`Executor.map()` exceptions lose the worker traceback and form a cycle

[fedonman]

Bug description:

Since gh-108518, the exception that Executor.map() re-raises for a failed call differs from 3.15 in two ways:

• ProcessPoolExecutor.map() no longer sets the worker-side traceback as its __cause__, so the traceback only shows raise exc in _MapResultIterator.__next__.
• For every executor, the exception is kept alive by a reference cycle through the __next__ frame, together with the frames and locals of the failed call, until the cyclic GC runs.

import gc, weakref
from concurrent.futures import ProcessPoolExecutor

class E(Exception):
    pass

def f(x):
    raise E(x)

if __name__ == '__main__':
    gc.disable()
    with ProcessPoolExecutor() as ex:
        try:
            next(ex.map(f, [1]))
        except E as e:
            print('cause:', type(e.__cause__).__name__)
            ref = weakref.ref(e)
    print('freed:', ref() is None)

On main:

cause: NoneType
freed: False

Expected, as on 3.15:

cause: _RemoteTraceback
freed: True

CPython versions tested on:

CPython main branch

Operating systems tested on:

Linux

Linked PRs

• gh-158497: Fix traceback loss and a reference cycle in Executor.map() #158499