Removed:
- isTrackingEnabled state and delays
- Complex composite keys
- Verbose debug logging
- isTrackingEnabledRef checks
Back to simple:
- isTextContent = basic check (loading, content exists, not video)
- Restore once per articleIdentifier
- Save on unmount
- Suppression during restore window
Much simpler, closer to original working version.
Split tracking enable logic into two effects:
1. Reset to false when article changes (selectedUrl)
2. Enable after 500ms if isTextContent is true AND not already enabled
Prevents isTextContent flipping from resetting the timer repeatedly, which was preventing isTrackingEnabled from ever becoming true.
Previously, if restore was skipped due to missing dependencies (content not loaded), it would never retry even after content loaded. Now resets the attempt tracker whenever articleIdentifier changes, allowing retry when dependencies become available.
- Check content length before enabling tracking (uses existing 1000 char minimum)
- Wait 500ms after content loads before enabling tracking (ensures stability)
- Prevents tracking on short notes and during page load transitions
- isTextContent now uses useMemo with comprehensive checks
saveNow() was bypassing suppression, causing 0% to overwrite saved positions during restore. Now checks suppressUntilRef before saving, just like the debounced auto-save.
- Suppress saves for 1700ms when restore starts (covers collection + render time)
- If no position found or delta too small, clear suppression immediately
- If restore happens, extend suppression for 1.5s after scroll
- Prevents 0% from overwriting saved 22% position during page load
Track whether we've already attempted restore for each article using a ref. Prevents the effect from restarting multiple times as html/markdown/loading state changes during initial page load, which was stopping the stabilization timer before it could complete.
Removed the check that prevented saving 0% positions. Now tracks when articles are opened, even if not read yet. Useful for engagement metrics and history.
- Use ref for suppressSavesFor to prevent restore effect from restarting on every position change
- Skip saving positions at 0% (meaningless start position)
- Restore effect now only restarts when article actually changes, not on every scroll
The unmount effect had saveNow in its dependency array. Since saveNow is a useCallback that depends on position, it was recreated on every scroll event, triggering the effect cleanup and calling saveNow() repeatedly (every ~14ms).
Now using a ref to store the latest saveNow callback, so the cleanup only runs when selectedUrl changes (i.e., when actually navigating away).
Replace complex interval logic with simple 2-second debounce. Every scroll event resets the timer, so saves only happen after 2s of no scrolling (or when reaching 100%). Much less aggressive than the previous 15s minimum interval.
Add detailed console logs to trace:
- Position collection and stabilization
- Save scheduling, suppression, and execution
- Restore calculations and decisions
- Scroll deltas and thresholds
Logs use [reading-position] prefix with emoji indicators for easy filtering and visual scanning.
Replace continuous restore with stabilized one-shot collector. Suppress saves for 1.5s after restore, skip tiny deltas (<48px or <5%), and use instant scroll (behavior: auto) to eliminate jumpy view behavior from conflicting relay updates.
Add collectReadingPositionsOnce() that buffers position updates for ~700ms, then emits the best one (newest timestamp, tie-break by highest progress). Prevents jumpy scrolling from conflicting relay updates.
Add suppressSavesFor(ms) API to temporarily block auto-saves after programmatic scrolls, preventing feedback loops where restore triggers save which triggers restore.
- Added third relay education article link in PWA settings
- Links to /a/naddr1qvzqqqr4gupzq3svyhng9ld8sv44950j957j9vchdktj7cxumsep9mvvjthc2pjuqq9hyetvv9uj6um9w36hq9mgjg8
- Updated punctuation to use commas for better readability (here, here, and here)
- Changed timestamp links in CardView and LargeView to use internal routes
- Articles (kind:30023) open in /a/{naddr}
- Notes (kind:1) open in /e/{eventId}
- External URLs open in /r/{encodedUrl}
- Removed unused eventNevent prop and neventEncode import
- Timestamp now uses Link component for client-side navigation
- Refactored VideoEmbedProcessor to process HTML and extract URLs in single pass
- Previously processedHtml and videoUrls were computed separately, causing index mismatches
- Now both are computed together ensuring placeholders match collected URLs
- Added check to skip empty HTML parts to prevent rendering stray characters
