Merge pull request ClickHouse#78565 from bigo-sg/opt_count_if

Trivial optimization: do not rewrite count(if()) to countIf if CAST is required

Author: Algunenano

src/Analyzer/Passes/RewriteAggregateFunctionWithIfPass.cpp

@@ -72,8 +72,8 @@ class RewriteAggregateFunctionWithIfVisitor : public InDepthQueryTreeVisitorWith
 
                 QueryTreeNodes new_arguments{2};
 
-                /// We need to preserve the output type from if()
-                if (if_arguments_nodes[1]->getResultType()->getName() != if_node->getResultType()->getName())
+                /// We need to preserve the output type from if(). Notice that the return type of count() is the same either way
+                if (if_arguments_nodes[1]->getResultType()->getName() != if_node->getResultType()->getName() && lower_name != "count")
                     new_arguments[0] = createCastFunction(std::move(if_arguments_nodes[1]), if_node->getResultType(), getContext());
                 else
                     new_arguments[0] = std::move(if_arguments_nodes[1]);
@@ -99,7 +99,8 @@ class RewriteAggregateFunctionWithIfVisitor : public InDepthQueryTreeVisitorWith
 
                 QueryTreeNodes new_arguments{2};
 
-                if (if_arguments_nodes[2]->getResultType()->getName() != if_node->getResultType()->getName())
+                /// We need to preserve the output type from if(). Notice that the return type of count() is the same either way
+                if (if_arguments_nodes[2]->getResultType()->getName() != if_node->getResultType()->getName() && lower_name != "count")
                     new_arguments[0] = createCastFunction(std::move(if_arguments_nodes[2]), if_node->getResultType(), getContext());
                 else
                     new_arguments[0] = std::move(if_arguments_nodes[2]);

tests/performance/rewrite_aggregate_if.xml

@@ -0,0 +1,7 @@
+<test>
+    <query>select count(if(rand() % 2 = 0, 1, null)) from numbers(100000000) settings optimize_rewrite_aggregate_function_with_if = true</query>
+    <query>select count(if(rand() % 2 = 0, null, 1)) from numbers(100000000) settings optimize_rewrite_aggregate_function_with_if = true</query>
+
+    <query>select count(if(rand() % 2 = 0, toNullable(1), null)) from numbers(100000000) settings optimize_rewrite_aggregate_function_with_if = true</query>
+    <query>select count(if(rand() % 2 = 0, null, toNullable(1))) from numbers(100000000) settings optimize_rewrite_aggregate_function_with_if = true</query>
+</test>