In Objects/genobject.c, gen_dealloc() calls PyObject_ClearWeakRefs() before trying to finalize the generator, during which the generator object can be resurrected. This is probably a bug, since weakrefs are supposed to be cleared (and their callbacks called) only when the object is really being destroyed.

(see also issue14933)

I'm still not brave enough to take on C code, but could this be handled by someone from the core-mentorship list?

In the docs https://docs.python.org/3/extending/newtypes.html#weak-reference-support points out:

> /* Clear weakrefs first before calling any destructors */

So maybe we should also update the docs after fix this bug. Anyway, I will try to understand/fix this bug.

The fixed looks easy, we call `PyObject_CallFinalizerFromDealloc` before PyObject_ClearWeakRefs. But I can't come up with the use case for testing when generator resurrects from `PyObject_CallFinalizer`. Any ideas?

If I recall correctly, it's the generator destructor that handles throwing in ``GeneratorExit`` to get the generator to terminate. So this code can resurrect a generator as it's being collected by the GC:

    def resurrecting(resurrected):
        self = yield
        try:
            yield
        finally:
            resurrected.append(self)

    storage = []
    g = resurrecting(storage)
    g.send(g) # Give the generator a reference to itself
    del g # Now the generator is in a cycle with itself
    gc.collect()
    gc.collect()
    gc.collect()
    # Generator has been added to the storage instead of collected
    assert len(storage) == 1
    # Clear the storage to kill it for real this time
    storage.clear()
    # Weakrefs shouldn't get called until here

Antoine, does that sound right to you?