Try to fix mypy issue with calculate_md5

[vfdev-5]

• https://app.circleci.com/pipelines/github/pytorch/vision/19736/workflows/98540a35-693d-4f68-92e4-cbdcc2601d81/jobs/1599504

and passing locally on python 3.9 without errors like:

# mypy --config-file mypy.ini
torchvision/datasets/utils.py:68: error: Argument 1 to "md5" has incompatible type "**Dict[str, bool]"; expected
"Union[bytes, Union[bytearray, memoryview, array[Any], mmap, _CData, PickleBuffer]]"  [arg-type]
        md5 = hashlib.md5(**dict(usedforsecurity=False) if sys.version_info >= (3, 9) else dict())

[NicolasHug]

IIRC this only fails on 3.10. @pmeier had an attempt to fix this at #5755 but we weren't too enthusiastic about the solution

[vfdev-5]

@NicolasHug yes, @pmeier pointed out to me his PR.

Let me recap the findings from my local tests using conda env on macosx :

code	python version	mypy check
main branch	3.8	failure with md5 = hashlib.md5(**dict(usedforsecurity=False) if sys.version_info >= (3, 9) else dict())
main branch	3.9	failure with md5 = hashlib.md5(**dict(usedforsecurity=False) if sys.version_info >= (3, 9) else dict())
main branch	3.10	failure with md5 = hashlib.md5(**dict(usedforsecurity=False) if sys.version_info >= (3, 9) else dict())
This PR	3.8	passing
This PR	3.9	passing
This PR	3.10	passing

mypy version: mypy 0.971 (compiled: yes)

[datumbox]

LGTM! It simplifies the code as well while not specifying Python to 3.10 as the previous attempt.

[pmeier]

It seems that mypy indeed recognizes the sys.version_info branch: https://mypy-play.net/?mypy=latest&python=3.10&gist=3f2a763676db35f08960e3bad08b62f3

while not specifying Python to 3.10 as the previous attempt.

My PR is not primarily about fixing this issue, but rather making these issues visible in CI. The change made in this PR cannot be verified by CI, but has to be done manually. Meaning, if someone introduces a change like I did in #5717, we will not know until someone on Python >= 3.7 is getting errors on main.

[NicolasHug]

mypy will always surprise me at how it behaves so differently on 2 pieces of code which look like the exact same thing externally...

Thanks @vfdev-5, I agree we can merge as this issue has been an inconvenience for local development

[github-actions]

Hey @vfdev-5!

You merged this PR, but no labels were added. The list of valid labels is available at https://github.com/pytorch/vision/blob/main/.github/process_commit.py