Fix test_print_to_correct_cell_from_child_thread

[davidbrochart]

The parent thread has to live for the duration of the child thread, otherwise the parent thread exits before the child thread and disappears from the active threads, causing it to be removed from thread_to_parent and eventually leading to messages not being routed to the right cell:

ipykernel/ipykernel/ipkernel.py

Lines 781 to 794 in 41a965e

	thread_to_parent_header = stream._thread_to_parent_header
	for identity in list(thread_to_parent_header.keys()):
	if identity not in active_threads:
	try:
	del thread_to_parent_header[identity]
	except KeyError:
	pass
	thread_to_parent = stream._thread_to_parent
	for identity in list(thread_to_parent.keys()):
	if identity not in active_threads:
	try:
	del thread_to_parent[identity]
	except KeyError:
	pass

@krassowski Could you confirm? I saw that while working on #1291.

[krassowski]

Does it also work if in addition to this sleep you keep the sleep({interval}) above? I wonder if that was here for a reason. In other words, I think that adding the sleep at the end is fine, I am less confident about the function of the sleep that was removed (as this can change some edge case related to execution order).

Further, should we save a reference to the thread in a local variable before calling sleep()?

[davidbrochart]

In 41b31c5 I added back the first sleep, and replaced the other sleep with a thread.join(), since it's essentially the same.

[minrk]

I think also related to this is #1289 . Most of the time, in my experience, routing output to the cell that happened to spawn the thread not the right thing to do, and results in quite a bit of mysteriously lost and out-of-order output.

[krassowski]

Thanks for linking #1289, I missed it when it was opened (despite the mention!). I see your point. I think that usually routing to the spawning cell is the right thing to do, but will happily agree that there should be a way to opt out for the scenarios that you present.

[davidbrochart]

Failure is unrelated:

FAILED tests/test_subshells.py::test_run_concurrently_sequence[True-are_subshells0] - AssertionError: assert 'error' == 'ok'

[krassowski]

LGTM!