Forge is a mocking library for Python. It draws most of its inspiration from Mox (http://code.google.com/p/pymox). It is aimed to be simple, but still feature-rich, and provide maximum flexibility for unit testing using the mock approach.
All of Forge's acceptance tests are in the tests/ directory under the root. They require either unittest2 or the built-in unittest module (2.7/3.2 and above).
Running the tests is recommended with the nosetests script, but in case you don't have it, the run_tests script can be used instead.
Installing forge is pretty much the same as most of the packages you already know
python setup.py install
Forge mostly creates mock objects and function stubs, but in a variety of flavors. Using Forge always starts with creating a "Mock Manager", with the Forge class:
>>> from forge import Forge >>> forge_manager = Forge()
There shouldn't be a real reason for keeping more than one forge manager. What it is typically used for is creating mocks:
>>> class SomeClass(object): ... def f(self, a, b, c): ... pass >>> mock = forge_manager.create_mock(SomeClass) >>> mock <Mock of 'SomeClass'>
Mock tests usually act in a record-replay manner. You record what you expect your mock to do, and then replay it, while Forge tracks what happens and makes sure it is correct:
>>> forge_manager.is_recording() True >>> mock.f(1, 2, 3) # doctest: +ELLIPSIS <...> >>> mock.f(3, 4, 5) # doctest: +ELLIPSIS <...> >>> forge_manager.replay() >>> forge_manager.is_recording() False >>> forge_manager.is_replaying() True >>> mock.f(1, 2, 3) >>> mock.f(3, 4, 5) >>> forge_manager.verify() # this verifies no more calls are expected
To start over working from scratch, you can always perform:
>>> forge_manager.reset()
Just like classes yield mocks, regular functions yield stubs, through the use of Forge.create_function_stub:
>>> def some_func(a, b, c): ... pass >>> stub = forge_manager.create_function_stub(some_func)
As methods and functions are recorded, their signature is verified against the recorded calls. Upon replay the call must match the original call, so you shouldn't worry too much about accidents concerning the function signature.
To promote niceness, two context managers provide syntactic sugar that structure the test code:
>>> with forge_manager.record_context(): ... mock.f(1, 2, 3) # doctest: +ELLIPSIS ... mock.f(3, 4, 5) # doctest: +ELLIPSIS <...> >>> with forge_manager.verified_replay_context(): ... mock.f(1, 2, 3) # doctest: +ELLIPSIS ... mock.f(3, 4, 5) # doctest: +ELLIPSIS
Whenever an event occurs that was not expected, an exception is raised, explaining what happened:
>>> stub = forge_manager.create_function_stub(some_func) >>> stub(1, 2, 3) # doctest: +ELLIPSIS <...> >>> forge_manager.replay() >>> stub(1, 2, 4) # doctest: +IGNORE_EXCEPTION_DETAIL Traceback (most recent call last): ... UnexpectedCall: Unexpected function called! (Expected: +, Got: -) - some_func(a=1, b=2, c=4) ? ^ + some_func(a=1, b=2, c=3) ? ^ >>> forge_manager.reset()
In some cases this is sufficient, but in case you want a bit more info as to where the calls were recorded and replayed, you can turn on debug info:
>>> forge_manager.debug.enable() >>> stub = forge_manager.create_function_stub(some_func) >>> stub(1, 2, 3) # doctest: +ELLIPSIS <...> >>> forge_manager.replay() >>> stub(1, 2, 4) # doctest: +IGNORE_EXCEPTION_DETAIL Traceback (most recent call last): ... UnexpectedCall: Unexpected function called! (Expected: +, Got: -) Recorded from ... Replayed from ... - some_func(a=1, b=2, c=4) ? ^ + some_func(a=1, b=2, c=3) ? ^ >>> forge_manager.reset() >>> forge_manager.debug.disable()
Since sometimes this is a very common pattern, you can also turn on debugging through environment variables, by setting the FORGE_DEBUG environment variable to anything when running your tests.
Setting attributes for mock object is allowed only during record mode. By default, attributes set during replay will trigger an exception.
However, in some cases you want to expect an attribute being set at some point of the replay. Due to the hackish nature of the Forge setattr/getattr mechanism, the way to do this is with a dedicated API through the __forge__ handle:
>>> mock = forge_manager.create_mock(SomeClass)
>>> mock.__forge__.expect_setattr("length", 20) # doctest: +ELLIPSIS
<...>
>>> forge_manager.replay()
>>> mock.length = 20
>>> forge_manager.verify()
>>> forge_manager.reset()
You can also set mock object to ignore attribute setting (thus allow all setattrs regardless of nature). This is not recommended, but might sometimes be useful:
>>> mock.__forge__.enable_setattr_during_replay() >>> forge_manager.replay() >>> mock.a = 2 # works! >>> forge_manager.reset()
If you want to simulate a mock structure, that is, an object with attributes which are in turn other objects, you can use the create_mock_with_attrs API. This is especially concise if you create a shortcut for it:
>>> class A(object): pass
>>> class B(object): pass
>>> class C(object): pass
>>> MOCK = forge_manager.create_mock_with_attrs
>>> result = MOCK(A, b=MOCK(B, c=MOCK(C)))
>>> result.b.c # doctest: +ELLIPSIS
<Mock of 'C'>
When expecting a call on a stub, you can control what happens when the call takes place. Supported cases are:
controlling the return value:
my_stub(1, 2, 3).and_return(666)
calling another function (no arguments):
my_stub(1, 2, 3).and_call(callback)
calling another function with certain arguments/keyword arguments:
my_stub(1, 2, 3).and_call(callback, args=(100, 200), kwargs={'some_arg':20})
calling another function (with the arguments of the call):
my_stub(1, 2, 3).and_call_with_args(callback)
raising an exception (happens after all callbacks are fired):
my_stub(1, 2, 3).and_raise(MyException())
If you don't know the exact value that the argument to a function is going to get, you sometimes have to use predicates to help you distinguish valid cases from invalid ones. For starters we'll mention that mock objects will only compare 'true' to themselves, so you shouldn't worry about any funky business as far as mock comparison goes.
To complete the picture, if you want to assert all sorts of checks on the arguments you are recording, you can use comparators. For instance, the following doesn't care about which argument is passed to 'name', as long as it is a string:
my_stub(name=IsA(basestring))
Many comparators exist in Forge:
Is(x)
: compares true only if the argument is xIsA(type)
: compares true only if the argument is of type typeRegexpMatches(regexp, [flags])
: compares true only if the argument is a string, and matches regexpFunc(f)
: compares true only if f returns True for the argumentIsAlmost(value, [places])
: compares true only if the argument is almost identical to value, by places digits after the floating pointContains(element)
: compares true only if element exists in the argumentStrContains(substring)
: compares true only if substring exists in the argument, and the argument is a stringHasKeyValue(key, value)
: compares true only if the argument has key as a key, whose value is valueHasAttributeValue(attr, value)
: same as HasKeyValue, but for attributesAnything()
: always compares trueAnd(...), Or(...), Not(c)
: and, or and a negator for other comparators
Forge includes a mechanism for installing (and later removing) stubs instead of ordinary methods and functions:
>>> import time >>> forge_manager.replace(time, "time") # doctest: +ELLIPSIS <...> >>> time.time().and_return(2) 2 >>> forge_manager.replay() >>> time.time() 2 >>> forge_manager.verify() >>> forge_manager.restore_all_replacements() >>> forge_manager.reset() This also works, of course, on methods: >>> class MyClass(object): ... def f(self): ... self.g() ... def g(self): ... raise NotImplementedError() >>> instance = MyClass() >>> forge_manager.replace(instance, "g") # doctest: +ELLIPSIS <...> >>> instance.g() # doctest: +ELLIPSIS <...> >>> forge_manager.replay() >>> instance.f() >>> forge_manager.verify() >>> forge_manager.restore_all_replacements() >>> forge_manager.reset()
One can also use the same install mechanism to set a custom value and have it restored along with all stubs:
>>> class SomeClass(object): ... x = 2 >>> forge_manager.replace_with(SomeClass, "x", 3) 3 >>> SomeClass.x 3 >>> forge_manager.restore_all_replacements() >>> SomeClass.x 2
Replacing is also supported within a context, restoring the installed stub upon exit from the context:
>>> with forge_manager.replacing_context(SomeClass, "x"): ... pass
By default, forge verifies that the order in which calls are made in practice is the same as the record flow. You can, however, control it and create groups in which order does not matter:
>>> class SomeClass(object): ... def func(self, arg): ... pass >>> mock = forge_manager.create_mock(SomeClass) >>> mock.func(1) # doctest: +ELLIPSIS <...> >>> mock.func(2) # doctest: +ELLIPSIS <...> >>> mock.func(3) # doctest: +ELLIPSIS ... # so far order must be kept <...> >>> with forge_manager.any_order(): # doctest: +ELLIPSIS ... mock.func(4) ... mock.func(5) <...> <...> >>> mock.func(6) # doctest: +ELLIPSIS <...> >>> forge_manager.replay() >>> mock.func(1) >>> mock.func(2) >>> mock.func(3) >>> mock.func(5) # ok! >>> mock.func(4) # also ok! >>> mock.func(6) >>> forge_manager.verify() >>> forge_manager.reset()
You can always nest ordering groups, by using ordered, any_order and interleaved_order (see below)
>>> with forge_manager.any_order(): # doctest: +ELLIPSIS ... mock.func(4) ... with forge_manager.ordered(): ... mock.func(5) ... mock.func(6) ... mock.func(7) <...> <...> <...> <...>
In the example above, func(5) and func(6) will be asserted to occur in this specific order, but the group can appear anywhere among func(4) and func(7).
>>> forge_manager.replay() >>> for i in (5, 6, 7, 4): ... _ = mock.func(i) >>> forge_manager.verify() >>> forge_manager.reset()
In the context of nested ordering groups, the interleaved ordering may come in handy when working with coroutines/greenlets:
>>> class SomeClass(object): ... def foo(self, arg): ... pass ... def bar(self, arg): ... pass >>> mock = forge_manager.create_mock(SomeClass) >>> with forge_manager.interleaved_order(): # doctest: +ELLIPSIS ... with forge_manager.ordered(): ... mock.foo(1) ... mock.foo(2) ... with forge_manager.ordered(): ... mock.bar(1) ... mock.bar(2) <...> <...> <...> <...> >>> forge_manager.replay() >>> mock.foo(1) >>> mock.bar(1) >>> mock.foo(2) >>> mock.bar(2) >>> forge_manager.verify() >>> forge_manager.reset()
The expectation above will work with following sequence as well:
>>> with forge_manager.interleaved_order(): # doctest: +ELLIPSIS ... with forge_manager.ordered(): ... mock.foo(1) ... mock.foo(2) ... with forge_manager.ordered(): ... mock.bar(1) ... mock.bar(2) <...> <...> <...> <...> >>> forge_manager.replay() >>> mock.bar(1) >>> mock.bar(2) >>> mock.foo(1) >>> mock.foo(2) >>> forge_manager.verify() >>> forge_manager.reset()
Sometimes you intend for a function to be called zero or more times, regardless of timing, and return a certain value (or raise an exception). There are ugly ways to do this:
>>> class MyObj(object): ... def f(self): ... pass >>> m = forge_manager.create_mock(MyObj) >>> m.f = lambda: 2 # yuck!
And of course the downside is that:
- the fact that f exists doesn't get verified. Also its signature is not verified with this method.
- lambda's are ugly, and it gets nastier when you want to use exceptions.
whenever() to the rescue - it is a method that can be called on expected methods, causing the call to be accepted, signature checked and acted upon. However, unlike regular recordings, it expects the call 0 or more times, at any point - so it achieves the same effect:
>>> m = forge_manager.create_mock(MyObj) >>> m.f().whenever().and_return(2) 2 >>> forge_manager.replay() >>> m.f() 2 >>> m.f() 2 >>> forge_manager.verify() >>> forge_manager.reset()
Multiple whenever() recordings can be specified with different parameters, which results in a form of "pattern matching" for the requested calls (each call signature will result in a different return value).
An alternative syntax exists for whenever() for easier readability:
>>> class Obj(object): ... def f(self, value): ... pass >>> m = forge_manager.create_mock(Obj) >>> m.f.when(2).then_return(3) 3 >>> forge_manager.replay() >>> m.f(2) 3 >>> forge_manager.verify() >>> forge_manager.reset()
Note
whenever() calls always apply to the ordering group in which they were recorded. This means that once an order group is cleared, all of the *whenever*s recorded in it are automatically "forgotten", and will no longer be accepted on replay.
Although not recommended, sometimes you just want a mock that accepts anything during record, and just verifies that you stick to it in replay. This is useful for prototyping an interface that doesn't exist yet. This is done in Forge by using wildcard mocks:
>>> mock = forge_manager.create_wildcard_mock() >>> mock <<Wildcard>> >>> stub = forge_manager.create_wildcard_function_stub() >>> stub <Stub for '<<Wildcard>>'> >>> mock.f() # doctest: +ELLIPSIS <...> >>> mock.g(1, 2, 3, d=4) # doctest: +ELLIPSIS <...> >>> stub() # doctest: +ELLIPSIS <...> >>> stub(1, 2, 3, d=4) # doctest: +ELLIPSIS <...> >>> forge_manager.replay() >>> mock.f() >>> mock.g(1, 2, 3, d=4) >>> stub() >>> stub(1, 2, 3, d=4) >>> forge_manager.reset()
Sometimes you would like to simulate the behavior of a class, and not an object. Forge allows to do this with the create_class_mock API:
>>> class MyClass(object): ... def __init__(self, a, b, c): ... pass ... def regular_method(self): ... pass ... @classmethod ... def some_class_method(cls): ... pass ... @staticmethod ... def some_static_method(): ... pass >>> class_mock = forge_manager.create_class_mock(MyClass) >>> class_mock <Class mock of 'MyClass'> >>> class_mock.regular_method() # doctest: +IGNORE_EXCEPTION_DETAIL Traceback (most recent call last): SignatureException: ... >>> class_mock.some_class_method() # doctest: +ELLIPSIS <...> >>> class_mock.some_static_method() # doctest: +ELLIPSIS <...> >>> fake_new_instance = forge_manager.create_mock(MyClass) >>> class_mock(1, 2, 3).and_return(fake_new_instance) # doctest: +ELLIPSIS <...> >>> forge_manager.replay() >>> class_mock.some_class_method() >>> class_mock.some_static_method() >>> assert class_mock(1, 2, 3) is fake_new_instance >>> forge_manager.verify() >>> forge_manager.reset()
Suppose you have a class like the following:
>>> class File(object): ... def __init__(self, filename): ... self.f = open(filename, "rb") ... def read(self, size): ... raise NotImplementedError() ... def log(self, buffer): ... raise NotImplementedError() ... def read_and_log(self, size): ... data = self.read(size) ... self.log(data) ... return data
Now, suppose you want to write a test for read_and_log, while mimicking the behavior of read() and log(). This is quite common, because sometimes methods in your classes have lots of side effects which are hard to plumb during test writing. One easy approach would be to create a File object and to replace read() and log() with stubs (see above). This is fine, but the problem is with the class construction, which opens a file for reading.
In some cases, constructors (especially in legacy code to which you add tests) do lots of things that are hard to stub, or that are likely to change thus breaking any stubbing work you might install. For this case Forge has hybrid mocks:
>>> mock = forge_manager.create_hybrid_mock(File) >>> mock.read(20).and_return("data") # doctest: +ELLIPSIS '...' >>> mock.log("data") # doctest: +ELLIPSIS <...> >>> forge_manager.replay() >>> assert mock.read_and_log(20) == "data" >>> forge_manager.verify() >>> forge_manager.reset()
Hybrid mocks are, well, hybrid. They behave as regular mocks during record, but calling any method during replay that hasn't been recorded will invoke the original method on the mock, thus testing it in an isolated environment.
A class equivalent also exists:
>>> class SomeClass(object): ... def __init__(self, parameter): ... raise NotImplementedError() ... @classmethod ... def constructor(cls): ... return cls(1) >>> mock = forge_manager.create_hybrid_class_mock(SomeClass) >>> expected_return_value = forge_manager.create_sentinel() >>> mock(1).and_return(expected_return_value) # doctest: +ELLIPSIS <...> >>> forge_manager.replay() >>> got_return_value = mock.constructor() >>> got_return_value is expected_return_value True