Images now load direct from images.04080616.xyz. Removes read-side
signing (signUrl/signCoverUrls + callers), unlocking browser and
edge caching since URLs are stable. Presigned upload kept for /api/upload.
PageReader retries failed loads via onError as a safety net.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Refreshing the chapter page was landing the user back at the top, because
only ?resume=1 triggered the saved-position restore. Added isPageReload()
helper (checks performance navigation entry type === 'reload') and OR'd
it with the resume flag. Refresh now restores to the last scroll
position; drawer/list clicks still go to top as intended.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
- Resume scroll position only when arriving via 继续阅读 (?resume=1).
Plain chapter-list / drawer clicks now actively scroll to top on mount.
- Progress format extended to {chapter, page, ratio} for within-page
precision; legacy bare-number and {chapter, page} still read correctly.
- Tappable skeleton logo (sunflower outline, spins) while a page loads;
tap force-fetches a fresh signed URL.
- Viewport-priority image loading: second IntersectionObserver at margin 0
marks truly-visible pages, drives <img fetchpriority="high"> and fires
immediate single-page fetches that cut the batch queue.
- Bounded image cache: unmount previous chapter's <img> elements when
currentPage > 5 into the new chapter; placeholders stay for layout.
One AbortController per live chapter; unmount aborts in-flight batches.
- Hashed chapter IDs on the wire via hashids; DB PKs unchanged.
- Origin/Referer allowlist + rate limiting on all /api/* routes via a
withGuards(opts, handler) wrapper (eliminates 6-line boilerplate x5).
- robots.txt allows Googlebot/Bingbot/Slurp/DuckDuckBot/Baiduspider/
YandexBot only; disallows /api/ for all UAs.
- Extract pure helpers for future TDD: lib/scroll-ratio.ts (calcScrollRatio,
scrollOffsetFromRatio), lib/progress.ts (parseProgress + injectable
StorageLike), lib/rate-limit.ts (optional { now, store, ipOf } deps),
lib/api-guards.ts.
- New env keys: HASHIDS_SALT, ALLOWED_ORIGINS (wired into docker-compose).
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
- Add width/height columns to Page (default 0, migration SQL committed).
- Backfill script ranges first 16KB of each R2 object and parses with
image-size. Probed all 26,209 existing pages successfully.
- Export keyFromPublicUrl from lib/r2 so the script reuses the existing
URL→key logic.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Reading progress
- New ReadingProgressButton client component that reads/writes
localStorage key "sunnymh:last-read:{slug}"
- PageReader writes the currently-visible chapter as the user scrolls
through continuous chapters
- Manga detail CTA now reads: "开始阅读" on first visit, or
"继续阅读 · #{N} {title}" when a prior chapter is stored (with
spacing and truncation for long titles)
Multiple genres
- lib/genres.ts with parseGenres() and collectGenres() helpers to
split "冒险, 恋爱, 魔幻" into a list and aggregate across a collection
- Manga detail renders one pill per genre
- GenreTabs filters via parseGenres(...).includes(activeGenre) so a
multi-genre manga appears under each of its genres
- Home / Genre pages compute the tab list from collectGenres(signedManga)
- Card captions (GenreTabs + TrendingCarousel) show "冒险 · 恋爱 · 魔幻"
with truncation
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
- Sign all image URLs server-side with 60s expiry presigned URLs
- Add /api/pages endpoint for batched page fetching (7 per batch)
- PageReader prefetches next batch when user scrolls to 3rd page
- Move chapter count badge outside overflow-hidden for 3D effect
- Fix missing URL signing on search and genre pages
- Extract signCoverUrls helper to reduce duplication
- Clamp API limit param to prevent abuse
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
