Add processed icons for glasnudeln, grünkohl, kokosnuss, lychee,
mangold, pak choi, pastinaken, reisnudeln, rettich, stroh 80, and
topinambur. Add ImageMagick script to remove Gemini watermark and
black background from raw icons. Update catalog and re-embed.
The CATEGORY_MAP was based on BLS 3.x letter codes which were completely
reshuffled in version 4.0. This caused wrong categories like Schwarztee
showing "Wurstwaren" instead of "Getränke". Remapped all 20 letter codes
to match actual BLS 4.0 Hauptlebensmittelgruppen and regenerated blsDb.
Real-time shopping list with SSE sync between multiple clients, automatic
item categorization using embedding-based classification + Bring icon
matching, and card-based UI with category grouping.
- SSE broadcast for live sync (add/check/remove items across tabs)
- Hybrid categorizer: direct catalog lookup → category-scoped embedding
search → per-category default icons, with DB caching
- 388 Bring catalog icons matched via multilingual-e5-base embeddings
- 170+ English→German icon aliases for reliable cross-language matching
- Move cospend dashboard to /cospend/dash, /cospend redirects to list
- Shopping icon on homepage links to /cospend/list
Scrape scripts for ExerciseDB v2 API (scrape-exercises.ts,
download-exercise-media.ts), raw data for 200 exercises with
images/videos, and a 1:1 mapping from ExerciseDB IDs to internal
kebab-case slugs (exercisedb-map.ts). 23 exercises matched to
existing internal IDs, 177 new slugs generated.
- Camera-based barcode scanning in FoodSearch using barcode-detector (ZXing WASM)
- Import script to load OFF MongoDB dump into lean openfoodfacts collection
with kJ→kcal fallback and dedup handling
- Barcode lookup API with live OFF API fallback that caches results locally,
progressively enhancing the local database
- Add 'off' source to food log, custom meal, and favorite ingredient models
- OpenFoodFact mongoose model for the openfoodfacts collection
resolve() uses CWD which in production (adapter-node) is dist/, not the
project root. Detect the correct data directory at startup and add a
postbuild step to copy the embedding JSON files into dist/data/.
The deployment server couldn't fetch transformer models at runtime due to
restricted network access and permission errors writing to node_modules.
Add a prebuild script to download models during build and document
TRANSFORMERS_CACHE env var for configuring a shared writable cache path.
Dual-source nutrition system using BLS (German, primary) and USDA (English, fallback)
with ML embedding matching (multilingual-e5-small / all-MiniLM-L6-v2), hybrid
substring-first search, and position-aware scoring heuristics.
Includes per-recipe and global manual ingredient overwrites, ingredient skip/exclude,
referenced recipe nutrition (base refs + anchor tags), section-name dedup,
amino acid tracking, and reactive client-side calculator with NutritionSummary component.
Adds a `debug` command that temporarily enables cleartext traffic and
points frontendDist at the local dev server, then restores release
config on exit via trap.
Move GPS collection from WebView JS (watchPosition) to native Android
LocationForegroundService, which survives screen-off. JS polls native
side for accumulated points. Also: auto-enable GPS for cardio exercises,
filter saved track to workout duration only, fix live map batch updates,
notification tap opens active workout, and fix build script for pnpm.
Wraps the web app in a Tauri Android shell that provides native GPS
via the geolocation plugin. Includes foreground service for background
tracking, live map display, GPS data storage in workout sessions,
and route visualization in workout history.
grep -oP '.' splits multi-byte emoji into individual bytes when the
locale is not UTF-8 (e.g. CI runners with LANG=C), causing pyftsubset
to fail on invalid codepoints.
Remove redundant `font-family: sans-serif` from 18 component-level
declarations — they now inherit the Helvetica/Arial/Noto Sans stack
from the global `*` selector in app.css.
Add self-hosted NotoColorEmoji subset (56 KB, down from 11 MB) as
fallback for systems without the Noto Color Emoji font installed.
The subset is generated at prebuild time via pyftsubset with a fixed
list of the ~32 emojis actually used on the site.
- Extract Bible lookup logic into shared src/lib/server/bible.ts module
- Add build script to pre-generate all 20 mystery verse lookups as static data,
eliminating runtime API calls on rosary page load
- Update Prayer.svelte to pass showLatin/urlLang as snippet parameters; all 14
prayer components now conditionally render only visible language elements
instead of hiding via CSS
- Extract 4 inline mystery selector SVGs into MysteryIcon.svelte component
- Remove unused CSS selectors from angelus page
Allow migration to run without browser session by using ADMIN_SECRET_TOKEN
environment variable. This enables running the migration directly on the
server via SSH.
Changes:
- Add ADMIN_SECRET_TOKEN support to migration endpoint
- Update shell script to read token from environment
- Improve script with better error handling and token validation
- Update documentation with admin token setup instructions
The endpoint now accepts authentication via either:
- Valid user session (browser-based)
- ADMIN_SECRET_TOKEN from environment (server-based)
Usage on server:
source .env && ./scripts/migrate-image-hashes.sh
Add content-based hashing to recipe images for proper cache invalidation
while maintaining graceful degradation through dual file storage.
Changes:
- Add imageHash utility with SHA-256 content hashing (8-char)
- Update Recipe model to store hashed filenames in images[0].mediapath
- Modify image upload endpoint to save both hashed and unhashed versions
- Update frontend components to use images[0].mediapath with fallback
- Add migration endpoint to hash existing images (production-only)
- Update image delete/rename endpoints to handle both file versions
Images are now stored as:
- recipe.a1b2c3d4.webp (hashed, cached forever)
- recipe.webp (unhashed, graceful degradation fallback)
Database stores hashed filename for cache busting, while unhashed
version remains on disk for backward compatibility and manual uploads.
- Replace 8 duplicate formatCurrency functions with shared utility
- Add comprehensive formatter utilities (currency, date, number, etc.)
- Set up Vitest for unit testing with 38 passing tests
- Set up Playwright for E2E testing
- Consolidate database connection to single source (src/utils/db.ts)
- Add auth middleware helpers to reduce code duplication
- Fix display bug: remove spurious minus sign in recent activity amounts
- Add path aliases for cleaner imports ($utils, $models)
- Add project documentation (CODEMAP.md, REFACTORING_PLAN.md)
Test coverage: 38 unit tests passing
Build: successful with no breaking changes
Alexander