mongofire 6.5.1 → 6.5.3

package/CHANGELOG.md

@@ -1,3 +1,70 @@
+## [6.6.0] - 2025-03-10
+
+### 🐛 Critical Bug Fixes
+
+#### BUG-1 — Delta cursor `$or` overwrite (data loss on every sync after first)
+**File:** `src/sync.js` — `_downloadCollection()` delta section  
+When `cursor.id` was set (every delta sync after the first), spreading `tsFilter`
+and `applyCondition` into the same object silently discarded the timestamp `$or`
+because JavaScript objects cannot have two `$or` keys. MongoDB received no
+timestamp constraint and re-fetched the entire changetrack collection on every
+delta sync. Fixed by using `$and: [tsCondition, applyCondition]` when both
+conditions contain a top-level `$or`.
+
+#### BUG-2 — Per-instance connection never wired (CRIT-3 was incomplete)
+**Files:** `src/index.cjs`, `src/sync.js`, `src/changetrack.js`, `src/state.js`, `src/reconcile.js`  
+`MongoFire` constructor created a new `Connection` instance (`this._conn`) but
+`sync.js`, `changetrack.js`, `state.js`, and `reconcile.js` all used the
+module-level singleton. `this._conn.local` was never `null.collection(...)` —
+it threw `TypeError` on every operation. Added `setConnection(conn)` to all four
+modules; `MongoFire._doStart()` now calls each after `connectLocal()`.
+
+#### BUG-3 — Bulk path non-idempotent on replica-set transaction retries
+**File:** `src/changetrack.js` — `_doRecordChangeBulk()`  
+`_doRecordChangeBulk()` used `insertOne` in its `bulkWrite` array. When a
+replica-set transaction was retried by `withTransaction()`, the same `opId` was
+re-inserted, hitting the unique sparse index and throwing `E11000 duplicate key`.
+Fixed by switching to `updateOne / $setOnInsert` (same pattern as the single
+`recordChange` path) — idempotent by design.
+
+#### BUG-4 — Bootstrap resume cursor silently lost on crash-resume
+**Files:** `src/sync.js`, `src/state.js`  
+Bootstrap resume was saved as `setCursor(..., new Date(0))` plus a separate raw
+`updateOne` writing a plain ObjectId string. `_normalizeCursor()` tried to parse
+the ObjectId string as a `Date`, got `Invalid Date`, and returned `null` —
+so resume always restarted from scratch. Fixed: `setCursor()` now accepts
+`{ ts: null, id: ObjectId-string }`; `_normalizeCursor()` recognises 24-char hex
+strings as ObjectId cursors and returns `{ ts: null, id }` correctly.
+
+### 🐛 Medium Bug Fixes
+
+#### BUG-5 — `conflictResolved` event declared but never emitted
+**File:** `src/index.cjs`  
+TypeScript types declared `conflictResolved` in `MongoFireEvents` but
+`retryConflict()` and `dismissConflict()` never called `this.emit(...)`. Any
+listener registered with `mongofire.on('conflictResolved', ...)` never fired.
+Both methods now emit `conflictResolved` with `{ opId, resolution }`.
+
+#### BUG-6 — Module-level `_deviceId` shared across multiple instances
+**File:** `src/device.js`  
+The module-level `_deviceId` singleton meant multiple `MongoFire` instances
+in the same process shared the same deviceId. The download de-dup filter
+(`op.deviceId === deviceId`) then skipped ops from sibling instances.
+Introduced `DeviceManager` class; each `MongoFire` instance creates its own
+via `newDeviceManager()`. Module-level singleton retained for backward compat.
+
+### ✨ New Features
+
+#### Interactive CLI (Inquirer.js)
+All `mongofire` CLI commands now run in interactive prompt mode when no flags
+are passed. Uses `inquirer` when available, falls back to `readline` so the
+CLI always works without extra deps.
+
+- `npx mongofire init` — Setup wizard: module system, collections, realtime, sync interval
+- `npx mongofire clean` — Interactive days selector with confirmation prompt
+- `npx mongofire conflicts` *(new)* — View, retry, or dismiss unresolved conflicts interactively
+- `npx mongofire reconcile` *(new)* — Run recovery scan and view per-collection results
+
 # Changelog
 
 All notable changes to MongoFire are documented here.