Preliminary python 3.6 port - development #2564

[phillxnet]

As part of our move to Python 3 we must first establish our
initial python 3 dependencies via Poetry.

Fixes #2564

Includes:

• Update pyproject.toml and poetry.lock for python 3.6.
• Remove int long designator from settings.py: Python 3
  no longer uses the "L" suffix.
• Use full path in initial import conversions. Python 3
  has more stringent import path requirements.
• Update imports for 3.6 email module. These differ in Python 3.
• Encode unicode line for system.osi.md5sum.
  TypeError: Unicode-objects must be encoded before hashing
• Tidy/update init_sftp_config() re Python 3.
  There was a subtle file pointer behaviour change
  between 2.7 and 3.6 which broke this mechanism.
• Address TODO re custom opener re Python 3.
• Services page & Users page - Python 3 style sorted():
  "cmp" sorted parameter and buildin function dropped
  in Python 3.
• init fix of groups page by adaptation to str type.
• Replace chardet dependency with charset-normalizer:
  also used by requests.
• Removes redundant encoding on already encoded groupname.
• Init dashboard fix via gevent/websocket update.
  Thanks to @FroggyFlox for investigations here.
• Init fix of get_users() - non admin users did not show.
  Move to more modern - time limited subprocess.run().
• fix samba, disk partition, rockon, re iteritems()
  In Python 3 we use iter(d.items()) instead:
  https://peps.python.org/pep-0469/
• Address TODO re ascii to utf-8 on SMART custom options.
• Untested bytecast in smart.py:
  custom options autodev related byte assertion.

Independent commits for attribution from @FroggyFlox

• Move UDEVADM and SHUTDOWN definitions to runtime constants.
  Removes now redundant and incompatible mechanism.
• Move passwd to str and use crypt.mksalt() to generate salt.
  Updates this mechanism re Python 3 capabilities.

[phillxnet]

N.B. Although these changes enable most functionality within a far newer Python 3.6 local .venv, they break rpmbuild as we require all tests to pass there. However the key purpose of this pull request is to enable spin-off issues to focus developer efforts on ease parallel tasks on fixing the following:

Ran 252 tests in 20.288s

FAILED (failures=12, errors=6)

Via clearly defined tasks to restore our testing channel to a build-able and thus release capable status.

[phillxnet]

@FroggyFlox & @Hooverdan96

Given this PR gets us to a source build state under python 3.6 (but not rpm buildable due to unit test failure) I propose that we merge it into testing and address the unit test failures as a matter of priority. As many of the tests fail due to Python 3.6 differences re unicode / string / byte i.e. u"some text" etc we likely need a dedicated test issue/s. However we also need to avoid stepping on each others toes. Some tests will of course reflect tested code changes regarding the underlying Python changes we have introduced here.

[FroggyFlox]

Agreed on merging this; that's already in an excellent state, relatively speaking of course. I was going to say "starting point" but that's actually much better than that.

[Hooverdan96]

I know, this is a very minor detail and I don't know where to put this, but since you were moving some of the paths from the settings.py to constants.py, I noticed the shutdown path. In the spirit of simplification, would it make sense to use systemctl poweroff/systemctl reboot instead of the shutdown command/path to get rid of another setting/path?

[phillxnet]

@Hooverdan96 Re:

In the spirit of simplification, would it make sense to use systemctl poweroff instead of the shutdown command/path to get rid of another setting/path?

If you could create an issue for this I think it would be a viable 'upgrade' as you say simplification is good and that may be the way to go, handing off yet more to systemd.

As for this pr I think I'll pop it in as we have much to do by way of follow-up and the sooner it's in the better.

N.B. we would normally be reluctant to merge any pr that breaks testing - as it in-turn breaks our rpmbuild (see %checks):

https://github.com/rockstor/rockstor-rpmbuild/blob/211ed89b89910afbf767459dc5933a52b1073038/rockstor.spec#L249-L251

But this is literally the first pr in our new testing run and we have changed our base language version (where that is significant in Python 2 to 3) so its merge assists with dividing up the remaining conversion work, some of which likely relates to why all tests are no passing.