3535: prevent subprocpool infinite loop

[oliver-sanders]

These changes close #3535

This is a potential fix for the infinite loop observed in the subprocpool process polling logic on darwin (BSD).

Not sure what is causing the bug, however, in the Cylc code it surfaces in _poll_proc_pipes where every time we go around the loop we get data == '' even long after the process has completed.

Requirements check-list

• I have read CONTRIBUTING.md and added my name as a Code Contributor.
• Contains logically grouped changes (else tidy your branch by rebase).
• Does not contain off-topic changes (use other PRs for other changes).
• Appropriate tests are included (impossible to test when you don't know what the underlying issue actually is).
• No documentation update required.

[kinow]

Here's what I did to test it:

• had a look at Travis CI, and it doesn't seem to affect Linux (the code added is not supposed to ever happen from what I understand)
• downloaded the FreeBSD 12 image
• installed on VirtualBox
  • resized disks
  • installed Python 3.7, pip, git, and pyzmq
  • installed Cylc Flow from master
  • created cylc-suites/five/suite.rc 😬

Happy to report Cylc Flow works on FreeBSD 12.1. But alas I couldn't reproduce it. Workflow five has been fine since ~4:30PM (it's ~5:30PM now).

Tried PureDarwin but too hard to install. If it fixes the issue on MacOS, then +1 from me, but probably needs @hjoliver and maybe others from MO to review it too.

[kinow]

Conflict in CHANGES.md @oliver-sanders

And screenshots of Cylc 8 on FreeBSD

[hjoliver]

The code looks fine to me, just wondering what it means exactly if the problem is triggered.

[hjoliver]

Here's what I did to test it: ...

Well done, that's some serious testing 😁

[oliver-sanders]

Tried PureDarwin but too hard to install

Wow @kinow that is dedication to the cause! Darwin is not the easiest OS to handle, I don't know anyone who has actually installed it standalone like that!

When we get this DNS muck sorted out we should use the MacOS platform with GitHub actions to save you the bother of all that!

[oliver-sanders]

Note, 363089e fixes the dependency section of setup.py which seems to have been broken during rebase/merge and will currently cause TB failures on master.

[kinow]

Good catch!

[hjoliver]

Tests passed after a kick.

[oliver-sanders]

Merging with two approvals.