BUG: Union of multi index with EA types can lose EA dtype

[ssche]

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

>>> from cyberpandas import IPArray
>>> import pandas as pd
>>> 
>>> df1 = pd.DataFrame({
...     'address': IPArray(['192.168.1.1', '192.168.1.10']),
...     'date': ['2022-01-01', '2022-01-02'],
...     'a': [1, 2]
... })
>>> df1 = df1.set_index(['address', 'date'])
>>> 
>>> 
>>> df2 = pd.DataFrame({
...     'address': IPArray(['192.168.1.1', '192.168.1.10']),
...     'date': pd.to_datetime(['2022-01-01', '2022-01-02']),
...     'a': [1, 2]
... })
>>> df2 = df2.set_index(['address', 'date'])
>>> 
>>> df1.index.dtypes
address        ip
date       object
dtype: object
>>> 
>>> df2.index.dtypes
address                ip
date       datetime64[ns]
dtype: object
>>> 
>>> df1.index.union(df2.index).dtypes
address            object   # <-- should be type "ip", not "object"
date       datetime64[ns]
dtype: object

Issue Description

The ExtensionType can get lost when two MultiIndex objects are combined by .union() (which becomes a problem when using df.combine_first(...) which relies on index.union(...)).

The problem occurs when both MIs share the same EA series, but the other series (assuming only 2-series MI) has a different type. In that case, the former EA dimension of the joined MI is losing its EA dtype.

Expected Behavior

EA type can be maintained after index.union(...).

Installed Versions

INSTALLED VERSIONS

commit : e8093ba
python : 3.8.13.final.0
python-bits : 64
OS : Darwin
OS-release : 21.5.0
Version : Darwin Kernel Version 21.5.0: Tue Apr 26 21:08:22 PDT 2022; root:xnu-8020.121.3~4/RELEASE_X86_64
machine : x86_64
processor : i386
byteorder : little
LC_ALL : None
LANG : en_AU.UTF-8
LOCALE : en_AU.UTF-8

pandas : 1.4.3
numpy : 1.23.1
pytz : 2022.1
dateutil : 2.8.2
setuptools : 62.3.2
pip : 22.1.1
Cython : None
pytest : None
hypothesis : None
sphinx : None
blosc : None
feather : None
xlsxwriter : None
lxml.etree : None
html5lib : None
pymysql : None
psycopg2 : None
jinja2 : None
IPython : None
pandas_datareader: None
bs4 : None
bottleneck : None
brotli : None
fastparquet : None
fsspec : None
gcsfs : None
markupsafe : None
matplotlib : None
numba : None
numexpr : None
odfpy : None
openpyxl : None
pandas_gbq : None
pyarrow : None
pyreadstat : None
pyxlsb : None
s3fs : None
scipy : None
snappy : None
sqlalchemy : None
tables : None
tabulate : None
xarray : None
xlrd : None
xlwt : None
zstandard : None

[phofl]

Hi, thanks for your report. As a note: This also happens for our own extension arrays.

[ssche]

This also happens for our own extension arrays.

Do you have an example?

[phofl]

Simply replace your Arrays with Series([1, 2], dtype="Int64")

[ssche]

Hmm... not quite. I get int64 instead of Int64, but not object.

>>> df1 = pd.DataFrame({
...     'address': pd.Series([1, 2], dtype="Int64"),
...     'date': ['2022-01-01', '2022-01-02'],
...     'a': [1, 2]
... })
>>> df1 = df1.set_index(['address', 'date'])
>>> 
>>> 
>>> df2 = pd.DataFrame({
...     'address': pd.Series([1, 2], dtype="Int64"),
...     'date': pd.to_datetime(['2022-01-01', '2022-01-02']),
...     'a': [1, 2]
... })
>>> df2 = df2.set_index(['address', 'date'])
>>> 
>>> df1.index.dtypes
address     Int64
date       object
dtype: object
>>> 
>>> df2.index.dtypes
address             Int64
date       datetime64[ns]
dtype: object
>>> 
>>> df1.index.union(df2.index).dtypes
address             int64
date       datetime64[ns]
dtype: object

[phofl]

Sorry, maybe my comment was not clear enough. I meant: This is buggy for our own extension arrays too.

[ssche]

>>> import pandas as pd
>>> 
>>> 
>>> df1 = pd.DataFrame({
...     'address': pd.Series([1, 2], dtype="Int64"),
...     'date': ['2022-01-01', '2022-01-02'],
...     'a': [1, 2]
... })
>>> df1 = df1.set_index(['address', 'date'])
>>> 
>>> df2 = pd.DataFrame({
...     'address': pd.Series([1, 2], dtype="Int64"),
...     'date': pd.to_datetime(['2022-01-01', '2022-01-02']),
...     'a': [1, 2]
... })
>>> df2 = df2.set_index(['address', 'date'])
>>> 
>>> df1.index.dtypes
address     Int64
date       object
dtype: object
>>> 
>>> df2.index.dtypes
address             Int64
date       datetime64[ns]
dtype: object
>>> 
>>> df1.index.union(df2.index).dtypes
<stdin>:1: RuntimeWarning: The values in the array are unorderable. Pass `sort=False` to suppress this warning.
address     Int64
date       object
dtype: object
>>> 
>>> pd.__version__
'1.6.0.dev0+328.g7aa391ead'

This seems to be resolved now. I might throw in a test case to peg down the behaviour.

[phofl]

Thanks for following up. This is fixed and tested :)