Fix needsPush for sharded categories and add debug logging

Author: ErcinDedeoglu

src/storage/repo/repo-client.ts

@@ -103,15 +103,18 @@ export class RepoStorageBackend implements StorageBackend {
 
   public async getFile(path: string): Promise<string | null> {
     const fullPath = `${SYNC_DIR}/${path}`;
-    const res = await this.fetchAllowNotFound(`/contents/${fullPath}`);
+    // Add cache-bust to avoid stale GitHub API responses
+    const cacheBust = Date.now();
+    const res = await this.fetchAllowNotFound(`/contents/${fullPath}?cb=${String(cacheBust)}`);
     if (!res?.ok) return null;
 
     const data = (await res.json()) as ContentFile;
 
     // For files >1MB, GitHub returns empty content and provides download_url
     if (!data.content && data.size > 1000000) {
       const branch = await this.getBranch();
-      const downloadUrl = `https://raw.githubusercontent.com/${this.owner}/${this.repo}/${branch}/${fullPath}`;
+      // Add cache-bust to avoid stale CDN responses
+      const downloadUrl = `https://raw.githubusercontent.com/${this.owner}/${this.repo}/${branch}/${fullPath}?cb=${String(cacheBust)}`;
       const downloadRes = await fetchWithRetry(
         downloadUrl,
         {

src/sync/operations/pull.ts

@@ -115,8 +115,15 @@ async function pullShardedCategoryToAcc(
   backend: StorageBackend,
   acc: PullAccumulator
 ): Promise<void> {
+  syncLog(`[PULL] Fetching shard: ${ref.shardFile} (itemCount: ${String(ref.itemCount)})`);
   const shard = await fetchCategoryShard(backend, ref.shardFile);
-  if (!shard) return;
+  if (!shard) {
+    syncLog(`[PULL] ${cat}: shard file not found or empty`);
+    return;
+  }
+  syncLog(
+    `[PULL] ${cat}: shard has ${String(Object.keys(shard.items).length)} items, ${String(Object.keys(shard.tombstones).length)} tombstones`
+  );
 
   // Convert shard to ItemCategoryInfo-like structure for pullItemCategory
   const info: ItemCategoryInfo = {

src/sync/operations/push.ts

@@ -178,6 +178,8 @@ function packItemCategoryData(
 
 /**
  * Check if local data needs pushing by comparing checksums.
+ * For sharded categories (sessions, messages), we skip checksum comparison
+ * since we don't have the shard content loaded. These rely on vector clock comparison.
  */
 export function needsPush(localData: CategoryData[], remoteManifest: Manifest | null): boolean {
   if (!remoteManifest) return true;
@@ -186,8 +188,11 @@ export function needsPush(localData: CategoryData[], remoteManifest: Manifest |
     const remoteInfo = remoteManifest.categories[catData.category];
 
     if (isItemCategoryData(catData)) {
-      // Per-item comparison
-      if (!remoteInfo || !isItemCatInfo(remoteInfo)) return true;
+      // Per-item comparison - but only if remote is also items type (not sharded)
+      if (!remoteInfo) return true;
+      // Skip comparison for sharded categories - they use vector clock comparison
+      if (remoteInfo.type === 'sharded') continue;
+      if (!isItemCatInfo(remoteInfo)) return true;
       const diff = diffItems(catData.checksums, remoteInfo.items);
       if (diff.toUpload.length > 0) return true;
     } else if (isBlobCategoryData(catData)) {