asyncio eager task inherits the creators exception state

Bug report

Bug description:

Eager tasks created from an except block inherits the creator's exception state, a bare raise re-raises it, and any other exception gets it as __context__.

Repro:

import asyncio


async def bare():
    raise


async def explicit():
    raise ValueError("child's own")


async def main(coro):
    try:
        raise ConnectionError("primary failed")
    except ConnectionError:
        t = asyncio.create_task(coro)
    try:
        await t
    except BaseException as e:
        print(f"  {type(e).__name__}: {e} | __context__: {e.__context__!r}")


for eager in (False, True):
    print("eager_task_factory:", eager)
    for coro_fn in (bare, explicit):
        loop = asyncio.new_event_loop()
        if eager:
            loop.set_task_factory(asyncio.eager_task_factory)
        try:
            loop.run_until_complete(main(coro_fn()))
        finally:
            loop.close()

expected:

eager_task_factory: False
  RuntimeError: No active exception to reraise | __context__: None
  ValueError: child's own | __context__: None
eager_task_factory: True
  RuntimeError: No active exception to reraise | __context__: None
  ValueError: child's own | __context__: None

actually:

eager_task_factory: False
  RuntimeError: No active exception to reraise | __context__: None
  ValueError: child's own | __context__: None
eager_task_factory: True
  ConnectionError: primary failed | __context__: None
  ValueError: child's own | __context__: ConnectionError('primary failed')

proposed fix - run the eager first step under an empty exception state in task_eager_start

    _PyErr_StackItem exc_state = { .exc_value = NULL, .previous_item = NULL };
    _PyErr_StackItem *prev_exc_info = ts->base.exc_info;
    ts->base.exc_info = &exc_state;
    Py_BEGIN_CRITICAL_SECTION(task);
    stepres = task_step_impl(state, task, NULL);
    Py_END_CRITICAL_SECTION();
    ts->base.exc_info = prev_exc_info;

This error also affects python implementation of task, but i think this needs a discussion in a separate issue

Have a fix ready

CPython versions tested on:

CPython main branch

Operating systems tested on:

macOS

Linked PRs

• gh-155578

Comece reproduzindo o comportamento de tarefas eager no exemplo de asyncio da issue e inspecione o ponto de entrada task_eager_start. Compare o tratamento de exceções eager e não eager e, em seguida, verifique se a saída esperada relatada é produzida sem alterar a implementação separada de tarefas em Python.