Siavash Sameni 12020b019c
Some checks failed
Build Release Binaries / build-amd64 (push) Failing after 3m20s
Mirror to GitHub / mirror (push) Failing after 28s
fix(video): normalize VideoToolbox plane strides to tight I420
Android-encoded H.264 decoded cleanly with ffmpeg but showed diagonal
green/magenta banding on macOS. Root cause: shiguredo_video_toolbox's
I420Frame exposes y/u/v planes as bytes_per_row * height, including
CoreVideo's stride padding. VideoToolboxDecoder concatenated those
slices verbatim, then downstream code indexed the buffer as tight I420,
producing per-row drift that wrapped one full row every 16 chroma rows
(32 luma rows) at 960x540.

Add i420_frame_to_tight() helper that copies each plane row-by-row at
width / chroma_width using the plane's actual stride. All three macOS
decoders (H.264, HEVC, AV1) now call it. On first decode each logs the
real plane dimensions and strides at target wzp_video::videotoolbox so
future stride bugs are diagnosable from logs.

Verified mathematically against the corrupted dump:
  band period = u_stride / (u_stride - chroma_width)
              = 512 / (512 - 480) = 16 chroma rows = 32 luma rows
which matches the measured spacing exactly. 640x360 was unaffected
because chroma_width 320 is already 64-aligned.

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
2026-05-26 15:22:40 +04:00

WarzonePhone

Custom lossy VoIP protocol built in Rust. E2E encrypted, FEC-protected, adaptive quality, designed for hostile network conditions.

Quick Start

# Build
cargo build --release

# Run relay
./target/release/wzp-relay --listen 0.0.0.0:4433

# Send a test tone
./target/release/wzp-client --send-tone 5 relay-addr:4433

# Web bridge (browser calls)
./target/release/wzp-web --port 8080 --relay 127.0.0.1:4433 --tls
# Open https://localhost:8080/room-name in two browser tabs

Architecture

See docs/ARCHITECTURE.md for the full system architecture with Mermaid diagrams covering:

  • System overview and data flow
  • Crate dependency graph (8 crates)
  • Wire formats (MediaHeader, MiniHeader, TrunkFrame, SignalMessage)
  • Cryptographic handshake (X25519 + Ed25519 + ChaCha20-Poly1305)
  • Identity model (BIP39 seed, featherChat compatible)
  • Quality profiles (GOOD/DEGRADED/CATASTROPHIC)
  • FEC protection (RaptorQ with interleaving)
  • Adaptive jitter buffer (NetEq-inspired)
  • Telemetry stack (Prometheus + Grafana)
  • Deployment topology

Features

  • 3 quality tiers: Opus 24k (28.8 kbps) / Opus 6k (9 kbps) / Codec2 1200 (2.4 kbps)
  • RaptorQ FEC: Recovers from 20-100% packet loss depending on tier
  • E2E encryption: ChaCha20-Poly1305 with X25519 key exchange
  • Adaptive jitter buffer: EMA-based playout delay tracking
  • Silence suppression: VAD + comfort noise (~50% bandwidth savings)
  • ML noise removal: RNNoise (nnnoiseless pure Rust port)
  • Mini-frames: 67% header compression for steady-state packets
  • Trunking: Multiplex sessions into batched datagrams
  • featherChat integration: Shared BIP39 identity, token auth, call signaling
  • Prometheus metrics: Relay, web bridge, inter-relay probes
  • Grafana dashboard: Pre-built JSON with 18 panels

Documentation

Document Description
ARCHITECTURE.md Full system architecture with diagrams
TELEMETRY.md Prometheus metrics specification
INTEGRATION_TASKS.md featherChat integration tracker
WZP-FC-SHARED-CRATES.md Shared crate strategy
grafana-dashboard.json Importable Grafana dashboard

Binaries

Binary Description
wzp-relay Relay daemon (SFU room mode, forward mode, probes)
wzp-client CLI client (send-tone, record, live mic, echo-test, drift-test, sweep)
wzp-web Browser bridge (HTTPS + WebSocket + AudioWorklet)
wzp-bench Component benchmarks

Linux Build

./scripts/build-linux.sh --prepare   # Create Hetzner VM + install deps
./scripts/build-linux.sh --build     # Build release binaries
./scripts/build-linux.sh --transfer  # Download to target/linux-x86_64/
./scripts/build-linux.sh --destroy   # Delete VM

Tests

cargo test --workspace   # 272 tests

License

MIT OR Apache-2.0

Description
No description provided
Readme 150 MiB
Languages
Rust 82.6%
Kotlin 6.3%
Shell 5.7%
TypeScript 1.8%
C++ 1.2%
Other 2.4%