Reuse a fixture with parameters from another fixture

[Wovchena]

Downloading images from the Internet is flaky. I'm trying to reuse the images in tests to reduce instability. I also want the images to be removed from memory after they aren't needed. I'm trying to achieve that with the following code

import pytest

@pytest.fixture(scope="module")
def image(request):
    print(f"image-{request.param}")
    return f"image-{request.param}"

@pytest.fixture(scope="module")
def images(request):
    print(f"images{request.param}")
    values = []
    for val in request.param:
        request._pyfuncitem.callspec.params["image"] = val
        request._pyfuncitem.callspec.indices["image"] = 0
        request._pyfuncitem.callspec._arg2scope["image"] = request._pyfuncitem.callspec._arg2scope["images"]
        values.append(request.getfixturevalue('image'))
    return "images[" + ' '.join(values) + "]"

@pytest.mark.parametrize("image", [1, 2], indirect=True)
def test_image(image):
    print(f"test_image: {image}")

@pytest.mark.parametrize("images", [[], [1], [1, 2, 3]], indirect=True)
def test_images(images):
    print(f"test_images: {images}")

test_images[images2] prints test_images: images[image-1 image-1 image-1] instead of 1, 2, 3. I'd like request.getfixturevalue() to have a parameter to be passed to the fixture. I've checked pytest-lazy-fixture but its fixtures cannot be parametrized (https://stackoverflow.com/a/62166129/25242963).

[RonnyPfannschmidt]

Due to the removal of relevant context all i can say with the code provided is make the parameter a own fixture and stop missing with the internals

[Wovchena]

What context is missing?
Can you link an example to make the parameter a own fixture

[webknjaz]

from dataclasses import dataclass

import pytest


@dataclass
class DataShape:
    param1: int
    param2: str


@pytest.fixture
def common_params(request: pytest.FixtureRequest) -> DataShape:
    return request.param


@pytest.fixture
def param1(common_params: DataShape) -> int:
    return common_params.param1


@pytest.fixture
def param2(common_params: DataShape) -> str:
    return common_params.param2


@pytest.fixture
def stash(param2: str) -> str:
    return param2


@pytest.mark.parametrize(
    'common_params',
    (DataShape(42, 'криївка'), ),
    indirect=True,
)
@pytest.mark.usefixtures('common_params')
def test_accessing_param2(stash: str) -> None:
    assert stash == 'криївка'

[Wovchena]

Suppose param1 is complex data derived from 42. And param2 is complex data derived from криївка. I'm trying to avoid constructing param1 if a test requires just param2 and if test_accessing_param2() is the only requested test to run. At the same time there's another test that requires the whole DataShape. If both tests are requested, param2 should be created only once. I tried to illustrate this here:

import pytest


def flaky_computation1(param1):
    return param1 + 1


def flaky_computation2(param2):
    return param2 + "!"


class DataShape:
    def __init__(self, param1: int, param2: str):
        self.param1 = flaky_computation1(param1)
        self.param2 = flaky_computation2(param2)


@pytest.fixture
def common_params(request: pytest.FixtureRequest) -> DataShape:
    return request.param


@pytest.fixture
def param1(common_params: DataShape) -> int:
    return common_params.param1


@pytest.fixture
def param2(common_params: DataShape) -> str:
    return common_params.param2


@pytest.fixture
def stash(param2: str) -> str:
    return param2

@pytest.mark.parametrize(
    'common_params',
    (DataShape(42, 'криївка'), ),  # shouldn't call flaky_computation1()
    indirect=True,
)
@pytest.mark.usefixtures('common_params')
def test_accessing_param2(stash: str) -> None:
    assert stash == 'криївка!'

@pytest.mark.parametrize(
    'common_params',
    (DataShape(42, 'криївка'), ),  # should be reused from the parameter of test_accessing_param2
    indirect=True,
)
def test_full(common_params) -> None:
    assert str(common_params.param1) + common_params.param2 == "43криївка!"

[The-Compiler]

Converting this into a discussion because I don't see anything actionable for pytest here.