refactor(voice): isolate gemini live provider core

src/voice/geminiDeferredInput.test.mjs

@@ -0,0 +1,285 @@
+import test from 'node:test';
+import assert from 'node:assert/strict';
+import {
+  nativeIntegrationFixture,
+  assertSuccessfulZoom,
+} from '../testSupport/geminiIntegrationFixture.mjs';
+
+const options = { concurrency: false, timeout: 5000 };
+const OLD_PCM = 'BAAFAA==';
+const TEXT = 'Zoom in for the owned typed successor';
+
+function exposeCapture(h) {
+  const createAudio = h.controller.createAudio;
+  h.controller.createAudio = (callbacks) => {
+    const audio = createAudio(callbacks);
+    audio.captureCallbacks = callbacks;
+    return audio;
+  };
+}
+
+async function startManual(h) {
+  await h.session.start({ pushToTalk: true });
+  assert.equal(h.session.state, 'listening');
+  assert.equal(h.controller.input.pushToTalkMode, true);
+  assert.equal(Boolean(h.controller.input.pushToTalkKeyHeld), false);
+  return h.sockets.at(-1);
+}
+
+// Explicit fake worklet callbacks retain the native physical-capture path;
+// this helper does not populate controller queues or ownership fields itself.
+function speakAndRelease(h, data) {
+  const c = h.controller;
+  const audio = h.audios.at(-1);
+  c.input.pushToTalkKeyHeld = true;
+  c.input.setMicrophoneEnabled(true);
+  const metadata = audio.metadata;
+  assert.ok(Number.isSafeInteger(metadata?.captureId));
+  audio.captureCallbacks.onAudioStart(metadata);
+  audio.captureCallbacks.onAudio(
+    { data, mimeType: 'audio/pcm;rate=16000' },
+    metadata,
+  );
+  c.input.pushToTalkKeyHeld = false;
+  c.input.setMicrophoneEnabled(false);
+  audio.captureCallbacks.onAudioEnd(metadata);
+  return metadata;
+}
+
+async function prepareDeferredPtt(h) {
+  exposeCapture(h);
+  const socket = await startManual(h);
+  speakAndRelease(h, 'AAABAA=='); // A is submitted and awaits its reply.
+  speakAndRelease(h, 'AgADAA=='); // B is submitted, retiring A.
+  const wireBeforeC = socket.sent.length;
+  const metadata = speakAndRelease(h, OLD_PCM); // C waits behind A's boundary.
+  const owner = h.controller.deferredInput;
+  assert.equal(owner?.kind, 'push-to-talk');
+  assert.equal(owner.captureId, metadata.captureId);
+  assert.equal(
+    socket.sent.length,
+    wireBeforeC,
+    'C must still be wholly deferred',
+  );
+  assert.deepEqual(
+    owner.messages.map((message) =>
+      message.realtimeInput?.activityStart
+        ? 'start'
+        : message.realtimeInput?.audio
+          ? 'audio'
+          : message.realtimeInput?.activityEnd
+            ? 'end'
+            : 'other',
+    ),
+    ['start', 'audio', 'end'],
+  );
+  return { socket, owner, metadata };
+}
+
+async function assertHealthySuccessor(h, socket, callId) {
+  const beforeHeight = h.state.height;
+  await h.tool(callId, socket);
+  assertSuccessfulZoom(h, socket, callId, beforeHeight);
+  assert.equal(h.controller.pendingTools, 0);
+  h.complete(socket); // Native post-tool response boundary, with no output PCM.
+  assert.equal(h.controller.narration.state().tools, 0);
+  assert.equal(h.session.state, 'listening');
+}
+
+for (const mode of ['typed replacement', 'Stop', 'Stop+restart']) {
+  test(
+    `F04-5 deferred PTT flush loses its owner on ${mode}`,
+    options,
+    async (t) => {
+      const h = nativeIntegrationFixture(t);
+      const { socket: oldSocket, owner } = await prepareDeferredPtt(h);
+      const oldAudio = h.audios.at(-1);
+      let observed = 0;
+      let sentAtObserverReturn;
+      let restart;
+      let stateAtObserverReturn;
+      const unsubscribe = h.session.subscribe((event) => {
+        if (
+          event.type !== 'transcript' ||
+          event.role !== 'user' ||
+          event.text !== '' ||
+          observed
+        )
+          return;
+        observed++;
+        if (mode === 'typed replacement') {
+          assert.equal(h.session.sendText(TEXT), true);
+        } else {
+          h.session.stop();
+          if (mode === 'Stop+restart')
+            restart = h.session.start({ pushToTalk: true });
+        }
+        sentAtObserverReturn = oldSocket.sent.length;
+        stateAtObserverReturn = {
+          epoch: h.controller.epoch,
+          turn: h.controller.turn,
+          replyPending: h.controller.replyPending,
+          forwardedActivityOpen: h.controller.forwardedActivityOpen,
+        };
+      });
+      t.after(unsubscribe);
+
+      h.complete(oldSocket); // A's boundary starts flushing C. C's start emits.
+      assert.equal(
+        observed,
+        1,
+        'must enter the actual flush/onset observer boundary',
+      );
+      assert.ok(
+        stateAtObserverReturn,
+        'observer must complete its replacement action',
+      );
+      assert.deepEqual(
+        oldSocket.sent.slice(sentAtObserverReturn),
+        [],
+        'retired C must send neither queued PCM nor an extra end after the observer returns',
+      );
+      assert.deepEqual(
+        {
+          epoch: h.controller.epoch,
+          turn: h.controller.turn,
+          replyPending: h.controller.replyPending,
+          forwardedActivityOpen: h.controller.forwardedActivityOpen,
+        },
+        stateAtObserverReturn,
+        'old flush must not write successor state',
+      );
+      assert.equal(h.controller.deferredInput, null);
+      assert.equal(
+        h.controller.typedIntent?.text ?? null,
+        mode === 'typed replacement' ? TEXT : null,
+      );
+
+      let successorSocket;
+      if (mode === 'typed replacement') {
+        assert.equal(h.controller.epoch, owner.epoch);
+        assert.notEqual(h.controller.turn, owner.turn);
+        assert.equal(
+          oldSocket.sent.some(
+            (message) => message.realtimeInput?.audio?.data === OLD_PCM,
+          ),
+          false,
+        );
+        const typedIndex = oldSocket.sent.findIndex((message) =>
+          message.clientContent?.turns?.some((turn) =>
+            turn.parts?.some((part) => part.text === TEXT),
+          ),
+        );
+        assert.ok(
+          typedIndex >= 0,
+          'owned replacement must actually reach transport',
+        );
+        // The replacement legitimately closes the already-forwarded C start once.
+        assert.ok(oldSocket.sent[typedIndex - 1]?.realtimeInput?.activityEnd);
+        h.complete(oldSocket); // B's remaining retired boundary, deliberately separate.
+        assert.equal(h.controller.retiredReplyPending, false);
+        assert.equal(h.controller.replyPending, true);
+        successorSocket = oldSocket;
+      } else {
+        assert.equal(oldAudio.closed, true);
+        assert.equal(oldSocket.closed, true);
+        if (!restart) restart = h.session.start({ pushToTalk: true });
+        await restart;
+        successorSocket = h.sockets.at(-1);
+        assert.notEqual(successorSocket, oldSocket);
+        assert.equal(
+          successorSocket.sent.some((message) => message.realtimeInput?.audio),
+          false,
+        );
+        assert.equal(
+          successorSocket.sent.some(
+            (message) => message.realtimeInput?.activityEnd,
+          ),
+          false,
+        );
+        assert.equal(h.session.sendText(TEXT), true);
+      }
+      await assertHealthySuccessor(
+        h,
+        successorSocket,
+        `healthy-after-deferred-${mode}`,
+      );
+    },
+  );
+}
+
+test(
+  'F04-5 deferred typed onSent cannot resurrect replyPending after Stop+restart',
+  options,
+  async (t) => {
+    const h = nativeIntegrationFixture(t);
+    const oldSocket = await h.start();
+    assert.equal(h.session.sendText('A'), true);
+    assert.equal(h.session.sendText('B'), true);
+    assert.equal(h.session.sendText('C deferred text'), true);
+    assert.equal(h.controller.deferredInput?.kind, 'user-text');
+    let restart;
+    let observed = 0;
+    const unsubscribe = h.session.subscribe((event) => {
+      if (
+        event.type !== 'transcript' ||
+        event.role !== 'user' ||
+        event.text !== 'C deferred text'
+      )
+        return;
+      observed++;
+      h.session.stop();
+      restart = h.session.start();
+    });
+    t.after(unsubscribe);
+    h.complete(oldSocket);
+    assert.equal(observed, 1);
+    assert.equal(
+      h.controller.replyPending,
+      false,
+      'old onSent wrapper must not resurrect a stopped reply',
+    );
+    await restart;
+    assert.equal(h.controller.replyPending, false);
+    const socket = h.sockets.at(-1);
+    assert.equal(
+      socket.sent.length,
+      1,
+      'replacement connection should contain setup only',
+    );
+    assert.equal(h.session.sendText(TEXT), true);
+    await assertHealthySuccessor(h, socket, 'healthy-after-deferred-text');
+  },
+);
+
+test(
+  'F04-5 ordinary deferred PTT flush preserves start, PCM and end exactly once',
+  options,
+  async (t) => {
+    const h = nativeIntegrationFixture(t);
+    const { socket } = await prepareDeferredPtt(h);
+    const before = socket.sent.length;
+    h.complete(socket);
+    assert.deepEqual(
+      socket.sent
+        .slice(before)
+        .map((message) =>
+          message.realtimeInput?.activityStart
+            ? 'start'
+            : message.realtimeInput?.audio?.data === OLD_PCM
+              ? 'audio'
+              : message.realtimeInput?.activityEnd
+                ? 'end'
+                : 'other',
+        ),
+      ['start', 'audio', 'end'],
+    );
+    assert.equal(h.controller.deferredInput, null);
+    assert.equal(h.controller.replyPending, true);
+    assert.equal(h.controller.retiredReplyPending, true);
+    h.complete(socket); // B's boundary; C remains the owned submitted input.
+    assert.equal(h.controller.retiredReplyPending, false);
+    assert.equal(h.controller.replyPending, true);
+    await assertHealthySuccessor(h, socket, 'healthy-ordinary-deferred');
+  },
+);