oidc_auto_redirect is managed-locked in the registry too, costs_mode is read
as the mode it is, the days of a dated trip are read through the factories,
and the dashboard's country order moves out of the atlas service into a
module of its own, so the service keeps its size.
The trips entity takes costs_mode as the generator writes it, the plugin
output contract publishes it, and the hand-built test schemas carry it. The
new MCP fields come from @trek/shared, the link tool tests join the collab
file that already mocks the database, oidc_auto_redirect is the operator's
on a managed install, and the tests follow the KML service, the dated trip
days and the zod pipe.
The place and endpoint reads behind the passport tile now carry the last day
of each started trip, and getTravelStats orders the countries by their latest
visit, with countries only marked by hand after them. The country set is
unchanged; repository, parity and service tests pin the order.
A new app setting oidc_auto_redirect (default on) is part of the admin form
and the public app config. Off, an instance without password login no longer
sends visitors straight to the IdP: the login page shows its SSO button,
also while the cached config is read. Admin toggle on desktop and phone.
Both feed routes and the one-off export.ics take ?days=0, validated by a
shared query contract, and then skip the all-day "Day N" event per day; the
default is unchanged. The subscribe dialog gets a Day overview events switch
that adds the parameter to its links, export_trip_ics takes dayEvents and the
feed tools explain the parameter.
list_collab_links, create_collab_link, update_collab_link and
delete_collab_link run the Links panel's REST routes as tools: the shared
body contracts, trip access, collab_edit on writes and the collab🔗*
broadcasts, behind the links sub-feature flag. Unit, e2e and broadcast
parity cases cover them; the wiki lists the tools.
dashboard_timezones joins the display-preference allow-list as a list of
IANA zone ids, each checked with Intl like the widget does, without repeats
and capped at 24. The write goes through saveMany, so open tabs refresh, and
both tool descriptions name the widget's fallback zones.
Google Maps takes no list from outside, but My Maps imports a KML layer.
The trip export dialog and the phone export sheet get "Places for Google
My Maps" (GET places/export.kml, MCP export_trip_kml), and a list's export
menu gets KML next to GPX. The KML codec gains a pin writer; the list file
mapping moves to collections.helpers to keep the service within its size.
A read tool that answers one list a page at a time (100 by default, up to
500) as compact one-line JSON rows, filtered by status, label or "no label",
with the list's labels, the matching total and a cursor. The cursor is the
last place's position (sort_order, created_at, id), so labelling a page
while paging through unlabelled places skips none. getCollection orders by
the same keys, id last, so both tools agree on ties.
The update contract takes google_place_id, google_ftid, osm_id, website,
phone, price and currency like saving does, website with the same
normalisation and refusal. REST and update_collection_place share it, so an
agent can correct a place instead of re-saving it and losing its labels,
ratings, position and id. The assistant's change card names the new fields.
A fourth status after visited, for a place already seen that the user wants
to go back to. It joins the one-tap cycle, the filters, the pickers, the
assistant cards and the MCP status tool, and counts as visited wherever
visits are counted (isVisitedCollectionStatus), so "visited everywhere"
keeps it. The status pickers go two by two where four would not fit.
Travellers who track what a trip costs but never split it saw an
Unfinished warning on every expense and a settle-up they had no use for.
A trip now has a costs mode (appended migration, trips.costs_mode, split
by default); totals leaves payers, shares and settling up out of Costs on
desktop and phone through the same gate as a trip of one, and saves new
expenses without a payer. PUT /api/trips/:id and update_trip take it on
budget_edit. The split chips move into their own component.
nativeApi goes to 3 with 5.1.4, whose UI follows a widget or Live Activity
tap that starts the app. The apps refuse an older server with the hint to
update, and check a server they remember once more.
The address and the key it may no longer carry now change together, as
they do for Dawarich, so the transaction ratchet drops the AirTrail entry.
The tests that pinned the old no-transaction shape now pin the new one.
Saving the AirTrail settings with a different URL and the key field left
masked or blank kept the stored key, so the next sync sent the old
instance's key to the new host. The key now goes with the host, as it
already did for the connection test and for Dawarich: a new origin without
a new key clears it, a path edit on the same origin keeps it. The key is
cleared before the new address is written.
- The SSO unlink answers with a response contract (adminSsoUnlinkResponseSchema).
- It reuses linkOidcIdentity with nulls instead of a new repository method,
so UsersRepository keeps its method count and size.
- The client calls the route through apiClient, so api/client.ts does not
grow, and its confirm dialog takes no async callback.
- The pager's preview step and pagesFor keep their complexity.
A sign-in with another subject of the same provider is refused, so an
account whose provider re-issued the subject (a migrated or rebuilt IdP, a
changed sub format) was locked out: editing the email, as the wiki advised,
never touched the stored link. The edit dialog now offers "Remove SSO link"
for other accounts while SSO is set up (DELETE /api/admin/users/:id/sso,
audit admin.user_sso_unlink). The next SSO sign-in links again under the
rules of a first one, so it still needs a verified, not self-set email.
The log line names the remedy, and the wiki tells the two refusals apart.
Fixes#3065
Sonnet 5.5 hands back the words written before a tool call as a hidden thinking block. After the ask for the answer, an empty step got the general push with tools open again; the model answered, offered follow-ups once more, and the answer was lost a second time. That push now asks for the answer as text and stays text-only.
DayPlanSidebar grew past its size baseline with the Firefox drag fix. Its drag state (what is dragged, the drop line, the day under the drag, whether a drag happened) and the effect that ends a lost drag move into useDayPlanDragState, typed, which also lowers the file's strict-mode count; size and strict baselines are lowered with it.
vacay.service.test.ts went past the 2000-line limit once the Nager v4 cases were back in it. They move into vacay.holiday-provider.test.ts, which takes VacayService from a Nest testing module over a snapshot db, so neither the db mock nor a hand-built service is needed.
- iOS: the navigation proxy reports an error page through a callback the controller sets; it called the controller's method on itself and did not build.
- keyboard: the done key reads the field's text in the capture phase, before the field's own handler, so a list that adds an item on Enter and clears itself keeps the keyboard; the tests wire it as the app does.
- Ireland migration: the queued writes use addSql instead of awaiting inside loops; only table constants and integer ids reach the SQL.
- tests: the Dawarich GL test types its two layer specs once (strict baseline lowered), the collab test reads the id off the listed message, the reasoning test runs its four servers in parallel, and a wasm copy covers the precompressed branch without a charset.
- en wording: the deviations agreement no longer calls TREK self-hosted.
geoBoundaries has no county layer for Ireland: its ADM1 is the four
provinces and its ADM2 the 166 local electoral areas, so the Atlas showed
Ireland as four provinces next to a UK of counties.
The Atlas builder can now take a country's regions from another source.
Ireland's come from Tailte Éireann's 26 counties (CC BY 4.0), with Dublin,
Cork, Galway and Limerick each one county, under their ISO 3166-2:IE
codes. Those outlines come at full detail, so the builder simplifies them
itself and keeps the borders two counties share identical on both sides,
so no point between them falls into neither. ATLAS_ONLY rebuilds just the
named countries without downloading the whole world.
A migration clears Ireland's cached place regions so they are worked out
again as counties, moves visited rows that name a county onto its code,
and hides each county of a province a user had hidden. Provinces a user
marked as visited are kept as they are.
Fixes#2922
Azure AI Foundry reasoning deployments behind an "Own server" connection
reject function tools on /v1/chat/completions at their default reasoning
effort and answer HTTP 400 naming reasoning_effort. The Responses API
switch from #2792 only covers direct OpenAI keys, so the assistant could
not run on these deployments at all.
The chat client still never sends reasoning_effort, so servers that do not
know the field (Ollama, LM Studio, vLLM, OpenRouter) see unchanged
requests. When a request with tools is refused over reasoning_effort, it
goes out once more with reasoning_effort "none", and the rest of the turn
keeps sending it. The refused field is now carried on AiProviderError
instead of only being logged.
Fixes#3043
The two new test files mocked the db module and built VacayService by hand,
which the test-mocks and new-service ratchets refuse for files without a
baseline entry. The cases move back into vacay.service.test.ts and
vacay.test.ts unchanged; VACAY-037 asks for a year no other case caches. The
unrelated admin addon test change is reverted and two comments name the new
provider.
A multi-day hike restocks on the way: bread in the morning, food for
lunch and supper every day or two. Neither could be searched for.
- Explore pill: new `bakery` and `supermarket` chips. `supermarket`
already existed on the server (shop=supermarket|convenience) but no
pill offered it; `bakery` is new (shop=bakery).
- Along the tour: `bakery` and `supermarket` are offered to a hike,
after its existing kinds so a new hike still starts on the same three.
- TREK places index: `bakery` maps to food_and_drink>casual_eatery>bakery;
`supermarket` is narrowed to grocery and convenience stores, since the
whole food_and_beverage_store branch also brought tobacconists and
wine shops. Both paths checked against a live answer of the index.
As the more specific term, a bakery is no longer listed as a place to
eat when only `food` is asked along a tour.
- Category guess: a bakery or grocer picked from the map was preselected
as Bar/Cafe or Shopping even where the admin had made a Bakery or
Groceries category. The guess now looks for such a category first, by
name in the usual languages or by icon, and keeps the broad one when
there is none. The category icon picker offers Croissant and
ShoppingBasket, the icons the map chips use.
- Labels in every locale; MCP tool description names the new kinds.
A point just off a simplified coastline lies in no admin0 polygon, and the
fallback then took the candidate with the smallest bounding box. Nice and
Menton sit ~80 m outside France's coast and 20+ km from Italy, but Italy's
box is the smaller one, so they resolved to IT. The fallback now ranks the
candidates by distance to their border. A deferred loose box (PS/XK) keeps
its priority. The parity test's reference lookup mirrors the new rule.
Closes#3048
The places list and the trip bundle resolve the country of every place
without a cached region on each read, and that lookup was about 90% of
their CPU time: it ray-cast every polygon of each candidate country, which
is thousands of rings for a country that is mostly islands.
A polygon whose outer ring's box does not hold the point is now skipped
without walking it, and an answer is kept by coordinate pair (20,000 pairs,
oldest out first), since the borders never change while the process runs.
The answers are the ones the ray-cast alone gives; ATLAS-SVC-051 and
ATLAS-SVC-055 compare every country against the reference lookup.
Measured on a production build with a 360-place trip: GET places
22.7 ms -> 3.5 ms, GET bundle 27.5 ms -> 6.2 ms. A first lookup of a
pair is about 1.8 times faster (90 us -> 50 us over 20 cities).
Every file under assets/ was compressed again on each download, at the low
level a request path can afford. They carry their content hash in the name
and never change, so the client build now writes a Brotli copy beside each
compressible one (postbuild, scripts/precompress.mjs) and production points
a browser that accepts Brotli at that copy before express.static serves it.
Measured on a production build of 5.1.2, fetching the 652 precached files
the way a service worker install does: 16.3 MB -> 13.9 MB on the wire and
about 1.0 s -> 0.2 s of server CPU per install. A client without Brotli, a
range request and a build without copies are served exactly as before.
From Node 25 the permission model covers the network, so a plugin child
forked with --permission and no --allow-net gets ERR_ACCESS_DENIED on every
lookup and connect, and no plugin can reach a host it was granted. Node 24
is unaffected and rejects the flag as a bad option, so it is passed only
where process.allowedNodeEnvironmentFlags knows it.
Which hosts a plugin may reach is still decided by the egress guard inside
the child.
Fixes#3045
Deleting a message set the flag and kept the text in the row and in every backup; only the client hid it. The delete now blanks the text with the flag, and a migration clears what earlier deletes left. The row stays for the placeholder.
A partial update that left day_id out (a notes-only PUT, an MCP
update_reservation or update_transport call) re-derived the day from the
stored reservation_time whenever one was set. A time-only value such as
"12:00" resolves to no day, so the booking's day_id was written as null and
it vanished from its day in the planner. The end_day_id branch did the same
when a time-only reservation_end_time was sent.
The day is now derived only when the time carries a date that resolves to a
trip day; otherwise an omitted day_id or end_day_id keeps the stored value.
A dated time still re-pins the day, an explicit day_id null still clears it,
and a cleared end time still clears end_day_id. REST and both MCP update
tools share updateInTx, so the fix covers all three.
Tests: service unit cases (RESV-SVC-044..048), an e2e case per behaviour
on PUT /api/trips/:tripId/reservations/:id, and an update_reservation MCP
case, all seeded with the ORM factories.
Fixes#3025
Both files had grown past their size limit with this round's fixes. The
two-factor step moves to pages/login/MfaCodeStep.tsx and the reservation row
types to reservation-rows.ts, re-exported from the repository, and both size
baselines come down to match.
The edit form fell back to the linked stay's times, the cards read only the
booking's metadata, so a hotel showed 15:00 and 11:00 when edited and no times
in the list. The reservation list now carries the stay's check-in window and
check-out, and the desktop facts and the phone card fall back to them. On the
phone they come first, so a price or a seat cannot push them off the card.
The fixes that keep the app's sign-in in the app live in the server's UI,
so an updated app on a 5.1.2 server still ended in the browser. The server now
reports nativeApi 2 and the app refuses anything below it, naming 5.1.3.
A server the app remembers from before is not loaded unchecked: the first
launch after the update shows the address screen filled in with it, says why
and checks it right away. A server that is new enough loads again with its
session; an older one gets the update hint.
The iOS app rewrites raster tiles to trek-tile:// and answers them from its
store. WebKit checks that scheme against img-src, which had only https:, so
every tile was refused and the app showed routes and markers on a blank map.
The home assistant wrote its answer and offered the trip's own chat in the
same step. The loop asked the model once more after the card, and it wrote
the same answer a second time below it. A step that has its answer and only
hands the conversation on now finishes the turn; a hand-on without an answer,
or beside a read, still asks the model again.
From #3032, adjusted where it broke:
- react and react-dom are pinned in the root overrides; the group raised the
workspaces to 19.3.1 and left the override at 19.3.0, so two react-dom
versions met in every client test. The override moves along.
- @mikro-orm/nestjs kept the old @mikro-orm/core at the root while the server
got 7.2.8 of its own, two copies whose types do not match; the lockfile now
holds one 7.2.8.
- mapbox-gl stays at 3.31: 3.34 drops setRTLTextPlugin from its types, which
the map engine uses.
- the musl pins of rollup stay at 4.62, the rollup version the lockfile holds.
@capacitor/cli 8.6 drops the old uuid 7 through its xcode dependency (#3012).
vite-plugin-pwa 2 keeps the registerSW.js in public/ in place of its own, as
1.x did (#3033). @nestjs/schedule 12 still runs on Nest 11 (#3035).
A browser that ran TREK before 5.1.1 keeps that version's app in its service
worker. When the app opened /native-handoff, the worker answered it from its
cache with an app that has no such page, which sent the user to /, and the
sign-in ended on the dashboard in the browser instead of in the app.
The app now opens /api/auth/native/start, which every TREK service worker
passes to the server. That page takes the workers off the browser and
forwards to the handoff page. Until the sign-in is handed back, the web app
registers no service worker and does not reload for a new version: a tab
still open on the older TREK keeps its worker alive, and a registration or a
reload in the middle of the sign-in brought it back to answer the next page.
Each photo's write follows its own network lookup, the repair's marker only
records that the pass ran, and a lookup may renew the Synology session on
its own; none of these can share one transaction.
The repair job wrote its marker rows through the AppSettings repository,
which the settings domain owns. It now registers both keys and goes through
SettingsService, as the import boundary check asks.
Maurice