API (string): str.center with pyarrow-backed string dtype

[jbrockmendel]

• closes API: behaviour for the "str.center()" string method for the pyarrow-backed string dtype #54807 (Replace xxxx with the GitHub issue number)
• Tests added and passed if fixing a bug or adding a new feature
• All code checks passed.
• Added type annotations to new arguments/methods/functions.
• Added an entry in the latest doc/source/whatsnew/vX.X.X.rst file if fixing a bug or adding a new feature.

Two bad options for older-pyarrow: different behavior or object fallback. I guessed that @mroeschke would prefer different behavior here, would be happy to be proved wrong.

[jorisvandenbossche]

Two bad options for older-pyarrow: different behavior or object fallback. I guessed that @mroeschke would prefer different behavior here, would be happy to be proved wrong.

I won't argue for ArrowDtype if there is a preference to not do object fallback there, but for StringDtype, we definitely want to fallback to object dtype, I think (as I proposed in #54807 (comment)). Otherwise this gives a breaking change for people with older pyarrow, and inconsistent behaviour depending on the pyarrow version you have installed. While we can very easily avoid this breaking change, so then we should just do that IMO

[jbrockmendel]

Using the lean_left_on_odd_padding=False breaks test_string_array:

import numpy as np
import pandas as pd

data = ["a", "bb", np.nan, "ccc"]
ser = pd.Series(data, dtype=pd.StringDtype(storage="pyarrow"))
alt = pd.Series(data, dtype=object)

res = ser.str.center(10)
exp = alt.str.center(10).astype(ser.dtype)

>>> res[0]
'     a    '
>>> exp[0]
'    a     '

The first explanation that comes to mind is that I misunderstood something and should be passing lean_left_on_odd_padding=True, but that in turn breaks test_center_ljust_rjust_fillchar:

import pandas as pd

ser = pd.Series(["a", "bb", "cccc", "ddddd", "eeeeee"], dtype=pd.StringDtype(storage="pyarrow"))
res = ser.str.center(5, fillchar="X")
exp = ser.astype(object).str.center(5, fillchar="X").astype(ser.dtype)

>>> res[2]
'ccccX'
>>> exp[2]
'Xcccc'

[jorisvandenbossche]

Looks good! Added a few small comments

[jorisvandenbossche]

Move to 2.3.0 file?

[jbrockmendel]

Done

[jorisvandenbossche]

Could also do obj_arr.array._str_pad(..) and then don't need this explicit construction?

[jorisvandenbossche]

Not sure how significant this is, but one small performance disadvantage of this instead of type(self)(pa.array(result)) is that _from_sequence has an (in this case) unnecessary call to lib.ensure_string_array.

The problem is that we don't know here which pyarrow type to use? But then could do something like pa.array(result, type=self.dtype.pyarrow_dtype) to preserve the type?

[jbrockmendel]

im not inclined to worry about this too much since its in a branch that will be going away eventually