thread_group::size complexity

[viboes]

The current implementation could have O(n) complexity due to the use of std::list. Even if we would have thread_group with no too many threads, it sounds extraneous to me to have a function size without constant complexity. As your implementation do not needs splice, what do you think about having a thread_group::size with constant complexity by adding an internal counter for example?

[viboes]

As discussed on the Boost ML, the number of threads in a group should be reduced, so the the cost of a linear size should be negligible. In addition the modification will add more complexity to the usual case.

So I don't think anymore this feature is interesting.