Improved memory footprint of IFC loading by avoiding keeping in cache duplicated indices.

code/Importer/IFC/IFCLoader.cpp

@@ -435,7 +435,7 @@ bool ProcessMappedItem(const Schema_2x3::IfcMappedItem& mapped, aiNode* nd_src,
 
     msrc = m*msrc;
 
-    std::vector<unsigned int> meshes;
+    std::set<unsigned int> meshes;
     const size_t old_openings = conv.collect_openings ? conv.collect_openings->size() : 0;
     if (conv.apply_openings) {
         IfcMatrix4 minv = msrc;
@@ -550,7 +550,7 @@ void ProcessProductRepresentation(const Schema_2x3::IfcProduct& el, aiNode* nd,
 
     // extract Color from metadata, if present
     unsigned int matid = ProcessMaterials( el.GetID(), std::numeric_limits<uint32_t>::max(), conv, false);
-    std::vector<unsigned int> meshes;
+    std::set<unsigned int> meshes;
 
     // we want only one representation type, so bring them in a suitable order (i.e try those
     // that look as if we could read them quickly at first). This way of reading