fix(athena): always generate string literal from FileFormatProperty

sqlglot/dialects/athena.py

@@ -71,6 +71,13 @@ def _partitioned_by_property_sql(self: Athena.Generator, e: exp.PartitionedByPro
     return f"{prop_name}={self.sql(e, 'this')}"
 
 
+def _file_format_property_sql(self: Athena.Generator, e: exp.FileFormatProperty) -> str:
+    this = e.args.get("this")
+    if not this:
+        return "format=''"
+    return f"format={exp.Literal.string(this.name)}"
+
+
 class Athena(Trino):
     """
     Over the years, it looks like AWS has taken various execution engines, bolted on AWS-specific modifications and then
@@ -148,7 +155,7 @@ class Generator(Trino.Generator):
 
         TRANSFORMS = {
             **Trino.Generator.TRANSFORMS,
-            exp.FileFormatProperty: lambda self, e: f"format={self.sql(e, 'this')}",
+            exp.FileFormatProperty: _file_format_property_sql,
             exp.PartitionedByProperty: _partitioned_by_property_sql,
             exp.LocationProperty: _location_property_sql,
         }

tests/dialects/test_athena.py

@@ -337,3 +337,22 @@ def test_parse_partitioned_by_returns_iceberg_transforms(self):
         assert isinstance(parsed.this, exp.Schema)
         assert next(n for n in parsed.this.expressions if isinstance(n, exp.PartitionedByBucket))
         assert next(n for n in parsed.this.expressions if isinstance(n, exp.PartitionByTruncate))
+
+    def test_ctas_uses_string_for_format(self):
+        # When transpiling, sometimes the FileFormatProperty is a Var, not a
+        # string literal and we need to ensure the athena dialect uses a string
+        # literal for the format
+        ctas_hive = exp.Create(
+            this=exp.to_table("foo.bar"),
+            kind="TABLE",
+            properties=exp.Properties(
+                expressions=[exp.FileFormatProperty(this=exp.Var(this="parquet"))]
+            ),
+            expression=exp.select("1"),
+        )
+
+        # Even if identify=True, the column names should not be quoted within the string literals in the partitioned_by ARRAY[]
+        self.assertEqual(
+            ctas_hive.sql(dialect=self.dialect, identify=True),
+            'CREATE TABLE "foo"."bar" WITH (format=\'parquet\') AS SELECT 1',
+        )

tests/dialects/test_dialect.py

@@ -1,4 +1,5 @@
 import unittest
+from typing import Optional
 
 from sqlglot import (
     Dialect,
@@ -16,7 +17,7 @@
 
 
 class Validator(unittest.TestCase):
-    dialect = None
+    dialect: Optional[str] = None
 
     def parse_one(self, sql, **kwargs):
         return parse_one(sql, read=self.dialect, **kwargs)