fix!: handle unnesting groups closes #3056 (#3058)

sqlglot/optimizer/unnest_subqueries.py

@@ -94,8 +94,20 @@ def unnest(select, parent_select, next_alias_name):
     else:
         _replace(predicate, join_key_not_null)
 
+    group = select.args.get("group")
+
+    if group:
+        if {value.this} != set(group.expressions):
+            select = (
+                exp.select(exp.column(value.alias, "_q"))
+                .from_(select.subquery("_q", copy=False), copy=False)
+                .group_by(exp.column(value.alias, "_q"), copy=False)
+            )
+    else:
+        select = select.group_by(value.this, copy=False)
+
     parent_select.join(
-        select.group_by(value.this, copy=False),
+        select,
         on=column.eq(join_key),
         join_type="LEFT",
         join_alias=alias,

tests/fixtures/optimizer/tpc-h/tpc-h.sql

@@ -1047,7 +1047,6 @@ WITH "_u_0" AS (
     "lineitem"."l_orderkey" AS "l_orderkey"
   FROM "lineitem" AS "lineitem"
   GROUP BY
-    "lineitem"."l_orderkey",
     "lineitem"."l_orderkey"
   HAVING
     SUM("lineitem"."l_quantity") > 300

tests/fixtures/optimizer/unnest_subqueries.sql

@@ -25,6 +25,7 @@ WHERE
   AND x.a > ALL (SELECT y.c FROM y WHERE y.a = x.a)
   AND x.a > (SELECT COUNT(*) as d FROM y WHERE y.a = x.a)
   AND x.a = SUM(SELECT 1)  -- invalid statement left alone
+  AND x.a IN (SELECT max(y.b) AS b FROM y GROUP BY y.a)
 ;
 SELECT
   *
@@ -155,6 +156,20 @@ LEFT JOIN (
     y.a
 ) AS _u_21
   ON _u_21._u_22 = x.a
+LEFT JOIN (
+  SELECT
+    _q.b
+  FROM (
+    SELECT
+      MAX(y.b) AS b
+    FROM y
+    GROUP BY
+      y.a
+  ) AS _q
+  GROUP BY
+    _q.b
+) AS _u_24
+  ON x.a = _u_24.b
 WHERE
   x.a = _u_0.a
   AND NOT _u_1.a IS NULL
@@ -212,6 +227,7 @@ WHERE
   AND x.a > COALESCE(_u_21.d, 0)
   AND x.a = SUM(SELECT
     1) /* invalid statement left alone */
+  AND NOT _u_24.b IS NULL
 ;
 SELECT
   CAST((