Require weaker Sphinx dependency for sagelib

[tobiasdiez]

The version constraint added in #38549 broke the conda ci workflow and added an unnecessarily strict version constraint for sphinx as a dependency of sagelib (as opposed to sage-the-docs).

This is fixed by providing the old version constraint for sphinx in setup.cfg, but with v8 now allowed.

Alternatively, the stricter version constraint should only be added to the sage-the-docs package.

📝 Checklist

• The title is concise and informative.
• The description explains in detail what this PR is about.
• I have linked a relevant issue or discussion.
• I have created tests covering the changes.
• I have updated the documentation and checked the documentation preview.

⌛ Dependencies

[kiwifb]

My update to sage_autodoc.py does require sphinx 7.4.x at the minimum otherwise building the doc breaks. Breaking doc building would also break CI (somewhere else). Now I had to walk back some changes in a follow up at #38619 to continue supporting python 3.9 :( If you are telling me that enables lower version of sphinx as well, then sure.

Is the problem that you do not an appropriate version of sphinx in conda or something like that?

[tobiasdiez]

My update to sage_autodoc.py does require sphinx 7.4.x at the minimum otherwise building the doc breaks. Breaking doc building would also break CI (somewhere else). Now I had to walk back some changes in a follow up at #38619 to continue supporting python 3.9 :( If you are telling me that enables lower version of sphinx as well, then sure.

The version constraint in setup.cfg is not for building the documentation but only for the very minimal usage of sphinx in sage-the-library (namely in https://github.com/sagemath/sage/blob/develop/src/sage/misc/sphinxify.py). That file didn't saw any serious updates in ages, so it should still work with v5 and 6.

[kiwifb]

I hear you on the sphinxify bit. But CI also builds documentation, there is not a process in sage's build system to have one sphinx to build sage and then update it to build the documentation.

[tobiasdiez]

Sage-the-distro (and the build system included in it), the ci doc build as well as the modularization sage-the-doc package will still use the stronger version constraint that you set. The changes in this PR only affect people trying to use only the library sagemath-standard (say via pypi).

[github-actions]

Documentation preview for this PR (built with commit ea94761; changes) is ready! 🎉
This preview will update shortly after each push to this PR.

[kiwifb]

OK, I am getting that argument. Although it is completely disjoint from the CI which would mean the CI label you added is misleading. But I am guessing you are also telling me the conda CI does not build the doc and therefore can use your lower constraint?

[tobiasdiez]

OK, I am getting that argument. Although it is completely disjoint from the CI which would mean the CI label you added is misleading. But I am guessing you are also telling me the conda CI does not build the doc and therefore can use your lower constraint?

Exactly, the conda workflow is still using sphinx v7.3 (but doesn't build the docs). The ci label is so that this PR get applied to the other PRs so that the "verify dependencies" step passes there.
https://github.com/sagemath/sage/actions/runs/10732023722/job/29763038707#step:9:9
vs
https://github.com/sagemath/sage/actions/runs/10734686493/job/29770380015#step:9:9

[kiwifb]

I see, and this particular file inherit my constraint. I must say I am always wary of dealing with sage-the-distro and avoid it as much as possible. This is clearly one of those case where I'd rather not be involved. But now that I have done all that digging on the details of your rationale, it'd be silly not to approve.

[kiwifb]

Good ahead if it really helps CI.

[tobiasdiez]

A bit late, but thanks @kiwifb for the review!