Replace the failure counter with a list of typed failures (fetch,
build, save, readme, unknown) carrying path, sha, message, and
affected doc ids so callers can surface diagnostics instead of just
a count.
Fetch blobs with a concurrency of 8 instead of one at a time, batch
PouchDB writes via bulkUpdate, filter already-cached files up front,
and run the README fetch in parallel with the worker loop.
The home page reads README via getCachedMainReadme, which uses
Note-{owner}-{repo}-README, not the blob SHA. Cache-all only wrote
SHA-keyed entries, so the homepage README was empty offline.
A single null from queryFileContent silently exited cacheAllNotes,
leaving every later note uncached while the success toast still fired.
Skip the failing file, count failures, and surface them via errorMessage.
All database reads and writes now run off the main thread via a
dedicated worker, eliminating IndexedDB overhead from the frame budget.
- Create data.worker.ts exposing the Data class via Comlink
- Refactor data.ts to export a Comlink-wrapped proxy and a standalone
generateId() pure function (workers can't expose sync methods cleanly)
- Update all 10 call sites to import generateId directly instead of
calling data.generateId()
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
