Fix a bug that zmq_poll's select backend spins when timeout=-1, due to

ptimeout not properly recalculated after first pass.
Signed-off-by: Chia-liang Kao <clkao@clkao.org>

AUTHORS

@@ -8,6 +8,7 @@ Bernd Prager <bernd@prager.ws>
 Bernd Melchers <melchers@ZEDAT.FU-Berlin.DE>
 Brian Buchanan <bwb@holo.org>
 Burak Arslan <burak-github@arskom.com.tr>
+Chia-liang Kao <clkao@clkao.org>
 Chris Wong <chris@chriswongstudio.com>
 Christian Gudrian <christian.gudrian@fluidon.com>
 Conrad D. Steenberg <conrad.steenberg@caltech.edu>

src/zmq.cpp

@@ -545,6 +545,8 @@ int zmq_poll (zmq_pollitem_t *items_, int nitems_, long timeout_)
     int nevents = 0;
     fd_set inset, outset, errset;
 
+    while (true) {
+
         //  Compute the timeout for the subsequent poll.
         timeval timeout;
         timeval *ptimeout;
@@ -561,8 +563,6 @@ int zmq_poll (zmq_pollitem_t *items_, int nitems_, long timeout_)
             ptimeout = &timeout;
         }
 
-    while (true) {
-
         //  Wait for events. Ignore interrupts if there's infinite timeout.
         while (true) {
             memcpy (&inset, &pollset_in, sizeof (fd_set));