Offline writes now queue + apply optimistically instead of throwing.
On reconnect, the queue drains automatically; conflicts and rejections
surface as one-time SnackBars so the user knows their edit didn't land.
- Drift schema v3: `pending_writes` table (verb + payload + baseline +
tries + last_error). Negative ids serve as cache placeholders for
offline-created rows until replay assigns the server id.
- Each write repository (notes, tasks, projects, milestones, events)
now catches NetworkException, applies the change to cache (with a
fresh `nextTempId()` for creates), and enqueues the API call.
- Edits to a still-queued offline-created row coalesce into the
original create payload — only one server call per row, in order.
- Deletes of still-queued offline-created rows drop the queue entry
and the cache row; no server call ever happens.
- New WriteQueue service drains the queue oldest-first.
Server-wins on `updated_at` baseline check (notes/tasks/projects);
last-writer-wins for milestones/events (no getOne available).
4xx → drop + surface as `rejected`; conflict → drop + `overwritten`;
404 on update → drop + `missing`; 5xx/network → keep + retry next
online cycle.
- Replay fires on AuthStatus → authenticated transitions
(cold-start, came-back-online, login).
- OfflineBanner shows "Retry (N)" with the pending-writes count.
- Distinct ServerException added so 4xx no longer masquerade as
NetworkException — the queue can drop them instead of looping.
flutter analyze clean; 21 tests pass (3 new for QueueFailure messaging).
Phase 4 (read-only UI indicators on cached/temp rows) still ahead.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
The global receiveTimeout is 30s but quick capture runs LLM inference
that can take much longer. receiveTimeout fired, fell through to the
NetworkException fallback in dioToApp, and the work queue queued it as
an offline item.
- quick_capture_api.dart: override receiveTimeout to 120s for the
/api/quick-capture request
- api_client.dart: handle receiveTimeout/sendTimeout explicitly in the
error interceptor, converting them to AppException (not NetworkException)
so slow responses are never mistaken for being offline
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
bvandeusen