Fix(bigquery): type-annotated array literal logic edge case (#4724)

Author: georgesittas

sqlglot/dialects/bigquery.py

@@ -1245,7 +1245,12 @@ def contains_sql(self, expression: exp.Contains) -> str:
         def cast_sql(self, expression: exp.Cast, safe_prefix: t.Optional[str] = None) -> str:
             this = expression.this
 
+            # This ensures that inline type-annotated ARRAY literals like ARRAY<INT64>[1, 2, 3]
+            # are roundtripped unaffected. The inner check excludes ARRAY(SELECT ...) expressions,
+            # because they aren't literals and so the above syntax is invalid BigQuery.
             if isinstance(this, exp.Array):
-                return f"{self.sql(expression, 'to')}{self.sql(this)}"
+                elem = seq_get(this.expressions, 0)
+                if not (elem and elem.find(exp.Query)):
+                    return f"{self.sql(expression, 'to')}{self.sql(this)}"
 
             return super().cast_sql(expression, safe_prefix=safe_prefix)

tests/dialects/test_bigquery.py

@@ -234,6 +234,10 @@ def test_bigquery(self):
         self.validate_identity(
             "CREATE OR REPLACE VIEW test (tenant_id OPTIONS (description='Test description on table creation')) AS SELECT 1 AS tenant_id, 1 AS customer_id",
         )
+        self.validate_identity(
+            "ARRAY(SELECT AS STRUCT e.x AS y, e.z AS bla FROM UNNEST(bob))::ARRAY<STRUCT<y STRING, bro NUMERIC>>",
+            "CAST(ARRAY(SELECT AS STRUCT e.x AS y, e.z AS bla FROM UNNEST(bob)) AS ARRAY<STRUCT<y STRING, bro NUMERIC>>)",
+        )
         self.validate_identity(
             "SELECT * FROM `proj.dataset.INFORMATION_SCHEMA.SOME_VIEW`",
             "SELECT * FROM `proj.dataset.INFORMATION_SCHEMA.SOME_VIEW` AS `proj.dataset.INFORMATION_SCHEMA.SOME_VIEW`",