From: Steven E. Harris (seh_at_[hidden])
Date: 2005-10-03 14:27:21
"Peter Dimov" <pdimov_at_[hidden]> writes:
> One argument against your enhancements is that (to the untrained
> eye) they seem to mandate a lock-based implementation. I wanted to
> permit a lock-free implementation, if possible.
Guilty as charged. At the time I designed this class, I had never
studied lock-free programming techniques.
The focus on these "front" handles came partly from thinking about
std::stack's and std::queue's top()/front() and pop() separation. Just
as exception safety requirements prevent these two operations from
being merged, concurrency requirements prevent them from being
separated -- without some kind of a lock.
I haven't thought it through much, but I doubt the same idea
(exclusive by-reference access to the queue head) could translate to a
lock-free interface.
-- Steven E. Harris