OpenSocialNet is a from-scratch real-time chat + voice + video desktop app — a Discord-style native client written in C++ where the entire audio/video pipeline (custom UDP packet format, jitter buffer, packet-loss concealment, threading, encoder/decoder wrappers, server-side UDP relay) is implemented by hand instead of delegating to WebRTC.

An earlier browser + libdatachannel-SFU experiment is archived on the archive/webrtc-sfu branch as a credential snapshot; it is no longer the deployment target. master is built around the native client.

Layout

• src/native_client/ — desktop GUI (SDL3 window + Dear ImGui docking branch). Entry point of the product.
• src/network/ — UDP transport: socket wrapper, sender/receiver, custom Packet header, jitter buffer, JitterStats, lock-free SpscQueue, sender-side LossSim. Loopback exe network exercises the whole pipeline on a single machine.
• src/audio_engine/ — Opus capture/encode/decode wrappers + SDL3 audio I/O.
• src/video_engine/ — V4L2 capture, x264 encode, FFmpeg decode, SDL3 render. Wiring into the native client is Phase 4.
• src/signaling_server/ — WebSocket gateway: Protobuf envelopes (Hello/Heartbeat/Ready/SendMessage/FetchHistory/JoinVoice/VoicePeerJoined/…), HMAC auth, Scylla chat persistence, Kafka pub/sub fan-out.
• src/relay/ — UDP relay that fans Packets out per voice room. Not written yet (Phase 2).
• docker-compose.yml / compose.prod.yml — runtime data services (Kafka, ZooKeeper, Scylla) used by signaling_server.

Build dependencies

Dep	How to get it
build-essential, cmake (>=3.24), pkg-config, git	apt install
libx264-dev	apt install
libavcodec-dev, libavutil-dev, libswscale-dev	apt install
libopus-dev	apt install
libv4l-dev	apt install
libssl-dev	apt install
SDL3	fetched by CMake (FetchContent)
Dear ImGui (docking branch)	fetched by CMake (FetchContent) — used by native_client
Protobuf, librdkafka, uWebSockets	apt install (only needed for signaling_server)

The full apt command list lives in the Dockerfile (treat it as executable documentation — anything missing from your host shows up as a CMake error).

Quick start

Option A: Docker (recommended for first build)

Builds a reproducible Ubuntu 24.04 image with every system dep preinstalled. Source is mounted from the host, so edits stay on your machine.

docker build -t opensocialnet-build .
docker run --rm -it -v "$PWD:/workspace" opensocialnet-build bash
# inside the container:
mkdir -p src/video_engine/build && cd src/video_engine/build
cmake .. && cmake --build . -j

Camera/audio passthrough into the container is host-specific (--device /dev/video0 --group-add video on Linux). For most local dev it's easier to run binaries directly on the host.

Option B: Native (Ubuntu 24.04 / WSL2)

# install system deps (subset of the Dockerfile)
sudo apt install -y build-essential cmake pkg-config git \
    libx264-dev libavcodec-dev libavutil-dev libswscale-dev \
    libopus-dev libv4l-dev libssl-dev v4l-utils

# build video_engine (SDL3 + libdatachannel are fetched + built automatically; first build is slow)
mkdir -p src/video_engine/build && cd src/video_engine/build
cmake .. && cmake --build . -j

Binaries land in src/video_engine/build/:

• video_sender [host] [port] — capture, encode, send. Defaults to 127.0.0.1:9001.
• video_receiver [port] — receive, decode, render in an SDL window. Defaults to port 9001.
• audio_video_client — both streams in one process (audio on 9000, video on 9001), self-loopback.
• test_video_loopback — capture → encode → UDP → decode → render in one process. Useful for sanity checks.
• test_capture_encode — dumps a few seconds of camera input to output.h264 (smoke-test capture).
• test_probe_camera — enumerates V4L2 formats and supported (size, fps) tuples for /dev/video0.

Camera on WSL2

/dev/video* doesn't exist by default on WSL2. To pass a host webcam through:

1. Install usbipd-win on Windows.

2. usbipd list (Windows) → find your camera's busid.

3. usbipd bind --busid <id> then usbipd attach --wsl --busid <id>.

4. WSL2 also needs UVC drivers in its kernel — stock WSL kernels are missing them. Build a custom WSL kernel with CONFIG_USB_VIDEO_CLASS=y (and the rest of CONFIG_MEDIA_*), drop the resulting bzImage somewhere on Windows, and point %USERPROFILE%/.wslconfig at it:

   [wsl2]
   kernel=C:\\Users\\<you>\\bzImage-wsl-custom

5. Restart WSL (wsl --shutdown), re-attach the camera, and /dev/video0 should appear.

6. Open device permissions for your user: sudo chmod a+rw /dev/video0 (or add yourself to the video group).

Signaling server

Data services come up via Compose:

docker compose up -d zookeeper kafka scylla

Then build and run src/signaling_server/. See that directory's own build.sh / scripts/ for vcpkg + Cassandra driver setup (the driver isn't in apt).