aiolibdatachannel 2026.5.7

asyncio-friendly Python wrapper for libdatachannel (WebRTC Data Channels)

aiolibdatachannel

An asyncio-friendly Python wrapper around libdatachannel — a lightweight C/C++ WebRTC stack. aiolibdatachannel exposes WebRTC Peer Connections and Data Channels through idiomatic async/await Python.

Scope: PeerConnection + DataChannel only. WebSocket and Media/RTP transport are deliberately disabled in this build to keep the wheel small and the API focused.

Install

pip install aiolibdatachannel

Wheels are published for Linux (manylinux_2_28 x86_64 / aarch64) and macOS (arm64, 14.0+). Intel Macs and Windows aren't currently covered — Intel Mac because Apple Silicon is the modern target, Windows because libdatachannel dynamically links OpenSSL and the wheel packaging for that on Windows is still open.

The Python↔C boundary uses nanobind's stable-ABI mode (Py_LIMITED_API): libdatachannel and its static dependencies (usrsctp, libjuice, OpenSSL) link directly into the extension, and a single cp312-abi3-<platform> wheel per architecture covers every CPython 3.12+ interpreter — no per-Python-version build. The binding layer itself is deliberately thin: native trampolines route every libdatachannel callback through a single Python dispatcher, and the asyncio semantics live in the pure-Python wrapper on top, not in C++.

Quickstart

import asyncio
from aiolibdatachannel import PeerConnection, RTCConfiguration

async def main() -> None:
    config = RTCConfiguration(ice_servers=["stun:stun.l.google.com:19302"])

    async with PeerConnection(config) as pc:
        dc = await pc.create_data_channel("chat")

        offer = await pc.create_offer()
        # ... ship `offer` to the peer via your signalling channel ...
        answer_sdp = await receive_answer_from_peer()
        await pc.set_remote_description(answer_sdp, "answer")

        await dc.wait_open()
        await dc.send(b"hello")

        async for message in dc:
            print("got", message)

asyncio.run(main())

See examples/offerer.py and examples/answerer.py for a runnable pair of scripts that negotiate over stdin/stdout.

Logging

Route libdatachannel's internal logs through Python's standard logging module:

import logging
from aiolibdatachannel import install_python_logger

logging.basicConfig(level=logging.INFO)
install_python_logger()  # logger name defaults to "aiolibdatachannel"

Severities are translated (FATAL→CRITICAL, ERROR→ERROR, WARNING→WARNING, INFO→INFO, DEBUG/VERBOSE→DEBUG) and the filter threshold on the native side is derived from the Python logger's effective level, so you don't pay to format lines that would be filtered out anyway. Pass install_python_logger(my_logger) or install_python_logger(level=LogLevel.DEBUG) to customise.

Development

Full native build (requires a C++17 toolchain, CMake ≥ 3.24, ninja, and OpenSSL headers):

git clone --recursive https://github.com/social-home-io/aiolibdatachannel.git
cd aiolibdatachannel
pip install -e .[dev]
pytest

If you only need to touch the Python wrapper (async semantics, async iterators, event plumbing), skip the C++ build and run against the pure-Python stand-in from tests/_fake_native.py:

pip install pytest pytest-asyncio
PYTHONPATH=. pytest -m "not native"

tests/conftest.py injects the fake into sys.modules whenever AIOLIB_REQUIRE_NATIVE is unset. Tests marked @pytest.mark.native (loopback, cancellation, logging, errors, stress) are auto-skipped in that mode.

Long-running soak tests (10 000-message loopback, 100 PC lifecycles, shutdown-leak subprocess checks) live behind a stress marker and are off by default — enable with pytest -m stress or AIOLIB_STRESS=1 pytest. They also run nightly in CI.

See docs/BUILDING.md for detailed build instructions, including how to switch TLS backends and how to bump the bundled libdatachannel version.

License

This project is licensed under the Mozilla Public License 2.0, matching the license of libdatachannel which it bundles and statically links. See LICENSE and vendor/libdatachannel/LICENSE.