How to show each page only once on search results?

[kaycebasques]

If you search for pw_unit_test on https://pigweed.dev/search.html you get results like this:

This happens because we have structured our changelog like this:

• 20240226
  • ...
  • pw_unit_test
  • ...
• 20240212
  • ...
  • pw_unit_test
  • ...

20240226 is an H2 containing the updates published on 2024 Feb 26. The pw_unit_test nested as a child is an H3 containing the updates for that module. Etc.

Is there a way to configure the local search engine to only show each page once?

[picnixz]

AFAIK, no. But there are a bunch of PRs trying to fix the search engine so maybe they can give you better results.

[jayaddison]

I think that this problem is a symptom of a (recently-resolved) bug #11943 : in Sphinx versions up-to-and-including 7.2.6, identical contextual snippet appeared despite search results linking to multiple matches within a document.

[jayaddison]

See also bugreport #11961 about search result duplicates - with two possible pull requests currently available (#11942, #12047) that perform corresponding de-duplication. Feedback on either of those would be much appreciated.

[jayaddison]

I think that this problem is a symptom of a (recently-resolved) bug #11943 : in Sphinx versions up-to-and-including 7.2.6, identical contextual snippet appeared despite search results linking to multiple matches within a document.

After further investigation: resolving #11943 does not de-duplicate the results mentioned in this thread, because each of them is a subtitle/subheading in the document, and we currently (up-to-and-including v7.4.6) include each distinct subtitle/subheading result (both pull requests mentioned addressed de-duplication of main-document-title results only).

I think there are two problems here:

• The results aren't useful when a user performs the search query.
• Each affected result creates a separate HTTP GET request to retrieve the same page for the purposes of displaying a summary (wasteful).

[jayaddison]

Although it might seem that we should de-duplicate the HTTP GET (or, strictly speaking: browser fetch operations) requests without making any changes to the eventual display of results, I think there is a user experience consideration to bear in mind:

We add each result one-by-one to the HTML search result display output. If we were to begin de-duplicating fetch requests without making any other changes, then the first request to retrieve the changelog page here would be slow -- but then all of the following subsection headings would rapidly appear and fill the user's display, and then slow fetch operations for other results would commence.

That might not be a great search experience, because each result could appear with a wildly differing delay, and whether a result was fast/slow to retrieve would seem erratic.

Perhaps we could group the results by page, or by the type of match -- and perhaps allow the content in the grouping to update (or not!) as subsequent results are received.