Python 3 on macOS/BSDs: Clush and library fails with no timeout

[volans-]

I wanted to test the latest release to start migrating my tool to Python 3 (thanks a lot for adding Py3 support!) and found that both clush and ClusterShell as a library fails on macOS (and other BSDs) when not setting a timeout, with ERROR: [Errno 22] Invalid argument.

After digging a bit I discovered that:

• in Python 3 the Engine.Factory.PreferredEngine factory returns an EnginePoll on macOS, while on Python 2 it was returning an EngineSelect because poll() was not supported.
• when polling for events in Engine.Poll.EnginePoll.runloop(), the timeout passed to the poll.poll() method is timeo * 1000.0 + 1.0, and when the timeout is not specified, timeo = -1, hence the passed timeout is -999.0.
• Python passes the timeout parameter of select.poll.poll([timeout]) basically as-is down to the poll() system call
• On macOS and other BSDs the poll() system call throws an error on arbitrary negative values. Only non-negative integers and -1 (or the value of INFTIM if defined) are accepted.

I've already opened an issue to Python [1] and sent a PR to hopefully fix it [2], but I'll follow up with a PR to ClusterShell too to workaround this issue passing directly -1 to the poll() method.

[1] https://bugs.python.org/issue31334
[2] python/cpython/pull/3277

[thiell]

@volans- thanks for the detailed issue, the PR and for the feedback to the Python community!

This problem is also present in Python 2 on the latest FreeBSD 11.1:

$ python2
Python 2.7.13 (default, Aug 15 2017, 01:13:59) 
[GCC 4.2.1 Compatible FreeBSD Clang 3.8.0 (tags/RELEASE_380/final 262564)] on freebsd11
Type "help", "copyright", "credits" or "license" for more information.
>>> import select
>>> p = select.poll()
>>> p.poll(-2)
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
select.error: (22, 'Invalid argument')

Looks like they added support for poll() there too.

I also noticed other errors with clush on this OS, but that's off-topic.

[volans-]

@thiell thanks for the additional info, I'll look into it. If/when I'll get some answer on the python issue I'll ask also to backport it.

On a side note, have you considered to use the selectors module for the polling?
It should allow to simplify a lot the polling-related code given that it abstracts the different polling systems, doing something similar of what was done here.
It looks like there are two Python 2 backward-compatible modules available: selectors2 and selectors34.
I didn't check if it covers all the use cases ClusterShell needs, so it's possible that the direct use of the lower level modules is still needed anyway.