on Fri Jul 29 2011, Gordon Woodhull <gordon-AT-woodhull.com> wrote:
On Jul 29, 2011, at 3:00 PM, Tim Blechmann wrote:
IMO the big question with a library like this is not so much the design as the correctness. What can we do as reviewers to verify the correctness, assuming we are not lockfree gurus ourselves? Go back to the original sources. So I did; see below.
well, there is one problem about the publications done by the lock-free gurus: they don't care about the implementation, but assume a sequencially consistent memory model, so they don't need to care about the required memory barriers or the like ...
Yes. Your rationale that the default options for atomic operations should ensure sequential consistency makes sense to me (from my limited experience).
Personally I never understood how sequential consistency could be much of a help in reasoning about multithreaded programs... but that's just me. -- Dave Abrahams BoostPro Computing http://www.boostpro.com