@DataJpaTest + Testcontainers exercises the V56 migration plus the
@ElementCollection wiring end-to-end. Saves a block with two PersonMention
entries, clears the persistence context, reloads, asserts both entries
return with their personId + displayName intact. Second test guards the
@Builder.Default — a block without explicit mentions reloads with an empty
list, not null.
Refs #362
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
@ElementCollection(LAZY) on List<PersonMention>, mapped to V56's
transcription_block_mentioned_persons via explicit @CollectionTable that
matches the migration name byte-for-byte (immune to Hibernate naming-strategy
changes). @Builder.Default keeps the field initialized to an empty list, so
existing transcription block construction stays untouched.
Refs #362
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
Carries personId + oldDisplayName + newDisplayName so transcription-side
listeners can rewrite block.text and sidecar entries when a person is
renamed. First custom application event in this codebase — the only prior
@EventListener consumes Spring's built-in ApplicationReadyEvent. Class doc
sets the convention for future cross-domain decoupling.
Refs #362
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
Value object held in TranscriptionBlock.mentionedPersons via @ElementCollection.
Carries the personId UUID (so renamed persons can be located) and the
displayName text (so block.text rewrites match exactly via "@" + name). Both
fields are non-null; displayName capped at 200 chars to match the V56 column
and bound the rename propagation cost.
Refs #362
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
Child table for @-mentions inside transcription block text. Each row binds
one block to one person via personId + displayName; the literal "@DisplayName"
stays in block.text. No FK on person_id so deleted persons degrade gracefully
to plain unlinked text rather than cascade-deleting the block. Indexed on
person_id for the future "blocks mentioning person X" query and on block_id
for the @ElementCollection load.
Schema choice diverges from document_comments.comment_mentions (many-to-many
to AppUser): the latter cascades, this one degrades. Mirrors the established
UserGroup.permissions / group_permissions @ElementCollection pattern.
Refs #362
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
Spring deserializes the enum directly; invalid values are caught by the
HttpMessageNotReadableException → 400 handler added in 99d00537, returning
a structured VALIDATION_ERROR. The manual parseType() helper is therefore
redundant and removed. Tests updated to construct requests with the enum.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
Proves the in-memory filter correctly drops edges where one Person is
not in findAllFamilyMembers(), preventing non-family relationships from
leaking into the graph.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
8 hops covers great-grandparents ↔ great-great-grandchildren and second
cousins — the practical horizon for a 1899–1950 archive. Prevents future
blind tuning of the constant.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Addresses @markus/@nora suggestion: makes explicit that the missing
@RequirePermission on read endpoints is intentional — all authenticated
family members may read the family graph; unauthenticated access is still
blocked by Spring Security's anyRequest().authenticated() rule.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Addresses @sara blocker: RelationshipControllerTest now has 6 tests covering
the two previously untested @RequirePermission(WRITE_ALL) endpoints. Prevents
silent permission regression if the controller is refactored.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Addresses @sara blocker: documents that Spring Security's anyRequest().authenticated()
guards these read endpoints and provides regression protection against accidental
@PermitAll additions in future.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
getRelationshipBetween now throws DomainException with RELATIONSHIP_NOT_FOUND
instead of ResponseStatusException, so the frontend receives a typed error code.
Removed redundant validateRelationType() guard — RelationshipService.parseType()
already handles this with the same DomainException/VALIDATION_ERROR path.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Removes direct PersonRepository injection from the relationship domain,
routing cross-domain person resolution through PersonService.getAllById()
per the layering rules.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Both /api/network and /api/persons/{id}/relationships threw
LazyInitializationException when toDTO read Person.getDisplayName():
the read-side service methods aren't @Transactional, so the session
closed before the proxy could initialize.
Eagerly fetch r.person and r.relatedPerson in the two queries used
by these endpoints, keeping the no-@Transactional convention for
read methods.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
Seven endpoints in one controller, two roots:
- GET /api/network → NetworkDTO
- GET /api/persons/{id}/relationships → List<RelationshipDTO>
- GET /api/persons/{id}/inferred-relationships
- GET /api/persons/{aId}/relationship-to/{bId} → 200 or 404
- POST /api/persons/{id}/relationships WRITE_ALL
- DEL /api/persons/{id}/relationships/{relId} WRITE_ALL, 204
- PATCH /api/persons/{id}/family-member WRITE_ALL
PersonController is intentionally untouched. Controller-boundary
validation via RelationType.valueOf catches unknown types as 400 before
the service is invoked. FamilyMemberPatchDTO is a one-field record for
the family-member toggle.
Refs #358.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
- Add PersonService.setFamilyMember (write, @Transactional) and
findAllFamilyMembers; PersonRepository gains the
findByFamilyMemberTrueOrderBy projection.
- RelationshipService orchestrates PersonService + the inference
service; never reaches into PersonRepository directly. addRelationship
guards self-relationship, year range, circular PARENT_OF (Nora B2),
and DataIntegrityViolation→DUPLICATE_RELATIONSHIP. deleteRelationship
enforces ownership from either side (Nora B1).
- Extend RelationshipDTO with personDisplayName + birth/death year so
the frontend can render rows from either viewpoint.
- 8 unit tests, written against a stub (red), then green: FORBIDDEN
delete, CIRCULAR add, DUPLICATE add, self-relationship, year range,
happy-path persistence, ownership-from-object, RELATIONSHIP_NOT_FOUND.
Full backend suite: 1399/1399 green.
Refs #358.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
RelationToken enum (UP/DOWN/SPOUSE/SIBLING) with reverse(), and
RelationshipInferenceService with:
- Bidirectional adjacency map: PARENT_OF emits UP and DOWN, SPOUSE_OF
and SIBLING_OF both directions.
- Virtual SIBLING edges derived from shared parents — no SIBLING_OF
row required for siblings to appear.
- BFS with MAX_DEPTH=8.
- 17-entry LABEL_MAP covering parent, child, spouse, sibling, grand*,
great-grand*, uncle/aunt, niece/nephew, great-uncle/aunt, great-niece/
nephew, in-law parent/child, sibling-in-law (both paths), cousin_1.
- "distant" fallback for any path not in LABEL_MAP.
- Two-sided labels via path reversal.
18 unit tests written first against a stub; all 18 confirmed red, then
green after implementation. PersonControllerTest's anonymous DTO updated
for the new isFamilyMember() projection.
Refs #358.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
- RelationType enum (9 values), PersonRelationship entity with
@ToString(exclude = "notes") and LAZY person FKs.
- PersonRelationshipRepository with the network bulk fetch, the
per-person subgraph fetch, and the existsBy check for the circular
PARENT_OF guard.
- Six DTO records: CreateRelationshipRequest, RelationshipDTO,
PersonNodeDTO, NetworkDTO, InferredRelationshipDTO,
InferredRelationshipWithPersonDTO. @Schema(REQUIRED) on every
always-populated field so OpenAPI/TS codegen stays accurate.
- Person entity gains familyMember, PersonSummaryDTO gains
isFamilyMember, both PersonRepository projections select
p.family_member.
- Three new ErrorCodes: RELATIONSHIP_NOT_FOUND, CIRCULAR_RELATIONSHIP,
DUPLICATE_RELATIONSHIP.
Refs #358.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Adds persons.family_member flag and person_relationships table with
ON DELETE CASCADE on both FKs, no_self_rel check, unique_rel composite,
indexes on both person columns, and partial unique index for symmetric
SIBLING_OF pairs (LEAST/GREATEST trick).
Refs #358.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Confirms that DELETE /api/documents/{id}/annotations/{id} requires at
least ANNOTATE_ALL; a user with only READ_ALL receives 403 Forbidden.
Closes the permission audit raised during PR review.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Adds a trash icon button (44×44 px touch target) directly on each annotation shape in transcription mode so users can delete a block without navigating through the sidebar. Includes keyboard support (Delete key), confirm dialog via ConfirmService, prop-chain wiring through DocumentViewer → PdfViewer → AnnotationLayer → AnnotationShape, and orphaned-annotation fallback (calls DELETE /annotations/{id} when no block is linked). Backend security regression test added for deleteBlock 403 on READ_ALL.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Regression guards verifying that Spring Security returns 401 (not 200) when
no credentials are provided, complementing the existing 403 permission tests.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Removes the wait+clear cycles that existed only to drain the audit events
emitted by createUserOrUpdate(null, ...). Timeouts increased 5 → 10 s to
reduce CI flakiness under load.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
createUserOrUpdate(UUID actorId, ...) is always called from the controller with
a real authenticated actor. createUserForBootstrap() handles seeding/test setup
without emitting an audit event, making the two contracts unambiguous.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Creates a real actor user first (needed for audit_log FK constraint),
then creates and deletes a target user, asserts USER_DELETED is newest
and USER_CREATED is second via findRecentUserManagementEvents.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Adds findRecentByKinds JPQL query to AuditLogQueryRepository and
findRecentUserManagementEvents(int limit) to AuditLogQueryService,
returning the N most recent USER_CREATED/USER_DELETED/GROUP_MEMBERSHIP_CHANGED
events ordered newest-first.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Adds actorId param to adminUpdateUser(), captures beforeGroups before
mutation, computes added/removed group names, emits logAfterCommit only
when the group set actually changes. Payload contains group names, not
permission strings.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Adds actorId param to deleteUser(), captures email before deletion,
emits logAfterCommit(USER_DELETED) with userId+email in payload.
Updates UserController to resolve and pass actorId.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Adds USER_CREATED, USER_DELETED, GROUP_MEMBERSHIP_CHANGED to AuditKind.
Injects AuditService into UserService; changes createUserOrUpdate to
accept actorId and emits logAfterCommit(USER_CREATED) only on the
new-user branch. Updates UserController to resolve and pass actorId.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Adds jsonPath("$.code").value("INVALID_PERSON_TYPE") to verify the full
error response shape, not just the HTTP status.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
- PersonController trims title (both create + update) matching the existing firstName/lastName trim pattern
- PersonControllerTest: verifies title is trimmed before service call (ArgumentCaptor)
- PersonControllerTest: verifies createPerson returns 400 when personType is SKIP
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Words like "Wille" stem to "will" via the German Snowball stemmer, which is
also a German stop word. The prefix-transform step (websearch_to_tsquery text →
regexp_replace → to_tsquery) was passing already-stemmed lexemes back through
the German dictionary, causing them to be silently dropped as stop words. Using
the 'simple' configuration skips stop-word processing entirely while the
tsvector @@ tsquery comparison still works because lexemes are matched by
string value, not by configuration.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
