Anyway, could you send the resulting files to make easier the review? The files are attached. These should be boost/thread/posix/once.hpp and libs/thread/src/posix/once.cpp, other files are intact.
Andrey, the provided patch goes too far for what I was expecting. You have made a lot of changes, C++11 interface has not been preserved, and I'm not sure the algorithm is the same. What I was asking for is just a patch to ensure that
typedef unsigned long uintmax_atomic_t;
is atomic in all platforms.
i suppose this is the case for most platforms. however according to the c++11 standard only atomic_flag is guaranteed to be lockfree, unfortunately we cannot guarantee this for boost.atomic (without some additional platform-specific code).
For the comments about the Jamfile, I deduce that the patch need to link with Boost.Atomic. I expected a patch that didn't need to link to Boost.Atomic. I should have missed something.
again, this is something we cannot guarantee for boost.atomic. during the review the original author wanted to provide an implementation of atomic<> that is address-free that should have become part of boost.interprocess ... well, that was before he disappeared. that implementation could have been header-only, as every atomic<> would have been associated with a spin-lock (or every non-lockfree atomic<> if we decide not to care about binary compatibility) but i wonder: is it a problem to link boost.thread with boost.atomic? cheers, tim