fix behavior with empty matches

[andreasvc]

Python 3.7 fixed bugs related to empty matches: https://bugs.python.org/issue1647489

[sarnold]

Okay, but I don't yet see anything pointing to this behavior change, or how using the patch you just merged exhibits the old behavior in ubuntu focal with python 3.9.6 but I get the new behavior in python 3.9.6 locally. WTH??

Anyway, I got bounced from the original "this should match perl's behavior" discussion from 2007:

https://bugs.python.org/issue1647489

to a closed bug and then a closed PR with no apparent fix:

python/cpython#4678

to another (later) closed bug on Python 3.7, Python 3.6, Python 2.7:

https://bugs.python.org/issue25054

and now too I'm tired to bisect and I think I'd rather push a few updates on another PR anyway.

[andreasvc]

It's an obscure issue. If Python cannot make up its mind about it, maybe we should just remove the test and document that the behavior is inconsistent.

[sarnold]

I have this one: sarnold#9 and everything is "fine" except coverage.py chokes with a parse error on the first cimport it sees. I tried the exact same config on the datrie extension and coverage works fine, so I'm not sure if it's a coverage.py bug or the extension code itself. It would be really nice to have some test coverage, but I'm not sue exactly what needs "fixing"...

[sarnold]

Oh, I also split out 2 problematic tests so they can be deselected/marked with pytest. Right now there are 2 places where the None vs '' test fails with the patch applied (which are accounted for in the workflows and tox.ini).

[sarnold]

Finally, this will fail hard in python 3.10: borgbackup/borg#5683

Fixing that ^^ and removing PY2 support should help getting the test coverage enabled.