No garbage collection between parametrized test instances.

[Vert53]

Using Pytest version: 3.8.0 on ubuntu 18 and python3.6

I encountered an issue where a parametrized test uses objects which define a destructor (del) that is needed to be called between test instances of the same test function. I noticed that the garbage collector doesn't collect these objects unless I explicitly del them. I assume this might be because of some inner behavior of pytest, but this is highly misleading and has potential for problems.

class Resource:
    def __init__(self):
        acquire_resource()
    def __del__(self):
        release_resource()


@pytest.mark.parametrize('param', (1,2,3))
def test_resource:
    res = Resource()

I except res to be collected when it is out of the scope of the function but it is not unless we call:
del res

Please let me know if this behavior is expected for some reason I missed.
If it is not perhaps we should fix it in the future

[RonnyPfannschmidt]

It's expected, use context managers

[asottile]

if you're using __del__ for resource freeing you're depending on borderline-undefined-cpython-implementation-details. as @RonnyPfannschmidt indicates you should adjust your api to use a context manager

[nicoddemus]

Just to complement the answers: this is not pytest specific, but how Python behaves; you will get the same behavior if you call test_resource() manually without using pytest at all.

[Vert53]

Thanks for the answers. Due to this being caused by garbage collector working in the background I found a possible workaround that might work:
Using the python gc module, gc.collect() can be run in the beginning of the test to block until the gc collects all objects from previous runs.
Do you think this is a good idea? If it is, perhaps some kind of "collect" flag could be added to ensure this happens.
Just a note, the reason in this case resource clean up is critical is due to working with a C++ library during the test using ctypes.

[RonnyPfannschmidt]

if you manage actual resources in c, use a context manager, anything else is bad practice and at best a hack

the garbage collector is not a resource manager, pretending it is, and making it work with g collect in tests, will in turn cause your library to be a absolute timebomb, its just asking for a horrendous error to happen

[nicoddemus]

I suggest to use an autouse fixture to ensure gc.collect() is called for each test:

@pytest.fixture(autouse=True)
def ensure_gc():
    gc.collect()

[Vert53]

The default (and most granular) scope is "function". Doesn't that mean that this fixture will not be called for each instance of the same test function? But rather once and then shared between the instances?

[nicoddemus]

It will be called before each test function is called, even parametrized ones.

[asottile]

running gc.collect like this does not mirror how python actually works though, so you're testing an environment which is unrealistic

[Vert53]

Thanks for the advice everyone