Hi, Once again I've uploaded the latest Boost.FSM version to the Vault. This update contains minor docs changes and (hopefully) fixes multithreading issues pointed out in the discussion earlier. The major addition in this update is a lightweight call_once implementation. I've tried my best to make it portable enough and fast, but I have absolutely no experience in some platforms I had to support. So, for example, BeOS and MacOS support is implemented exclusively by docs I've found for these OSes. I have no chance neither compile nor test it on a real platform like that. That said, I would be glad if someone experienced would have a look at that code or even try to compile it in the real environment. As a side note, I was thinking if this lightweight call_once implementation and memory barrier macros it incorporates should be extracted from the Boost.FSM library to some more common place, like boost/detail. This would make it more convenient to use the code from other libs, for example, a singleton implementation library that was under discussion not long ago. Is it reasonable? Thank you. -- Best regards, Andrey mailto:andysem@mail.ru