Pieter,
> - Martin points out that the calling thread is sleeping, not
> processing commands, so it's normal that the mailbox backs up.
Here's a patch that should alleviate the problem in this specific use case.
It make zmq_init_t object (the one that handles initial exchange of
identities) a child of zmq_listener_t object (one that waits for new
connections) instead of being direct child of the socket object.
Thus the init's startup and teardown handshaking happens between init
object and listener object. Both of those live in I/O thread and are
able to process the commands immediately -- there's no way user can
block the processing.
The application thread's mailbox would thus get *no* commands when new
connection is created, identities are exchanged and the connection is
dropped subsequently because of duplicit identity.
diff --git a/src/zmq_listener.cpp b/src/zmq_listener.cpp
index 522ab3f..2a7f1eb 100644
--- a/src/zmq_listener.cpp
+++ b/src/zmq_listener.cpp
@@ -72,6 +72,6 @@ void zmq::zmq_listener_t::in_event ()
zmq_init_t *init = new (std::nothrow) zmq_init_t (io_thread, socket,
NULL, fd, options);
zmq_assert (init);
- launch_sibling (init);
+ launch_child (init);
}
Martin
_______________________________________________
zeromq-dev mailing list
[email protected]
http://lists.zeromq.org/mailman/listinfo/zeromq-dev
