Phase 4 of the collaboration plan:
- The sync engine listens for DOC_SEQ in the note's realtime room and syncs
right away. Without WebSocket/SSE it long-polls /docs/{id}/changes (25 s);
after three failures it falls back to polling every 3 s and retries long
polling when the browser comes back online.
- With push or long polling the engine only syncs every 7 s while holding
leases (renewal), otherwise every 30 s as a safety net. Focusing a block
announces its lease immediately (otherwise it could take up to 30 s).
- Realtime connection: SSE counts as connected only after its first event;
a stream that stays silent for 5 s (buffering proxy) switches SSE off for
the tab. App-level PING/PONG detects WebSockets that are accepted but
forward nothing and switches to SSE after 35 s.
- The editor shows the active channel for shared notes.
- Tests: engine channel tests (push, heartbeat, lease on focus, long
polling, interval fallback, switching), realtime fallback tests with fake
WebSocket/EventSource, Go tests for PONG and DOC_SEQ delivery, and
e2e/channels.e2e.mjs measuring latency behind six simulated proxy setups
(0.2-0.3 s on all live channels, ~3 s with plain polling).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Phase 3 of the collaboration plan:
- Blocks held by someone else show a lock with the name and are read-only;
Enter in such a block adds a new block below. Delete, merge, indent and
drag are not started when a involved block (or one in its branch) is
locked; a short notice explains why (fixes FLAW-25).
- Presence: the editor header lists who else has the note open.
- Conflict copies are highlighted ("Konflikt – Version von …") with
Übernehmen (replaces the original or the title) and Verwerfen. Both check
the locks of the original and of the copy first.
- Undo/redo of own changes (Ctrl/Cmd+Z, Ctrl+Shift+Z, Ctrl+Y, beforeinput
history events): typing is grouped per block until a 1 s pause, up to 200
steps per tab; a step is skipped with a notice if someone else changed or
holds the block meanwhile. New clearConflict props patch on the server.
- New blocks get focus synchronously on mount (fixes FLAW-26).
- Fix: after reopening a note from local storage the engine polled every
30 s instead of 3 s until the next change, because the delay was decided
before the first response told it the note is shared.
- Tests: undo/conflict engine tests, lock/undo/conflict UI tests and
e2e/lock.e2e.mjs (two browsers); all three browser tests pass against a
freshly built image.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The share dialog for notes and shared folders now offers "Bearbeiten" and
"Nur Lesen" and stores write/read accordingly. Editors already edit through
the sync engine since phase 2.
Adds e2e/share.e2e.mjs: shares a note as editor through the dialog and checks
that the recipient can edit it; e2e/run.sh runs both browser tests.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Phase 2 of the collaboration plan:
- frontend/src/lib/docSync: TypeScript port of the block tree, operations,
effects and order keys (kept identical to Go via shared test vectors),
diff from the editor's flat list to operations, IndexedDB persistence,
HTTP transport and DocSyncEngine (one request in flight, idempotent
resends, backoff, 3 s/30 s polling, keepalive flush and leave on exit,
block leases for focus, 20 s grace and new blocks).
- Text changes carry the version the user saw when typing (first change
wins when coalescing; a stale focused field keeps its old base), so a
concurrent edit becomes a conflict copy instead of a silent overwrite.
- DocumentEditor uses the engine instead of the 800 ms whole-document PUT;
editors (not only owners) can edit, remote changes appear live.
- Server: task IDs of linked blocks are delivered again, linked tasks follow
ticked/deleted blocks after sync, note tags are derived, title changes
notify other users; block IDs already used in another note are renamed
(starter content with fixed IDs was empty for every further user).
- Fix: unlinking tasks from blocks wrote to non-existent columns and never
worked (error was swallowed).
- Tests: Go/TS order key vectors, diff round trips, engine races (lost
responses, offline + reload, edits arriving between typing and sending,
stale focused field, duplicate first block), plus a Playwright test with
two browsers against the real image (e2e/run.sh).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Phase 1 of the collaboration plan (server side):
- internal/collab: block tree with parent IDs and fractional order keys,
operations with conflict rules (version check, block leases, conflict
copies instead of lost text, restore on late edits, cycle checks),
in-memory leases/sessions and long-poll notifier.
- Storage: per-note lock + transaction, append-only effect log (doc_ops),
idempotent results per opId (doc_op_results), snapshot, log paging,
central DocRole. Old flat block data is converted to the tree.
- The existing whole-document save (PUT/PATCH) runs as an adapter through
the same engine, so the current editor keeps working and writes to the log.
- API: GET /docs/{id}/snapshot, POST /docs/{id}/sync, GET /docs/{id}/changes
(long poll), POST /docs/{id}/leave; DOC_SEQ poke for WebSocket/SSE rooms.
- Backend accepts editor (write) shares for notes and folders; the forced
read-only migration is gone. The share dialog still offers read only.
- Tests: conflict matrix, order keys, leases, storage adapter round trip,
idempotency, API protocol incl. long poll, and a randomized simulation
(4 clients, lost/duplicated requests and responses, offline phases,
lease expiry) checking convergence, no lost text and exactly-once results.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Phase 0 of the collaboration plan:
- Remove BroadcastToAll; task, task list and workspace folder events now
only reach users who can see the resource. Task list events carry only
the list ID, clients reload with their own permissions.
- Reject unauthenticated WebSocket/SSE connections (401). Every room join
(URL room, JOIN_ROOM, SSE rooms) is checked via a room authorizer;
/events/rooms requires login and only accepts the caller's own sseId.
- Reorder rejects tasks the caller cannot access or edit (IDOR).
- Lift the server WriteTimeout for SSE streams so they are not cut after 30s.
- Fix folder share removal sending DOC_DELETED to the share ID.
- Tests for all of the above; docs and flaw list updated.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Server-authoritative operation log with per-block versions and 20s
block leases, conflict copies instead of data loss, IndexedDB queue
for offline use, HTTP-based sync with WebSocket/SSE/long-poll wakeups,
unified note/folder permissions, and per-user undo.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- Add vitest, jsdom and Testing Library as dev dependencies,
replace the placeholder test:run script with vitest run.
- Add outliner tests for Backspace merging and caret placement on click.
- Run the frontend tests in the Dockerfile before the Vite build.
- Update docs (README, architecture, FLAW-19).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Remove outdated claims (CRDT, DocRooms, presence, .md files on disk) and
document the actual architecture, API, permissions and known flaws.
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Task list shares now carry a permission (editor/viewer). The server rejects
writes from viewers with 403, and the UI hides or disables edit controls
for read-only lists.
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
sandrodachandClaude Opus 5.5