Peter Dimov wrote:
Use aligned_storage<> and placement new.
Sorry, I didn't know what aligned_storage was for, boost is really huge. I've seen it uses a union using alignment information from other boost tools. Doing an placement new invalidates automatic destructor, so it sounds to me like the famous ugly pimpl hack to avoid dynamic memory and header dependences, based on max_align. Anyway, it is true that dynamic memory is the best way to make thread lifetime independent from boost::thread object's lifetime, so maybe in those special systems, a singleton memory pool should be used for this, if "new" function is not available. What I don't know if that singleton lifetime can lead to problems because I don't know if threads must exit before static object destructors are called, but this is another topic. I think embedded C++ should be taken in care because C++ is used a lot in this area (for example, eCos embedded OS is written in C++ with a C wrapper API). The C++ phylosophy ("you don't pay for what you don't use") is very nice in these systems, if you know what is happening behind the scenes with C++ (inheritance, constructor/destructors, etc...). It would be nice to have some brainstorming regarding boost threads if it is possible. I don't know if C++0x wants only to include basic functionality in threads but I think developers are waiting for some more complex utilities, like thread pools. Regards, Ion