fix: prevent stale personalized cache and preserve page edits

[cursor]

Summary

• Prevent the service worker from caching /, /auth, private/no-store/no-cache responses, or Set-Cookie responses; bump the SW cache version to evict previously cached personalized home HTML.
• Render moderation page-fixer thumbnails against the saved edited PDF baseline with only the pending reorder/rotation delta applied.
• Include saved pageEdits in course paper list data and pass them into split-view PDF rendering/markdown extraction.

Bug and impact

• A signed-in user's personalized home HTML could be cached by the service worker and shown after sign-out or account switch, leaking stale user identity.
• Page-fixer thumbnails for papers with saved reorder/rotation fixes could show the wrong page or double-rotated page, leading moderators to save corrupt page metadata.
• Split-view PDF rendering omitted saved page fixes, so users could see and extract markdown from the unfixed paper layout.

Root cause

• The service worker cached credentialed navigation/prefetch responses without honoring route or response cacheability.
• Thumbnail rendering used source page indices/absolute rotations even though the active PDF buffer already had saved page edits applied.
• Course paper list rows did not carry pageEdits into the split-view item.

Validation

• pnpm exec tsx -e ... targeted page-edit render mapping regression check passed.
• node -e ... targeted service-worker cache guard check passed.
• pnpm exec tsc --noEmit --pretty false passed.
• pnpm build passed with local placeholder env and PostgreSQL service.

  

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated (UTC)
examcooker-dev	Error		Jun 13, 2026 11:14am

[greptile-apps]

Greptile Summary

This PR fixes three related bugs: stale personalized HTML being served by the service worker after sign-out, incorrect thumbnails in the moderation page-fixer when saved page edits exist, and split-view PDF rendering ignoring saved page fixes.

• Service worker hardening: adds isCacheableResponse to block responses with Cache-Control: private/no-store/no-cache or Set-Cookie headers from being written to the page cache; excludes / and /auth exactly via NO_CACHE_PATH_EXACT; bumps the cache version to v4 to evict previously cached personalized HTML.
• Thumbnail rendering fix: introduces getPdfPageRenderEntry in lib/pdf/page-edits.ts to compute the correct page index and incremental rotation delta against the already-processed (baseline-edited) PDF buffer, preventing double-rotation and wrong-page thumbnails in the moderator UI.
• Split-view data flow: threads pageEdits from the DB query through CoursePaperListItem → PaperSplitItem → PDFViewerClient, with a Redis cache key bump (course-paper-rows-v2) to invalidate entries missing the new field.

Confidence Score: 4/5

Safe to merge — all three bugs are addressed by well-scoped, targeted changes with no regressions in the data flow or rendering paths.

The service worker changes correctly guard all write sites except one: the background-refresh branch inside cacheFirst's cache-hit path still uses the old response.ok && !opaque check. For static assets this is benign in practice, but it leaves a small gap in the consistency of the new isCacheableResponse contract. The PDF page-edit and split-view changes look correct and the Redis cache key bump ensures old entries are evicted.

public/sw.js — the cacheFirst background-refresh write at line 185 was not updated to use isCacheableResponse.

Important Files Changed

Filename	Overview
public/sw.js	Adds isCacheableResponse to block private/no-store/set-cookie responses from the page cache; adds NO_CACHE_PATH_EXACT for / and /auth; bumps CACHE_VERSION to v4. The cacheFirst background-refresh path (hit branch, line 185) still uses the old ok+!opaque guard instead of isCacheableResponse — minor inconsistency for static assets.
lib/pdf/page-edits.ts	Adds getPdfPageRenderEntry and toPdfPageRotation helpers to compute the correct page index and rotation delta against a baseline-edited PDF buffer. Logic correctly handles negative deltas via modular normalization and falls back safely when baselineEntry is not found.
app/components/moderation/past-paper-page-editor.tsx	Threads baselinePageEdits down to PageEntryCard and uses getPdfPageRenderEntry to derive the correct pageIndex/rotation for PdfPageThumbnail against the already-processed PDF buffer. Change is well-scoped and memoized correctly.
lib/data/course-papers.ts	Adds pageEdits to the DB select projection and CoursePaperListItem type; bumps Redis cache key to course-paper-rows-v2 to invalidate entries missing the new field. Clean and correct.
app/components/past_papers/paper-split-view.tsx	Adds pageEdits to PaperSplitItem and forwards it to PDFViewerClient, completing the data flow from the DB through to split-view rendering. Straightforward type and prop addition.
app/components/past_papers/course-paper-grid.tsx	Passes paper.pageEdits into the PaperSplitItem mapping. One-line addition that correctly connects the data layer field to the split-view consumer.
app/components/past_papers/course-paper-card.tsx	Adds pageEdits to the Paper prop type to match the updated CoursePaperListItem. Type-only change, no logic impact.

Flowchart

%%{init: {'theme': 'neutral'}}%%
flowchart TD
    A["SW fetch event (navigate/HTML)"] --> B{isUncacheable?}
    B -- "/ or /auth or /api/..." --> C[Pass through — no cache]
    B -- No --> D[staleWhileRevalidate]
    D --> E[Fetch network response]
    E --> F{isCacheableResponse?}
    F -- "private/no-store/no-cache/set-cookie" --> G[Discard — not cached]
    F -- Yes --> H[Write to PAGE_CACHE]

Loading

Comments Outside Diff (1)

1. public/sw.js, line 183-191 (link)

   Inconsistent cache guard in cacheFirst background-refresh path

   The cacheFirst hit path still uses the old response.ok && response.type !== "opaque" guard on its background-refresh write (line 185), while all other write sites in this PR were updated to use isCacheableResponse. If a static asset somehow starts returning a private or set-cookie response, this path would cache it. The two miss-path writes (staleWhileRevalidate and the miss branch of cacheFirst) were correctly updated, but this hit-path branch was overlooked.

   

_{Reviews (1): Last reviewed commit: "fix: prevent stale personalized cache an..." | Re-trigger Greptile}

[greptile-apps]

Commit message style

The PR title reads well as a conventional commit, but individual commit messages in this branch would benefit from being more descriptive — stating why each change was made, not just what changed (e.g., "bump SW cache version to v4 to evict stale user-personalized page cache" rather than just a generic description). This makes bisecting and blame much easier for the team.

Context Used: Encourage people to write better commit messages (source)

Note: If this suggestion doesn't match your team's coding style, reply to this and let me know. I'll remember it for next time!