On 1 August 2013 15:09, Paul Martz <[email protected]> wrote: > You can pass references to shared_ptr objects. You only need to copy them > for long-term storage, which is typically not a performance critical > operation in a scene graph.
It occurred to me that a reference to a shared_ptr<> is actually a pointer to a pointer, and depending upon the implementation details of shared_ptr<> it could well be a pointer to pointer to pointer just to get access to the object and it's methods. So efficiency wise I don't think a returning a shared_ptr<>& is going to be as efficient as just passing back a C* like the OSG does at present, if fact I think there is probably quite a big difference in performance. And as Mathias mentions that's without considering the extra memory allocations that associated with shared_ptr<> over a intrusive ref_ptr<> design. My conclusion would be the shared_ptr<> is clever, but it's ability to be a general purpose smart pointer impacts the performance of it usage over intrusive ref_ptr<>. For a scene graphs with a consistent class hierachy and with this ability to provide intrusive ref/unref() makes intrusive smart pointers a better performing and lighter weight solution. Like Mathias also suggested, using of std::atomic looks to be very useful and will simplify codes. Same applies to other std::thread related functionality. Robert. _______________________________________________ osg-users mailing list [email protected] http://lists.openscenegraph.org/listinfo.cgi/osg-users-openscenegraph.org
