Event-loop ownership

The application that starts the process should own the event loop. For a command-line application with no loop already running in the current thread, miniproto.event_loop.run(...) creates an asyncio.Runner, runs one coroutine, closes its loop, and clears the current-loop reference when it finishes.

from miniproto import event_loop


async def main() -> None: ...


event_loop.run(main())

Do not call event_loop.run(...) from a thread that already has a running event loop. In an async web server, notebook, GUI, test runner, or library callback, await the client work from the loop the host already owns instead of nesting a runner.

Scoped optimized loops

On Windows, miniproto selects winloop; on other platforms it selects uvloop. If that selected package is absent, the helpers fall back to the standard asyncio event loop. A broken backend import is propagated rather than silently treated as unavailable, which helps distinguish an absent optional dependency from an installation problem.

For an application that owns a runner but needs explicit configuration, use the helper as its loop factory:

import asyncio

from miniproto import event_loop


with asyncio.Runner(loop_factory=event_loop.new_event_loop) as runner:
    runner.run(main())

new_event_loop() registers the new loop as current for the calling thread. The caller owns its lifecycle when using it directly; asyncio.Runner closes the loop it receives from its factory.

Debug runners intentionally use the standard loop for the known debug-mode incompatibilities with uvloop through 0.22.1 and winloop through 0.6.3. This is a debugging trade-off, not evidence that the optimized backend is absent.

Avoid the legacy installer

event_loop.install() attempts a deprecated process-global policy installation, always raises a DeprecationWarning, and returns False when a backend/hook is unavailable, installation fails, or Python 3.16+ does not support policy installation. It is race-unsafe with other tasks that create loops. Prefer event_loop.run(...) for a top-level command or asyncio.Runner(loop_factory=event_loop.new_event_loop) for an owned service boundary.

The package metadata requires Python 3.13 or newer. Its declared optimized-loop dependencies target Linux/Darwin (uvloop) and Windows/Cygwin (winloop); verify the actual interpreter and installed distribution in the environment that runs the service before assuming an optimized backend is active.