fix: pandas 3.0 compatibility for strict string dtype enforcement

[lukepayyapilli]

Summary

@tompollard Fixes #207 - Incompatibility with Pandas 3.0+ string dtype.

Pandas 3.0 enforces strict string dtype, which raises TypeError when assigning non-string values to string-typed columns.

Approach

Per the pandas 3.0 migration guide, the recommended approach for library maintainers is:

1. Detection: Use pd.api.types.is_string_dtype() to detect string columns (works for both object and string dtypes).
2. Mixed-type assignments: Convert to object dtype with .astype(object) when flexible typing is needed.
3. Null handling: Use fillna() instead of replace('nan', ...) since pandas 3.0 NA values are actual NA objects, not the string 'nan'.

Changes

tableone/tableone.py

• In _insert_n_row(): Convert string dtype columns to object dtype before assigning integer counts.

tableone/preprocessors.py

• In handle_categorical_nulls(): Use fillna() instead of replace('nan', ...) for pandas 3.0 compatibility.

tests/unit/test_tableone.py

• Fix data_mixed fixture: Convert column to object dtype before assigning string to numeric column.
• Add two new tests for pandas 3.0 string dtype compatibility.

Testing

• All 46 tests pass with pandas 3.0.0.

Related

Issue #208 discusses adding version ceilings to dependencies to protect against future breaking changes.

[tompollard]

thanks luke!