fix: don't call create_task in ClientSession.__del__ without a running loop

[noron12234]

The problem

ClientSession.__del__ closes the session by scheduling a task:

def __del__(self, _warnings: Any = warnings) -> None:
    """Close the session."""
    if not self.closed:
        asyncio.create_task(self.close())

__del__ runs during garbage collection, and that can happen with no running event loop — a sync context, or interpreter shutdown. asyncio.create_task() raises RuntimeError there:

>>> s = ClientSession(); del s        # no running loop
RuntimeError: no running event loop
sys:1: RuntimeWarning: coroutine 'ClientSession.close' was never awaited

Exceptions raised inside __del__ are swallowed by Python and printed to stderr rather than propagated, so this does not crash anything — the session simply never closes and the connection leaks, with a stray warning as the only sign.

There is a second, smaller issue on the path that does work: when a loop is running, the task is discarded, and the loop only keeps a weak reference, so the close can be collected before it finishes.

What aiohttp itself does

Worth noting because this class subclasses aiohttp.ClientSession — upstream's __del__ never schedules anything (aiohttp/client.py:427):

def __del__(self, _warnings: Any = warnings) -> None:
    if not self.closed:
        _warnings.warn(f"Unclosed client session {self!r}", ResourceWarning, source=self)
        context = {"client_session": self, "message": "Unclosed client session"}
        ...
        self._loop.call_exception_handler(context)

It warns and hands off to the exception handler, precisely because __del__ cannot rely on a loop being available. The override here departs from that.

The change

Check for a running loop first; fall back to the same ResourceWarning upstream emits. When a loop is available, keep the close task referenced until it completes.

try:
    loop = asyncio.get_running_loop()
except RuntimeError:
    _warnings.warn(f"Unclosed client session {self!r}", ResourceWarning, source=self)
    return

task = loop.create_task(self.close())
_BACKGROUND_TASKS.add(task)
task.add_done_callback(_BACKGROUND_TASKS.discard)

Verification

Both paths, with the patched __del__:

	before	after
no running loop	RuntimeError: no running event loop + coroutine … was never awaited	ResourceWarning: Unclosed client session — same as aiohttp
running loop	task discarded, collectable mid-close	task held until done

$ ruff check openbb_platform/core/openbb_core/provider/utils/client.py    # ruff 0.12.12, pinned in .pre-commit-config.yaml
All checks passed!
$ black --check openbb_platform/core/openbb_core/provider/utils/client.py # black 25.1.0
1 file would be left unchanged.

No test added: the no-loop path is straightforward to test, but the loop path is a garbage-collection race and a test for it would be flaky by construction. Happy to add a test for the RuntimeError case alone if you'd like one.

Found with an AST scan for create_task / ensure_future results discarded as bare expression statements. This was the only hit under openbb_platform/.

[CLAassistant]

All committers have signed the CLA.