@tldraw/sync-core 5.5.0-canary.d64d005dcf06 → 5.5.0-canary.d7142fa77bb5

package/src/lib/ClientWebSocketAdapter.test.ts

@@ -1,4 +1,4 @@
-import { warnOnce } from '@tldraw/utils'
+import { promiseWithResolve, warnOnce } from '@tldraw/utils'
 import { TLRecord, sleep } from 'tldraw'
 import { afterEach, beforeEach, describe, expect, it, vi } from 'vitest'
 
@@ -527,6 +527,46 @@ describe('ClientWebSocketAdapter', () => {
 			adapter.close()
 			expect(() => adapter.close()).not.toThrow()
 		})
+
+		it('[CW9][RM5] closing immediately does not start another getUri call', async () => {
+			const getUri = vi.fn(() => 'ws://localhost:2233')
+			const testAdapter = new ClientWebSocketAdapter(getUri)
+			testAdapter.close()
+			const callsAtClose = getUri.mock.calls.length
+
+			await vi.advanceTimersByTimeAsync(INACTIVE_MAX_DELAY)
+
+			expect(getUri).toHaveBeenCalledTimes(callsAtClose)
+			expect(testAdapter._ws).toBeNull()
+		})
+
+		it('[CW9][RM5] close reports offline without notifying listeners or reconnecting', async () => {
+			let uriCallCount = 0
+			const testAdapter = new ClientWebSocketAdapter(() => {
+				uriCallCount++
+				return 'ws://localhost:2233'
+			})
+			const onStatusChange = vi.fn()
+			testAdapter.onStatusChange(onStatusChange)
+			await waitFor(() => testAdapter._ws?.readyState === WebSocket.OPEN)
+			const callsBeforeClose = uriCallCount
+			onStatusChange.mockClear()
+
+			testAdapter.close()
+			expect(testAdapter.connectionStatus).toBe('offline')
+			// let the socket's own close event land, then run out any (leaked) reconnect timer
+			vi.useRealTimers()
+			await sleep(50)
+			vi.useFakeTimers()
+			vi.advanceTimersByTime(INACTIVE_MAX_DELAY)
+			vi.useRealTimers()
+			await sleep(20)
+			vi.useFakeTimers()
+
+			expect(onStatusChange).not.toHaveBeenCalled()
+			expect(uriCallCount).toBe(callsBeforeClose)
+			expect(testAdapter._ws).toBeNull()
+		})
 	})
 
 	describe('orphaned sockets (CW10)', () => {
@@ -710,7 +750,7 @@ describe('ReconnectManager', () => {
 		// it's necessary to close the socket, as otherwise the websocket might stay half-open
 		connectedServerSocket.close()
 		wsServer.close()
-		await waitFor(() => adapter._ws?.readyState !== WebSocket.OPEN)
+		await waitFor(() => adapter.connectionStatus === 'offline')
 		expect(adapter._reconnectManager.intendedDelay).toBeGreaterThanOrEqual(INACTIVE_MIN_DELAY)
 
 		hiddenMock.mockReturnValue(false)
@@ -776,6 +816,32 @@ describe('ReconnectManager', () => {
 		expect(adapter._ws).toBeNull()
 	})
 
+	it.each(['throws', 'rejects'])(
+		'[RM1][CW1] getUri that %s is retried on the backoff instead of stranding the connection',
+		async (failure) => {
+			const consoleErrorSpy = vi.spyOn(console, 'error').mockImplementation(() => {})
+			let attempts = 0
+			const testAdapter = new ClientWebSocketAdapter(() => {
+				attempts++
+				if (attempts < 3) {
+					const error = new Error('token endpoint unavailable')
+					if (failure === 'throws') throw error
+					return Promise.reject(error)
+				}
+				return 'ws://localhost:2234'
+			})
+			try {
+				await waitFor(() => testAdapter._ws?.readyState === WebSocket.OPEN)
+				expect(attempts).toBe(3)
+				expect(consoleErrorSpy).toHaveBeenCalledTimes(2)
+				expect(testAdapter.connectionStatus).toBe('online')
+			} finally {
+				testAdapter.close()
+				consoleErrorSpy.mockRestore()
+			}
+		}
+	)
+
 	it('[RM5] close cancels timers and removes the reconnect event listeners', () => {
 		const testAdapter = new ClientWebSocketAdapter(() => 'ws://localhost:2234')
 		const manager = testAdapter._reconnectManager
@@ -794,3 +860,57 @@ describe('ReconnectManager', () => {
 		expect(() => manager.close()).not.toThrow()
 	})
 })
+
+describe('URI failure boundaries', () => {
+	it.each(['resolves', 'rejects'])('ignores getUri that %s after close', async (outcome) => {
+		const uri = promiseWithResolve<string>()
+		const getUri = vi.fn(() => uri)
+		const testAdapter = new ClientWebSocketAdapter(getUri)
+		const onStatusChange = vi.fn()
+		testAdapter.onStatusChange(onStatusChange)
+		const consoleError = vi.spyOn(console, 'error').mockImplementation(() => {})
+		try {
+			await Promise.resolve()
+			testAdapter.close()
+			if (outcome === 'resolves') uri.resolve('ws://localhost:2234')
+			else uri.reject(new Error('token endpoint unavailable'))
+			await vi.advanceTimersByTimeAsync(INACTIVE_MAX_DELAY)
+
+			expect(getUri).toHaveBeenCalledTimes(1)
+			expect(testAdapter._ws).toBeNull()
+			expect(testAdapter.connectionStatus).toBe('offline')
+			expect(onStatusChange).not.toHaveBeenCalled()
+			expect(consoleError).not.toHaveBeenCalled()
+		} finally {
+			testAdapter.close()
+			consoleError.mockRestore()
+		}
+	})
+
+	it.each([
+		{ uri: 'not a URL', openSocket: false, error: 'Invalid URL' },
+		{
+			uri: 'ws://localhost:2234',
+			openSocket: true,
+			error: 'There should be no connection attempts while already connected',
+		},
+	])('does not treat $error as a getUri failure', async ({ uri, openSocket, error }) => {
+		let nextUri: string | Promise<string> = new Promise(() => {})
+		const testAdapter = new ClientWebSocketAdapter(() => nextUri)
+		const consoleError = vi.spyOn(console, 'error').mockImplementation(() => {})
+		try {
+			await Promise.resolve()
+			nextUri = uri
+			if (openSocket) testAdapter._ws = mockSocket(WebSocket.OPEN)
+
+			const manager = testAdapter._reconnectManager as unknown as {
+				scheduleAttempt(): Promise<void>
+			}
+			await expect(manager.scheduleAttempt()).rejects.toThrow(error)
+			expect(consoleError).not.toHaveBeenCalled()
+		} finally {
+			testAdapter.close()
+			consoleError.mockRestore()
+		}
+	})
+})

package/src/lib/ClientWebSocketAdapter.ts

@@ -93,8 +93,13 @@ export class ClientWebSocketAdapter implements TLPersistentClientSocket<
 	close() {
 		this.isDisposed = true
 		this._reconnectManager.close()
+		// orphan the socket before closing it (as _closeSocket does) so its onclose can't run
+		// _handleDisconnect after disposal — that would notify listeners and schedule a reconnect
+		const ws = this._ws
+		this._ws = null
 		//  WebSocket.close() is idempotent
-		this._ws?.close()
+		ws?.close()
+		this._connectionStatus.set('offline')
 	}
 
 	/**
@@ -240,8 +245,6 @@ export class ClientWebSocketAdapter implements TLPersistentClientSocket<
 		this._handleDisconnect('manual')
 	}
 
-	// TLPersistentClientSocket stuff
-
 	_connectionStatus: Atom<TLPersistentClientSocketStatus | 'initial'> = atom(
 		'websocket connection status',
 		'initial'
@@ -513,21 +516,33 @@ export class ReconnectManager {
 		}
 	}
 
-	private scheduleAttempt() {
+	private async scheduleAttempt() {
 		assert(this.state === 'pendingAttempt')
+		if (this.isDisposed) return
 		debug('scheduling a connection attempt')
-		Promise.resolve(this.getUri()).then((uri) => {
-			// this can happen if the promise gets resolved too late
+
+		let uri: string
+		try {
+			uri = await this.getUri()
+		} catch (error) {
+			// A failed auth-token fetch must retry instead of leaving us with no socket or timer.
 			if (this.state !== 'pendingAttempt' || this.isDisposed) return
-			assert(
-				this.socketAdapter._ws?.readyState !== WebSocket.OPEN,
-				'There should be no connection attempts while already connected'
-			)
+			console.error('Failed to get the websocket URI, retrying', error)
+			this.state = 'delay'
+			this.reconnectTimeout = setTimeout(() => this.disconnected(), this.intendedDelay)
+			return
+		}
 
-			this.lastAttemptStart = Date.now()
-			this.socketAdapter._setNewSocket(new WebSocket(httpToWs(uri)))
-			this.state = 'pendingAttemptResult'
-		})
+		// this can happen if the promise gets resolved too late
+		if (this.state !== 'pendingAttempt' || this.isDisposed) return
+		assert(
+			this.socketAdapter._ws?.readyState !== WebSocket.OPEN,
+			'There should be no connection attempts while already connected'
+		)
+
+		this.lastAttemptStart = Date.now()
+		this.socketAdapter._setNewSocket(new WebSocket(httpToWs(uri)))
+		this.state = 'pendingAttemptResult'
 	}
 
 	private getMaxDelay() {
@@ -641,6 +656,7 @@ export class ReconnectManager {
 		debug('ReconnectManager.disconnected')
 		// This either means we're freshly disconnected, or the last connection attempt failed;
 		// either way, time to try again.
+		if (this.isDisposed) return
 
 		// Guard against delayed notifications and recheck synchronously
 		if (

package/src/lib/SQLiteSyncStorage.test.ts

@@ -1,7 +1,7 @@
 import { DatabaseSync } from 'node:sqlite'
 import { DocumentRecordType, PageRecordType, TLDOCUMENT_ID, TLRecord } from '@tldraw/tlschema'
 import { ZERO_INDEX_KEY } from '@tldraw/utils'
-import { describe, expect, it } from 'vitest'
+import { describe, expect, it, vi } from 'vitest'
 import {
 	contractRecords,
 	contractSchema,
@@ -9,6 +9,7 @@ import {
 	makePage,
 	registerStorageContractTests,
 } from '../test/storageContractSuite'
+import { MAX_TOMBSTONES } from './InMemorySyncStorage'
 import { NodeSqliteWrapper } from './NodeSqliteWrapper'
 import { SQLiteSyncStorage } from './SQLiteSyncStorage'
 
@@ -263,4 +264,62 @@ describe('SQLiteSyncStorage', () => {
 			expect(row?.migrationVersion).toBe(3)
 		})
 	})
+
+	describe('tombstone pruning robustness', () => {
+		const makeTombstones = (count: number) => {
+			const tombstones: Record<string, number> = {}
+			for (let i = 0; i < count; i++) tombstones[`shape:doc${i}`] = i + 1
+			return tombstones
+		}
+		function makeOverfullStorage() {
+			const db = new DatabaseSync(':memory:')
+			const storage = new SQLiteSyncStorage<TLRecord>({
+				sql: new NodeSqliteWrapper(db),
+				snapshot: makeContractSnapshot(contractRecords, {
+					tombstones: makeTombstones(MAX_TOMBSTONES + 500),
+					documentClock: MAX_TOMBSTONES + 501,
+					tombstoneHistoryStartsAtClock: 0,
+				}),
+			})
+			return { db, storage, prune: (storage as any).pruneTombstones as { (): void; flush(): void } }
+		}
+
+		it('[SS18] a prune that fires after the database was closed does not throw', () => {
+			const { db, storage, prune } = makeOverfullStorage()
+			const consoleSpy = vi.spyOn(console, 'error').mockImplementation(() => {})
+			try {
+				// a delete schedules the throttled prune; the host then closes everything
+				storage.transaction((txn) => txn.delete(contractRecords[1].id))
+				db.close()
+				// the throttle timer fires later, from outside any caller that could catch it
+				expect(() => prune.flush()).not.toThrow()
+				expect(consoleSpy).toHaveBeenCalledWith('Failed to prune tombstones', expect.anything())
+			} finally {
+				consoleSpy.mockRestore()
+			}
+		})
+
+		it('[SS18] a prune whose delete fails leaves the history clock where it was', () => {
+			const { storage, prune } = makeOverfullStorage()
+			const consoleSpy = vi.spyOn(console, 'error').mockImplementation(() => {})
+			try {
+				const stmts = (storage as any).stmts
+				const originalRun = stmts.deleteTombstonesBefore.run
+				stmts.deleteTombstonesBefore.run = () => {
+					throw new Error('SQLITE_BUSY: database is locked')
+				}
+				prune()
+				expect(() => prune.flush()).not.toThrow()
+				stmts.deleteTombstonesBefore.run = originalRun
+
+				// neither half of the prune applied: the history start was not advanced past
+				// tombstones that still exist
+				const snapshot = storage.getSnapshot()
+				expect(snapshot.tombstoneHistoryStartsAtClock).toBe(0)
+				expect(Object.keys(snapshot.tombstones!).length).toBe(MAX_TOMBSTONES + 500)
+			} finally {
+				consoleSpy.mockRestore()
+			}
+		})
+	})
 })

package/src/lib/SQLiteSyncStorage.ts

@@ -394,7 +394,6 @@ export class SQLiteSyncStorage<R extends UnknownRecord> implements TLSyncStorage
 			const documentClock = snapshot.documentClock ?? snapshot.clock ?? 0
 			const tombstoneHistoryStartsAtClock = snapshot.tombstoneHistoryStartsAtClock ?? documentClock
 
-			// Clear existing data
 			this.sql.exec(`
 				DELETE FROM ${documentsTable};
 				DELETE FROM ${objectsTable};
@@ -410,14 +409,12 @@ export class SQLiteSyncStorage<R extends UnknownRecord> implements TLSyncStorage
 				table.insert.run(doc.state.id, encodeState(doc.state), doc.lastChangedClock)
 			}
 
-			// Insert tombstones
 			if (snapshot.tombstones) {
 				for (const [id, clock] of objectMapEntries(snapshot.tombstones)) {
 					this.stmts.insertTombstone.run(id, clock)
 				}
 			}
 
-			// Insert metadata row
 			this.stmts.updateMetadata.run(
 				documentClock,
 				tombstoneHistoryStartsAtClock,
@@ -513,18 +510,29 @@ export class SQLiteSyncStorage<R extends UnknownRecord> implements TLSyncStorage
 	/** @internal */
 	pruneTombstones = throttle(
 		() => {
-			const tombstoneCount = this.stmts.countTombstones.all()[0].count as number
-			if (tombstoneCount > MAX_TOMBSTONES) {
-				// Get all tombstones sorted by clock ascending (oldest first)
-				const tombstones = this.stmts.iterateTombstones.all()
-
-				const result = computeTombstonePruning({ tombstones, documentClock: this.getClock() })
-				if (result) {
-					this.stmts.setTombstoneHistoryStartsAtClock.run(result.newTombstoneHistoryStartsAtClock)
-					// Delete all tombstones with clock < newTombstoneHistoryStartsAtClock in one operation.
-					// This works because computeTombstonePruning ensures we never split a clock value.
-					this.stmts.deleteTombstonesBefore.run(result.newTombstoneHistoryStartsAtClock)
+			// Runs from a timer, so a host that closed the database right after the last delete
+			// (e.g. `room.close(); db.close()` in onSessionRemoved) would otherwise get an uncaught
+			// throw from here; pruning is best-effort and simply runs again on the next delete.
+			try {
+				const tombstoneCount = this.stmts.countTombstones.all()[0].count as number
+				if (tombstoneCount > MAX_TOMBSTONES) {
+					// Get all tombstones sorted by clock ascending (oldest first)
+					const tombstones = this.stmts.iterateTombstones.all()
+
+					const result = computeTombstonePruning({ tombstones, documentClock: this.getClock() })
+					if (result) {
+						this.sql.transaction(() => {
+							this.stmts.setTombstoneHistoryStartsAtClock.run(
+								result.newTombstoneHistoryStartsAtClock
+							)
+							// Delete all tombstones with clock < newTombstoneHistoryStartsAtClock in one operation.
+							// This works because computeTombstonePruning ensures we never split a clock value.
+							this.stmts.deleteTombstonesBefore.run(result.newTombstoneHistoryStartsAtClock)
+						})
+					}
 				}
+			} catch (e) {
+				console.error('Failed to prune tombstones', e)
 			}
 		},
 		1000,

package/src/lib/TLSocketRoom.ts

@@ -273,13 +273,16 @@ export class TLSocketRoom<R extends UnknownRecord = UnknownRecord, SessionMeta =
 				})
 			)
 		}
+		// The default logger belongs to both layers: TLSyncRoom is where authorizer throws are
+		// logged, and a host that passes no `log` would otherwise never see them.
+		this.log = 'log' in opts ? opts.log : { error: console.error }
 		this.room = new TLSyncRoom<R, SessionMeta>({
 			onPresenceChange: opts.onPresenceChange,
 			onCommittedChanges: opts.onCommittedChanges,
 			objectTypes: opts.objectTypes,
 			authorizeRecord: opts.authorizeRecord,
 			schema: opts.schema ?? (createTLSchema() as any),
-			log: opts.log,
+			log: this.log,
 			storage,
 			clientTimeout: opts.clientTimeout,
 		})
@@ -295,7 +298,6 @@ export class TLSocketRoom<R extends UnknownRecord = UnknownRecord, SessionMeta =
 				})
 			}
 		})
-		this.log = 'log' in opts ? opts.log : { error: console.error }
 	}
 
 	/**
@@ -854,12 +856,11 @@ export class TLSocketRoom<R extends UnknownRecord = UnknownRecord, SessionMeta =
 		if (this.isClosed()) {
 			throw new Error('Cannot update store on a closed room')
 		}
+		// read through the transaction rather than getSnapshot(): the document snapshot excludes the
+		// object-store lane, which would make object-lane records invisible (and undeletable) here
+		const records = this.storage.transaction((txn) => Object.fromEntries(txn.entries())).result
 		// eslint-disable-next-line @typescript-eslint/no-deprecated
-		const ctx = new StoreUpdateContext<R>(
-			// eslint-disable-next-line @typescript-eslint/no-deprecated
-			Object.fromEntries(this.getCurrentSnapshot().documents.map((d) => [d.state.id, d.state])),
-			this.room.schema
-		)
+		const ctx = new StoreUpdateContext<R>(records, this.room.schema)
 		try {
 			await updater(ctx)
 		} finally {

package/src/test/TLSocketRoom.test.ts

@@ -210,6 +210,45 @@ describe('28. TLSocketRoom (SR)', () => {
 			consoleSpy.mockRestore()
 		})
 
+		it('[SR3] the default logger also reaches TLSyncRoom, so authorizer failures are reported', () => {
+			const consoleSpy = vi.spyOn(console, 'error').mockImplementation(() => {})
+			try {
+				const room = new TLSocketRoom({
+					authorizeRecord: {
+						page: () => {
+							throw new Error('authorizer exploded')
+						},
+					},
+				})
+				const socket = createMockSocket()
+				connectSession(room, 'test-session', socket)
+
+				const pageId = PageRecordType.createId('new-page')
+				room.handleSocketMessage(
+					'test-session',
+					JSON.stringify({
+						type: 'push',
+						clientClock: 1,
+						diff: {
+							[pageId]: [
+								RecordOpType.Put,
+								PageRecordType.create({ id: pageId, name: 'New Page', index: 'a2' as any }),
+							],
+						},
+					})
+				)
+
+				// the authorizer threw, and the host heard about it without passing a logger
+				expect(room.getRecord(pageId)).toBeUndefined()
+				expect(consoleSpy).toHaveBeenCalledWith(
+					'record authorizer threw; rejecting the write',
+					expect.objectContaining({ message: 'authorizer exploded' })
+				)
+			} finally {
+				consoleSpy.mockRestore()
+			}
+		})
+
 		it('[SR3] uses custom logger when provided', () => {
 			const mockLog: TLSyncLog = {
 				warn: vi.fn(),

package/src/test/objectStore.test.ts

@@ -8,6 +8,7 @@ import {
 	TLSocketServerSentEvent,
 	getTlsyncProtocolVersion,
 } from '../lib/protocol'
+import { TLSocketRoom } from '../lib/TLSocketRoom'
 import { TLSyncErrorCloseEventCode, TLSyncErrorCloseEventReason } from '../lib/TLSyncClient'
 import { RoomSnapshot, TLRoomSocket, TLSyncRoom } from '../lib/TLSyncRoom'
 
@@ -354,6 +355,40 @@ describe('object store lane', () => {
 			const connectMsg = late.__messages.find((m: any) => m.type === 'connect') as any
 			expect(Object.keys(connectMsg.diff).sort()).toEqual([doc.id, note.id].sort())
 		})
+
+		it('[US1][US3] updateStore sees and can delete object-lane records', async () => {
+			// eslint-disable-next-line @typescript-eslint/no-deprecated
+			const socketRoom = new TLSocketRoom<R, void>({
+				schema,
+				storage: new InMemorySyncStorage<R>({
+					snapshot: { documents: [], clock: 0, documentClock: 0, schema: schema.serialize() },
+					objectTypes: ['note'],
+				}),
+				objectTypes: ['note'],
+			})
+			disposables.push(() => socketRoom.close())
+			const doc = Doc.create({ title: 'canvas' })
+			const note = Note.create({ text: 'annotation' })
+			socketRoom.storage.transaction((txn) => {
+				txn.set(doc.id, doc)
+				txn.set(note.id, note)
+			})
+
+			// the update context reads both lanes, not just the document snapshot
+			// eslint-disable-next-line @typescript-eslint/no-deprecated
+			await socketRoom.updateStore((store) => {
+				expect(store.get(note.id)).toEqual(note)
+				expect(
+					store
+						.getAll()
+						.map((r) => r.id)
+						.sort()
+				).toEqual([doc.id, note.id].sort())
+				store.delete(note.id)
+			})
+			expect(socketRoom.getRecord(note.id)).toBeUndefined()
+			expect(socketRoom.getRecord(doc.id)).toEqual(doc)
+		})
 	})
 
 	describe('connect message', () => {