float_format in .to_string can be str

[zmoon]

Describe the bug
The change in #721 broke this for mypy. Although the current docs say "one-parameter function, optional, default None", you are allowed to pass a string that specifies the format like in .to_csv.

To Reproduce
These usages all work with both frame and series.

# t.py
import pandas as pd

df = pd.DataFrame({"a": [0.1, 1.0, 3.14]})

print(df.to_string(float_format="%.3g"))
print(df.to_string(float_format="{:.3g}".format))
print(df.to_string(float_format=lambda x: f"{x:.3e}"))
print(df.to_string(float_format=pd.io.formats.format.EngFormatter(3, True)))

print(df.a.to_string(float_format="%.3g"))
print(df.a.to_string(float_format="{:.3g}".format))
print(df.a.to_string(float_format=lambda x: f"{x:.3e}"))
print(df.a.to_string(float_format=pd.io.formats.format.EngFormatter(3, True)))

> mypy t.py
t.py:5: error: No overload variant of "to_string" of "DataFrame" matches argument type "str"  [call-overload]
t.py:5: note: Possible overload variants:
t.py:5: note:     def [HashableT1 <: Hashable, HashableT2 <: Hashable] to_string(self, buf: Union[str, PathLike[str], WriteBuffer[str]], columns: Union[List[HashableT1], Index, Series[Any], None] = ..., col_space: Union[int, List[int], Dict[HashableT2, int], None] = ..., header: Union[bool, List[str], Tuple[str, ...]] = ..., index: bool = ..., na_rep: str = ..., formatters: Union[List[Callable[..., Any]], Tuple[Callable[..., Any], ...], Mapping[Union[str, int], Callable[..., Any]], None] = ..., float_format: Optional[Callable[[float], str]] = ..., sparsify: Optional[bool] = ..., index_names: bool = ..., justify: Optional[str] = ..., max_rows: Optional[int] = ..., max_cols: Optional[int] = ..., show_dimensions: bool = ..., decimal: str = ..., line_width: Optional[int] = ..., min_rows: Optional[int] = ..., max_colwidth: Optional[int] = ..., encoding: Optional[str] = ...) -> None
t.py:5: note:     def [HashableT <: Hashable] to_string(self, buf: None = ..., columns: Union[List[HashableT], Index, Series[Any], None] = ..., col_space: Union[int, List[int], Dict[Hashable, int], None] = ..., header: Union[bool, Sequence[str]] = ..., index: bool = ..., na_rep: str = ..., formatters: Union[List[Callable[..., Any]], Tuple[Callable[..., Any], ...], Mapping[Union[str, int], Callable[..., Any]], None] = ..., float_format: Optional[Callable[[float], str]] = ..., sparsify: Optional[bool] = ..., index_names: bool = ..., justify: Optional[str] = ..., max_rows: Optional[int] = ..., max_cols: Optional[int] = ..., show_dimensions: bool = ..., decimal: str = ..., line_width: Optional[int] = ..., min_rows: Optional[int] = ..., max_colwidth: Optional[int] = ..., encoding: Optional[str] = ...) -> str
t.py:10: error: No overload variant of "to_string" of "Series" matches argument type "str"  [call-overload]
t.py:10: note: Possible overload variants:
t.py:10: note:     def to_string(self, buf: Union[str, PathLike[str], WriteBuffer[str]], na_rep: str = ..., float_format: Callable[[float], str] = ..., header: bool = ..., index: bool = ..., length: bool = ..., dtype: bool = ..., name: bool = ..., max_rows: Optional[int] = ..., min_rows: Optional[int] = ...) -> None
t.py:10: note:     def to_string(self, buf: None = ..., na_rep: str = ..., float_format: Callable[[float], str] = ..., header: bool = ..., index: bool = ..., length: bool = ..., dtype: bool = ..., name: bool = ..., max_rows: Optional[int] = ..., min_rows: Optional[int] = ...) -> str
Found 2 errors in 1 file (checked 1 source file)

Please complete the following information:

• OS: [e.g. Windows, Linux, MacOS] Windows
• OS Version [e.g. 22] 10
• python version 3.10.11
• version of type checker mypy 1.3.0
• version of installed pandas-stubs 2.0.2.230605
• pandas 2.0.2

Additional context
Maybe FloatFormatType can be used (again) but could be updated to be more specific about the callable (Callable[[float], str]).

And perhaps the pandas docs should be updated? Unless support for passing format string was supposed to have been removed?

[Dr-Irv]

@twoertwein can you take a look at this?

[twoertwein]

And perhaps the pandas docs should be updated? Unless support for passing format string was supposed to have been removed?

In the past, it apparently did not work. Since it works now, it probably is intended to pass formatting strings. Do you mind opening an issue at pandas to clarify the intend/update the docs?

[zmoon]

In the past, it apparently did not work.

Issue that seems related: pandas-dev/pandas#9448

[zmoon]

It still doesn't work for extension dtypes.

pd.Series([1.1, 2.2, 3.3]).to_string(float_format="%.3e")  # works
pd.Series([1.1, 2.2, 3.3]).convert_dtypes().to_string(float_format="%.3e")  # `TypeError: 'str' object is not callable`

[twoertwein]

Thank you, @zmoon ! In that case, I would be inclined to not add str to the annotations.

[zmoon]

That seems fair until the docs are clarified and it works for extension dtypes too. Maybe we could leave this open though?

[twoertwein]

Feel free to open also a new pandas issue to highlight the discrepancy with the extension types (and xref the existing issue about str/no str).

edit: should we use the label "upstream issue" for this (and similar issues which first need to be confirmed/fixed in pandas)?