So, do we yank read_write_mutex from 1.33.1 and, err, start over?
What about avoiding policies, or making independent read-write locks, one for reader priority, other for writers priority and other for fifo priority? Or a templatized priority policy. Or just pick a default implementation like pthreads and don't guarantee anything. There are simple implementations of read-write locks available in books that we can use without thinking very much, for example, I can remember one from "Programming with posix threads" and another one from "Unix network programming vol.2", one with reader prioriy and another with writer priority. All these where made using posix conditions and mutexes, and the implementation is quite simple. ACE has its own implementation if I'm not wrong. I think that runtime policy has an overhead and complexity. But this would require changing, Boost.Threads read-write lock interface, something that I don't know we like to do. On the other hand, I don't know if reader/writer priority is very important, since I think posix does not define anything about this. Regards, Ion