- Added disney_park TEXT DEFAULT NULL to assets table in _create_v2_tables
- Created service_entrances table with id, location_id, name, latitude,
longitude, notes, created_at, updated_at; location_id FK to locations
with ON DELETE CASCADE; name/lat/lon are NOT NULL
- Added v4 migration in init_db: ALTER TABLE for disney_park on existing
DBs, schema-aware recreation of service_entrances if name column lacks
NOT NULL constraint
- 44/45 tests pass (1 pre-existing 404 failure on stats endpoint)
- Retake full set of p9a_admin_*.png screenshots (login, dashboard, settings, users, customers, activity, export, import, sidebar)
- Each screenshot inspected for layout errors
- Proper 1280px desktop viewport for all pages, clean empty states
- Save to docs/images directory
- Replace p9a_admin_sync.png with p9a_admin_import.png (new Import page)
- Add p9a_admin_sidebar.png (mobile sidebar with hamburger open)
- Update ADMIN_GUIDE.md with mobile navigation section
- Add screenshot reference in Import section
- Replace Cantaloupe Sync section with Excel Import section
- Update troubleshooting tips for import (not sync)
- Mention hamburger menu in mobile description
The test suite login as 'admin'/'changeme' but only 'shawn' was seeded.
Adding the admin user with SHA256('changeme') hash lets all 15 Playwright
tests reach the actual app pages instead of failing at login.
- Removed hardcoded display:none from GPS center chip on map tab
- Chip now shows when GPS position is obtained in startVisitTracking
- Added leaflet-draw CSS and JS to HTML head
- Added draw control (polygon, rectangle, circle) to map init
- Emits geofenceDrawn custom event for external hooks
- Manual photo area now shows two clear choices:
📁 Choose from Gallery — plain accept=image/* (EXIF preserved)
📸 Take Photo — calls openManualCamera() for quick capture
- Each has a label explaining EXIF behavior
- openManualCamera() now appends input to DOM before click()
(required by mobile Chrome for programmatic file input clicks)
- Dynamically created camera input is cleaned up after use
- camera-area overflow:visible to not clip the taller placeholder
- manPhotoInput now uses accept=image/* without capture attribute
-> opens gallery/file picker which gives raw file bytes (EXIF preserved)
- Camera shortcut preserved via openManualCamera() creates a temp
capture=environment input for quick field capture (no EXIF but available)
- Label updated: 'Tap to take photo' -> 'Tap to choose photo'
- Manual photo capture row references openManualCamera() instead
GPS:
- initGPS() now shows tappable GPS badge instead of failing silently
- Badge pulses amber with '📍 Tap for GPS' — click triggers permission prompt
- fillGpsParking() auto-triggers badge click if GPS not yet available
- requestLocation() extracted as reusable Promise
EXIF bypass:
- submitManualAsset() now uses server-extracted GPS from raw upload bytes
- Server returns exif_gps in /api/upload/photo response (was already working)
- Client was ignoring it — now fills parking + map link fields automatically
- Bypasses Android content provider EXIF stripping since server reads raw bytes
Admin separation:
- admin.canteen.ourpad.casa now served by systemd service (canteen-admin.service)
- Main server also systemd (canteen-main.service) — both auto-restart on crash/boot
- extractExifData() now async — reads real EXIF tags (Make, Model,
DateTimeOriginal, ISO, FNumber, ExposureTime, FocalLength, GPS)
via exifr.parse() and displays them in the UI
- handlePickedPhoto() + reader.onload made async to support this
- GPS extraction already auto-triggers on photo pick (tryExtractGpsFromPicked)
- Server-side exif_data round-trip already functional (verified)
- DNG/RAW-02: .dng allowed, 20MB limit
- Added .dng to PHOTO_ALLOWED_EXTS
- Bumped PHOTO_MAX_SIZE from 10MB to 20MB
- DNG files (~13MB from Pixel 9a) now accepted by all upload paths:
/api/upload/photo, /api/ocr, connect_labels
- Server validates extension+size only; DNG EXIF preserved naturally
by write_bytes(); _re_embed_exif() correctly skips non-JPEG;
_extract_gps_from_bytes() opens DNG as TIFF via PIL (already works)
Verified: .dng upload via /api/upload/photo and /api/ocr both succeed,
file saved byte-identical.
- Removed duplicate clearPickedPhoto() at old line ~1772
(kept newer version at line ~3857 with pickedPhotoGps clearing)
- Removed duplicate extractGpsFromPicked() at old line ~1831
(kept newer async version at line ~3766 that uses exifr)
- Both ocrPickedPhoto() and createAssetFromPicked() were already
singular — they were refactored in a prior edit
Verified: no duplicate function names, JS syntax OK, no regressions
- Set manualPhotoFile = pickedPhotoFile directly (preserves EXIF)
- Set manualPhotoBlob = pickedPhotoFile (preserves EXIF)
- Keep FileReader for preview display only
- Trigger EXIF GPS extraction via extractGpsFromPhoto()
- Fill GPS coordinates into manual form (parking + map link)
Previously the function used readAsDataURL -> atob -> Blob
which stripped all EXIF metadata. Now the original File
flows through to submitManualAsset where exifr.parse()
re-embeds EXIF on the server.
- Added global let map/heatLayer/heatVisible declarations
- Leaflet map initializes on first visit to Map tab (OSM tiles)
- Implemented loadAssetPins() - fetches assets with coords from API,
creates Leaflet markers with popups, auto-fits bounds
- Guard toggleHeatmap() with proper declared globals
- Closes#31 H1, #32 H2
Also confirmed M1 (empty asset list) is not a bug: API returns [] because
DB has no assets yet. Empty state renders correctly.
- Client: exifr.parse() reads full EXIF before upload, sent as exif_data
- Server: piexif re-embeds EXIF into saved JPEG after upload
- Auto-extract GPS from picked gallery photos on selection
- Removed Google Photos button (needs OAuth, not feasible)
- Replaced getUserMedia() + canvas.toBlob() with <input type=file capture=environment>
→ native camera app preserves full EXIF data including GPS
- Defined missing extractGpsFromPhoto() using exifr.gps() library
- Manual mode now uploads the original file (with EXIF) instead of a canvas-encoded blob
- Server-side exif_gps fallback if client-side extraction fails
- Renamed manualPhotoBlob → manualPhotoFile for clarity
- backup-db.sh: copies assets.db to /home/oplabs/backups/canteen-db/
- start.sh: runs backup automatically before every launch
- Daily cron at 3am for unattended backups
- Keeps last 30 backups, prunes oldest
- Replaced loadSettingsCache() with hardcoded seed data (categories, makes, keys, badges)
- Removed tabActivity lifecycle hook (activity log moved to admin app)
- Updated version to v3.0 + Admin Panel link in drawer footer
- Updated PROJECT.md with v3 project structure (3 repos, admin server docs)
- Removed backup files (.bak, .orig)
- Main index.html reduced from 7645 to 4228 lines
- _extract_gps_from_bytes helper reads GPSInfo IFD via PIL get_ifd()
- Converts DMS to decimal degrees, returns {lat, lng} or None
- /api/upload/photo: reads bytes first, extracts GPS before saving
- /api/ocr: extracts GPS from raw bytes before OCR processing
- exif_gps returned alongside path/machine_id in both endpoints
- Photos without EXIF GPS omit the exif_gps key entirely
- Raw file bytes preserved (no re-compression, no EXIF stripping)
Two places referenced old photoPicker element ID that was removed:
1. connectFromGallery() — switched to window._connectMode flag + galleryInput.click()
2. Connect mode hook IIFE — now hooks both cameraInput and galleryInput via
shared hookPicker() helper using window._connectMode instead of DOM attr
This was crashing the entire inline script (TypeError on null) ~line 4040,
preventing initAuth()/checkAuthGate() from ever running — no login screen.
- Two dedicated <input> elements: cameraInput (capture=environment) for
direct camera, galleryInput (no capture) for gallery picker
- Both wired to handlePickedPhoto() via a shared wireInput helper
- refreshGPS() fires on button click (before picker opens) for a fresh
high-accuracy GPS fix — no maximumAge, so it pulls current location
instead of relying on cached values from page-load initGPS()
- Workaround for browser EXIF GPS stripping: camera captures may preserve
EXIF GPS, gallery picks fall through to device geolocation API
- Removed extension whitelist — no more 'unsupported format' errors
- Falls back to .jpg temp file for unknown/no extension
- PIL detects actual image format from file bytes (handles DNG as TIFF)
- Android browsers often strip DNG extensions when picking from gallery
- Tested: works with .dng, .jpg, no extension, and .bogus
- 6s was too short for 13MB DNG upload on mobile
- Bumped to 20s to handle large RAW photos
- Added 'Uploading to server OCR...' status message during upload
- Server OCR handles DNG correctly (tested via curl), so the
timeout was the root cause of gallery upload failures
- Server: accept .dng in OCR endpoint (PIL opens as TIFF)
- Server: bump OCR size limit to 20MB (DNG files can be large)
- Client: gallery upload tries server OCR first (with 6s timeout),
then falls back to client-side Tesseract.js — matches 📸 capture flow
- DNG photos now work with 'Pick from Gallery'
- Catches 'unknown error' from Tesseract.js (unsupported
image formats like RAW/DNG) and shows helpful guidance:
'Use the 📸 live camera instead'
- Falls through to actual error message for other failures
- Better null-safety on error object
- Login lag: camera/GPS/assets no longer start before login.
Only initAuth() + initOfflineSupport() run on pageload.
Camera/GPS/assets start after successful login (or auto-login
from stored session).
- 📸 capture overlay: was defined in HTML but never shown.
Now shown when camera starts, hidden when scanning stops.
- Status text: updated from 'Point camera at a barcode' to
'Point camera at a barcode, or tap 📸 for text stickers'
- GPS: when EXIF stripped by mobile browser, falls back to
navigator.geolocation.getCurrentPosition() for device location
- OCR result now populates scanMachineId hidden field so Create works
- createFromScanPhoto also checks AppState.gpsLat as fallback
- Gallery OCR now auto-lookups via handleBarcode() — routes to confirm
card (existing asset) or new asset form (not found)
- Manual entry button also calls handleBarcode() after prompt
- Existing assets now show a confirm card with machine ID + ✓ Confirm & Check In / ✏️ Edit buttons
- Confirm triggers auto check-in then shows the result card
- Edit opens inline machine ID editor with Re-lookup
- Cancel button returns to original confirmation
- New asset flow unchanged (form already editable)
- Removed separate OCR and Connect mode toggles — merged into Scan
- Renamed Barcode mode to Scan with OCR fallback + gallery upload
- Added captureScanOcr() — tap to capture frame for OCR when barcode fails
- handleBarcode() now compares barcode vs OCR result, picks valid ID
- Gallery photo picker integrated into Scan mode (OCR + EXIF GPS)
- Removed old photoPreviewCard and connect-mode HTML sections
- Removed orphaned functions referencing removed elements
- Added missing #offlineBanner HTML element with queueCountBadge
(showOfflineBanner/hideOfflineBanner were referencing it but it didn't exist)
- Enhanced SW: API GET calls now use network-first with cache fallback
(POST/PUT/DELETE still pass through — main thread handles offline queuing)
- Bumped SW cache from canteen-v1 to canteen-v2
shawn