On Fri, 2017-01-20 at 10:10 -0500, Ted Ross wrote: > Adel, > > Yes, that would be ok. The switch to eventfd is an optimization for > performance. > > This may become a non-issue with the proactor update I mentioned on > the > other thread.
It is a non-issue for the libuv proactor. If we to do a plain-posix proactor it will use only POSIX-portable stuff - poll() and pipes. I'm still not sure if Adel's situation be better off with libuv (native kqueue, probably faster than plain-posix) or if the libuv dependency is a fatal problem and we need plain-posix. Once the proactor is in place, testing and swapping between different proactors will be trivial. Writing the plain-posix proactor should be easy - we've done epoll, libuv and IOCP now with no unpleasant surprises. > > -Ted > > On 01/20/2017 09:57 AM, Adel Boutros wrote: > > Hello Ted, > > > > > > With DISPATCH-518, you have replaced the pipe with eventfd. > > Unfortunately, there is no eventfd on Solaris. > > > > > > What do you think if I revert the fix but just for Solaris with an > > "#ifdef". Does it have any impact? > > > > > > Regards, > > > > Adel > > > > --------------------------------------------------------------------- > To unsubscribe, e-mail: [email protected] > For additional commands, e-mail: [email protected] > --------------------------------------------------------------------- To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]
