Drop support for Python 3.6

[sethmlarson]

This PR only effects the v2.x development branch, v1.26.x will continue to support Python 3.6.

The reason for this PR is two-fold:

• We're unlikely to release v2.0 before Python 3.6 being EOL in December of 2021.
• There are multiple features in the v2.0 timeline that will be dramatically easier to implement without having to think about Python 3.6 (see Start using SSLContext.minimum|maximum_version when available #2220 for an in-progress example)

I think removing now is best to not impede development and if by some stroke of luck we complete all v2.0 deliverables in the Python 3.6 timeline we'll re-evaluate.

[codecov]

Codecov Report

Merging #2332 (bcdddc5) into main (e772731) will not change coverage.
The diff coverage is 100.00%.

@@            Coverage Diff            @@
##              main     #2332   +/-   ##
=========================================
  Coverage   100.00%   100.00%           
=========================================
  Files           25        25           
  Lines         2439      2452   +13     
=========================================
+ Hits          2439      2452   +13     

Impacted Files	Coverage Δ
src/urllib3/_collections.py	100.00% <100.00%> (ø)
src/urllib3/connection.py	100.00% <100.00%> (ø)
src/urllib3/util/ssl_.py	100.00% <100.00%> (ø)
src/urllib3/response.py	100.00% <0.00%> (ø)
src/urllib3/exceptions.py	100.00% <0.00%> (ø)
src/urllib3/util/retry.py	100.00% <0.00%> (ø)
src/urllib3/poolmanager.py	100.00% <0.00%> (ø)
src/urllib3/connectionpool.py	100.00% <0.00%> (ø)

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update e772731...bcdddc5. Read the comment docs.

[pquentin]

Deleted my previous comment as I hadn't seen all the code, sorry.

Barring the codecov issue, this looks good but it would be nice if our type experts @haikuginger and @hramezani could take a look as types are evolving rapidly. I think @haikuginger mentioned in the past that 3.6 made things more difficult type wise, and that RLock wasn't the only issue.

And I guess mentioning @hugovk is always a good idea when you stop supporting a Python version :p

[shazow]

Looks sensible. I'm not intimately familiar with these details but let me know if you require my approval. :)

[haikuginger]

This looks correct in general; I'm just going to re-acquaint myself with why the RLock compat code was 3.6-only before approving.

[haikuginger]

LGTM; this is exciting! Found the doc where threading was made non-optional in 3.7.

[pquentin]

I'll look into the hostname_checks_common_name PyPy issue. (Its availability depends on OpenSSL, even on CPython.)

[hramezani]

I think we have typing.OrderedDict in Python > 3.7.
So, I think we can typing.OrderedDict use in

urllib3/src/urllib3/_collections.py

Line 112 in 8c905b0

_container: "OrderedDict[_KT, _VT]"

[hramezani]

I think we can remove nox > Session test-3.6 was successful. and a line for 3.10 maybe from https://github.com/urllib3/urllib3/blob/5b61732f2403c378bb60b451d95ce2835f6b5485/docs/contributing.rst

[hramezani]

There is an Optimized for Python 3.6+ section in https://github.com/urllib3/urllib3/blob/2ef05adfcf001c85f1f95ae3fa078d16190851b0/docs/v2-roadmap.rst
which may need some change.

[pquentin]

There's another error message that changed on PyPy, that I still need to fix, but the hostname_checks_common_name bug on our end is fixed.

[pquentin]

CI is green again, we just need to address the comments from @hramezani now.

[hramezani]

@SethMichaelLarson @pquentin I can address them if you want.

[pquentin]

Yes, please do! You should be allowed to push to Seth's branch directly.

I also noticed another change needed in test_ssltransport.py (search for 3.7)

[hramezani]

@sethmlarson It seems I don't have permission to push. I've created a new PR #2336

[hramezani]

Closing in favor of #2336