error: No overload variant of "DataFrame" matches argument types "Sequence[int]" [call-overload]

[randolf-scholz]

import pandas as pd
from collections.abc import Sequence

index: Sequence[int] = [3, 2, 7]
columns = ["a", "b", "c"]
df = pd.DataFrame(index=index, columns=columns)  # ✘ error: [call-overload]
print(df)

If the annotation Sequence[int] is removed, the file parses.

Please complete the following information:

• Ubuntu 22.04.1
• Python 3.10.6
• mypy 0.982
• pandas 1.5.0
• pandas-stubs 1.5.0.221012

[Dr-Irv]

The types for index and columns are Axes, which are defined as:

AnyArrayLike: TypeAlias = Union[Index, Series, np.ndarray]
Axes: TypeAlias = Union[AnyArrayLike, list, dict, range]

We left out Sequence on purpose as an Axes argument, because a plain string is a Sequence. If you change your code to use List[int] instead of Sequence[int], it will work fine.

Because of this unfortunate aspect of python typing, we avoid using Sequence as the type of an argument when we the argument would accept allow a list of strings, but not accept a pure string.

[randolf-scholz]

Seems like this is another case where the absence of a NOT type bites.

Potentially one could discuss allowing at least MutableSequence to catch user types, as strings are immutable.

[bashtage]

Sounds like a good way forward for all of the places where we really want Sequence[Scalar] but can't because of str.

[gandhis1]

That would exclude things that should be included though, tuple, FrozenSet, etc.

[bashtage]

That would exclude things that should be included though, tuple, FrozenSet, etc.

Mostly we are using list[ScalarT] as a work around though, so it seems like it is likely an improvement in many cases. There are also quite a few places where tuple are poorly handled (e.g., anything that indexes).

[randolf-scholz]

@gandhis1 using tuple[int, int, int] or tuple[int, ...] in my example currently also raises [call-overload] errors.

[Dr-Irv]

I think that if we add tuple to the definition of Axes in _typing.pyi, we should be OK, as Axes is only used as a type for the argument for the Series and DataFrame constructors, and for reindex().

We just can't support Sequence