BUG: Remove complex floor divide

[ganesh-k13]

Changes:

1. Removed floor_divide for all complex types
2. Modified UFuncs floor_divde TC to check for TypeError. Added same for divmod and remainder.
3. Modified test_core to check TypeError raised for unsupported types.
4. Removed Typing Tests

TODO:

• Add release note (Need PR number for this)

closes: #13236
finishes: #13245 (work done by @kaivu1999)

[ganesh-k13]

I really need to change my .vimrc one day

[ganesh-k13]

@kaivu1999 didn't cross my mind to check with you before raising this. If you see this before the review for this code is done, please feel free to re-open your PR and add the extra changes I have made to fix the tests.

[BvB93]

There are a couple of places in the stub files that still contain references to complex floordivision:

numpy/numpy/__init__.pyi

Lines 2310 to 2311 in 9fcc132

	@overload
	def __floordiv__(self: _ArrayComplex_co, other: _ArrayLikeComplex_co) -> NDArray[complexfloating[Any, Any]]: ... # type: ignore[misc]

numpy/numpy/__init__.pyi

Lines 2338 to 2339 in 9fcc132

	@overload
	def __rfloordiv__(self: _ArrayComplex_co, other: _ArrayLikeComplex_co) -> NDArray[complexfloating[Any, Any]]: ... # type: ignore[misc]

numpy/numpy/__init__.pyi

Lines 3258 to 3259 in 9fcc132

	__floordiv__: _ComplexOp[_NBit1]
	__rfloordiv__: _ComplexOp[_NBit1]

The same holds for the typing tests in reveal/arithmetic.py and pass/arithmetic.py.

[ganesh-k13]

Thanks @BvB93, I'll remove those as well. I wanted to understand how and where the three references in numpy/numpy/__init__.pyi are being used. When I tried removing them for unsigned int, I was still able to go forward with the division. So where are they used?

[BvB93]

When I tried removing them for unsigned int, I was still able to go forward with the division. So where are they used?

Yeah, that makes sense actually. Each of these _ArrayLike<x>_co variables represents an union of all types that can be cast into <x>, according to the rules of casting="same_kind". So removing the overload for unsigned integers means that, next time, they'll be handled by the overload for signed integers. Remove that overload as well and they'll be handled by the float overload, etc..

Complex is at the end of this tree though (as it cannot be safely cast into any other type), so removing it means that complex operations are no longer allowed.

numpy/numpy/__init__.pyi

Lines 2016 to 2021 in 9fcc132

	@overload
	def __matmul__(self: _ArrayUInt_co, other: _ArrayLikeUInt_co) -> NDArray[unsignedinteger[Any]]: ... # type: ignore[misc]
	@overload
	def __matmul__(self: _ArrayInt_co, other: _ArrayLikeInt_co) -> NDArray[signedinteger[Any]]: ... # type: ignore[misc]
	@overload
	def __matmul__(self: _ArrayFloat_co, other: _ArrayLikeFloat_co) -> NDArray[floating[Any]]: ... # type: ignore[misc]

[ganesh-k13]

Thanks! Always wondered where the casting order is taken care for arrays.

[ganesh-k13]

Hey folks, any changes needed here?

[mattip]

Do we know if this will have implications for scipy/astropy/pandas?

[ganesh-k13]

Will building them with this branch of NumPy help? Will be happy to do so.

[charris]

Thanks @ganesh-k13 .Lets get this in for testing at least.