Eh, we've all been there. My record is patching my own patch two hours after commit for switching a && with ||.

On the side, have one final test running on CI that should tell us whether it was the test that's faulty or the patch.

FYI, @jbergstroem has helped me confirm that the test is what's wrong. Specifically when run with test.py. When run on its own seems to complete fine. Here's the output: https://gist.github.com/jbergstroem/9b9f11f0e43e21cf5cb7

IOW let's give this at least another day of troubleshooting before reverting the change.

LGTM. If nothing is found in the next day, feel free to land.

Why not just to fix the test?

@Olegas that is the plan. The revert is a fallback.

@cjihrig They are already fixed here in #1934. This is how they are originally implemented. There are some weird timing issues I think we can address in another issue later.

I'm still working on troubleshooting the test so don't land this yet. I'll report back a little later today.

@Olegas Unfortunately if a test fails then it's considered faulty. If it can't be figured out then the entire patch will be reverted. This is just standard protocol to allow the patch to be hammered out without causing any failures on CI. Re-landing it again once it's more solid won't be an issue at all.

Unnecessary, test fixed in #1953