Hello, i just uploaded a new version of Join library to Vault/ Concurrent programming. A brief summary of Join library: an asynchronous message coordination and concurrency library based on JoCaml and Cw. It provides a high level api of asynchronous ports, synchronous ports, chords and joints which support multithreaded applications and orchestration of asynchronous messages. There are some major changes for this new version: 1. originally the library is structured as 2 layers: a port api at bottom and a function api at top. In this new version, function level api is dropped to simplify the interface and implementation. There are some benefits coming with it: 1> the signature of chord's function can be directly mapped from the types of its ports, there is no need for argument wrappers (async_o, synch_o) anymore. so a thread safe message queue can be defined as: template <typename T> class msg_que : public joint { public: async<T> send; synch<T,void> recv; msg_que() { chord(recv, send, &msg_que::proc); } T proc(void_t, T s) { return s; } }; 2> we can use boost::lambda or phoenix to define chord's function; so the above class becomes: template <typename T> class msg_que { public: async<T> send; synch<T,void> recv; msg_que() { joins().chord(recv, send, _2); } }; 2. apply "pimpl" idiom to major port classes and joint class. the benefits of this: 1> port/joint classes play well with standard c++ containers and other libraries 2> better control of lifetime of port/joint objects 3> we can do JoCaml (functional) style "Join" programming, instead of Cw's class based design. For example the msg_que can be created using "factory" function as in JoCaml: template <typename T> tuple<async<T>, synch<T,void> > create_msg_que() { async<T> send; synch<T,void> recv; joins().chord(recv,send,_2); return make_tuple(send,recv); } So the code which uses this "factory" function look like this: ... async<int> send; synch<int,void> recv; tie(send,recv) = create_msg_que<int>(); //spawn 2 threads, thread1 creates data and send it, thread2 receives and use data spawn(bind(thread1_func, send)); spawn(bind(thread2_func, recv)); The documents and samples are updated. You can view them from the following link. http://channel.sourceforge.net/boost_join/libs/join/doc/boost_join_design.ht... You can download it from Vault / Concurrent Programming or the project website http://channel.sourceforge.net Comments and corrections are highly appreciated. Thanks Yigong