BUG: Some ExtensionArrays can return 0-d Elements

[ilan-gold]

Pandas version checks

• I have checked that this issue has not already been reported.

• I have confirmed this bug exists on the latest version of pandas.

• I have confirmed this bug exists on the main branch of pandas.

Reproducible Example

import pandas as pd


for arr in [pd.arrays.PeriodArray(pd.PeriodIndex(['2023-01-01','2023-01-02'], freq='D')), pd.Categorical(["a", "b"])]:
    subset = arr[(0, Ellipsis)]
    assert isinstance(subset, type(arr))
    assert subset.shape == ()

Issue Description

Given what is stated on https://pandas.pydata.org/docs/reference/api/pandas.api.extensions.ExtensionArray.html, I would expect this not to be possible at all.

Expected Behavior

The reason I care is that arrow arrays do not have a 0d version, which makes it tough to develop over all ExtensionArray classes:

pd.array([1, 2], dtype="int64[pyarrow]")[(0, Ellipsis)]

gives simply the number 1.

Installed Versions

INSTALLED VERSIONS

commit : 0691c5c
python : 3.11.11
python-bits : 64
OS : Darwin
OS-release : 24.1.0
Version : Darwin Kernel Version 24.1.0: Thu Oct 10 21:03:15 PDT 2024; root:xnu-11215.41.3~2/RELEASE_ARM64_T6000
machine : arm64
processor : arm
byteorder : little
LC_ALL : None
LANG : en_US.UTF-8
LOCALE : en_US.UTF-8

pandas : 2.2.3
numpy : 2.2.5
pytz : 2024.1
dateutil : 2.9.0.post0
pip : 25.0.1
Cython : None
sphinx : None
IPython : 8.32.0
adbc-driver-postgresql: None
adbc-driver-sqlite : None
bs4 : 4.12.3
blosc : None
bottleneck : 1.4.2
dataframe-api-compat : None
fastparquet : None
fsspec : 2025.3.2
html5lib : None
hypothesis : 6.131.6
gcsfs : None
jinja2 : 3.1.5
lxml.etree : 5.3.2
matplotlib : 3.10.1
numba : 0.61.2
numexpr : 2.10.2
odfpy : None
openpyxl : None
pandas_gbq : None
psycopg2 : None
pymysql : None
pyarrow : 19.0.1
pyreadstat : None
pytest : 8.3.5
python-calamine : None
pyxlsb : None
s3fs : None
scipy : 1.15.2
sqlalchemy : None
tables : None
tabulate : None
xarray : 2025.4.1.dev3+gd998eac1.d20250509
xlrd : None
xlsxwriter : None
zstandard : None
tzdata : 2025.1
qtpy : None
pyqt5 : None

[rhshadrach]

Thanks for the report!

The reason I care is that arrow arrays do not have a 0d version, which makes it tough to develop over all ExtensionArray classes

Can you give a little more detail here as to why it makes it tough? In particular, does

pd.array([1, 2], dtype="int64[pyarrow]")[0]

also cause difficulties?

[rhshadrach]

Ah - I see it now. It's that with just [0], you get the type of the scalar and not the ExtensionArray type.

[rhshadrach]

Thanks for the report. NumPy appears to have decided the user is asking to get back a 0-dim ndarray rather than a scalar.

print(type(np.array([1, 2, 3])[1, ...]))
# <class 'numpy.ndarray'>
print(type(np.array([1, 2, 3])[1]))
# <class 'numpy.int64'>

pandas would be consistent to agree that the user is asking for a 0-dim ExtensionArray here, and hence to raise as these are not supported.

@jbrockmendel - do you have any thoughts here?

[jbrockmendel]

raising would make sense, but im a bit concerned about the performance hit of adding that check (that __getitem__ method was optimized pretty hard IIRC). are there user-facing methods of reaching this?