On Tue, Aug 17, 2010 at 6:43 PM, Pieter Hintjens <[email protected]> wrote: > Not particularly, poll() would have to make the same assumption, but > honestly I don't think treating the null case as "everything" is > clean. It's a hack. Better to be 100% consistent.
Yes, of course. Having poll() make the same assumption would be even worse, though. Consider the case you've described earlier (polling a command socket and a subscription socket and applying commands as SUBSCRIBE/UNSUBSCRIBE): if poll() on a socket with no subscriptions assumes the socket is subscribed to everything, end users would have to code extremely dirty logic in the command handler to avoid catching all messages (create the sockets, but don't connect()/bind() the subscribing one until the first subscription, close the socket and recreate it instead of removing the last one, etc). Having recv() just fail (and poll() to do nothing) would be a lot saner, IMO. _______________________________________________ zeromq-dev mailing list [email protected] http://lists.zeromq.org/mailman/listinfo/zeromq-dev
