[Python] PyArrow allows creating non-nullable columns containing nulls

[adamreeve]

Describe the bug, including details regarding any error messages, version, and platform.

Reproduced using PyArrow 16.1.0 and Python 3.12 on Fedora 39 Linux. The null values are still displayed as null when printing, but if round tripped via Parquet they are lost:

import pyarrow as pa
import pyarrow.parquet as pq

schema = pa.schema([
    pa.field("x", pa.int64(), nullable=False)])

table = pa.Table.from_pydict({
        "x": [1, None, 3],
    }, schema=schema)

print(f"Original table:\n{table}\n")

pq.write_table(table, 'data.parquet')
read = pq.read_table('data.parquet')

print(f"Table from Parquet:\n{read}")

This outputs:

Original table:
pyarrow.Table
x: int64 not null
----
x: [[1,null,3]]

Table from Parquet:
pyarrow.Table
x: int64 not null
----
x: [[1,0,3]]

Is this expected behaviour? I would have thought this should raise an exception when the table is created.

Component(s)

Python

[adamreeve]

Closing this as a duplicate of #41667