Uh oh!
There was an error while loading.Please reload this page.
- Notifications
You must be signed in to change notification settings - Fork33.7k
Description
Bug report
I found a bug that seems to be code corruption.
While working on an example project with ~70 threads, I occasionally (once every hour or so) get the following exception from various locks:
Exception in thread Sequence 2:Traceback (most recent call last): File "/Volumes/RAMDisk/installed-nogil-main/Library/Frameworks/Python.framework/Versions/3.14/lib/python3.14/threading.py", line 1039, in _bootstrap_inner self.run() ~~~~~~~~^^ ... File "/Volumes/RAMDisk/installed-nogil-main/Library/Frameworks/Python.framework/Versions/3.14/lib/python3.14/threading.py", line 656, in wait with self._cond: ^^^^^^^^^^ File "/Volumes/RAMDisk/installed-nogil-main/Library/Frameworks/Python.framework/Versions/3.14/lib/python3.14/threading.py", line 304, in __enter__ return self._lock.__enter__() ~~~~~~~~~~~~~~~~~~~~^^TypeError: descriptor '__exit__' for '_thread.RLock' objects doesn't apply to a '_thread.lock' objector
Exception in thread Clock:Traceback (most recent call last): File "/Volumes/RAMDisk/installed-nogil-main/Library/Frameworks/Python.framework/Versions/3.14/lib/python3.14/threading.py", line 1039, in _bootstrap_inner self.run() ~~~~~~~~^^ File "europython.py", line 113, in run for message in input: File "/Volumes/RAMDisk/nogil-ep-temp/lib/python3.14/site-packages/mido/ports.py", line 243, in __iter__ yield self.receive() ~~~~~~~~~~~~^^ File "/Volumes/RAMDisk/nogil-ep-temp/lib/python3.14/site-packages/mido/ports.py", line 215, in receive with self._lock: ^^^^^^^^^^TypeError: descriptor '__exit__' for '_thread.lock' objects doesn't apply to a '_thread.RLock' objectThis looks like it's a bug related to locks but it isn't. It's not even related to descriptors, only descriptors nicely refuse running invalid code.
This issue is also externally reported inPyWavelets/pywt#758 with the same Lock descriptor error message I've seen, and I can reproduce the failure locally, albeit with a different exception:
TypeError: descriptor 'sort' for 'numpy.ndarray' objects doesn't apply to a 'ThreadPoolExecutor' objectTo reproduce this with cpythonmain, do the following:
- make a venv with a free-threaded build of Python
- install Cython from main with
pip install -e . - install Numpy from main with
pip install . --no-build-isolation(important: no-ein this case) - install pywt from main with
pip install -e . --no-build-isolation(important: youDO need-ein this case) - run pytest in a loop (or withautoclave) like this:
PYTHON_GIL=0 pytest pywt/tests/test_concurrent.py
You will need to run this for a longer while to get to a failure.
By doing this, I managed to find this particular failure case:
self = <concurrent.futures.thread.ThreadPoolExecutor object at 0x225be124750>, fn = functools.partial(<function dwtn at 0x225be6d3b40>, wavelet='haar')args = (array([[1., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.], [0., 1., 0., 0., 0., 0., 0., 0., 0., ...., 0., 0., 0., 0., 0., 0., 0., 0., 1., 0.], [0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 1.]]),)kwargs = {}, f = <Future at 0x225bea44310 state=finished returned dict>, w = <concurrent.futures.thread._WorkItem object at 0x225bc875bf0> def submit(self, fn, /, *args, **kwargs): with self._shutdown_lock, _global_shutdown_lock: if self._broken: raise BrokenThreadPool(self._broken) if self._shutdown: raise RuntimeError('cannot schedule new futures after shutdown') if _shutdown: raise RuntimeError('cannot schedule new futures after ' 'interpreter shutdown') f = _base.Future() w = _WorkItem(f, fn, args, kwargs) self._work_queue.put(w)> self._adjust_thread_count()E TypeError: Future.set_result() missing 1 required positional argument: 'result'args = (array([[1., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.], [0., 1., 0., 0., 0., 0., 0., 0., 0., ...., 0., 0., 0., 0., 0., 0., 0., 0., 1., 0.], [0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 1.]]),)f = <Future at 0x225bea44310 state=finished returned dict>fn = functools.partial(<function dwtn at 0x225be6d3b40>, wavelet='haar')kwargs = {}self = <concurrent.futures.thread.ThreadPoolExecutor object at 0x225be124750>w = <concurrent.futures.thread._WorkItem object at 0x225bc875bf0>../installed-nogil-main/Library/Frameworks/Python.framework/Versions/3.14/lib/python3.14/concurrent/futures/thread.py:179: TypeError======================================================================================================== short test summary info ========================================================================================================FAILED pywt/tests/test_concurrent.py::test_concurrent_dwt - TypeError: Future.set_result() missing 1 required positional argument: 'result'====================================================================================================== 1 failed, 3 passed in 0.44s ======================================================================================================-- 863 runs, 862 passes, 1 failure, 734486 msecObserve how Python wants to callself._adjust_thread_count() (with no arguments) but ends up callingf.set_result(), which causes an exception due to no arguments being passed.
Tested on macOS Sonoma on M1 Max with Python 3.14.0a0 experimental free-threading build (heads/main:7a807c3efaa, Jul 2 2024, 11:58:38).
AFAICT the problem only occurs with the GIL actually disabled.