From 8e007b01306a1647e5bfe84c2695f882652c8fd4 Mon Sep 17 00:00:00 2001
From: dijunkun <junkun.di@hotmail.com>
Date: Thu, 29 Oct 2026 15:08:56 +0800
Subject: [PATCH] [fix] release idle audio playback devices to allow Windows
 sleep, refs #131

---
 .../devices/session_device_manager.cpp        | 23 ++++++++++++++++---
 .../features/devices/session_device_manager.h |  4 +++-
 2 files changed, 23 insertions(+), 4 deletions(-)

diff --git a/apps/desktop/src/gui/features/devices/session_device_manager.cpp b/apps/desktop/src/gui/features/devices/session_device_manager.cpp
index 0e16c2e7..7040088c 100644
--- a/apps/desktop/src/gui/features/devices/session_device_manager.cpp
+++ b/apps/desktop/src/gui/features/devices/session_device_manager.cpp
@@ -123,7 +123,6 @@ void SessionDeviceManager::RecordCaptureCadence(
 }
 
 void SessionDeviceManager::Initialize() {
-  if (!owner_.unattended_host_) InitializeAudioOutput();
   screen_capturer_factory_ = new ScreenCapturerFactory();
   device_controller_factory_ = new DeviceControllerFactory();
   keyboard_capturer_ =
@@ -573,6 +572,12 @@ int SessionDeviceManager::StopKeyboardCapturer() {
 }
 
 int SessionDeviceManager::InitializeAudioOutput() {
+  if (output_device_) return 0;
+  const auto now = std::chrono::steady_clock::now();
+  if (now < next_audio_output_retry_) return -1;
+  // A missing/unavailable playback device must not be reopened for every frame.
+  next_audio_output_retry_ = now + std::chrono::seconds(1);
+
   SDL_AudioSpec desired_out{};
   desired_out.freq = 48000;
   desired_out.format = SDL_AUDIO_S16;
@@ -610,7 +615,12 @@ int SessionDeviceManager::InitializeAudioOutput() {
 #endif
   }
 
-  SDL_ResumeAudioDevice(output_device_);
+  if (!SDL_ResumeAudioDevice(output_device_)) {
+    LOG_ERROR("Failed to resume output device: {}", SDL_GetError());
+    DestroyAudioOutput();
+    return -1;
+  }
+  next_audio_output_retry_ = {};
   return 0;
 }
 
@@ -630,11 +640,15 @@ void SessionDeviceManager::ClearAudio(const std::string& remote_id) {
   if (found == output_streams_.end()) return;
   SDL_DestroyAudioStream(found->second);
   output_streams_.erase(found);
+  // An open SDL playback device can keep the Windows audio driver active even
+  // without queued samples. Release it when no peer needs playback anymore.
+  if (output_streams_.empty()) DestroyAudioOutput();
 }
 
 void SessionDeviceManager::PushAudio(const std::string& remote_id,
                                      const char *data, size_t size) {
-  if (!output_device_) return;
+  if (!data || size == 0 || owner_.unattended_host_) return;
+  if (InitializeAudioOutput() != 0) return;
   auto found = output_streams_.find(remote_id);
   if (found == output_streams_.end()) {
     SDL_AudioSpec spec{};
@@ -644,17 +658,20 @@ void SessionDeviceManager::PushAudio(const std::string& remote_id,
     auto* stream = SDL_CreateAudioStream(&spec, &spec);
     if (!stream) {
       LOG_ERROR("Failed to create output stream: {}", SDL_GetError());
+      if (output_streams_.empty()) DestroyAudioOutput();
       return;
     }
     if (!SDL_BindAudioStream(output_device_, stream)) {
       LOG_ERROR("Failed to bind output stream: {}", SDL_GetError());
       SDL_DestroyAudioStream(stream);
+      if (output_streams_.empty()) DestroyAudioOutput();
       return;
     }
     found = output_streams_.emplace(remote_id, stream).first;
   }
   if (!SDL_PutAudioStreamData(found->second, data, static_cast<int>(size))) {
     LOG_ERROR("Failed to push audio data: {}", SDL_GetError());
+    ClearAudio(remote_id);
   }
 }
 
diff --git a/apps/desktop/src/gui/features/devices/session_device_manager.h b/apps/desktop/src/gui/features/devices/session_device_manager.h
index 114864d1..faa42b75 100644
--- a/apps/desktop/src/gui/features/devices/session_device_manager.h
+++ b/apps/desktop/src/gui/features/devices/session_device_manager.h
@@ -79,10 +79,10 @@ public:
   int StartKeyboardCapturer();
   int StopKeyboardCapturer();
 
-  int InitializeAudioOutput();
   int DestroyAudioOutput();
   // Serialized by GuiRuntime::audio_delivery_mutex_. Each peer has its own
   // stream so SDL mixes simultaneous playback and muting clears only that peer.
+  // Open the output on the first frame and close it after the last stream ends.
   void ClearAudio(const std::string& remote_id);
   void PushAudio(const std::string& remote_id, const char *data, size_t size);
 
@@ -102,6 +102,7 @@ public:
   std::vector<HostDisplay> host_display_list() const;
 
 private:
+  int InitializeAudioOutput();
   unsigned ReleaseMouseButtonMask(unsigned mask);
   void DrainCapturedKeyboardInput();
   bool ShouldSendCapturedFrame(std::chrono::steady_clock::time_point now,
@@ -117,6 +118,7 @@ private:
   GuiRuntime &owner_;
   uint32_t output_device_ = 0;
   std::unordered_map<std::string, SDL_AudioStream*> output_streams_;
+  std::chrono::steady_clock::time_point next_audio_output_retry_{};
   ScreenCapturerFactory *screen_capturer_factory_ = nullptr;
   ScreenCapturer *screen_capturer_ = nullptr;
   ScreenSharingPower screen_sharing_power_;