Merge pull request #52438 from ClickHouse/fix_remove_redundant_distinct_with_view

Fix: remove redundant distinct with views

src/Interpreters/ActionsDAG.cpp

@@ -2515,11 +2515,21 @@ FindOriginalNodeForOutputName::FindOriginalNodeForOutputName(const ActionsDAGPtr
         /// find input node which refers to the output node
         /// consider only aliases on the path
         const auto * node = output_node;
-        while (node && node->type == ActionsDAG::ActionType::ALIAS)
+        while (node)
         {
-            /// alias has only one child
+            if (node->type == ActionsDAG::ActionType::ALIAS)
-            chassert(node->children.size() == 1);
+            {
-            node = node->children.front();
+                node = node->children.front();
+            }
+            /// materiailze() function can occur when dealing with views
+            /// TODO: not sure if it should be done here, looks too generic place
+            else if (node->type == ActionsDAG::ActionType::FUNCTION && node->function_base->getName() == "materialize")
+            {
+                chassert(node->children.size() == 1);
+                node = node->children.front();
+            }
+            else
+                break;
         }
         if (node && node->type == ActionsDAG::ActionType::INPUT)
             index.emplace(output_node->result_name, node);

tests/queries/0_stateless/02810_fix_remove_dedundant_distinct_view.reference

@@ -0,0 +1,21 @@
+-- { echoOn }
+set query_plan_remove_redundant_distinct=1;
+-- DISTINCT has to be removed since the view already has DISTINCT on the same column
+SELECT count()
+FROM
+(
+    EXPLAIN SELECT DISTINCT x FROM tab_v
+)
+WHERE explain ILIKE '%distinct%';
+2
+SELECT DISTINCT x FROM tab_v ORDER BY x;
+1
+2
+-- explicitly checking that materialize() doesn't affect the result, - redundant DISTINCT is still removed
+SELECT count()
+FROM
+(
+    EXPLAIN SELECT DISTINCT x FROM (SELECT materialize(x) as x FROM (select DISTINCT x from tab))
+)
+WHERE explain ILIKE '%distinct%';
+2

tests/queries/0_stateless/02810_fix_remove_dedundant_distinct_view.sql

@@ -0,0 +1,28 @@
+drop table if exists tab_v;
+drop table if exists tab;
+create table tab (x UInt64, y UInt64) engine MergeTree() order by (x, y);
+insert into tab values(1, 1);
+insert into tab values(1, 2);
+insert into tab values(2, 1);
+
+create view tab_v as select distinct(x) from tab;
+
+-- { echoOn }
+set query_plan_remove_redundant_distinct=1;
+-- DISTINCT has to be removed since the view already has DISTINCT on the same column
+SELECT count()
+FROM
+(
+    EXPLAIN SELECT DISTINCT x FROM tab_v
+)
+WHERE explain ILIKE '%distinct%';
+
+SELECT DISTINCT x FROM tab_v ORDER BY x;
+
+-- explicitly checking that materialize() doesn't affect the result, - redundant DISTINCT is still removed
+SELECT count()
+FROM
+(
+    EXPLAIN SELECT DISTINCT x FROM (SELECT materialize(x) as x FROM (select DISTINCT x from tab))
+)
+WHERE explain ILIKE '%distinct%';