Fix announce timers on Solaris (and others, given the right timing) (...)

In the successfull case, we start the timer with NewTimer(0), then do a
bunch of stuff during which time it can fire, then reset it with
Reset(0). The result is that two timer firings are queued when we enter
the select loop, so we do two announcements back to back and fail the
tests.

internal/discover/client_udp.go

@@ -91,16 +91,14 @@ func (d *UDPClient) Start(uri *url.URL, pkt *Announce) error {
 
 func (d *UDPClient) broadcast(pkt []byte) {
 	defer d.wg.Done()
-	timer := time.NewTimer(0)
 
 	conn, err := net.ListenUDP(d.url.Scheme, d.listenAddress)
 	for err != nil {
-		timer.Reset(d.errorRetryInterval)
 		l.Warnf("Global UDP discovery (%s): %v; trying again in %v", d.url, err, d.errorRetryInterval)
 		select {
 		case <-d.stop:
 			return
-		case <-timer.C:
+		case <-time.After(d.errorRetryInterval):
 		}
 		conn, err = net.ListenUDP(d.url.Scheme, d.listenAddress)
 	}
@@ -108,18 +106,16 @@ func (d *UDPClient) broadcast(pkt []byte) {
 
 	remote, err := net.ResolveUDPAddr(d.url.Scheme, d.url.Host)
 	for err != nil {
-		timer.Reset(d.errorRetryInterval)
 		l.Warnf("Global UDP discovery (%s): %v; trying again in %v", d.url, err, d.errorRetryInterval)
 		select {
 		case <-d.stop:
 			return
-		case <-timer.C:
+		case <-time.After(d.errorRetryInterval):
 		}
 		remote, err = net.ResolveUDPAddr(d.url.Scheme, d.url.Host)
 	}
 
-	timer.Reset(0)
+	timer := time.NewTimer(0)
-
 	for {
 		select {
 		case <-d.stop: