fix: set pyarrow field nullable to False for a BigQuery field in REPEATED mode (#1999)

Co-authored-by: Lingqing Gan <[email protected]>

Author: xyloid

google/cloud/bigquery/_pandas_helpers.py

@@ -200,7 +200,7 @@ def bq_to_arrow_field(bq_field, array_type=None):
             # local NULL values. Arrow will gladly interpret these NULL values
             # as non-NULL and give you an arbitrary value. See:
             # https://github.com/googleapis/python-bigquery/issues/1692
-            nullable=True,
+            nullable=False if bq_field.mode.upper() == "REPEATED" else True,
             metadata=metadata,
         )
 

tests/unit/test__pandas_helpers.py

@@ -2002,6 +2002,23 @@ def test_bq_to_arrow_field_type_override(module_under_test):
     )
 
 
+@pytest.mark.skipif(isinstance(pyarrow, mock.Mock), reason="Requires `pyarrow`")
+def test_bq_to_arrow_field_set_repeated_nullable_false(module_under_test):
+    assert (
+        module_under_test.bq_to_arrow_field(
+            schema.SchemaField("name", "STRING", mode="REPEATED")
+        ).nullable
+        is False
+    )
+
+    assert (
+        module_under_test.bq_to_arrow_field(
+            schema.SchemaField("name", "STRING", mode="NULLABLE")
+        ).nullable
+        is True
+    )
+
+
 @pytest.mark.parametrize(
     "field_type, metadata",
     [