fix(optimizer): convert INNER JOINs to LEFT JOINs when merging LEFT JOIN subqueries (#5980)

* fix(optimizer): convert INNER JOINs to LEFT JOINs when merging LEFT JOIN subqueries

* change comment

* refactor names

Author: geooo109

sqlglot/optimizer/merge_subqueries.py

@@ -282,7 +282,14 @@ def _merge_joins(outer_scope: Scope, inner_scope: Scope, from_or_join: FromOrJoi
     new_joins = []
 
     joins = inner_scope.expression.args.get("joins") or []
+
+    outer_is_left_join = isinstance(from_or_join, exp.Join) and from_or_join.side == "LEFT"
+
     for join in joins:
+        if outer_is_left_join and not join.method and join.kind in ("", "INNER"):
+            join.args.pop("kind", None)
+            join.set("side", "LEFT")
+
         new_joins.append(join)
         outer_scope.add_source(join.alias_or_name, inner_scope.sources[join.alias_or_name])
 

tests/fixtures/optimizer/merge_subqueries.sql

@@ -511,3 +511,86 @@ with cte as (
 )
 select cte.mult from cte;
 SELECT x.a * x.b AS mult FROM x AS x;
+
+# title: replace INNER JOIN with LEFT JOIN when it exists in a subquery used as the RHS of a LEFT JOIN
+WITH 
+t0 AS (
+  SELECT 5 as id
+),
+t1 AS (
+  SELECT 1 AS id, 'US' AS cid
+),
+t2 AS (
+  SELECT 1 AS id, 'US' AS cid
+)
+SELECT
+  t0.id,
+  t3.cid AS cid
+FROM t0
+LEFT JOIN (
+  SELECT
+    t1.id,
+    t2.cid
+  FROM t1
+  INNER JOIN t2
+    ON t1.cid = t2.cid
+) AS t3
+  ON t0.id = t3.id;
+WITH t0 AS (SELECT 5 AS id), t1 AS (SELECT 1 AS id, 'US' AS cid), t2 AS (SELECT 1 AS id, 'US' AS cid) SELECT t0.id AS id, t2.cid AS cid FROM t0 AS t0 LEFT JOIN t1 AS t1 ON t0.id = t1.id LEFT JOIN t2 AS t2 ON t1.cid = t2.cid;
+
+# title: preserve INNER JOIN when it exists in a subquery used as the RHS of a RIGHT JOIN
+WITH 
+t0 AS (
+  SELECT 5 as id
+),
+t1 AS (
+  SELECT 1 AS id, 'US' AS cid
+),
+t2 AS (
+  SELECT 1 AS id, 'US' AS cid
+)
+SELECT
+  t0.id,
+  t3.cid AS cid
+FROM t0
+RIGHT JOIN (
+  SELECT
+    t1.id,
+    t2.cid
+  FROM t1
+  INNER JOIN t2
+    ON t1.cid = t2.cid
+) AS t3
+  ON t0.id = t3.id;
+WITH t0 AS (SELECT 5 AS id), t1 AS (SELECT 1 AS id, 'US' AS cid), t2 AS (SELECT 1 AS id, 'US' AS cid) SELECT t0.id AS id, t2.cid AS cid FROM t0 AS t0 RIGHT JOIN t1 AS t1 ON t0.id = t1.id INNER JOIN t2 AS t2 ON t1.cid = t2.cid;
+
+# title: replace multiple INNER JOINS with LEFT JOINS when they exist in a subquery used as the RHS of a LEFT JOIN
+WITH
+t0 AS (
+  SELECT 5 as id
+),
+t1 AS (
+  SELECT 1 AS id, 'US' AS cid
+),
+t2 AS (
+  SELECT 1 AS id, 'US' AS cid
+),
+t3 AS (
+  SELECT 1 AS id, 'CA' AS cid
+)
+SELECT
+  t0.id,
+  t4.cid AS cid
+FROM t0
+LEFT JOIN (
+  SELECT
+    t1.id,
+    t3.cid
+  FROM t1
+  INNER JOIN t2
+    ON t1.cid = t2.cid
+  INNER JOIN t3
+    ON t2.id = t3.id
+) AS t4
+  ON t0.id = t4.id;
+WITH t0 AS (SELECT 5 AS id), t1 AS (SELECT 1 AS id, 'US' AS cid), t2 AS (SELECT 1 AS id, 'US' AS cid), t3 AS (SELECT 1 AS id, 'CA' AS cid) SELECT t0.id AS id, t3.cid AS cid FROM t0 AS t0 LEFT JOIN t1 AS t1 ON t0.id = t1.id LEFT JOIN t2 AS t2 ON t1.cid = t2.cid LEFT JOIN t3 AS t3 ON t2.id = t3.id;