BUG: (pandas 3.0rc0) `str.cat` is not retaining `index` dtype

[galipremsagar]

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

In [1]: import pandas as pd

In [2]: pd.__version__
Out[2]: '3.0.0rc0'

In [3]: s = pd.Series(["a", "b", "c"])

In [4]: s.dtype
Out[4]: <StringDtype(na_value=nan)>

In [5]: s = pd.Series(["a", "b", "c"], index=['1', '2', '3'])

In [6]: s.index
Out[6]: Index(['1', '2', '3'], dtype='str')

In [7]: s.index.dtype
Out[7]: <StringDtype(na_value=nan)>

In [9]: res = s.str.cat(["A", "B", "C"], sep=",", na_rep="-")

In [10]: res
Out[10]: 
1    a,A
2    b,B
3    c,C
dtype: str

In [11]: res.dtype
Out[11]: <StringDtype(na_value=nan)>

In [12]: res.index
Out[12]: Index(['1', '2', '3'], dtype='str')

In [13]: res.index.dtype
Out[13]: <StringDtype(na_value=nan)>

In [14]: res = s.str.cat(pd.Series(["A", "B", "C"]), sep=",", na_rep="-")

In [15]: res
Out[15]: 
1    a,-
2    b,-
3    c,-
dtype: str

In [16]: res.index
Out[16]: Index(['1', '2', '3'], dtype='object')


In [24]: res.index.dtype
Out[24]: dtype('O')

Issue Description

I was expecting res.index to have same dtype instead of str vs object. Is this expected?

Expected Behavior

In [16]: res.index
Out[16]: Index(['1', '2', '3'], dtype='str')

Installed Versions

Details

In [21]: pd.show_versions()

INSTALLED VERSIONS

commit : 1a3230d
python : 3.13.11
python-bits : 64
OS : Linux
OS-release : 6.14.0-36-generic
Version : #36~24.04.1-Ubuntu SMP PREEMPT_DYNAMIC Wed Oct 15 15:45:17 UTC 2
machine : x86_64
processor : x86_64
byteorder : little
LC_ALL : None
LANG : en_US.UTF-8
LOCALE : en_US.UTF-8

pandas : 3.0.0rc0
numpy : 2.2.6
dateutil : 2.9.0.post0
pip : 25.3
Cython : 3.1.7
sphinx : 8.2.3
IPython : 9.8.0
adbc-driver-postgresql: None
adbc-driver-sqlite : None
bs4 : 4.14.3
bottleneck : None
fastparquet : None
fsspec : 2025.12.0
html5lib : None
hypothesis : 6.148.7
gcsfs : None
jinja2 : 3.1.6
lxml.etree : None
matplotlib : None
numba : 0.61.2
numexpr : None
odfpy : None
openpyxl : 3.1.5
psycopg2 : None
pymysql : None
pyarrow : 21.0.0
pyiceberg : None
pyreadstat : None
pytest : 9.0.2
python-calamine : None
pytz : 2025.2
pyxlsb : None
s3fs : 2025.12.0
scipy : 1.16.3
sqlalchemy : 2.0.45
tables : None
tabulate : 0.9.0
xarray : None
xlrd : None
xlsxwriter : None
zstandard : 0.25.0
qtpy : None
pyqt5 : None

[galipremsagar]

cc: @mroeschke

[jorisvandenbossche]

This seems to be specifically an issue in align when the other operand is a DataFrame instead of a series

>>> ser1 = pd.Series(index=["a", "b"])
>>> ser2 = pd.Series(index=range(3))
>>> res1, res2 = ser1.align(ser2, join="left")
>>> res1.index
Index(['a', 'b'], dtype='str')
>>> res1, res2 = ser1.align(pd.DataFrame({0: ser2}), join="left")
>>> res1.index
Index(['a', 'b'], dtype='object')

[jorisvandenbossche]

And the align behaviour boils down to:

>>> idx_str = pd.Index(["a", "b"], dtype=str)
>>> idx_obj = pd.Index(["a", "b"], dtype=object)
>>> 
>>> idx_str.join(idx_obj, how="left")
Index(['a', 'b'], dtype='object')

[jorisvandenbossche]

So the question here is whether this is actually the expected behaviour. In doing res = s.str.cat(pd.Series(["A", "B", "C"]), sep=",", na_rep="-"), you are passing a Series with an integer range index to cat(). At that point we have an operation between an object with str index vs integer range index. The common dtype for those is object dtype. In that sense, this is the expected behaviour.

However, since this is doing a "left" index join, one could expect that it is doing a bit more work to preserve the left index' dtype if there is no overlap. At the moment we don't special case such a left join without overlap, though. For example also an integer index gets cast to object:

>>> pd.Index([1, 2, 3])
Index([1, 2, 3], dtype='int64')
>>> pd.Index([1, 2, 3]).join(pd.Index(["a", "b"]))
Index([1, 2, 3], dtype='object')

[rhshadrach]

Closely related: #62949