Update changelog

for edges with the split flag false and in_backward_direction true
if in_forward_direction is also true backward segments can be incorrectly duplicated

CHANGELOG.md

@@ -1,3 +1,7 @@
+# 5.6.5
+  - Changes from 5.6.4
+    - Bugfixes
+      - Fix #3873 excessive usage of disk and memory during osrm-extract because edge deduplication was broken
 
 # 5.6.4
   - Changes from 5.6.3

src/extractor/extraction_containers.cpp

@@ -397,21 +397,24 @@ void ExtractionContainers::PrepareEdges(ScriptingEnvironment &scripting_environm
             BOOST_ASSERT(edge_iterator->source_coordinate.lon !=
                          util::FixedLongitude{std::numeric_limits<std::int32_t>::min()});
 
-            const util::Coordinate target_coord{node_iterator->lon, node_iterator->lat};
-            const double distance = util::coordinate_calculation::greatCircleDistance(
-                edge_iterator->source_coordinate, target_coord);
+            util::Coordinate source_coord(edge_iterator->source_coordinate);
+            util::Coordinate target_coord{node_iterator->lon, node_iterator->lat};
 
-            auto weight = edge_iterator->weight_data(distance);
-            auto duration = edge_iterator->duration_data(distance);
+            // flip source and target coordinates if segment is in backward direction only
+            if (!edge_iterator->result.forward && edge_iterator->result.backward)
+                std::swap(source_coord, target_coord);
 
-            ExtractionSegment extracted_segment(
-                edge_iterator->source_coordinate, target_coord, distance, weight, duration);
-            scripting_environment.ProcessSegment(extracted_segment);
+            const auto distance =
+                util::coordinate_calculation::greatCircleDistance(source_coord, target_coord);
+            const auto weight = edge_iterator->weight_data(distance);
+            const auto duration = edge_iterator->duration_data(distance);
+
+            ExtractionSegment segment(source_coord, target_coord, distance, weight, duration);
+            scripting_environment.ProcessSegment(segment);
 
             auto &edge = edge_iterator->result;
-            edge.weight =
-                std::max<EdgeWeight>(1, std::round(extracted_segment.weight * weight_multiplier));
-            edge.duration = std::max<EdgeWeight>(1, std::round(extracted_segment.duration * 10.));
+            edge.weight = std::max<EdgeWeight>(1, std::round(segment.weight * weight_multiplier));
+            edge.duration = std::max<EdgeWeight>(1, std::round(segment.duration * 10.));
 
             // assign new node id
             auto id_iter = external_to_internal_node_id_map.find(node_iterator->node_id);

src/extractor/extractor_callbacks.cpp

@@ -330,7 +330,7 @@ void ExtractorCallbacks::ProcessWay(const osmium::Way &input_way, const Extracti
                              (turn_lane_id_forward != turn_lane_id_backward));
 
     if (in_forward_direction)
-    {
+    { // add (forward) segments or (forward,backward) for non-split edges in backward direction
         util::for_each_pair(
             nodes.cbegin(),
             nodes.cend(),
@@ -355,8 +355,8 @@ void ExtractorCallbacks::ProcessWay(const osmium::Way &input_way, const Extracti
             });
     }
 
-    if (in_backward_direction || split_edge)
-    {
+    if (in_backward_direction && (!in_forward_direction || split_edge))
+    { // add (backward) segments for split edges or not in forward direction
         util::for_each_pair(
             nodes.cbegin(),
             nodes.cend(),