Various `Index` operators missing

[adamchainz]

Describe the bug
I discovered that Index.__rpow__ is missing when trying to lint my code.

To Reproduce

import pandas

df = pandas.DataFrame({"nums": [1, 2]})
print(2**df.index)

Runs:

$ python example.py
0    2
1    4
Name: nums, dtype: int64

But Mypy not happy:

$ pre-commit run mypy --file example.py
mypy.....................................................................Failed
- hook id: mypy
- exit code: 1

example.py:4: error: No overload variant of "__pow__" of "int" matches argument type "Index"  [operator]
example.py:4: note: Possible overload variants:
example.py:4: note:     def __pow__(self, Literal[0]) -> Literal[1]
example.py:4: note:     def __pow__(self, Literal[0], None) -> Literal[1]
example.py:4: note:     def __pow__(self, Literal[1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25], None = ...) -> int
example.py:4: note:     def __pow__(self, Literal[-1, -2, -3, -4, -5, -6, -7, -8, -9, -10, -11, -12, -13, -14, -15, -16, -17, -18, -19, -20], None = ...) -> float
example.py:4: note:     def __pow__(self, int, None = ...) -> Any
example.py:4: note:     def __pow__(self, int, Literal[0]) -> NoReturn
example.py:4: note:     def __pow__(self, int, int) -> int
Found 1 error in 1 file (checked 1 source file)

It seems that various dunder funcs on Index are missing.

Please complete the following information:

• OS: macOS
• OS Version 12
• python version: 3.10
• version of type checker: 0.982
• version of installed pandas-stubs: 1.5.1.221024

Additional context
n/a

[adamchainz]

In the implementation, the methods are inherited from OpsMixin, through IndexOpsMixin.

But in the stubs, IndexOpsMixin does not inherit from OpsMixin. I guess this is because OpsMixin is hardcoded right now to only return DataFrames.

[Dr-Irv]

Since we want to be more specific about what types are returned for these operators when applied to Series, DataFrame and Index, probably the best solution is to define all of the relevant methods for Index in typing, as is done for Series. Some of them, e.g., __add__(), __sub__() are already defined in core/indexes/base.pyi, so we just need to add the missing ones.

PR with tests welcome!

[adamchainz]

I think you could use Self in OpsMixin, if using such ops always returns the same type as the index/dataframe/series?

Like:

def __pow__(self, other: Any) -> Self: ...

Or one could be specific about the other types that are supported:

def __pow__(self, other: int | float | Decimal | ...) -> Self: ...

Not sure what types pandas allows here though - I don't really use it, I am just helping a project that uses it to roll out type hints.

WDYT?

[Dr-Irv]

Not sure what types pandas allows here though - I don't really use it, I am just helping a project that uses it to roll out type hints.

WDYT?

In Series, we try to do some better type checking based on the type of the Series, e.g.:

    @overload
    def __add__(self, other: TimestampSeries) -> TimestampSeries: ...
    @overload
    def __add__(self, other: DatetimeIndex) -> TimestampSeries: ...
    @overload
    def __add__(self, other: Timestamp) -> TimestampSeries: ...
    @overload
    def __add__(
        self, other: num | _str | Timedelta | _ListLike | Series[S1]
    ) -> Series: ...

So for Index (and things like DatetimeIndex), we also want to be more restrictive on the allowed types for the operators.

So I think the best way to go is to just copy the various operators over to Index, or create an internal hidden type equivalent to OpsMixin that would be used by Index and DataFrame, but not by Series

[ramvikrams]

@Dr-Irv Sir then pow,rmul,mul,rdiv etc. has to be added