PROBLEM:
When streaming Radio Paradise blocks via HTTP using FFPlay, the buffer
would cycle between 0KB and ~130KB approximately once per second, causing
audio dropouts and interruptions. VLC worked fine, but FFPlay was sensitive
to the burst transmission pattern.
ROOT CAUSE:
In StreamingFlacSink::broadcast_flac_stream(), the broadcaster was reading
8KB (8192 bytes) at a time from the FLAC encoder and sending the entire
chunk at once to all HTTP clients. This created a "burst" pattern:
- Read 8KB from encoder
- Send entire 8KB chunk to all clients
- Sleep if ahead of real-time pacing
- Repeat
FFPlay's buffer would fill rapidly with each 8KB burst, then drain completely
before the next burst arrived, causing the observed cycling behavior.
SOLUTION:
Reduced the HTTP broadcast buffer size from 8KB to 512 bytes in
StreamingFlacSink::broadcast_flac_stream(). This creates a much smoother,
more continuous data flow that FFPlay can handle without buffer cycling.
The 512-byte buffer size is:
- Small enough to prevent burst transmission
- Large enough to avoid excessive overhead
- Sufficient for smooth streaming with real-time pacing
CHANGES:
- Restore stream_block.rs example from git history
- Restore StreamingFlacSink and StreamingOggFlacSink from git history
- Add "streaming" feature to pmoaudio-ext
- Reduce broadcast buffer from 8192 to 512 bytes
- Update pmoparadise to use pmoaudio-ext streaming feature
TESTING:
Test with FFPlay to verify smooth buffering:
```bash
cargo run --example stream_block --features full -- 0
# In another terminal:
ffplay http://localhost:8080/test/stream
```
Watch the "aq=" value in FFPlay output. It should now remain stable
instead of cycling between 0KB and 130KB.
Fix "PlaylistManager not initialized" error without creating circular
dependency between pmoparadise and pmoupnp.
Changes:
- Add AUDIO_CACHE static to pmoplaylist/manager.rs
- Add register_audio_cache() function to register cache
- Export register_audio_cache from pmoplaylist lib.rs
- Update audio_cache() to check local registry first, then pmoupnp
- Update play_and_cache example to use pmoplaylist::register_audio_cache
- Remove pmoupnp::register_*_cache functions (not needed)
The example now calls pmoplaylist::register_audio_cache() to make
the cache available for pk validation in WriteHandle::push().
Fix "PlaylistManager not initialized" error in play_and_cache example.
The error occurred because pmoplaylist's WriteHandle calls
pmoupnp::get_audio_cache() to validate cache pks, but the global
cache registry wasn't initialized.
Changes:
- Add register_audio_cache() and register_cover_cache() functions
- Export them from pmoupnp lib.rs
- Call them in play_and_cache example after creating caches
This mirrors how UpnpServer initializes the cache registry.
- Add pmoconfig to the 'full' feature so PlaylistManager can auto-initialize
- Update play_and_cache example to use AudioSink::new() without volume
- Remove unused PlaylistManager import
Without pmoconfig feature enabled, PlaylistManager::get() fails with
"PlaylistManager not initialized" error.
Corrections pour que l'exemple utilise les bonnes API:
- Utilisation directe de Cache::new() au lieu de méthodes de config
- Suppression des appels à root() qui n'existent pas
- Utilisation du singleton PlaylistManager() au lieu de new()
- Ajout de cache-sink comme dépendance de playlist feature dans pmoaudio-ext
L'exemple devrait maintenant compiler correctement avec:
cargo run --example play_and_cache --features full -- <channel_id>
Crée un nouvel exemple complet qui démontre l'utilisation de tout le pipeline:
- Téléchargement d'un bloc Radio Paradise
- Cache FLAC via FlacCacheSink
- Playlist alimentée automatiquement
- Lecture en temps réel via PlaylistSource et AudioSink
Architecture à deux pipelines :
Pipeline 1 (Download & Cache):
RadioParadiseStreamSource → FlacCacheSink (avec playlist abonnée)
Pipeline 2 (Playback):
PlaylistSource (lit la playlist) → AudioSink (joue l'audio)
Les deux pipelines s'exécutent en parallèle, permettant la lecture pendant le
téléchargement.
Modifications:
- Ajout pmoaudio-ext avec feature playlist dans pmoparadise
- Nouvelle feature "full" combinant pmoaudio + pmoaudio-ext
- Logs détaillés à tous les niveaux (DEBUG)
Usage: cargo run --example play_and_cache --features full -- <channel_id>
Ajoute la possibilité de configurer le channel par défaut de Radio Paradise
de manière persistante et user-friendly.
Fonctionnalités :
- get_paradise_default_channel() : récupère le channel configuré (défaut: 0/main)
- set_paradise_default_channel(u8) : définit le channel par défaut
- Accepte DEUX formats dans le fichier YAML :
* Noms conviviaux : "main", "mellow", "rock", "eclectic"
* IDs numériques : 0, 1, 2, 3
- Stocke les valeurs comme chaînes conviviales pour la lisibilité
- Validation automatique avec fallback sur "main" si invalide
- Persistence automatique de la valeur par défaut lors du premier accès
Exemple de configuration YAML :
```yaml
sources:
radio_paradise:
enabled: true
default_channel: mellow # ou 1
```
Cette amélioration rend la configuration plus accessible aux utilisateurs
qui préfèrent un channel autre que Main Mix par défaut.
Corrections :
1. Supprimé le commentaire obsolète sur block_base (ligne 277)
2. Créé la constante DEFAULT_CHANNEL pour éviter de coder "0" en dur
3. Utilisé DEFAULT_CHANNEL dans with_client(), ClientBuilder::default() et tests
4. Amélioré la documentation de with_client() pour guider vers le builder
Bien que with_client() et ClientBuilder::default() aient encore une structure
similaire, ils utilisent maintenant les mêmes constantes, réduisant ainsi
le risque d'incohérence lors de modifications futures.
Supprime complètement la duplication d'information en transformant
block_base d'un champ stocké en une méthode calculée dynamiquement.
Changements:
- Supprimé le champ block_base de RadioParadiseClient
- Ajouté la constante BLOCK_BASE_URL pour éviter la duplication de l'URL
- Transformé block_base en méthode publique qui calcule à partir de channel
- Simplifié with_client() et clone_with_channel()
- Simplifié le builder qui n'a plus besoin d'initialiser block_base
Cette approche garantit que block_base est toujours cohérent avec channel,
éliminant définitivement toute possibilité de bug de synchronisation.
Le bug identifié était que le block_base n'était pas synchronisé avec
le channel dans le ClientBuilder, causant le téléchargement du même
bloc pour différents channels.
Changements:
- Supprimé le champ block_base du ClientBuilder (duplication)
- Supprimé la constante DEFAULT_BLOCK_BASE (plus nécessaire)
- Supprimé la méthode .block_base() du builder (complexité inutile)
- Le block_base est maintenant calculé dynamiquement dans build()
à partir du channel, éliminant toute possibilité de désynchronisation
Cette approche suit le principe DRY et élimine une source de bugs.
Add a new example that demonstrates downloading a complete Radio Paradise
block and saving each track as a separate FLAC file.
The example:
- Takes a channel ID as argument (0-3)
- Fetches current block metadata
- Creates an output directory ./rp_channel_{id}block{blockid}
- Uses RadioParadiseStreamSource to stream and decode the block
- Uses FlacFileSink to automatically detect TrackBoundary markers
- Saves each track as a separate FLAC file with metadata
Example usage:
cargo run --example download_block --features=pmoaudio -- 0
This demonstrates the full pipeline integration between pmoparadise
and pmoaudio, showing how RadioParadiseStreamSource and FlacFileSink
work together to handle multi-track FLAC blocks seamlessly.
Add async_trait annotation to RadioParadiseExt trait and its implementation
to suppress the "async fn in public traits" warning.
This is the recommended approach for traits with async methods as it ensures
proper Future bounds (Send) are generated.
The old streaming API has been completely replaced by RadioParadiseStreamSource
which integrates directly with the pmoaudio pipeline.
Removed:
- src/stream.rs (179 lines) - BlockStream, stream_block(), download_block()
- src/track.rs - Per-track extraction functionality
- examples/stream_block.rs - Obsolete streaming example
- examples/extract_track.rs - Per-track extraction example
- Feature "per-track" and dependencies (hound, tempfile)
Updated:
- Cargo.toml: Removed per-track feature and obsolete examples
- lib.rs: Removed module declarations and re-exports
The new RadioParadiseStreamSource provides:
- Direct integration with pmoaudio pipeline
- FLAC decoding via pmoflac
- Automatic TrackBoundary insertion
- Better performance and lower latency
Changes:
1. Updated pmoserver_ext.rs to use channels module:
- Changed imports from paradise:: to channels::
- Now uses ChannelDescriptor from channels module
- No longer depends on paradise orchestration code
2. Updated lib.rs:
- Added pub mod channels
- Maintains existing modules for now (will evaluate removal later)
3. Verification:
- All tests pass (25/25)
- No dead code warnings
- Compilation successful with all features
The REST API is now simplified to provide only direct Radio Paradise
API access without the heavy orchestration layer.
Changes:
1. Removed dead code from paradise/worker.rs:
- Unused process_song() method
- Unused DecodedBlock struct
- Unused helper functions: song_duration_ms, ms_to_frames, decode_block_audio
2. Simplified pmoserver_ext.rs (840 → 383 lines):
- Removed complex orchestration endpoints (status, playlist, history, streaming)
- Kept only simple API access endpoints:
* /now-playing
* /block/current
* /block/{event_id}
* /channels
- Removed dependencies on RadioParadiseSource and ParadiseChannel
3. Created channels.rs:
- Extracted channel definitions from paradise/channel.rs
- Pure data module with no orchestration logic
- Contains: ParadiseChannelKind, ChannelDescriptor, ALL_CHANNELS
Note: This is work in progress. Still need to update lib.rs and remove
unused modules once dependencies are fully resolved.
Add support for 32-bit integer audio samples to match FileSource and
HttpSource capabilities, ensuring complete bit depth coverage.
Changes:
- Add I32 case to pcm_to_audio_segment() for 32-bit stereo samples
- Update output_type() comment to document 16/24/32-bit support
- Note that bit depth is auto-detected from FLAC header via pmoflac
The implementation now supports the full range of FLAC bit depths:
- 16-bit: AudioChunk::I16 (most common)
- 24-bit: AudioChunk::I24 (high quality)
- 32-bit: AudioChunk::I32 (maximum precision)
pmoflac reads bits_per_sample from the FLAC STREAMINFO header
(decoder.rs:97), so the actual bit depth is determined by the source
stream, not hardcoded.
Verified: cargo check passes successfully.
Remove unused FFmpeg-based progressive streaming implementation that was
never completed and is not used anywhere in the codebase.
Changes:
- Delete src/ffmpeg_streaming.rs (173 lines of unfinished code with TODOs)
- Remove ffmpeg module import from lib.rs
- Remove ffmpeg feature from Cargo.toml
- Remove ffmpeg-next dependency from Cargo.toml
The current implementation uses claxon (StreamingPCMDecoder) and symphonia
(decode_block_audio) for FLAC decoding, which are fully functional.
Verified: cargo check passes successfully after removal.
Problem:
- Used `let _ = metadata.set_title(...)` which creates unawaited Future
- Futures were never executed → metadata fields never set!
- Ignored Result<(), MetadataError> which could contain errors
Solution:
- Spawn tokio task to configure metadata asynchronously
- Properly await all set_*() calls
- Handle errors with eprintln! warnings instead of silent ignore
- Clone all data upfront for the async task
Type info:
- metadata: MemoryTrackMetadata (concrete type)
- Returns: Arc<RwLock<dyn TrackMetadata>> (trait object)
- Methods: async fn set_*(&mut self) -> MetadataResult<()>
All 7 tests still pass ✅
Refactored RadioParadiseStreamSource to use current pmoaudio API:
Audio Segment Creation:
- Replaced AudioSegment::new_audio() with manual construction using _AudioSegment
- Convert PCM to Vec<[i16; 2]> or Vec<[I24; 2]> stereo pairs
- Use AudioChunkData::new(stereo, sample_rate, gain_db) → Arc
- Wrap in AudioChunk::I16() or AudioChunk::I24()
- Create AudioSegment with order, timestamp_sec, and _AudioSegment::Chunk()
Sync Markers:
- Replaced AudioSegment::new_sync() with AudioSegment::new_track_boundary()
- Created TopZeroSync manually with _AudioSegment::Sync()
- Use AudioSegment::new_end_of_stream() for EOF
Stream Handling:
- Changed from decoder.next() (doesn't exist) to decoder.read()
- Added AsyncReadExt import
- Use buffered read approach like http_source
- Changed decoder.stream_info() to decoder.info()
I24 Construction:
- Changed I24::from_i32() to I24::new_clamped()
- Properly handles 24-bit PCM conversion with sign extension
Metadata:
- Fixed RwLock usage - write().await returns guard directly, no Result
- Added `let _` for Future return values
Testing:
✅ All 7 unit tests pass (cache FIFO behavior)
✅ RadioParadiseStreamSource compiles successfully with pmoaudio feature
Fixed test issues:
- Changed EventId(i) to plain i (EventId is type alias for u64)
- Added create_test_client() helper using RadioParadiseClient::with_client()
- Cast DEFAULT_CHUNK_DURATION_MS to u32 as expected by constructor
Outstanding API incompatibility issues:
- AudioSegment API has evolved (new_audio/new_sync no longer exist)
- AudioChunkData::from_interleaved() doesn't exist
- I24::from_i32() should be I24::new() or I24::new_clamped()
- Need to understand current pmoaudio API for creating audio segments
Tests compile but RadioParadiseStreamSource implementation needs
significant refactoring to match current pmoaudio API.
Problem:
- With `if >= CACHE_SIZE`, only ONE element removed per call
- If cache ever had >10 elements (abnormal state), would stay oversized
- Example: 12 elements → if removes 1 → 11 elements → add 1 → 12 elements ❌
Solution:
- Use `while >= CACHE_SIZE` to remove ALL excess elements
- Example: 12 elements → while removes 2 → 10 elements → add 1 → 10 elements ✅
- Guarantees exactly ≤10 elements regardless of initial state
Changes:
- mark_block_downloaded(): changed `if` to `while`
- Updated comment to reflect "tous les éléments excédentaires"
- Documentation updated with robustness guarantee
Problem:
- Previous logic: push_back() first, then pop_front() if len > 10
- This temporarily creates 11 elements, exceeding VecDeque capacity of 10
- Wastes the benefit of with_capacity() pre-allocation
Solution:
- Check capacity BEFORE adding: if len >= 10, pop_front() first
- Then push_back() new element
- Guarantees never exceeding 10 elements at any time
Changes:
- mark_block_downloaded(): inverted order (pop before push)
- Changed condition from `> CACHE_SIZE` to `>= CACHE_SIZE`
- Documentation updated with correct logic and benefits
Problem:
- HashSet doesn't maintain insertion order
- iter().next() returns arbitrary element, not the oldest
- Cache eviction was unpredictable
Solution:
- Use VecDeque for FIFO ordering
- push_back() adds new block
- pop_front() removes oldest block when cache exceeds 10 elements
- contains() is O(n) but performant for 10 elements
Changes:
- RadioParadiseStreamSourceLogic: recent_blocks now VecDeque<EventId>
- mark_block_downloaded(): simplified with guaranteed FIFO eviction
- Documentation updated with VecDeque usage and advantages
- Add comprehensive technical documentation (RADIO_PARADISE_STREAM_SOURCE.md)
- Add practical usage example (examples/radio_paradise_stream.rs)
- Document architecture, timing algorithm, and API
- Include both basic and advanced usage patterns with nowplaying stream
Implement a new pmoaudio source node that streams Radio Paradise blocks
with automatic TrackBoundary insertion at the correct timing.
Features:
- Downloads and decodes FLAC blocks from Radio Paradise API
- Queue management for block IDs via push_block_id()
- Recent blocks cache (10 blocks) to avoid re-downloads
- Automatic TrackBoundary insertion based on sample count timing
- Converts Song metadata to TrackMetadata with cover URLs
- Timeout of 3 seconds for new block IDs (radio real-time)
- Support for 16-bit and 24-bit FLAC audio
Architecture:
- RadioParadiseStreamSourceLogic: Pure business logic implementing NodeLogic
- RadioParadiseStreamSource: Wrapper using Node<> pattern
- Uses logic_mut() for push_block_id() configuration
The node emits:
- TopZeroSync at block start
- TrackBoundary before each song (with same order as next chunk)
- Audio chunks (I16 or I24)
- EndOfStream on timeout or completion
New pmoaudio feature gate with dependencies on:
- pmoaudio, pmoflac, pmometadata, futures-util
