fix(athena): treat DECIMAL(p) as scale 0, not a default non-zero scale (#2403)

core/wren/src/wren/connector/athena.py

@@ -91,7 +91,13 @@ def _trino_data_type_to_arrow(node) -> pa.DataType:
         return _TRINO_DATA_TYPE_TO_ARROW[kind]
 
     if kind == T.DECIMAL:
-        precision, scale = 38, 9
+        # Trino/Athena DECIMAL semantics: DECIMAL(p) means scale 0 (an integer
+        # of p digits), NOT an unspecified scale. Bare DECIMAL defaults to
+        # (38, 0). Only DECIMAL(p, s) carries a non-zero scale. Defaulting the
+        # scale to a non-zero value here both mistyped DECIMAL(p) columns and,
+        # for small precisions (e.g. DECIMAL(5)), produced scale > precision,
+        # which PyArrow rejects with an ArrowInvalid.
+        precision, scale = 38, 0
         params = node.expressions
         if len(params) >= 1:
             with contextlib.suppress(AttributeError, ValueError, TypeError):

core/wren/tests/unit/test_athena_connector.py

@@ -78,6 +78,22 @@ def test_parse_athena_type_decimal():
     assert _parse_athena_type("decimal(12,4)") == pa.decimal128(12, 4)
 
 
+def test_parse_athena_type_decimal_precision_only_is_scale_zero():
+    # Trino/Athena: DECIMAL(p) means scale 0, not an unspecified/default scale.
+    assert _parse_athena_type("decimal(10)") == pa.decimal128(10, 0)
+
+
+def test_parse_athena_type_decimal_bare_defaults():
+    # Bare DECIMAL is DECIMAL(38, 0) per the SQL standard / Trino.
+    assert _parse_athena_type("decimal") == pa.decimal128(38, 0)
+
+
+def test_parse_athena_type_decimal_small_precision_only():
+    # Regression: DECIMAL(5) must not yield scale > precision. A non-zero
+    # default scale (e.g. 9) produced decimal128(5, 9), which PyArrow rejects.
+    assert _parse_athena_type("decimal(5)") == pa.decimal128(5, 0)
+
+
 def test_parse_athena_type_array():
     assert _parse_athena_type("array(varchar)") == pa.list_(pa.string())