feat(optimizer)!: bq annotate type for NULL

sqlglot/dialects/bigquery.py

@@ -371,6 +371,7 @@ class BigQuery(Dialect):
     EXCLUDES_PSEUDOCOLUMNS_FROM_STAR = True
     QUERY_RESULTS_ARE_STRUCTS = True
     JSON_EXTRACT_SCALAR_SCALAR_ONLY = True
+    DEFAULT_NULL_TYPE = exp.DataType.Type.BIGINT
 
     # https://docs.cloud.google.com/bigquery/docs/reference/standard-sql/string_functions#initcap
     INITCAP_DEFAULT_DELIMITER_CHARS = ' \t\n\r\f\v\\[\\](){}/|<>!?@"^#$&~_,.:;*%+\\-'

sqlglot/dialects/dialect.py

@@ -703,6 +703,13 @@ class Dialect(metaclass=_Dialect):
     so we map the ExplodingGenerateSeries expression to "generate_series" string.
     """
 
+    DEFAULT_NULL_TYPE = exp.DataType.Type.UNKNOWN
+    """
+    The default type of NULL for producing the correct projection type.
+
+    For example, in BigQuery the default type of the NULL value is INT64.
+    """
+
     # --- Autofilled ---
 
     tokenizer_class = Tokenizer

sqlglot/optimizer/annotate_types.py

@@ -261,15 +261,16 @@ def annotate(self, expression: E, annotate_scope: bool = True) -> E:
         # This takes care of non-traversable expressions
         self._annotate_expression(expression)
 
-        # Replace NULL type with UNKNOWN, since the former is not an actual type;
+        # Replace NULL type with the default type of the targeted dialect, since the former is not an actual type;
         # it is mostly used to aid type coercion, e.g. in query set operations.
         for expr in self._null_expressions.values():
-            expr.type = exp.DataType.Type.UNKNOWN
+            expr.type = self.dialect.DEFAULT_NULL_TYPE
 
         return expression
 
     def annotate_scope(self, scope: Scope) -> None:
         selects = {}
+
         for name, source in scope.sources.items():
             if not isinstance(source, Scope):
                 continue

tests/fixtures/optimizer/annotate_types.sql

@@ -122,3 +122,7 @@ DATETIME;
 # dialect: bigquery
 CASE WHEN TRUE THEN TIMESTAMP '2020-02-02 00:00:00' ELSE '2010-01-01' END;
 TIMESTAMP;
+
+# dialect: bigquery
+NULL;
+INT64;

tests/test_optimizer.py

@@ -1937,3 +1937,15 @@ def test_deep_ast_type_annotation(self):
         annotated = annotate_types(parse_one(binary_sql), schema={"t": {"a": "INT"}})
         self.assertEqual(annotated.sql(), binary_sql)
         self.assertEqual(annotated.selects[0].type.this, exp.DataType.Type.INT)
+
+    def test_null_coerce_annotation(self):
+        null_sql = "SELECT t.foo FROM (SELECT CAST(1 AS BIGDECIMAL) AS foo UNION ALL SELECT NULL AS foo) AS t"
+        annotated = parse_and_optimize(annotate_types, null_sql, "bigquery", dialect="bigquery")
+
+        self.assertEqual(annotated.sql(), null_sql)
+        self.assertEqual(annotated.selects[0].type.this, exp.DataType.Type.BIGDECIMAL)
+
+        null_sql = "SELECT t.foo FROM (SELECT NULL AS foo UNION ALL SELECT CAST(1 AS BIGDECIMAL) AS foo) AS t"
+        annotated = parse_and_optimize(annotate_types, null_sql, "bigquery", dialect="bigquery")
+        self.assertEqual(annotated.sql(), null_sql)
+        self.assertEqual(annotated.selects[0].type.this, exp.DataType.Type.BIGDECIMAL)