On Sun, Nov 1, 2009 at 10:03 AM, Leif Hedstrom <[email protected]> wrote: > On 10/30/2009 03:46 PM, Paul Querna wrote: >> >> Hi, >> >> I think we need to use some kind of library to abstract IO polling, >> rather than completely reinventing the wheel. >> > > How about we take a two-step approach here? My worry is to break something > major like this int he first week of being OpenSource. So, my suggestion is: > > 1) We add a --with-libev configure option. > 2) If this option is not specified, we use the existing epoll() code on > Linux. > 3) On other platforms, the --with-libev becomes implicit (since there's no > other choice anyways) > > > Later on, we can decide if libev becomes default on all platforms, but at > least this way we have some options. Thoughts?
I believe this will prove to be quite difficult. Currently libev uses callbacks for each fired event, rather than returning to you an array of fired events, as our use (or at least from what I could see quickly) is built around processing these arrays, it would be definitely more complicated than just a simple ifdef around epoll_wait(). Thanks, Paul
