docs(specs): search-row recording_date + softened estimated count (#61)

This commit is contained in:
2026-06-08 10:03:15 +02:00
parent d082836529
commit 150ca63fc0
@@ -0,0 +1,128 @@
# Search-Result Date Meta + Estimated-Count Copy — Design
**Date:** 2026-06-08
**Status:** Approved (brainstorming) — ready for implementation planning.
**Issue:** #61 (scoped: add a date to result rows + soften the estimated count; object type/facet is
declined — no such field exists in the domain).
## Context
Search result rows show name + object_number + visibility badge + snippet, but no date — power users
disambiguate hits by period. The count copy reads "{{count}} results" though it's Meilisearch's
`estimated_total`. **There is no object type/category/classification field anywhere in the domain**
(`CatalogueObject` = object_number/object_name/number_of_objects/brief_description/current_location/
current_owner/recorder/**recording_date**/visibility + flexible `fields` + timestamps), so a "type"
meta/facet is impossible without a new domain concept (out of scope). The only disambiguating date
available is `recording_date` (`YYYY-MM-DD`) — but it is **not indexed** into the search document or
returned in the hit, so surfacing it needs a backend change.
**Facts:** `SearchDocument` (`crates/search/src/lib.rs:30`) and `SearchHit` (`:52`) carry
`id/object_number/object_name/brief_description/[current_owner/recorder]/visibility/[snippet]` — no
date. `build_document` (`:302`) projects a `CatalogueObject``SearchDocument` (does not copy
`recording_date`). `search_objects` (`:185`) maps Meili results → `SearchHit`. `SearchHitView`
(`crates/api/src/admin_search.rs:26`) mirrors `SearchHit`. The frontend type is
`components["schemas"]["SearchHitView"]`; `schema.d.ts` is generated by `pnpm gen:api`. `AdminObjectView`
already serializes `recording_date` as `Option<String>` (`YYYY-MM-DD`, via `format_date`). On-write
`sync_object` re-projects an object after each catalogue write; `reindex_all` is the full rebuild path.
### Decisions (from brainstorming)
1. Thread `recording_date` (`YYYY-MM-DD`) through `SearchDocument``SearchHit``SearchHitView`;
show it on the row when present.
2. Soften the count to `~{{count}}` (it's an estimate).
3. Decline object type/facet (no domain field).
## Backend (`crates/search`, `crates/api`)
### `SearchDocument` + `build_document` (`crates/search/src/lib.rs`)
- Add field: `pub recording_date: Option<String>,` to `SearchDocument`.
- In `build_document`'s returned struct: `recording_date: object.recording_date.map(|d| d.to_string()),`
(`domain::Date`'s `Display` is ISO `YYYY-MM-DD`, matching `AdminObjectView`). Index it as a plain
string (Meili stores/returns it; it need not be filterable).
### `SearchHit` + `search_objects` mapping (`crates/search/src/lib.rs`)
- Add `pub recording_date: Option<String>,` to `SearchHit`.
- In `search_objects`'s hit map: `recording_date: doc.recording_date,` (the executed
`SearchDocument` deserialization carries it; old index docs missing the field deserialize to `None`
because it's `Option` — graceful).
### `SearchHitView` (`crates/api/src/admin_search.rs`)
- Add `pub recording_date: Option<String>,` to `SearchHitView`.
- In the map closure (`:100`): `recording_date: h.recording_date,`.
### Indexing / backfill
- New & edited objects get `recording_date` automatically via the existing on-write `sync_object`.
- Already-indexed objects return `recording_date: None` until a `reindex_all` (the existing rebuild
path) runs — a graceful, opt-in backfill; **a `reindex` CLI command is a follow-up**, not in scope.
### Tests (Rust — need the docker stack: Postgres :5442, Meilisearch :7700)
- Update any direct `SearchDocument`/`SearchHit` struct literals in `crates/search/tests/*` /
`crates/api/tests/*` to include `recording_date: None` (compile fix). (Object fixtures construct
`CatalogueObject`, which already has `recording_date` — unaffected.)
- Extend `crates/search/tests/search.rs` (the `search_objects_returns_hits_…` test): give one seeded
object a `recording_date` and assert the returned hit's `recording_date` is the `YYYY-MM-DD` string
(proves it flows index → hit).
## Frontend (`web`)
### `web/src/api/schema.d.ts` (generated)
After the backend change, regenerate via `pnpm gen:api` (stack + server up) **or** hand-add
`recording_date?: string | null;` to the `SearchHitView` block (mirroring `brief_description?: string |
null` / `snippet?: string | null`) — a later `gen:api` reproduces it identically. Either is acceptable;
the manual edit avoids running the server purely for codegen.
### `search-result-row.tsx`
Add `recording_date` to the meta line (the `flex items-center gap-2 text-xs text-muted-foreground`
row), after `object_number`, when present:
```tsx
<span>{hit.object_number}</span>
{hit.recording_date && <span>· {hit.recording_date}</span>}
<VisibilityBadge visibility={hit.visibility} />
```
(Render the `YYYY-MM-DD` string as-is — it's a plain recording date, not a UTC timestamp, so no
timezone formatting. A separator dot before it keeps the row scannable.)
### `search-panel.tsx` + i18n — soften the count
The count uses `t("search.resultCount", { count: total })` with `resultCount_one`/`resultCount_other`.
Change the i18n values to flag the estimate (en + sv, parity preserved):
- en: `"resultCount_one": "~{{count}} result"`, `"resultCount_other": "~{{count}} results"`
- sv: `"resultCount_one": "~{{count}} träff"`, `"resultCount_other": "~{{count}} träffar"`
No code change in `search-panel.tsx` (the key already interpolates `count`).
### Fixtures + tests
- `web/src/test/fixtures.ts` `searchHits`: add `recording_date` to the first hit (e.g. `"1962-04-03"`);
the rest can stay `recording_date: null` (or omit — it's optional).
- `web/src/search/search.test.tsx`: assert the first result row shows the date (`getByText("1962-04-03")`
or within the row); assert the count copy includes the `~` (`getByText(/~\s*25 results/i)` — note the
existing `/25 results/i` assertion still matches the new copy; tighten it to require the `~`).
## Data flow
DB object → `build_document` (now copies `recording_date`) → Meili doc → `search_objects` hit → API
`SearchHitView``useSearch``SearchResultRow` renders the date. The count is the same
`estimated_total`, now labelled `~`.
## Error handling / edges
- `recording_date` is `Option` end-to-end → absent on objects without a recording date and on
not-yet-reindexed docs; the row simply omits it.
- `Date::to_string()` is ISO `YYYY-MM-DD` (matches `AdminObjectView`); no locale/timezone formatting.
- The `~` is cosmetic; pluralization (`_one`/`_other`) is unchanged.
## Testing
- **Rust:** `cargo build --workspace`; `cargo nextest run -p search -p api` (stack up) green, incl. the
new recording_date assertion + any literal compile-fixes.
- **Frontend:** `typecheck`/`lint`/`test`/`build`/`check:size`/`check:colors` green; the new row-date +
count tests pass; en/sv parity (the #60 parity test guards the changed values); no codename.
- No new dependency.
## Acceptance criteria
1. `recording_date` is projected into the search index (`SearchDocument`/`build_document`) and returned
through `SearchHit``SearchHitView``schema.d.ts` (`SearchHitView.recording_date?: string|null`).
2. The search result row shows the recording date (when present) on its meta line.
3. The result-count copy reads `~{{count}} …` to flag the Meilisearch estimate (en + sv).
4. Rust (`cargo build` + search/api tests) green; frontend gate green; en/sv parity; no codename; no new
dependency.
## Out of scope → follow-ups
- An object **type/classification** domain concept + a **type facet** (no such field exists today).
- A `reindex` CLI command to backfill `recording_date` onto already-indexed objects (new/edited objects
index it automatically; `reindex_all` is the existing rebuild path).
- Richer faceting; making `recording_date` filterable/sortable in search.