Nullable dtypes for Series.__new__ and astype

[cmp0xff]

At the moment, nullable Pandas dtypes in Series are recognised as their non-nullable counterparts.

Can we make the stubs revealing the distinctions?

from typing import assert_type
import pandas as pd

assert_type(pd.Series([1], dtype="Int64"), "pd.Series[pd.Int64Dtype]")  # recognised as pd.Series[int]
assert_type(pd.Series([True], dtype="boolean"), "pd.Series[pd.BooleanDtype]")  # recognised as pd.Series[bool]

assert_type(pd.Series([1]).astype("Int64"), "pd.Series[pd.Int64Dtype]")  # recognised as pd.Series[int]
assert_type(pd.Series([True]).astype("boolean"), "pd.Series[pd.BooleanDtype]")  # recognised as pd.Series[bool]

[Dr-Irv]

At the moment, nullable Pandas dtypes in Series are recognised as their non-nullable counterparts.

Can we make the stubs revealing the distinctions?

So I am sure we could recognize this in the Series.__new__() and Series.astype() methods. But, if we did, then we'd have to do a lot of work to make sure methods that have Series[int] parameters or return types use Series[Int] instead. Also, care is required when handling arithmetic between nullable and non-nullable types.

[cmp0xff]

Currently, S1 in Series[S1] does not agree with the Series.dtype property. For example, for Series[int], the dtype is actuallly an instance of numpy.dtypes.Int64DType, but we are using Python builtin int to indicate. This has several benefits, among others simplicity. The underlying element type is numpy.int64.

In the mean time, when dtype is pandas.core.arrays.integer.Int64Dtype, the underlying element type is Python builtin int (or pandas.NA).

We need to make a plan here, naively go for pd.Series[pd.Int64Dtype] may not be ideal.