Feat: transpile multi-arg DISTINCT expression (#2936)

sqlglot/dialects/duckdb.py

@@ -333,6 +333,7 @@ class Generator(generator.Generator):
         IGNORE_NULLS_IN_FUNC = True
         JSON_PATH_BRACKETED_KEY_SUPPORTED = False
         SUPPORTS_CREATE_TABLE_LIKE = False
+        MULTI_ARG_DISTINCT = False
 
         TRANSFORMS = {
             **generator.Generator.TRANSFORMS,

sqlglot/dialects/postgres.py

@@ -384,6 +384,7 @@ class Generator(generator.Generator):
         JSON_TYPE_REQUIRED_FOR_EXTRACTION = True
         SUPPORTS_UNLOGGED_TABLES = True
         LIKE_PROPERTY_INSIDE_SCHEMA = True
+        MULTI_ARG_DISTINCT = False
 
         SUPPORTED_JSON_PATH_PARTS = {
             exp.JSONPathKey,

sqlglot/dialects/presto.py

@@ -292,6 +292,7 @@ class Generator(generator.Generator):
         LIMIT_ONLY_LITERALS = True
         SUPPORTS_SINGLE_ARG_CONCAT = False
         LIKE_PROPERTY_INSIDE_SCHEMA = True
+        MULTI_ARG_DISTINCT = False
 
         PROPERTIES_LOCATION = {
             **generator.Generator.PROPERTIES_LOCATION,

sqlglot/generator.py

@@ -296,6 +296,10 @@ class Generator(metaclass=_Generator):
     # Whether or not the LikeProperty needs to be specified inside of the schema clause
     LIKE_PROPERTY_INSIDE_SCHEMA = False
 
+    # Whether or not DISTINCT can be followed by multiple args in an AggFunc. If not, it will be
+    # transpiled into a series of CASE-WHEN-ELSE, ultimately using a tuple conseisting of the args
+    MULTI_ARG_DISTINCT = True
+
     # Whether or not the JSON extraction operators expect a value of type JSON
     JSON_TYPE_REQUIRED_FOR_EXTRACTION = False
 
@@ -2837,6 +2841,13 @@ def addconstraint_sql(self, expression: exp.AddConstraint) -> str:
 
     def distinct_sql(self, expression: exp.Distinct) -> str:
         this = self.expressions(expression, flat=True)
+
+        if not self.MULTI_ARG_DISTINCT and len(expression.expressions) > 1:
+            case = exp.case()
+            for arg in expression.expressions:
+                case = case.when(arg.is_(exp.null()), exp.null())
+            this = self.sql(case.else_(f"({this})"))
+
         this = f" {this}" if this else ""
 
         on = self.sql(expression, "on")

tests/dialects/test_spark.py

@@ -277,6 +277,21 @@ def test_spark(self):
             "SELECT STR_TO_MAP('a:1,b:2,c:3', ',', ':')",
         )
 
+        self.validate_all(
+            "WITH tbl AS (SELECT 1 AS id, 'eggy' AS name UNION ALL SELECT NULL AS id, 'jake' AS name) SELECT COUNT(DISTINCT id, name) AS cnt FROM tbl",
+            write={
+                "clickhouse": "WITH tbl AS (SELECT 1 AS id, 'eggy' AS name UNION ALL SELECT NULL AS id, 'jake' AS name) SELECT COUNT(DISTINCT id, name) AS cnt FROM tbl",
+                "databricks": "WITH tbl AS (SELECT 1 AS id, 'eggy' AS name UNION ALL SELECT NULL AS id, 'jake' AS name) SELECT COUNT(DISTINCT id, name) AS cnt FROM tbl",
+                "doris": "WITH tbl AS (SELECT 1 AS id, 'eggy' AS name UNION ALL SELECT NULL AS id, 'jake' AS name) SELECT COUNT(DISTINCT id, name) AS cnt FROM tbl",
+                "duckdb": "WITH tbl AS (SELECT 1 AS id, 'eggy' AS name UNION ALL SELECT NULL AS id, 'jake' AS name) SELECT COUNT(DISTINCT CASE WHEN id IS NULL THEN NULL WHEN name IS NULL THEN NULL ELSE (id, name) END) AS cnt FROM tbl",
+                "hive": "WITH tbl AS (SELECT 1 AS id, 'eggy' AS name UNION ALL SELECT NULL AS id, 'jake' AS name) SELECT COUNT(DISTINCT id, name) AS cnt FROM tbl",
+                "mysql": "WITH tbl AS (SELECT 1 AS id, 'eggy' AS name UNION ALL SELECT NULL AS id, 'jake' AS name) SELECT COUNT(DISTINCT id, name) AS cnt FROM tbl",
+                "postgres": "WITH tbl AS (SELECT 1 AS id, 'eggy' AS name UNION ALL SELECT NULL AS id, 'jake' AS name) SELECT COUNT(DISTINCT CASE WHEN id IS NULL THEN NULL WHEN name IS NULL THEN NULL ELSE (id, name) END) AS cnt FROM tbl",
+                "presto": "WITH tbl AS (SELECT 1 AS id, 'eggy' AS name UNION ALL SELECT NULL AS id, 'jake' AS name) SELECT COUNT(DISTINCT CASE WHEN id IS NULL THEN NULL WHEN name IS NULL THEN NULL ELSE (id, name) END) AS cnt FROM tbl",
+                "snowflake": "WITH tbl AS (SELECT 1 AS id, 'eggy' AS name UNION ALL SELECT NULL AS id, 'jake' AS name) SELECT COUNT(DISTINCT id, name) AS cnt FROM tbl",
+                "spark": "WITH tbl AS (SELECT 1 AS id, 'eggy' AS name UNION ALL SELECT NULL AS id, 'jake' AS name) SELECT COUNT(DISTINCT id, name) AS cnt FROM tbl",
+            },
+        )
         self.validate_all(
             "SELECT TO_UTC_TIMESTAMP('2016-08-31', 'Asia/Seoul')",
             write={