Improve handling of IDNA/Unicode domains

[felixfontein]

SUMMARY

Improve IDNA/Unicode handling.

Fixes #426.

ISSUE TYPE

• Feature Pull Request

COMPONENT NAME

various

[felixfontein]

One problem with the current approach is that Python's IDNA encoding apparently is kind of outdated, see https://pypi.org/project/idna/ (project description) and the official docs https://docs.python.org/3/library/codecs.html#module-encodings.idna. It might be better to add an optional dependency on the idna library and use it to do the conversion, and fail if a conversion is needed and the library is not present.

[felixfontein]

Note that idna is only required if IDNs are used. If IDNs worked before with an older cryptography / PyOpenSSL version, then it was already using IDNA to do the conversion (and thus IDNA was already installed), so this isn't a breaking change for any existing configuration. (Assuming nobody patched their cryptography to use Python's IDNA2003 implementation instead :) )

[felixfontein]

The Python 3.5 version included in the Ansible 2.9 default test container has a stupid bug which causes the above error. This has been fixed in later versions of Python 3.5 (the default test container coming with devel doesn't have it for example). (This was fixed in python/cpython@4655d57#diff-b3712475a413ec972134c0260c8f1eb1deefb66184f740ef00c37b4487ef873e as part of https://bugs.python.org/issue36742, and the fix first appeared in 3.5.8 rc 1.)

[felixfontein]

ready_for_review

[felixfontein]

Confirmed to be working as expected in #426 (comment)

[briantist]

Looks great! Love to see i18n support

[felixfontein]

@briantist thanks a lot for reviewing this!