CLN: Fix typing in pandas\tests\arrays\test_datetimelike.py (#28926)

[jjlkant]

• partially adresses Fix mypy errors for pandas\tests\* #28926
• passes black pandas
• passes git diff upstream/master -u -- "*.py" | flake8 --diff

[TomAugspurger]

Merging master should fix the CI failures.

[simonjayhawkins]

@jjlkant Thanks for the PR. looks good.

[simonjayhawkins]

@jjlkant lgtm @WillAyd

[WillAyd]

Can you put an alias for DatetimeLikeIndex in pandas._typing and use that here instead? I could see this being generally useful and we already have a DatetimeLikeScalar defined there, so would be consistent

@jbrockmendel in case he has other thoughts

[jjlkant]

Very fair comment! Went ahead with an alias like:

DatetimeLikeIndex = Union["DatetimeIndex", "PeriodIndex", "TimedeltaIndex"]

Tried the route of using TypeVar as used for DatetimeLikeScalar, but this sent me down a road with unbound type variables as a result of index_cls being a class variable. In the case of DatetimeLikeScalar, the type hint was used on a @property instead of a class variable, and without hinting inside of a comment. Do we want to go that way as well with this case?

[WillAyd]

Yea let's go with a TypeVar. Did that come when doing Type[DatetimeLikeIndex]? If so not sure off hand of solution so curious to see what you can find, but in any case the TypeVar is the right way to go here

[jjlkant]

Reading up on python/mypy#4590 as a result of looking into changing the class variables to @property, I think it is not possible to actually use a TypeVar? The @property looks as follows for example:

class SharedTests:
    @property
    def index_cls(self):
        # type: (SharedTests) -> Optional[Type[DatetimeLikeIndex]]
        return None

    ...

I'm currently not seeing a route that will lead to the TypeVar behaving as expected. Will dive deeper into the current usage of other TypeVar entries in pandas._typing like ArrayLike tomorrow, but any pointers are welcome in meantime!

[WillAyd]

What error would you get with TypeVar?

[jjlkant]

error: Incompatible return value type for every entry that's not matching:

pandas\tests\arrays\test_datetimelike.py:236: error: Incompatible return value type (got "Type[DatetimeIndex]", expected "Type[PeriodIndex]")
pandas\tests\arrays\test_datetimelike.py:236: error: Incompatible return value type (got "Type[DatetimeIndex]", expected "Type[TimedeltaIndex]")
pandas\tests\arrays\test_datetimelike.py:489: error: Incompatible return value type (got "Type[TimedeltaIndex]", expected "Type[DatetimeIndex]")
pandas\tests\arrays\test_datetimelike.py:489: error: Incompatible return value type (got "Type[TimedeltaIndex]", expected "Type[PeriodIndex]")
pandas\tests\arrays\test_datetimelike.py:602: error: Incompatible return value type (got "Type[PeriodIndex]", expected "Type[DatetimeIndex]")
pandas\tests\arrays\test_datetimelike.py:602: error: Incompatible return value type (got "Type[PeriodIndex]", expected "Type[TimedeltaIndex]")

With for example at L230-L238:

class TestDatetimeArray(SharedTests):
    array_cls = DatetimeArray

    @property
    def index_cls(self):
        # type: (TestDatetimeArray) -> Type[DatetimeLikeIndex]
        return pd.DatetimeIndex

    ...

[WillAyd]

Hmm OK I suppose these tests aren't fully generic so maybe the TypeVar doesn't work. I guess best to just revert to what you had, i.e. the Union def directly in the test module. Will leave the generic DatetimeLikeArray to a separate exercise as required

Sorry for back and forth but appreciate your thoroughness

[jjlkant]

Yeah, if the regular __init__(self) would be available, the TypeVar was no issue at all, however pytest doesn't take classes that actually use an __init__(self). Rolled back to the Union[] version for now

[WillAyd]

Thanks @jjlkant