Drop Python 2.7

[hynek]

We have paid our dues.

The tooling situation around Python 2 has deteriorated to such a degree that
upholding compatibility is not tenable anymore for a volunteer-run project.

Here's the stats for March so far:

| category | percent |  downloads |
|:---------|--------:|-----------:|
| 3.7      |  33.67% | 18,582,627 |
| null     |  19.82% | 10,936,104 |
| 3.8      |  18.75% | 10,344,976 |
| 3.9      |  11.20% |  6,180,529 |
| 3.6      |  10.88% |  6,006,746 |
| 3.10     |   2.95% |  1,625,923 |
| 2.7      |   1.87% |  1,030,024 |
| 3.5      |   0.84% |    463,005 |
| 3.4      |   0.02% |     10,044 |
| 3.11     |   0.01% |      7,899 |
| 3.3      |   0.00% |         25 |
| 2.6      |   0.00% |          8 |
| 3.2      |   0.00% |          2 |
| Total    |         | 55,187,912 |

Date range: 2022-03-01 - 2022-03-18

The main impetus being that Python 2.7 is as of macOS 12.3 not usable with virtualenv anymore. I will probably drop 3.5 next since nobody uses it and I can't install it anymore.

There's a lot of code that could probably be simplified (e.g. when just aliases are imported from _compat), but we can keep that for another PR. Same as ridding docs of (object)s.

[hynek]

I'm gonna need some guidance re __qualname__.

We've got partial coverage in

attrs/src/attr/_make.py

Lines 864 to 866 in b093bc7

	qualname = getattr(self._cls, "__qualname__", None)
	if qualname is not None:
	cd["__qualname__"] = qualname

(looks like qualname's never None)

and

attrs/src/attr/_make.py

Lines 1945 to 1950 in 2047cfa

	qualname = getattr(real_cls, "__qualname__", None)
	if qualname is not None:
	# This case only happens on Python 3.5 and 3.6.
	class_name = qualname.rsplit(">.", 1)[-1]
	else:
	class_name = real_cls.__name__

Here:

Line 1950 is never executed so qualname appears to be never None, contradicting the comment??? Or is the comment so old that it just meant "new Pythons"?

---

Qualified names have been added in 3.3, but maybe there's some weird way to create classes without a qualname?

[ssbarnea]

Bit late but drop it! Better late than never.

[sscherfke]

🎉 🎉 🎉

[hynek]

So I've decided:

1. the second instance is a non-issue, since the new f-string-based repr never cared about it'
2. the first one is a risk we'll take. i mean i've got a tradition of quickly releasing a hot fix release for edge cases to uphold

[hynek]

Less emoji, more reviewing gentlepeople! ;)

[ssbarnea]

I rarely see people that do not have review rights (write) doing format review approvals on github. Even when I explicitly asked them to do it, it got very poor results. Maybe you will be more successful. Another issue is the size of the PR, usually I would have only modify metadata to require newer python and remove the CI jobs for py27, before gradually remove the code in follow-ups, so the reviews are more humanely reviewable.

[sscherfke]

I skimmed the changes and haven't noticed any problems. Also ran the tests on my machine and checked that it does not break Typed Settings.

[wbolster]

left a few comments, mostly nits

didn't spot anything (too) weird 🙃

[Tinche]

Cattrs tests pass ;)