"Server side search" gets an not-useful wall of text

Details

• Read the Docs project URL: https://readthedocs.org/projects/test-ddev/

• Build URL (if applicable): https://readthedocs.org/projects/test-ddev/builds/17382495/

• Read the Docs username: https://readthedocs.org/profiles/ddev/

• Source: https://github.com/rfay/ddev/tree/20220707_docs2

Expected Result

Simple and appropriate search results. This is what I see with mkdocs serve:

Actual Result

On RTD at https://readthedocs.org/projects/test-ddev/ I get a huge wall of text that’s completely not useful:

I assume this is a result of Server side search because I get very decent results with mkdocs serve.

Is there a way to disable server-side search or fix this somehow?

[squidfunk]

Author of Material for MkDocs here. This happens when you feed a search_index.json that was built with Insiders into the non-Insiders version of Material for MkDocs. Insiders features a completely rewritten search plugin. You only have to make sure that you’re not mixing inside the same environment (be it through caching or other means).

I suspect that search_index.json is served with cache headers that don’t force the browser to re-validate. Now, when you deploy your documentation with non-Insiders afterwards, your browser will use the old search_index.json, but the application bundle (which has content hashes) comes from the community edition, thus the mismatch.

[squidfunk]

@stsewd adding must-revalidate when serving search_index.json solves the problem. Also, it would mean that users will receive the new search index immediately after a deployment, and not after a day. It’s current set to max-age=86400.

Unless there’s a reason I don’t see why data should be cached for so long. GitHub Pages uses max-age=600.