Well, we do check all this via configure... if the platforms supports 64 "native" atomics then we could use those; if not, we could use the portable versions as a backup (or simply return NOTIMPL).
I don't see how this is different from how we handle atomics currently, but I could be mistaken. > On Sep 17, 2018, at 10:03 AM, Nick Kew <n...@apache.org> wrote: > > Apologies to Jim. Sent this to him, meant for the dev list of course. > >>> On 17 Sep 2018, at 14:18, Jim Jagielski <j...@jagunet.com> wrote: >>> >>> FYI: Both clang and GCC support both __sync and __atomic which support >>> 64bit ints. We could add that functionality to APR... >> >> We could indeed. >> >> But does that not potentially leave a nasty gotcha? Where a developer uses >> it and expects >> atomic operations, and their application is subsequently built on a platform >> that >> doesn't support those qualifiers. >> >> With the obvious fallback it breaks silently. With a more >> sophisticated/heavyweight >> fallback, we should consider whether it's maintainable or likely to fall >> into disrepair. >> >> I wouldn't want to stop you, but it needs some thought. >> >> -- >> Nick Kew >
