BUG: Possible bad api side effect of replacing FrozenList with tuple

[dalejung]

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

df = pd.DataFrame({'name': ['dale', 'frank'], 'age': [10, 20]})
df = df.set_index('name')

index_names = df.index.names

df = df.reset_index()
df.set_index(index_names)

# KeyError: "None of [('name',)] are in the columns"

Issue Description

This might not be considered a bug, but I would expect there to be symmetry in df.set_index and df.index.names. Previously when index.names was a FrozenList, this would work. However a tuple has different semantics than a list throughout pandas.

Expected Behavior

I would expect Index.names to work for set_index.

Installed Versions

INSTALLED VERSIONS

commit : f01ae20
python : 3.11.7.final.0
python-bits : 64
OS : Linux
OS-release : 6.7.4-arch1-1
Version : #1 SMP PREEMPT_DYNAMIC Mon, 05 Feb 2024 22:07:49 +0000
machine : x86_64
processor :
byteorder : little
LC_ALL : None
LANG : en_US.UTF-8
LOCALE : en_US.UTF-8

pandas : 3.0.0.dev0+431.gf01ae209da
numpy : 1.24.4
pytz : 2023.3
dateutil : 2.8.2
setuptools : 69.1.1
pip : 24.0
Cython : 3.0.8
pytest : 7.4.0
hypothesis : 6.98.11
sphinx : 6.2.1
blosc : None
feather : None
xlsxwriter : 3.1.0
lxml.etree : 4.9.2
html5lib : 1.1
pymysql : 1.0.3
psycopg2 : 2.9.6
jinja2 : 3.1.2
IPython : 8.22.1
pandas_datareader : None
adbc-driver-postgresql: None
adbc-driver-sqlite : None
bs4 : 4.12.2
bottleneck : 1.3.7
fastparquet : 2023.4.0
fsspec : 2023.5.0
gcsfs : 2023.5.0
matplotlib : 3.7.1
numba : 0.57.0
numexpr : 2.8.4
odfpy : None
openpyxl : 3.1.2
pyarrow : 16.0.0.dev175+g2d44818d0
pyreadstat : 1.2.1
python-calamine : None
pyxlsb : 1.0.10
s3fs : 2023.5.0
scipy : 1.10.1
sqlalchemy : 2.0.14
tables : 3.8.0
tabulate : 0.9.0
xarray : 2023.5.0
xlrd : 2.0.1
zstandard : 0.21.0
tzdata : 2023.3
qtpy : 2.4.0
pyqt5 : None

[rhshadrach]

cc @mroeschke @jorisvandenbossche

[dalejung]

https://github.com/search?q=%22index.names+%3D%3D+%5BNone%5D%22&type=code

fwiw checking for df.index.names == [None] happens pretty frequently. Variants of this type of error is how I ran into the change originally in private repo. Just ran into it with hvplot. The fixes to these are obviously simple, you just wrap index.names in a list. But unsure what the upside of the change is.

[mroeschke]

Maybe we could return a list here instead of a tuple since FrozenList was originally a list subclass.

Since codes/names/levels that returned this before are not writable, maybe we don't need the mutability anymore

[jorisvandenbossche]

maybe we don't need the mutability anymore

What do you mean exactly with this?

[jorisvandenbossche]

Consider something like this

df.index.names[0] = "a"

Currently such code fails, because the FrozenList is not mutable (and also with the tuple it fails). But if this would return a list instead, that would actually "work", in the sense of not raising an error anymore, but I suppose it would not actually update the names?

[mroeschke]

What do you mean exactly with this?

Sorry I meant immutability.

Consider something like this

It appears for MultiIndex that codes/levels/names are backed by private _codes/_level/_names attributes. If we have the public attributes return lists and the private attributes use tuples we can still achieve "immutability" as before. Here's an example from a branch I have that implements this

In [5]: df = pd.DataFrame([1], index=pd.MultiIndex(codes=[[0]], levels=[["a"]], names=[0]))

In [6]: df
Out[6]: 
   0
0   
a  1

In [9]: df.index.names[0] = "a"

In [10]: df.index
Out[10]: 
MultiIndex([('a',)],
           names=[0])

[jorisvandenbossche]

Yes, but so that has the behaviour that I mentioned above of appearing to work (because it no longer raises an error about being immutable), while not actually working.

I personally think it is perfectly fine that this doesn't work (that's how it always has been, and there are other APIs to change the name), but IMO it would be a UX regression if it no longer fails loudly.

[WillAyd]

Maybe we also just restore the FrozenList? Not a strong preference but I think the pattern the OP established is pretty reasonable. I would really never expect df.set_index(index_names) to act like a tuple indexer