From 87492aab89dabebad782d0cbc25030777933f849 Mon Sep 17 00:00:00 2001
From: klopez4212 <klopez4212@gmail.com>
Date: Mon, 6 Jul 2026 17:20:26 +0100
Subject: [PATCH] fix(dictation): guard cross-composer cancel, stop engine on
 unmount, await flush before stop
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

Addresses 3 new review comments:

1. **Avoid cancelling another composer's dictation** — the draftKey effect
   now checks isRecordingRef before calling cancelRecording(), so switching
   channels in one composer won't kill a recording in another.

2. **Stop native dictation on unmount** — cleanup() now calls
   invoke('stop_dictation') so the native SttEngine doesn't linger when
   the hook unmounts mid-recording (navigation, composer close).

3. **Await final audio flush before stopping STT** — stopRecording() and
   cleanup() now await the flushAudioBatch() promise before invoking
   stop_dictation, ensuring the last batch of audio arrives at the native
   engine before it shuts down and flushes its speech buffer.
---
 .../dictation/hooks/useComposerDictation.ts   |  8 ++++-
 .../dictation/hooks/useLocalDictation.ts      | 31 +++++++++++--------
 2 files changed, 25 insertions(+), 14 deletions(-)

diff --git a/desktop/src/features/dictation/hooks/useComposerDictation.ts b/desktop/src/features/dictation/hooks/useComposerDictation.ts
index 564db7fe3..709662ccb 100644
--- a/desktop/src/features/dictation/hooks/useComposerDictation.ts
+++ b/desktop/src/features/dictation/hooks/useComposerDictation.ts
@@ -89,9 +89,15 @@ export function useComposerDictation({
 
   // Cancel dictation when the channel/thread changes so that transcript events
   // from a stale local STT session don't leak into the wrong draft.
+  // Only cancel if this instance is actually recording — avoids killing another
+  // composer's session since the native engine is a singleton.
+  const isRecordingRef = useRef(false);
+  isRecordingRef.current = dictation.isRecording || dictation.isStarting;
   // biome-ignore lint/correctness/useExhaustiveDependencies: draftKey is the sole trigger
   useEffect(() => {
-    dictation.cancelRecording();
+    if (isRecordingRef.current) {
+      dictation.cancelRecording();
+    }
   }, [draftKey]);
 
   // Auto-cancel dictation when the composer becomes disabled mid-recording
diff --git a/desktop/src/features/dictation/hooks/useLocalDictation.ts b/desktop/src/features/dictation/hooks/useLocalDictation.ts
index 0d992b92b..b5a938fdf 100644
--- a/desktop/src/features/dictation/hooks/useLocalDictation.ts
+++ b/desktop/src/features/dictation/hooks/useLocalDictation.ts
@@ -111,10 +111,11 @@ export function useLocalDictation({
     };
   }, []);
 
-  /** Flush accumulated audio batch to the native STT engine. */
-  const flushAudioBatch = useCallback(() => {
+  /** Flush accumulated audio batch to the native STT engine. Returns a promise
+   *  that resolves once the IPC call completes (or immediately if nothing to flush). */
+  const flushAudioBatch = useCallback((): Promise<void> => {
     const batch = audioBatchRef.current;
-    if (batch.length === 0) return;
+    if (batch.length === 0) return Promise.resolve();
 
     // Calculate total byte length and merge into a single buffer.
     let totalSamples = 0;
@@ -134,12 +135,16 @@ export function useLocalDictation({
       merged.byteOffset,
       merged.byteLength,
     );
-    invokeRawBinary("push_dictation_audio", bytes).catch(() => {});
+    return invokeRawBinary("push_dictation_audio", bytes)
+      .then(() => {})
+      .catch(() => {});
   }, []);
 
   const cleanup = useCallback(() => {
-    // Flush any remaining audio before teardown.
-    flushAudioBatch();
+    // Flush any remaining audio before teardown, then stop the native engine.
+    void flushAudioBatch().then(() => {
+      invoke("stop_dictation").catch(() => {});
+    });
     // Stop batch timer.
     if (batchTimerRef.current) {
       clearInterval(batchTimerRef.current);
@@ -307,16 +312,12 @@ export function useLocalDictation({
   }, [cleanup, flushAudioBatch, isEnabled, isRecording, isStarting]);
 
   const stopRecording = useCallback(() => {
-    // Flush remaining audio so the native engine can transcribe the tail.
-    flushAudioBatch();
-    // Stop batch timer.
+    // Stop batch timer immediately.
     if (batchTimerRef.current) {
       clearInterval(batchTimerRef.current);
       batchTimerRef.current = null;
     }
-    // Stop mic and audio pipeline immediately so the user gets visual feedback,
-    // but keep isTranscribing=true until the native `stopped` event arrives
-    // (which fires only after the final transcript has been forwarded).
+    // Stop mic and audio pipeline immediately so the user gets visual feedback.
     if (streamRef.current) {
       for (const track of streamRef.current.getTracks()) {
         track.stop();
@@ -331,9 +332,13 @@ export function useLocalDictation({
       void audioContextRef.current.close();
       audioContextRef.current = null;
     }
-    invoke("stop_dictation").catch(() => {});
     setIsRecording(false);
+    // Flush remaining audio and THEN stop the native engine, ensuring the
+    // final batch arrives before the engine shuts down and flushes its buffer.
     // isTranscribing stays true — cleared when `dictation-state: stopped` arrives.
+    void flushAudioBatch().then(() => {
+      invoke("stop_dictation").catch(() => {});
+    });
   }, [flushAudioBatch]);
 
   const cancelRecording = useCallback(() => {