chore: release v0.25.8

internal/cluster/cluster.go

@@ -18,6 +18,7 @@ import (
 	"strconv"
 	"strings"
 	"sync"
+	"sync/atomic"
 	"time"
 
 	"hamlog/internal/applog"
@@ -355,6 +356,18 @@ func (s *session) run() {
 // that gets skipped.
 var idleTick = 30 * time.Second
 
+// promptTick is the read deadline used until the login handshake is finished.
+// Short, because a node's "login:" / "password:" carries no newline and is only
+// visible when the read times out — see the read loop. It costs a few wake-ups
+// during the first seconds of a connection and nothing afterwards.
+const promptTick = 700 * time.Millisecond
+
+// handshakeWindow bounds how long the fast promptTick applies. A node that has
+// a password configured but never asks for one would otherwise keep the loop
+// waking every 700 ms for the life of the connection, for a prompt that is never
+// coming. Any real login exchange is over in a second or two.
+const handshakeWindow = 20 * time.Second
+
 // quietNotice is the silence after which the log says so, once.
 const quietNotice = 10 * time.Minute
 
@@ -398,16 +411,37 @@ func (s *session) runOnce() (time.Time, error) {
 
 	// Login: send on first prompt OR blindly after 1.5s. Many DXSpider
 	// nodes accept the callsign without re-prompting.
-	loginSent := false
+	//
+	// Atomic because the blind-login timer below and the read loop both touch
+	// these. loginSent used to be a plain bool that the timer NEVER SET: the
+	// callsign went out and nothing recorded it, so the password branch — gated on
+	// loginSent — was dead code, and the session could only reach "connected" by
+	// recognising a welcome banner. On a node whose entire greeting is a bare
+	// "login:" and which then demands a password (f5mzn.org:9000), that left the
+	// server stuck at "connecting" for ever while telnet logged in by hand fine.
+	var loginSent, pwdSent atomic.Bool
+	// CompareAndSwap, not a plain store: the timer and the loop can reach these at
+	// the same moment, and the callsign must be written exactly once.
+	sendLogin := func() {
+		if s.login != "" && loginSent.CompareAndSwap(false, true) {
+			_, _ = conn.Write([]byte(s.login + "\r\n"))
+		}
+	}
+	// Sent ONCE per connection. A node that re-prompts is refusing the password,
+	// and answering with the same one again only loops — better to let the
+	// refusal show in the console than to hide it behind a retry.
+	sendPassword := func() {
+		if s.cfg.Password != "" && loginSent.Load() && pwdSent.CompareAndSwap(false, true) {
+			_, _ = conn.Write([]byte(s.cfg.Password + "\r\n"))
+		}
+	}
 	if s.login != "" {
 		go func() {
 			select {
 			case <-s.stopCh:
 				return
 			case <-time.After(1500 * time.Millisecond):
-				if !loginSent {
-					_, _ = conn.Write([]byte(s.login + "\r\n"))
-				}
+				sendLogin()
 			}
 		}()
 	}
@@ -453,7 +487,21 @@ func (s *session) runOnce() (time.Time, error) {
 
 	var connectedAt time.Time
 	var quiet time.Duration // how long the node has said nothing
+	var quietNoticed bool   // the long-silence line is said once
 	var pending string      // a line cut in half by a read deadline
+	markConnected := func() {
+		if s.snapshot().State == StateConnected {
+			return
+		}
+		connectedAt = time.Now()
+		s.mu.Lock()
+		s.status.State = StateConnected
+		s.status.ConnectedAt = connectedAt
+		s.status.Error = ""
+		s.mu.Unlock()
+		s.emitStatus()
+		fireInitCommands()
+	}
 	rd := bufio.NewReader(conn)
 	for {
 		select {
@@ -471,7 +519,21 @@ func (s *session) runOnce() (time.Time, error) {
 		// Only a REAL error ends the session. A dead peer still gets caught:
 		// TCP keepalive probes an idle connection and its failure arrives here
 		// as an error, not as a timeout.
-		_ = conn.SetReadDeadline(time.Now().Add(idleTick))
+		// SHORT deadline until the handshake is done, the long idle tick after.
+		//
+		// A cluster writes its prompts WITHOUT a trailing newline, and ReadString
+		// only returns on one — so a prompt is never a "line" at all, it is whatever
+		// is sitting in the buffer when the read deadline expires. At the ordinary
+		// 30 s tick that made a bare "login:" invisible for half a minute and a
+		// following "password:" invisible for another, which is long enough for the
+		// node to give up on us. Only the handshake needs the fast tick; once logged
+		// in, a long deadline is exactly what we want (see idleTick).
+		tick := idleTick
+		if time.Since(linkUpAt) < handshakeWindow &&
+			(!loginSent.Load() || (s.cfg.Password != "" && !pwdSent.Load())) {
+			tick = promptTick
+		}
+		_ = conn.SetReadDeadline(time.Now().Add(tick))
 		chunk, err := rd.ReadString('\n')
 		if err != nil {
 			var ne net.Error
@@ -480,16 +542,48 @@ func (s *session) runOnce() (time.Time, error) {
 				// Keep it: a spot line straddling the deadline would otherwise lose
 				// its first half and arrive as nonsense, or vanish entirely.
 				pending += chunk
-				quiet += idleTick
+				// …but a newline-less PROMPT is not half a line, it is a question,
+				// and this is the only place it can ever be seen. Answer it, show it
+				// in the console (an operator watching a stuck server has a right to
+				// see what the node actually asked), and drop it so it is not glued
+				// onto the front of the next real line.
+				if p := strings.TrimSpace(pending); p != "" {
+					switch {
+					case !loginSent.Load() && s.login != "" && isLoginPrompt(p):
+						s.emitLine(p, false)
+						pending = ""
+						sendLogin()
+						if s.cfg.Password == "" {
+							markConnected()
+						}
+						continue
+					case !pwdSent.Load() && isPasswordPrompt(p):
+						s.emitLine(p, false)
+						pending = ""
+						if s.cfg.Password == "" {
+							// Nothing to answer with. Shown in the console rather than
+							// swallowed: an unanswered "password:" sitting there IS the
+							// explanation for a server that never finishes connecting,
+							// and it is something the operator can act on.
+							continue
+						}
+						sendPassword()
+						markConnected()
+						continue
+					}
+				}
+				quiet += tick
 				// Said once at the first long silence, so a genuinely mute node is
 				// visible without a line every tick.
-				if quiet == quietNotice {
+				if !quietNoticed && quiet >= quietNotice {
+					quietNoticed = true
 					applog.Printf("cluster[%s] no traffic for %s — still connected", s.cfg.Name, quiet)
 				}
 				continue
 			}
 			return connectedAt, fmt.Errorf("read: %w", err)
 		}
+		quietNoticed = false
 		quiet = 0
 		line := pending + chunk
 		pending = ""
@@ -519,28 +613,23 @@ func (s *session) runOnce() (time.Time, error) {
 			}
 		}
 
-		// Login on explicit prompt.
-		if !loginSent && s.login != "" && isLoginPrompt(line) {
-			_, _ = conn.Write([]byte(s.login + "\r\n"))
-			loginSent = true
+		// Login on explicit prompt — the case where the node DID terminate it with
+		// a newline. The newline-less form is handled on the timeout path above.
+		if !loginSent.Load() && s.login != "" && isLoginPrompt(line) {
+			s.emitLine(line, false)
+			sendLogin()
 			continue
 		}
-		// Password on prompt (rare).
-		if loginSent && s.cfg.Password != "" && isPasswordPrompt(line) {
-			_, _ = conn.Write([]byte(s.cfg.Password + "\r\n"))
+		// Password on prompt.
+		if !pwdSent.Load() && isPasswordPrompt(line) {
+			s.emitLine(line, false)
+			sendPassword()
 			continue
 		}
 
 		// Mark connected once we've sent login OR seen a welcome banner.
-		if s.snapshot().State != StateConnected && (loginSent || isWelcome(line)) {
-			connectedAt = time.Now()
-			s.mu.Lock()
-			s.status.State = StateConnected
-			s.status.ConnectedAt = connectedAt
-			s.status.Error = ""
-			s.mu.Unlock()
-			s.emitStatus()
-			fireInitCommands()
+		if loginSent.Load() || isWelcome(line) {
+			markConnected()
 		}
 
 		// EVERY line goes to the console — spot or not. This is the whole point:
@@ -701,6 +790,38 @@ func parseSpot(line string) (Spot, bool) {
 	}, true
 }
 
+// NewLocalSpot builds the Spot for a DX announcement WE just sent, so it lands
+// in the operator's own list at once.
+//
+// A node does not necessarily broadcast a spot back to the station that sent it:
+// DXSpider suppresses the echo to the originator, and a node-side filter can eat
+// it too. So an operator spotted a station, watched their own spot list stay
+// empty, and concluded the spot had never gone out — when it had. This is the
+// spot the echo would have carried, built from what we sent, deliberately the
+// same shape so the UI's call+band de-dupe folds the two into one row on the
+// nodes that DO echo.
+func NewLocalSpot(srv ServerConfig, spotter, dxCall string, freqKHz float64, comment string) Spot {
+	freqHz := int64(freqKHz*1000 + 0.5)
+	now := time.Now()
+	sp := Spot{
+		SourceID:   srv.ID,
+		SourceName: srv.Name,
+		Spotter:    strings.ToUpper(strings.TrimSpace(spotter)),
+		DXCall:     strings.ToUpper(strings.TrimSpace(dxCall)),
+		FreqKHz:    freqKHz,
+		FreqHz:     freqHz,
+		Band:       bandFromHz(freqHz),
+		Comment:    strings.TrimSpace(comment),
+		TimeUTC:    now.UTC().Format("1504") + "Z",
+		ReceivedAt: now,
+	}
+	// Raw reads like the node's own broadcast — it is what anything showing the
+	// source line expects, and it keeps a local spot legible in the log.
+	sp.Raw = fmt.Sprintf("DX de %s: %9.1f  %-12s %-30s %s",
+		sp.Spotter, sp.FreqKHz, sp.DXCall, sp.Comment, sp.TimeUTC)
+	return sp
+}
+
 func isLoginPrompt(s string) bool {
 	low := strings.ToLower(s)
 	return strings.Contains(low, "login:") ||