Errors with map/apply with pyright "basic"

[PhanAlexandre]

Hi,
I have some errors with pyright and pandas. I have ask the question there and they send me here.
microsoft/pyright#8985

Its seems the installing pandas-stubs resolve some of those errors (all but .apply(Path) on one computer but for the other its still showing errors on all line.
Here is some line of code that show this errors

import pandas as pd
from pathlib import Path

df = pd.DataFrame({"a":["a","a","b","b","c"]})

df['b'] = df['a'].map(Path) #No overloads for "map" match the provided arguments
df['c'] = df['b'].apply(Path) #No overloads for "map" match the provided arguments
df['b'] = df['a'].map(list) #No overloads for "map" match the provided arguments
df['c'] = df['b'].apply(list)
df['b'] = df['a'].map(set) #No overloads for "map" match the provided arguments
df['c'] = df['b'].apply(set)

Is it normal ?

I am on pandas 2.2.2
pylance v2024.9.1
on the computer with less errors pandas-stubs-2.1.4.231227
and the other pandas-stubs-2.2.2.240909

[twoertwein]

Pandas-stubs has made the trade-off of allowing only a subset of common types on Series to provide better operator/method support for those types. Whether this trade-off was a good decision is a different question :). Unfortunately, the Python static type system cannot model operators of container types properly (see for example microsoft/pyright#6549), except if you enumerate all of them which is not maintainable.

I think it might be possible to at least allow the call but return Series[Unknown/Any].

Similar to #1001

[PhanAlexandre]

Thanks for the answer. However, its still strange that the newest version of pandas-stud show more errors than pandas-stubs-2.1.4.231227 (which show only 'df['c'] = df['a'].apply(Path)' as an errors)
And that map and apply don't show all the same errors.

I fond a workaround by creating a function likes, def x: return path with typing ->Any.

[twoertwein]

Thank you for pointing out that this behavior changed! I will later today/tomorrow look through recent PRs that might have changed that behavior to see whether that was an (un)intended change.

[twoertwein]

apply and map used to have some un-annotated parameters so you didn't see errors in older versions.

I think we should add a fallback overload for Series.apply and Series.map, that accepts Any and returns Series | DataFrame (or Any). This would also fix #1001 @Dr-Irv

For issues in pandas, please open an issue there.