Pathlib everywhere

[gforcada]

While looking at the code I noticed that we were using plenty of os.sep and os.path.join that since Python 3.4 is so much more nicer to use pathlib.Path instead.

The diff turned out quite big, but it's only about replacing old with new usages, no functionality changes.

As an extra bonus there are a couple of minor improvements added 😄

[gforcada]

Interestingly, on all python versions works but on 3.10 🤷🏾

[reinout]

The 3.10 error is caused by python/cpython#88227
Though.... I mis-looked and took 3.8.10 to be 3.10. It looks similar, though.

[coveralls]

coverage: 99.967% (-0.001%) from 99.968%
when pulling b2acd51 on pathlib-everywhere
into 131da6a on master.

[reinout]

3.10 is still a problem. Weird. Looks like the bug I mentioned.
Barbaric idea: remove 3.10 from the test? We're testing 3.9 and 3.11, if those two work...

[gforcada]

Yes, we could do that indeed, maybe with a big disclaimer, or directly a sys.exit(1) if the user tries to run it on Python 3.10? 🤔

[reinout]

Hm, I updated the PR from master as I wanted to get the update to the "setup-python" action merged. But it also grabbed some other changes from master. Anyway, I see it still fails on 3.10.

[reinout]

I've removed 3.10 from the tests and added a note to the changelog. I assume most people using this tool use a more recent python version. And otherwise they'll probably look at the changelog when they see a sudden error.

[reinout]

@gforcada , I did some mocking in 582fd4f, can you check that?

Weird, I must have done something wrong, as the code isn't reached?? Anyway, I'm off home to dinner :-)

[reinout]

Coverage went down 0.001% because we removed a few lines somewhere. Failing on that is too restrictive to my taste, so I've set the threshold for going down to a generous 0.5% :-)

@gforcada, if you're happy with my mucking-about with the tests, you can merge it.