Boost C++ Libraries: Ticket #8593: enable std::atomic with clang / libc++

attachment set

reakinator@… — Sun, 19 May 2013 23:51:28 GMT

attachment → lockfree_atomic_libcpp.patch

lockfree/detail/atomic.hpp patch enabling <atomic> for libc++

timblechmann — Mon, 20 May 2013 08:22:57 GMT

unfortunately this does not work on linux:

/usr/include/c++/v1/atomic:623:58: error: no viable conversion from 'boost::lockfree::detail::tagged_index' to '_Atomic(boost::lockfree::detail::tagged_index)'
    _LIBCPP_CONSTEXPR __atomic_base(_Tp __d) _NOEXCEPT : __a_(__d) {}
                                                         ^    ~~~
/usr/include/c++/v1/atomic:727:51: note: in instantiation of member function 'std::__1::__atomic_base<boost::lockfree::detail::tagged_index, false>::__atomic_base' requested here
    _LIBCPP_CONSTEXPR atomic(_Tp __d) _NOEXCEPT : __base(__d) {}
                                                  ^
../../../boost/lockfree/detail/freelist.hpp:404:9: note: in instantiation of member function 'std::__1::atomic<boost::lockfree::detail::tagged_index>::atomic' requested here
        pool_(tagged_index(static_cast<index_t>(count), 0))
        ^

reakinator@… — Mon, 20 May 2013 16:59:29 GMT

Agh, as luck would have it I'm only using lockfree::spsc_queue, which doesn't have this problem (both lockfree stack and queue do).

Can you tell why the conversion can't be made? libc++ is known to be pretty strict with standards (more than gcc), but hopefully we can narrow down what needs to be fixed.

reakinator@… — Mon, 20 May 2013 19:58:40 GMT

I found the following thread when doing some research:

http://clang-developers.42468.n3.nabble.com/RFC-atomic-support-for-gcc-4-7-compatibility-td3900609.html

My hunch is that point 2 on that list ("relax the requirement to the type being either _Atomic or trivially copyable") was not implemented, or tagged_ptr is not trivially copyable.

Tue, 21 May 2013 04:15:25 GMT

Hm, appears it is a clang bug, H.Hinnant submitted a patch a few hours after I reported it...

http://lists.cs.uiuc.edu/pipermail/cfe-dev/2013-May/029771.html

I guess this has to go on the back burner for now... when there is a way to verify we are using the updated version of libc++, I'll update.

timblechmann — Tue, 21 May 2013 07:36:04 GMT

gcc used to have the same problem ... hopefully this patch makes it into a libc++ release in the near future ... until then, boost.atomic is the way to go ...