**Test classes, modules or whole projects can make use of one or more fixtures**. All required fixture functions will execute before a test from the specifying context executes. As You can use this to make tests operate from a pre-initialized directory or with certain environment variables or with pre-configured global application settings. For example, the Django_ project requires database initialization to be able to import from and use its model objects. For that, the `pytest-django`_ plugin provides fixtures which your project can then easily depend or extend on, simply by referencing the name of the particular fixture. Fixture functions have limited visilibity which depends on where they are defined. If they are defined on a test class, only its test methods may use it. A fixture defined in a module can only be used from that test module. A fixture defined in a conftest.py file can only be used by the tests below the directory of that file. Lastly, plugins can define fixtures which are available across all projects. Python, Java and many other languages support a so called xUnit_ style for providing a fixed state, `test fixtures`_, for running tests. It typically involves calling a autouse function ahead and a teardown function after test execute. In 2005 pytest introduced a scope-specific model of automatically detecting and calling autouse and teardown functions on a per-module, class or function basis. The Python unittest package and nose have subsequently incorporated them. This model remains supported by pytest as :ref:`classic xunit`. One property of xunit fixture functions is that they work implicitely by preparing global state or setting attributes on TestCase objects. By contrast, pytest provides :ref:`funcargs` which allow to dependency-inject application test state into test functions or methods as function arguments. If your application is sufficiently modular or if you are creating a new project, we recommend you now rather head over to :ref:`funcargs` instead because many pytest users agree that using this paradigm leads to better application and test organisation. However, not all programs and frameworks work and can be tested in a fully modular way. They rather require preparation of global state like database autouse on which further fixtures like preparing application specific tables or wrapping tests in transactions can take place. For those needs, pytest-2.3 now supports new **fixture functions** which come with a ton of improvements over classic xunit fixture writing. Fixture functions: - allow to separate different autouse concerns into multiple modular functions - can receive and fully interoperate with :ref:`funcargs `, - are called multiple times if its funcargs are parametrized, - don't need to be defined directly in your test classes or modules, they can also be defined in a plugin or :ref:`conftest.py ` files and get called - are called on a per-session, per-module, per-class or per-function basis by means of a simple "scope" declaration. - can access the :ref:`request ` object which allows to introspect and interact with the (scoped) testcontext. - can add cleanup functions which will be invoked when the last test of the fixture test context has finished executing. All of these features are now demonstrated by little examples. test modules accessing a global resource ------------------------------------------------------- .. note:: Relying on `global state is considered bad programming practise `_ but when you work with an application that relies on it you often have no choice. If you want test modules to access a global resource, you can stick the resource to the module globals in a per-module autouse function. We use a :ref:`resource factory <@pytest.fixture>` to create our global resource:: # content of conftest.py import pytest class GlobalResource: def __init__(self): pass @pytest.fixture(scope="session") def globresource(): return GlobalResource() @pytest.fixture(scope="module") def setresource(request, globresource): request.module.globresource = globresource Now any test module can access ``globresource`` as a module global:: # content of test_glob.py def test_1(): print ("test_1 %s" % globresource) def test_2(): print ("test_2 %s" % globresource) Let's run this module without output-capturing:: $ py.test -qs test_glob.py FF ================================= FAILURES ================================= __________________________________ test_1 __________________________________ def test_1(): > print ("test_1 %s" % globresource) E NameError: global name 'globresource' is not defined test_glob.py:3: NameError __________________________________ test_2 __________________________________ def test_2(): > print ("test_2 %s" % globresource) E NameError: global name 'globresource' is not defined test_glob.py:5: NameError The two tests see the same global ``globresource`` object. Parametrizing the global resource +++++++++++++++++++++++++++++++++++++++++++++++++ We extend the previous example and add parametrization to the globresource factory and also add a finalizer:: # content of conftest.py import pytest class GlobalResource: def __init__(self, param): self.param = param @pytest.fixture(scope="session", params=[1,2]) def globresource(request): g = GlobalResource(request.param) def fin(): print "finalizing", g request.addfinalizer(fin) return g @pytest.fixture(scope="module") def setresource(request, globresource): request.module.globresource = globresource And then re-run our test module:: $ py.test -qs test_glob.py FF ================================= FAILURES ================================= __________________________________ test_1 __________________________________ def test_1(): > print ("test_1 %s" % globresource) E NameError: global name 'globresource' is not defined test_glob.py:3: NameError __________________________________ test_2 __________________________________ def test_2(): > print ("test_2 %s" % globresource) E NameError: global name 'globresource' is not defined test_glob.py:5: NameError We are now running the two tests twice with two different global resource instances. Note that the tests are ordered such that only one instance is active at any given time: the finalizer of the first globresource instance is called before the second instance is created and sent to the autouse functions.