Clark S. Cox III wrote:

A couple of questions:

Is there any rational behind the fact that boost::detail::atomic_count doesn't have a default constructor?

The interface is specified in a way that makes 'int' or 'long' meet its requirements, so that one can typedef atomic_count to int/long in ST builds.

For what reason is atomic_count an implementation detail, and not available for general use? (i.e. why is it in boost::detail, and not boost::)

It wasn't considered ready for public consumption. An atomic operations library is hard to get right, as both interface and implementation, which is why we still don't have one in Boost. I think that we've now reached a degree of understanding that will make it possible to specify and implement one. So we should have one by 1.36, maybe even 1.35 if we hurry up. :-)