Help needed understanding a test failure in jupyter_client + pypy

[mattip]

It seems there is a problem somewhere releasing a resource so that the jupyter kernel can die. This manifests itself both in a PR to ipykernel and a PR to jupyter_client. I wonder if it has something to do with pyzmq?

If someone who understands the different layers could take a look that would be greatly appreciated.

[minrk]

I poked around, and my strong suspicion is that most if not all of these are related to the difference in garbage collection. this one in particular is exactly that - a direct test of the consequences of del which shouldn't be assumed. I'd argue that test should be updated to test less - just verify that msg['tracker'] is present at all, since the rest is really testing internal pyzmq machinery already covered in this repo.

I was able to infrequently reproduce the uuid-in-stdout failure, and that's another race - there's a time.sleep(1) which is a race, and I guess startup is slower on pypy. That can be fixed with a more direct wait-for-stdout.