Dark mode

[bartekpacia]

Did you consider creating the dark version of the website?

[janbrasna]

Not sure how much of the underlying RTD template is still actively used here, but there's a long running upstream issue: readthedocs/sphinx_rtd_theme#224 — with a couple smart hacks as readthedocs/sphinx_rtd_theme#224 (comment) and/or reference to a separate implementation as https://github.com/MrDogeBro/sphinx_rtd_dark_mode or alternative https://github.com/jucacrispim/sphinx_pdj_theme that might serve as inspiration how involved this might be.

[janbrasna]

(There used to be an attempt to move from RTD template to Material template #696 — and that one indeed does have a dark theme: https://squidfunk.github.io/mkdocs-material/setup/changing-the-colors/ … however that move was cut short at some point as the template work for mkdocs upgrade was based on the current original that was kept.)

[janbrasna]

The actual design is quite similar to what Furo theme https://github.com/pradyunsg/furo has out-of-the-box, and that one DOES HAVE the dark theme functionality. Maybe that could be the upgrade path? EDIT: sorry that's sphinx not mkdocs

[Brett-Best]

I've opened draft PR #1343 as an explicitly AI-driven experiment to explore dark mode as an accessibility improvement.

It keeps the customized Read the Docs theme rather than reviving the Material migration from #696. MkDocs' native color-mode controls only support its built-in mkdocs theme, while this site relies heavily on RTD-specific templates, navigation, Algolia search, and several thousand lines of custom CSS. The draft therefore needs a small persisted Auto / Light / Dark controller plus targeted styling for code, tables, admonitions, search, plugins, mobile navigation, branding, and print.

The experiment includes desktop/mobile/search/print browser checks and an APCA comparison against the current light theme. The comparison found contrast parity and no identified dark-mode regression, but this is not a claim of WCAG or APCA compliance. The PR documents the remaining contrast clusters and implementation tradeoffs, and I would especially welcome human accessibility and design review before it is considered ready.