back to back flac playback is working :)

src/audio/audio_decoder.cpp

@@ -36,37 +36,27 @@ AudioDecoder::AudioDecoder()
       current_codec_(),
       current_input_format_(),
       current_output_format_(),
+      has_prepared_output_(false),
       has_samples_to_send_(false),
       has_input_remaining_(false) {}
 
 AudioDecoder::~AudioDecoder() {}
 
 auto AudioDecoder::ProcessStreamInfo(const StreamInfo& info) -> bool {
+  has_prepared_output_ = false;
+  current_codec_.reset();
+  current_input_format_.reset();
+  current_output_format_.reset();
+
   if (!std::holds_alternative<StreamInfo::Encoded>(info.format)) {
     return false;
   }
-  ESP_LOGI(kTag, "got new stream");
-  const auto& encoded = std::get<StreamInfo::Encoded>(info.format);
 
-  // Reuse the existing codec if we can. This will help with gapless playback,
-  // since we can potentially just continue to decode as we were before,
-  // without any setup overhead.
-  // TODO(jacqueline): Reconsider this. It makes a lot of things harder to smash
-  // streams together at this layer.
-  /*
-  if (current_codec_ != nullptr && current_input_format_) {
-    auto cur_encoding = std::get<StreamInfo::Encoded>(*current_input_format_);
-    if (cur_encoding.type == encoded.type) {
-      ESP_LOGI(kTag, "reusing existing decoder");
-      current_input_format_ = info.format;
-      return true;
-    }
-  }
-  */
+  const auto& new_format = std::get<StreamInfo::Encoded>(info.format);
   current_input_format_ = info.format;
 
   ESP_LOGI(kTag, "creating new decoder");
-  auto result = codecs::CreateCodecForType(encoded.type);
+  auto result = codecs::CreateCodecForType(new_format.type);
   if (result.has_value()) {
     current_codec_.reset(result.value());
   } else {
@@ -86,18 +76,31 @@ auto AudioDecoder::Process(const std::vector<InputStream>& inputs,
   auto input = inputs.begin();
   const StreamInfo& info = input->info();
 
-  // Check the input stream's format has changed (or, by extension, if this is
-  // the first stream).
-  if (!current_input_format_ || *current_input_format_ != info.format) {
-    has_samples_to_send_ = false;
+  // Is this a completely new stream?
+  if (!current_input_format_) {
     if (!ProcessStreamInfo(info)) {
+      // We couldn't handle the new stream. Signal to the producer that we don't
+      // have anything to do.
+      input->mark_consumer_finished();
       return;
     }
-    ESP_LOGI(kTag, "beginning new stream");
+  }
+
+  // Have we determined what kind of samples this stream decodes to?
+  if (!current_output_format_) {
     auto res = current_codec_->BeginStream(input->data());
     input->consume(res.first);
+
     if (res.second.has_error()) {
-      // TODO(jacqueline): Handle errors.
+      auto err = res.second.error();
+      if (err == codecs::ICodec::Error::kOutOfInput) {
+        // We didn't manage to clear whatever front matter is before this
+        // stream's header. We need to call BeginStream again with more data.
+        return;
+      }
+      // Somthing about the stream's header was malformed. Skip it.
+      ESP_LOGE(kTag, "error beginning stream");
+      input->mark_consumer_finished();
       return;
     }
 
@@ -116,59 +119,82 @@ auto AudioDecoder::Process(const std::vector<InputStream>& inputs,
     }
   }
 
-  while (seek_to_sample_) {
+  if (seek_to_sample_) {
     ESP_LOGI(kTag, "seeking forwards...");
     auto res = current_codec_->SeekStream(input->data(), *seek_to_sample_);
     input->consume(res.first);
+
     if (res.second.has_error()) {
       auto err = res.second.error();
       if (err == codecs::ICodec::Error::kOutOfInput) {
         return;
       } else {
         // TODO(jacqueline): Handle errors.
-        seek_to_sample_.reset();
       }
-    } else {
-      seek_to_sample_.reset();
     }
+
+    seek_to_sample_.reset();
   }
 
   has_input_remaining_ = true;
   while (true) {
+    // Make sure the output buffer is ready to receive samples in our format
+    // before starting to process data.
     // TODO(jacqueline): Pass through seek info here?
-    if (!output->prepare(*current_output_format_)) {
+    if (!has_prepared_output_ && !output->prepare(*current_output_format_)) {
       ESP_LOGI(kTag, "waiting for buffer to become free");
-      break;
-    }
-
-    auto res = current_codec_->ContinueStream(input->data(), output->data());
-    input->consume(res.first);
-    if (res.second.has_error()) {
-      if (res.second.error() == codecs::ICodec::Error::kOutOfInput) {
-        ESP_LOGW(kTag, "out of input");
-        ESP_LOGW(kTag, "(%u bytes left)", input->data().size_bytes());
-        has_input_remaining_ = false;
-        // We can't be halfway through sending samples if the codec is asking
-        // for more input.
-        has_samples_to_send_ = false;
-        input->mark_incomplete();
-      } else {
-        // TODO(jacqueline): Handle errors.
-        ESP_LOGE(kTag, "codec return fatal error");
-      }
       return;
     }
+    has_prepared_output_ = true;
 
-    codecs::ICodec::OutputInfo out_info = res.second.value();
-    output->add(out_info.bytes_written);
-    has_samples_to_send_ = !out_info.is_finished_writing;
+    // Parse frames and produce samples.
+    auto res = current_codec_->ContinueStream(input->data(), output->data());
+    input->consume(res.first);
 
-    if (has_samples_to_send_) {
-      // We weren't able to fit all the generated samples into the output
-      // buffer. Stop trying; we'll finish up during the next pass.
-      break;
+    // Handle any errors during processing.
+    if (res.second.has_error()) {
+      // The codec ran out of input during processing. This is expected to
+      // happen throughout the stream.
+      if (res.second.error() == codecs::ICodec::Error::kOutOfInput) {
+        ESP_LOGI(kTag, "codec needs more data");
+        has_input_remaining_ = false;
+        has_samples_to_send_ = false;
+        if (input->is_producer_finished()) {
+          ESP_LOGI(kTag, "codec is all done.");
+
+          // We're out of data, and so is the producer. Nothing left to be done
+          // with the input stream.
+          input->mark_consumer_finished();
+
+          // Upstream isn't going to give us any more data. Tell downstream
+          // that they shouldn't expact any more samples from this stream.
+          output->mark_producer_finished();
+          break;
+        }
+      } else {
+        // TODO(jacqueline): Handle errors.
+        ESP_LOGE(kTag, "codec returned fatal error");
+      }
+      // Note that a codec that returns an error is not allowed to write
+      // samples. So it's safe to skip the latter part of the loop.
+      return;
+    } else {
+      // Some samples were written! Ensure the downstream element knows about
+      // them.
+      codecs::ICodec::OutputInfo out_info = res.second.value();
+      output->add(out_info.bytes_written);
+      has_samples_to_send_ = !out_info.is_finished_writing;
+
+      if (has_samples_to_send_) {
+        // The codec wasn't able to finish writing all of its samples into the
+        // output buffer. We need to return so that we can get a new buffer.
+        return;
+      }
     }
   }
+
+  current_codec_.reset();
+  current_input_format_.reset();
 }
 
 }  // namespace audio

src/audio/audio_task.cpp

@@ -125,8 +125,13 @@ void AudioTaskMain(std::unique_ptr<Pipeline> pipeline, IAudioSink* sink) {
     InputStream sink_stream(&raw_sink_stream);
 
     if (sink_stream.info().bytes_in_stream == 0) {
+      if (sink_stream.is_producer_finished()) {
+        sink_stream.mark_consumer_finished();
+      } else {
+        // The user is probably about to hear a skip :(
+        ESP_LOGW(kTag, "!! audio sink is underbuffered !!");
+      }
       // No new bytes to sink, so skip sinking completely.
-      ESP_LOGW(kTag, "no bytes to sink");
       continue;
     }
 

src/audio/fatfs_audio_input.cpp

@@ -43,6 +43,7 @@ FatfsAudioInput::FatfsAudioInput()
       pending_path_(),
       current_file_(),
       is_file_open_(false),
+      has_prepared_output_(false),
       current_container_(),
       current_format_() {}
 
@@ -57,10 +58,13 @@ auto FatfsAudioInput::OpenFile(const std::string& path) -> bool {
   if (is_file_open_) {
     f_close(&current_file_);
     is_file_open_ = false;
+    has_prepared_output_ = false;
   }
+
   if (pending_path_) {
     pending_path_ = {};
   }
+
   ESP_LOGI(kTag, "opening file %s", path.c_str());
 
   database::TagParserImpl tag_parser;
@@ -112,13 +116,11 @@ auto FatfsAudioInput::NeedsToProcess() const -> bool {
 auto FatfsAudioInput::Process(const std::vector<InputStream>& inputs,
                               OutputStream* output) -> void {
   if (pending_path_) {
-    ESP_LOGI(kTag, "waiting for path");
     if (!pending_path_->valid()) {
       pending_path_ = {};
     } else {
       if (pending_path_->wait_for(std::chrono::seconds(0)) ==
           std::future_status::ready) {
-        ESP_LOGI(kTag, "path ready!");
         auto result = pending_path_->get();
         if (result) {
           OpenFile(*result);
@@ -131,9 +133,11 @@ auto FatfsAudioInput::Process(const std::vector<InputStream>& inputs,
     return;
   }
 
-  if (!output->prepare(*current_format_)) {
+  if (!has_prepared_output_ && !output->prepare(*current_format_)) {
+    ESP_LOGI(kTag, "waiting for buffer to free up");
     return;
   }
+  has_prepared_output_ = true;
 
   std::size_t max_size = output->data().size_bytes();
   if (max_size < output->data().size_bytes() / 2) {
@@ -152,14 +156,11 @@ auto FatfsAudioInput::Process(const std::vector<InputStream>& inputs,
   output->add(size);
 
   if (size < max_size || f_eof(&current_file_)) {
+    ESP_LOGI(kTag, "file finished. closing.");
     f_close(&current_file_);
     is_file_open_ = false;
-
-    // HACK: libmad requires an 8 byte padding at the end of each file.
-    if (current_container_ == database::Encoding::kMp3) {
-      std::fill_n(output->data().begin(), 8, std::byte(0));
-      output->add(8);
-    }
+    has_prepared_output_ = false;
+    output->mark_producer_finished();
 
     events::Dispatch<InputFileFinished, AudioState>({});
   }

src/audio/include/audio_decoder.hpp

@@ -43,6 +43,7 @@ class AudioDecoder : public IAudioElement {
   std::optional<StreamInfo::Format> current_input_format_;
   std::optional<StreamInfo::Format> current_output_format_;
   std::optional<std::size_t> seek_to_sample_;
+  bool has_prepared_output_;
   bool has_samples_to_send_;
   bool has_input_remaining_;
 

src/audio/include/fatfs_audio_input.hpp

@@ -52,6 +52,7 @@ class FatfsAudioInput : public IAudioElement {
   std::optional<std::future<std::optional<std::string>>> pending_path_;
   FIL current_file_;
   bool is_file_open_;
+  bool has_prepared_output_;
 
   std::optional<database::Encoding> current_container_;
   std::optional<StreamInfo::Format> current_format_;

src/audio/include/stream_info.hpp

@@ -26,10 +26,9 @@ struct StreamInfo {
   // stream's buffer.
   std::size_t bytes_in_stream{0};
 
-  // The total length of this stream, in case its source is finite (e.g. a
-  // file on disk). May be absent for endless streams (internet streams,
-  // generated audio, etc.)
-  std::optional<std::size_t> length_bytes{};
+  bool is_producer_finished = true;
+
+  bool is_consumer_finished = true;
 
   //
   std::optional<uint32_t> seek_to_seconds{};
@@ -62,10 +61,8 @@ class RawStream {
  public:
   StreamInfo* info;
   cpp::span<std::byte> data;
-  bool is_incomplete;
 
-  RawStream(StreamInfo* i, cpp::span<std::byte> d)
-      : info(i), data(d), is_incomplete(false) {}
+  RawStream(StreamInfo* i, cpp::span<std::byte> d) : info(i), data(d) {}
 };
 
 /*
@@ -78,7 +75,9 @@ class InputStream {
 
   void consume(std::size_t bytes) const;
 
-  void mark_incomplete() const;
+  bool is_producer_finished() const;
+
+  void mark_consumer_finished() const;
 
   const StreamInfo& info() const;
 
@@ -100,7 +99,9 @@ class OutputStream {
 
   cpp::span<std::byte> data() const;
 
-  bool is_incomplete() const;
+  bool is_consumer_finished() const;
+
+  void mark_producer_finished() const;
 
  private:
   RawStream* raw_;

src/audio/stream_info.cpp

@@ -28,8 +28,12 @@ void InputStream::consume(std::size_t bytes) const {
   raw_->info->bytes_in_stream = new_data.size_bytes();
 }
 
-void InputStream::mark_incomplete() const {
-  raw_->is_incomplete = true;
+void InputStream::mark_consumer_finished() const {
+  raw_->info->is_consumer_finished = true;
+}
+
+bool InputStream::is_producer_finished() const {
+  return raw_->info->is_producer_finished;
 }
 
 const StreamInfo& InputStream::info() const {
@@ -46,17 +50,12 @@ void OutputStream::add(std::size_t bytes) const {
 }
 
 bool OutputStream::prepare(const StreamInfo::Format& new_format) {
-  if (std::holds_alternative<std::monostate>(raw_->info->format)) {
-    raw_->info->format = new_format;
-    raw_->info->bytes_in_stream = 0;
-    return true;
-  }
-  if (new_format == raw_->info->format) {
-    return true;
-  }
-  if (raw_->is_incomplete) {
+  if (std::holds_alternative<std::monostate>(raw_->info->format) ||
+      raw_->info->is_consumer_finished) {
     raw_->info->format = new_format;
     raw_->info->bytes_in_stream = 0;
+    raw_->info->is_producer_finished = false;
+    raw_->info->is_consumer_finished = false;
     return true;
   }
   return false;
@@ -70,8 +69,12 @@ cpp::span<std::byte> OutputStream::data() const {
   return raw_->data.subspan(raw_->info->bytes_in_stream);
 }
 
-bool OutputStream::is_incomplete() const {
-  return raw_->is_incomplete;
+void OutputStream::mark_producer_finished() const {
+  raw_->info->is_producer_finished = true;
+}
+
+bool OutputStream::is_consumer_finished() const {
+  return raw_->info->is_consumer_finished;
 }
 
 }  // namespace audio

src/codecs/mad.cpp

@@ -44,6 +44,7 @@ MadMp3Decoder::~MadMp3Decoder() {
 }
 
 auto MadMp3Decoder::GetInputPosition() -> std::size_t {
+  assert(stream_.next_frame >= stream_.buffer);
   return stream_.next_frame - stream_.buffer;
 }
 
@@ -51,7 +52,7 @@ auto MadMp3Decoder::BeginStream(const cpp::span<const std::byte> input)
     -> Result<OutputFormat> {
   mad_stream_buffer(&stream_,
                     reinterpret_cast<const unsigned char*>(input.data()),
-                    input.size());
+                    input.size_bytes());
   // Whatever was last synthesized is now invalid, so ensure we don't try to
   // send it.
   current_sample_ = -1;
@@ -65,11 +66,11 @@ auto MadMp3Decoder::BeginStream(const cpp::span<const std::byte> input)
       // Recoverable errors are usually malformed parts of the stream.
       // We can recover from them by just retrying the decode.
       continue;
-    } else {
-      // Don't bother checking for other errors; if the first part of the stream
-      // doesn't even contain a header then something's gone wrong.
-      return {GetInputPosition(), cpp::fail(Error::kMalformedData)};
     }
+    if (stream_.error == MAD_ERROR_BUFLEN) {
+      return {GetInputPosition(), cpp::fail(Error::kOutOfInput)};
+    }
+    return {GetInputPosition(), cpp::fail(Error::kMalformedData)};
   }
 
   uint8_t channels = MAD_NCHANNELS(&header);

src/system_fsm/running.cpp

@@ -38,7 +38,6 @@ void Running::entry() {
   vTaskDelay(pdMS_TO_TICKS(250));
 
   ESP_LOGI(kTag, "opening database");
-  database::Database::Destroy();
   auto database_res = database::Database::Open();
   if (database_res.has_error()) {
     ESP_LOGW(kTag, "failed to open!");