fix: model TO_DATE return type — DATE without format, DATETIME with format

TO_DATE(str) returns DATE → maps to TsOrDsToDate (unchanged).
TO_DATE(str, fmt) returns DATETIME → maps to StrToTime so cross-dialect
consumers see a datetime type; generator emits TO_DATE(str, fmt) for MaxCompute.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

Author: azurechen97

src/sqlglot_maxcompute/generator.py

@@ -73,6 +73,8 @@ class MaxComputeGenerator(Hive.Generator):
         exp.Variance: rename_func("VAR_SAMP"),
         # String position: MaxCompute uses INSTR(str, substr), not LOCATE(substr, str)
         exp.StrPosition: lambda self, e: self.func("INSTR", e.this, e.args.get("substr")),
+        # TO_DATE(str, fmt) returns DATETIME — modeled as StrToTime; emit TO_DATE in MaxCompute
+        exp.StrToTime: lambda self, e: self.func("TO_DATE", e.this, e.args.get("format")),
     }
 
     def _dateadd_sql(

src/sqlglot_maxcompute/parser.py

@@ -73,9 +73,13 @@ class MaxComputeParser(Hive.Parser):
         "DATE_FORMAT": lambda args: exp.TimeToStr(
             this=seq_get(args, 0), format=seq_get(args, 1)
         ),
-        # Hive override: MaxCompute TO_DATE accepts date types directly (no TimeStrToTime wrap)
-        "TO_DATE": lambda args: exp.TsOrDsToDate(
-            this=seq_get(args, 0), format=seq_get(args, 1)
+        # Hive override: TO_DATE return type depends on args:
+        #   TO_DATE(str)       → DATE   → TsOrDsToDate (no format)
+        #   TO_DATE(str, fmt)  → DATETIME → StrToTime (format present)
+        "TO_DATE": lambda args: (
+            exp.StrToTime(this=seq_get(args, 0), format=seq_get(args, 1))
+            if seq_get(args, 1) is not None
+            else exp.TsOrDsToDate(this=seq_get(args, 0))
         ),
         # Hive override: MaxCompute FROM_UNIXTIME takes 1 arg and returns DATETIME, not STRING
         "FROM_UNIXTIME": lambda args: exp.UnixToTime(this=seq_get(args, 0)),

tests/test_maxcompute.py

@@ -241,11 +241,29 @@ def test_date_conversion(self):
             },
         )
 
-        # TO_DATE: parses without Hive's TimeStrToTime wrapping (format stored as-is)
-        expr = self.parse_one("TO_DATE('2024-01-01', 'yyyy-mm-dd')")
+        # TO_DATE without format → DATE (TsOrDsToDate)
+        expr = self.parse_one("TO_DATE('2024-01-01')")
         self.assertIsInstance(expr, exp.TsOrDsToDate)
-        # Format should be stored as Oracle style, not strftime
-        self.assertEqual(expr.args.get("format").this, "yyyy-mm-dd")
+        self.assertIsNone(expr.args.get("format"))
+        self.validate_all(
+            "TO_DATE('2024-01-01')",
+            write={
+                "maxcompute": "TO_DATE('2024-01-01')",
+                "spark": "TO_DATE('2024-01-01')",
+            },
+        )
+
+        # TO_DATE with format → DATETIME (StrToTime); format stored as MaxCompute style, not strftime
+        expr = self.parse_one("TO_DATE('20240101', 'yyyymmdd')")
+        self.assertIsInstance(expr, exp.StrToTime)
+        self.assertEqual(expr.args.get("format").this, "yyyymmdd")
+        self.validate_all(
+            "TO_DATE('20240101', 'yyyymmdd')",
+            write={
+                "maxcompute": "TO_DATE('20240101', 'yyyymmdd')",
+                "spark": "TO_TIMESTAMP('20240101', 'yyyymmdd')",
+            },
+        )
 
         # TO_CHAR (untyped arg → ToChar)
         self.assertIsInstance(self.parse_one("TO_CHAR(dt, 'yyyy-mm-dd')"), exp.ToChar)