directions: ask the car for its speakers before each turn when the music isn't ours

gearslip/src/main/java/app/seb3thehacker/gearslip/GearslipRunner.kt

@@ -274,7 +274,12 @@ class GearslipRunner(
             MSG_PING_REQUEST -> onPingRequest(body)
             MSG_BYEBYE_REQUEST -> onByeByeRequest(body)
             MSG_UNEXPECTED_MESSAGE, MSG_FRAMING_ERROR -> onProtocolError(messageId)
-            AudioLink.MSG_FOCUS_RESPONSE -> audioLink?.onFocus(Protobuf.readInt32Field(body, 1) ?: 0)
+            AudioLink.MSG_FOCUS_RESPONSE -> {
+                // One answer for both sinks: it is the directions' while they wait on one.
+                val state = Protobuf.readInt32Field(body, 1) ?: 0
+                val guidance = guidanceLink
+                if (guidance != null && guidance.expectsFocusAnswer()) guidance.onFocus(state) else audioLink?.onFocus(state)
+            }
             else -> log.i("unhandled control message id=$messageId (${body.size} bytes) - continuing")
         }
     }
@@ -587,7 +592,7 @@ class GearslipRunner(
             tag = "guidance",
             sessionId = GUIDANCE_SESSION_ID,
             asksFocus = false,
-            onReady = { DirectionsVoice.attach(it) },
+            onReady = { DirectionsVoice.attach(it) { audioLink?.streaming == true } },
             onClose = { DirectionsVoice.detach(it) },
         )
         openChannel(guidance.serviceId)

gearslip/src/main/java/app/seb3thehacker/gearslip/audio/AudioLink.kt

@@ -4,6 +4,8 @@ import app.seb3thehacker.gearslip.GearslipLog
 import app.seb3thehacker.gearslip.Protobuf
 import app.seb3thehacker.gearslip.ServiceDiscovery
 import app.seb3thehacker.gearslip.Wire
+import java.util.concurrent.CountDownLatch
+import java.util.concurrent.TimeUnit
 
 /**
  * The phone end of the head unit's media audio sink, the audio twin of the video channel:
@@ -25,7 +27,10 @@ class AudioLink(
     /** Log name: "audio" for the media sink, "guidance" for the directions sink. */
     private val tag: String = "audio",
     private val sessionId: Int = SESSION_ID,
-    /** Media asks the car for focus; directions just start, and the car ducks music under them. */
+    /**
+     * Media asks the car for focus when it starts. Directions ask only around each announcement,
+     * through [askForSpeakers], and only when the music isn't already ours.
+     */
     private val asksFocus: Boolean = true,
     private val onReady: (AudioLink) -> Unit = { CarAudio.attach(it) },
     private val onClose: (AudioLink) -> Unit = { CarAudio.detach(it) },
@@ -58,6 +63,47 @@ class AudioLink(
 
     val channelId get() = service.serviceId
 
+    /** Sending audio now, which for media means the car has handed us the speakers. */
+    val streaming get() = started
+
+    @Volatile private var holdsSpeakers = false
+    @Volatile private var focusAnswer: CountDownLatch? = null
+    @Volatile private var answerDueBy = 0L
+
+    /** True just after [askForSpeakers] or [handBackSpeakers], when the car's next focus answer is ours. */
+    fun expectsFocusAnswer() = System.currentTimeMillis() < answerDueBy
+
+    /**
+     * For directions when nothing of ours is playing, so the car's speakers belong to the radio
+     * or a Bluetooth phone. Android Auto asks for them before each announcement, letting the
+     * other source duck, and hands them back afterwards. Speaking without asking left a 2025
+     * Nissan's Bosch head unit rebooting mid-drive (issue #35).
+     *
+     * Waits briefly for the car's answer and speaks anyway without one, as some cars never answer.
+     */
+    fun askForSpeakers() {
+        if (holdsSpeakers) return
+        val answer = CountDownLatch(1)
+        focusAnswer = answer
+        answerDueBy = System.currentTimeMillis() + FOCUS_WAIT_MS
+        holdsSpeakers = true
+        sendControl(MSG_FOCUS_REQUEST, Protobuf.varintField(1, FOCUS_GAIN_TRANSIENT_MAY_DUCK.toLong()))
+        GearslipLog.i("$tag: -> AudioFocusRequest(GAIN_TRANSIENT_MAY_DUCK)")
+        if (!answer.await(FOCUS_WAIT_MS, TimeUnit.MILLISECONDS)) {
+            GearslipLog.i("$tag: no focus answer after ${FOCUS_WAIT_MS}ms - speaking anyway")
+        }
+        focusAnswer = null
+    }
+
+    /** Gives back what [askForSpeakers] took, so the other source comes back to full volume. */
+    fun handBackSpeakers() {
+        if (!holdsSpeakers) return
+        holdsSpeakers = false
+        answerDueBy = System.currentTimeMillis() + FOCUS_WAIT_MS
+        runCatching { sendControl(MSG_FOCUS_REQUEST, Protobuf.varintField(1, FOCUS_RELEASE.toLong())) }
+        GearslipLog.i("$tag: -> AudioFocusRequest(RELEASE)")
+    }
+
     fun opened(status: Int) {
         GearslipLog.i("$tag: <- ChannelOpenResponse: status=$status")
         if (status != 0) {
@@ -161,6 +207,12 @@ class AudioLink(
     fun onFocus(state: Int) {
         val name = FOCUS_STATES[state] ?: "unknown($state)"
         GearslipLog.i("$tag: <- AudioFocusResponse state=$name")
+        if (!asksFocus) {
+            // Directions speak whatever the answer: a lost turn is worse than a loud one.
+            answerDueBy = 0L
+            focusAnswer?.countDown()
+            return
+        }
         val had = focusGranted
         focusGranted = state == FOCUS_STATE_GAIN || state == FOCUS_STATE_GAIN_TRANSIENT ||
             state == FOCUS_STATE_GAIN_MEDIA_ONLY
@@ -286,6 +338,7 @@ class AudioLink(
         configTimer?.interrupt()
         configTimer = null
         stop()
+        handBackSpeakers()
         ready = false
         onClose(this)
     }
@@ -303,7 +356,9 @@ class AudioLink(
         private const val MSG_SETUP_ANSWER = 255
         private const val STATUS_READY = 2
         private const val FOCUS_GAIN = 1
+        private const val FOCUS_GAIN_TRANSIENT_MAY_DUCK = 3
         private const val FOCUS_RELEASE = 4
+        private const val FOCUS_WAIT_MS = 500L
         private const val FOCUS_STATE_GAIN = 1
         private const val FOCUS_STATE_GAIN_TRANSIENT = 2
         private const val FOCUS_STATE_GAIN_MEDIA_ONLY = 6

gearslip/src/main/java/app/seb3thehacker/gearslip/car/DirectionsVoice.kt

@@ -27,7 +27,8 @@ import kotlin.math.roundToInt
  * same turn the instrument cluster gets ([NavTurn]) and says it itself: once when the turn comes
  * up, again a kilometre or so out on a long stretch, and once more just before it. The words are
  * rendered with the phone's text-to-speech to a file, never played on the phone, and streamed to
- * the car, which ducks the music under them the way it does for real Android Auto.
+ * the car, which ducks the music under them the way it does for real Android Auto. When the music
+ * isn't Gearslip's, each announcement first asks the car for its speakers, as Android Auto does.
  *
  * Nothing spoken is logged: road names are where the driver is going.
  */
@@ -42,7 +43,10 @@ object DirectionsVoice {
 
     @Volatile private var link: AudioLink? = null
     @Volatile private var running = false
-    private val speaker = Executors.newSingleThreadExecutor { Thread(it, "gearslip-directions").apply { isDaemon = true } }
+    @Volatile private var musicIsOurs: () -> Boolean = { false }
+    private val speaker = Executors.newSingleThreadScheduledExecutor { Thread(it, "gearslip-directions").apply { isDaemon = true } }
+    /** Counts announcements, so a hand-back scheduled after one is dropped if another has begun. */
+    private var spoken = 0
     @Volatile private var latest = 0
 
     private var turnKey: String? = null
@@ -62,8 +66,9 @@ object DirectionsVoice {
     }
 
     /** The car opened its directions sink: start following the map app's turns. */
-    fun attach(link: AudioLink) {
+    fun attach(link: AudioLink, musicIsOurs: () -> Boolean) {
         this.link = link
+        this.musicIsOurs = musicIsOurs
         GearslipLog.i("guidance: the car offered a directions sink at ${link.sampleRate}Hz x${link.channels}")
         if (running) return
         running = true
@@ -223,7 +228,16 @@ object DirectionsVoice {
             GearslipLog.w("guidance: text-to-speech gave audio Gearslip can't read")
             return
         }
-        stream(sink, pcm)
+        // Logged before speaking too, so a car that dies mid-announcement shows what it was doing.
+        GearslipLog.i("guidance: speaking a turn")
+        // Held through back-to-back prompts rather than asked for and given back each time.
+        if (!musicIsOurs()) sink.askForSpeakers()
+        val turn = ++spoken
+        try {
+            stream(sink, pcm)
+        } finally {
+            speaker.schedule({ if (turn == spoken) sink.handBackSpeakers() }, HOLD_MS, TimeUnit.MILLISECONDS)
+        }
         GearslipLog.i("guidance: spoke a turn (${pcm.size / (2 * sink.channels) * 1000 / sink.sampleRate} ms)")
     }
 
@@ -321,6 +335,7 @@ object DirectionsVoice {
     private const val PREPARE_FROM_METERS = 2_500
     private const val PREPARE_AT_METERS = 1_000
     private const val LEAD_MS = 100L
+    private const val HOLD_MS = 2_000L
     private const val SYNTH_TIMEOUT_S = 10L
     private val ORDINALS = listOf("first", "second", "third", "fourth", "fifth", "sixth", "seventh", "eighth")
 }