mypy compliance

[Zeitsperre]

Pull Request Checklist:

• This PR addresses an already opened issue (for bug fixes / features)
  • This PR addresses issues raised in Adopt mypy and stronger enforcement of pylint #1719
• Tests for the changes have been added (for bug fixes / features)
  • (If applicable) Documentation has been added / updated (for bug fixes / features)
• CHANGES.rst has been updated (with summary of main changes)
  • Link to issue (:issue:number) and pull request (:pull:number) has been added

What kind of change does this PR introduce?

• Addresses several issues (around 250/1250 errors) associated with (re-)casting of variables and output typing.
• The Indicator.__getitem__ method has been retired (to reduce instances of class methods returning Any)
• PEP8 rule N802 is now enforced (with noqa exceptions, mostly for functions based on sfcWind variable).
• The wrapped_partial function wrapper has been removed.

Does this PR introduce a breaking change?

Yes. The Indicator __getitem__ method has been removed. This has downstream impacts on finch and possibly other projects. Additionally, the obsolete wrapped_partial function (was previously used in Indicator construction) has also been removed.

Other information:

@aulemahal

We currently use the Quantified type alias in almost all the threshold-based indicators to specify that this variable should conform to something resembling a string with units (str | pint.Quantity), but this alias also includes xarray.DataArray. Does this mean that we can pass (or would like to enable passing) DataArrays of grids denoting thresholds? If not, we may need to create a new, more constrained custom variable (str2pint often takes variables of this type but does not support pint.Quantity or xarray.DataArray). Solving this issue alone will address +250 errors.

[Zeitsperre]

There seems to be a handful of trip-falls that we keep running into:

1. Many of the decorators that we employ do not constrain the types of variables passed in/out of them; This makes it so that regardless of the typing identified by a decorated function, the end results will always return Any. This can be fixed by using -> TypeVar[...] signatures.
2. Mentioned above, Quantified is used too often in places where it doesn't actually work (some functions demand str-like variables, but claim to support Quantified, which includes DataArray). This could become a breaking change but would address many errors.

Changes as of this comment have us hovering at around 1050 errors total.

[Zeitsperre]

Additional comment:

Due to the way that numpy is typed, any time that we pass xarray.DataArray to a numpy.ufunc-like operation (e.g. np.logical_and(da1, da2)) we need to specify that the output is still a xarray.DataArray (using: output: xarray.DataArray; expected output is an ndarray otherwise).

[aulemahal]

This looks good, but there are a lot of cases where it seems an annotation shouldn't be necessary ? I've commented on a few, but not all.

[Zeitsperre]

This is just missing the updates to the changelog about breaking changes, but I think we're better off now. Lots of small improvements.

[aulemahal]

Thanks @Zeitsperre! This was a lot of work.

I have a last conservative comment, but this looks like a very good start for type-checked xclim.

[review-notebook-app]

Check out this pull request on 

See visual diffs & provide feedback on Jupyter Notebooks.

---

Powered by ReviewNB