feat(duckdb): convert ARRAY_AGG IGNORE NULLS to FILTER clause

sqlglot/generator.py

@@ -5668,7 +5668,15 @@ def _add_arrayagg_null_filter(
 
     def arrayagg_sql(self, expression: exp.ArrayAgg) -> str:
         array_agg = self.function_fallback_sql(expression)
-        return self._add_arrayagg_null_filter(array_agg, expression, expression.this)
+        # The argument inside ARRAY_AGG may be wrapped in Order and/or Limit
+        # nodes.  Unwrap to get the bare column expression so the FILTER
+        # clause is just "WHERE x IS NOT NULL".
+        column_expr = expression.this
+        if isinstance(column_expr, exp.Limit):
+            column_expr = column_expr.this
+        if isinstance(column_expr, exp.Order):
+            column_expr = column_expr.this
+        return self._add_arrayagg_null_filter(array_agg, expression, column_expr)
 
     def slice_sql(self, expression: exp.Slice) -> str:
         step = self.sql(expression, "step")

sqlglot/generators/duckdb.py

@@ -3953,6 +3953,11 @@ def unnest_sql(self, expression: exp.Unnest) -> str:
 
         return super().unnest_sql(expression)
 
+    def arrayagg_sql(self, expression: exp.ArrayAgg) -> str:
+        if isinstance(expression.this, exp.Limit):
+            self.unsupported("LIMIT inside ARRAY_AGG is not supported in DuckDB")
+        return super().arrayagg_sql(expression)
+
     def ignorenulls_sql(self, expression: exp.IgnoreNulls) -> str:
         this = expression.this
 
@@ -3961,6 +3966,14 @@ def ignorenulls_sql(self, expression: exp.IgnoreNulls) -> str:
             # window functions that accept it e.g. FIRST_VALUE(... IGNORE NULLS) OVER (...)
             return super().ignorenulls_sql(expression)
 
+        # For ARRAY_AGG(expr IGNORE NULLS ...), convert IGNORE NULLS to a
+        # FILTER(WHERE expr IS NOT NULL) clause by setting nulls_excluded on
+        # the ArrayAgg.  The existing _add_arrayagg_null_filter method will
+        # emit the FILTER clause during arrayagg_sql / withingroup_sql.
+        if isinstance(this, exp.ArrayAgg):
+            this.set("nulls_excluded", True)
+            return self.sql(this)
+
         if isinstance(this, exp.First):
             this = exp.AnyValue(this=this.this)
 

tests/dialects/test_bigquery.py

@@ -737,18 +737,10 @@ def test_bigquery(self):
             "SELECT ARRAY_AGG(DISTINCT x IGNORE NULLS ORDER BY a, b DESC LIMIT 10) AS x",
             write={
                 "bigquery": "SELECT ARRAY_AGG(DISTINCT x IGNORE NULLS ORDER BY a, b DESC LIMIT 10) AS x",
-                "duckdb": "SELECT ARRAY_AGG(DISTINCT x ORDER BY a NULLS FIRST, b DESC LIMIT 10) AS x",
+                "duckdb": UnsupportedError,
                 "spark": "SELECT COLLECT_LIST(DISTINCT x ORDER BY a, b DESC LIMIT 10) IGNORE NULLS AS x",
             },
         )
-        self.validate_all(
-            "SELECT ARRAY_AGG(DISTINCT x IGNORE NULLS ORDER BY a, b DESC LIMIT 1, 10) AS x",
-            write={
-                "bigquery": "SELECT ARRAY_AGG(DISTINCT x IGNORE NULLS ORDER BY a, b DESC LIMIT 1, 10) AS x",
-                "duckdb": "SELECT ARRAY_AGG(DISTINCT x ORDER BY a NULLS FIRST, b DESC LIMIT 1, 10) AS x",
-                "spark": "SELECT COLLECT_LIST(DISTINCT x ORDER BY a, b DESC LIMIT 1, 10) IGNORE NULLS AS x",
-            },
-        )
         self.validate_all(
             "SELECT * FROM Produce UNPIVOT((first_half_sales, second_half_sales) FOR semesters IN ((Q1, Q2) AS 'semester_1', (Q3, Q4) AS 'semester_2'))",
             read={