Revert "singlehtml: deprecate the 'fix_refuris' helper function (#13037)"

[jayaddison]

Purpose

As reported in #14221, the fix_refuris method does appear to prevent duplicate fragment delimiters (the # character) from appearing in the hyperlinks output by the singlehtml builder.

In my opinion we should still consider fix_refuris deprecated -- I'd still like to remove it, since it seems like an unnecessary post-processing of the documents; but additional investigation is required to confirm what it does and whether a better alternative mechanism can be found.

References

• Accompanies Tests: add reference URI test coverage for 'singlehtml' builder #14240.
• Resolves Incorrect HTML anchors generated by singlehtml builder #14221.
• Reverts singlehtml: deprecate the 'fix_refuris' helper function #13037

[jayaddison]

Self-quoting:

In my opinion we should still consider fix_refuris deprecated -- I'd still like to remove it, since it seems like an unnecessary post-processing of the documents; but additional investigation is required to confirm what it does and whether a better alternative mechanism can be found.

Please refer to this comment for the results of this investigation: #14221 (comment)

[nvetrini]

Hi,

just checking if there is any feedback on this PR.

Thanks

[kreuzberger]

It would be helpful to fix the singlethml builds for sphinx 9.x therefore 👍 for this PR.
But as long as the checks are failing i think nobody will merge this?

[nvetrini]

Hi @kreuzberger . Thank you for the feedback. I might try to fix those if I have time.

For the Python 3.12 check only one failed test, as far as I can see. It has been reported to be unreliable in #14104

[jayaddison]

This depends on #14240, that is ready for review.

[jayaddison]

The latest unit test failures (e.g. on cf5af53) seem to relate to HTML encoding differences - single-quotes and double-quotes are no-longer HTMLescaped by Sphinx (to ' and " respectively), for example, while the test suite continues to expect that escaping in a few places.

I'm not sure what has caused that behaviour to change, though. Perhaps the Python version used by the GitHub runners has updated with a security fix?

I'll report a CI failure issue, and have some local changes prepared that should fix it -- but it would be good to figure out why it happened.

[jdillard]

This should be the last review comment, I don't think I see anything else in the code.

Related: The Purpose section of the PR description is now out of sync with the implementation and says that fix_refuris should remain deprecated.

[jdillard]

Thanks!