ENH: Can pandas.Series.str.len returns a nullable pd.Int64Dtype rather than float64

[chelsea-lin]

Feature Type

• Adding new functionality to pandas

• Changing existing functionality in pandas

• Removing existing functionality in pandas

Problem Description

The [pandas.Series.str.len](https://pandas.pydata.org/docs/reference/api/pandas.Series.str.len.html) documents the returned value is Series or Index of int, but the codes actually return a float64 dtype because it allows nullable integers. However, float64 may loss the precision. Ideally, can we return pd.Int64Dtype(), the nullable integer data type here? Same request applied to pandas.Series.str.find etc.

Feature Description

Add use_nullable_dtype parameter to return a pd.Int64Dtype() in the pandas.Series.str.len() API

Alternative Solutions

pandas.Series.str.len().astype(pd.Int64Dtype()) can explicitly change the dtype from float64 but may loss the precision.

Additional Context

No response

[molsonkiko]

This is more of a documentation bug that anything else. The largest positive float that can be precisely represented as an integer is 2**53, which would correspond to a 9 quadrillion character string (9000 terabytes assuming UTF-8 encoding) as a single entry in a pandas series.

[tswast]

True that there's no loss of precision for the immediate return value for realistically sized strings, though it could make some unexpected types bubble through in subsequent operations.

IMO, if someone is using the new StringDtype() it makes sense to use the other nullable extension types in methods such as this.