col_summary_tbl() fails with a ShapeError when a boolean column has no False values #465

Description

col_summary_tbl() raises an error whenever a boolean column contains no False values: a column that is all True, True plus missing values, or any boolean column in a one-row table. This happens on Polars, Pandas and DuckDB (Ibis) tables.

Minimal reproducible example

import polars as pl
import pointblank as pb

pb.col_summary_tbl(pl.DataFrame({"is_active": [True, True, True]}))

narwhals.exceptions.ShapeError: can't broadcast Series '__freq_true' of length 2 to length 1

The same happens with:

import ibis
import pandas as pd

pb.col_summary_tbl(pl.DataFrame({"flag": [True]}))                    # one-row table
pb.col_summary_tbl(pl.DataFrame({"flag": [True, None, True]}))        # True plus missing values
pb.col_summary_tbl(pd.DataFrame({"flag": [True, True]}))              # ShapeError: Expected object of length 1, got: 2.
pb.col_summary_tbl(ibis.memtable({"flag": [True, True], "x": [1, 2]}))  # ShapeError: Expected object of length 2, got: 3.

A related case gives wrong output with no error. When the first boolean column has values that the next one lacks, the next column loses its T/F summary or shows the wrong False share:

pb.col_summary_tbl(pl.DataFrame({"a": [True, True, False], "b": [True, True, True]}))
# `b` shows a plain unique count instead of "T 1 / F 0"

pb.col_summary_tbl(pl.DataFrame({"a": [True, True, True], "b": [True, False, None]}))
# `b` shows "F 0" even though one third of its values are False

Expected

Every boolean column is summarized as the share of True and False values, e.g. "T 1 / F 0" for an all-True column.

Actual

Either a ShapeError is raised, or the summary for the later boolean column is missing or wrong.

The frequencies stored for a boolean column (freqs) only have keys for the values that are present, e.g. {"True": 3} for an all-True column. In the report, the T/F extraction then either reads a key that doesn't exist or builds one struct column for all boolean columns using only the first column's keys.

Environment

• pointblank main (6ba7841)
• Python 3.12.13, polars 2.0.0, pandas 2.3.3, narwhals 2.27.1, ibis-framework 12.0.0, duckdb 1.3.2, great_tables 1.0.0
• macOS (arm64)