feat(default-layout): sortable search results, search phrases, experimental config

[robinpyon]

trafficstars

Description

Features

• Order search results by relevance, created and last updated date.
• Search for documents containing an exact phrase by wrapping text inside double quotes.
• Add __experimental_omnisearch_visibility: true to your document schemas to hide them from omnisearch. Perfect for hiding workflow-related documents or configuration singletons that don’t need to be accessed frequently.
• View relevance scores for search results via a debug flag (#_debug_search_score)

Bug fixes

• Fixes an issue where the scroll position of search results wasn’t being correctly retained in some instances.
• Fixes an issue where search would cause the studio to crash if Local Storage is unavailable.

What to review

Notable changes

1. Taking cues from the reference and cross dataset reference input components – the function to generate weighted search is now colocated within default-layout rather than being an export of @sanity/base/search

   • Similar to reference / cross reference input components, this now imports from @sanity/base/_internal
   • ~~Given that part:@sanity/base/search and part:@sanity/base/search/weighted are not being used anywhere else in the studio, they’ve been removed~~

2. Logic for how we extract search queries and calculate scores has been expanded to accommodate searching for phrases in wrapped quotes:

   • Fundamentally, this is handled in a very similar way to how search worked prior. My understanding of search, and how phrases come into play can be found here (internal doc)
   • The important thing to note is that how we handle non-quoted text searches hasn’t changed. If you don’t engage with this feature then you won’t notice a difference here.

3. A few more comments are now attached with each search query for measurement. Full example:

   // findability-mvi:2
   // findability-recent-search:4
   // findability-selected-types:2
   // findability-sort:relevance
   

   In the above example, findability-recent-search:4 indicates that this was the 5th item clicked in recent searches. All these comments will be removed once client-side instrumentation is in place.

4. In the test studio: a debug schema experimentalOmnisearchVisibilityTest has been added. These documents have __experimental_omnisearch_visibility:false and won’t appear in omnisearch.

5. Sorting by relevance will now sort by _updatedAt desc in GROQ before applying scoring on the client, leaning towards more recent documents in general. Previously, omnisearch was sorting on _id asc.

   • Search within reference + cross dataset reference fields are unchanged, and will continue to sort on _id asc.

6. Notable refactoring:

   • Simplified createWeightedSearch signature slightly, and have moved more into the SearchOptions type
   • Some unused dev dependencies were removed from default-layout (mocha, nyc) and it now uses react-testing-library .
   • default-layout was also added to the root level jest.config.js
   • Search related tests in @sanity/base are now colocated and have been expanded on slightly.
   • Quite a bit of code was touched to ensure this is valid once we can enable strictNullChecks: true.

Things to test

1. Change the sorting of search results:
   • Does this work and make sense to you? Bearing in mind scoring isn’t taken into consideration when not sorting on relevance.
2. Search on phrases with quoted text:
   • View relevance scoring in debug mode – do these scores seem reasonable to you?
   • Try search for strings with special characters – some good candidates to compare and contrast (test studio):
     1. "winnie-the-pooh" vs winnie-the-pooh
     2. "go, team. go!" vs go, team. go!

Notes for release

TBD

[vercel[bot]]

The latest updates on your projects. Learn more about Vercel for Git ↗︎

Name	Status	Preview	Updated
studio-workshop	✅ Ready (Inspect)	Visit Preview	Sep 20, 2022 at 0:54AM (UTC)
test-studio	✅ Ready (Inspect)	Visit Preview	Sep 20, 2022 at 0:54AM (UTC)

[robinpyon]

Thanks @snorrees!

Re: scores: I agree! Especially with how search will use preview fields (which can be automatically inferred if you don't have a preview configuration) .... mix in the interplay between __experimental_search and it can all seem a bit much. It begs the question of whether this needs to be surfaced in documentation more.