[sendspin] Add codec preference list to the media source (#19047)

Co-authored-by: Jesse Hills <3060199+jesserockz@users.noreply.github.com>

esphome/components/audio/audio_reader.cpp

@@ -58,6 +58,9 @@ esp_err_t AudioReader::add_sink(const std::weak_ptr<ring_buffer::RingBuffer> &ou
   if (current_audio_file_ != nullptr) {
     // A transfer buffer isn't ncessary for a local file
     this->file_ring_buffer_ = output_ring_buffer.lock();
+    if (this->file_ring_buffer_ == nullptr) {
+      return ESP_ERR_INVALID_STATE;
+    }
     return ESP_OK;
   }
 

esphome/components/audio/audio_transfer_buffer.cpp

@@ -51,14 +51,14 @@ void AudioTransferBuffer::increase_buffer_length(size_t bytes) { this->buffer_le
 
 void AudioTransferBuffer::clear_buffered_data() {
   this->buffer_length_ = 0;
-  if (this->ring_buffer_.use_count() > 0) {
+  if (this->ring_buffer_ != nullptr) {
     this->ring_buffer_->reset();
   }
 }
 
 void AudioSinkTransferBuffer::clear_buffered_data() {
   this->buffer_length_ = 0;
-  if (this->ring_buffer_.use_count() > 0) {
+  if (this->ring_buffer_ != nullptr) {
     this->ring_buffer_->reset();
   }
 #ifdef USE_SPEAKER
@@ -69,7 +69,7 @@ void AudioSinkTransferBuffer::clear_buffered_data() {
 }
 
 bool AudioTransferBuffer::has_buffered_data() const {
-  if (this->ring_buffer_.use_count() > 0) {
+  if (this->ring_buffer_ != nullptr) {
     return ((this->ring_buffer_->available() > 0) || (this->available() > 0));
   }
   return (this->available() > 0);
@@ -144,7 +144,7 @@ size_t AudioSourceTransferBuffer::transfer_data_from_source(TickType_t ticks_to_
   size_t bytes_to_read = AudioTransferBuffer::free();
   size_t bytes_read = 0;
   if (bytes_to_read > 0) {
-    if (this->ring_buffer_.use_count() > 0) {
+    if (this->ring_buffer_ != nullptr) {
       bytes_read = this->ring_buffer_->read((void *) this->get_buffer_end(), bytes_to_read, ticks_to_wait);
     }
 
@@ -161,7 +161,7 @@ size_t AudioSinkTransferBuffer::transfer_data_to_sink(TickType_t ticks_to_wait,
       bytes_written = this->speaker_->play(this->data_start_, this->available(), ticks_to_wait);
     } else
 #endif
-        if (this->ring_buffer_.use_count() > 0) {
+        if (this->ring_buffer_ != nullptr) {
       bytes_written =
           this->ring_buffer_->write_without_replacement((void *) this->data_start_, this->available(), ticks_to_wait);
     } else if (this->sink_callback_ != nullptr) {
@@ -186,7 +186,7 @@ bool AudioSinkTransferBuffer::has_buffered_data() const {
     return (this->speaker_->has_buffered_data() || (this->available() > 0));
   }
 #endif
-  if (this->ring_buffer_.use_count() > 0) {
+  if (this->ring_buffer_ != nullptr) {
     return ((this->ring_buffer_->available() > 0) || (this->available() > 0));
   }
   return (this->available() > 0);

esphome/components/esphome/ota/ota_esphome.cpp

@@ -41,7 +41,10 @@ const noise::NoiseContext &ESPHomeOTAComponent::noise_context_() const {
 #endif
 static constexpr uint16_t OTA_BLOCK_SIZE = 8192;
 static constexpr uint32_t OTA_SOCKET_TIMEOUT_HANDSHAKE = 20000;  // milliseconds for initial handshake
-static constexpr uint32_t OTA_SOCKET_TIMEOUT_DATA = 90000;       // milliseconds for data transfer
+// Milliseconds for data transfer. Covers the lwIP retransmit run seen in
+// practice for a lost chunk ack (1.5 + 3 + 6 + 12 + 24 + 48 s); the CLI waits
+// longer (espota2.DATA_PHASE_TIMEOUT) so the device is free before it retries
+static constexpr uint32_t OTA_SOCKET_TIMEOUT_DATA = 105000;
 
 // Single-instance pointer — multi-port configs are rejected in final_validate.
 // NOLINTNEXTLINE(cppcoreguidelines-avoid-non-const-global-variables)

esphome/components/i2s_audio/speaker/i2s_audio_speaker.cpp

@@ -118,21 +118,24 @@ void I2SAudioSpeakerBase::loop() {
         break;
       }
 
+      // Still starting up or winding down from a previous run
+      if ((this->tx_handle_ != nullptr) || (this->speaker_task_handle_ != nullptr)) {
+        break;
+      }
+
       if (this->start_i2s_driver(this->audio_stream_info_) != ESP_OK) {
         ESP_LOGE(TAG, "Driver failed to start; retrying in 1 second");
         this->status_momentary_error("driver-failure", 1000);
         break;
       }
 
-      if (this->speaker_task_handle_ == nullptr) {
+      xTaskCreate(I2SAudioSpeakerBase::speaker_task, "speaker_task", TASK_STACK_SIZE, (void *) this, TASK_PRIORITY,
-        xTaskCreate(I2SAudioSpeakerBase::speaker_task, "speaker_task", TASK_STACK_SIZE, (void *) this, TASK_PRIORITY,
+                  &this->speaker_task_handle_);
-                    &this->speaker_task_handle_);
 
-        if (this->speaker_task_handle_ == nullptr) {
+      if (this->speaker_task_handle_ == nullptr) {
-          ESP_LOGE(TAG, "Task failed to start, retrying in 1 second");
+        ESP_LOGE(TAG, "Task failed to start, retrying in 1 second");
-          this->status_momentary_error("task-failure", 1000);
+        this->status_momentary_error("task-failure", 1000);
-          this->stop_i2s_driver_();  // Stops the driver to return the lock; will be reloaded in next attempt
+        this->stop_i2s_driver_();  // Stops the driver to return the lock; will be reloaded in next attempt
-        }
       }
       break;
     case speaker::STATE_RUNNING:   // Intentional fallthrough
@@ -218,8 +221,8 @@ size_t I2SAudioSpeakerBase::play(const uint8_t *data, size_t length, TickType_t
 }
 
 bool I2SAudioSpeakerBase::has_buffered_data() const {
-  if (this->audio_ring_buffer_.use_count() > 0) {
+  std::shared_ptr<ring_buffer::RingBuffer> temp_ring_buffer = this->audio_ring_buffer_.lock();
-    std::shared_ptr<ring_buffer::RingBuffer> temp_ring_buffer = this->audio_ring_buffer_.lock();
+  if (temp_ring_buffer != nullptr) {
     return temp_ring_buffer->available() > 0;
   }
   return false;

esphome/components/micro_wake_word/micro_wake_word.cpp

@@ -129,7 +129,7 @@ void MicroWakeWord::setup() {
       return;
     }
     std::shared_ptr<ring_buffer::RingBuffer> temp_ring_buffer = this->ring_buffer_.lock();
-    if (this->ring_buffer_.use_count() > 1) {
+    if (temp_ring_buffer != nullptr) {
       // Producer-only write: never touches consumer state. If the buffer is full, ask the inference task
       // to drain it - reset() is a consumer operation and must run on the inference task's thread.
       // Disable partial writes so audio chunks are either fully accepted or rejected and handled below.
@@ -446,9 +446,9 @@ void MicroWakeWord::loop() {
     xEventGroupClearBits(this->event_group_, EventGroupBits::TASK_STOPPING);
   }
 
-  if ((event_group_bits & EventGroupBits::TASK_STOPPED)) {
+  // Retries on a subsequent loop if the task is still running on the other core
+  if ((event_group_bits & EventGroupBits::TASK_STOPPED) && this->inference_task_.deallocate()) {
     ESP_LOGD(TAG, "Inference task is finished, freeing task resources");
-    this->inference_task_.deallocate();
     xEventGroupClearBits(this->event_group_, ALL_BITS);
     xQueueReset(this->detection_queue_);
     this->set_state_(State::STOPPED);

esphome/components/microphone/microphone_source.h

@@ -48,7 +48,7 @@ class MicrophoneSource final {
   template<typename F> void add_data_callback(F &&data_callback) {
     this->mic_->add_data_callback([this, data_callback](const std::vector<uint8_t> &data) {
       if (this->enabled_ || this->passive_) {
-        if (this->processed_samples_.use_count() == 0) {
+        if (this->processed_samples_ == nullptr) {
           // Create vector if its unused
           this->processed_samples_ = std::make_shared<std::vector<uint8_t>>();
         }

esphome/components/mixer/speaker/mixer_speaker.cpp

@@ -218,7 +218,7 @@ size_t SourceSpeaker::play(const uint8_t *data, size_t length, TickType_t ticks_
   }
   size_t bytes_written = 0;
   std::shared_ptr<ring_buffer::RingBuffer> temp_ring_buffer = this->ring_buffer_.lock();
-  if (temp_ring_buffer.use_count() > 0) {
+  if (temp_ring_buffer != nullptr) {
     // Only write to the ring buffer if the reference is valid
     bytes_written = temp_ring_buffer->write_without_replacement(data, length, ticks_to_wait);
     if (bytes_written > 0) {
@@ -250,14 +250,14 @@ esp_err_t SourceSpeaker::start_() {
   // avoids unnecessary single-frame splices.
   const size_t ring_buffer_size =
       (this->audio_stream_info_.ms_to_bytes(this->buffer_duration_ms_) / bytes_per_frame) * bytes_per_frame;
-  if (this->audio_source_.use_count() == 0) {
+  if (this->audio_source_ == nullptr) {
     std::shared_ptr<ring_buffer::RingBuffer> temp_ring_buffer = this->ring_buffer_.lock();
-    if (!temp_ring_buffer) {
+    if (temp_ring_buffer == nullptr) {
       temp_ring_buffer = ring_buffer::RingBuffer::create(ring_buffer_size);
       this->ring_buffer_ = temp_ring_buffer;
     }
 
-    if (!temp_ring_buffer) {
+    if (temp_ring_buffer == nullptr) {
       return ESP_ERR_NO_MEM;
     }
 
@@ -278,7 +278,7 @@ void SourceSpeaker::stop() { this->send_command_(SOURCE_SPEAKER_COMMAND_STOP); }
 void SourceSpeaker::finish() { this->send_command_(SOURCE_SPEAKER_COMMAND_FINISH); }
 
 bool SourceSpeaker::has_buffered_data() const {
-  return ((this->audio_source_.use_count() > 0) && this->audio_source_->has_buffered_data());
+  return ((this->audio_source_ != nullptr) && this->audio_source_->has_buffered_data());
 }
 
 void SourceSpeaker::set_mute_state(bool mute_state) {
@@ -382,8 +382,8 @@ void MixerSpeaker::loop() {
     ESP_LOGV(TAG, "Stopping");
     xEventGroupClearBits(this->event_group_, MIXER_TASK_STATE_STOPPING);
   }
-  if (event_group_bits & MIXER_TASK_STATE_STOPPED) {
+  // Retries on a subsequent loop if the task is still running on the other core
-    this->task_.deallocate();
+  if ((event_group_bits & MIXER_TASK_STATE_STOPPED) && this->task_.deallocate()) {
     ESP_LOGD(TAG, "Stopped");
     xEventGroupClearBits(this->event_group_, MIXER_TASK_ALL_BITS);
     this->all_stopped_since_ms_ = 0;
@@ -496,7 +496,7 @@ void MixerSpeaker::audio_mixer_task(void *params) {
         if (speaker->is_running() && !speaker->get_pause_state()) {
           // Speaker is running and not paused, so it possibly can provide audio data
           std::shared_ptr<audio::RingBufferAudioSource> audio_source = speaker->get_audio_source().lock();
-          if (audio_source.use_count() == 0) {
+          if (audio_source == nullptr) {
             // No audio source allocated, so skip processing this speaker
             continue;
           }

esphome/components/noise/__init__.py

@@ -88,12 +88,12 @@ def encryption_schema(config: ConfigType | None) -> ConfigType:
 
 async def to_code(config: ConfigType) -> None:
     cg.add_define("USE_NOISE")
-    cg.add_library("esphome/noise-c", "0.1.24")
+    cg.add_library("esphome/noise-c", "0.1.26")
     # noise-c depends on libsodium, but declaring it here too lets the
     # library manager see the full set up front instead of discovering
     # libsodium only after noise-c has downloaded, so the two can download
     # in parallel. The version must match noise-c's library.json.
-    cg.add_library("esphome/libsodium", "1.10021.6")
+    cg.add_library("esphome/libsodium", "1.10021.8")
     # Enable optimized memzero/memcmp in libsodium instead of volatile byte loops
     cg.add_build_flag("-DHAVE_WEAK_SYMBOLS=1")
     cg.add_build_flag("-DHAVE_INLINE_ASM=1")

esphome/components/resampler/speaker/resampler_speaker.cpp

@@ -153,8 +153,8 @@ void ResamplerSpeaker::loop() {
     ESP_LOGV(TAG, "Stopping");
     xEventGroupClearBits(this->event_group_, ResamplingEventGroupBits::STATE_STOPPING);
   }
-  if (event_group_bits & ResamplingEventGroupBits::STATE_STOPPED) {
+  // Retries on a subsequent loop if the task is still running on the other core
-    this->task_.deallocate();
+  if ((event_group_bits & ResamplingEventGroupBits::STATE_STOPPED) && this->task_.deallocate()) {
     ESP_LOGD(TAG, "Stopped");
     xEventGroupClearBits(this->event_group_, ResamplingEventGroupBits::ALL_BITS);
   }
@@ -235,7 +235,7 @@ size_t ResamplerSpeaker::play(const uint8_t *data, size_t length, TickType_t tic
     bytes_written = this->output_speaker_->play(data, length, ticks_to_wait);
   } else {
     std::shared_ptr<ring_buffer::RingBuffer> temp_ring_buffer = this->ring_buffer_.lock();
-    if (temp_ring_buffer) {
+    if (temp_ring_buffer != nullptr) {
       // Only write to the ring buffer if the reference is valid
       bytes_written = temp_ring_buffer->write_without_replacement(data, length, ticks_to_wait);
     } else {
@@ -299,7 +299,7 @@ bool ResamplerSpeaker::has_buffered_data() const {
   bool has_ring_buffer_data = false;
   if (this->requires_resampling_()) {
     std::shared_ptr<ring_buffer::RingBuffer> temp_ring_buffer = this->ring_buffer_.lock();
-    if (temp_ring_buffer) {
+    if (temp_ring_buffer != nullptr) {
       has_ring_buffer_data = (temp_ring_buffer->available() > 0);
     }
   }
@@ -342,7 +342,7 @@ void ResamplerSpeaker::resample_task(void *params) {
       std::shared_ptr<ring_buffer::RingBuffer> temp_ring_buffer = ring_buffer::RingBuffer::create(
           this_resampler->audio_stream_info_.ms_to_bytes(this_resampler->buffer_duration_ms_));
 
-      if (!temp_ring_buffer) {
+      if (temp_ring_buffer == nullptr) {
         err = ESP_ERR_NO_MEM;
       } else {
         this_resampler->ring_buffer_ = temp_ring_buffer;

esphome/components/sendspin/__init__.py

@@ -30,6 +30,7 @@ CONF_SENDSPIN_ID = "sendspin_id"
 CONF_INITIAL_STATIC_DELAY = "initial_static_delay"
 CONF_FIXED_DELAY = "fixed_delay"
 CONF_DECODE_MEMORY = "decode_memory"
+CONF_CODECS = "codecs"
 
 # Matches ARTWORK_MAX_SLOTS in sendspin-cpp.
 MAX_ARTWORK_SLOTS = 4
@@ -44,6 +45,20 @@ CODEC_FORMAT_OPUS = SendspinCodecFormat.enum("OPUS")
 CODEC_FORMAT_PCM = SendspinCodecFormat.enum("PCM")
 CODEC_FORMAT_UNSUPPORTED = SendspinCodecFormat.enum("UNSUPPORTED")
 
+CODEC_FLAC = "flac"
+CODEC_OPUS = "opus"
+CODEC_PCM = "pcm"
+
+CODECS = {
+    CODEC_FLAC: CODEC_FORMAT_FLAC,
+    CODEC_OPUS: CODEC_FORMAT_OPUS,
+    CODEC_PCM: CODEC_FORMAT_PCM,
+}
+
+# Opus only supports 48 kHz audio, so it is left out of the default list at other rates.
+DEFAULT_CODECS = [CODEC_FLAC, CODEC_OPUS, CODEC_PCM]
+OPUS_SAMPLE_RATE = 48000
+
 SendspinImageFormat = sendspin_library_ns.enum("SendspinImageFormat", is_class=True)
 IMAGE_FORMAT_JPEG = SendspinImageFormat.enum("JPEG")
 IMAGE_FORMAT_PNG = SendspinImageFormat.enum("PNG")
@@ -286,16 +301,13 @@ async def to_code(config: ConfigType) -> None:
     if data.player_support:
         cg.add_define("USE_SENDSPIN_PLAYER", True)
 
-        # Configures the player role. We always assume support for 16 bits per sample mono and stereo FLAC, Opus, and PCM at the configured sample rate
+        # Configures the player role. Each configured codec is advertised for 16 bits per sample
-        # (with Opus only supported at 48 kHz since that's the only sample rate it supports). Users can configure the specific formats via the Sendspin server
+        # mono and stereo at the configured sample rate. The order is a preference order, both for
+        # the codecs themselves and for stereo over mono.
         player_cfg = data.player_config
         sample_rate = player_cfg[CONF_SAMPLE_RATE]
 
-        # OPUS only supports 48 kHz audio
+        codecs = player_cfg[CONF_CODECS]
-        codecs = [CODEC_FORMAT_FLAC]
-        if sample_rate == 48000:
-            codecs.append(CODEC_FORMAT_OPUS)
-        codecs.append(CODEC_FORMAT_PCM)
 
         def _audio_format(codec: MockObj, channels: int) -> cg.StructInitializer:
             return cg.StructInitializer(

esphome/components/sendspin/media_source/__init__.py

@@ -13,11 +13,16 @@ from esphome.cpp_generator import MockObj, TemplateArgsType
 from esphome.types import ConfigType
 
 from .. import (
+    CODEC_OPUS,
+    CODECS,
+    CONF_CODECS,
     CONF_DECODE_MEMORY,
     CONF_FIXED_DELAY,
     CONF_INITIAL_STATIC_DELAY,
     CONF_SENDSPIN_ID,
+    DEFAULT_CODECS,
     MEMORY_LOCATIONS,
+    OPUS_SAMPLE_RATE,
     SendspinHub,
     register_player_config,
     request_controller_support,
@@ -49,10 +54,32 @@ DisableStaticDelayAdjustmentAction = sendspin_ns.class_(
 )
 
 
+def _resolve_codecs(config: ConfigType) -> ConfigType:
+    """Validate the codec preference list, filling in the default when it is not set."""
+    sample_rate = config[CONF_SAMPLE_RATE]
+    if (codecs := config.get(CONF_CODECS)) is None:
+        config[CONF_CODECS] = [
+            codec
+            for codec in DEFAULT_CODECS
+            if codec != CODEC_OPUS or sample_rate == OPUS_SAMPLE_RATE
+        ]
+        return config
+
+    if len(set(codecs)) != len(codecs):
+        raise cv.Invalid("Each codec may only be listed once", path=[CONF_CODECS])
+    if CODEC_OPUS in codecs and sample_rate != OPUS_SAMPLE_RATE:
+        raise cv.Invalid(
+            f"Codec '{CODEC_OPUS}' requires a {CONF_SAMPLE_RATE} of {OPUS_SAMPLE_RATE}",
+            path=[CONF_CODECS],
+        )
+    return config
+
+
 def _register(config: ConfigType) -> ConfigType:
     request_controller_support()
     register_player_config(
         {
+            CONF_CODECS: config[CONF_CODECS],
             CONF_SAMPLE_RATE: config[CONF_SAMPLE_RATE],
             CONF_BUFFER_SIZE: config[CONF_BUFFER_SIZE],
             CONF_INITIAL_STATIC_DELAY: config[CONF_INITIAL_STATIC_DELAY],
@@ -85,9 +112,13 @@ CONFIG_SCHEMA = cv.All(
                 min=16000, max=96000
             ),
             cv.Optional(CONF_DECODE_MEMORY): cv.one_of(*MEMORY_LOCATIONS, lower=True),
+            cv.Optional(CONF_CODECS): cv.All(
+                cv.ensure_list(cv.enum(CODECS, lower=True)), cv.Length(min=1)
+            ),
         }
     ),
     cv.only_on_esp32,
+    _resolve_codecs,
     _register,
 )
 

esphome/components/speaker/media_player/audio_pipeline.cpp

@@ -202,8 +202,15 @@ AudioPipelineState AudioPipeline::process_state() {
     if (!this->is_playing_) {
       // The tasks have been stopped for two ``process_state`` calls in a row, so delete the tasks
       if (this->read_task_.is_created() || this->decode_task_.is_created()) {
-        this->read_task_.deallocate();
+        // Both are attempted every time; a task that is still running on the other core is freed by a
-        this->decode_task_.deallocate();
+        // subsequent call, and freeing an already freed task succeeds without doing anything
+        bool read_task_freed = this->read_task_.deallocate();
+        bool decode_task_freed = this->decode_task_.deallocate();
+        if (!read_task_freed || !decode_task_freed) {
+          // A task is still running on the other core, so keep the pipeline in its current state and try
+          // again on the next call
+          return AudioPipelineState::PLAYING;
+        }
         if (this->hard_stop_) {
           // Stop command was sent, so immediately end the playback
           this->speaker_->stop();
@@ -315,17 +322,17 @@ void AudioPipeline::read_task(void *params) {
       if (err == ESP_OK) {
         size_t file_ring_buffer_size = this_pipeline->buffer_size_;
 
-        std::shared_ptr<ring_buffer::RingBuffer> temp_ring_buffer;
+        std::shared_ptr<ring_buffer::RingBuffer> temp_ring_buffer = this_pipeline->raw_file_ring_buffer_.lock();
 
-        if (!this_pipeline->raw_file_ring_buffer_.use_count()) {
+        if (temp_ring_buffer == nullptr) {
           temp_ring_buffer = ring_buffer::RingBuffer::create(file_ring_buffer_size);
           this_pipeline->raw_file_ring_buffer_ = temp_ring_buffer;
         }
 
-        if (!this_pipeline->raw_file_ring_buffer_.use_count()) {
+        if (temp_ring_buffer == nullptr) {
           err = ESP_ERR_NO_MEM;
         } else {
-          reader->add_sink(this_pipeline->raw_file_ring_buffer_);
+          err = reader->add_sink(temp_ring_buffer);
         }
       }
 
@@ -396,7 +403,9 @@ void AudioPipeline::decode_task(void *params) {
           make_unique<audio::AudioDecoder>(this_pipeline->transfer_buffer_size_, this_pipeline->transfer_buffer_size_);
 
       esp_err_t err = decoder->start(this_pipeline->current_audio_file_type_);
-      decoder->add_source(this_pipeline->raw_file_ring_buffer_);
+      if (err == ESP_OK) {
+        err = decoder->add_source(this_pipeline->raw_file_ring_buffer_);
+      }
 
       if (err != ESP_OK) {
         // Send specific error message

esphome/core/static_task.cpp

@@ -40,16 +40,31 @@ bool StaticTask::create(TaskFunction_t fn, const char *name, uint32_t stack_size
   return true;
 }
 
-void StaticTask::destroy() {
+bool StaticTask::destroy() {
-  if (this->handle_ != nullptr) {
+  if (this->handle_ == nullptr) {
-    TaskHandle_t handle = this->handle_;
+    return true;
-    this->handle_ = nullptr;
-    vTaskDelete(handle);
   }
+
+  // Suspending takes the task off the ready and event lists, so nothing can schedule it again. It only asks
+  // the other core to yield though, so the task may still be running on it for a moment.
+  vTaskSuspend(this->handle_);
+  if (eTaskGetState(this->handle_) != eSuspended) {
+    // The task is still running on the other core and using its stack. Deleting it now would only put it on
+    // the termination list and return, so the caller has to try again once it has been swapped out.
+    return false;
+  }
+
+  // The task cannot run again, so the delete completes right away instead of being left to the idle task.
+  TaskHandle_t handle = this->handle_;
+  this->handle_ = nullptr;
+  vTaskDelete(handle);
+  return true;
 }
 
-void StaticTask::deallocate() {
+bool StaticTask::deallocate() {
-  this->destroy();
+  if (!this->destroy()) {
+    return false;
+  }
   if (this->stack_buffer_ != nullptr) {
     RAMAllocator<StackType_t> allocator(this->use_psram_ ? RAMAllocator<StackType_t>::ALLOC_EXTERNAL
                                                          : RAMAllocator<StackType_t>::ALLOC_INTERNAL);
@@ -57,6 +72,7 @@ void StaticTask::deallocate() {
     this->stack_buffer_ = nullptr;
     this->stack_size_ = 0;
   }
+  return true;
 }
 
 }  // namespace esphome

esphome/core/static_task.h

@@ -11,6 +11,7 @@ namespace esphome {
 
 /** Helper for FreeRTOS static task management.
  * Bundles TaskHandle_t, StaticTask_t, and the stack buffer into one object with create/destroy methods.
+ * Call destroy() and deallocate() from another task: a task cannot free the stack it is still running on.
  */
 class StaticTask {
  public:
@@ -23,7 +24,7 @@ class StaticTask {
   /// @brief Allocate stack and create task.
   /// @param fn         Task function
   /// @param name       Task name (for debug)
-  /// @param stack_size Stack size in StackType_t words
+  /// @param stack_size Stack size in bytes (StackType_t is a byte on ESP-IDF)
   /// @param param      Parameter passed to task function
   /// @param priority   FreeRTOS task priority
   /// @param use_psram  If true, allocate stack in PSRAM; otherwise internal RAM
@@ -31,11 +32,17 @@ class StaticTask {
   bool create(TaskFunction_t fn, const char *name, uint32_t stack_size, void *param, UBaseType_t priority,
               bool use_psram);
 
-  /// @brief Delete the task but keep the stack buffer allocated for reuse by a subsequent create() call.
+  /// @brief Delete the task, keeping the stack buffer allocated for reuse by a subsequent create() call.
-  void destroy();
+  /// The task must have finished its work and parked itself, either suspended or blocked indefinitely: it is
+  /// suspended here so that it cannot be scheduled again, and it is given no chance to clean up.
+  /// @return true if the task was deleted; false if it is still running on another core, in which case the
+  /// caller should try again later.
+  bool destroy();
 
-  /// @brief Delete the task (if running) and free the stack buffer.
+  /// @brief Delete the task (if created) and free the stack buffer.
-  void deallocate();
+  /// @return true if the stack buffer was freed; false if the task is still running on another core, in
+  /// which case the caller should try again later.
+  bool deallocate();
 
  protected:
   TaskHandle_t handle_{nullptr};

esphome/espota2.py

@@ -96,6 +96,10 @@ UPLOAD_BUFFER_SIZE = UPLOAD_BLOCK_SIZE * 8
 # across the addresses on top of that.
 EXTRA_UPLOAD_ATTEMPTS = 2
 UPLOAD_RETRY_DELAY = 5.0
+# Data phase timeout; must stay longer than the device's OTA_SOCKET_TIMEOUT_DATA
+# (105 s) so a stalled session is gone before a retry, and long enough for lwIP
+# to get a lost chunk ack through after the retransmit run seen in practice
+DATA_PHASE_TIMEOUT = 160.0
 
 _LOGGER = logging.getLogger(__name__)
 
@@ -694,8 +698,7 @@ def perform_ota(
 
     _LOGGER.info("Handshake complete")
 
-    # Timeout must match device-side OTA_SOCKET_TIMEOUT_DATA to prevent premature failures
+    sock.settimeout(DATA_PHASE_TIMEOUT)
-    sock.settimeout(90.0)
 
     if extended_proto:
         send_check(sock, ota_type, "ota type")
@@ -854,7 +857,7 @@ def run_ota_impl_(
     # clean up a half-open connection (its handshake watchdog runs at 20s);
     # moving on to the next address family stays immediate. Known limitation:
     # a silent mid-transfer drop with no reset can wedge the device until its
-    # 90s data timeout, which outlasts this budget; the retries target the
+    # 105s data timeout, which outlasts this budget; the retries target the
     # common failures where the device resets or closes the link promptly.
     total_attempts = len(res) + EXTRA_UPLOAD_ATTEMPTS
     last_error = ""

platformio.ini

@@ -45,7 +45,7 @@ lib_deps_base =
 lib_deps =
     ${common.lib_deps_base}
     https://github.com/dudanov/MideaUART.git#eeea6c3e9b4474f067054592b435be1c4e466815 ; midea
-    esphome/noise-c@0.1.24                 ; noise (api, ota)
+    esphome/noise-c@0.1.26                 ; noise (api, ota)
     improv/Improv@1.2.7                    ; improv_serial / esp32_improv
     kikuchan98/pngle@1.1.0                 ; online_image
     ; Using the repository directly, otherwise ESP-IDF can't use the library
@@ -244,7 +244,7 @@ lib_deps =
     ${common:idf-component-libs.lib_deps}
     ESP32Async/ESPAsyncWebServer@3.9.6    ; web_server_base
     droscy/esp_wireguard@0.4.5            ; wireguard
-    esphome/noise-c@0.1.24                ; noise (api, ota)
+    esphome/noise-c@0.1.26                ; noise (api, ota)
     ESP32Async/AsyncTCP@3.4.5             ; async_tcp
     DNSServer                             ; captive_portal
     heman/AsyncMqttClient-esphome@2.0.0   ; mqtt
@@ -641,7 +641,7 @@ build_unflags =
 extends = common
 platform = platformio/native
 lib_deps =
-    esphome/noise-c@0.1.24  ; used by noise (api, ota)
+    esphome/noise-c@0.1.26  ; used by noise (api, ota)
     lvgl/lvgl@9.5.0         ; lvgl
 build_flags =
     ${common.build_flags}

requirements.txt

@@ -10,7 +10,7 @@ tzlocal==5.4.4    # from time
 tzdata>=2026.3  # from time
 pyserial==3.5
 platformio==6.1.19
-esptool==5.3.1
+esptool==5.4.0
 click==8.3.3
 aioesphomeapi==46.3.0
 aiohappyeyeballs==2.7.1  # Happy Eyeballs for requests downloads; already pulled in by aioesphomeapi

tests/component_tests/sendspin/test_media_source.py

@@ -0,0 +1,90 @@
+"""Validation tests for the sendspin media_source platform.
+
+These cover the codec preference list, whose rejection branches a compile test
+cannot reach: a `test*.yaml` can only assert that a configuration is accepted.
+"""
+
+from typing import Any
+
+import pytest
+
+from esphome import config_validation as cv
+from esphome.components.sendspin import CONF_CODECS, _get_data
+from esphome.components.sendspin.media_source import CONFIG_SCHEMA
+from esphome.const import PlatformFramework
+from esphome.types import ConfigType
+from tests.component_tests.types import SetCoreConfigCallable
+
+
+def _media_source_config(**overrides: Any) -> ConfigType:
+    """Build a minimal valid media source config, allowing field overrides."""
+    config: ConfigType = {
+        "id": "sendspin_media_source",
+        "sendspin_id": "sendspin_hub",
+    }
+    config.update(overrides)
+    return config
+
+
+def test_default_codecs_at_48_khz(set_core_config: SetCoreConfigCallable) -> None:
+    """Every codec is advertised when the sample rate suits all of them."""
+    set_core_config(PlatformFramework.ESP32_IDF)
+
+    config = CONFIG_SCHEMA(_media_source_config())
+
+    assert config[CONF_CODECS] == ["flac", "opus", "pcm"]
+
+
+def test_default_codecs_drop_opus_at_other_rates(
+    set_core_config: SetCoreConfigCallable,
+) -> None:
+    """Opus only supports 48 kHz, so it leaves the default list at other rates."""
+    set_core_config(PlatformFramework.ESP32_IDF)
+
+    config = CONFIG_SCHEMA(_media_source_config(sample_rate=44100))
+
+    assert config[CONF_CODECS] == ["flac", "pcm"]
+
+
+def test_configured_order_is_preserved(set_core_config: SetCoreConfigCallable) -> None:
+    """The list is a preference order, so it reaches the player role as written."""
+    set_core_config(PlatformFramework.ESP32_IDF)
+
+    CONFIG_SCHEMA(_media_source_config(codecs=["pcm", "flac"]))
+
+    assert _get_data().player_config[CONF_CODECS] == ["pcm", "flac"]
+
+
+def test_empty_codec_list_rejected(set_core_config: SetCoreConfigCallable) -> None:
+    """A player with no codecs at all could never be given a stream."""
+    set_core_config(PlatformFramework.ESP32_IDF)
+
+    with pytest.raises(cv.Invalid, match="length of value must be at least 1"):
+        CONFIG_SCHEMA(_media_source_config(codecs=[]))
+
+
+def test_duplicate_codec_rejected(set_core_config: SetCoreConfigCallable) -> None:
+    """A repeated codec has no meaning in a preference order."""
+    set_core_config(PlatformFramework.ESP32_IDF)
+
+    with pytest.raises(cv.Invalid, match="may only be listed once"):
+        CONFIG_SCHEMA(_media_source_config(codecs=["flac", "flac"]))
+
+
+def test_unknown_codec_rejected(set_core_config: SetCoreConfigCallable) -> None:
+    """Only codecs the player role can decode are accepted."""
+    set_core_config(PlatformFramework.ESP32_IDF)
+
+    with pytest.raises(cv.Invalid, match="Unknown value"):
+        CONFIG_SCHEMA(_media_source_config(codecs=["mp3"]))
+
+
+def test_opus_at_wrong_sample_rate_rejected(
+    set_core_config: SetCoreConfigCallable,
+) -> None:
+    """Asking for Opus at a rate it cannot handle fails rather than silently
+    dropping the stated preference."""
+    set_core_config(PlatformFramework.ESP32_IDF)
+
+    with pytest.raises(cv.Invalid, match="requires a sample_rate of 48000"):
+        CONFIG_SCHEMA(_media_source_config(codecs=["opus"], sample_rate=44100))

tests/components/sendspin/common-media_source.yaml

@@ -9,3 +9,4 @@ media_source:
     static_delay_adjustable: true
     fixed_delay: 480us
     decode_memory: internal
+    codecs: [pcm, opus, flac]

tests/script/test_platformio_install_deps.py

@@ -35,8 +35,8 @@ def _load_script():
 def test_spec_key_collapses_destinations() -> None:
     """Two specs delivering one package share a directory and one key."""
     mod = _load_script()
-    assert mod.spec_key("esphome/noise-c @ 0.1.24") == "noise-c"
+    assert mod.spec_key("esphome/noise-c @ 0.1.26") == "noise-c"
-    assert mod.spec_key("esphome/noise-c@0.1.24") == "noise-c"
+    assert mod.spec_key("esphome/noise-c@0.1.26") == "noise-c"
     assert mod.spec_key("ESP32Async/AsyncTCP @ ^3.4.10") == mod.spec_key(
         "esp32async/asynctcp @ 3.5.0"
     )
@@ -54,23 +54,23 @@ def test_parse_specs_and_cli_args(tmp_path: Path) -> None:
         "[env:a]\n"
         "platform = fake/platform@1\n"
         "lib_deps =\n"
-        "    esphome/noise-c @ 0.1.24\n"
+        "    esphome/noise-c @ 0.1.26\n"
         "    ${common.lib_deps}\n"
         "    internal_lib\n"
         "[env:b]\n"
         "lib_deps =\n"
-        "    esphome/noise-c @ 0.1.24\n"
+        "    esphome/noise-c @ 0.1.26\n"
     )
     mod = _load_script()
     args = Namespace(libraries=True, platforms=True, tools=False)
     libs, platforms, tools = mod.parse_specs(str(ini), args)
     # exact-string duplicates collapse; distinct version pins survive
-    assert libs == ["esphome/noise-c @ 0.1.24"]
+    assert libs == ["esphome/noise-c @ 0.1.26"]
     assert platforms == ["fake/platform@1"]
     assert tools == []
     assert mod.build_cli_args(libs, platforms, tools) == [
         "-l",
-        "esphome/noise-c @ 0.1.24",
+        "esphome/noise-c @ 0.1.26",
         "-p",
         "fake/platform@1",
     ]
@@ -162,13 +162,13 @@ def test_parallel_install_behavior(tmp_path: Path) -> None:
     mod.parallel_install(
         cls,
         [
-            "esphome/noise-c @ 0.1.24",
+            "esphome/noise-c @ 0.1.26",
-            "esphome/noise-c @ 0.1.24",
+            "esphome/noise-c @ 0.1.26",
             "esphome/already @ 1.0",
             "https://x/framework.tar.xz",
         ],
     )
-    assert cls.calls == ["esphome/noise-c @ 0.1.24"]
+    assert cls.calls == ["esphome/noise-c @ 0.1.26"]
     assert cls.lock_events == ["lock", "unlock"]
 
 
@@ -205,7 +205,7 @@ def test_parallel_install_runs_dependency_waves(tmp_path: Path) -> None:
     mod = _load_script()
     cls = _reset_fake(str(tmp_path))
     cls.deps = {
-        "esphome/noise-c @ 0.1.24": [
+        "esphome/noise-c @ 0.1.26": [
             {"owner": "esphome", "name": "libsodium", "version": "^1.0"},
             {"name": "SPI"},
         ],
@@ -213,12 +213,12 @@ def test_parallel_install_runs_dependency_waves(tmp_path: Path) -> None:
             {"owner": "esphome", "name": "libsodium", "version": "^1.0"},
         ],
     }
-    mod.parallel_install(cls, ["esphome/noise-c @ 0.1.24", "esphome/wg @ 1.0"])
+    mod.parallel_install(cls, ["esphome/noise-c @ 0.1.26", "esphome/wg @ 1.0"])
     assert len(cls.calls) == 3  # the shared dep installs exactly once
     assert {mod.spec_key(c) for c in cls.calls} == {"noise-c", "wg", "libsodium"}
     # Wave-1 strings carry no compatibility; the dependency wave does
     compats = dict(cls.compat_calls)
-    assert compats["esphome/noise-c @ 0.1.24"] is None
+    assert compats["esphome/noise-c @ 0.1.26"] is None
     dep_compat = next(v for k, v in cls.compat_calls if "libsodium" in k)
     assert dep_compat is not None  # mirrors pio's install_dependency
 
@@ -229,11 +229,11 @@ def test_dependency_wave_excludes_url_specs(tmp_path: Path) -> None:
     mod = _load_script()
     cls = _reset_fake(str(tmp_path))
     cls.deps = {
-        "esphome/noise-c @ 0.1.24": [
+        "esphome/noise-c @ 0.1.26": [
             {"name": "vendored", "version": "https://github.com/x/y.git"},
         ],
     }
-    mod.parallel_install(cls, ["esphome/noise-c @ 0.1.24"])
+    mod.parallel_install(cls, ["esphome/noise-c @ 0.1.26"])
     assert {mod.spec_key(c) for c in cls.calls} == {"noise-c"}
 
 
@@ -348,13 +348,13 @@ def test_warm_store_still_walks_dependencies(tmp_path: Path) -> None:
     """Already-installed top-level packages still feed the dependency
     wave; a warm store can be missing a transitive dep."""
     mod = _load_script()
-    cls = _reset_fake(str(tmp_path), installed={"esphome/noise-c @ 0.1.24"})
+    cls = _reset_fake(str(tmp_path), installed={"esphome/noise-c @ 0.1.26"})
     cls.deps = {
-        "esphome/noise-c @ 0.1.24": [
+        "esphome/noise-c @ 0.1.26": [
             {"owner": "esphome", "name": "libsodium", "version": "^1.0"},
         ],
     }
-    mod.parallel_install(cls, ["esphome/noise-c @ 0.1.24"])
+    mod.parallel_install(cls, ["esphome/noise-c @ 0.1.26"])
     assert [mod.spec_key(c) for c in cls.calls] == ["libsodium"]
 
 

tests/unit_tests/test_espota2.py

@@ -416,6 +416,9 @@ def test_perform_ota_no_auth(
         "Update took 14.00 seconds (prepare 2.00, upload 5.00, commit 7.00)"
         in caplog.text
     )
+    # The data phase timeout must outlast the device's 105 s data timeout
+    mock_socket.settimeout.assert_any_call(espota2.DATA_PHASE_TIMEOUT)
+    assert espota2.DATA_PHASE_TIMEOUT > 105.0
 
 
 @pytest.mark.usefixtures("mock_time")

tests/unit_tests/test_platformio_prefetch.py

@@ -1663,7 +1663,7 @@ def test_preinstall_runs_dependency_waves(tmp_path: Path) -> None:
         {"name": "SPI"},
     ]
     m.dependency_to_spec.side_effect = lambda dep: _FakeSpec(name=dep["name"])
-    pf._preinstall(m, [("noise-c@0.1.24", _FakeSpec(name="noise-c"))])
+    pf._preinstall(m, [("noise-c@0.1.26", _FakeSpec(name="noise-c"))])
     assert installed == ["noise-c", "libsodium"]  # dep deduped, SPI left out
     # The dep wave carries its compatibility so _install searches qualified
     dep_call = m._install.call_args_list[-1]
@@ -1683,7 +1683,7 @@ def test_preinstall_dependency_wave_skips_seen_names(tmp_path: Path) -> None:
     m._install.side_effect = lambda spec, skip_dependencies, compatibility=None: (
         installed.append(getattr(spec, "name", str(spec)))
     )
-    pf._preinstall(m, [("noise-c@0.1.24", _FakeSpec(name="noise-c"))])
+    pf._preinstall(m, [("noise-c@0.1.26", _FakeSpec(name="noise-c"))])
     assert installed == ["noise-c"]