error: Argument 1 to "mul" of "Series" has incompatible type "timedelta64" [arg-type]

[randolf-scholz]

Describe the bug

Series.mul has incompatible signature with Series.__mul__

To Reproduce

import pandas as pd
import numpy as np

s = pd.Series([1, 2, 3])
x = s * np.timedelta64(1, "s")  # ✔
y = s.mul(np.timedelta64(1, "s") )  # ✘ [arg-type]

Please complete the following information:

• OS: Ubuntu 22.04
• python version 3.10
• mypy 1.4.1
• pandas 2.0.3
• pandas-stubs 2.0.2.230605

[Dr-Irv]

Need to coordinate the definitions of Series.mul() and Series.__mul__(), as well as tthe other operators (e.g., add() and __add__(), etc.

[mutricyl]

Any kind of multiplication of a series with Timedeltas induce a TimedeltaSeries. I have updated mul, __mul__, rmul and __rmul__ to be coherent.

I have quickly tested + / // & | and they all raise an error when dealing with Series and Timedeltas. So I wonder if I should add a bunch of Never or keep as is managing only *. @Dr-Irv could you advise ?

[Dr-Irv]

I have quickly tested + / // & | and they all raise an error when dealing with Series and Timedeltas. So I wonder if I should add a bunch of Never or keep as is managing only *. @Dr-Irv could you advise ?

Let's leave the scope as is. The challenge here is that you could get a Series that contains timedeltas, but from a typing perspective, it is Series[Any], but addition is allowed at runtime.

So make a PR that just aligns the arguments for things like SOMEOP and __SOMEOP__ , where SOMEOP is mul, add, etc., with appropriate tests. I'm pretty sure that the arguments we use for the __SOMEOP__ versions are well tested so if you just copy those declarations over to SOMEOP and add tests for the SOMEOP versions, you'll be in good shape.

Separately, if you want to experiment with using Never to prevent operations, you could do a separate PR, and as long as none of the existing tests fail, it's something we could consider.

Thanks for your contributions!

[mutricyl]

Series[int] or Series[float] works fine of course for * but fails for +

>>> s2 = pd.Series([1, 2, 3])                         
>>> s2 * np.timedelta64(1, "s")     
0   0 days 00:00:01
1   0 days 00:00:02
2   0 days 00:00:03
dtype: timedelta64[s]
>>> s2 + np.timedelta64(1, "s")
(...)
TypeError: Concatenation operation is not implemented for NumPy arrays, use np.concatenate() instead. Please do not rely on this error; it may not be given on all Python implementations.

however with Series[object] (as we mix int and timedelta):

>>> s = pd.Series([1, 2, 3, np.timedelta64(1, "s")]) 
>>> s
0            1
1            2
2            3
3    1 seconds
>>> s * np.timedelta64(1, "s")
(...)
numpy.core._exceptions._UFuncBinaryResolutionError: ufunc 'multiply' cannot use operands with types dtype('O') and dtype('<m8[s]')
>>> s + np.timedelta64(1, "s") 
(...)
TypeError: unsupported operand type(s) for +: 'Timedelta' and 'int'
>>> s[0:2] * np.timedelta64(1, "s")
(...)
numpy.core._exceptions._UFuncBinaryResolutionError: ufunc 'multiply' cannot use operands with types dtype('O') and dtype('<m8[s]')

+ and * are failling. (you can not multiply to timedeltas anyway)

I can not find any case where Series + Timedeltas are allowed. So I'll go for Never.

I will PR what I have done for * so far and give a try with Never all the other operators in a separate branch.