Httpx: Type-checking our tests

[florimondmanca]

I know this is not a standard thing to do across Encode projects, but I've been wondering if it would be worth starting to type-hint our tests.

I've seen at least two instances of this recently:

• In HTTPX: https://github.com/encode/httpx/pull/648#discussion_r359862603
• In Starlette: https://github.com/encode/starlette/issues/722

My rationale is based on two aspects:

• It improves our upfront knowledge about how users will actually use HTTPX — currently their usage of type hints in the wild is not reflected anywhere.
• It helps us catch type hint inconsistencies we wouldn't see in the core package.

The main counter-argument, I suppose, is that type hinting tests is tedious. I think that's fair, but I believe the two pro's above make it compelling.

Thoughts?

[florimondmanca]

Alternatively, we could tweak the mypy config:

[mypy]
files = httpx, tests
ignore_missing_imports = True
disallow_untyped_defs = False
check_untyped_defs = True

[mypy-httpx]
disallow_untyped_defs = True
check_untyped_defs = False

Running this locally raises 141 errors, but mostly on internal APIs, so, hmm.

[aalvrz]

I think this is a great idea, would like to work on this. Since I am pretty new to httpx, adding these type hints to the tests will give me a better picture of how httpx works under the hood.

[florimondmanca]

I think this is still a good idea, though it's a bit of busy work to upgrade our existing test code.

Let's perhaps just remind ourselves to add types whenever we add/touch existing tests?

[Pradhvan]

@florimondmanca I see a lot of merged PR. I guess you should close this ? 😄

[florimondmanca]

I think there’s one small last step which would be to add tests/ to the mypy arguments in the check script. :-) Once this is enforced then yup, we can close this off.

[Pradhvan]

I am currently working on a different Good First Issue PR and when that is done. I would be happy to pick up what's remaining in this PR so we can close this one too. 😄