fix: render data frames with non-string column names (#2115)

[eeshsaxena]

Fixes #2115.

Problem

Rendering a render.DataGrid / render.data_frame from a DataFrame whose column names are non-string (e.g. numeric) fails with an obscure AttributeError:

from shiny.express import render
import pandas as pd

@render.data_frame
def table():
    df = pd.DataFrame([["a", 1], ["b", 2], ["c", 3], ["d", 4]], columns=[0, 1])
    return render.DataGrid(df)
# AttributeError: 'DataFrame' object has no attribute 'dtype'

Cause

In shiny/render/_data_frame_utils/_tbl_data.py, columns were accessed with data[col_name] on the narwhals frame:

type_hints = [serialize_dtype(data[col_name]) for col_name in data.columns]

For a numeric column name, narwhals interprets data[0] as positional/row access and returns a row frame rather than the column, so serialize_dtype receives a DataFrame instead of a Series. Thanks to @… (the issue reporter) for the precise diagnosis.

The same ambiguous access exists in the cell-patch scatter path (apply_frame_patches), where the column name comes from nw_data.columns[...] and can likewise be numeric:

nw_data[column_name].scatter(...)

Fix

Use data.get_column(col_name) / nw_data.get_column(column_name) in both places, which unambiguously accesses a column by name.

>>> serialize_frame(as_data_frame(pd.DataFrame([["a", 1], ["b", 2]], columns=[0, 1])))
# now returns columns=[0, 1], typeHints=["string", "numeric"], data=[["a", 1], ["b", 2]]

Testing

• Added test_serialize_frame_numeric_column_names in tests/pytest/test_render_data_frame_tbl_data.py. It fails on main with the AttributeError and passes with this change.
• Full test_render_data_frame_tbl_data.py passes (42 tests); ruff check + format clean.
• Added a CHANGELOG entry under Bug fixes.

An AI assistant helped implement this fix (following the root-cause and suggested approach from the issue); I reviewed the change and ran the tests locally.