Problem

Collections let people search and filter documents by their attributes. That works for plain text attributes like Status, but not for attributes that store Hypermedia links, such as Owner.

An Owner value is saved as a link like hm://z6Mkvz9TgGtv9zsGsdrksfNk1ajbFancgHREJEz3Y2HsAVdk. The table shows it as a profile avatar, but search only sees the raw link. So:

    Typing "horac" finds nothing, even though that link points to the profile named "horacio".

    The Owner filter's IS picker lists raw hm://… strings that nobody can read.

    Pasting a web link to the same profile (for example https://hyper.media/hm/<uid>) or a link with ?v=… doesn't match, because the text is different.

As more collections use link attributes (Owner, Assignee, related documents), search gets less useful in exactly the places teams rely on it, like the Issue Tracker.

Solution

Make search and filters understand what a link points to, not just its text.

User stories

    As a teammate, I type part of someone's name ("horac") in a collection's search and see every document they own.

    As a teammate, I paste a link to a profile or document into search and see the documents that reference it, whatever form the link takes.

    As a teammate, I filter Owner IS "horacio" by picking a name from a list, not a raw link.

    As a teammate, I search and find matching documents even when they aren't on the visible page.

How it works

    Search already runs on the server over the whole collection, not only the rows on screen. The server loads every document, keeps the matches, then returns the first page and the total count. That stays the same.

    New step before filtering: the server collects every link stored in the attributes of all documents and looks up the names in one batch, using profile names for accounts and titles for documents. Collections usually reference only a few distinct people, so this is cheap.

    Search and the contains filter match the name and the raw link. Nothing that matched before stops matching.

    If the search text is itself a Hypermedia link (hm://…, hm://…?v=…, hm://…/:profile, or a https://…/hm/… gateway link), the server matches documents whose attributes point to the same profile or document, ignoring the version.

    The IS / IS NOT picker shows and searches by name, but still filters on the link. If someone renames their profile, or two people share a name, the filter still picks the right person.

    If a linked document is missing or private, only that link loses its title. The rest of the request keeps working.

Implementation notes

    The name lookup happens in the shared Query Block API (api-query-block.ts), so the web app, the desktop app and the editor's collection block all get it.

    A new helper (getQueryBlockRefKey) turns any form of a link into one canonical key, so links compare by identity.

    Filter options gain an optional labels field (link → name), so older servers and clients keep working.

    No Go backend, protocol or database changes.

Scope

    Phase 1: Implementation (done locally, not yet merged). Server-side name lookup, search and filter matching, filter picker labels, and the in-browser fallback. Covered by unit tests in the shared and UI packages. Typechecks pass for client, shared, ui, editor, web and desktop.

    Phase 2: Verification (next). Check it against a real collection (the Issue Tracker) in the web and desktop apps, take before/after screenshots of the filter picker, and run local CI.

    Phase 3: Release. Merge and ship with the next release. Depends on phase 2.

Phases 2 and 3 are small: verification and shipping only, with no new design work.

Rabbit Holes

    A database search index. Search already covers the whole collection, so an index only helps speed for very large collections. It would mean moving search into the Go backend and re-indexing every document whenever a profile name changes. Only worth doing if the [QueryBlock perf] timings show slow loading.

    Custom-domain links. A pasted link like https://seedteamtalks.hyper.media/tasks needs a site lookup to know which account it belongs to. Gateway links (/hm/<uid>/…) already work.

    Document titles in the in-browser fallback. The fallback only has profile names, not document titles. It's only used when the server didn't run the search.

    Fuzzy matching, typo tolerance, or ranking. Plain "contains" matching is enough for now.

No Gos

    Don't change how link attributes are stored. They stay as hm:// links.

    Don't make the IS filter store a name instead of the link. Names change and can collide.

    Don't limit search to the visible page.

    Don't let one missing or private linked document break the collection.

    No backend, protocol or database changes in this project.

Do you like what you are reading? Subscribe to receive updates.

Unsubscribe anytime