Multiprocessing with "fork" fails with Python 3.14 but works with Python 3.13

[LostInDarkMath]

Bug report

Bug description:

The following code works fine on Python 3.13, but fails on Python 3.14 with a RuntimeError related to asyncio tasks.

Minimal reproducible example

import asyncio
import inspect
from functools import wraps
from typing import Any, Awaitable, Callable, Union

import pytest
from multiprocess import Pipe, Process
from multiprocess.connection import Connection

import multiprocess as mp
mp.set_start_method(method="fork", force=True)  # "spawn" works fine


class SubprocessError:
    def __init__(self, ex: Exception) -> None:
        self.exception = ex


def in_subprocess[T](func: Callable[..., Union[T, Awaitable[T]]]) -> Callable[..., Awaitable[T]]:
    @wraps(func)
    async def wrapper(*args: Any, **kwargs: Any) -> T:
        return await calculate_in_subprocess(func, *args, **kwargs)

    return wrapper


async def calculate_in_subprocess[T](func: Callable[..., Union[T, Awaitable[T]]], *args: Any, **kwargs: Any) -> T:
    rx, tx = Pipe(duplex=False)  # receiver & transmitter ; Pipe is one-way only
    process = Process(target=_inner, args=(tx, func, *args), kwargs=kwargs)
    process.start()

    event = asyncio.Event()
    loop = asyncio.get_event_loop()
    loop.add_reader(fd=rx.fileno(), callback=event.set)

    if not rx.poll():  # do not use process.is_alive() as condition here
        await event.wait()

    loop.remove_reader(fd=rx.fileno())
    event.clear()

    result = rx.recv()
    process.join()  # this blocks synchronously! make sure that process is terminated before you call join()
    rx.close()
    tx.close()

    if isinstance(result, SubprocessError):
        raise result.exception

    return result


def _inner[T](tx: Connection, fun: Callable[..., Union[T, Awaitable[T]]], *a, **kw_args) -> None:
    event_loop = None
    if inspect.iscoroutinefunction(fun):
        event_loop = asyncio.new_event_loop()
        asyncio.set_event_loop(event_loop)

    try:
        if event_loop is not None:
            res = event_loop.run_until_complete(fun(*a, **kw_args))
        else:
            res = fun(*a, **kw_args)
    except Exception as ex:
        tx.send(SubprocessError(ex=ex))
    else:
        tx.send(res)


@pytest.mark.asyncio
async def test_in_subprocess_simple_async():
    @in_subprocess
    async def f() -> int:
        return 42

    assert await f() == 42

---

Error message with Python 3.14

-------------------------------- live log call ---------------------------------
ERROR    asyncio:base_events.py:1875 Exception in callback <_asyncio.TaskStepMethWrapper object at 0x7e71ba729ff0>()
handle: <Handle <_asyncio.TaskStepMethWrapper object at 0x7e71ba729ff0>()>
Traceback (most recent call last):
  File "/usr/lib/python3.14/asyncio/events.py", line 94, in _run
    self._context.run(self._callback, *self._args)
    ~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
RuntimeError: Cannot enter into task <Task pending name='Task-2' coro=<test_in_subprocess_simple_async.<locals>.f() running at foo.py:73> cb=[_run_until_complete_cb() at /usr/lib/python3.14/asyncio/base_events.py:181]> while another task <Task pending name='Task-1' coro=<test_in_subprocess_simple_async() running at foo.py:77> cb=[_run_until_complete_cb() at /usr/lib/python3.14/asyncio/base_events.py:181]> is being executed.

---

Environment

Installed packages (note: multiprocess must be installed from GitHub):

certifi==2025.10.5
charset-normalizer==3.4.3
dill==0.4.0
docker==7.1.0
idna==3.10
iniconfig==2.1.0
multiprocess @ git+https://github.com/uqfoundation/multiprocess.git@02ea4bd36cac5013d70847815c92e1a736ef4a05
packaging==25.0
pluggy==1.6.0
Pygments==2.19.2
pytest==8.4.2
pytest-asyncio==1.2.0
pytest_docker_tools==3.1.9
requests==2.32.5
urllib3==2.5.0

---

And before you say, that the error is in the third-party lib multiprocess and I should use multiprocessing instead of multiprocess: The bug even occurs with multiprocessing (just replace the lib in the code example above). In Python 3.13 it will fail with a PicklingError (which is expected for this example) but in Python 3.14 the RuntimeError: Cannot enter into task will be thrown (see above). So this issue must be in Python 3.14.

CPython versions tested on:

3.14

Operating systems tested on:

Linux

Linked PRs

• gh-139894: fix incorrect sharing of current task while forking in asyncio  #139897
• [3.14] gh-139894: fix incorrect sharing of current task while forking in asyncio (GH-139897) #139913

[kumaraditya303]

Thanks for the report, I have a fix at #139897 which fixes it.

While forking works somewhat in asyncio, its usage is still discouraged because of other issues of fork like #130442