resource tracker dies due to unsafe late-forking?

[srepmub]

tqdm triggers multiprocessing.resource_tracker to fork off in tqdm.create_mp_lock. and forking is unsafe for multi-threaded programs (when not using forkserver/spawn start methods). we very reliably get this under OSX for our multi-hreaded program:

"/python3.10/multiprocessing/resource_tracker.py:105: UserWarning: resource_tracker: process died unexpectedly, relaunching. Some resources might leak."

we tried to use set_start_method('forkserver'), but for some reason this led to the same error. so we now add this at import time to our code:

import tqdm
from multiprocessing import RLock
RLock()

which seems to work around the problem for us.. of course it would be nicer if tqdm could fix/work around the issue itself..? thanks!