error: Argument 1 to "astype" of "DataFrame" has incompatible type "Dict[str, str]" [arg-type]

[randolf-scholz]

import pandas as pd

df = pd.DataFrame({"a": [1, 2, 3], "b": [4, 5, 6]})
columns_types = {"a" : "int", "b" : "float"}
df = df.astype(columns_types)  # ✘ raises [arg-type]

Context:

error: Argument 1 to "astype" of "DataFrame" has incompatible type "Dict[str, str]"; expected
"Union[str, Union[ExtensionDtype, Union[str, dtype[generic], Type[str], Type[complex], Type[bool], Type[object]]], Dict[str, Union[str, Union[ExtensionDtype, Union[str, dtype[generic], Type[str], Type[complex], Type[bool], Type[object]]]]]]"  [arg-type]
    df = df.astype(columns_types)  # ✘ raises [arg-type]
                   ^~~~~~~~~~~~~
note: "Dict" is invariant -- see https://mypy.readthedocs.io/en/stable/common_issues.html#variance
note: Consider using "Mapping" instead, which is covariant in the value type

Suggestion:

pandas-stubs/pandas-stubs/core/frame.pyi

Lines 1360 to 1365 in 001c0af

	def astype(
	self,
	dtype: _str | Dtype | dict[_str, _str | Dtype],
	copy: _bool = ...,
	errors: _str = ...,
	) -> DataFrame: ...

needs to be replaced with

    def astype(
        self,
        dtype: _str | Dtype | Mapping[_str, _str | Dtype],
        copy: _bool = ...,
        errors: _str = ...,
    ) -> DataFrame: ...

Please complete the following information:

• python 3.10.6
• mypy 0.982
• pandas 1.5.1
• pandas-stubs 1.5.1.221024

[gandhis1]

IMO we should be using Mapping pretty much anywhere a Dict is required. You can't pass a TypedDict in otherwise

[randolf-scholz]

I think it's unfortunately not super straight forward. One needs to check:

• in which cases the Mapping sufficies
• in which cases MutableMapping is required
• in which cases dict is required. Unfortunately, MutableMapping lacks {'__or__', 'copy', '__ror__', '__ior__', 'fromkeys'}.
  • In particular, for some reason MutableMapping lacks the dictionary merge operator | (PEP-584)

[randolf-scholz]

I created an issue for this in the cpython repo python/cpython#99327

Won't happen.

[gandhis1]

I think it's unfortunately not super straight forward. One needs to check:

• in which cases the Mapping sufficies

• in which cases MutableMapping is required

• in which cases dict is required. Unfortunately, MutableMapping lacks {'__or__', 'copy', '__ror__', '__ior__', 'fromkeys'}.

  • In particular, for some reason MutableMapping lacks the dictionary merge operator | (PEP-584)

I think the baseline expectation should be that Pandas is not mutating any of the passed dictionaries. Nor is there an intuitive reason for Pandas to do so.