Use compaction request as signal for full state update

Author: chriszarate

lib/compat/wordpress-7.0/class-wp-http-polling-sync-server.php

@@ -434,17 +434,14 @@ private function get_updates( string $room, int $client_id, int $cursor, bool $i
 			}
 
 			// Determine if this client should perform compaction.
-			$compaction_request = null;
-			if ( $is_compactor && $total_updates > self::COMPACTION_THRESHOLD ) {
-				$compaction_request = $updates_after_cursor;
-			}
+			$should_compact = $is_compactor && $total_updates > self::COMPACTION_THRESHOLD;
 
 			return array(
-				'compaction_request' => $compaction_request,
-				'end_cursor'         => $this->storage->get_cursor( $room ),
-				'room'               => $room,
-				'total_updates'      => $total_updates,
-				'updates'            => $typed_updates,
+				'end_cursor'     => $this->storage->get_cursor( $room ),
+				'room'           => $room,
+				'should_compact' => $should_compact,
+				'total_updates'  => $total_updates,
+				'updates'        => $typed_updates,
 			);
 		}
 	}

packages/sync/src/providers/http-polling/README.md

@@ -55,7 +55,7 @@ Updates are tagged with a type to enable different server-side handling:
 | `sync_step1` | State vector announcement    | Stored, delivered to other clients   |
 | `sync_step2` | Missing updates response     | Stored, delivered to other clients   |
 | `update`     | Regular document change      | Stored until compacted               |
-| `compaction` | Merged updates via Y.mergeUpdates | Clears older updates, then stored |
+| `compaction` | Full document state via Y.encodeStateAsUpdate | Clears older updates, then stored |
 
 ## Data Flow
 
@@ -109,13 +109,9 @@ To prevent unbounded message growth, the server coordinates compaction:
 
 1. **Threshold reached**: Server detects >50 stored updates for a room
 2. **Client selection**: Server nominates the lowest active client ID
-3. **Compaction request**: Server sends all updates to the nominated client via `compaction_request`
-4. **Client merges**: Uses `Y.mergeUpdates()` to combine all updates, preserving operation metadata
-5. **Client sends compaction**: The merged update replaces older updates on the server
-
-**Why Y.mergeUpdates instead of Y.encodeStateAsUpdate?**
-
-`Y.mergeUpdates()` preserves the original operation metadata (client IDs, logical clocks). This allows Yjs to correctly deduplicate when a compaction is applied to a document that already contains some of those operations. Using `Y.encodeStateAsUpdate()` would create fresh metadata, causing content duplication on clients that already have overlapping state.
+3. **Compaction request**: Server sends `should_compact: true` to the nominated client
+4. **Client encodes**: Uses `Y.encodeStateAsUpdate()` to capture the full document state
+5. **Client sends compaction**: The encoded state replaces older updates on the server
 
 ### 5. Awareness
 
@@ -164,7 +160,7 @@ Single endpoint for bidirectional sync including awareness. Clients send their u
       "updates": [
         { "type": "update", "data": "base64-encoded-yjs-update" }
       ],
-      "compaction_request": null
+      "should_compact": false
     }
   ]
 }
@@ -177,7 +173,7 @@ Single endpoint for bidirectional sync including awareness. Clients send their u
 - `after`: Cursor timestamp; only receive updates newer than this
 - `awareness`: Client's awareness state (or null to disconnect)
 - `end_cursor`: New cursor to use in next request
-- `compaction_request`: Array of all updates if this client should compact (null otherwise)
+- `should_compact`: Boolean indicating whether this client should compact
 - `updates`: Array of typed updates with base64-encoded Yjs data
 
 ## Permissions

packages/sync/src/providers/http-polling/polling-manager.ts

@@ -50,6 +50,7 @@ interface RegisterRoomOptions {
 
 interface RoomState {
 	clientId: number;
+	createCompactionUpdate: () => SyncUpdate;
 	endCursor: number;
 	localAwarenessState: LocalAwarenessState;
 	log: LogFunction;
@@ -62,31 +63,6 @@ interface RoomState {
 
 const roomStates: Map< string, RoomState > = new Map();
 
-/**
- * Create a compaction update by merging existing updates. This preserves
- * the original operation metadata (client IDs, logical clocks) so that
- * Yjs deduplication works correctly when the compaction is applied.
- *
- * @param updates The updates to merge
- */
-function createCompactionUpdate( updates: SyncUpdate[] ): SyncUpdate {
-	// Extract only compaction and update types for merging (skip sync-step updates).
-	// Decode base64 updates to Uint8Array for merging.
-	const mergeable = updates
-		.filter( ( u ) =>
-			[ SyncUpdateType.COMPACTION, SyncUpdateType.UPDATE ].includes(
-				u.type
-			)
-		)
-		.map( ( u ) => base64ToUint8Array( u.data ) );
-
-	// Merge all updates while preserving operation metadata.
-	return createSyncUpdate(
-		Y.mergeUpdates( mergeable ),
-		SyncUpdateType.COMPACTION
-	);
-}
-
 /**
  * Create sync step 1 update (announce our state vector).
  *
@@ -306,11 +282,11 @@ function poll(): void {
 				roomState.updateQueue.addBulk( responseUpdates );
 
 				// Respond to compaction requests from server. The server asks only one
-				// client at a time to compact (lowest active client ID). We merge the
-				// received updates (the server has given us everything it has).
-				if ( room.compaction_request ) {
+				// client at a time to compact (lowest active client ID). We encode our
+				// full document state to replace all prior updates on the server.
+				if ( room.should_compact ) {
 					roomState.updateQueue.add(
-						createCompactionUpdate( room.compaction_request )
+						roomState.createCompactionUpdate()
 					);
 				}
 			} );
@@ -387,6 +363,11 @@ function registerRoom( {
 
 	const roomState: RoomState = {
 		clientId: doc.clientID,
+		createCompactionUpdate: () =>
+			createSyncUpdate(
+				Y.encodeStateAsUpdate( doc ),
+				SyncUpdateType.COMPACTION
+			),
 		endCursor: 0,
 		localAwarenessState: awareness.getLocalState() ?? {},
 		log,

packages/sync/src/providers/http-polling/types.ts

@@ -31,8 +31,8 @@ interface SyncEnvelopeFromClient {
 
 interface SyncEnvelopeFromServer {
 	awareness: AwarenessState;
-	compaction_request?: SyncUpdate[];
 	end_cursor: number; // use as `after` in next request
+	should_compact?: boolean;
 	room: string;
 	updates: SyncUpdate[];
 }