bug: dependency ceiling markdown>=3.3,<3.6 acts as a timebomb

[ssbarnea]

Description of the bug

This dependency "time-bomb" was recently introduced in cd93ee3 - i supposed as it was known that the new ones cased some failures.

Expected behavior

Be compatible with latest versions of its dependencies, especially as other mkdocs plugins might rely newer versions, so a conflict could happen.

Usually I prefer to have a different pipeline that tests unreleased versions of major major dependencies, so I know when they are about the break before they release. This sorts the problems of having to add ceiling when problems are encountered.

Environment information

python -m mkdocstrings_handlers.python.debug  # | xclip -selection clipboard

PASTE OUTPUT HERE

Additional context

[pawamoy]

From https://github.com/ansible/ansible-navigator/actions/runs/8408879265/job/23025573920?pr=1721:

The user requested (constraint) markdown==3.6

Does this constraint come from dependabot or from the project itself? If the latter, do you actually require markdown==3.6? This version is indeed breaking some features of mkdocstrings-python, this is why I added the cap.

In any case, I agree that mkdocstrings-python should support markdown 3.6. I just didn't get to it yet 🙂

[ssbarnea]

Yep, it should support 3.6. I only mentioned the other things for the future. To avoid similar problem when 3.7 is released.

I know how to bypass it but better not to have.

[pawamoy]

Yup, I usually never cap my dependencies.

Usually I prefer to have a different pipeline that tests unreleased versions of major major dependencies, so I know when they are about the break before they release. This sorts the problems of having to add ceiling when problems are encountered.

In this particular case, I knew 3.6 would break mkdocstrings-python before it was released, so I added the cap before the release (in sync with @waylan, Python-Markdown's maintainer). Since this won't change in 3.7+, I didn't just exclude 3.6 with !=3.6. Also, the breakage is not easily detected, at least not with our current test suite. I have it in my backlog to add end-to-end tests to facilitate detecting such breakages. Lots of work 😄

[pawamoy]

Fixed. For reference: Python-Markdown/markdown#1441, mkdocstrings/mkdocstrings@c0d0090.