fix(parser): robust CONCAT_WS (#7544)

* fix(parser): robust CONCAT_WS

* fix double value

* revert comment

* fix

* remove sf

* add trino

Author: geooo109

sqlglot/dialects/dialect.py

@@ -399,6 +399,9 @@ class Dialect(metaclass=_Dialect):
     CONCAT_COALESCE = False
     """A `NULL` arg in `CONCAT` yields `NULL` by default, but in some dialects it yields an empty string."""
 
+    CONCAT_WS_COALESCE = False
+    """A `NULL` arg in `CONCAT_WS` yields `NULL` by default, but in some dialects it is skipped."""
+
     HEX_LOWERCASE = False
     """Whether the `HEX` function returns a lowercase hexadecimal string."""
 

sqlglot/dialects/dremio.py

@@ -9,6 +9,7 @@
 class Dremio(Dialect):
     SUPPORTS_USER_DEFINED_TYPES = False
     CONCAT_COALESCE = True
+    CONCAT_WS_COALESCE = True
     TYPED_DIVISION = True
     NULL_ORDERING = "nulls_are_last"
     SUPPORTS_VALUES_DEFAULT = False

sqlglot/dialects/drill.py

@@ -17,6 +17,7 @@ class Drill(Dialect):
     SUPPORTS_USER_DEFINED_TYPES = False
     TYPED_DIVISION = True
     CONCAT_COALESCE = True
+    CONCAT_WS_COALESCE = True
 
     TIME_MAPPING = {
         "y": "%Y",

sqlglot/dialects/duckdb.py

@@ -19,6 +19,7 @@ class DuckDB(Dialect):
     SAFE_DIVISION = True
     INDEX_OFFSET = 1
     CONCAT_COALESCE = True
+    CONCAT_WS_COALESCE = True
     SUPPORTS_ORDER_BY_ALL = True
     SUPPORTS_FIXED_SIZE_ARRAYS = True
     STRICT_JSON_PATH_SYNTAX = False

sqlglot/dialects/hive.py

@@ -20,6 +20,7 @@ class Hive(Dialect):
     IDENTIFIERS_CAN_START_WITH_DIGIT = True
     SUPPORTS_USER_DEFINED_TYPES = False
     SAFE_DIVISION = True
+    CONCAT_WS_COALESCE = True
     ARRAY_AGG_INCLUDES_NULLS = None
     REGEXP_EXTRACT_DEFAULT_GROUP = 1
     ALTER_TABLE_SUPPORTS_CASCADE = True

sqlglot/dialects/postgres.py

@@ -11,6 +11,7 @@ class Postgres(Dialect):
     INDEX_OFFSET = 1
     TYPED_DIVISION = True
     CONCAT_COALESCE = True
+    CONCAT_WS_COALESCE = True
     NULL_ORDERING = "nulls_are_large"
     TIME_FORMAT = "'YYYY-MM-DD HH24:MI:SS'"
     TABLESAMPLE_SIZE_IS_PERCENT = True

sqlglot/dialects/trino.py

@@ -9,6 +9,7 @@
 class Trino(Presto):
     SUPPORTS_USER_DEFINED_TYPES = False
     LOG_BASE_FIRST = True
+    CONCAT_WS_COALESCE = True
 
     class Tokenizer(Presto.Tokenizer):
         KEYWORDS = {

sqlglot/dialects/tsql.py

@@ -15,6 +15,7 @@ class TSQL(Dialect):
     LOG_BASE_FIRST = False
     TYPED_DIVISION = True
     CONCAT_COALESCE = True
+    CONCAT_WS_COALESCE = True
     NORMALIZATION_STRATEGY = NormalizationStrategy.CASE_INSENSITIVE
     ALTER_TABLE_ADD_REQUIRED_FOR_EACH_COLUMN = False
 

sqlglot/generator.py

@@ -3397,7 +3397,13 @@ def convert_concat_args(self, expression: exp.Func) -> list[exp.Expr]:
         if self.dialect.STRICT_STRING_CONCAT and expression.args.get("safe"):
             args = [exp.cast(e, exp.DType.TEXT) for e in args]
 
-        if not self.dialect.CONCAT_COALESCE and expression.args.get("coalesce"):
+        concat_coalesce = (
+            self.dialect.CONCAT_WS_COALESCE
+            if isinstance(expression, exp.ConcatWs)
+            else self.dialect.CONCAT_COALESCE
+        )
+
+        if not concat_coalesce and expression.args.get("coalesce"):
 
             def _wrap_with_coalesce(e: exp.Expr) -> exp.Expr:
                 if not e.type:
@@ -3432,8 +3438,8 @@ def concat_sql(self, expression: exp.Concat) -> str:
         return self.func("CONCAT", *expressions)
 
     def concatws_sql(self, expression: exp.ConcatWs) -> str:
-        if self.dialect.CONCAT_COALESCE and not expression.args.get("coalesce"):
-            # Dialect's CONCAT_WS function coalesces NULLs to empty strings, but the expression does not.
+        if self.dialect.CONCAT_WS_COALESCE and not expression.args.get("coalesce"):
+            # Dialect's CONCAT_WS function skips NULL args, but the expression does not.
             # Wrap the entire call in a CASE expression that returns NULL if any input IS NULL.
             all_args = expression.expressions
             expression.set("coalesce", True)

sqlglot/parser.py

@@ -342,7 +342,7 @@ class Parser:
         "CONCAT_WS": lambda args, dialect: exp.ConcatWs(
             expressions=args,
             safe=not dialect.STRICT_STRING_CONCAT,
-            coalesce=dialect.CONCAT_COALESCE,
+            coalesce=dialect.CONCAT_WS_COALESCE,
         ),
         "CONVERT_TIMEZONE": build_convert_timezone,
         "DATE_TO_DATE_STR": lambda args: exp.Cast(