- Add PR template with description, type of change, and contributing checklist
- Enforce target branch: label + comment + 24h auto-close for PRs targeting main
- Flag bad issue titles: label + comment + 24h auto-close instead of instant close
- Redirect feature requests to Discussions (instant close, unchanged)
- Add two scheduled workflows to close stale labeled issues and PRs after 24h
- Update CONTRIBUTING.md with tests and branch up-to-date requirements
Three distinct bugs caused infinite OIDC redirect loops:
1. After logout, navigating to /login with no signal to suppress the
auto-redirect caused the login page to immediately re-trigger the
OIDC flow. Fixed by passing `{ state: { noRedirect: true } }` via
React Router's navigation state (not URL params, which were fragile
due to async cleanup timing) from all logout call sites.
2. On the OIDC callback page (/login?oidc_code=...), App.tsx's
mount-level loadUser() fired concurrently with the LoginPage's
exchange fetch. The App-level call had no cookie yet and got a 401,
which (if it resolved after the successful exchange loadUser()) would
overwrite isAuthenticated back to false. Fixed by skipping loadUser()
in App.tsx when the initial path is /login.
3. React 18 StrictMode double-invokes useEffect. The first run called
window.history.replaceState to clean the oidc_code from the URL
before starting the async exchange, so the second run saw no
oidc_code and fell through to the getAppConfig auto-redirect, firing
window.location.href = '/api/auth/oidc/login' before the exchange
could complete. Fixed by adding a useRef guard to prevent
double-execution and moving replaceState into the fetch callbacks so
the URL is only cleaned after the exchange resolves.
Also adds login.oidcLoggedOut translation key in all 14 languages to
show "You have been logged out" instead of the generic OIDC-only
message when landing on /login after an intentional logout.
Closes#491
- Exclude place-assigned reservations from timeline to prevent duplicate display
- Use selected day's date instead of today when entering time without date
- Pass day_id when updating reservations, not only when creating
- Clear reservation_time fields when switching booking type to hotel (#459)
- Parse date-only reservation_end_time correctly on edit (#455)
- Show end date on booking cards for date-only values (#455)
- Add auth token to file download links in bookings (#454)
- Account for timezone offsets in flight time validation (#456)
Bidirectional substring matching in isVisitedFeature caused unrelated
regions to be highlighted as visited (e.g. selecting Nordrhein-Westfalen
also marked Nord France due to "nord" being a substring match).
Replace the fuzzy loop with an additional exact check against the Natural
Earth name_en property to cover English-vs-native name mismatches.
Also fix Nominatim field priority to prefer state over county so
reverse-geocoded places resolve to the correct admin-1 level.
Adds integration tests ATLAS-009 through ATLAS-011 covering mark/unmark
region endpoints and user isolation.
Fixes#446
Timed places now auto-sort chronologically when a time is set.
Untimed places can be freely dragged between timed items.
Transports are inserted by time with per-day position override.
Fixes regression from multi-day spanning PR that removed timed/untimed split.
When a client disconnects mid-stream, headers are already sent so the
catch block now calls response.end() before returning, preventing the
socket from being left open and crashing the server. Fixes#445.
Reordering places on one day of a multi-day reservation no longer
affects the order on other days. Transport positions are now stored
per-day in a new reservation_day_positions table instead of a single
global day_plan_position on the reservation.
If the budget category stored in reservation metadata was deleted, the
form would re-submit it on next save, resurrecting the deleted category.
Now validates against live budget items on form init and falls back to
auto-generation when the stored category is gone.
Closes#442
If a user's last visited tab belongs to an addon that gets disabled while
they are away, re-opening the trip now resets the active tab to 'plan'
instead of rendering the inaccessible addon page.
Closes#441
The navigateTo function was clearing lightboxInfo without re-fetching it,
causing the EXIF sidebar to disappear and nav button placement to break.
Mirrors the fetch logic already present in the thumbnail click handler.
Fixes#439
Replace node-fetch v2 with Node 22's built-in fetch API across the entire server.
Add undici as an explicit dependency to provide the dispatcher API needed for
DNS pinning (SSRF rebinding prevention) in ssrfGuard.ts. All seven service files
that used a plain `import fetch from 'node-fetch'` are updated to use the global.
The ssrfGuard safeFetch/createPinnedAgent is rewritten as createPinnedDispatcher
using an undici Agent, with correct handling of the `all: true` lookup callback
required by Node 18+. The collabService dynamic require() and notifications agent
option are updated to use the dispatcher pattern. Test mocks are migrated from
vi.mock('node-fetch') to vi.stubGlobal('fetch'), and streaming test fixtures are
updated to use Web ReadableStream instead of Node Readable.
Fix several bugs in the Synology and Immich photo integrations:
- pipeAsset: guard against setting headers after stream has already started
- _getSynologySession: clear stale SID and re-login when decrypt_api_key returns null
instead of propagating success(null) downstream
- _requestSynologyApi: return retrySession error (not stale session) on retry failure;
also retry on error codes 106 (timeout) and 107 (duplicate login), not only 119
- searchSynologyPhotos: fix incorrect total field type (Synology list_item returns no
total); hasMore correctly uses allItems.length === limit
- _splitPackedSynologyId: validate cache_key format before use; callers return 400
- getImmichCredentials / _getSynologyCredentials: treat null from decrypt_api_key as
a missing-credentials condition rather than casting null to string
- Synology size param: enforce allowlist ['sm', 'm', 'xl'] per API documentation
Times like 09:00 were exported as YYYYMMDDTHHMM (13 chars) instead of
YYYYMMDDTHHMMSS (15 chars). Google Calendar couldn't parse the short
format and defaulted all events to 12:00 AM. Closes#432
- Add quantity field to packing items (persisted, visible per item)
- Bags are now renamable (click to edit in sidebar)
- Bags support multiple user assignments with avatar display
- New packing_bag_members table for multi-user bag ownership
- Save current packing list as reusable template
- Add bag members API endpoint (PUT /bags/:bagId/members)
- Migration 74: quantity on packing_items, user_id on packing_bags, packing_bag_members table
- Link budget items to reservations via reservation_id column
- Update budget entry when reservation price changes (not create duplicate)
- Delete budget entry when reservation price is cleared
- Sync price back to reservation when edited in budget panel
- Lock budget item name when linked to a reservation
- Add migration 73 for reservation_id on budget_items
- Add TREK description and "Made with heart" text to About tab (all 13 languages)
- Add Report Bug, Feature Request and Wiki cards to About tab and Admin GitHub panel
- Version shown as inline badge
- TRIP-002 now expects null dates and 7 placeholder days instead of forced date window
- Add TRIP-002b to verify custom day_count creates correct number of days
- Show day count input in trip form when no start/end date is set
- Backend accepts day_count param for create and update
- Remove forced date assignment for dateless trips (was always setting tomorrow + 7)
- Fix off-by-one: single-date fallback now creates 7 days instead of 8
- Add dayCount/dayCountHint translations for all 13 languages
GPX files containing both <wpt> and <trk> elements would only import
waypoints, silently discarding track geometry. The fallback chain only
parsed <trkpt> when no waypoints were found.
Replaced all regex-based XML parsing helpers with fast-xml-parser for
correctness (namespaces, CDATA, attribute ordering). Tracks are now
always parsed independently of waypoints, with each <trk> element
becoming its own place with route geometry. Fixes#427.
The search_place MCP tool was hardcoding a direct Nominatim call, ignoring
any configured Google Maps API key and never returning google_place_id despite
the tool description advertising it. Replace the inline fetch with the existing
searchPlaces() service which already switches between Google and Nominatim.
Update unit tests to mock mapsService instead of global fetch, and add a
dedicated test case for the Google path returning google_place_id.
Closes#424
