fix(pyarrow): properly support round tripping of fixed size list types (#11330)

Author: cpcloud

ibis/formats/pyarrow.py

@@ -90,11 +90,10 @@ def to_ibis(cls, typ: pa.DataType, nullable=True) -> dt.DataType:
             return dt.Interval(typ.unit, nullable=nullable)
         elif pa.types.is_interval(typ):
             raise ValueError("Arrow interval type is not supported")
-        elif (
-            pa.types.is_list(typ)
-            or pa.types.is_large_list(typ)
-            or pa.types.is_fixed_size_list(typ)
-        ):
+        elif pa.types.is_fixed_size_list(typ):
+            value_dtype = cls.to_ibis(typ.value_type, typ.value_field.nullable)
+            return dt.Array(value_dtype, length=typ.list_size, nullable=nullable)
+        elif pa.types.is_list(typ) or pa.types.is_large_list(typ):
             value_dtype = cls.to_ibis(typ.value_type, typ.value_field.nullable)
             return dt.Array(value_dtype, nullable=nullable)
         elif pa.types.is_struct(typ):
@@ -196,7 +195,10 @@ def from_ibis(cls, dtype: dt.DataType) -> pa.DataType:
                 cls.from_ibis(dtype.value_type),
                 nullable=dtype.value_type.nullable,
             )
-            return pa.list_(value_field)
+            if dtype.length is None:
+                return pa.list_(value_field)
+            else:
+                return pa.list_(value_field, dtype.length)
         elif dtype.is_struct():
             fields = [
                 pa.field(name, cls.from_ibis(dtype), nullable=dtype.nullable)

ibis/formats/tests/test_pyarrow.py

@@ -41,6 +41,9 @@ def assert_dtype_roundtrip(arrow_type, ibis_type=None, restored_type=None):
         | past.binary_type
         | past.timestamp_types
         | st.builds(pa.list_, roundtripable_types)
+        | st.builds(
+            pa.list_, roundtripable_types, st.integers(min_value=1, max_value=2**31 - 1)
+        )
         | past.struct_types(roundtripable_types)
         | past.map_types(roundtripable_types, roundtripable_types)
     )
@@ -75,11 +78,6 @@ def test_roundtripable_types(arrow_type):
             dt.Array(dt.Int64(nullable=True), nullable=False),
             pa.list_(pa.int64()),
         ),
-        (
-            pa.list_(pa.int64(), list_size=3),
-            dt.Array(dt.Int64(nullable=True), nullable=False),
-            pa.list_(pa.int64()),
-        ),
     ],
 )
 def test_non_roundtripable_types(arrow_type, ibis_type, restored_type):