pytest-mock mock的高层封装

pytest-mock

pytest-mock是一个pytest的插件，安装即可使用。 它提供了一个名为mocker的fixture，仅在当前测试function或method生效，而不用自行包装。

object

mock一个object，是最常见的需求。 由于function也是一个object，以下以function举例。

import os


def rm(filename):
    os.remove(filename)


def test_rm(mocker):
    filename = 'test.file'
    mocker.patch('os.remove')
    rm(filename)
    os.remove.assert_called_once_with(filename)

这里在给os.remove打了一个patch，让它变成了一个MagicMock。 然后利用assert_called_once_with，查看它是否被调用一次，并且参数为filename。

注意：只能对已经存在的东西使用mock。

method

有时，仅仅需要mock一个object里的method，而无需mock整个object。 例如，在对当前object的某个method进行测试时。 这时，可以用patch.object。

class ForTest:
    field = 'origin'

    def method():
        pass


def test_for_test(mocker):
    test = ForTest()
    mock_method = mocker.patch.object(test, 'method')
    test.method()
    assert mock_method.called

    assert 'origin' == test.field
    mocker.patch.object(test, 'field', 'mocked')
    assert 'mocked' == test.field

上例中，分别对field和method进行了mock。 当然，对一个给定module的function，也能使用。

def test_patch_object_listdir(mocker):
    mock_listdir = mocker.patch.object(os, 'listdir')
    os.listdir()
    assert mock_listdir.called

用spy包装

如果只是想用MagicMock包装一个东西，而又不想改变其功能，可以用spy。

def test_spy_listdir(mocker):
    mock_listdir = mocker.spy(os, 'listdir')
    os.listdir()
    assert mock_listdir.called

与上例中的patch.object不同的是，上例的os.listdir()不会真的执行，而本例中则会真的执行。

pytest-mock

This plugin installs a mocker fixture which is a thin-wrapper around the patching API provided by the mock package, but with the benefit of not having to worry about undoing patches at the end of a test:

import os

class UnixFS:

    @staticmethod
    def rm(filename):
        os.remove(filename)

def test_unix_fs(mocker):
    mocker.patch('os.remove')
    UnixFS.rm('file')
    os.remove.assert_called_once_with('file')

Professionally supported pytest-mock is now available

Usage

The mocker fixture has the same API as mock.patch, supporting the same arguments:

def test_foo(mocker):
    # all valid calls
    mocker.patch('os.remove')
    mocker.patch.object(os, 'listdir', autospec=True)
    mocked_isfile = mocker.patch('os.path.isfile')

The supported methods are:

• mocker.patch
• mocker.patch.object
• mocker.patch.multiple
• mocker.patch.dict
• mocker.stopall
• mocker.resetall(): calls reset_mock() in all mocked objects up to this point.

These objects from the mock module are accessible directly from mocker for convenience:

• Mock
• MagicMock
• PropertyMock
• ANY
• DEFAULT (Version 1.4)
• call (Version 1.1)
• sentinel (Version 1.2)
• mock_open

Spy

The spy acts exactly like the original method in all cases, except it allows use of mock features with it, like retrieving call count. It also works for class and static methods.

def test_spy(mocker):
    class Foo(object):
        def bar(self):
            return 42

    foo = Foo()
    mocker.spy(foo, 'bar')
    assert foo.bar() == 42
    assert foo.bar.call_count == 1

Since version 1.11, it is also possible to query the return_value attribute to observe what the spied function/method returned.

Since version 1.13, it is also possible to query the side_effect attribute to observe any exception thrown by the spied function/method.

Stub

The stub is a mock object that accepts any arguments and is useful to test callbacks, for instance. May be passed a name to be used by the constructed stub object in its repr (useful for debugging).

def test_stub(mocker):
    def foo(on_something):
        on_something('foo', 'bar')

    stub = mocker.stub(name='on_something_stub')

    foo(stub)
    stub.assert_called_once_with('foo', 'bar')

Improved reporting of mock call assertion errors

This plugin monkeypatches the mock library to improve pytest output for failures of mock call assertions like Mock.assert_called_with() by hiding internal traceback entries from the mock module.

It also adds introspection information on differing call arguments when calling the helper methods. This features catches AssertionError raised in the method, and uses py.test's own advanced assertions to return a better diff:

mocker = <pytest_mock.MockFixture object at 0x0381E2D0>

    def test(mocker):
        m = mocker.Mock()
        m('fo')
>       m.assert_called_once_with('', bar=4)
E       AssertionError: Expected call: mock('', bar=4)
E       Actual call: mock('fo')
E
E       pytest introspection follows:
E
E       Args:
E       assert ('fo',) == ('',)
E         At index 0 diff: 'fo' != ''
E         Use -v to get the full diff
E       Kwargs:
E       assert {} == {'bar': 4}
E         Right contains more items:
E         {'bar': 4}
E         Use -v to get the full diff


test_foo.py:6: AssertionError
========================== 1 failed in 0.03 seconds ===========================

This is useful when asserting mock calls with many/nested arguments and trying to quickly see the difference.

This feature is probably safe, but if you encounter any problems it can be disabled in your pytest.ini file:

[pytest]
mock_traceback_monkeypatch = false

Note that this feature is automatically disabled with the --tb=native option. The underlying mechanism used to suppress traceback entries from mock module does not work with that option anyway plus it generates confusing messages on Python 3.5 due to exception chaining

Use standalone "mock" package

New in version 1.4.0.

Python 3 users might want to use a newest version of the mock package as published on PyPI than the one that comes with the Python distribution.

[pytest]
mock_use_standalone_module = true

This will force the plugin to import mock instead of the unittest.mock module bundled with Python 3.4+. Note that this option is only used in Python 3+, as Python 2 users only have the option to use the mock package from PyPI anyway.

Note about usage as context manager

Although mocker's API is intentionally the same as mock.patch's, its use as context manager and function decorator is not supported through the fixture:

def test_context_manager(mocker):
    a = A()
    with mocker.patch.object(a, 'doIt', return_value=True, autospec=True):  # DO NOT DO THIS
        assert a.doIt() == True

The purpose of this plugin is to make the use of context managers and function decorators for mocking unnecessary.

Requirements

• Python 2.7, Python 3.4+
• pytest
• mock (for Python 2)

Install

Install using pip:

$ pip install pytest-mock

Changelog

Please consult the changelog page.

Why bother with a plugin?

There are a number of different patch usages in the standard mock API, but IMHO they don't scale very well when you have more than one or two patches to apply.

It may lead to an excessive nesting of with statements, breaking the flow of the test:

import mock

def test_unix_fs():
    with mock.patch('os.remove'):
        UnixFS.rm('file')
        os.remove.assert_called_once_with('file')

        with mock.patch('os.listdir'):
            assert UnixFS.ls('dir') == expected
            # ...

    with mock.patch('shutil.copy'):
        UnixFS.cp('src', 'dst')
        # ...

One can use patch as a decorator to improve the flow of the test:

@mock.patch('os.remove')
@mock.patch('os.listdir')
@mock.patch('shutil.copy')
def test_unix_fs(mocked_copy, mocked_listdir, mocked_remove):
    UnixFS.rm('file')
    os.remove.assert_called_once_with('file')

    assert UnixFS.ls('dir') == expected
    # ...

    UnixFS.cp('src', 'dst')
    # ...

But this poses a few disadvantages:

• test functions must receive the mock objects as parameter, even if you don't plan to access them directly; also, order depends on the order of the decorated patch functions;
• receiving the mocks as parameters doesn't mix nicely with pytest's approach of naming fixtures as parameters, or pytest.mark.parametrize;
• you can't easily undo the mocking during the test execution;

An alternative is to use contextlib.ExitStack to stack the context managers in a single level of indentation to improve the flow of the test:

import contextlib
import mock

def test_unix_fs():
    with contextlib.ExitStack() as stack:
        stack.enter_context(mock.patch('os.remove'))
        UnixFS.rm('file')
        os.remove.assert_called_once_with('file')

        stack.enter_context(mock.patch('os.listdir'))
        assert UnixFS.ls('dir') == expected
        # ...

        stack.enter_context(mock.patch('shutil.copy'))
        UnixFS.cp('src', 'dst')
        # ...

But this is arguably a little more complex than using pytest-mock.

Contributing

Contributions are welcome! After cloning the repository, create a virtual env and install pytest-mock in editable mode with dev extras:

$ pip install --editable .[dev]
$ pre-commit install

Tests are run with tox, you can run the baseline environments before submitting a PR:

$ tox -e py27,py36,linting

Style checks and formatting are done automatically during commit courtesy of pre-commit.

License

Distributed under the terms of the MIT license.

Security contact information

To report a security vulnerability, please use the Tidelift security contact. Tidelift will coordinate the fix and disclosure.