BUG: indexing a MultiIndex-ed DataFrame (only) with a scalar on a level and a slice on another doesn't drop a level

[toobaz]

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 numpy as np

In [2]: import pandas as pd

In [3]: df = pd.DataFrame(np.arange(20).reshape((4, 5))).set_index([0, 1, 2])

In [4]: df.loc[(slice(None), 1,), :].index
Out[4]: 
MultiIndex([(0, 1, 2)],
           names=[0, 1, 2])

In [5]: df[3].loc[(slice(None), 1,)].index
Out[5]: 
MultiIndex([(0, 2)],
           names=[0, 2])

Issue Description

This is basically #12827 for DataFrames, since it was fixed (quite ironically, only) for Series.

Expected Behavior

Out[4] should be the same as Out[5].

Compare to

In [6]: df.loc[(0,), :].index
Out[6]: 
MultiIndex([(1, 2)],
           names=[1, 2])

... which correctly drops a level.

Installed Versions

INSTALLED VERSIONS

commit : 9c8bc3e
python : 3.11.2
python-bits : 64
OS : Linux
OS-release : 6.10.11+bpo-amd64
Version : #1 SMP PREEMPT_DYNAMIC Debian 6.10.11-1~bpo12+1 (2024-10-03)
machine : x86_64
processor :
byteorder : little
LC_ALL : None
LANG : it_IT.UTF-8
LOCALE : it_IT.UTF-8

pandas : 2.3.3
numpy : 2.3.4
pytz : 2025.2
dateutil : 2.9.0.post0
pip : 23.0.1
Cython : None
sphinx : None
IPython : 9.6.0
adbc-driver-postgresql: None
adbc-driver-sqlite : None
bs4 : 4.14.2
blosc : None
bottleneck : None
dataframe-api-compat : None
fastparquet : None
fsspec : None
html5lib : None
hypothesis : None
gcsfs : None
jinja2 : 3.1.6
lxml.etree : None
matplotlib : 3.10.7
numba : None
numexpr : None
odfpy : None
openpyxl : None
pandas_gbq : None
psycopg2 : None
pymysql : None
pyarrow : None
pyreadstat : None
pytest : None
python-calamine : None
pyxlsb : None
s3fs : None
scipy : None
sqlalchemy : None
tables : None
tabulate : None
xarray : None
xlrd : None
xlsxwriter : None
zstandard : None
tzdata : 2025.2
qtpy : None
pyqt5 : None

[rhshadrach]

Seems related to #46704

[toobaz]

Seems related to #46704

Indeed, apologies for missing that, it's closely related, except that the OP there complains about the "good" behavior, expecting the buggy behavior (likely led to this by the inconsistency with other cases).

Anyway, #46704 made me realize the behavior is not limited to first level, it happens (at least) whenever a slice is used:

In [1]: import pandas as pd

In [2]: import numpy as np

In [3]: df = pd.DataFrame(np.arange(20).reshape((4, 5))).set_index([0, 1, 2])

In [4]: df.loc[(0, slice(None),), :].index
Out[4]: 
MultiIndex([(0, 1, 2)],
           names=[0, 1, 2])

In [5]: df.loc[(0,), :].index
Out[5]: 
MultiIndex([(1, 2)],
           names=[1, 2])

In [6]: df[3].loc[(0,)].index
Out[6]: 
MultiIndex([(1, 2)],
           names=[1, 2])

... which is even more inconsistent than I thought.

(In general, I found it mindblowing that after all the refactoring, a MultiIndex behaves differently when attached to a DataFrame and to a Series.)

Anyway, I'm editing the title of this bug accordingly, and I think it better describes the issue, but it is true that the same problem is reported in this comment, so if @rhshadrach you prefer to close this rather than the other, fine.

[litang99]

I would like give it a try
take.

[toobaz]

I would like give it a try

Great!

My best guess is that ideally, the code that maps a (tuple of) key(s) to one or more positions and the code that determines what will the new index be should all be MultiIndex or _LocationIndexer (not Series or DataFrame) code.