HIK-01. The details of the hiking planner gain an optional Winter activity
switch, off by default, so the normal hiking flow asks for nothing more.
Switched on, the hike is a winter hike or a snowshoe hike; the choice stays
dimmed and out of reach while the switch is off. The label is saved with the
tour, comes back when it is opened, survives a draft reload, a copy and a
take-over, and switching it off takes only the label away: the route, its
grade and the days stay as they were.
The label is a label only. Routing, the walking time and the 3D terrain view
work as for any hike, and the info dot and the head pill's tooltip say that
nothing was checked for snow, avalanches or whether a path can be walked in
winter. The saved label shows as a frosted snowflake on the corner of the
tour's tile in the tour list and the day plan, and as a pill beside the type
in the tour's detail.
The server keeps it in tours.winter_activity (CHECK list, NULL for every tour
saved before and for every other type), refuses it on a bike tour, a city walk
and a paddle tour, and keeps the stored label when an older client saves a
hike without the field. A snowshoe hike exports to GPX as snowshoeing.
list_tours names the field, with the same words on what it does not say.
A search for sights along a tour now also asks the places index for the best
known places near the line, ranked by how often their Wikipedia articles are
read (QRank). A row the index already found is marked as a highlight; a
famous sight no kind names, like a square, is added as a sight of its own.
Stations, hotels and restaurants rank a row they match but are never added.
Highlights are listed first with a star in the list and a badge in the map
card. A place counts as a highlight from 40000 views a year, or the best
three from 5000 where fewer reach that. An index without ranks, or one older
than the endpoint, leaves the search as it was.
An assistant connected over MCP can now prepare a city walk from what the
traveller tells it. propose_city_walk takes a start (and an end), a time
frame on a day of the trip, the themes of CW-05, the stations the
traveller wishes for, the places of the trip it may use and fixed
appointments with their times, and saves nothing. Places for the themes
come from the search along a tour, at most eight, each with the theme,
category and source that brought it; nothing is made up. The stations are
ordered on estimates over the straight distance, and a place of a theme or
of the trip goes in only where it is open on arrival, keeps the frame and
the appointments, and leaves every station already on the walk open when
the walk gets there. The walk in that order is then planned on foot by the
router the road trip uses (the traveller's own routing server where they
set one), and its legs give the walking distance and time from one station
to the next. Where the router cannot be reached, or a station lies more
than 300 m off every way (the planner's limit, now shared with the
server), the times are estimates over the straight distance with an
allowance for streets at 4.5 km/h, and the answer says so and why. The
answer gives the arrival and leave times, the stay where the traveller
gave it and unknown otherwise, which makes every later arrival the
earliest one, and the opening hours checked against the arrival on that
day with the helpers the planner uses. Walking and stays are counted
apart, the total is complete only when every stay is known, and the
minutes the frame leaves for the unknown stays are named. Free time before
an appointment or at the end, wishes and appointments that cannot be kept
as asked, and places left out with their reason are listed; a tour of the
trip offered as a station is named as a tour. A wish or an appointment is
never dropped; an appointment is a must and protected. A place's default
duration is not read as a stay: every place carries one, so it says
nothing.
Once the traveller confirms, create_city_walk saves the walk through the
same creation the planner's save uses: one write, a key for every station,
the links to the trip's places checked, the trip told, with place editing
permission on the trip. Its way on foot is planned by the same router and
saved with the walk, together with its distance and walking time. Where
the router gives no way, the walk is saved with its stations only and the
answer says why: no line, no distance and no walking time. The tours list
says so (route_pending), the tour detail says in one line that the way is
planned in the planner, and opening the walk in the planner plans its way
at once and offers to save it, with a note once it is planned that it was
just planned. Such a walk can be saved as a new tour and taken over into
another trip, starting at its first stop, and is never offered as a place
to stay. A tour without a line needs two points of its own.
get_tour_plan now spells out every field of a city walk's stations: key,
name, stay (null when not known), importance (null for normal), protection
(false when not set) and opening hours (null when none are known);
list_tours names the station count and whether a walk waits for its way,
get_city_walk_progress gives each station's key and opening hours, and
take_over_tour says a walk keeps its stations with their protection. No
tool changes a walk's stations; deleting a walk's place takes its stations
along, protected ones included, and both delete tools now ask the
assistant to name those and ask the traveller first. REST answers the
stations as before.
A tour create refused for a type that is switched off, a description that
is too long or planned times out of range is a domain refusal now, so an
assistant gets the same text REST sends; REST still answers it as an error
at 400.
"Add tour" in the planner now offers "Import walk from Google Maps". The
dialog takes the link of a route shared in Google Maps, either a
maps.app.goo.gl short link or the long google.com/maps/dir link, or the
share text of the Google Maps app with the one link in it. It shows the
stops in their order with their names, marks a stop the link gives no
position as such, says that the route comes from Google Maps with a link to
open it there, and that TREK plans the way between the stops itself and
does not take the line drawn in Google Maps over. Nothing is written until
"Load as walk": that opens a new, unsaved city walk in the planner with the
positioned stops and their names, routed on foot by TREK, and warns first
when it replaces unsaved work. A stop without a position is left out and
never looked up or guessed; with fewer than two positions nothing can be
loaded. While the walk is not saved, a calm notice in the planner says
where it came from, also after a reload of the draft. Cancelling or a
failure leaves every tour as it was.
Only a route the link marks unambiguously as on foot is read. A car, bike
or transit route, a link that does not say, or one that says two things is
refused with a reason the dialog gives in the reader's language. Each stop
takes its position from its own block of the link's data, so a line
dragged through extra points or a stop given as coordinates no longer
leaves the stops without positions or shifts them. The road trip and
places imports read links as before.
A short link is followed by a resolver of its own: every hop is checked
against the Google Maps short link shapes and the strict SSRF guard before
it is looked up or asked, so ALLOW_INTERNAL_NETWORK and
ALLOW_LINK_LOCAL_IPS do not open it. Each request goes to the vetted
addresses only, follows nothing by itself, reads no body and carries no
credentials; the chain has one deadline including name lookups, at most
three hops and a cap on the Location header, and the long link it ends on
must be Google Maps and is never fetched. The preview is POST
/api/trips/:tripId/tours/import/google-maps/preview for members who may
edit places, with a 502 when Google could not be reached, and the MCP tool
preview_google_maps_walk reads the same.
The day row on the phone now shows how many stops of a city walk were
visited on that day, "2 of 5 visited", like the desktop row does. Both rows
read it from one place, and a tap on the phone names it in full without
opening the row.
A walk's stops can be checked off, skipped and taken back without a
connection. The mark shows at once, waits in the offline queue and goes out
when the connection is back, in the order the marks were made and behind any
older mark of the same day that still waits. The day plan, the walk's detail
and a reload read the visits kept on this device with the waiting marks on
top, so a mark made offline stays on screen until the server has it. A mark
the server answers with "not found", because the stop, the day or the walk
was removed meanwhile, counts as done and holds back no later mark of that
day. Any other refusal, the Tours addon turned off among them, keeps the mark
as a failed change to try again from the offline settings. The visits are
read again when the connection comes and goes, and once more after the
waiting marks went out, so a mark the server dropped or refused leaves the
day plan. A trip's visits that answer after another trip was opened no longer
show there.
The offline sync keeps each trip's walk visits and reads the stops of a walk
this device has never opened, so it can be checked off offline too. A walk
whose stops this device could not read lists no stand-in stops and no count,
and says to open it again while online. A walk put on a day while offline
says its stops can be checked off once it has synced, instead of offering
buttons that cannot work yet. A mark that fails says to open the walk again
and try once more, in every language.
Marks carry no version, so the last one to reach the server wins, as with
the other changes sent after being offline. The offline guide says so.
The search along a tour found nothing. It asked the places index for all
kinds of a stretch at once, above the 100 rows and 200 characters of terms
the index takes, so the index refused and every stretch went to the public
Overpass mirrors, which do not answer from this server. Now the index is
asked in small circles strung along the line, one request per kind and
circle within its limits, and an empty circle counts as an answer, so a
city walk lists its sights, museums, parks, cafes, restaurants and markets
in well under a second. The kinds the index does not know (viewpoints,
huts, drinking water, toilets and the like) are asked of Overpass along the
whole line at the same time, one query per stretch of up to 50 km, three at
a time, each with a short deadline of its own; a line past 400 km is
searched along its start and says so. When the index answers no circle at
all, its kinds go to Overpass beside the others instead of after them. A
kind no source could be searched for is named in a calm notice above the
list instead of reporting nothing found, and the reason is written to the
log. Unnamed taps, toilets, shelters, picnic spots, viewpoints, summits,
slipways and repair stands stay in the list and on the map under the name
of their kind; an unnamed garden, cafe or memorial is left out. The places
nearest the line are kept when there are more than sixty, listed from the
start, with the opening hours the source has.
A walk's parks are parks and gardens only, and its museums are museums
and galleries only, no longer the beaches, summits, theatres and cinemas of
the map's explore buttons; a water park the index files under parks is not
one of them. The index also answers markets and summits. The explore
buttons keep their single request, and a search of several kinds that
would exceed the index limits is asked one kind at a time instead of being
refused. The MCP tool search_tour_pois and its wiki row describe the walk
kinds and the new unavailable list, and the OVERPASS_TIMEOUT_MS row says
the search along a tour keeps its own shorter deadline.
The list at the top of test.yml now names every step group its jobs hold,
the contributing guide names the parity job as it is called now, and
db-mock.ts says test-db reaches the auth controller only when a rate limit
is reset.
The test jobs waited for the quality jobs (~9 min) and the client suite ran
on one runner (20-35 min), so a run took 45-60 min before ci-ok answered.
Every test job now starts right after changes. The client suite runs in
eight slices and the server suite in six, each with coverage into a blob
report; client-coverage and server-coverage merge the blobs, check the
thresholds against the whole and upload the lcov the scan reads. The scan
needs the gates directly, so a failed gate still stops it.
POST /api/trips/:tripId/tours/poi-search looks for what a hike or a paddle
tour stops for along a line: viewpoints, peaks, shelters, drinking water,
food, toilets, picnic spots, campsites and sights, and on the water
slipways and boat rental. The line comes in the body, so a draft not yet
saved can be searched, thinned to at most 1000 points; the corridor is
50 m to 2 km wide.
The line is cut into stretches as the road trip's corridor search does,
each searched through the map search (the TREK places index first,
OpenStreetMap after), three at a time; a long line is searched along its
first twelve stretches and the answer says so (capped). The places kept
lie within the corridor of the line itself, each once, nearest the start
first, up to 60, each with its source, kind, address, website, phone and
opening hours as the source has them, its distance straight across to the
line (distance_m, never called a detour) and how far along it lies. A
stretch that fails leaves the others' places and marks the answer partial;
a cut answer is truncated. Nothing is written. It is a read, so anyone who
sees the trip may ask.
The map search's OSM filters gain those outdoor categories; the explore
pill does not offer them. MCP: search_tour_pois, along a saved tour's line.
GET /api/tours/library lists the tours of every trip the user owns or is a
member of, with the trip each is in and how many tours each trip holds, to
choose one to take over into the trip being planned. It is narrowed by a
part of the name (matched ignoring case, a "%" or "_" just a character), a
type or one trip, and the asking trip can be left out; up to 100 tours, and
the answer says when there were more. GET /api/tours/library/:placeId
previews one: its facts and its line thinned to 400 points, its ends kept.
Neither is trip-scoped, so every answer is bounded by the user's own trips,
read anew each call: a tour of a trip out of reach is "not found", nothing
hints that it exists. Measures stay null where the tour has none.
POST /api/trips/:tripId/tours/take-over copies one into this trip, which the
guard checks with 'place_edit'; the source trip is checked against the
user's trips as they are at that moment, and the source is read again inside
the write. The copy is a tour of its own: its line, measures, settings and
control points, without days, without its stops' links to places of the
other trip and without the obstacle decisions taken there; it keeps the
import record of its line. A copy that fails leaves nothing behind.
Where it came from is kept as text (tours.taken_over_from_trip_title and
taken_over_from_tour_name, migration 045100), not as a pointer, and read
back in the provenance as takenOverFrom; a copy of such a copy names it too.
The in-trip copy (P2C-07) and the take-over share one copy routine.
MCP: list_tour_library and take_over_tour, over the same service and
checks. The repository tests follow the hiking preferences column of MF-08.
A hike can ask the router for more than its difficulty limit: avoid steps,
avoid ferries, prefer footpaths, shorter over faster. They sit under the
limit in the planner's details as switches, each with what it really does,
and the info dot says they weigh the way and give way where there is no
other, and that none grades a path. Switching one plans the route again as
one step of the history, and the choice is kept in the draft.
Only preferences the public Valhalla was seen to honour are offered, with
the measured cases beside hikingCostingOptions: step_penalty took Würzburg
to the Festung from four flights to one, use_ferry 0 sent Kiel round the
Förde and Hamburg through the Elbe tunnel, walkway_factor moved Garmisch to
the Partnachklamm onto the paths, shortest cut the way to the Eckbauer.
use_hills and use_tracks showed no effect on foot and are left out, as are
marked, paved and rough ways, for which the router has no option.
tours.hiking_preferences (migration 045000) holds them as JSON, NULL for
none, so no tour saved before routes differently. Only a hike keeps them;
a paddle tour naming one is refused. A copy carries them, and list_tours
and get_tour_plan read them.
The trip copy's waypoint links now follow the copied places inside
TourWaypointsRepository.insertForPlace.
A saved tour now has a technical revision, raised inside the transaction of
every write that changes it: the route and its controls, the plan laid onto a
recorded paddle track, going back to that track. A write that names the
revision it was based on (expected_revision) is refused with a 409
{ code: 'tour_revision_conflict', revision } before anything of it is
written, so a stale draft never overwrites a newer tour unseen. A write that
names none writes as before, so older clients and queued offline writes keep
working; the queue never carries a base. A copy starts at 1, a source's
version stays provenance. Migration 044800 adds the column; list_tours and
get_tour_plan read it.
The editor follows the trip's tours live: a newer revision of the open tour
loads at once while the editor has no changes of its own, and shows a notice
while it has. A refused save keeps the draft and asks: save it as a new tour,
load the newer version, or overwrite on purpose.
useTourPlanner was at the size limit; its derived facts, undo history and
draft storage move into tourPlannerFacts, usePlannerHistory and
useTourDraftStorage, unchanged in behaviour.
A saved tour can be handed over as a GPX file on purpose: its whole saved
line as one track with the heights it has and none it lacks, its name,
description and link, and the activity in <type>. The file is named after
the tour. A tour without a line of two points is refused with 422 rather
than written as a file that imports as nothing; the export changes nothing.
GET /api/trips/:tripId/tours/:placeId/export.gpx, readable by anyone who
sees the trip, and the MCP tool export_tour_gpx with the same rules. The
tour detail gets a "GPX" action beside the navigation, the planner head a
download button that waits while the tour has unsaved changes.
The tours rail head now uses the same plain icon buttons as the editor.
Paddle tours are planned on the water by TREK's own router: a water network
per paddle region, built from the same Geofabrik extract as the obstacles and
searched in a worker thread of the server (no sidecar). Rivers, canals and
lakes (direct or along the shore), portages round weirs, locks by policy,
whitewater by boat, gaps marked. Found portages and locks become points of
their own; an unplannable leg says why and offers a way out.
- Shaping points: a click on the line sets a point that steers the route
without being a stop (hikes and paddle tours); recompute a single leg.
- Lake style direct/shore, suggested from the wind forecast.
- Regions load by themselves where a paddler plans (admin setting: automatic,
on request, off; size cap and a daily count), with requests for the admins,
a notification, and the load followed on the leg.
- Portage from the last point, portages follow footways via the walking
router (or stay straight), "put in here" on slipways and put-ins.
- MCP plan_paddle_route with the REST route's rules.
The OpenAI-compatible client always asked for 4096 tokens, and a document
whose bookings needed more came back as "no reservations" with nothing that
pointed at the cap. LLM_MAX_OUTPUT_TOKENS now sets the cap for every
provider; unset keeps 4096 and 8192 so a small local model's context still
fits. A finish_reason of "length" is reported like Anthropic's max_tokens
stop, naming the limit and the variable that raises it.
Closes#2677
How an admin adds a region from Geofabrik's list, what the card shows while it is read, the monthly refresh on the 3rd, what a failed read keeps, and the OpenStreetMap attribution under the ODbL; a record in the data is no check on site.
- Paddle as a plannable tour type: one boat (kayak, canoe, SUP, other), river or
lake, a pace that a boat change never overwrites; time is distance over pace
- Hand-drawn legs beside routed ones, drawn dotted and labelled as not routed,
kept through save, reopen and edits elsewhere
- Reverse a tour and close it back to its start, each one undo step
- GPX import shows a preview first (line, measures, duplicates, type) and only
the confirmation stores the tour
- Base maps and overlays per tour type in the planner: hike keeps topo and the
via ferrata layer, paddle gets the OpenSeaMap seamark overlay
The column showed the last sign-in, which a renewing cookie leaves weeks
behind for someone using TREK daily. It now shows the later of the sign-in
and the last request. Saving an edit no longer blanks the row's avatar,
activity and online dot, and the list refreshes every 30 seconds while open.
Cloud models got the first 4,000 characters, so totals and lodging sections
further down never reached them, without a word. They now get 24,000 with the
start and the end kept and a warning when the middle is left out; a server
whose context is too small is asked once more with 8,000. A total the model
repeated on every flight of one booking stays on the first flight only, when
the document prints it once as a total, and the prompt says so too.
Journeys from the transit search are routing, not bookings, and a day of them
pushed the real bookings out of the six slots. Booked trains and buses keep
their own types and stay. The MCP tool reads the same feed.
The OIDC page explains when a login by email links an account, what
email_link_blocked means and how an admin lets a blocked account link by
entering its address in the admin panel. The audit log page lists
user.email_change.
A guest's payer rows and recorded payments hung on the users row by ON DELETE
CASCADE, so deleting the guest wiped money that had really moved: expenses
they paid were left without a payer, settled payments reopened, and a custom
or receipt split lost the guest's share so the balances no longer summed to
zero.
Deleting a guest who paid for an expense or is part of a recorded payment is
now refused with a 409 (REST and the delete_trip_guest MCP tool, same text),
and the share dialog says to change the payer or delete the payment first.
Otherwise the guest's share is handed on: an equal split re-splits as before,
a custom amount goes to the others by largest remainder in the currency's
smallest step (members without an amount get one, so the expense stays a
custom split), and a receipt drops the guest from its lines and is split
again like the cost editor does. A legacy paid_by_user_id is cleared as on
account deletion. Account deletion goes through the same re-split. The
confirmation text in all languages and the wiki say what really happens.
Refs #2604
The undo wrote a new place from nine of its fields and put it on its days as a
bare stop, so notes, website, phone, hours, picture, the visit times, day notes
and travellers were gone while the toast said it was undone. It now writes every
field the place had and plans each visit again with its times, note, end of day,
route flag, travellers and leg modes, and casts the own rating again.
The server takes a booked night, its booking and expense, linked expenses, file
links, other members' ratings and an uploaded picture down with the place, and
writing the place again brings none of that back. Such a delete is no longer
offered for undo. Undo steps that name a deleted place are dropped at the delete
instead of failing later, and adding a place, moving categories and the phone
reorder and optimize now carry their place ids for that.
Taking a stop off a day and clearing a day undo with the visit details too. The
stays are read again when a deleted place had one, here and for collaborators.
Refs #2604
API keys were split off from MCP to work without any addon, but the tab
that holds them still hid unless a photo provider, MCP, AirTrail, AI
Parsing or Dawarich was enabled. On a fresh install all of those are off,
so there was no way to create a key. The desktop tab now always shows;
the addon backed sections inside it keep their own gates. The phone layout
has no API Keys section and keeps its gate.
The wiki now says where keys are created and that no addon is needed.
Refs #2561
Marking a region wrote its country with source 'manual', the same as ticking
the country off by hand, and removing the last region then took the country
away whatever its origin, Dawarich confirmations included. A country that
only comes in through a region is now stored as 'region', marking it visited
afterwards turns it into the user's own mark, and the cascade in unmarkRegion
only runs for a country without such a mark. Readers still see 'manual' for
those rows, so the country detail, the MCP resource and the plugin RPC answer
as before. Rows written before this change stay 'manual' and keep their
country.
REST, MCP and plugin RPC share the service. The atlas popup on desktop and
phone now asks the server whether the country is still marked before it
drops it from the map instead of guessing.
Refs #2597
NODE_ENV="production " passed the boot check but matched none of the
exact production comparisons, so the web UI was never served and every
page answered {"error":"Cannot GET /"}, while the banner showed a clean
"production". The boot import now trims these three variables in
process.env before anything else reads them, so Express, the plugin
children and readEnv() agree, and reports each change as a process
warning. Casing stays as it was. A padded "test" is only reported, so an
instance cannot switch to an in-memory database after an update. The
banner prints NODE_ENV as set instead of lowercasing it.
Refs #2565
Moving a saved place to another list takes it out of the source list, but
updatePlace only asked for edit rights there, so an editor could empty a
shared list they may not delete from. The source now takes owner or admin,
like a delete, and the target still takes edit rights. REST and the
update_collection_place tool share the service path. The Move to list
action on desktop and phone follows the delete permission.
theme:lint only reported a total and exited 0, skipped whole files by name and
never ran in CI. It now counts per file under src/ against
scripts/theme-baseline.json, which only shrinks, like lint:rtl: the colour
literals, arbitrary colour classes and numeric font sizes it already saw, plus
palette classes (bg-gray-100, text-indigo-600, bg-white), raw text sizes
(text-sm, text-[13px]) instead of the type tiers, z-index literals above z-50
instead of a step of the layering scale, and reads of the dark_mode setting.
The file exemptions are gone; the map, PDF and brand files carry their count in
the baseline instead. theme-lint-disable markers stay valid in a comment and
are counted against scripts/theme-disable-baseline.json, like rtl-lint-disable
in lint:rtl; a new one raises its file's entry by hand, where review sees it.
The comment handling lint:rtl already had moves into scripts/lib/source.mjs so
both read source the same way.
The check runs in the client-quality job.
server/CLAUDE.md now says which API new repository code uses: Kysely for join, aggregate and set-based statements, the entity methods for single-table CRUD, no new QueryBuilder calls. The TrekRepository docstring no longer calls Kysely an escape hatch. lint:query-api counts this.qb()/createQueryBuilder() per repository file against a baseline that may only shrink and runs in the server CI job.
The same file documents the JSON column and timestamp rules from the two changes before.
tsconfig.strict.json is the build config with strictNullChecks and
noImplicitAny on. lint:strict runs tsc over it and holds the errors per file
to scripts/strict-baseline.json (204 in 109 files today): a file may only
lose errors, a new file starts at none, and an entry above its count fails
until --update lowers it. An error outside any file or an unreadable tsc run
fails closed. It runs in the server job; the build itself stays on
strict: false.
shared's lint ran eslint --fix with prettier/prettier as an error, so in
lint-prettier.yml the format:check after it only ever saw files it had just
rewritten, and the server's format scripts pointed at a test/ directory that
does not exist. Now lint is check-only in both (lint:fix applies fixes), the
server scripts cover src/ and tests/, and a lint:format ratchet on the shared
library holds every file outside scripts/format-baseline.json to Prettier in
both workspaces: in the server job of test.yml and in lint-prettier.yml,
which keeps providing the lint check. Formatting leaves ESLint in shared, and
the unused eslint-plugin-prettier goes. Once the tree is reformatted in one
commit, --update empties the baselines and the check is a full prettier
--check.
Server lint ran without --max-warnings or a baseline, so the rules left on
warn could grow without limit. lint:warnings runs ESLint with the server's
own config through the shared tally, fails on any error or inline rule
config, and holds the warnings per rule for src, tests and eslint-disable'd
messages to scripts/eslint-baseline.json, which only falls. It replaces
lint:check in CI. no-await-in-loop, no-non-null-assertion and no-console come
in as warnings held at today's counts. npm run lint no longer rewrites files;
lint:fix does.
server/src/nest/README.md drops from 988 lines to under 200: the module
shape with weather as the reference, the cross-cutting pieces, the guards,
parity and the test layout, without the bridge playbook and the services
recipe that contradicted the code. The migration log moves verbatim to
server/docs/nest-migration-history.md, marked historical.
scripts/ci/doc-paths.mjs resolves every backtick path in the CLAUDE.md files,
the root README, the blueprint and the wiki against the tracked files. The
paths already missing are listed in a baseline that only shrinks, and the
changes job runs it so docs-only changes are checked too. The wiki no longer
says the Playwright flows stay out of CI.
Most features have a desktop view and a phone view, and until now each
carried its own copy of the feature's logic, so a fix on one side drifted
away from the other. With the logic moved into one shared hook per
feature, scripts/feature-pairs.json names every feature with its hooks
and the desktop and phone files that run on them.
lint:pairs fails when a listed view stops importing one of its pair's
hooks (a type only import does not count), when a listed file is gone,
when a hook sits in a view layer rather than under components, hooks,
utils or the folder of a page under pages/, and when a file under
src/mobile/screens is in no pair and not under mobileOnly with the
reason it has no twin. A new phone screen is so either wired to the
desktop feature's hook or marked as phone only on purpose. lint:dup
catches the copies that look alike; this one catches the copies written
anew.
The manifest starts with 120 features and 62 phone only screen files,
each with the reason it stays on the phone alone.
client/CLAUDE.md says how to change a feature, add one with both views
and add a phone screen.
Most features have a desktop view and a phone view, and each carried its
own copy of the feature's logic, so SonarCloud's duplication condition
was the gate a pull request touching both failed most often, and only
after a scan of twenty minutes. lint:dup finds the same copies before
the push.
It runs jscpd's detector (@jscpd/core with @jscpd/tokenizer) over the
source files under src/ at the thresholds the SonarCloud gate applies to
TypeScript, 100 tokens over at least 10 lines, and counts per file the
lines inside a block that repeats elsewhere or in the same file. Tests
and declaration files are left out as Sonar leaves them out, and .ts and
.tsx are read as one language, so a hook that repeats a view's code is a
copy too. The files are read sorted and with readText, so every machine
counts the same. scripts/dup-baseline.json holds the files that still
hold copies with their count; a file may hold no more than its entry, a
file without an entry none, and an entry above its file or for a file
that is gone fails until --update lowers it. --list prints every block
with its other side.
The engine packages are used instead of the jscpd command, which walks
the tree in file system order, skips files past its size limits without
a word and drops every result once it meets a file it has no format for.
The server suite seeds and reads its database through some 5,500 raw
better-sqlite3 statements on the sync handle. They restate the schema by
hand, skip the repositories the app goes through and run on SQLite only,
so nothing in the suite could ever exercise the dialect layer on another
engine.
tests/helpers/factories/ adds typed factories per domain (users, trips and
days, places, assignments, bookings, budget, packing, to-dos, files,
journeys, tours, collections, notifications, settings, plugins, tokens,
atlas, vacay, photos, collab) and generic readers (findRow, findRows,
countRows, updateRows, deleteRows) that go through MikroORM in a request
context of their own, against the app's ORM or a createTestOrm handle.
They refuse a write to a read-only foreign-key twin, which MikroORM would
otherwise drop without a word. The README there is the conversion guide.
The categories integration suite and the tags and calendar-feed e2e suites
are converted in full as worked examples; the tags suite also moves from
hand-rolled tables to the migrated snapshot.
lint:test-sql (scripts/test-sql-ratchet.mjs) holds the raw statements left
to a per-file baseline: growth and stale entries fail, files whose subject
is raw SQL are exempt with a reason, and a single statement can be allowed
with a test-sql-allow comment. CI runs it next to lint:boundaries.
A password change ends every session of the user and re-issues one for
the device it was made on, handed out as the session cookie. A caller
that authenticated with a Bearer token never reads that cookie, so the
new session sat in the session list as a live device nobody held until
it expired.
The controller now asks for the replacement session only when the
request came with the session cookie, the same check the sliding
renewal makes. The response body is unchanged.
A session renewed from a token issued before sessions were tracked has
an id derived from that token, and the renewal inserts it only when no
row by that id exists. The nightly purge deleted revoked rows straight
away, so the next request with the still valid old token inserted the
session again as an active one and got a fresh tracked token with a new
lifetime. A logout, an ended session or a sign out of the others held
only until the purge ran.
The purge now deletes expired rows only, and a revoked row stays until
its own expiry. A derived row is recorded with an expiry no earlier
than the old token's, so by the time it goes the old token is refused
by its signature check anyway.
The token in the sign-in bodies was declared deprecated and "read by
nothing of ours", but the web client still decided an SSO sign-in by
reading data.token from GET /api/auth/oidc/exchange, an endpoint the
deprecation notice left out. Removing the field later, as documented,
would have broken SSO sign-in.
The exchange now answers { success: true, token } and the login hook
signs in when the response is OK and success is true, so it never reads
token. The field stays for API clients and is now on the deprecation
list in the API docs, the controller comments, server/CLAUDE.md and the
wiki (Security-Hardening, OIDC-SSO). Adding a field to the body breaks
no existing client.
A token without a jti past half its life is renewed on every request
its cookie authenticates, and each renewal minted a session with a
random id. A first page load sends many requests at once, so one browser
ended up with several active sessions; only the last cookie survived,
and the rest showed up in GET /api/auth/sessions as separate devices
until they expired, up to 30 days for remember me.
The interceptor now hands the token to SessionsService.renew, which
derives the session id from it (a SHA-256 of the token, laid out as a
version 8 UUID so it passes the same id check) and inserts the row only
if it is not there yet. Every renewal of the same token lands on the
same session; if that session was ended in the meantime nothing is
renewed, just as for a tracked token. A caller that passes no token gets
the old behaviour.
On a demo instance every visitor signs in as the one shared demo
account, so its sessions are other people's browsers. The session
routes had no demo check, unlike the password, MFA and account
changes, so any visitor could read the User-Agent of everyone else
signed in and sign them all out with revoke-others.
For the shared demo account in demo mode, GET /api/auth/sessions now
lists only the session the request came with, and DELETE
/api/auth/sessions/:id and POST /api/auth/sessions/revoke-others answer
403 "Sessions cannot be ended in demo mode." without ending anything.
Logout still ends the visitor's own session. Other accounts on a demo
instance keep the full routes. No MCP tool covers sessions, so there is
no parallel tool to change.
The demo sign-in now records the device's User-Agent on its session
like every other sign-in, which the session list needs to tell the
visitor's own entry apart.
Both swap the whole database file. The demo reset copies a baseline
saved at first seed, before anyone logged in, so it brings back an empty
user_sessions table; a backup restore brings the backup's own rows, or
none for a backup from before sessions were tracked. A session token
whose row is missing is refused, so the hourly reset signed every demo
visitor and the demo admin out, and the admin who ran a restore got a
401 on the next request. Before session tracking only the password
version applied, and those tokens survived.
Both paths now read the active rows with their owner's email before the
swap and put them back after the reopen, in a fresh fork, for every
account the new file still holds under the same id and email. A row the
new file already has is left as it is, so a session that ended there
stays ended, and a session whose id now names somebody else (a backup
from another install) is dropped instead of handed to them. The password
version check still refuses a token the restored account no longer
matches. Reading or writing them is best effort and never fails the
reset or the restore.
A session token was a signed JWT and nothing else. Logout only cleared
the cookie, a 30-day "remember me" token stayed valid wherever a copy of
it had ended up, and nobody could sign out one device short of changing
the password or rotating the JWT secret for everyone.
Every session token now carries an id (jti) naming a row in the new
user_sessions table (id, user_id, created_at, last_seen_at, expires_at,
revoked_at, user_agent cut to 256 characters; no address is kept). The
rows live behind UserSessionsRepository and the new sessions domain,
whose SessionsService is the one issuer: AuthService.generateToken and
the OIDC sign-in both go through it, and sliding renewal re-signs a
tracked token under the same id and moves its expiry. verifyJwtAndLoadUser
takes the repository and refuses a token whose session is revoked,
expired or gone, refreshing last_seen_at at most every five minutes.
A token without a jti was issued before this change; it passes until its
own expiry (or a password change), so the upgrade signs nobody out, and
renewal turns it into a tracked session.
Sessions end on logout (the one the request names, verified even when
expired), on a password change (all, the device gets a new one), a
password reset, an admin setting a password, reset-admin.js, an MFA
disable (all but the current one), an admin clearing a user's MFA, and
with the account. GET /api/auth/sessions lists the active sessions,
DELETE /api/auth/sessions/:id ends one and POST
/api/auth/sessions/revoke-others ends all but the current, with their
contracts in @trek/shared and two new audit actions. There is no UI for
them yet. MCP exposes no account or session tools, so there is no
parallel tool to change; an MCP call made with a session JWT as bearer
goes through the same check.
The token in the login, register, demo-login, MFA and passkey bodies
stays for API clients but is marked deprecated in the API docs and in
the client type, which never read it.
A nightly session-purge job on CronRegistrarService deletes the expired
and revoked rows. The migration is 042500. The sessions domain gets its
own coverage entry at the 80 floor until a full run measures it.
The rule that a multi-statement write runs in uow.transactional was
written down but nothing held it. scripts/tx-writes.mjs reads every class
under src/ with the TypeScript parser and counts, per method, the writes
it issues outside a transactional callback: repository calls that build
an INSERT, UPDATE or DELETE, calls to methods that write (on the class
itself or on a constructor-injected one), and a transaction holding
writes as one. A write in a loop counts twice, of two branches only the
larger counts, local helpers count where they are called, and a helper
whose every call sits inside a transaction is not reported. --explain
lists the line of every call that counted.
Today's 102 reported methods sit in scripts/tx-writes-baseline.json and
may only shrink; most are syncs with network I/O between their writes,
plugin lifecycle steps and adapters calling two service writes. A new
method fails CI, and an entry above what its method holds fails until
--update lowers it, the same ratchet as lint:size and lint:boundaries.
Two JSDoc tags mark the deliberate cases with their reason in the tag:
@txStandalone for a write that stands on its own whatever the caller does
(the audit row, notification sends, the booking price mirror, the
assignment journey refresh, the quota retention purge), @txIndependent
for writes complete one by one on purpose (the OAuth token retention
sweep, the notification claim and its release). Runs in CI after
lint:boundaries; documented in server/CLAUDE.md, the root CLAUDE.md and
the wiki's dev setup page.
The time budget for an exec script was read after every statement, the
last one included. A single statement without args, or a script whose
closing COMMIT crossed the budget, had already run and committed, yet the
call threw a time budget error. A plugin that retries on error then wrote
the same rows a second time.
The script now reads the clock before each statement after the first, so
it still stops before the next statement once the time is up and rolls
back a transaction it opened, but a script whose statements have all run
is reported as done. query() likewise reads the clock only once another
row has come back, so a result read to its end is returned.
The ctx.db doc in the SDK said a single statement was not timed, which
contradicted the query budget. It now names what is really untimed: an
exec with bound args and migrate. README and wiki describe the script
stopping before its next statement.
The time budget covered query() and tx() only. exec() without bound args
handed the plugin's script to Database#exec, which runs any number of
statements in one call with nothing checked between them, so a long
script held the host loop as long as it liked.
An exec script now runs one statement at a time. sql-script.ts cuts it
after every semicolon outside a string, a quoted identifier or a
comment; a piece SQLite reports as incomplete input (a trigger body cut
at its inner semicolon) is joined with the next until it prepares. The
clock is read after every statement, and past the budget the call
throws. As with Database#exec, the statements before that point stay
applied, and an error is SQLite's own; a transaction the script opened
itself is rolled back so the connection is not left inside it.
exec() with bound args is one statement and stays as it was. A single
statement cannot be bounded at all on the host thread, a write such as
INSERT ... SELECT over a cross join included, and the SDK README and the
wiki now say so next to the cross-join read. migrate() keeps running
unbudgeted on purpose: a migration stopped part of the way is rolled back
and fails the same way on every later start, leaving the plugin unable to
update its schema.
The ctx.db rows of the SDK README and the plugin wiki pages describe the
budget for scripts, and the lines touched there and in plugin-sdk's db
comment lose their em dashes.
Maurice