ENH: Multiplication of pd.ArrowDtype(pa.string()) and int should support the integral type being a series

[rohanjain101]

Feature Type

• Adding new functionality to pandas

• Changing existing functionality in pandas

• Removing existing functionality in pandas

Problem Description

Follow up of #51970

The current implementation of multiplying pd.ArrowDtype(pa.string()) by an int, requires the int to be a scalar value. It would be nice to allow the int to be a series, that would be similar to pd.StringDtype() as well.

Feature Description

>>> binary = pd.Series(["abc", "defg"], dtype=ArrowDtype(pa.string()))
>>> repeat = pd.Series([2, -2], dtype="int64[pyarrow]")
>>> binary * repeat
0    abcabc
1
dtype: string[pyarrow]
>>>

pd.StringDtype support this:

>>> binary = pd.Series(["abc", "defg"], dtype=pd.StringDtype())
>>> repeat = pd.Series([2, -2], dtype="int64")
>>> binary * repeat
0    abcabc
1
dtype: string
>>>

Alternative Solutions

NA

Additional Context

No response

[rohanjain101]

Linked PR should add this support, if this change is not desired, I can abandon that PR, but it seemed useful, especially since pd.StringDtype() allows it. Also, I believe this implementation may perform better, since it uses pyarrow compute functions without falling back to pure python.

[rohanjain101]

strings = Series([''.join(random.choices(string.ascii_uppercase + string.digits, k=20)) for _ in range(150_000)], dtype=ArrowDtype(pa.string()))


def func(strings):
    a = strings * 5

pd_time = min(
    timeit.repeat(
        stmt="func(strings)",
        repeat=3,
        number=5,
        globals={"func": func, "strings": strings},
    )
)
print(pd_time)

Used the script above to compare performance, on my machine, time on main is:

0.04619359999196604

Time in proposed PR is

0.012037000036798418

About 4x improvement.