Merge pull request #68099 from GrahamCampbell/patch-2

Do not apply redundant sorting removal when there's an offset

src/Processors/QueryPlan/Optimizations/removeRedundantSorting.cpp

@@ -8,6 +8,7 @@
 #include <Processors/QueryPlan/JoinStep.h>
 #include <Processors/QueryPlan/LimitByStep.h>
 #include <Processors/QueryPlan/LimitStep.h>
+#include <Processors/QueryPlan/OffsetStep.h>
 #include <Processors/QueryPlan/Optimizations/Optimizations.h>
 #include <Processors/QueryPlan/QueryPlanVisitor.h>
 #include <Processors/QueryPlan/ReadFromMergeTree.h>
@@ -59,9 +60,10 @@ public:
 
         if (typeid_cast<LimitStep *>(current_step)
             || typeid_cast<LimitByStep *>(current_step) /// (1) if there are LIMITs on top of ORDER BY, the ORDER BY is non-removable
-            || typeid_cast<FillingStep *>(current_step) /// (2) if ORDER BY is with FILL WITH, it is non-removable
-            || typeid_cast<SortingStep *>(current_step) /// (3) ORDER BY will change order of previous sorting
-            || typeid_cast<AggregatingStep *>(current_step)) /// (4) aggregation change order
+            || typeid_cast<OffsetStep *>(current_step) /// (2) OFFSET on top of ORDER BY, the ORDER BY is non-removable
+            || typeid_cast<FillingStep *>(current_step) /// (3) if ORDER BY is with FILL WITH, it is non-removable
+            || typeid_cast<SortingStep *>(current_step) /// (4) ORDER BY will change order of previous sorting
+            || typeid_cast<AggregatingStep *>(current_step)) /// (5) aggregation change order
         {
             logStep("nodes_affect_order/push", current_node);
             nodes_affect_order.push_back(current_node);

tests/queries/0_stateless/02496_remove_redundant_sorting.reference

@@ -465,6 +465,37 @@ Expression ((Projection + Before ORDER BY))
             ReadFromStorage (SystemOne)
 -- execute
 Float64	9007199254740994
+-- presence of an inner OFFSET retains the ORDER BY
+-- query
+WITH
+  t1 AS (
+    SELECT a, b
+    FROM
+      VALUES (
+        'b UInt32, a Int32',
+        (1, 1),
+        (2, 0)
+      )
+  )
+SELECT
+  SUM(a)
+FROM (
+  SELECT a, b
+  FROM t1
+  ORDER BY 1 DESC, 2
+  OFFSET 1
+) t2
+-- explain
+Expression ((Projection + Before ORDER BY))
+  Aggregating
+    Expression (Before GROUP BY)
+      Offset
+        Expression (Projection)
+          Sorting (Sorting for ORDER BY)
+            Expression ((Before ORDER BY + (Projection + Before ORDER BY)))
+              ReadFromStorage (Values)
+-- execute
+0
 -- disable common optimization to avoid functions to be lifted up (liftUpFunctions optimization), needed for testing with stateful function
 -- neighbor() as stateful function prevents removing inner ORDER BY since its result depends on order
 -- query

tests/queries/0_stateless/02496_remove_redundant_sorting.sh

@@ -302,6 +302,27 @@ FROM
 )"
 run_query "$query"
 
+echo "-- presence of an inner OFFSET retains the ORDER BY"
+query="WITH
+  t1 AS (
+    SELECT a, b
+    FROM
+      VALUES (
+        'b UInt32, a Int32',
+        (1, 1),
+        (2, 0)
+      )
+  )
+SELECT
+  SUM(a)
+FROM (
+  SELECT a, b
+  FROM t1
+  ORDER BY 1 DESC, 2
+  OFFSET 1
+) t2"
+run_query "$query"
+
 echo "-- disable common optimization to avoid functions to be lifted up (liftUpFunctions optimization), needed for testing with stateful function"
 ENABLE_OPTIMIZATION="SET query_plan_enable_optimizations=0;$ENABLE_OPTIMIZATION"
 echo "-- neighbor() as stateful function prevents removing inner ORDER BY since its result depends on order"

tests/queries/0_stateless/02496_remove_redundant_sorting_analyzer.reference

@@ -464,6 +464,36 @@ Expression ((Project names + Projection))
             ReadFromStorage (SystemOne)
 -- execute
 Float64	9007199254740994
+-- presence of an inner OFFSET retains the ORDER BY
+-- query
+WITH
+  t1 AS (
+    SELECT a, b
+    FROM
+      VALUES (
+        'b UInt32, a Int32',
+        (1, 1),
+        (2, 0)
+      )
+  )
+SELECT
+  SUM(a)
+FROM (
+  SELECT a, b
+  FROM t1
+  ORDER BY 1 DESC, 2
+  OFFSET 1
+) t2
+-- explain
+Expression ((Project names + Projection))
+  Aggregating
+    Expression ((Before GROUP BY + (Change column names to column identifiers + Project names)))
+      Offset
+        Sorting (Sorting for ORDER BY)
+          Expression ((Before ORDER BY + (Projection + (Change column names to column identifiers + (Project names + (Projection + Change column names to column identifiers))))))
+            ReadFromStorage (Values)
+-- execute
+0
 -- disable common optimization to avoid functions to be lifted up (liftUpFunctions optimization), needed for testing with stateful function
 -- neighbor() as stateful function prevents removing inner ORDER BY since its result depends on order
 -- query