fix(timeline): resolve a multi-event letter link deterministically

A letter whose document is in more than one curated event's `documents`
set was linked by `putIfAbsent` over `eventRepository.findAll()`, whose
iteration order JPA does not guarantee — so the same letter could cluster
under a different event across re-seeds/VACUUM with no data change.

resolveLetterEventLinks now runs over a stably-ordered copy (earliest
event date, undated last, then lowest id), so the link is a deterministic
property of the data. Fixes review finding #9 (Architect-3).

Refs #850
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
This commit is contained in:
Marcel
2026-06-15 21:57:07 +02:00
committed by marcel
parent 7ccc4c8896
commit 2151cfad18
3 changed files with 57 additions and 6 deletions

View File

@@ -269,17 +269,28 @@ public class TimelineService {
* event whose {@code documents} set contains the letter (REQ-009). A single doc→event map is
* built over the already-loaded events — no per-letter query ({@code TimelineEvent.documents}
* carries {@code @BatchSize(50)}). When a document is referenced by more than one curated
* event, the first by repository iteration order wins ({@code putIfAbsent}). The map is built
* from <em>all</em> events (not just the year/type-filtered ones) so the link is a stable
* property of the data; the frontend's filter-then-cluster decides whether the linked event is
* actually on screen (#850). Logs nothing — the assembly path keeps UUIDs out of logs (§2.7).
* event, the earliest-dated event wins (then lowest {@code eventId}); the pass runs over a
* stably-ordered copy so the link is a deterministic property of the data, not a coin-flip on
* the undefined repository iteration order ({@code putIfAbsent} keeps the first = winner). The
* map is built from <em>all</em> events (not just the year/type-filtered ones) so the link is a
* stable property of the data; the frontend's filter-then-cluster decides whether the linked
* event is actually on screen (#850). Logs nothing — the assembly path keeps UUIDs out of logs (§2.7).
*/
private Map<UUID, UUID> resolveLetterEventLinks(List<Document> letters, List<TimelineEvent> events) {
Set<UUID> letterDocIds = letters.stream().map(Document::getId).collect(Collectors.toSet());
if (letterDocIds.isEmpty()) return Map.of();
// Stable order so a multi-event letter links deterministically: earliest event date
// (undated last), then lowest id. putIfAbsent below then keeps the winner (REQ-009).
List<TimelineEvent> ordered = events.stream()
.sorted(Comparator
.comparing(TimelineEvent::getEventDate,
Comparator.nullsLast(Comparator.naturalOrder()))
.thenComparing(TimelineEvent::getId))
.toList();
Map<UUID, UUID> eventByDocId = new HashMap<>();
for (TimelineEvent ev : events) {
for (TimelineEvent ev : ordered) {
Set<Document> linkedDocs = ev.getDocuments();
if (linkedDocs == null) continue;
for (Document linked : linkedDocs) {

View File

@@ -549,6 +549,46 @@ class TimelineServiceTest {
assertThat(entry.linkedEventId()).isNull();
}
@Test
void multi_event_letter_links_deterministically_to_the_earliest_event() {
// REQ-009: a document referenced by >1 curated event links to the earliest-dated event
// (then lowest id), independent of repository iteration order — not a coin-flip on
// findAll()'s undefined order.
Document shared = docWithDate(LocalDate.of(1916, 5, 1), DatePrecision.MONTH);
TimelineEvent earlier = TimelineEvent.builder()
.id(UUID.fromString("00000000-0000-0000-0000-000000000001"))
.title("Frühes Ereignis").type(EventType.PERSONAL)
.eventDate(LocalDate.of(1913, 3, 1)).precision(DatePrecision.MONTH)
.documents(new HashSet<>(Set.of(shared)))
.build();
TimelineEvent later = TimelineEvent.builder()
.id(UUID.fromString("00000000-0000-0000-0000-000000000002"))
.title("Spätes Ereignis").type(EventType.PERSONAL)
.eventDate(LocalDate.of(1914, 3, 1)).precision(DatePrecision.MONTH)
.documents(new HashSet<>(Set.of(shared)))
.build();
when(timelineEventService.assembleDerivedEvents()).thenReturn(List.of());
when(documentService.getAllForTimeline()).thenReturn(List.of(shared));
// `later` first in iteration: a naive putIfAbsent would wrongly pick it.
when(eventRepository.findAll()).thenReturn(List.of(later, earlier));
assertThat(theLetter(timelineService.assemble(noFilters())).linkedEventId())
.isEqualTo(earlier.getId());
// Reversed order yields the same winner — the link is order-independent.
when(eventRepository.findAll()).thenReturn(List.of(earlier, later));
assertThat(theLetter(timelineService.assemble(noFilters())).linkedEventId())
.isEqualTo(earlier.getId());
}
private static TimelineEntryDTO theLetter(TimelineDTO result) {
return java.util.stream.Stream.concat(
result.years().stream().flatMap(y -> y.entries().stream()),
result.undated().stream())
.filter(e -> e.kind() == Kind.LETTER)
.findFirst().orElseThrow();
}
private static TimelineEntryDTO onlyLetterEntry(TimelineDTO result) {
assertThat(result.years()).hasSize(1);
return result.years().get(0).entries().get(0);