Skip to content

Parametrized testing

Parametrization is a process of running the same test with varying sets of data. Each combination of a test and data is counted as a new test case.

There are multiple ways to parametrize your tests, each differs in complexity and flexibility.

Parametrize the test

The most simple form of parametrization is at test level:

@pytest.mark.parametrize("number", [1, 2, 3, 0, 42])
def test_foo(number):
    assert number > 0

In this case we are getting five tests: for number 1, 2, 3, 0 and 42. Each of those tests can fail independently of one another (if in this example the test with 0 will fail, and four others will pass).

Parametrize the fixtures

Fixtures may have parameters. Those parameters are passed as a list to the argument params of @pytest.fixture() decorator.

Those parameters must be iterables, such as lists. Each parameter to a fixture is applied to each function using this fixture. If a few fixtures are used in one test function, pytest generates a Cartesian product of parameters of those fixtures.

To use those parameters, a fixture must consume a special fixture named request. It provides the special (built-in) fixture with some information on the function it deals with. request also contains request.param which contains one element from params. The fixture called as many times as the number of elements in the iterable of params argument, and the test function is called with values of fixtures the same number of times. (basically, the fixture is called len(iterable) times with each next element of iterable in the request.param).

@pytest.fixture(params=["one", "uno"])
def fixture1(request):
    return request.param

@pytest.fixture(params=["two", "duo"])
def fixture2(request):
    return request.paramdef test_foobar(fixture1, fixture2):
    assert type(fixture1) == type(fixture2)

The output is:

#OUTPUT 3
collected 4 itemstest_3.py::test_foobar[one-two] PASSED  [ 25%]
test_3.py::test_foobar[one-duo] PASSED  [ 50%]
test_3.py::test_foobar[uno-two] PASSED  [ 75%]
test_3.py::test_foobar[uno-duo] PASSED  [100%]

Parametrization with pytest_generate_tests

There is an another way to generate arbitrary parametrization at collection time. It’s a bit more direct and verbose, but it provides introspection of test functions, including the ability to see all other fixture names.

At collection time Pytest looks up for and calls (if found) a special function in each module, named pytest_generate_tests. This function is not a fixture, but just a regular function. It receives the argument metafunc, which itself is not a fixture, but a special object.

pytest_generate_tests is called for each test function in the module to give a chance to parametrize it. Parametrization may happen only through fixtures that test function requests. There is no way to parametrize a test function like this:

def test_simple(): assert 2+2 == 4

You need some variables to be used as parameters, and those variables should be arguments to the test function. Pytest will replace those arguments with values from fixtures, and if there are a few values for a fixture, then this is parametrization at work.

metafunc argument to pytest_generate_tests provides some useful information on a test function:

  • Ability to see all fixture names that function requests.
  • Ability to see the name of the function.
  • Ability to see code of the function.

Finally, metafunc has a parametrize function, which is the way to provide multiple variants of values for fixtures.

The same case as before written with the pytest_generate_tests function is:

def pytest_generate_tests(metafunc):
    if "fixture1" in metafunc.fixturenames:
        metafunc.parametrize("fixture1", ["one", "uno"])
    if "fixture2" in metafunc.fixturenames:
        metafunc.parametrize("fixture2", ["two", "duo"])

def test_foobar(fixture1, fixture2):
    assert type(fixture1) == type(fixture2)

This solution is a little bit magical, so I'd avoid it in favor of pytest-cases.

Use pytest-cases

pytest-case gives a lot of power when it comes to tweaking the fixtures and parameterizations.

Check that file for further information.

Customizations

Change the tests name

Sometimes you want to change how the tests are shown so you can understand better what the test is doing. You can use the ids argument to pytest.mark.parametrize.

File tests/unit/test_func.py

tasks_to_try = (
    Task('sleep', done=True),
    Task('wake', 'brian'),
    Task('wake', 'brian'),
    Task('breathe', 'BRIAN', True),
    Task('exercise', 'BrIaN', False),
)

task_ids = [
    f'Task({task.summary}, {task.owner}, {task.done})'
    for task in tasks_to_try
]

@pytest.mark.parametrize('task', tasks_to_try, ids=task_ids)
def test_add_4(task):
    task_id = tasks.add(task)
    t_from_db = tasks.get(task_id)
    assert equivalent(t_from_db, task)
$ pytest -v test_func.py::test_add_4
===================== test session starts ======================
collected 5 items

test_add_variety.py::test_add_4[Task(sleep,None,True)] PASSED
test_add_variety.py::test_add_4[Task(wake,brian,False)0] PASSED
test_add_variety.py::test_add_4[Task(wake,brian,False)1] PASSED
test_add_variety.py::test_add_4[Task(breathe,BRIAN,True)] PASSED
test_add_variety.py::test_add_4[Task(exercise,BrIaN,False)] PASSED

=================== 5 passed in 0.04 seconds ===================

Those identifiers can be used to run that specific test. For example pytest -v "test_func.py::test_add_4[Task(breathe,BRIAN,True)]".

parametrize() can be applied to classes as well.

If the test id can't be derived from the parameter value, use the id argument for the pytest.param:

@pytest.mark.parametrize('task', [
    pytest.param(Task('create'), id='just summary'),
    pytest.param(Task('inspire', 'Michelle'), id='summary/owner'),
])
def test_add_6(task):
    ...

Will yield:

$ pytest-v test_add_variety.py::test_add_6

=================== test session starts ====================
collected 2 items

test_add_variety.py::test_add_6[justsummary]PASSED
test_add_variety.py::test_add_6[summary/owner]PASSED

================= 2 passed in 0.05 seconds =================

References