tonio 0.1.0a4

A multi-threaded async runtime

TonIO

TonIO is a multi-threaded async runtime for free-threaded Python, built in Rust on top of the mio crate, and inspired by tinyio and trio.

Warning: TonIO is currently a work in progress and very pre-alpha state. The APIs are subtle to breaking changes.

Note: TonIO is available on free-threaded Python and on Unix systems only.

In a nutshell

import tonio

def wait_and_add(x: int) -> int:
    yield tonio.sleep(1)
    return x + 1

def foo():
    four, five = yield tonio.spawn(wait_and_add(3), wait_and_add(4))
    return four, five

out = tonio.run(foo())
assert out == (4, 5)

Usage

Entrypoint

Every TonIO program consist of an entrypoint, which should be passed to the run method:

import tonio

def main():
    yield
    print("Hello world")

tonio.run(main())

TonIO also provides a main decorator, thus we can rewrite the previous example as:

import tonio

@tonio.main
def main():
    yield
    print("Hello world")

main()

Runtime options

Both run and main accept options, specifically:

option name	description	default
context	enable contextvars usage in coroutines	False
threads	Number of runtime threads	# of CPU cores
threads_blocking	Maximum number of blocking threads	128
threads_blocking_timeout	Idle timeout for blocking threads (in seconds)	30

Events

The core object in TonIO is Event. It's basically a wrapper around an atomic boolean flag, initialised with False. Event provides the following methods:

• is_set(): return the value of the flag
• set(): set the flag to True
• clear(): set the flag to False
• wait(timeout=None): returns a coroutine you can yield on that unblocks when the flag is set to True or the timeout expires. Timeout is seconds.
• __call__(timeout=None): same of wait, but returns a coroutine you can await on.

import tonio

@tonio.main
def main():
    event = tonio.Event()

    def setter():
        yield tonio.sleep(1)
        event.set()

    tonio.spawn(setter())
    yield event.wait()

Spawning tasks

TonIO provides the spawn method to schedule new coroutines onto the runtime:

import tonio

def doubv(v):
    yield
    return v * 2

@tonio.main
def main():
    parallel = tonio.spawn(doubv(2), doubv(3))
    v3 = yield doubv(4)
    v1, v2 = yield parallel
    print([v1, v2, v3])

Coroutines passed to spawn get schedule onto the runtime immediately. Using yield on the return value of spawn just waits for the coroutines to complete and retreive the results.

Blocking tasks

TonIO provides the spawn_blocking method to schedule blocking operations onto the runtime:

import tonio

def read_file(path):
    with open(file, "r") as f:
        return f.read()

@tonio.main
def main():
    file_data = yield tonio.spawn_blocking(read_file, "sometext.txt")

Scopes and cancellations

TonIO provides a scope context, that lets you cancel work spawned within it:

import tonio

def slow_push(target, sleep):
    yield tonio.sleep(sleep)
    target.append(True)

@tonio.main
def main():
    values = []
    with tonio.scope() as scope:
        scope.spawn(_slow_push(values, 0.1))
        scope.spawn(_slow_push(values, 2))
        yield tonio.sleep(0.2)
        scope.cancel()
    yield scope()
    assert len(values) == 1

When you yield on the scope, it will wait for all the spawned coroutines to end. If the scope was canceled, then all the pending coroutines will be canceled.

Note: the colored version of scope, doesn't require to be awaited, as it will yield on exit.

Time-related functions

• tonio.time.time(): a function returning the runtime's clock
• tonio.time.sleep(delay): a coroutine you can yield on to sleep (delay is in seconds)
• tonio.time.timeout(coro, timeout): a coroutine you can yield on returning a tuple (output, success). If the coroutine succeeds in the given time then the pair (output, True) is returned. Otherwise this will return (None, False).

Note: time.sleep is also exported to the main tonio module.

Synchronization primitives

Synchronization primitives are exposed in the tonio.sync module.

Lock

Implements a classic mutex, or a non-reentrant, single-owner lock for coroutines:

import tonio
import tonio.sync

@tonio.main
def main():
    # counter can't go above 1
    counter = 0

    def _count(lock):
        nonlocal counter
        with (yield lock()):
            counter += 1
            yield
            counter -= 1
    
    lock = tonio.sync.Lock()
    yield tonio.spawn(*[_count(lock) for _ in range(10)])

Semaphore

A semaphore for coroutines:

import tonio
import tonio.sync

@tonio.main
def main():
    # counter can't go above 2
    counter = 0

    def _count(lock):
        nonlocal counter
        with (yield lock()):
            counter += 1
            yield
            counter -= 1
    
    lock = tonio.sync.Semaphore(2)
    yield tonio.spawn(*[_count(lock) for _ in range(10)])

Barrier

A barrier for coroutines:

import tonio
import tonio.sync

@tonio.main
def main():
    barrier = tonio.sync.Barrier(3)
    count = 0

    def _start_at_3():
        nonlocal count
        count += 1
        i = yield barrier.wait()
        assert count == 3
        return i

    yield tonio.spawn(*[_start_at_3() for _ in range(3)])

Channels

Multi-producer multi-consumer channels for inter-coroutine communication.
The tonio.sync.channel module provides both a channel and unbounded constructors:

import tonio
import tonio.sync
import tonio.sync.channel as channel

def producer(sender, barrier, offset):
    for i in range(20):
        message = offset + 1
        yield sender.send(message)
    yield barrier.wait()

def consumer(receiver):
    while True:
        try:
            message = yield receiver.receive()
            print(message)
        except Exception:
            break

@tonio.main
def main():
    def close(sender, barrier):
        yield barrier.wait()
        sender.close()

    sender, receiver = channel.channel(2)
    barrier = tonio.sync.Barrier(3)
    yield tonio.spawn(*[
        producer(sender, barrier, 100),
        producer(sender, barrier, 200),
        consumer(receiver),
        consumer(receiver),
        consumer(receiver),
        consumer(receiver),
        close(sender, barrier),
    ])

Network module

Network primitives are exposed under the tonio.net module.

Low-level sockets

The tonio.net.socket module provides TonIO's basic low-level networking API.
Generally, the API exposed by this module mirrors the standard library socket module.

TonIO socket objects are overall very similar to the standard library socket objects, with the main difference being that blocking methods become coroutines.

import tonio
from toio.net import socket

def server():
    sock = socket.socket()
    with sock:
        yield sock.bind(('127.0.0.1', 8000))
        sock.listen()

        while True:
            client, _ = yield sock.accept()
            tonio.spawn(server_handle(client))

def server_handle(connection):
    with connection:
        # receive some data
        data = yield connection.recv(4096)

def client():
    sock = socket.socket()
    with sock:
        yield sock.connect(('127.0.0.1', 8000))
        yield sock.send("message")

Using async/await notation

All TonIO primitives ships with an async/await syntax compatible variant under the tonio.colored module.

Warning: despite the fact TonIO supports async and await notations, it's not compatible with any asyncio object like futures and tasks.

import tonio.colored as tonio

@tonio.main
async def main():
    event = tonio.Event()

    async def setter():
        await tonio.sleep(1)
        event.set()

    tonio.spawn(setter())
    await event()

The only major syntax difference between the yield and async/await notation is around with blocks:

from tonio.sync import Lock

lock = Lock()

def yield_lock():
    with (yield lock()):
        # do something

async def async_lock():
    async with lock:
        # do something

Also, the colored module provides the additional yield_now awaitable function, a quick way to define a suspension point:

import tonio.colored as tonio

@tonio.main
async def main():
    await tonio.yield_now()
    print("hello world")

License

TonIO is released under the BSD License.