodCallError
Multiple allowances can be specified for the same method with different arguments and return values:
from doubles import allow from myapp import User def test_returns_different_values_for_different_arguments(): user = User('Carl') allow(user).speak.with_args('hello').and_return('Carl says hello') allow(user).speak.with_args('thanks').and_return('Carl says thanks') assert user.speak('hello') == 'Carl says hello' assert user.speak('thanks') == 'Carl says thanks'
To specify that a method can only be called with no arguments, use with_no_args:
from doubles import allow from myapp import User def test_allows_greet_with_no_args(): user = User('Carl') allow(user).greet.with_no_args().and_return('Hello!') user.greet() # Returns 'Hello!' user.greet('Henry') # Raises an UnallowedMethodCallError
Without the call to with_no_args, user.greet('Henry') would have returned 'Hello!'.
Mocks and expectations¶
Stubs are useful for returning predetermined values, but they do not verify that they were interacted with. To add assertions about double interaction into the mix, create a mock object by declaring an expectation. This follows a very similar syntax, but uses expect instead of allow:
from doubles import expect from myapp import User def test_allows_get_name(): user = User('Carl') expect(user).get_name
The above test will fail with a MockExpectationError exception, because we expected user.get_name to be called, but it was not. To satisfy the mock and make the test pass:
from doubles import expect from myapp import User def test_allows_get_name(): user = User('Carl') expect(user).get_name user.get_name()
Mocks support the same interface for specifying arguments that stubs do. Mocks do not, however, support specification of return values or exceptions. If you want a test double to return a value or raise an exception, use a stub. Mocks are intended for verifying calls to methods that do not return a meaningful value. If the method does return a value, write an assertion about that value instead of using a mock.
Doubling top-level functions¶
The previous sections have shown examples where methods on classes are stubbed or mocked. It’s also possible to double a top-level function by importing the module where the function is defined into your test file. Pass the module to allow or expect and proceed as normal. In the follow example, imagine that we want to stub a function called generate_user_token in the myapp.util module:
from doubles import allow from myapp import util, User def test_get_token_returns_a_newly_generated_token_for_the_user(): user = User('Carl') allow(util).generate_user_token.with_args(user).and_return('dummy user token') assert user.get_token() == 'dummy user token'
Fakes are doubles that have special logic to determine their return values, rather than returning a simple static value. A double can be given a fake implementation with the and_return_result_of method, which accepts any callable object:
from doubles import allow from myapp import User def test_fake(): user = User('Carl') allow(user).greet.and_return_result_of(lambda: 'Hello!') assert user.greet() == 'Hello!'
Although this example is functionally equivalent to calling and_return('Hello!'), the callable passed to and_return_result_of can be arbitrarily complex. Fake functionality is available for both stubs and mocks.
Both stubs and mocks allow a method call to raise an exception instead of returning a result using the and_raise method. Simply pass the object you want to raise as an argument. The following test will pass:
from doubles import allow from myapp import User def test_raising_an_exception(): user = User('Carl') allow(user).get_name.and_raise(StandardError) try: user.get_name() except StandardError: pass else: raise AssertionError('Expected test to raise StandardError.')
If the exception to be raised requires arguments, they can be passed to the Exception constructor directly before and_raises is invoked:
from doubles import allow from myapp import User def test_raising_an_exception(): user = User('Carl') allow(user).get_name.and_raise(NonStandardError('an argument', arg2='another arg')) try: user.get_name() except NonStandardError: pass else: raise AssertionError('Expected test to raise NonStandardError.')
Limits can be set on how many times a doubled method can be called. In most cases, you’ll specify an exact call count with the syntax exactly(n).times, which will cause the test to fail if the doubled method is called fewer or more times than you declared:
from doubles import expect from myapp import User def test_expect_one_call(): user = User('Carl') expect(user).get_name.exactly(1).time user.get_name() user.get_name() # Raises a MockExpectationError because it should only be called once
The convenience methods once, twice and never are provided for the most common use cases. The following test will pass:
from doubles import expect from myapp import User def test_call_counts(): user = User('Carl') expect(user).get_name.once() expect(user).speak.twice() expect(user).not_called.never() user.get_name() user.speak('hello') user.speak('good bye')
To specify lower or upper bounds on call count instead of an exact number, use at_least and at_most:
from doubles import expect from myapp import User def test_bounded_call_counts(): user = User('Carl') expect(user).get_name.at_least(1).time expect(user).speak.at_most(2).times user.get_name # The test would fail if this wasn't called at least once user.speak('hello') user.speak('good bye') user.speak('oops') # Raises a MockExpectationError because we expected at most two calls
Call counts can be specified for allowances in addition to expectations, with the caveat that only upper bounds are enforced for allowances, making at_least a no-op.
In all of the examples so far, we added stubs and mocks to an instance of our production User class. These are called a partial doubles, because only the parts of the object that were explicitly declared as stubs or mocks are affected. The untouched methods on the object behave as usual. Let’s take a look at an example that illustrates this.:
from doubles import allow class User(object): @classmethod def find_by_email(cls, email): pass @classmethod def find_by_id(cls, user_id): pass def test_partial_double(): dummy_user = object() allow(User).find_by_email.and_return(dummy_user) User.find_by_email('firstname.lastname@example.org') # Returns <object object at 0x100290090> User.find_by_id(1) # Returns <User object at 0x1006a8cd0>
For the sake of the example, assume that the two class methods on User are implemented to return an instance of the class. We create a sentinel value to use as a dummy user, and stub User to return that specific object when User.find_by_email is called. When we then call the two class methods, we see that the method we stubbed returns the sentinel value as we declared, and User.find_by_id retains its real implementation, returning a User object.
After a test has run, all partial doubles will be restored to their pristine, undoubled state.
One of the trade offs of using test doubles is that production code may change after tests are written, and the doubles may no longer match the interface of the real object they are doubling. This is known as “API drift” and is one possible cause of the situation where a test suite is passing but the production code is broken. The potential for API drift is often used as an argument against using test doubles. Doubles provides a feature called verifying doubles to help address API drift and to increase confidence in test suites.
All test doubles created by Doubles are verifying doubles. They will cause the test to fail by raising a VerifyingDoubleError if an allowance or expectation is declared for a method that does not exist on the real object. In addition, the test will fail if the method exists but is specified with arguments that don’t match the real method’s signature.
In all the previous examples, we added stubs and mocks for real methods on the User object. Let’s see what happens if we try to stub a method that doesn’t exist:
from doubles import allow from myapp import User def test_verification(): user = User('Carl') allow(user).foo # Raises a VerifyingDoubleError, because User objects have no foo method
Similarly, we cannot declare an allowance or expectation with arguments that don’t match the actual signature of the doubled method:
from doubles import allow from myapp import User def test_verification_of_arguments(): user = User('Carl') # Raises a VerifyingDoubleArgumentError, because set_name accepts only one argument allow(user).set_name.with_args('Henry', 'Teddy')
Disabling builtin verification¶
Some of the objects in Python’s standard library are written in C and do not support the same introspection capabilities that user-created objects do. Because of this, the automatic verification features of Doubles may not work when you try to double a standard library function. There are two approaches to work around this:
Recommended: Create a simple object that wraps the standard library you want to use. Use your wrapper object from your production code and double the wrapper in your tests. Test the wrapper itself in integration with the real standard library calls, without using test doubles, to ensure that your wrapper works as expected. Although this may seem heavy handed, it’s actually a good approach, since it’s a common adage of test doubles never to double objects you don’t own.
Alternatively, use the no_builtin_verification context manager to disable the automatic verification. This is not a recommended approach, but is available if you must use it:
from doubles import allow, InstanceDouble, no_builtin_verification with no_builtin_verification(): date = InstanceDouble('datetime.date') allow(date).ctime assert date.ctime() is None
Often it’s useful to have a test double that represents a real object, but does not actually touch the real object. These doubles are called pure doubles, and like partial doubles, stubs and mocks are verified against the real object. In contrast to partial doubles, pure doubles do not implement any methods themselves, so allowances and expectations must be explicitly declared for any method that will be called on them. Calling a method that has not been allowed or expected on a pure double will raise an exception, even if the object the pure double represents has such a method.
There are three different constructors for creating pure doubles, depending on what type of object you’re doubling and how it should be verified:
InstanceDouble creates a pure test double that will ensure its usage matches the API of an instance of the provided class. It’s used as follows:
from doubles import InstanceDouble, allow def test_verifying_instance_double(): user = InstanceDouble('myapp.User') allow(user).foo
The argument to InstanceDouble is the fully qualified module path to the class in question. The double that’s created will verify itself against an instance of that class. The example above will fail with a VerifyingDoubleError exception, assuming foo is not a real instance method.
ClassDouble is the same as InstanceDouble, except that it verifies against the class itself instead of an instance of the class. The following test will fail, assuming find_by_foo is not a real class method:
from doubles import ClassDouble, allow def test_verifying_class_double(): User = ClassDouble('myapp.User') allow(User).find_by_foo
ObjectDouble creates a pure test double that is verified against a specific object. The following test will fail, assuming foo is not a real method on some_object:
from doubles import ObjectDouble, allow from myapp import some_object def test_verifying_object_double(): something = ObjectDouble(some_object) allow(something).foo
There is a subtle distinction between a pure test double created with ObjectDouble and a partial double created by passing a non-double object to allow or expect. The former creates an object that does not accept any method calls which are not explicitly allowed, but verifies any that are against the real object. A partial double modifies parts of the real object itself, allowing some methods to be doubled and others to retain their real implementation.
If you ever want to to clear all allowances and expectations you have set without verifying them, use teardown:
from doubles import teardown, expect def test_clearing_allowances(): expect(some_object).foobar teardown()
If you ever want to to clear all allowances and expectations you have set on an individual object without verifying them, use clear:
from doubles import clear, expect def test_clearing_allowances(): expect(some_object).foobar clear(some_object)
patch is used to replace an existing object:
from doubles import patch import doubles.testing def test_patch(): patch('doubles.testing.User', 'Bob Barker') assert doubles.testing.User == 'Bob Barker'
Patches do not verify against the underlying object, so use them carefully. Patches are automatically restored at the end of the test.
patch_class is a wrapper on top of patch to help you patch a python class with a ClassDouble. patch_class creates a ClassDouble of the class specified, patches the original class and returns the ClassDouble:
from doubles import patch_class, ClassDouble import doubles.testing def test_patch_class(): class_double = patch_class('doubles.testing.User') assert doubles.testing.User is class_double assert isinstance(class_double, ClassDouble)
By default ClassDoubles cannot create new instances:
from doubles import ClassDouble def test_unstubbed_constructor(): User = ClassDouble('doubles.testing.User') User('Teddy', 1901) # Raises an UnallowedMethodCallError
Stubbing the constructor of a ClassDouble is very similar to using allow or expect except we use: allow_constructor or expect_constructor, and don’t specify a method:
from doubles import allow_constructor, ClassDouble import doubles.testing def test_allow_constructor_with_args(): User = ClassDouble('doubles.testing.User') allow_constructor(User).with_args('Bob', 100).and_return('Bob') assert User('Bob', 100) == 'Bob'
The return value of allow_constructor and expect_constructor support all of the same methods as allow/expect. (e.g. with_args, once, exactly, .etc).
NOTE: Currently you can only stub the constructor of ClassDoubles
Stubbing Asynchronous Methods¶
Stubbing asynchronous methods requires returning futures and_return_future and and_raise_future do it for you.
Stubbing a method with and_return_future is similar to using and_return, except the value is wrapped in a Future:
from doubles import allow, InstanceDouble def test_and_return_future(): user = InstanceDouble('doubles.testing.User') allow(user).instance_method.and_return_future('Bob Barker') result = user.instance_method() assert result.result() == 'Bob Barker'
Stubbing a method with and_raise_future is similar to using and_raise, except the exceptions is wrapped in a Future:
from doubles import allow, InstanceDouble from pytest import raises def test_and_raise_future(): user = InstanceDouble('doubles.testing.User') exception = Exception('Bob Barker') allow(user).instance_method.and_raise_future(exception) result = user.instance_method() with raises(Exception) as e: result.result() assert e.value == exception