remove-redundant-sorts, initial

arangod/Aql/AstNode.cpp

@@ -814,6 +814,31 @@ void AstNode::append (triagens::basics::StringBuffer* buffer) const {
   }
 }
 
+////////////////////////////////////////////////////////////////////////////////
+/// @brief stringify a node
+/// note that currently stringification is only supported for certain node types
+////////////////////////////////////////////////////////////////////////////////
+
+std::string AstNode::stringify () const {
+  switch (type) {
+    case NODE_TYPE_ATTRIBUTE_ACCESS: {
+      auto member = getMember(0);
+      return member->stringify() + std::string(".") + getStringValue();
+    }
+
+    case NODE_TYPE_REFERENCE: {
+      auto variable = static_cast<Variable*>(getData());
+      TRI_ASSERT(variable != nullptr);
+      return variable->name;
+    }
+
+    default: {
+    }
+  }
+
+  THROW_ARANGO_EXCEPTION_MESSAGE(TRI_ERROR_INTERNAL, "stringification not supported for node type");
+}
+
 // -----------------------------------------------------------------------------
 // --SECTION--                                                   private methods
 // -----------------------------------------------------------------------------

arangod/Aql/AstNode.h

@@ -493,6 +493,13 @@ namespace triagens {
 
         void append (triagens::basics::StringBuffer*) const;
 
+////////////////////////////////////////////////////////////////////////////////
+/// @brief stringify a node
+/// note that currently stringification is only supported for certain node types
+////////////////////////////////////////////////////////////////////////////////
+
+        std::string stringify () const;
+
 // -----------------------------------------------------------------------------
 // --SECTION--                                                   private methods
 // -----------------------------------------------------------------------------

arangod/Aql/ExecutionNode.cpp

@@ -27,6 +27,7 @@
 
 #include "Aql/ExecutionNode.h"
 #include "Aql/Collection.h"
+#include "Aql/ExecutionPlan.h"
 #include "Aql/WalkerWorker.h"
 #include "Aql/Ast.h"
 
@@ -70,7 +71,7 @@ std::unordered_map<int, std::string const> const ExecutionNode::TypeNames{
 /// @brief returns the type name of the node
 ////////////////////////////////////////////////////////////////////////////////
 
-const std::string& ExecutionNode::getTypeString () const {
+std::string const& ExecutionNode::getTypeString () const {
   auto it = TypeNames.find(static_cast<int>(getType()));
   if (it != TypeNames.end()) {
     return (*it).second;
@@ -918,6 +919,44 @@ std::vector<std::pair<ExecutionNode*, bool>> SortNode::getCalcNodePairs ()
   return findExp._myVars;
 }
 
+////////////////////////////////////////////////////////////////////////////////
+/// @brief returns all sort information 
+////////////////////////////////////////////////////////////////////////////////
+
+SortInformation SortNode::getSortInformation (ExecutionPlan* plan) const {
+  SortInformation result;
+
+  auto elements = getElements();
+  for (auto it = elements.begin(); it != elements.end(); ++it) {
+    auto variable = (*it).first;
+    TRI_ASSERT(variable != nullptr);
+    auto setter = plan->getVarSetBy(variable->id);
+
+    if (setter == nullptr) {
+      result.isValid = false;
+      break;
+    }
+
+    if (setter->getType() == ExecutionNode::CALCULATION) {
+      // variable introduced by a calculation
+      auto expression = static_cast<CalculationNode*>(setter)->expression();
+
+      if (! expression->isAttributeAccess() &&
+          ! expression->isReference()) {
+        result.isComplex = true;
+        break;
+      }
+
+      result.criteria.emplace_back(std::make_tuple(setter, expression->stringify(), (*it).second));
+    }
+    else {
+      result.criteria.emplace_back(std::make_tuple(setter, variable->name, (*it).second));
+    }
+  }
+
+  return result;
+}
+
 // -----------------------------------------------------------------------------
 // --SECTION--                                          methods of AggregateNode
 // -----------------------------------------------------------------------------

arangod/Aql/ExecutionNode.h

@@ -154,7 +154,7 @@ namespace triagens {
 /// @brief return the type name of the node
 ////////////////////////////////////////////////////////////////////////////////
 
-        const std::string &getTypeString () const;
+        std::string const& getTypeString () const;
 
 ////////////////////////////////////////////////////////////////////////////////
 /// @brief checks whether we know a type of this kind; throws exception if not.
@@ -1413,6 +1413,20 @@ namespace triagens {
 
     };
 
+// -----------------------------------------------------------------------------
+// --SECTION--                                            struct SortInformation
+// -----------------------------------------------------------------------------
+
+////////////////////////////////////////////////////////////////////////////////
+/// @brief this is an auxilliary struct for processed sort criteria information
+////////////////////////////////////////////////////////////////////////////////
+    
+    struct SortInformation {
+      std::vector<std::tuple<ExecutionNode const*, std::string, bool>> criteria;
+      bool isValid   = true;
+      bool isComplex = false;
+    };
+
 // -----------------------------------------------------------------------------
 // --SECTION--                                                    class SortNode
 // -----------------------------------------------------------------------------
@@ -1493,10 +1507,16 @@ namespace triagens {
 /// @brief get Variables Used Here including ASC/DESC
 ////////////////////////////////////////////////////////////////////////////////
 
-        std::vector<std::pair<Variable const*, bool>> getElements () {
+        std::vector<std::pair<Variable const*, bool>> getElements () const {
           return _elements;
         }
 
+////////////////////////////////////////////////////////////////////////////////
+/// @brief returns all sort information 
+////////////////////////////////////////////////////////////////////////////////
+
+        SortInformation getSortInformation (ExecutionPlan*) const;
+
         std::vector<std::pair<ExecutionNode*, bool>> getCalcNodePairs ();
 
 // -----------------------------------------------------------------------------

arangod/Aql/Expression.cpp

@@ -347,14 +347,11 @@ AqlValue Expression::executeSimpleExpression (AstNode const* node,
 }
 
 ////////////////////////////////////////////////////////////////////////////////
-/// @brief check whether this is a simple access to a variable with n attributes.
+/// @brief check whether this is an attribute access of any degree (e.g. a.b, 
+/// a.b.c, ...)
 ////////////////////////////////////////////////////////////////////////////////
 
-bool Expression::isMultipleAttributeAccess () const {
+bool Expression::isAttributeAccess () const {
-  if (!isSimple()) {
-    return false;
-  }
-
   auto expNode = _node;
 
   if (expNode->type != triagens::aql::NODE_TYPE_ATTRIBUTE_ACCESS) {
@@ -362,15 +359,23 @@ bool Expression::isMultipleAttributeAccess () const {
   }
 
   while (expNode->type == triagens::aql::NODE_TYPE_ATTRIBUTE_ACCESS) {
-    expNode = expNode->getMember (0);
+    expNode = expNode->getMember(0);
   }
   
   return (expNode->type == triagens::aql::NODE_TYPE_REFERENCE);
 }
 
+////////////////////////////////////////////////////////////////////////////////
+/// @brief check whether this is a reference access
+////////////////////////////////////////////////////////////////////////////////
+
+bool Expression::isReference () const {
+  return (_node->type == triagens::aql::NODE_TYPE_REFERENCE);
+}
+
 ////////////////////////////////////////////////////////////////////////////////
 /// @brief this gives you ("variable.access", "Reference")
-/// call isSimpleAccessReference in advance to enshure no exceptions.
+/// call isSimpleAccessReference in advance to ensure no exceptions.
 ////////////////////////////////////////////////////////////////////////////////
 std::pair<std::string, std::string> Expression::getMultipleAttributes() const {
   if (!isSimple()) {
@@ -411,6 +416,14 @@ std::pair<std::string, std::string> Expression::getMultipleAttributes() const {
 
 }
 
+////////////////////////////////////////////////////////////////////////////////
+/// @brief stringify an expression
+/// note that currently stringification is only supported for certain node types
+////////////////////////////////////////////////////////////////////////////////
+
+std::string Expression::stringify () const {
+  return _node->stringify();
+}
 
 // -----------------------------------------------------------------------------
 // --SECTION--                                                       END-OF-FILE

arangod/Aql/Expression.h

@@ -151,18 +151,32 @@ namespace triagens {
         }
 
 ////////////////////////////////////////////////////////////////////////////////
-/// @brief check whether this is a simple access to a variable with n attributes.
+/// @brief check whether this is an attribute access of any degree (e.g. a.b, 
+/// a.b.c, ...)
 ////////////////////////////////////////////////////////////////////////////////
 
-        bool isMultipleAttributeAccess () const;
+        bool isAttributeAccess () const;
+
+////////////////////////////////////////////////////////////////////////////////
+/// @brief check whether this is only a reference access
+////////////////////////////////////////////////////////////////////////////////
+
+        bool isReference () const;
 
 ////////////////////////////////////////////////////////////////////////////////
 /// @brief this gives you ("variable.access", "Reference")
-/// call isSimpleAccessReference in advance to enshure no exceptions.
+/// call isSimpleAccessReference in advance to ensure no exceptions.
 ////////////////////////////////////////////////////////////////////////////////
 
         std::pair<std::string, std::string> getMultipleAttributes() const;
 
+////////////////////////////////////////////////////////////////////////////////
+/// @brief stringify an expression
+/// note that currently stringification is only supported for certain node types
+////////////////////////////////////////////////////////////////////////////////
+
+        std::string stringify () const;
+
 // -----------------------------------------------------------------------------
 // --SECTION--                                                 private functions
 // -----------------------------------------------------------------------------

arangod/Aql/Optimizer.cpp

@@ -170,10 +170,14 @@ void Optimizer::setupRules () {
   TRI_ASSERT(_rules.empty());
 
   // List all the rules in the system here:
+  // lower level values mean earlier rule execution
+  // if two rules have the same level value, they will be executed in declaration order
 
   // try to find sort blocks which are superseeded by indexes
   registerRule("use-index-for-sort", useIndexForSort, 2000);
-
+  
+  // remove redundant sort blocks
+  registerRule("remove-redundant-sorts", removeRedundantSorts, 1000);
 
   // try to find a filter after an enumerate collection and find an index . . . 
   registerRule("use-index-range", useIndexRange, 999);

arangod/Aql/OptimizerRules.cpp

@@ -49,6 +49,38 @@ int triagens::aql::dummyRule (Optimizer*,
   return TRI_ERROR_NO_ERROR;
 }
 
+////////////////////////////////////////////////////////////////////////////////
+/// @brief remove redundant sorts
+/// this rule modifies the plan in place:
+/// - sorts that are covered by earlier sorts will be removed
+////////////////////////////////////////////////////////////////////////////////
+
+int triagens::aql::removeRedundantSorts (Optimizer* opt, 
+                                         ExecutionPlan* plan, 
+                                         int level, 
+                                         Optimizer::PlanList& out) {
+  // should we enter subqueries??
+  std::vector<ExecutionNode*> nodes = plan->findNodesOfType(triagens::aql::ExecutionNode::SORT, true);
+  
+  for (auto n : nodes) {
+    auto sortInfo = static_cast<SortNode*>(n)->getSortInformation(plan);
+
+    if (sortInfo.isValid) {
+      /*
+      TODO: finalize
+      std::cout << "FOUND SORT:\n";
+      for (auto it = sortInfo.criteria.begin(); it != sortInfo.criteria.end(); ++it) {
+        std::cout << "- " << std::get<1>(*it) << ", " << std::get<2>(*it) << "\n";
+      }
+      */
+    }
+  }
+  
+  out.push_back(plan, level);
+
+  return TRI_ERROR_NO_ERROR;
+}
+
 ////////////////////////////////////////////////////////////////////////////////
 /// @brief remove all unnecessary filters
 /// this rule modifies the plan in place:
@@ -61,6 +93,7 @@ int triagens::aql::removeUnnecessaryFiltersRule (Optimizer* opt,
                                                  int level,
                                                  Optimizer::PlanList& out) {
   std::unordered_set<ExecutionNode*> toUnlink;
+  // should we enter subqueries??
   std::vector<ExecutionNode*> nodes = plan->findNodesOfType(triagens::aql::ExecutionNode::FILTER, true);
   
   for (auto n : nodes) {
@@ -594,11 +627,11 @@ public:
       d->ASC = sortParams[n].second;
       d->calculationNodeID = sortParams[n].first->id();
 
-      if (sortParams[n].first->getType() ==  EN::CALCULATION) {
+      if (sortParams[n].first->getType() == EN::CALCULATION) {
         auto cn = static_cast<triagens::aql::CalculationNode*>(sortParams[n].first);
         auto oneSortExpression = cn->expression();
         
-        if (oneSortExpression->isMultipleAttributeAccess()) {
+        if (oneSortExpression->isAttributeAccess()) {
           auto simpleExpression = oneSortExpression->getMultipleAttributes();
           d->variableName = simpleExpression.first;
           d->attributevec = simpleExpression.second;
@@ -813,7 +846,6 @@ int triagens::aql::useIndexForSort (Optimizer* opt,
                                     ExecutionPlan* plan,
                                     int level,
                                     Optimizer::PlanList& out) {
-  std::cout << plan->toJson(TRI_UNKNOWN_MEM_ZONE, false).toString() << "\n";
   std::vector<ExecutionNode*> nodes
     = plan->findNodesOfType(triagens::aql::ExecutionNode::SORT, true);
   for (auto n : nodes) {

arangod/Aql/OptimizerRules.h

@@ -46,6 +46,14 @@ namespace triagens {
 
     int dummyRule (Optimizer*, ExecutionPlan*, int level, Optimizer::PlanList&);
 
+////////////////////////////////////////////////////////////////////////////////
+/// @brief remove redundant sorts
+/// this rule modifies the plan in place:
+/// - sorts that are covered by earlier sorts will be removed
+////////////////////////////////////////////////////////////////////////////////
+
+    int removeRedundantSorts (Optimizer*, ExecutionPlan*, int level, Optimizer::PlanList&);
+
 ////////////////////////////////////////////////////////////////////////////////
 /// @brief remove all unnecessary filters
 /// this rule modifies the plan in place: