The 4.3.0 notes drop the plates and the tile index for the Keep a Changelog layout. This banner is the one image left, at the top. Scaled from 5292 to 2400 px wide so the body does not pull 4.7 MB.
The card beside the headline was three places stacked, which is what the
planner already had before this addon existed. It is the rail now: one
line down the drive, a filled node for a place you are going to and a
hollow one for a halt on the way, the time and the distance on the line
between them, and the one thing that has to be seen before it is read,
a stop past the daily limit, in the only colour on the plate that is not
its own pair.
Two stays beside a tall pane left a gap under the card that read as a
rendering fault rather than as space. Four fill it, and the list stretches
instead of centring.
The fourth figure was the nine scopes an API key can be narrowed to, and
the section of the release that explained them is gone; it says what the
page still says, that the addon is off until an admin turns it on.
The lede stopped at a width picked by hand, which broke the line early
and left a wedge of empty pane beside it; it runs to the pane's own
padding now, the same gap it keeps on the left. The bars title sits on
the ground rather than in a pane, so it takes more air under it than two
panes take between them.
It sat in a glass panel of its own with nothing else in it, which read as
a card that had lost its contents. It is the plate's heading now, on the
ground at the plate's own margin, the way the eyebrows sit elsewhere.
The winning row drops "alternating"; the bar beside it already says what
the two sources add up to.
The paragraph under it explained the corpus, which the sentence beside
the image on the release page already does. The plate is the claim and
the four bars behind it.
Three plates for the road trip addon in a third palette, violet and
teal, so the page's three subjects tell themselves apart before a word is
read: the drive as one chain with a day's own card beside it, what it
finds along the road against what a list of days could offer, and the
four ways to send the drive where you want it.
The Places API set loses two headings that repeated the sentence above
them, and the half of each comparison that is history is drained of
colour and dimmed, by brightness rather than opacity so the pane keeps
hiding the route behind it. "Index" is gone everywhere in favour of TREK
API, and the bars drop the index-alone row.
Both halves of the release page were built on the same blue and green,
which are Dawarich's own two brand colours and mean something on its
plates. On the Places API ones they meant nothing and made the page read
as one long subject.
The aurora and the accents are four tokens now, overridden per plate and
nothing else with them: the Places API set is amber and rose, Dawarich
keeps the cool pair it takes from its mark.
The mark was assembled here out of the plain wordmark and a hand-built
API chip, which is not the lockup: the official one has no leading glyph
and its own badge geometry. It ships as one file now, black on
transparent as delivered, inverted on the plate because that is the
dark-background variant.
The plate is 16:5 and its right half was empty ground while the lockup
stacked above a two-line title and pushed the lede down the card. The
mark sits on the far side of the headline now, at the size that space
deserves.
The fit check was reading .stage, which is `height: 100%` and therefore
always reported the plate's own height; it measures the first and last
block instead, which is how the hero turns out to have been carrying 94px
of empty ground under it.
The gaps were whatever each plate happened to pick: 16 here, 22 there,
and a `margin-top: auto` that pushed a tile row flush against the card
above it. They are one token now, used between the stacked blocks,
between the tiles, and between a card and the numbers under it.
Every plate is also as tall as its content instead of a guess, so nothing
is clipped and no plate publishes a band of empty ground. The Places API
hero drops the release pill and carries its API badge at the size the
lockup is now, and one subtext in the comparison is a line shorter so it
stops wrapping.
The release page carried two visual languages: flat cards for the Places
API and glass for Dawarich, one above the other. All five plates are the
same build now, from the same stylesheet, with the wording, the figures,
the bars and the row order of the originals left exactly as they were.
Each plate also carries the rim it was missing, a lit 1px edge and a
sheen along the top, so the image reads as a pane rather than as a dark
rectangle with rounded corners.
TREK's wordmark sat bare beside a chipped Dawarich, which read as one
brand hosting another rather than two meeting. Both are in the same glass
pill now, at a size the lockup deserves, with the multiplication sign
white between them instead of grey.
The plates are rounded now and carry transparency outside the radius, so
they sit on the release page as cards rather than as two black rectangles
with square corners against it.
The secondary lines were the same grey as the labels around them and read
as switched off at the size GitHub renders these; they are close to white
now. The second plate loses its heading, which repeated the sentence
above it, and the route line is one line instead of two.
The flat cards read as a spec sheet. Both plates now sit on the same
aurora, mesh and recorded route the app's own glass panels imply, with
rounded frosted panes over it; the hero carries a real From Dawarich row
so the first thing anybody sees is the product rather than a claim.
Two plates, not three: the third said in pictures what the text beside it
already says in words.
Three plates in the same dark palette and type scale as the Places API
set, built from the tokens in client/src/index.css, the TREK wordmark and
Dawarich's own mark: the lockup and the four numbers that matter, the
line between what stays in Dawarich and what reaches TREK as a question,
and the four places it turns up once it is connected.
Blanking the two lines left a band of empty plate between the title and the
first bar, which read as a missing element rather than as breathing room.
The band is cut out instead, so the plate is 56px shorter and the spacing
above the bars matches the spacing below them.
The two grey lines explained where the 128 places came from. In a release
note that reads as a caveat before the numbers rather than context after
them, and the corpus is already described in the text around the image.
72.7% and 58 ms take the success token, and their boxes a trace of it.
Only those two: tinting the licence line underneath as well would turn a
statement of fact into a claim.
The same treatment the repo's own tiles carry: a hairline ring at 12%
white, a brighter line along the top where the light would catch, and a
diagonal sheen falling off from the top left. The radius matches the
mask the PNG is cut with, so the ring follows the corners instead of
stopping at them.
The pills were pushed to the bottom of the plate, which left a hand's
width of nothing between them and the sentence they belong to. The plate
is as tall as its content now.
73.6 million is a number nobody can picture, so the plate shows a field
instead of only naming one: 520 seeded dots weighted to the right so the
headline keeps a quiet ground, five of them lit and strung together like
a drive across it. The mark lifts off that on its own light.
min-height only set a floor, and the right card's two wrapping notes
pushed past it, so the pair ended seventeen pixels apart. Both fill the
grid row now.
The past column now carries a faint diagonal hatch and one step less
contrast on every word in it, so the eye sorts the two sides before it
reads either. Every figure sits in its own box, the three closing
numbers are centred, and the caption above them lost the sentence that
pushed it onto a third line. All four plates get rounded corners with
real transparency, so they sit on either GitHub theme.
The subtext ran on across three ragged lines and the four figures sat in
a bare grid under a rule. They are pills now, and the sentence breaks
where it means to. The mark moves out of the corner into a plate beside
the headline, where it is large enough to read as the subject.
The chart carried a fourth bar for the index on its own, which is a step
in the reasoning rather than an option anybody can pick. Nominatim, what
ships, and Google is the comparison a reader is actually making.
The arrow sat in a column that included the metrics under the panels, so
it centred below them. It sits on the panel row now.
Four plates in the TREK dark palette, built from the real tokens in
client/src/index.css, the wordmark component and the Lucide set the app
already uses: the hero, the before-and-after against Nominatim and
Overpass, the settings card at poster size, and the measured top-5 bars.
The schemas had no spec of their own. The two things worth pinning are the
ones the feature rests on: the anchor is a position and never an id, so a
fraction or a negative is not a leg of anything; and the batch route's
track has three states, because absent and null mean different things to a
day that already follows a line.
The road trip handlers in useTripPlanner are unreachable until the addon is
on for the instance and the mode is on for the trip, so the existing suite
never entered them: 360 of the file's new lines and branches had never run.
Turns both switches on and drives what the rail, the corridor panel and the
map call. The cases that matter are the ones where a via has to be
re-anchored: a via records which stop it follows by position, so a reorder,
a move between days or an insert that forgets the correction silently
re-shapes the drive.
Five files, about 570 lines, that shipped without one. The cases are chosen for
what a later change could break quietly rather than for the line count.
useRouteAlternatives carries the decision the feature rests on: when a leg
already has vias, the road actually being driven is NOT among what the router
offers for the two bare endpoints, because it was asked a different question. It
is fetched separately and put at the top — without that, opening the picker on a
leg you have reshaped shows three roads, none of which is the one you are on and
no way back to it. Also pinned: a via that barely moves the route produces one
entry rather than two, a second ask abandons the answer still in flight, and only
driving, walking and cycling ever reach the router.
RoadtripStayModal writes `places.duration_minutes`, which the schedule has read
for years and nothing could write. The case that matters is that clearing sends a
zero and never a null: the update statement folds a null into "leave it alone",
so a null could give a stop a stay and never take one away. The clock cases pin
the thing a slider is for over a text box, which is watching the departure move.
StopKindPicker is a portal, so it gets the cases a portal makes easy to get
wrong: nothing without an anchor, outside-click and Escape both closing, the
anchor itself not counting as outside, and picking the kind a stop already has
taking it back off rather than setting it twice.
RoadtripCategoryPicker stays open on a click, which is the whole reason it is
multi-select, and names the kinds rather than counting them.
RoadtripAlternativesBar reports each offer by its own index rather than by its
position — a position drifts the moment the driven road is prepended — and shows
loading, failed and "only one way" as three different answers instead of an
empty list.
Four small things, and one that was not this branch's fault but is now its
neighbour.
The driving limits saved on every keystroke. Typing "180" sent three settings
writes carrying 1, 18 and 180, unordered and concurrent over HTTP/2, so
whichever the server committed last won: the number that decides every
over-budget warning could quietly end up a tenth of what was typed, and only
show it after the next reload. Typed locally now, committed on blur or Enter,
and a value that did not actually change writes nothing.
The empty-result message always blamed the name box. Four controls narrow that
list and only one of them is the name, so picking a section with no hits
produced `Nothing on the way matches ""` — empty quotes naming a filter the
reader never set. The count line above it was gated the same way, so the section
case lost its "x of y" as well. Both follow whatever actually narrowed the list.
The check-out dropdown printed a bare ISO date, the only place in the planner
that did; it goes through the same helper as the rail beside it.
Switching an addon off left open MCP sessions holding its write tools. A `when:`
gate runs once per session, at attach time, so the tools stay registered and
callable while the REST half answers 404 for the same user in the same moment —
and the list that decides whether to tear those sessions down had drifted:
airtrail and collections gate tools and were missing from it, which this addon
would have inherited. The list now sits beside ADDON_IDS and a parity test reads
the gates out of the source and holds it there. Read out of the source rather
than out of a runtime registry on purpose: a runtime set only holds the gates
whose modules the caller happened to import, so it would pass in a suite that
imports none — which is the shape of test that hid this in the first place.
RoadtripLimitsCard gets the test it never had, including the one case that
matters: three keystrokes are not three writes.
Five things, all of them a click or a drag doing something the traveller did not
ask for.
The GL via handle committed a move on every pointerup. Any button started the
drag and there was no movement threshold, so a plain click wrote the via to
wherever inside the 17px handle the pointer happened to be — at a low zoom that
is kilometres, and the write re-routes the whole trip. A right-click ran the
drag AND the delete, so one gesture sent two writes; on the platforms where
contextmenu comes first, the move landed on an id that no longer existed and
reported a failure for a deletion that had worked. Leaflet gets this right for
free from `dragend`, which is why the two renderers behaved differently for the
same gesture. Primary button only now, with a threshold, and pointercancel puts
the pin back instead of committing — the browser fires it when the captured
element is torn out of the DOM, which is exactly what a delete does.
With the alternatives picker open, one click on the route fired both handlers.
Layer registrations are independent in MapLibre, and every alternative starts
and ends at the same two stops — when the leg already carries vias the current
road is itself one of the offers, so the overlap is the whole leg. The click
dropped a via AND chose a different road: two writes and two trip reloads for a
gesture made once.
The corridor search could be started before the day had routed. The corridor is
then built from the straight line between the stops rather than the roads
driven, and the routing answer landing a second later changes the line, which
clears the search mid-flight and drops every result — no error, no partial
result, just the idle empty state again. Routing runs one day at a time about a
second apart, so the window is wide open on a long trip.
Leaving road trip mode left the alternatives drawn on the map. The switch is in
the sidebar and reachable while the bar is open, and the overlay depends only on
the picker, so pale blue routes with drive-time pills stayed on an ordinary
planner map with no road trip UI left to dismiss them from.
Clearing a leg's vias swallowed its failure and closed the picker anyway. Every
neighbouring write in that hook reports one. The leg still carried its via and
still routed the old way, so the traveller believed they had undone the detour.
Walking and cycling legs were counted as driving. A day may legitimately mix
modes — splitIntoRuns exists for that — and a two hour hike was reported as
"over the longest drive allowed", its minutes went into the day's driving time
and its kilometres onto the tank, where a ferry crossing could trip a range
warning outright. A leg with no mode still counts, because every leg was a drive
before the field existed.
The road-trip domain also says out loud that it is online-only, which the repo
asks for rather than leaving it to be discovered. A via anchors by POSITION, so a
queued write replayed after somebody else moved a stop would re-pin the drive
onto a leg nobody chose, silently and hours later; the offline story needs the
anchor to be an assignment id first. Until then the surfaces disable the gesture
instead of accepting one and dropping it.
Six theme-lint markers move onto the line they exempt. The linter suppresses per
line, so a marker in the comment block above suppressed nothing and the six
road-signage colours the authors documented as intentional were counted anyway.
The field was rendered in every non-managed user's own Map settings and written
to their personal settings row. The CSP is built from something else: bootstrap
reads the ADMIN default and puts that origin into `connect-src`. So filling the
field in was the one path the product offered, and it could not work — the route
calculator happily built its URLs against the typed origin and the browser
refused every one of them with a bare TypeError, not a status, so no fallback and
no retry could see it. Day-plan routes, transport legs, the mobile day route,
Studio's road route and the whole road-trip rail stopped routing, app-wide, with
a generic "route could not be calculated". Nothing in the UI could repair it
except emptying the field again, and there was no admin field to set the value
that would have made the origin allowed.
The repo has decided this shape once already: `llm_base_url` is admin-only
because it names where the server sends its own requests and only whoever runs
the instance can judge what is reachable. This is the same class for a different
reason, so it joins the same guard, which is now named for what it actually
covers. Clearing stays open to everyone, so a row written before the rule cleans
itself up rather than blocking the save.
The field moves to the admin's default-user settings, next to the tile URL and
the Carto key, where a value takes effect.
The domain arrived with 557 lines, a service test, and nothing else. No
controller test, no e2e, and no MCP test — so the guard chain, the trip scoping
on every route, the cross-trip track check and the day_edit parity were all
unasserted, and the whole domain sat in the 80 catch-all where it could shed
twenty points against the headroom of the domains beside it.
The e2e is the one that earns its place. `@RequireAddon` is metadata and does
nothing on its own, and AddonGuard is not global, so the ordering the
controller's own comment calls load-bearing is exactly the kind of thing a later
change removes without noticing — it had already happened once. Deleting the
guard now turns three cases red, which I checked by deleting it.
The MCP tools had never executed at all: the shared harness seeds the addon
disabled, so `when: roadtripAddonOn` kept all five from attaching. They now run
against the same checks the REST routes make, including the demo account and a
member without day_edit, because the repo requires the two surfaces to move
together.
Two client modules that had no test get one: the overlay builder, where two
decisions live that are not cosmetic — which offered road is drawn as the one you
are on, and that a label hangs where its own route is unlike every other, rather
than at a midpoint that can fall on a stretch two routes share — and the service
marker, where the thing worth pinning is that an ordinary place gets nothing
back, and that the anchor arithmetic includes the border.
The domain carries its own coverage floor now instead of hiding in the catch-all.
Pinned under the measured figure on purpose; the exact ratchet gets regenerated
with the repo's own script after a full run.
Along the way, roadtrip_vias and roadtrip_day_tracks join the reset list every
other domain's tables are on. Without that a via written by one case was still
there for the next one, which is how the first version of the MCP suite passed
an assertion it should have failed.
Search along the route went to Overpass and only to Overpass, on the one path
where that hurts most: one search fans out over sixteen boxes, and each of those
races four public mirrors. The index answers the map's explore pills in
milliseconds where those mirrors take seconds, and the corridor was the reason
the index exists.
Two independent things kept it out. The corridor sends its categories as a
comma-separated list, and the index branch looked the whole string up as one key,
so it missed every time. And the four kinds a drive needs, fuel, charging, rest
areas and campsites, had no Overture terms mapped at all. They do now:
gas_station and fueling_station, ev_charging_station, rest_stop, and campground
with rv_park.
The branch is list-aware the way searchOverpassPois already was, and it labels
each hit with the category that produced it rather than with the list that was
asked for, because the client colours and groups its markers by that field and
"fuel,charging" is not a category. It answers all-or-nothing: a category the
index has no terms for sends the whole query to Overpass, since answering the
half it knows would silently drop the rest while the panel still reported a
finished search. The caller's limit is passed through instead of a hardcoded 50,
so a mixed search gets the same per-category budget the Overpass path spends.
The other outbound path of this addon changes with it. Importing a shared Google
route geocoded its waypoints by calling searchNominatim directly, which walked
past geocodeQuery and its index lookup. Through geocodeQuery an index hit costs
no throttle slot at all, and what the index does not know still falls through on
the background lane.
A place answered from the index also carries its chain and that chain's Wikidata
item, so a branch gets its logo on the pin here exactly as it does on the
Overpass path. Sockets stay an OSM field: a charging station answered from the
index reports "not stated" rather than claiming it offers nothing.
A via records which stop it follows by POSITION, so anything that renumbers a
day's stops moves the ground under it. The road-trip rail corrects for that;
almost nothing else did.
reanchor() never renumbered `sequence`. It is allocated per leg and starts at 0
in each, and re-anchoring MERGES two legs whenever the stop between them goes
away — so leg A's 0,1 ended up beside leg B's 0,1, and everything that draws the
route orders by sequence. The drive ran A0, B0, A1, B1: forward, back, and
forward again. Persisted, surviving a reload, and unrepairable except by deleting
the vias. The old anchor is read before the updates and used as the primary sort
key, so the vias of the earlier leg stay ahead of the ones that merged into it.
Reordering the only two stops of a day deleted every via on it. A reorder is
modelled as remove-then-insert, and the removal half refuses to treat a two-stop
day as still having a drive — right when a stop really goes away, catastrophic
when nothing was removed. Two stops is exactly the shape "follow this track"
produces, up to nine vias on that one leg, gone to a single drag with no prompt
and no undo.
Removing a stop, reordering within a day, and assigning a place mid-day never
re-anchored at all, though the helpers exist and three other paths use them. The
detour then sat on a leg the traveller never chose, or on a leg that no longer
existed and vanished — until a later stop brought it back bending a road nobody
asked for. Reordering hands a whole new ordering rather than one move, so the
anchors there follow the stop they were pinned behind instead of arithmetic on
the index, which is correct for any permutation.
Duplicating a trip dropped all of it: `stop_type` was missing from the place
insert, and neither via nor day-track table was copied. The duplicate looked
complete and quietly drove somewhere else.
A failed read of the via list emptied it. That read runs after every write and
feeds the routing key, so a 502, a dropped connection or a tab going offline
re-routed the whole trip along the roads the traveller had steered away from,
with nothing said and nothing retried — indistinguishable from having deleted
them. Only a 404 or 403 empties now; anything else keeps the last known list and
says it may be stale.
Placing one triggers a re-route, the re-route lands about a second later, and
the render it causes rebuilt every via handle — which is exactly when somebody
reaches for the dot they just made. On the GL renderer that rebuild replaces
the DOM element outright, and a pointerdown belongs to the element it happened
on, so the drag went to an element that no longer existed. The second attempt
worked because by then nothing was pending, and an older via never showed it
because nothing was re-routing.
Two causes, both identity: the vias arrive as a fresh object on every reload
even when the rows came back identical, and `onMoveVia` is a fresh closure on
every render of the planner. Both were in the effect's dependencies.
The list is now keyed by what a handle is built from — id and position — so it
only changes when a via does, and the callbacks are read through a ref. A via
that genuinely moved still gets a new list, which is what redraws it where it
now is; the test pins both halves, because holding the element still must not
mean holding it in the wrong place.
Leaflet's markers had the same churn without the same symptom: a fresh
`position` array called `setLatLng` on every marker on every render, and a
fresh `eventHandlers` object unbound and rebound every listener. Same fix.
The vias already made the day follow the road; nothing said which road. After a
reload the badge could only say "shaped by hand", and re-fitting after the stops
changed meant finding the track again by eye.
One row per day, written in the same transaction as the chain it describes: a
day claiming to follow a road whose vias never landed would be worse than a day
claiming nothing. Three states rather than two, which is why the field is an
object and not an id — absent leaves whatever the day followed alone, null says
it follows nothing, an object records the road it now takes. A bare nullable id
could not tell "do not touch this" apart from "forget it", and a hand-dragged
via on the same day must not wipe the label.
The foreign key is the point: delete the imported track and the label cascades
with it, so no day is left claiming to follow a line nobody can see. The vias
stay, because they are the drive.
Both write paths check that the place is a track of this trip before it becomes
a label — permission alone would let another trip's place id, or an ordinary
place, become a name that can never be drawn.
Also clears three reliability findings the scanner raised on this branch: two
default sorts given an explicit comparator, and a trailing-slash regex the
scanner is right to call backtracking, walked instead.
Pasting a /maps/dir/ link into the list-import box answered "could not
extract list ID", which is the whole of the complaint: people share a drive
far more often than they share a list, and it is the same gesture — somebody
pressed Share in Google Maps. Which screen they were on is the URL's business,
so the box takes both and the controller decides.
No API key and no call to Google for the link itself: the stops are in the URL.
Three shapes arrive — the path form with its data blob, the documented
?api=1&origin= form, and a short link that is one of those after a redirect —
and a stop the link spells out in coordinates needs no geocoder at all. One
that is only a name is geocoded, keeping the name from the link rather than
the one the geocoder answers with: somebody who typed a nickname into Google
should not find a street address on their trip.
A name nobody can place is left out and counted rather than failing the
import, because a route of six stops with five findable is five stops more
than the traveller had. Coordinates are read as 1d/2d, longitude before
latitude, which is the reverse of a place link and the one way to put Berlin
in Somalia. A blob that does not line up with the stops is ignored entirely:
half a link read confidently is worse than one read as names.
An imported GPX track was a line drawn beside the day: the map showed the
scenic route while the day kept driving the motorway, so the distance, the
drive time and every warning belonged to a road nobody was taking. Norway
publishes its scenic routes that way, and so does every touring club worth
importing from.
A track becomes a route the day takes by refinement, not by conversion. Route
through what we have, find the point of the track the road misses by the most,
make that point a via, go again. Each round roughly halves the worst miss,
because the new via lands in the middle of what the last one left unfollowed,
so a scenic route usually fits inside ten vias — and the loop stops when the
road is close enough rather than after a fixed count. Every point of the track
would route exactly and take a hundred requests; two would route in seconds
and follow nothing.
The day is routed whole per round, not leg by leg: it is a single request
already, and per-leg would multiply every round by the number of legs against
a host that answers one request a second. The track is oriented to the day
first, because half of Norway's routes are published southbound, and the legs
it covers are cleared before the new chain lands, so applying a track twice
does not stack.
Offered from the day's own badges, and only once something has been imported.
Tinted while the day carries vias: the rail draws none of them, so that badge
is the only place a drive shaped by hand differs from one the router picked.
A nowrap label reports its whole width as its min-content contribution, so the
place-name select sat at its default `min-width: auto` and refused to shrink —
pushing the ± kilometres control past the panel's right edge instead of
ellipsing the name.
The menu that select opens now grows to its options rather than inheriting the
narrowed trigger, bounded by the panel it sits in so it never hangs over the
map: a list of identical shortened prefixes is not a list you can pick from.
A seven-hundred-kilometre day turns up seventy hits spread over the whole of
it, and the traveller wanting a break near Berlin has to read all seventy and
work out from "245 km along the route" which ones are anywhere near. That is
the wrong way round: you decide roughly where to stop, and then look at what is
there.
Picking a point narrows the list to what lies within 25, 50 or 100 km of it.
This is a subtraction rather than a second search, because the stops and the
hits are already measured along the same thinned line.
The midpoints of the legs are offered alongside the stops, and they are the
point of the feature: a break on a five-hour drive belongs in the middle of the
drive, not at either end. Asking only about stops would answer "what is near
Berlin" when the question is "what is near the halfway mark".
The chosen point is stored as which anchor it is rather than as a distance.
alongKm is measured against a line thinned with a tolerance derived from the
corridor width, so a kilometre figure kept here would quietly drift when the
width goes from 5 to 10; resolved through the stops it lands on the same place
again. It also names its day, so switching days cannot apply yesterday's break
to today's drive.
The corridor query has always ended in `out center tags`, and the projection
has always kept six keys out of the whole tag set. Everything an electric car
needs to know was in the answer and went in the bin: which plug families the
station has, how many of each, at what power, how many vehicles fit, and whether
it charges. Reading them costs no extra request, no key and no second query.
The filters treat "not stated" as its own answer, which is the only honest way
to build them: OSM states a socket type on roughly a third of charging stations
and a power on fewer still, so a power filter keeps every station that says
nothing rather than reading silence as too slow. The plug dropdown offers only
the families this search actually turned up, because CHAdeMO in a list for a
corridor with three Type 2 posts on it is a filter that can only return nothing.
Socket names stay as they are printed on the plug. "Type 2" is not a string
anyone translates, and the lowercase tag key is not what is written on the
charger.
The tool description follows, as parity requires, including the coverage caveat
— a model that reads a missing power as "no fast charging here" would be wrong
about most of the map.
An ordinary place could not become a fuel stop without going through the full
form, and the number in the rail is exactly what changes when it does: a service
stop has none. So the number is the control. Click the 3, pick the pump, and the
3 becomes an orange disc while everything below renumbers itself. The disc of a
service stop opens the same picker the other way, with a row to make it an
ordinary destination again.
Six coloured discs rather than six lines of text: at that size the colour and
the shape are the label, and anyone who has seen one petrol stop on the map
knows the orange pump before reading anything.
Centred on the column rather than on the number, which sits hard against the
left edge and would hang the panel half over the map. The click stops at the
disc, so picking a kind never also selects the place underneath.
Also in this change, from the same round: the corridor search offers hotels, a
hit somebody could sleep at opens the popup in an overnight mode with a check-out
day instead of a dwell, and booking one writes the place, its day and the row in
day_accommodations together. A hotel keeps its number, because sleeping there
ends the day rather than interrupting the drive, which is why the category list
and the stop-kind list are now two lists instead of one.
Reported off a real day: Berlin said 245 km since the last fill-up and Sanssouci
said 121, as if the tank had been filled in between. It had not. The budget was
being zeroed at every warning, so the first figure was right and every one after
it was the length of one leg rather than the distance on the tank.
A warning is not a fill-up. Only a fuel or charging stop resets it now, which is
what makes putting one in the answer rather than a formality: the same day reads
125, 370, 519 without one, and collapses to a single mark with one.
Nothing is flagged on a fuel stop either. Reaching a petrol station on an empty
tank is the plan working, and a warning there would sit on the one stop that
answers it.
Also: the stop number sits level with the middle of its row rather than at the
top, the way the corridor list already places its badges. A number pinned to the
first line drifts off as soon as a stop carries a stay and a walk under its name.
The three figures under the button were one run-on string joined by dots, which
says nothing about which number is which. Each takes the badge shell the rail
already uses, with its own icon: a clock, a calendar, a pump. The icon tile in
front of the title goes, because it was eating exactly the width the three
badges need to sit on one line.
Distances there go through formatDistance, so an imperial traveller reads miles
on the button and types miles in the dialog while storage stays kilometres.
EmptyState gains a compact variant. A sidebar state sits under its own controls,
where a line at page size competes with them; at the caption tier it reads as
what it is, a note about why the list below is blank.
Reported from a real trip: Hamburg, A1, HEM, Berlin. HEM to Berlin is 244.7 km,
and the badge on Berlin said 121 — which is the length of the NEXT leg, Berlin
to Sanssouci. The finding was anchored to the stop the leg departs from while
its wording ("since the last fill-up") is a running total, so the number never
matched what anyone adds up off the drive bands above it.
Both findings are true on arrival, so that is where they belong. The same trip
now reads 125 at HEM (21.5 + 103.4) and 245 at Berlin, which is exactly what you
get counting the bands.
A service stop carries them too. A charging halt is a stop like any other as far
as the tank is concerned, and HEM was the very stop whose figure went missing.
The corridor's empty state gets the mascot the rest of TREK uses: searching,
idle, and a confused look when a filter left nothing standing, so the picture
says which of the three it is before the sentence is read.
Eight native title attributes across the rail and the corridor panel, which
render as the browser's own grey box after its own delay and ignore every
appearance setting. All of them are the shared Tooltip now, the same one the
rest of the app uses.
The warning badge had its border in the soft tint of the warning colour, which
at sixteen pixels disappears into the card and left it as the only badge in the
row without a shell. It takes the warning colour itself.
Two tests moved with it: a tooltip only enters the DOM on hover, so what they
pin now is the visible difference between a pinned and a computed arrival —
weight and ink — rather than an attribute that no longer exists.
Three holes, one seam. Leaving the corridor popup by "more details" dropped the
kind and the dwell it had just worked out, so a fuel stop silently became a
numbered destination that counted in every total, in the badge and in the PDF.
Once set, the kind could never be changed back through the web UI at all. And
the REST route did not validate the field: both request schemas are open
objects, so stop_type: 'banana' was written straight to the column, where
isServiceStopType read it as false and the stop counted as a destination again.
The form now carries the field, so the popup's answer survives the trip through
it, an ordinary place can become a service stop, and a wrong kind can be taken
off. Editing keeps stop_type and deliberately leaves duration_minutes out: how
long a stay takes belongs to the rail's own dialog, and sending it from here
would overwrite what was set there.
The schemas gain the one field rather than closing: roughly 190 call sites post
whatever their form holds, and a closed object would reject all of them at once.
The doc comment claiming the list was already enforced is corrected — it was
describing an intention.
The two driving findings move into the stop's badge row in warning colours,
beside the stay and the walk from the road. All three say what this stop costs,
and one row reads as one answer rather than as notes stacked underneath.
Two things the column was getting wrong. The limits sat open as a card, three
rows of a narrow sidebar for something set once on the first day of planning
and left alone after; they are behind a proper full-width button now, with what
is currently set riding on the button so the form does not have to be open to
read it. And the walk from the road was its own line under the stay, where it
looked like an unrelated note; it wears the stay's two-part badge and sits
beside it, because both answer the same question about a stop — one in time,
one in a walk.
Roel and gfRally asked for three things: how far the car goes on a tank, how
long you are willing to sit behind the wheel in one go, and how much driving
belongs in a day. They are one feature. All three are a threshold, a walk over
the legs, and a finding.
Two things they deliberately are not. They are durations and distances rather
than times of day, because with no stop pinned to a clock the cascade produces
no times at all, for any stop, and a limit on an hour would simply never fire.
And the day limit is called driving time, not travel time, because the figure
it is measured against is exactly that: the badge beside it, which counts no
time spent at stops and no leg to a hotel. The card says so out loud rather
than leaving it to be discovered.
The range budget carries across midnight, since a tank does not empty
overnight, and resets after a finding: without that, 1800 km on a 600 km range
flags every leg after the first overrun. Only fuel and charging reset it, never
the other four service kinds — a two-hour lunch fills nothing, and nobody here
knows whether the restaurant has a charger in its car park.
Kilometres in storage, the traveller's own unit on screen. Skipping that round
trip means an imperial user types 400 meaning miles, 400 km is stored, and the
warnings arrive a third too early from then on.
No stop is suggested and no vehicle is modelled. OSM does not know what car
this is and neither do we, so the range is a number you type. Furkot does the
same, for the same reason.
A plugin may already route a day in road trip mode, and the charging stops it
answered with were thrown away: the map was handed an empty list and the rail
never looked. That made "build it as a plugin" — the answer given to charging
prices, to ABRP and to the AFIR data — untrue in the one mode where it matters.
Each halt is measured along the run once and filed on the leg it is actually
driven on, using the leg lengths the router reported rather than by projecting
the stops too: leg distances are exact by construction, while two stops close
together project onto each other's stretch and would put a charge the wrong
side of a town.
The rail row is deliberately the poorest of the four. No number, no editable
stay, no arrival time, a hollow disc: the halt belongs to the provider, not to
the traveller, and everything filled on that rail is something they put there.
Nor is it ever written back — a plugin halt stored as a via would go out as a
waypoint on the next run, and the plugin would plan around its own charging
stop.
No time on the row for a second reason: the SDK contract folds the stop into
the leg duration, so the arrivals already account for it. Adding dwellSeconds
to the schedule would push every arrival after a charge back by the charging
time.
A router snaps every waypoint to the nearest road before it starts, and OSRM
does it with no distance limit at all. A viewpoint above a valley, a farmhouse
down a track, a marina at the end of a pier: the drive to it really begins
somewhere else, and the line said nothing about that. It either stopped short
for no visible reason or appeared to cut across open ground.
OSRM reports the snap position and its distance in every single answer, and
this client has thrown both away since the first route was drawn. Reading them
costs no request and no key.
The gap is drawn as a dashed spur in the route's own blue, over the route
rather than under it, because it is the one piece of the line that is not
driving. Thirty metres is the floor, which is about a dual carriageway and its
verge, so a place pinned off an aerial photo does not earn a dash. From a
quarter of a kilometre the rail prints the distance too, because that is where
"the car park is over there" turns into luggage, a gate, or a track a hire car
should not be on.
Vias are deliberately left out: a via is a shape handle, not a destination, and
a dash hanging off one reads as a fault.
Both public hosts TREK ships with reject the parameter with HTTP 400: their car
profile is built without excludable classes, and only a self-hosted OSRM built
through the MLD pipeline has them. Every long leg was therefore paying two extra
requests, against a host that allows one per second, to be refused twice.
The first real attempt is the probe. Only a 400 is remembered, because only a
400 is about the host: a 429 is a rate limit, an abort is the user moving on,
and a NoRoute means this leg has no way round the motorway while the next one
might. Kept per base URL, so pointing the instance at an own router that does
support it starts asking again without a reload.
The comment above the fallback said none of this and read as a working feature,
which is why "only one sensible way to drive this" keeps appearing on legs that
plainly have another road.
The single-via route is right for a detour dragged by hand and wrong for
anything that derives its anchors from a line. Pinning a day onto a recorded
track or a signed scenic road means a couple of dozen points, and one request
each re-routes the entire trip once per point, spaced by the routing host's own
rate limit — twenty anchors would spend half a minute drawing routes nobody
asked to see.
replace_legs names the legs to clear rather than taking a flag for the day, and
that is the whole difference: a traveller who bent two legs by hand and then
adopts a scenic road for a third has not asked to lose the two.
The tools follow in the same change, as parity requires. Undoing a detour picks
this up too, where the vias of one leg were deleted one request at a time.
Three separate features write into the same two places here: the leg merge and
the final memo. The access lines want the snapped waypoints, plugin routing
wants the via points it was given, and the driving limits want findings that
belong to a leg rather than to a stop. Built one after another, the same two
lines get rewritten three times.
So the shapes land once, empty. RoutedLeg gains snapped and vias, RoadtripDay
gains driveWarnings and dayWarning, RoadtripRoutes gains accessLines, and the
memo becomes a walk rather than a map, because a range budget carried past
midnight needs somewhere to live between two days.
Schedule stays exactly as it was. Its own test compares the empty result with
toEqual, which checks both directions, so a third field there would fail it —
and a finding about the driving does not belong to a stop anyway.
The kinds were written out in nine places: the enum in @trek/shared, the
service list and the colours in the model, the corridor's category keys, the
map's road-trip categories, the popup's own table, two lookups in the rail, one
in the map markers and one in the corridor panel. Each carried a comment asking
the next person to keep the others in step, and it came apart anyway.
A rest area was a parking sign in the popup, on the map pill and in the panel,
and a coffee cup on the rail disc and the map marker, for the same stop. The
parking sign wins: the colour has been documented since it was written as "the
blue every parking sign in Europe is printed in".
stopKinds.ts holds one row per kind and everything else reads from it. It sits
beside the model rather than inside it because an icon is a React component and
roadtripModel.ts is deliberately React-free so the arithmetic can be tested on
its own; the colours and the service list stay there and are read from here.
Two flags rather than one, because they are not the same question: a hotel is
something the corridor can find, but sleeping somewhere ends the day rather
than interrupting the drive.
Seven tests pin the table against the two lists that cannot import it, so the
next kind added in one place and forgotten in another fails a test instead of
quietly counting as a destination.
The registry the MCP tests build had every other domain and not this one, so
the four via tools were absent from every assertion that walks it. A test that
asked whether they were reachable, or gated, would have passed without them
ever being there.
@RequireAddon has sat on the controller since the domain was written and did
nothing: it is metadata, AddonGuard was not in the chain, and it is not one of
the three global guards either. The addon ships disabled (seeds sets enabled 0),
so on any instance that never switched it on, all six via routes answered
anyway. Two doc comments in this repo already describe the behaviour this
commit creates rather than the one it replaces.
AddonGuard leads the chain, the way reservation-import and collections do it,
because a disabled addon owes an anonymous caller a 404 rather than a 401.
The page reaches into routeAlternatives.open, roadtripRoutes.lines and
roadtripVias.byDay before it ever asks whether road trip mode is on, so a
fixture without them took all 47 wiring cases down with "Cannot read properties
of undefined". The mode stays off, which is the state every assertion here is
written against.
routing_base_url arrived on this branch and was assigned to neither managed
set, which MANAGED-KEYS-001 exists to catch. It is locked, alongside the other
infrastructure endpoints: the host it names is folded into the CSP once at boot
(routingCspOrigins), so a customer who repoints it lands in a policy that does
not know their router and the browser blocks the requests with nothing to
catch. MANAGED-KEYS-005 pins the list verbatim and is updated with it.
The client-side budget moved to twenty-five seconds when the query timeout it
has to outlast was written down as a constant, and this assertion stayed on the
old twelve. Pointing it at OVERPASS_TIMEOUT_DEFAULT_MS is what the test was
after in the first place: the budget must be longer than the timeout the query
carries, and a literal here is exactly how the two came apart.
Three e2e suites build their own trimmed places table with a plain CREATE TABLE
rather than going through the schema and the migrations, so the column this
branch added never reached them. Every assignment, day and place case that
touched a place failed on "no such column: stop_type", thirteen in total, and
none of it showed because the checks do not run on a draft.
create_place learned stop_type when the column arrived; the three tools that
could change a place did not. Over MCP the kind was therefore writable exactly
once and never afterwards: a place saved as fuel fell out of every stop count,
wore the wrong disc for good, and could only be put right through the REST API.
update_place and bulk_update_places take it nullable, which is how the service
already reads "turn this back into an ordinary place" (an explicit null rather
than COALESCE). create_and_assign_place takes it the way create_place does.
The description also named four kinds where the enum holds six, so the model
had no way to reach restaurant or sights.
The controller carries @RequireAddon(ROADTRIP), the four MCP tools carried
nothing. On an instance that left the addon off, REST answered 404 while
list_route_vias, add_route_via, reanchor_route_vias and remove_route_via kept
reading and writing, which is exactly the parity the repo refuses to break.
Same shape as budget: an addonGate constant, AddonsService held publicly so
the module-level gate can reach it, AddonsModule in the imports.
The car branch prints the endpoint chain whenever it holds two entries, which
an ordinary rental always does: picked up and returned at the same desk. The
printout then reads "Munich Airport -> Munich Airport", where before there was
no car branch at all and the line simply stayed empty.
Consecutive repeats now collapse and a chain naming one place only is dropped.
A genuine round trip keeps both ends, because Dresden sits between them.
The road trip rail reads `duration_minutes` as a stay, and a place created
without one claimed an hour nobody had entered. The branch answered that by
binding NULL instead, which is right for the rail and wrong for everything
else: the column declares DEFAULT 60, three tests assert it, and the comment
on one of them says so in as many words. That default reaches every place in
TREK, not only the ones on a drive.
So the hour stays, and the road trip solves its own problem where it arises.
A stop added from the corridor search already brings its own figure with it
(RoadtripStopPopup's STOP_KINDS: ten minutes for fuel, twenty for a rest area,
thirty for a charger), and saveStopDraft passes it through, so the kinds that
would be misread as an hour never take the default in the first place.
The road-trip addon's second half: the parts that let a planned drive be
argued with rather than only read.
Reshaping. A via records which stop it follows by position, and that position
is resolved fresh every time the day is routed — so inserting, reordering or
removing a stop moved the ground under every via after it, and the drive
silently snapped back to the road it had been steered away from while the
handle still sat on the map where it was left. The correction is computed
against the day as it stands and written with the same change, in one
transaction: a half-applied re-anchoring mixes two numbering schemes in one
day. The one judgement that needs geometry — which half of a split leg a via
belongs to when a stop lands in the middle of it — is measured on the road
actually driven, so a fuel stop added to a bent leg keeps the bend on the side
it belongs to.
Alternatives. OSRM answers most long legs with exactly one route, which read as
"there is no other way" when a perfectly good slower road existed. When it
offers nothing, the same drive is asked for again without the motorway, and
failing that without the tolls — one question at a time, stopping as soon as
one lands. The offer says what it is rather than how much slower.
Stops along the way. Everything the corridor search can find is a pause, not a
destination: fuel, charging, rest areas, campsites, food and sights all sit
inside the leg they were found on, carry their own colour and icon, and are
left out of every stop count. That colour is defined once and read by the rail,
the corridor list, the map pin and the picker — and it replaced the chains'
logos, which turned a corridor full of petrol stations into a row of
advertisements and left the brands with no logo on file looking like something
else entirely.
How long a stop takes was readable in the rail and writable nowhere; new places
were given an invented hour that nobody had entered. It is now set from the
rail, cleared as a zero because the update statement folds a null into "leave
it alone", and new places start with nothing.
Both columns are rebuilt on one type scale, anchored to the day plan's own
sizes so switching tabs does not change the size of the text beside it. The
rail reads as trip totals, day cards and a chain of stops with the drives
between them; the corridor column as what to look for, how the search is
getting on, and what it found. Six category pills over three lines became one
dropdown, which is the space the results wanted.
Also: the offered routes are framed on the map when the picker opens (`fitKey`
carries no coordinates and each renderer reads it as "the selected day"), road
trip mode does not cluster at all so the whole drive stays readable zoomed out,
and the drive-time labels finally render in the app's own typeface instead of
Leaflet's Helvetica. All of it in both renderers.
The query carries `[timeout:20]`, so the mirror is allowed twenty seconds of
work on it — and the fetch around it was aborted after twelve. A mirror under
load that would have answered at fifteen never got the chance, and the caller
saw a failed area.
That is what makes a corridor search along a long drive report half its
stretches as unsearchable: the stretches were being searched, we just hung up.
Both numbers now come from one constant, the client budget is the query's own
plus room to hand the answer back, and a test holds the relationship rather
than the two literals.
Three read paths built the same nested place object by hand, and two of them
left `stop_type` out of it while their SQL was already selecting the column.
The effect was visible: a petrol station added along a route drew correctly
for a moment from the optimistic write, then the server's answer replaced it
with a place that had no kind, and the road-trip rail redrew it as an ordinary
numbered destination.
`getAssignmentWithPlace` now goes through `formatAssignmentWithPlace` like the
list path already did, which removes the third copy rather than adding the
missing field to it. The two queries in the days service gain the column, and
its hand-built shape gains the field it cannot yet be spared.
The addon shipped with only its two pure modules under test, which left the parts
that actually talk to the network — and the panel a user reads the answer from —
resting on nothing.
Covers the routing hook (one request per travel-mode run, day default over trip
fallback, retry then give up, no re-fetch on a rename, abort on leave), the
corridor search (one query per box carrying every kind, the perpendicular filter,
dedupe, the retry pass before a stretch counts as unsearched), the corridor state
(roads driven over the line between stops, day follow-along) and both panels.
Two findings from writing them, fixed here:
- A dead straight leg is reduced to its two ends by the simplifier, and tiling
walks point to point — so a 250 km motorway stretch produced a box at each end
and nothing in between, while the search still reported itself complete. Steps
longer than half a box are now split before tiling.
- projectOnSegment guarded a degenerate segment with `lenSq === 0`, which two
ends that came through a chain of floating point arithmetic never hit.
Also splits the anchor resolution out of computeSchedule, which was three levels
of nesting for one question, and sorts the POI cache key with a comparator.
The road trip corridor asks for fuel, charging, rest areas and campsites over a
dozen boxes, and asking per category was four times the requests for the same
answer — slow, and four times the chance a shared Overpass mirror times out and
reports the stretch as unsearched.
`category` now takes a comma-separated list. Each OSM selector belongs to exactly
one category, so a hit is still labelled with the kind it answered, and the
result cap is shared out per category so one dense kind can't crowd the rest.
The MCP tool takes an array for the same reason.
Also puts the POI, plugin and route via-point pins on the map's render clock,
the same fix the planned-place pins already had: a library marker repositions on
every pointer sample while the canvas draws once a frame, so the suggestions
swam during a drag and snapped back on release.
A corridor full of petrol stations was a row of identical coloured dots, and the one
thing that tells them apart is the brand. OSM carries brand:wikidata on most of them —
88% of the fuel stations in a Berlin box — Wikidata carries the logo as property P154,
and Commons serves it.
The bytes are proxied and cached rather than linked: one self-hosted instance asking
Wikimedia for a handful of logos is a very different egress profile from every
visitor's browser announcing which petrol stations they are looking at. The cache is in
memory because a GET must not write to the database, and there are only so many fuel
brands; a brand with no logo is remembered too, or every pan would ask again.
A logo-less or unreachable brand answers 204 and the pin keeps its category icon, so
nothing depends on Wikidata being reachable. Both renderers show it, since both read
the same category table.
A petrol station and a campsite both came back as the same nameless grey dot, because
the road categories only existed on the server: the map falls back to grey for a
category it does not know. They now sit in the client's category table too — with the
pump, the bolt, the parking sign and the tent, in their own colours — which both map
renderers and the popup read from the same place.
They stay out of the explore pill's own list on purpose: that pill offers what a city
visit needs, and while road trip mode is on it is not on screen anyway.
The search also stayed quiet when only some boxes failed, which reads as "no fuel on
this stretch" when the truth is that nobody looked. A partial failure now says how many
stretches went unanswered.
The corridor was built from the day's stops, so it followed straight lines across
country the car never drives. Between two cities that line runs through open fields
while the motorway swings wide of it, which is why every hit came back sitting on a
stop rather than on the way.
The days now carry the routed geometry, and the corridor is cut from that. Same trip,
same search: 192 hits spread over the whole 482 km — first fifty, middle, and the last
stretch — where before all of them reported kilometre zero.
Following real roads takes more boxes to cover the same drive, so they go out three at
a time instead of strictly one after another, and the budget rises from ten to sixteen.
The panel head loses the sentence explaining what the sections below already say, and
the day's distance and driving time, which the day list has been showing all along.
Three things the first cut got wrong.
Switching into road trip mode took its time: every leg was its own routing request,
paced a second apart so the shared hosts would answer at all. The router already
returns a leg per consecutive pair of the waypoints it is handed, so a day is now one
request — split only where the travel mode changes, because one request carries one
mode. Answers that come from the cache are no longer paced at all, which is what made
returning to the view feel broken.
A list of petrol stations is half an answer: which one is on your side of the road is
the other half. The corridor search therefore moved out of the panel into the planner,
and its hits are handed to the map like any other place suggestion — grey, because they
are not part of the trip until you add one. The explore pill stands down while the mode
is on: it draws into the same channel and answers the same question for a drive.
The panel itself was a stack of controls without an order. It now reads as header,
search, results: what is being searched (with the day's distance and driving time),
what is being looked for, how far off the route, then the hits grouped by kind with a
count — and a progress bar while the boxes are being worked through.
The road trip mode could show the route but not what sits beside it, which is
half of what planning a drive is about (#1797).
The right column becomes a corridor search: pick what you need, pick how far off
the route you will go for it, and the day's drive is searched box by box. The
boxes come from the route itself rather than the viewport, so the answer is
"on the way", not "on screen"; each hit reports how far off the route it lies
and how far into the drive it comes, and adding one opens the normal place form
so it lands on the day like any other stop.
Overpass is asked in plain bounding boxes rather than with a polyline filter,
because that is what the existing /api/maps/pois route accepts and what the
mirrors can answer; the geometry work is pure and tested. Long routes stop at ten
boxes and say so instead of quietly searching half the drive.
Fuel, charging, rest areas and campsites join the server's POI categories, so the
explore pill and the MCP search_pois tool gain them too. A car booking's stops now
also reach the PDF, which printed pick-up and return with everything in between
missing.
The trip planner could show one day's route at a time and nothing across days,
so a road trip had to be pieced together in Google Maps (#1797, #435).
The addon adds a mode to the plan view rather than a tab of its own: the map
stays the workspace and gains the route over the whole trip, while the left rail
becomes one chain of stops with the driving distance and time between them and
arrival times that follow from them. A time pinned on a stop is kept and restarts
the chain, so a ferry or a museum ticket stays where the user put it.
Car bookings can finally carry the stops the schema has always allowed: the
endpoints table has held role/sequence for every transport type since #384, but
only flights and trains had an editor for them. The MCP tools already accepted
them, so this closes the gap the other way round.
Legs are fetched one at a time and spaced out — the public routing hosts answer a
whole trip's worth of requests with 429 — and the distance cap now looks at the
single leg instead of the sum, which is what dropped long drives entirely.
Two halves of one mistake, both found by the suite.
Letting each side of the merged search catch its own failure meant the index
could contribute nothing while the answer still called itself
"trek-places+openstreetmap". That label is not decoration: the search log writes
it into the corpus a candidate index gets scored against later, so a list that is
entirely OpenStreetMap must not be recorded as though the index had a hand in
it. The source now names what actually contributed.
And maps.service.test.ts drove the index for real. Its fetch stubs answer with
json() and no text(), which is not what the index client reads, so the branch was
entered and then failed on the stub — "uses Nominatim when user has no API key"
was really testing that a stub is incomplete. With the new circuit breaker those
failures also accumulated, and the breaker is process state, so it leaked into
whatever file ran next. trekPlacesSearch is stubbed there now, beside the nearby
stub that was already there for the same reason. The index paths keep their own
files.
Attribution first, because this branch argues the point itself: the card names
the sources with the comment that naming them is a licence obligation rather
than decoration, and the copy beside it promises that every field says where it
came from. The POI path then stamped Overture rows as OpenStreetMap. Overture is
not OpenStreetMap — it carries OSM among other sources under other licences — so
that is a wrong attribution rather than a rounded one. The wire contract keeps
`source` an open string and nothing on the client switches on it, so saying
`trek-places` costs nothing. The MCP tool description no longer calls itself an
OpenStreetMap tool either, and says which source answered.
Adding one place cost two round trips to the index. The details lookup fetches
the place and its answer carries the description; toPlaceRecord dropped that
field, so the enrichment fetched the same place again a moment later to read it.
It is carried through now and read from what the caller already has, with the
second lookup left as the fallback for when there is nothing to read.
That description's source URL becomes an href on the client and was taken on
`typeof === 'string'` alone. It now goes through the same allow-list a place's
website goes through — the file already does this for menu_url, with the comment
explaining why — so anything but http(s) loses the link and keeps the text.
The instance token was hashed from raw APP_URL, which the shipped compose file
leaves commented out. Every default install therefore sent the same token, and a
token every caller shares is not a caller: one instance running a large import
would rate-limit all the others, who would fall back to Nominatim with nothing
to point at. getAppUrl() falls through to ALLOWED_ORIGINS and then to the port,
so unconfigured instances still differ.
Two client-side leaks. A pick made out of the offline cache was recorded into
the shadow corpus, where it moved the two figures the corpus exists to
measure — and offline is a setting, so the network is usually reachable and the
row landed. And the prefetch never re-checked the session after its request came
back: a logout in that window repoints the handle at the anonymous database, so
one person's trip area was written where the next person's offline search reads.
The cache path also ignored the abort signal, so an answer for a keystroke two
keystrokes old could overwrite a newer suggestion list. It rejects the way both
surfaces already know to swallow.
Left alone, and said in a comment rather than hidden: the index cannot localise.
The Overpass path picks name:<lang> and falls back to int_name; the service has
no language parameter at all, so a German reader exploring Tokyo gets the
Japanese primary names on this path.
Three things an install feels rather than reads.
A service that cannot be reached cost the full timeout on every call. Each call
site falls back to the old path, but pays 3.5 s first — so an instance behind a
firewall that drops rather than refuses, or one that simply has no route to the
service, pays it on every keystroke of autocomplete and on all three queries a
booking import runs per venue. Ten venues is over a hundred seconds of waiting
for an answer that was never coming. A process-wide breaker now fails instantly
after four consecutive failures and tries again a minute later, so an
unreachable index costs one timeout instead of one per request. A 404 does not
count against it, because "no such place" is an answer, and neither does an
oversized body, because the service replied.
The offline place cache hung off the map-tile switch. That setting is about
tiles — its own text says "off keeps the cache to trip data + documents only" —
and it gets turned off for space, tens of megabytes of it. A trip's places are
about one. Somebody who turned tiles off therefore got an empty offline search
with no explanation, and "prepare for offline" reported success having stored no
place at all. The two run independently now; the tile progress phase stays gated
on the tile setting, so a run without tiles still reports no tile phase.
Tests for the merged search path, which had none: the wiring around
mergeSearchResults was uncovered, so which list goes in first and what `source`
says were both free to change. Neither is cosmetic — index-first is the whole
point of the feature, and the search log writes `source` into the corpus a
candidate index gets scored against later. Two of the eight cases pin the fix
from the previous commit: an index failure keeps the OpenStreetMap answer, and
neither source is asked twice.
place-shadow lands on its own coverage entry rather than the 80 catch-all, where
a new domain could shed twenty points unnoticed. Pinned under its measured
figure on purpose; the exact ratchet gets regenerated with the repo's own script
after a full run.
trek_places_enabled was read by six call paths — search, autocomplete, the POI
pills, booking-import geocoding, the offline area and details resolution — and
written by nothing. No admin route, no MCP tool, no switch anywhere in the UI,
and the key is not in the settings allow-lists either, so the only way to set it
was an INSERT against app_settings by hand on the server.
That matters more than the four switches beside it, which trade away quota. This
one decides whether every search and every keystroke in the place box leaves the
instance at all. The flag reads fail-open, so an install that upgrades without
visiting the admin panel starts sending them, which is the right default and
exactly why the operator needs somewhere to say no.
Adds the pair symmetrically to place-shadow from the same branch: getTrekPlaces
/updateTrekPlaces on AddonsService, GET/PUT /api/admin/trek-places with an audit
row, and a real switch on the card. The read stays fail-open so it keeps
agreeing with MapsService.trekPlacesEnabled(), and a test pins that agreement
across every stored value — if those two ever diverge the panel shows "off"
while the queries keep leaving, which is worse than either default.
The card said this itself: `available` was a display-only prop whose comment
read "False until the client is wired to the service; the card then says so
instead of offering a switch that would do nothing." It is wired now, so the
prop is a switch. Its two badge strings had been shipped to all 23 locales and
rendered nowhere; they make way for the switch label and a line naming what
answers while the index is off.
Six things the merged search and the offline cache got wrong.
An index failure took the OpenStreetMap answer down with it. Only the OSM half
of the Promise.all had a catch, so a rejection on the index side discarded ten
results that had already arrived and sent the same query to Nominatim a second
time, behind its 1.1 s process-wide throttle. The index refusing a query is
ordinary traffic rather than an outage — a common single word without
coordinates is turned down upstream as too expensive — so this fired in normal
use. Each side now catches its own failure, and the keyless fallback reuses the
answer OpenStreetMap already gave instead of asking again for a list that
cannot have changed.
Dedup never fired for a non-Latin name. namesOverlap splits on [^a-z0-9], so
長谷寺, Кремль and مسجد الشيخ زايد all tokenise to nothing and the function
returned false — the same place came back once from each source, one row under
the other, and picking either gave it a different id. That is the corpus this
whole feature was measured on. Two names with no Latin words are now compared
strictly, folded through NFKC so the width variants the sources disagree on
still match; sharing a character is deliberately not enough, because deduping
東京タワー against 東京タワー駅前郵便局 would swallow a real result.
expand=1 answered nothing for an index place while expand=0 answered in full.
The expanded lookup degrades OSM ids to the plain one and had no branch for the
new family, so the richer request was the poorer one, over REST and MCP alike.
The bias circle could exceed Google's 50 km ceiling. A trip spread over more
than that produced a radius Google rejects outright, turning a search into an
error toast. Clamped on both sides — the client because it draws the box, the
server because the radius arrives over the wire.
The offline cache was rebuilt on every sync and never deleted. syncTrip wrote
its meta row with put(), which replaced the whole record and dropped the
fingerprint saying the cached area was current, so every login, reconnect and
manual sync re-downloaded megabytes that had not changed. And clearTripData
never touched areaPlaces, so a trip switched off for offline use kept answering
offline searches and its rows stayed for the life of the install.
clearCachedPlaces, which claimed to do that job, was called from nowhere; its
only test asserted with .every() on an empty table and passed on nothing.
Picking an offline suggestion ended in an error toast. The details lookup had
no cache path, so it failed and the callers fell back to searching for
"name, address" — which the cache matches on the folded name alone, so it found
nothing for a place that was sitting right there. A miss stays a miss.
Four things the index changed and nothing followed:
`gers:` was missing from NON_GOOGLE_PLACE_ID, so isGooglePlaceId said true
for every id the index hands out. Five call sites gate on that function to
keep unresolvable ids away from Google, which answers them with a billable
400 INVALID_ARGUMENT: photo refs, the editorial summary, the place-photo
route, the expanded details lookup and the enrichment photo pass. Since the
index answers first for search and autocomplete, a GERS id is what an
ordinary new place now carries, so this was the normal case rather than an
edge one — three billed lookups per added place, none of which could ever
return anything.
The area download shared the search path's megabyte cap and its 3.5 s
timeout. Both are right for a keystroke and wrong for the offline prefetch,
which asks for 3000 rows at once: a city-sized answer sits on that megabyte,
readCapped threw, and the caller could only report "index unavailable" — the
trip then went offline with an empty place cache and nothing said so. The cap
now follows the row count the caller asked for, with a floor, and the bulk
call gets a budget that fits a bulk download. It is still capped, so a broken
upstream cannot hand us an endless body.
searchPlaces returned before the Google branch as soon as the index or
OpenStreetMap had anything, and both set google_place_id to null. That is
right for a search and wrong for the one caller that wants an identity rather
than an answer: list-import enrichment (#886) exists to attach a
google_place_id to a bare imported pin, and pickEnrichmentMatch discards every
candidate without one. So the import silently stayed unenriched on an instance
that pays for a key. That caller now asks for the Google identity explicitly.
Nothing changes for an instance without a key — enrichment could never resolve
anything there, before or after.
Tests cover the three functions this branch had left untested, which is what
put the maps ratchet under its functions threshold: the autocomplete index
path (a new file, MAPS-AUTO-001..007), the gers details branch when Overpass
is down (MAPS-GERS-008), and the client's abort timeout — including that the
signal exists at all, which nothing asserted before.
The coverage ratchet measures per domain, and this branch put new code into
four domains pinned well above the floor without tests reaching it.
addons (99/99/99): getPlaceShadow and updatePlaceShadow had none at all, and
the parity the service comment claims with PlaceShadowService.enabled() is now
asserted rather than asserted in prose.
maps (93/86/97/96): geocodeQuery was invisible - its only consumer stubs it
away - so the new suite drives the gate, an accepted hit, a hit with an
unusable coordinate, an index failure falling through, and both ends of the
Nominatim tail. It sits on the booking-import path, where it decides whether
thirty sequential lookups leave the instance.
admin (86/70/89/88) and place-enrichment (96/95/94/97): the place-shadow
handlers and their audit entry, the getAppConfig read, and nearbyWouldMislead,
whose lines ran incidentally while nothing asserted the result - the guard
could have been inverted and the suite would have stayed green.
toPlaceRecord wrote the brand's Wikidata item into `wikidata`, and
getPlaceDetails spreads the record over the OSM details, so for a chain branch
picked from the index that value replaced the building's own tag - and nulled a
real one when the place had no brand. resolveIdentity reads that field as the
place's own identity, so the photo ladder returned the chain's Commons files as
pictures of this branch and the description came back as the company's article
without the aboutBrand flag.
That is precisely the failure WIKI_IDENTITY_TAGS and readBrandIdentity were
written to prevent, in their own words: "a chain's logo passed off as a photo of
the restaurant would be a plain lie". The item now travels under brand:wikidata,
the key OSM uses and readBrandIdentity already reads, and resolveIdentity picks
it up from the payload for a place OSM does not know - so a branch still gets
the chain's description, flagged as such, and never its logo.
The client test also covers trekPlacesArea, the streaming half of readCapped
including the size cap, the configured base URL and the nearby defaults, none
of which any test reached.
The index-first branch hardcoded opening_hours: null while the payload it maps
carries them, and search_pois advertises "opening hours wherever OSM carries
them". It also capped the radius at 20km and then answered clamped: false, so a
wide viewport came back narrowed and said it was not - the same field the
Overpass path fills honestly.
The three red MCP cases were the same code telling the truth from the other
side: they stub Overpass one level below the facade, so pois() ran for real,
and because trekPlacesEnabled() fails open on a database with no row, a unit
test was sending live requests to places.liketrek.com from CI. That is what
answered with 50 places instead of the stubbed one. The index client is now
mocked in that suite, with the gate left open so the cases still run on the
shipping configuration, and the index-first path gets cases of its own here and
in a new maps.pois suite: field mapping, the empty and the failing index both
dropping through to Overpass, and an outage under both providers still reaching
the caller as an error rather than an empty list.
maps.service.test.ts drove pois() with no stub at all and leaked the same way.
When details fails to resolve a picked suggestion the sheet falls back to a
text search (#1192). That call was the one search left in the sheet without a
location hint, so the fallback could land on an identically named place in
another country - exactly what the hint exists to prevent, on the path taken
when the first hop already failed.
useLocationBias reads useTripStore; the sheet's harness only ever built a
planner prop. So the hook saw an empty store, sent no hint, and the one case
that asserts a box went red while its three siblings passed vacuously - they
expect undefined, which an empty store produces no matter what they seed.
capturedBias now seeds places, assignments and selectedDayId the way the
running app has them, which puts all four cases back to work. The 500km case
is retargeted from Tokyo/Sapporo to Tokyo/Nagoya: 264km across, biased under
the old diagonal rule and dropped under the 60km radius one, so it guards the
rule that actually shipped. A new 029b pins the day-first precedence - a trip
too wide to bias still hints at the day being planned.
The three AdminPage cases located the card with closest('.bg-white'), which
returned null once the card became bg-surface-card, and within(null) reports
that a line later as "container is null". Anchor on .rounded-xl, the same way
the green AdminSettingsTab suite already does, and ask for the single Test
button rather than the first of several: weather left the card when it stopped
needing a key.
FE-ADMSET-031 asserted the Open-Meteo panel inside that card. The panel was
deliberately removed from the desktop tab - it needs no key and had nothing to
configure - so the case now pins what took its place, the TREK index heading
the card with the recommendation the Google key used to carry. The panel itself
still lives in the mobile section, where FE-MOB-ASET-019 asserts it.
MapsSearchDto validates a locationBias and MapsService passes it through, but
no caller ever sent one. That gap is the whole difference between the numbers
in this PR and what the software does. Measured against a live instance over
the 127 places of a real trip, searching each by name:
without a hint 50.0% -> 54.8% top-5
with a hint 50.0% -> 73.0% top-5 (this PR reports 72.7%)
mess/eval_merge.mjs sends lat/lng on every index call and a viewbox to
Nominatim; the client sent neither. Without coordinates the index refuses
common queries as too expensive and contributes to 41% of searches instead of
100%, so the merge degenerates to Nominatim alone.
The hint has to be NARROW, which is the part that is easy to get wrong. Same
corpus, same instance, varying only how the box is derived:
no hint 54.8% top-5, 38.1% at rank 1
box around the whole trip 57.1% top-5, 17.5% at rank 1
box around the nearest few places 70.6% top-5, 34.9% at rank 1
A box spanning a round trip has its centre in open country between the cities.
It points at nothing and costs more than half the rank-1 hits. useLocationBias
therefore prefers the day being planned, falls back to the trip only while it
still fits inside one region, and drops the hint past a 60km radius rather
than sending a misleading one.
This also replaces the two hand-rolled copies of the bounding box in the
planner dialog and the phone sheet, which shared a defect: Number(null) is 0,
so a place without coordinates entered the box as the Gulf of Guinea and
stretched it across the planet.
Measured on 128 places out of a real trip to Japan, all saved through
the old search. Index alone: 51.6 percent inside the top five. Index and
OpenStreetMap alternating: 72.7 percent.
Two cleverer orderings were tried first and both lost. A score threshold
cannot separate the cases at all: at 0.75 it catches 19 of 62 misses and
throws away 6 of 66 hits. Promoting results whose name matches what was
typed is worse than it sounds, because "Hase Station" shares a word with
"Hase-dera" while OpenStreetMap answers in the local language and
returns the right temple under a name that shares nothing: that rule put
seven weak index matches ahead of the correct answer, which landed at
position eight. Alternating puts it at position two.
So the merge does not rank. Each source keeps its own ordering and they
take turns, index first. An exact business match still opens the list;
a landmark Overture does not carry is second rather than eleventh.
Measured against 128 places out of a real trip to Japan, all saved
through the old search: the index answered 51.6 percent inside its top
five. For most of the rest it returned the shops AROUND the landmark
rather than nothing, which is worse than nothing, because ten plausible
wrong answers look like an answer.
Overture Places is a dataset of businesses and is very good at those. It
largely does not carry temples, bridges, riverside walks, viewpoints or
observation decks, which is a fair share of what somebody planning a
trip searches for.
Two ways to detect "the index does not have this one" were measured and
both failed. Its own score does not separate the cases: a 0.75 threshold
catches 19 of 62 misses and throws away 6 of 66 hits. Name overlap does
not either: 57 false positives, because "Hase Station" shares a word
with "Hase-dera".
So this does not classify. Explicit search asks both concurrently and
merges, name matches first and the index ahead of OpenStreetMap within
each group. Autocomplete is untouched, because Nominatim's policy names
it as unacceptable use. Nominatim also gets the location bias it never
had, without which "Hase-dera" returns the temple in Nara rather than
the one in Kamakura the user is standing beside.
Two booking-import cases still stubbed searchNominatim after geocoding
moved to geocodeQuery, and stubbed it with the old array shape, which
made "has coordinates" pass on an undefined lat.
The index can now carry opening hours, from two sources: an
OpenStreetMap layer the service keeps as a separate database file, and
the schema.org data a place publishes on its own site, read in the same
page fetch the description comes from.
OSM stays the first choice where it has them. Measured across seven
countries it covers 27.5 percent of gastronomy, and an OSM entry
describes this exact object and gets corrected by people who walked past
it, where a chain's website often carries one set of hours for every
branch. The rest is what the operator publishes, which measured 8.3
percent on the same kind of place.
Both go through the same expansion into weekday lines, so a card reading
`opening_hours` gets one shape and not two.
Five maps.service cases went red when search started answering from the
index ahead of Google; they are about the shape of the Google request,
so they now switch the index off first.
An offline instance had the trip's own places and nothing else, so "find
a pharmacy near the hotel" had nothing to answer from: every lookup went
to a server.
The index can hand over a bounding box in one request, so it does, once,
next to the map tiles. A trip to Rostock is roughly a megabyte, refreshed
only when the trip's area actually moves. Search and autocomplete fall
back to it at the API layer, which is why all ten call sites got the
offline path without any of them learning about it, and the answer says
`source: 'offline-cache'` rather than pretending to be live.
Country packages stay what they were: a SQLite file for anyone who would
rather run their own copy of the service. A browser cannot query one
without shipping a WASM SQLite, and Germany alone would not fit in
IndexedDB, so that door stays where it belongs. /v1/packages lists them,
because "you can leave" only means something if the door is visible.
Overture has no descriptions, and the encyclopaedias only describe the
places that have an article, which is a fraction of a percent. What
Overture does have is the address of the page the description is on, for
43 percent of places, and those pages publish a summary in JSON-LD or
og:description precisely so machines can read it.
The API fetches each page once and keeps the summary; every instance
reads it from there. TREK never opens a business's website itself, which
would mean one request per instance per place against small servers.
In the ladder it sits below the encyclopaedias and above Google: an
article about this exact place beats the operator's own copy, but for a
restaurant or a shop the operator's summary is the only one that exists.
Credited by host and linked back, with no licence claimed.
The corpus behind "is our own index good enough to replace Nominatim".
When somebody searches and then picks a result, the client posts the query,
the rank and the pick. Nothing else: no user, no trip, no session, and
coordinates only to three decimals.
Deliberately decoupled from the search endpoint. Correlating a response
with a later pick would mean handing out a ticket on POST /api/maps/search
and threading it through all fifteen call sites; the client already holds
everything the row needs at the moment of the pick.
Evaluation happens offline afterwards: export the corpus, run it against a
candidate index, count how often the pick came back in the top five. That
is why collection can start before such an index exists, which is the whole
point, because the corpus needs weeks of real use and the index needs a
server, and those two waits should not be serial.
Off unless an admin switches it on, and it reads === 'true' rather than
!== 'false'. The three older places switches fail open because a migration
backfilled a row for installs already using them; nothing uses this one, so
an instance that upgrades and never visits the admin panel logs nothing.
Reading is admin only, retention is 180 days, and the switch state is on the
unauthenticated app-config endpoint on purpose: whether an instance records
what its users search for is something a visitor may know before logging in.
The card was a flat list: two key fields, four Google switches and a large
weather panel, all at the same level. That gave a switch you set once the
same weight as the key it depends on, and gave the loudest treatment on the
page to the one thing with nothing to configure.
Now three comparable provider blocks. The TREK index carries the accent and
the recommendation, because a key should be the exception rather than the
expected setup. Google carries no border badge, so it does not stand out,
and says plainly what it costs beyond money. Weather has left the card: it
needs no key and had nothing to set.
The "recommended" badge moved from Google to the index, which is the whole
point of the change.
Also here, and both are their own thing:
- The settings tab is two columns from xl up. Nearly every card is "label
left, switch right", so the middle stayed empty while the page scrolled
for ages. Explicit columns rather than a grid over the flat list, because
a grid pairs cards row by row and leaves a hole under the shorter one,
and these differ by hundreds of pixels.
- Pictures: the nearby fallback drops from 300m to 60m and is skipped
entirely for shops and gastronomy. Measured across 600 places in six
cities, 2 percent of ordinary businesses have a picture on Wikimedia
against 70 percent of churches, so for a cafe the curated rungs never
fire and the fallback always does. That is how the town hall ended up
over the doner shop. It fails open on an unknown category, so landmarks
keep the fallback that works for them.
vite.config reads the dev port and the API target from the environment,
defaulting to what they were. Several worktrees could not run side by side.
Adds a provider for the project's own place index and puts it in front of
the existing ones for search, autocomplete, details and the explore pills.
It needs no key, has no quota, and what it returns may be stored, which is
true of neither of the two providers it sits in front of.
Nothing about the Google path changes. Every call site falls through to
exactly what it did before when the index has no answer or the service is
unreachable, so a working search keeps working.
Two details worth knowing while reading:
- Details for a place from the index also resolve its OpenStreetMap record
by name and coordinate, because the index has no opening hours yet and
never has cuisine or a menu link. Without that the hours a user saw while
adding a place vanished from its card afterwards.
- The explore pills keep reporting source 'openstreetmap'. It is a typed
contract the client switches on, and the places are the same places.
TREK_PLACES_URL points at a different instance; unset means the public one.
Transitous has no GTFS coverage for much of Asia, so a search between two
Osaka subway stops finds nothing — the stops are not in the feed at all.
Add Google as an optional backend behind an admin setting, reusing the
install's existing Places key, and fall back to Transitous whenever no
key resolves.
Uses the Routes API rather than the legacy Directions endpoint the
request named: every other Google call here already speaks the
v1/X-Goog-Api-Key family, and Google closed the legacy endpoint to Cloud
projects created after March 2025.
Google returns walking as turn-by-turn navigation steps where MOTIS
returns one leg per walking segment, so consecutive walk steps are
coalesced into a single leg — otherwise a finely-sliced walk pushed a
real journey past the 20-leg cap and the whole itinerary was dropped.
Both responses now name the backend that answered. The fallback is
silent by design, which left an empty result ambiguous between "no data
here" and "the provider you picked never ran", and the admin panel warns
when Google is selected but no key resolves — or when only the admin's
own key is set, which serves them and nobody else.
Cost shaped the implementation, since Google bills where Transitous does
not: both field masks are cut to the cheapest SKU tier that still
answers the question, cache TTLs are longer than the Transitous path's,
and route alternatives ride along inside the already-billed request.
Addresses #1699
A probe map is exactly the value someone reaches for when their cluster is
unhappy, so the two ways of reaching for it should not break the render.
`probes: null` hit a nil pointer on `.Values.probes.liveness`; guarding with
`with (default dict ...)` also makes dropping a single probe work, which is what
you want on a pod whose liveness check is the thing killing it.
The rest is documentation the chart already gives every other value: a wiki
section next to Resource Limits, a Notes bullet, and a values comment naming the
one trap here, that a handler switch has to clear the shipped httpGet in the same
override or the API refuses the pod.
Verified with helm template: defaults render byte for byte what the chart
produced before, probes=null renders no probes, probes.liveness=null leaves one,
and helm lint passes.
The version window already said 4.0.0 to 5.0.0, and the test above it is even
called "covers the whole 4.x line", but that was only half the story: without
`recurring: 'per-version'` a notice falls back to a one-time dismissal, so
everyone who closed the modal on 4.0.0 never saw it again. It did not come back
on 4.1, 4.2 or 4.2.1, which is the opposite of what the window says.
The mechanism was already there and in use one entry below, on thank-you-support.
The dismissal path records the app version it was closed at, so the notice
re-surfaces on the next upgrade and stays down within the same version. Rows
dismissed before that column existed read as 0.0.0 and simply come back once.
A grid item defaults to min-width:auto, so the two date cells never went below
the width of their contents: a long locale date plus the keyboard button beside
it pushed each cell past its track, and on a phone the end date came to sit on
top of the start date with its own button clipped off the edge. Measured on a
412px viewport, the row overflowed its container by 30px.
Both cells get min-w-0 so the track wins, and the picker's own trigger gets it
too, which is what finally lets the ellipsis it already carries do its job
rather than the button simply growing. "12. Sept. 2026" now reads "12. Sept…"
in a narrow column instead of escaping it.
Keeping the locale format: the report suggests ISO 8601, but the date is
formatted from the user's language everywhere in the app, and the overlap is
the layout, not the wording.
Two reports, both about an import that fails without saying why.
#2262 — temperature: 0 sat in the shared request body, so it went out on every
attempt including both retries. Reasoning models (the gpt-5 family) reject any
explicit temperature, and the json_object retry fires on any 400, so the import
spent two doomed requests and could not succeed at all. The parameter is now
dropped once a server has explicitly said it will not take it.
Not as a third link in the chain, though. A gpt-5 rejects max_tokens AND
temperature, the API names one parameter per response, and it may name either
first — a chain of one-shot ifs converges on only one of the two orders. The
retries are a loop over what the server actually said instead: each remedy
applies at most once and they compose. Detection is a phrase, not the bare word,
because dropping temperature costs the deterministic sampling that small local
models depend on, and an unfamiliar wording still falls through to the
json_object retry exactly as before. NuExtract and Ollama are untouched: the
first attempt always carries temperature, and it is the only one they ever see.
#2261 — the bundled extractor is Debian trixie's 24.12.3, roughly eighteen
months behind upstream, and the reporter is right that this is invisible: a
stale extractor, a mistyped KITINERARY_EXTRACTOR_PATH and a genuinely
unsupported provider all look the same from the UI, which is nothing imported
and no reason given. The version now appears in the startup log and in a new
admin-only GET /api/admin/system-info, and LOG_LEVEL=debug passes the
extractor's raw stderr through — the JS ERROR and Invalid-result-type lines the
filter drops are the only evidence that a vendor script is missing or failing.
Capped at 200 lines, and opt-in because those lines can carry document
fragments.
Not bumping the package: the filter the reporter needs first ships in KDE Gear
26.08 and no Debian suite carries it, and sid's 26.04.3 wants KF6 6.22 and Qt6
6.10.2 against trixie's 6.6 and 6.8.2 — a partial upgrade of Qt6, KF6 and libc6
from unstable into the production image, on two architectures, that still would
not fix the reported carrier. It comes for free with the base-image move after
Debian 14.
Two suites encoded the behaviour this branch changes: MReservationSheet
asserted the literal metadata: null the forms no longer send, and the new
Your-share case reached for a label that is a card footer rather than a
heading.
Five findings from auditing the release branch, none of them a blocker for it,
all of them things it would have shipped with.
The mirrored booking price only survived an edit when the form happened to fill
metadata of its own. Both reservation forms and the phone sheet sent
`metadata: null` whenever the form had no metadata key to send, and null clears
the column outright — so restaurant, event, tour, parking, other, a hotel with
no check-in times and a transport with no carrier details all lost the price
again on the next save. They send the object now: it still clears what the form
dropped, and the server carries the price across (#2233).
Your share counted expenses nobody has paid for, while the balances right below
it stopped counting them. The tile contradicted the panel it sits in (#2225).
App.tsx kept its own list of routes that must not mount authed-only widgets, and
it had already drifted from the one the response interceptor uses: this one
named /public/journey/, that one covers all of /public/. It reads
isAuthPublicPath now, so a route cannot be public to one and not the other.
LLM_TIMEOUT_MS and TREK_PLUGINS_IGNORE_TREK_RANGE shipped in .env.example but
not in the compose file most installs actually read.
And a typo in the Russian offline copy: Завершёна → Завершена.
The credential strip and the 303/302 body downgrade are new behaviour in a
security helper, and src/utils sits on a 96% line ratchet that the untested
branches took under.
Two reported advisories, and a third path neither of them covers.
GHSA-7pqc-fj3c-9346 — the ntfy token fell back to the operator's whenever a
request did not carry one, so an authenticated user who named their own server
was handed the operator's decrypted token as `Authorization: Bearer`. The report
covers `test-ntfy`, but the live send path in channels/builtins.ts does the same
thing off a per-user `ntfy_server` setting: no test route needed, just wait for
the next notification. Gating the route on `role === 'admin'` would have closed
one of the two and broken the documented shared-server setup, where the operator
supplies server plus token and users only set a topic. The invariant is the
target, not the role: the operator's token travels to the operator's server and
nowhere else, and that rule fits both paths unchanged. Scheme is part of the
comparison — an http twin of an https server would put it on the wire in clear.
GHSA-8mw6-xphx-886m — both transports checked only the first URL and then let
the platform follow redirects on a dispatcher pinned to that first IP. Pinning
does not help: Node skips the pinned lookup for an IP-literal host, so a 307 to
127.0.0.1 or 169.254.169.254 connected straight through. They go through
safeFetchFollow now, which re-checks and re-pins per hop. safeFetch itself had
the same shape and carried it to sixteen more callers, several with a URL out of
a per-user setting, and one (pipeAsset) streams the response back — which would
have made a redirect into an internal service readable rather than blind. It
now routes through the same follower.
The third: following by hand opts out of the platform's own rules, and both
manual followers replayed the caller's init verbatim. undici drops
`Authorization` across origins itself, so converting ntfy to safeFetchFollow
would have re-opened the first advisory through a redirect. Credential headers
are now dropped on an origin change (same-host http to https stays, self-hosted
setups depend on it), and a 303 — or a 301/302 on a non-idempotent method —
drops the body, so a plugin's client_secret is not re-POSTed at a host the
first one merely pointed at. plugin-oauth pins maxRedirects to 0 like its OIDC
twin, and the Google Places calls get the timeout none of their nine call sites
passed.
The phone map filters down to one day and had no way back out. The chip
rail can only swap one day for another, the second tap on the active chip
is the day sheet, and the seed guard deliberately does not re-seed — so a
day picked on the map stayed picked until you left the trip and came back
(#2257). Desktop has had the deselect since the day header got it.
Its own round control beside the rail, map view only: the plan timeline is
single-day, so there is nothing to widen there, and the day sheet keeps the
one route into it that also exists on the map. It toggles — the first press
parks the day it came from and frames the whole trip, the second puts that
day back with its route, or falls back to the focus day if a remote delete
retired it in between. Leaving the map restores it too, since an empty
timeline would be the worse answer.
The Leaflet mock had no getCenter, so the new guard bailed out of every
draw and FE-COMP-JOURNEYMAP-046/047 went red. Give the mock a loaded map
by default and add FE-COMP-JOURNEYMAP-050, which takes the view away for
exactly the window #2254 hit: render must survive it drawing nothing, and
the moveend listener must still put the thumbnail up once fitBounds lands.
Public journey pages ≥1024px render the sticky sidebar map immediately with photos, before the deferred fitBounds/setView lands, causing latLngToContainerPoint to throw "Set map center and zoom first."
SystemNoticeHost, BackgroundTasksWidget, and save-to-collection were gated only on login/register/password routes, so anonymous visitors to /shared/:token and /public/journey/:token still mounted them and hit system-notices/active with no session, logging a 401 on every visit.
Three reports from the same week, all "the layout is in my way".
#2247 — the Plan tab shows both side panels from 768px up, unconditionally.
A Pixel 10 Pro Fold unfolded is ~860 CSS px, so 340 + 300 of panel left the
map a 200px strip; those same widths became fit padding, shrinking the
framing window to 140px and opening the camera two zoom levels out over half
of East Asia. The floating POI/compass cluster is centred on the viewport at
z-25, which at that width put it on top of both collapse tabs (z-1 inside
z-20 containers) and on the Add Place/Activity button — and its pills set
pointer-events: auto, so it ate the taps meant for them. The resize handles
are mousedown-only, so a touch user could not narrow anything either.
Between 768 and 1023px only one panel is open now, the day plan first, the
places list one tap away — the Google Maps shape the reporter asked for. The
collapse flags stay the wide-layout state, so a fold back restores what was
there, and a marker tap reopening "both" no longer re-crowds the screen. The
control cluster centres on the corridor the panels leave (the formula
PlaceInspector and DayDetailPanel already use), the tabs finally have
accessible names, the dead resize handles go, and a panel dragged to 520px on
a desktop is clamped so the map keeps 360px without touching the stored
width. computeMapViewport refuses to frame for less than a third of the box,
so absurd padding can never pick the zoom again.
#2248 — the public journey hero clips its own overflow to keep its
decorative circles from widening the document, which also amputated the
language menu at the hero's bottom edge: 8 of 23 locales, the rest
unreachable. Both share pages inlined the same menu, so it is one component
now, portalled to the body and anchored to its trigger the way CustomSelect
is, with a height cap and its own scroll. It closes on Escape and on an
outside click, which it never did, and `zh-TW` stops labelling itself 简体中文.
#2249 — the note row's pencil/trash pill is absolutely positioned over the
row's own reorder chevrons, and a coarse-pointer `opacity: 1 !important`
pinned it permanently visible on tablets. On a mouse it was no better: the
chevrons are revealed on hover, which is exactly when the opaque pill covers
them, so note reordering has never been clickable on any pointer. The pill is
gone; the row opens the note the way the phone shell already does, and delete
lives in that dialog behind the same confirmation. A link in the note body
keeps its own tap, on both shells.
Closes#2247Closes#2248Closes#2249
POST /api/addons/collections answers with the collection itself — the
controller returns the service's `Collection` straight through, and the e2e
suite pins that shape. The client declared a `{ collection }` envelope and
read `.collection` off it, so `createCollection()` resolved `undefined` and
handed back `null` every single time.
The caller needs that id. In the list editor and its phone sheet the id is
what the cover upload and `onCreated` hang off, so creating a list with a
cover picked in the same dialog silently dropped the image, the new list was
never opened or selected afterwards, and the retry guard that exists to stop
a failed cover step from minting a duplicate was dead. The list itself still
appeared, because `loadAll()` runs first — which is why this went unnoticed.
Nothing caught it because both test layers stubbed the shape the client
believed in rather than the one the server sends: the api suite's MSW
handlers wrapped the collection in an envelope, and the store suite's mock
resolved one too. Both now answer the way the controller does, `update()`
included — its declared type was wrong for the same reason, though no caller
reads its result.
The fix is on the client. That wire shape has shipped for as long as
Collections has existed and the server e2e test pins it, so changing that
side would be the breaking one.
The envelope fix routes the created place through the store, which offline
mints a NEGATIVE temp id: reservations have no offline write path, the
mutation queue rewrites temp ids in a URL but never inside another entity's
body, and day_accommodations.place_id carries a foreign key. So a
forced-offline save (the Settings toggle sets the flag while the network is
still live) reached the server with place_id: -1758..., failed the FK inside
the insert transaction and rolled the whole booking back into a 500 — where
dev merely saved it unlinked. Worse, the temp place is in `places` by then,
so every retry matched the same negative id and the booking stayed unsavable
for the rest of the session.
Offline the function now links nothing at all and skips the geocode too: the
booking cannot be written either way, so minting a place would only leave an
orphan behind on the next flush. The name-match arm rejects temp ids for the
same reason. `as never` is gone from the addPlace argument, and
placesApi.create finally declares its `{ place }` return type — an untyped
`r.data` is what let the envelope be misread in the first place.
resolveImportedPlace created the hotel's place through placesApi.create and
read .id off the response, but that endpoint answers { place }, so the id was
undefined: the accommodation was saved without a place_id and the new place
was orphaned. Because the planner's places state never learned about it, the
name match failed on the next edit too, and every save of the same hotel
produced another copy (five in a row on one trip).
Create the place through the store's addPlace instead. It unwraps the
envelope, so the accommodation is linked, and it pushes the place into
`places`, so a later save of the same hotel reuses it. The mobile sheet goes
through the same handler and is fixed with it.
The hook tests mocked placesApi.create with a bare { id } and therefore
never saw the envelope; they now assert on addPlace, and a new case saves the
same hotel twice and expects one place.
Closes#2243
FE-PAGE-LOGIN-014 failed intermittently in CI while passing in isolation:
twice in a row on unrelated PRs, each time costing a full run, and once on
one of two runs of the identical commit.
The takeoff overlay renders LoginWorld, which starts three
requestAnimationFrame loops on mount. Under the load of the full suite that
was enough to push the overlay assertion past waitFor's default timeout.
Nothing in this file asserts on the animation, and LoginWorld has its own
tests, so the file stubs it.
The guard in rtlText.ts had tests; the two lines that call it did not.
setRTLTextPlugin appears at exactly two places in the client and no test
touched either, so deleting a call, swapping an argument or dropping the
deferred flag left the whole suite green. The engines are mocked because
importing them for real builds a worker pool under jsdom, and the plugin
URL is mocked so the payload never loads.
The precache comment in vite.config.js was three releases stale, and it is
the only record in the repo of what a service-worker install costs. A real
build says 463 entries / 23,292 KiB, the plugin among them: it matches the
js glob, so it is fetched at install by every user rather than on first RTL
text as the branch description claims. That trade is the right one, since
the alternative is a CDN the production CSP refuses and an unshaped map for
anyone offline. Only the description of it was wrong.
Measured against a real client build, which no CI workflow performs: none
of them runs vite build or check:gl-split, so nothing here was covered
before.
Vector basemaps rendered Arabic, Hebrew, Persian and Urdu place names as
isolated letters in logical order — "القاهرة" drawn as "ة ر ه ا ق ل ا" —
because neither GL engine loads its optional RTL text plugin by default and
TREK never registered one. Raster basemaps hid the problem, since their
labels are drawn server-side, so it only became visible for everyone when
the default basemap became a vector style.
Both engine entry points now register the plugin lazily, so the payload is
only fetched once a map actually paints RTL text and Latin-only users
download nothing. The call is left with each engine because the two
signatures genuinely differ: MapLibre takes (url, lazy) and returns a
promise, Mapbox takes (url, callback, deferred).
The plugin is bundled rather than loaded from a CDN: TREK is offline-first
and the production CSP is script-src 'self', so a remote plugin URL would be
refused by the worker that imports it. Bundling also puts it in the Workbox
precache, so shaping keeps working offline. It needs a resolve alias because
the package publishes only its ESM/WASM entry point through "exports" while
setRTLTextPlugin requires the classic script in dist/ — the alias is shared
by the build and test configs so the two cannot drift.
The carry-over was conditional on a budget_items row still being linked to
the booking, which drops a real price in two shipped cases. The booking
importer writes metadata.price unconditionally but only creates the linked
cost when the Costs addon is on and the price is above zero, so an instance
with that addon off has bookings carrying a price and no budget row at all;
an MCP booking created with a price of 0 is in the same position. Both lost
the price on the first edit.
The condition is gone: a payload that does not name the key keeps what is
stored, and naming it, null included, is taken at face value so a price can
still be removed on purpose.
What stops a price from outliving its expense is now the other end.
deleteBudgetItem is the single chokepoint for all four delete paths, and it
clears the mirror there, inside the same transaction as the delete. That
also removes an ordering hazard the linked-row check had: the controller
runs update() before it syncs the budget, so on an explicit clear the row
still existed and the price would have come back a moment before it went.
The predicate moves into reservation-metadata.ts so it can be read and
tested on its own, and the update_transport tool description no longer
promises that metadata is replaced wholesale.
A booking's linked expense price lives in reservations.metadata.price
(written by BudgetService.syncReservationPrice). The booking form rebuilds
metadata from its fields on every save and only carries transit /
airtrail_ids over, so PUT /api/trips/:tripId/reservations/:id overwrote the
stored metadata and the price vanished from the card until the expense was
re-saved.
ReservationsService.update() now carries price / priceCurrency from the
stored metadata into an incoming metadata object that does not set them,
as long as a budget item is still linked to the booking. A payload that
sets price itself, a booking without a linked expense, and an explicit
metadata: null behave exactly as before.
Closes#2233
LLM_TIMEOUT_MS was validated as "a positive number", which lets two shapes
through that then degrade into an immediate abort rather than a long one.
A fractional value passes the check and floors to 0 in derive, which makes
the dispatcher fall back to undici's default and setTimeout(abort, 0) kill
every model call on the next tick. Anything past 2^31-1 does the same,
because Node clamps the delay to 1 ms. Both now refuse at boot, which is
what the wiki row already promised.
Also pins the seam the three deleted constants used to occupy: the clients'
abort deadline has to come from the configured ceiling. Nothing else in the
suite would have noticed a literal coming back.
Three shipped sentences became false with LLM_TIMEOUT_MS: two claiming the
addon needs no environment variables (one of them in the very file the
variable was added to), and one naming the old five-minute ceiling that the
deleted constant used to set.
The timeout for a model call was spelled three times and adjustable
nowhere: 120s in the Anthropic client, 300s in the OpenAI-compatible one,
300s in the Ollama one. None of them was the ceiling that actually
applied, either — every one of those calls goes through the pinned undici
dispatcher, and an undici Agent stops waiting for response headers after
five minutes by default, so a client constant set above that never got
the chance to fire.
- LLM_TIMEOUT_MS (default 900000) replaces the three constants and is
passed to the dispatcher as headersTimeout/bodyTimeout, so the abort
signal and undici are governed by the same setting. The default is
deliberately generous: heavier parsing work should fit without a code
change, and an admin who would rather fail fast can lower it.
- The ceiling stays on the model lane. safeFetchAdminConfigured is also
the OIDC lane (discovery, token, userinfo, JWKS) and the plugin OAuth
token exchange, and that one sends no AbortSignal of its own, so it
keeps undici's five minutes rather than inheriting a model setting.
- The derived value is floored: the schema accepts any positive number
and undici rejects a fractional headersTimeout.
Four separate defects sat behind "offline mode is broken".
The app shell. On the first online launch after a version bump, App.tsx
deleted EVERY Cache Storage bucket and unregistered EVERY service worker,
then reloaded. That is exactly wrong for a Workbox autoUpdate build, which
swaps its precache atomically on install and never leaves an empty one.
Between the purge and the completion of a fresh 22 MB precache the origin
had no shell at all; anyone who closed the app or lost signal inside that
window could no longer start the PWA offline. It also threw away the map
tiles and documents the user had downloaded on purpose. It now asks the
worker to fetch the new build and reloads once it has taken over. The
version marker also moves inside the guarded path, because it was written
after the purge and outside its try: a throwing setItem left the caches
gone, the marker unwritten and the reload unreached, repeating on every
launch.
The reported version is validated before it reaches this device's storage and
before it is compared against the stored marker, so a value that is not a
release tag can neither be persisted nor start an update on every launch.
The session. loadUser's offline branch kept the persisted session but
never called onAuthSuccess, the only caller of setAuthed(true), and the
mutation queue's flush, syncAll and prepareForOffline are all gated on it.
A session that launched without network could therefore never sync again
for the rest of its life, even after the signal returned: the offline
settings buttons spun and returned instantly having done nothing, and
queued edits never uploaded. It also left reads pointed at the anonymous
database instead of the user's own.
The date rule. A trip is only cached when it has not ended, and one that
ended over a week ago is actively evicted. The per-trip switch was an
opt-out list, so ticking a finished trip changed nothing at all. For
anyone whose trips are all in the past, "Download for offline use" stored
none of them, reported success, and the cache stayed at zero. Switching a
trip on now pins it, which overrides both the skip and the eviction, and
the switch shows the honest position rather than promising storage that
never happens.
What the screen says. A sync that never started (already running, offline,
signed out) was indistinguishable from one that stored everything: the
progress bar painted full and a green "Ready for offline use" appeared over
a run of nothing. Both syncs now report why they stopped and the screen
says so. A rejected Dexie read no longer renders as "No trips cached yet",
and the clear button stays reachable when the database holds data that no
bookkeeping row points at. A run that rejects outright says so too: the trip
list it starts from is a plain fetch, so a network blip took the whole thing
down and left the last progress frame on screen looking like a finished
download. The 401-with-HTML branch in the API client now
confirms a reachable proxy before tearing the worker down, since several of
TREK's own routes answer 401 with a text/html body.
Both settings shells move onto one shared hook rather than growing the same
logic twice, and the tile prefetch caps its depth when the browser refused
persistent storage. Opaque tiles are billed at about 7 MB each, and an
origin that is not exempt from eviction loses its whole bucket, precached
shell included, which Workbox then never refills because sw.js has not
changed.
Closes#2228
POST /oauth/register refused every single-label private-use scheme, so an
MCP client registering workbuddy:// could not connect at all, while the
error it got back advertised "a private custom scheme" as supported.
The cause was two validators that had drifted apart, both of which run on
registration. The provider required a dot in the scheme name, which RFC
8252 section 7.1 only recommends and which protected nothing
(javascript.evil:// passed it). The service behind it had no
custom-scheme branch whatsoever and no [::1], which is what actually
answered the reporter's second symptom, and its localhost exemption was
not gated on http:. So javascript://localhost/%0aalert(1), blob:, about:,
ftp: and wss: were all accepted through the settings UI, and the consent
flow assigns the stored URI to window.location.href. Only the CSP stood
between that and a navigation. Both now decide through one policy in
oauth.helpers.ts.
Widening the scheme set means the deny-list has to carry its own weight:
it grows by the browser, extension and OS-handler schemes that were
previously blocked only as a side effect of the dot rule, including
ms-msdt (Follina) and the ms- namespace by prefix. intent: is denied,
because Chrome follows its embedded fallback URL, which turns it into an
open redirect the authorization server cannot inspect.
Fixing registration alone would have moved the reporter's [::1]:{port}
failure one step later: TREK matched redirect URIs byte for byte while
RFC 8252 section 7.3 frees the port for loopback, so a native client on
an OS-assigned port got a 302 to consent and then invalid_redirect_uri.
The matcher now relaxes the port, and only the port, for http loopback.
Stored URIs are never canonicalised, so the %3A round-trips untouched.
Note this tightens POST /api/oauth/clients for inputs it accepts today.
That is the security half of the change and applies at create time only,
so no already-registered client changes behaviour.
Closes#2227
The note website round-trips correctly through the API and the desktop
grid renders it, but the mobile notes tab never carried the field: the
form had no input, the payload had no key, and neither the card nor the
read-only sheet showed a stored link. Since an editor only ever opens the
form and a viewer only ever opens the sheet, no role on a phone could see
or reach a website saved from a desktop.
Editing a note on mobile did not wipe an existing link. The key was
omitted rather than nulled, and the server's CASE WHEN guard kept the
stored value, so this was a visibility gap rather than a destructive one.
The form now always sends the field, null included, because the server
only clears on an explicit non-undefined value. Every render path goes
through safeExternalHref: the REST contract accepts any string, so an
unguarded href would execute a stored javascript: value in this origin.
The desktop expanded view had the same omission and is fixed alongside it,
so a member without collab_edit no longer loses the link there either.
Closes#2222
An expense saved with "No one paid yet" was still charging its split
members in calculateSettlement while crediting nobody, so the ledger
drifted off zero by the unpaid total. The greedy simplifier that turns
balances into payments has no memory of which expense produced which
debt, so it handed that phantom amount to whoever happened to be a
creditor from an unrelated expense. In the report, two people each owed
about 80 to someone who never paid that bill. Booking every payment it
offered then left one member permanently short, next to an "everyone
square" line.
Both cost panels already promise the opposite: they label such a row
"Unfinished, total only, not settled yet" and count it into the
Outstanding amount card instead. The server now agrees with them.
The guard deliberately does not also require a non-zero total, unlike the
panels' label predicate. PUT /budget/:id/payers with an empty array zeroes
total_price without re-applying it, which would otherwise leave a custom
split debiting its members against no credit at all.
The #1543 fixture and BUDGET-SVC-DB-004/005 were re-baselined: they pinned
the old arithmetic through a deliberately payer-less 9 000 RUB item. The
wiki said the old behaviour was intended, so it is updated too.
Closes#2225
* feat(plugins): TREK_PLUGINS_IGNORE_TREK_RANGE turns the version gate into a warning
A plugin whose author has not updated its `trek` range for the running TREK can
now be installed, updated and activated when the operator sets
TREK_PLUGINS_IGNORE_TREK_RANGE to a truthy value. Every gate (registry picker,
post-extraction re-check, sideload, dev-link, activation) warns instead of
refusing; a missing range is tolerated the same way; "install latest" takes the
newest published version. The plugin-API version gate is not lifted and the
Discover compatibility verdict stays truthful.
The admin is warned at every step: each bypass is logged, install/upload/link/
update responses carry a `trekRangeBypassed` marker, and both admin panels show
a "Version checks off" pill, an "Install anyway" confirm dialog before a
registry install, the same notice after a path that could not ask first, and a
persistent chip on the row while the plugin runs outside its range.
Nine new i18n keys in every locale; wiki pages updated (Environment-Variables,
Admin-Plugins, Plugin-Development, Security-Hardening).
* refactor(plugins): one useRangeBypass hook behind both admin shells
SonarQube flagged PR 2226 at 10.8% duplicated new code: the range-bypass
logic had been added to the desktop and phone plugin panels as identical
copies. The marker/warning types, the row chip, the "Install anyway" offer,
the dialog copy and the confirm-or-notice state now live in
useRangeBypass (next to useInstanceSettings, the existing shared-logic
precedent); each panel keeps only its own dialog/sheet markup.
* docs(claude): record the SonarQube gate rules and the no-English-placeholder i18n rule
The CI SonarCloud gate (Sonar way, on new code: >=80% coverage, <=3%
duplication, A ratings, hotspots reviewed) was only discoverable by waiting
20-25 minutes for the workflow. The root CLAUDE.md now states the conditions,
the desktop/mobile shell-mirror duplication trap and its shared-hook fix, and
how to check a PR with the sonarqube MCP first; client and server CLAUDE.md
point at it. shared/CLAUDE.md says every locale needs a real translation —
parity cannot tell an English placeholder from a translation.
* docs(claude): keep CLAUDE.md to commands, invariants, gates and philosophy
Verified every concrete claim in the five CLAUDE.md files against the code
and fixed the stale ones: the websocket server is the Nest realtime gateway,
MCP tools live in <domain>.mcp.ts files, the client "native shell" section
described code that was never written, offline maps also prefetch vector
basemaps, plus several renamed files and wrong table names.
Then trimmed everything that drifts on its own: dates, migration history,
counts, and inventories of stores, slices, domains, sync files and CLI
commands. What stays is the commands, the invariants with the file that
enforces each one, what CI actually runs, the reference implementations to
copy, and the rules for new code. Each file now opens with that rule.
In an installed app, opening a maps target in a new context left an empty
window: the platform switches to the maps application before that context
paints anything, and coming back lands on a blank page the user has to
dismiss before TREK is usable again. The handover now happens from the page
the user is already on when TREK runs as an installed app, which is where
there is no tab strip to close a stray window from. A browser tab still opens
a tab, because there a link should.
Editing a booking on the phone erased the stop it was linked to. The sheet
seeded its assignment from the create-for-a-stop flag, which is null on an
edit, and saved that straight back over a link the user had made on the
desktop; the server writes the field whenever the key is present. It now
seeds from the booking being edited, and the picker the desktop dialog has
always had sits in the sheet too, so a stop can be seen, changed and cleared
from a phone. The option list moved to its own module rather than being
copied.
Two more of the same shape, found on the way. The endpoint set was sent as an
empty list on every save, and the server swaps the whole set when that key is
present, so editing a transit booking dropped its stations; neither form
edits endpoints, so an edit no longer sends the key at all. And the sheet's
Files block listed only what was picked in this session, so a file uploaded
from the desktop was invisible on the phone. It lists what is attached now,
resolved the way the desktop resolves it.
A place also shows the files of the bookings that sit on it, on the phone and
on the desktop: a confirmation lives on the booking, and the place card was
the one screen that could not reach it.
The #2196 timeouts were applied to one shared transport, so the 30-second
inactivity bound landed on the password-reset and notification mails as well
as on the admin's test send. That bound starts counting after the greeting,
which means it bounds the transfer: a relay that scans the message can sit on
DATA for minutes, and a reset mail dropped for that reason fails where nobody
can see it. Nodemailer's ten minutes are back for a real send, the connect and
greeting cuts stay, and the test send keeps the 30 seconds it needs to answer
inside the client's 40-second budget. The source comment already described this
split; only the value disagreed.
Three things Sonar was right about. The content panel had grown a filter, a
menu and an upload into one function of thirty-two branches: the filtering
is four pure rules in photoFilter.ts, the row that drives it is its own
component, and what is left in the panel is the grid and the drop zone. The
gallery rows behind an entry upload move out of the store's setter into a
named function, since five levels of nesting is a place bugs hide. And the
ids for placed elements come from getRandomValues rather than Math.random,
in one module instead of four copies of the same line: two editors minting
the same id for two elements is one of them losing their work.
The new paths carry tests now, which is what the gate was really asking
for: the filter's rules, the ids, the upload helper, the store's derived
gallery rows, the canvas file drop, and the shell's upload, drop and stop
switch.
The stops list was the longest thing in the travel panel by a distance: a
fortnight is fourteen rows, and it pushed the maps and the country tiles
below the fold on every journey. It is a folding section now, closed to
begin with, with the count of what still counts in its head, so the closed
section answers the question anyone opens it for. The rows are one line
each, the day loses its weekday, the country appears only on a journey with
more than one, and the switch is a mark rather than a word, which is what
lets the place name fit.
A picture uploaded onto an entry also goes into the gallery, because that is
where the row lives. The entry upload only answers with the entry's view of
it, so the store derives the gallery row it did not get and the content
browser shows a new picture without the journey being fetched again.
The content browser gets a filter over the journey's pictures: all of them,
the ones no entry holds, or one entry at a time, with the day and the place
under each name. An entry card in the Entries tab reaches its own pictures
the same way. Pictures can now come in without leaving Studio, into the
entry the filter names or into the gallery, from the panel or dropped
straight onto the page, and a photo element can fill its page or the whole
spread in one press.
Stops can be taken out of the arithmetic. A journal entry carries a
stats_excluded flag; an entry that has it is off the route, out of the
distance, out of the countries and not a step, while staying in the
journal. Studio lists every stop in the travel panel with a switch, the
entry editor carries the same switch on desktop and phone, and the entry
card says when a day is off the route.
The hotel chips next to the day pill shared its target, so tapping a stay
opened the day sheet. A chip now carries its stay and opens it the way the
stay card in the day sheet does: the accommodation editor for members who
may edit days, otherwise the hotel's place, and the day sheet only when
neither applies. Closing the editor returns to the timeline it was opened
from instead of surfacing the day sheet, and the chip's accessible name
says whether it is a check-in, a check-out or an ongoing stay.
Prompt arguments cross the wire as strings, but the three trip prompts
declared tripId as z.number(), so every prompts/get failed with -32602
"expected number, received string", MCP Inspector included. The id is now
parsed from the string through a shared tripIdPromptArg (still an integer
in the safe range), PromptOptions only admits string-input schemas so the
mistake cannot come back unnoticed, and a prompt declared with an empty
argsSchema no longer demands an empty arguments object from the client.
The prompt tests run through a real in-memory MCP client instead of
calling the callbacks directly.
* feat(shared): scope plugin action descriptors
* feat(plugin-sdk): instance-scoped settings actions
* feat(plugins): run instance-scoped actions from the admin settings dialog
* feat(admin): instance-scoped plugin actions in the settings dialog
* fix(plugins): surface instance actions without instance fields and harden the action gate
* docs(plugin-sdk): note that hosts before 4.2.0 ignore action scope
* fix(admin): open the dialog for instance actions, explain inactive 404s, guard the save window
* docs(wiki): plugin pages for instance actions and the admin panel changes
The list held its manual order, so an item due tomorrow sat behind one created earlier but due later. The sort-by section gains a due-date toggle next to the priority one, on desktop and in the phone tab: nearest deadline first, undated tasks after all dated ones, ties keep the manual order. One sort at a time, and manual drag-reorder pauses while either is active, exactly like the priority sort.
The farthest-stop reduce gets its explicit seed (the guard above already refuses an empty list, the seed makes that local), and the toast pill moves into its own component so the host stays below the complexity ceiling.
Without stay times the bookend leg drew unless a carrier or a 2000 km edge stop disproved it, which is exactly the reported car trip: home five hundred kilometres away, routed back to the hotel after check-out. The disproof threshold is a day-trip distance now, and the route optimizer anchors through the same rule on desktop and mobile instead of unconditionally, so drawing and optimizing agree.
failed was set once and never cleared, so a single hiccup kept the badge on its error dash while later fetches quietly delivered data. It resets with every new anchor.
On wider phones the dashboard filter pill stretched across the leftover width instead of wrapping its chips. The collections list switcher could not scroll, so its lower entries and the new-list button hid behind the dock.
The instance-settings save button paired bg-accent with a hard text-white and vanished in dark mode; it uses the accent pair token now. A failed activate() after saving no longer reports success over a dead plugin: the config stays, the response says what happened. The allowed-hosts dialog scrolls like the ones e70d7397 fixed, and the discover cards get the press-scale hardening from #2158, on desktop and on the phone panel.
rehype-sanitize silently dropped anything that parsed as an unknown tag, eating text people had saved long before the sanitizer arrived; disallowed markup renders as literal text now. GFM footnotes pointed at a doubly prefixed anchor and never landed. One shared plugin set drives the polls, the notes, the note card and both mobile tabs.
Disabling Journey cascades the photo providers off on the server, but the panel kept its local rows, so after off and on the providers showed active while being disabled. The desktop and the mobile panel re-read the list after a journey toggle; the write itself still rolls back a single row on failure.
Same trap the task-list fix hardened: the press animation shrinks the container mid-click and the release lands next to the target. The collections rows, the journey gallery tiles and the shared vacay calendar cards carry the established no-press attribute now.
Without any configuration getAppUrl() invents http://localhost as the relying party, and the UI advertised passkeys that every click answered with not configured. A localhost RP only counts as configured now when the operator declared one. The origin check also binds to the expected origins again instead of accepting every subdomain of the RP id; that widening never shipped in a release.
The mirror of the account language stayed in localStorage past logout, so the next account on the device booted in the previous user's language forever. It is cleared on logout and rewritten on login, and the rate-limit message follows the same order the boot language does.
The split and receipt fields rejected the third decimal a KWD or BHD amount seeds, on desktop and in the mobile sheet, which share the currency-aware guard now. Signed amounts get a sign toggle on the numeric input because the iOS decimal pad has no minus key, and the under/over hint no longer swaps direction for a negative total.
Deleting a bag over MCP moves every figure, the items fall into the unassigned pile, yet it sent no totals ping; it pings the room now like REST and the plugin RPC do. The client refetches totals after a room re-join and when coming back online, through the same coalescing window. The plugin SDK types learn the create fields the routes accept since #2154, on both hand-kept declarations.
Between 768 and 1023px the view-controls row is hidden and the provider and upload buttons had moved into it, leaving no way to add photos; the gallery actions render below the floating bar there now. The mobile journey settings sheet gets the linked-trip tracks switch the desktop dialog got, and the request gate and the fitBounds behaviour are pinned by tests.
The phone toast was a single truncated pill, which cut the denied message exactly before the settings hint it exists for; long messages wrap up to four lines now. An insecure context is reported as its own error instead of masquerading as a blocked device permission.
Leaflet's wheel handler on the map container cancels the native scroll, so the wheel zoomed the world under the note; the tooltip stops scroll and click propagation now. Hovering back to the marker no longer closes the tooltip mid-way, the flip threshold matches the real tooltip height, and a touch swipe scrolls the note instead of panning the map.
The @fontsource packages for Inter, Lora, EB Garamond, Playfair and Bebas were declared but never imported, so every pick fell through to its fallback stack, in the print path too. The faces load with the lazy Studio chunk now, in the exact weights BOOK_FONTS uses, including the MuseoModerno 500/600 the picker offered and nobody shipped. A guard test derives the required imports from the registry.
connect-src never listed tile.openstreetmap.de or tiles.stadiamaps.com, so every prefetch fetch was refused before the service worker saw it, while the swallowed error still counted as a cached tile. Both hosts are allowed now and only reached tiles count. The prefetch rectangle is also widened to the screen the fitBounds view fills, on the raster and the vector path alike; the places extent alone left grey bands beside a north-south trip.
The markers replaced the booking's block but carried none of what it said: no DESCRIPTION (type, confirmation, notes) and no LOCATION fallback. The carrying booking is resolved once now, and the markers inherit its description and location exactly when the block is dropped. A check-in window ending before it starts reads as a late arrival past midnight instead of swallowing the event.
Four render sites resolved the day assignment's booking with find(), while the booking dialog happily links several to the same stop. A shared selector in dayMerge returns them all, time-sorted and stable, and the desktop plan tab, the place inspector, the mobile timeline and the mobile place sheet render the full list; the single-booking case is unchanged pixel for pixel.
journey_photos.sort_order is nothing but the insert counter, so photos added late landed behind days they belong before. Both gallery readers (the app and the shared journey) order by capture time now, falling back to the timestamp of the stop the photo belongs to and then to upload time, through one shared SQL fragment; ties keep the old order, so existing galleries only move where chronology demands it. The order inside an entry stays the user's manual arrangement.
testSmtp logged nothing and the transport ran on nodemailer's two-minute defaults while the client gave up after eight seconds, so every failure was a generic toast over an empty log. The transport gets real timeouts, every outcome is logged without credentials, and a classified, scrubbed cause (auth, DNS, refused, filtered port, TLS mismatch, certificate) goes into the response the admin sees; the channel test calls wait long enough to receive it. The password field also rendered its mask as value, so replacing a stored password silently saved the bullets as part of the new one; it is a placeholder now, on desktop and the phone admin.
* feat(plugin-sdk): settings-field default + unknown-attribute warning
* feat(shared): default on the plugin settings-field contract
* feat(server): persist and serve plugin settings-field defaults
* feat(client): pre-fill plugin settings defaults in the admin and user forms
* feat(server): enforce required plugin settings on both save paths
* feat(client): block plugin-settings save on an empty required field
* feat(server): resolve plugin settings defaults at runtime
A settings-field `default` was stored and served to the form but never
reached the plugin: the child's init envelope, `ctx.settings.get()`, the
channel dispatch's readAll and the OAuth provider config all read the
stored blob only. Fold `stored ?? default` in on every path that hands
config to a plugin (settings-defaults.ts), and let a default satisfy
`required` on both save gates — refusing a save the runtime would have
accepted was a contradiction.
Also: one `isFilled` rule shared by the save gate and `hasRequired`
(checkbox exempt, whitespace empty); the install parser drops a
non-boolean checkbox default or a select default outside its options,
matching what the SDK already refuses; export SETTING_FIELD_KEYS so the
SDK's mirror can be parity-tested.
* feat(client): share the plugin settings form rules across all three forms
The mobile user settings screen was a third consumer of the fields/config
contract that never got `default` pre-fill or the required check, and
the seed/patch loops lived in five hand copies. Extract
seedSettingsValues / findMissingRequired / settingsPatch into one module
and use it from the admin instance form, the desktop user tab and the
mobile screen. The user forms now surface a 4xx `{ error }` from the
server (the required-field refusal) instead of the generic toast.
* feat(plugin-sdk): seed dev fixtures from manifest defaults
`trek-plugin dev` handed the plugin only the fixture values, so a plugin
relying on its manifest defaults read undefined in dev and only in dev.
Seed ctx.config / ctx.settings from the declared defaults (a fixture
value still wins) via a new exported settingDefaults(manifest, scope).
SETTING_FIELD_KEYS is now parity-tested against the host's list.
* docs(plugins): document settings defaults as runtime fallbacks
A `default` is the field's effective value wherever nobody set one, not
just a form pre-fill; `required` is enforced on both ends; unknown
attributes are dropped at install and warned on by validate. Add a
Settings section with the attribute table to the SDK README and update
the ctx type comments, the shared schema, and the wiki guide.
* add missing exports
The quality gate came in at 79.5% coverage on new code against a floor of 80%.
The gaps were real ones rather than a threshold quibble: three pieces of new
behaviour had no test at all.
- `useBagTotalsPing` was a new file with no test, and it holds the logic that
stops one write turning into a storm of listBags requests on every connected
client: the coalescing window, the in-flight guard, and the ping that arrives
mid-flight and must still be served.
- The Journey Settings tracks switch had no test, including the part the review
caught: that it refreshes rather than calling onSaved, which would close the
dialog and drop unsaved edits.
- Neither bag panel had ever been rendered in a test. They now pin what the
panels actually print: the server total rather than the sum of what the
viewer can see, the grand total following the same rule as the rows above it,
and the offline fallback.
Also covers the two ping call sites that were still unmeasured, the MCP
apply-template path and the plugin RPC bag delete.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01S5VwmiBQ9sPCJwfmpv64r3
Three plugin-RPC tests and the MCP bulk-import test assert the exact list of
broadcasts a write produces, so adding the #2191 weight ping to those two
surfaces made them fail — correctly: they are the contract for who hears what.
Updated to include it, and asserting it rather than just tolerating it. The
private-item cases are the interesting ones: the item event still reaches only
its viewers while the weight ping goes to the whole room, which is precisely
what makes a content-free ping safe there.
The bulk import gets one ping for the whole import rather than one per item.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01S5VwmiBQ9sPCJwfmpv64r3
An adversarial review of the diff found sixteen defects that survived
verification. The substantive ones:
- The new `packing:bag-totals` event was never classified in wsEventPolicy,
so the registry-parity ratchet was red. It belongs in
HANDLED_OUTSIDE_TRIP_STORE: it carries no state to apply, which is the
point of it.
- The Journey Settings tracks switch called onSaved(), which closes the
dialog. That destroyed the switch the moment it was flipped AND skipped
the unsaved-changes guard, so a title typed but not saved was silently
lost. It refreshes instead.
- Deleting a bag sends no totals ping, yet packing_items.bag_id is
ON DELETE SET NULL: everything in the bag lands in the unassigned pile and
both figures move. The comment claiming otherwise was wrong.
- Only the REST handlers pinged. The MCP tools and the plugin RPC are two
more live write paths, and since the displayed number now comes from the
server rather than from the item events, weights simply stopped updating
for anything written through them.
- Every item write pinged, including checking an item off — the most
frequent packing write there is — and clearing twelve checked items sent
twelve. The ping now fires only when a weight-relevant field changed, and
the clients coalesce bursts and never overlap two refetches.
- Bags have no repo and no Dexie table, so offline the server totals froze
at the last online read and went blind to the mutation queue. A stale
absolute number measured against an airline limit is worse than an honest
partial one, so offline the surfaces sum what they can see.
- SIGTERM arriving mid-bootstrap called server.close() on undefined, and the
fire-and-forget promise turned that into an unhandled rejection: on Node 22
a crash, which is the opposite of the fix.
- Promise.allSettled swallowed a rejected nestApp.close(), reporting a clean
exit 0 while the crons and the plugin children were never torn down. This
bug class is diagnosed from container logs.
Also: the track fetch gate read a stale `current` during journey-to-journey
navigation and fetched the previous journey's tracks; listBags ran the weight
aggregate twice per request, on a route the ping now fires constantly; the
grand total counted unassigned weight whose row was not rendered; the client
declared show_trip_tracks boolean while the wire carries SQLite 0|1; and the
six broadcast calls had no test, so removing all of them left the suite green.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01S5VwmiBQ9sPCJwfmpv64r3
A bag's weight was summed in the browser over the item list the viewer is
allowed to see. Personal items belonging to other members are filtered out of
that list by design (#858), so their weight vanished with them: two people
looking at the same shared bag saw two different totals, and neither saw the
real one.
That number is then measured against `weight_limit_grams`, an absolute airline
limit. A per-viewer subtotal there is not merely incomplete but wrong — a
shared bag could sit over its limit and warn nobody.
The sum now happens in SQL over every row and ships as `total_weight_grams` on
the bag, with `unassigned_weight_grams` alongside so the "no bag" pile and the
grand total follow the same rule; a screen mixing true totals with per-viewer
ones would be worse than either. Only integers cross the wire: a member learns
that a bag is heavier than the items they can see, never a name, category,
quantity or owner. That disclosure is deliberate and bounded, and it is what
the issue asks for.
The item lists stay viewer-scoped, so #1767 ("someone shared it with you, but
they are the one bringing it") is unchanged — it just no longer decides what a
bag weighs.
Totals also had to stay live, because the reported repro is two people looking
at the same screen. A private item's event reaches only its owner, which is the
very rule that made the totals wrong, so item writes now also ping the room
with a content-free `packing:bag-totals` and everyone re-reads the numbers.
All three surfaces — the bag sidebar, the bag modal and the mobile sheet — read
the server figure and fall back to the local sum only for a bag cached offline
before this change.
Closes#2191
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01S5VwmiBQ9sPCJwfmpv64r3
#1260 drew the GPX tracks of every trip behind a journey onto its map, with
nothing to switch off — the reasoning being that a route imported into a
linked trip is part of the journey's story. For a trip carrying a season of
recorded drives it is instead a map nobody asked for, drawn from places that
never became an entry.
So it becomes a journey-level setting, owner-only like the other Journey
Settings, and it starts off: a journal should show what its author put in it,
and #1260's set is everything the linked trips happen to contain. The switch
gates the request, not just the rendering — the endpoint ships full unthinned
geometry, which is megabytes fetched on every journey open for a map nobody
wanted.
`journeyTracks()` itself is deliberately untouched, so TREK Studio keeps the
set its saved books were laid out with.
Also fixes what the reporter's screenshot actually shows: JourneyMap fed track
coordinates into `fitBounds`, so one drive across a country zoomed the journey
map out until the entries were specks. Tracks now only frame the map when
there is nothing else to frame, which is what JourneyMapGL already did — the
same journey used to open at two different zooms depending on the renderer.
Closes#2194
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01S5VwmiBQ9sPCJwfmpv64r3
Every `docker stop` of a TREK instance with a browser attached exited 137,
exactly ten seconds after the stop.
`http.Server.close()` stops new connections and then waits for the open ones
to end by themselves — and a WebSocket never does. TREK holds one per open
tab, so the close callback never ran. `closeDb()` and `process.exit(0)` lived
inside that callback, which left the only remaining exit a fallback timer set
to 10000ms: exactly when Docker sends SIGKILL. Docker starts counting when it
sends the signal and Node only once the handler is dispatched, so that timer
started strictly later and could never win. It was unreachable code in every
default deployment.
Waiting for Nest was not an option either: `WsAdapter.close()` is the one
place that terminates clients, and Nest reaches it only after every
onModuleDestroy hook has resolved — including the plugin supervisor's SIGTERM
grace for its forked children. The sockets holding the server open would be
hostage to the slowest hook in the container.
So the sockets are now released deliberately: ws clients get 1001 "going
away", idle keep-alive sockets are dropped at once, and anything still
attached after a 2s drain is destroyed. Nest is closed alongside that, the
database is closed on both exit paths rather than only the unreachable one,
and the fallback fires at 5s — inside Docker's grace, so a forced exit is
still ours.
The sequence moved into src/shutdown.ts behind injected dependencies. It was
untestable by construction before, which is part of why this went unnoticed;
it now has nine tests, including the signal that beats the bootstrap.
Closes#2193
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01S5VwmiBQ9sPCJwfmpv64r3
The three stacked sections become one grid of cards per group, with the
type shown on the tile itself. Bag tracking, the collab features and the
photo providers hang off their parent addon as compact shelf rows
instead of full-width siblings, and each group header counts how many of
its addons are on.
Along the way:
- naver_list_import, airtrail and llm_parsing get catalog entries, so
their names and descriptions translate like everyone else
- dark detection moves into a shared useIsDark hook that watches the
.dark class instead of re-deriving settings + matchMedia locally
- the ToggleSwitch knob uses --accent-text so a light accent in dark
mode no longer renders white-on-white
It has no trip tab and no page of its own — it pulls places from an
external service, which is what the integration group is for. The type
only drives grouping in the admin UI (and the global nav check), so
existing installs just see the tile move; migration appended for them,
seed fixed for fresh ones.
Closes#2151
The desktop Costs view mixes recorded settle-up payments into the day-grouped
expense ledger, but the mobile screen only ever fetched balances/flows and
silently dropped the settlements array the server already returns — a
recorded payment cleared the balance but then left no trace anywhere on
mobile.
The store boots from localStorage, but 'app_language' is only written
on an explicit in-app choice; a language that lives in the account's
server settings never reaches the device. A PWA cold start with no
network then runs in English even though every online session ran in
the user's language, the same stranding #1618 fixed for currency and
units.
loadSettings now mirrors the fetched language into its own key
('app_language_server'), and the boot chain falls back to it between
the explicit choice and 'en'. A separate key on purpose: 'app_language'
means an explicit choice, and the login page's detection chain must
keep running for users who never made one.
Closes#2153
Bucket-list marker tooltips rendered as a single nowrap line, so long
notes overflowed off both edges of the viewport instead of wrapping.
Wrap and cap the tooltip's size, add a scrollbar for notes that still
exceed it (only shown when content actually overflows, and scoped to
this tooltip so short region/country tooltips are unaffected), and
make the tooltip interactive with a delayed close so the pointer can
reach it to scroll instead of Leaflet closing it the instant it leaves
the marker's tiny hit area. Also compute the tooltip's direction and
offset from the marker's on-screen position before it opens, so it
flips below the marker and stays clear of the left/right edges when
there's no room to render it centered above.
The positioning/scroll math is pure and covered by unit tests in
atlasModel.test.ts. Updated the L.marker mocks in AtlasPage.test.tsx
and useAtlas.test.tsx with on/off/getTooltip/closeTooltip so the new
marker event wiring doesn't crash under test.
* chore: bump version to 4.1.0 [skip ci]
* fix(collections): stop findDuplicateCollectionPlace merging places by coordinates alone
Three places answered "is this a place the trip already has?", not two:
PlacesService.findDuplicatePlace and isPlaceDuplicate got the shared strategy
list in #2130; CollectionsService.findDuplicateCollectionPlace was missed and
kept its own hand-rolled copy of the pre-#2130 order.
Unlike findDuplicatePlace, this one had no isPlaceDuplicate() guard in front of
it — savePlace calls it directly — so the coordinate fallback for a NAMED
candidate was live, not latent: two distinct places sharing an address (a
restaurant and the bar upstairs) would report each other as duplicates the
moment their coordinates round to the same ~11 m box, regardless of name.
It also never read google_place_id/google_ftid/osm_id at all, even though
every collection_places row stores them, so a place renamed since it was
saved could be saved again instead of being recognised by its provider id.
Now walks placeMatchStrategies() from @trek/shared, same as
findMatchingPlaceId: provider id, then name, then coordinates only when there
is no name. savePlace passes the provider id fields through since they are
already on the request body; the two bulk call sites (importablePlaces,
saveFromTripPlaces) are unchanged, since their source rows don't carry
provider ids to pass.
Closes#2137
* fix(collections): ask the same question at all three call sites
savePlace was not the only caller of findDuplicateCollectionPlace. The bulk copy
already carries google_place_id/google_ftid/osm_id into the row it writes but
asked without them, and the import picker never selected them at all, so a
renamed place was recognised on one path and not on the other two. The picker
answering differently from the import is the drift the method exists to prevent:
a row shown as new would come back refused.
COLLECTIONS-SVC-102 and -103 cover the bulk copy and the picker.
* fix(map): give the Leaflet map its own zoom ceiling, so a vector basemap cannot take the planner down
Opening a trip on 4.1.0 could land on the error boundary with "Map has no maxZoom
specified", and so could the map settings tab, which left no screen from which to
pick a different basemap.
Leaflet answers getMaxZoom() from the map options first and otherwise from a
layer that brought one, and only a GridLayer ever contributes via its beforeAdd
hook. A vector basemap is a GL canvas on a plain L.Layer, so it contributes
nothing, and MapContainer set no ceiling of its own. MarkerClusterGroup.onAdd
refuses an infinite one by throwing, before any clustering, so an empty trip hit
it too.
That path opened when the default basemap became a MapLibre style. It reaches
further than an operator choosing OpenFreeMap: resolveTileUrl sends a keyless
CARTO template to the app default, which is now a vector style, so every
instance still carrying the old CARTO basemap fell into it on upgrade.
The ceiling belongs on the map rather than on the base layer, because it has to
hold whichever of the three branches renders. 19 matches what the raster and
satellite layers already carried, so nothing changes for the maps that worked.
SharedTripPage gets the same treatment: it has no cluster to throw, but it draws
the same vector basemap and its fitBounds asks for a ceiling.
mapZoomCeiling.test.ts pins Leaflet's own semantics against the real library
rather than a mock, since a mocked react-leaflet would have kept passing through
all of this.
* fix(map): put the CARTO key back on the template the settings preview draws
The map settings fields hold what the user is editing rather than what
useTileUrl already resolved, so the preview received a bare template with no key
on it. resolveTileUrl then read it as a keyless CARTO url and did what it is
meant to do with one: fall back to the app default. The preview drew
OpenFreeMap while the fields said CARTO, whether or not a key was saved.
Same reason the tab crashed for anyone on a CARTO basemap even after entering a
key, which is the loop that made this hard to escape: the key is entered on the
page the fallback had taken down.
FE-COMP-MAP-033b pins the template the preview is handed, with and without a key.
* fix(admin): put the CARTO key back on the default-settings map preview too
Same shape as the user-facing map tab, and reported alongside it: the field holds
what the admin is editing, so the preview resolved a keyless CARTO template and
drew the app default instead of the basemap being set.
* fix(mobile): put the CARTO key back on the phone map settings preview too
The third preview with the same shape, and the one that matters most while the
crash is live: the phone layout is where someone would go to change the basemap
when the desktop tab will not open.
* fix(calendar): a fully timed stay is its two hand-overs, not a week-long bar
Closes#2136
A stay was always exported as an all-day block across its whole range, and the
check-in and check-out markers came on top of it. When the stay records both
ends of its clock the markers already say everything the block does, and the
block is the half nobody can act on: it takes a week of calendar to repeat what
two one-hour events state precisely. Same shape as the parking and rental
windows in #2068.
Only where the markers actually stand in for the booking, though. They are
emitted once per stay and titled from its lowest-id reservation, so a second
room on the same stay would otherwise be left with nothing at all. A stay that
knows only one end keeps its block too, since nothing else carries the other
end's date.
CAL-025b covers the drop, -025c the one-sided stay, -025d the second room.
* fix(mobile): the phone admin map preview needs the CARTO key too
The fourth preview with the same shape, missed when the other three were fixed.
Sonar pointed at it: the desktop and phone admin pages are near-identical, so the
duplication report named the file the change had not reached.
* chore: bump version to 4.1.1 [skip ci]
* feat(plugins): admin editor for instance-wide plugin settings
The scope:'instance' settings a plugin declares had API endpoints but no UI.
- shared: plugins.schema.ts — the settings-field descriptor and the admin
instance-config GET/PUT wire contracts, imported by both sides
- server: GET /api/admin/plugins/:id/config now returns the declared fields
alongside the (masked) values; PUT re-spawns an ACTIVE plugin (its child
reads config once, at init) and reports it via 'restarted'; the admin list
carries instanceSettingsCount so the UI gates its menu item without a fetch
- client: 'Instance settings' row action opening a form modal (desktop) /
MSheet (phone) in both admin shells — checkbox/select/secret rendering as
in the user settings form; an untouched secret mask is never sent back
- i18n: three new admin.plugins.* keys in every locale
* refactor(plugins): move the admin body contracts to @trek/shared
The remaining plugin DTO schemas (install, link, activate, uninstall,
retrust, update, egress-hosts, user-settings update, route) now live in
shared/src/plugins/plugins.schema.ts with their inferred types, alongside
the instance-config contracts. The journey-contract doctrine notes (#1842)
moved with them — each body stays deliberately loose so the handlers keep
the rejections and fallbacks they already own, and new spec tests pin that
looseness so a well-meant tightening breaks a test that explains itself.
plugins.dto.ts is now thin createZodDto wrappers over the shared schemas
(same class names, no controller changes), and the client's pluginInstall
types its options and body against PluginInstallRequest.
* fix(admin): offer Allowed hosts only to plugins that declared operatorEgress
The row's egress chip was already gated on the declaration, but the kebab
menu (desktop) and row action sheet (phone) offered Allowed hosts on every
installed plugin — inviting the admin to widen egress for plugins that
never asked for it. Both entries now render only when the plugin declares
operatorEgress; the dialog's unsupported-notice fallback stays as the
truth-teller for stale list data.
* test(admin): close the Sonar gaps on the instance-settings form
The PR gate flagged new-code coverage (76.4%) and duplication (10.9%).
- The two identical 39-line handler blocks were logic, not markup — they
move into one shared useInstanceSettings hook; each shell keeps its own
markup (the deliberate desktop/phone divergence stays)
- New tests per shell cover the untested branches: every field type
renders and round-trips into the save payload (select, checkbox/MToggle,
number, hint, no-label key fallback), a failed fetch is a toast, a
rejected save shows the server reason and keeps the edits, and the
dialog/sheet closes without saving
- INS-009 covers the controller's no-body '|| {}' fallback
---------
Co-authored-by: github-actions[bot] <github-actions[bot]@users.noreply.github.com>
Co-authored-by: michael-bohr <263459669+michael-bohr@users.noreply.github.com>
Co-authored-by: Maurice <mauriceboe@icloud.com>
tags and pros_cons are JSON held in TEXT columns. Three read paths decoded
them by hand and createEntry and updateEntry did not, so an edit answered
with the raw row: tags came back as '["beach","sunset"]', the store spread
it into an entry the client types as string[], and the journey page threw
"tags.map is not a function" behind an error boundary.
Any edit was enough to arm it. The update path writes JSON.stringify(val)
with no length check, so even clearing an entry's tags stores '[]' and
answers with that string.
Neither compiler could see it: JourneyEntry describes the row, tags and
all, and both sides read the same field name while only one of them was
right. So the decoded shape gets a type of its own and one decoder, which
is what turned up the third case - the early return for an update that
changes nothing handed back a raw row too.
Fixes#2085
The fourth preview with the same shape, missed when the other three were fixed.
Sonar pointed at it: the desktop and phone admin pages are near-identical, so the
duplication report named the file the change had not reached.
Closes#2136
A stay was always exported as an all-day block across its whole range, and the
check-in and check-out markers came on top of it. When the stay records both
ends of its clock the markers already say everything the block does, and the
block is the half nobody can act on: it takes a week of calendar to repeat what
two one-hour events state precisely. Same shape as the parking and rental
windows in #2068.
Only where the markers actually stand in for the booking, though. They are
emitted once per stay and titled from its lowest-id reservation, so a second
room on the same stay would otherwise be left with nothing at all. A stay that
knows only one end keeps its block too, since nothing else carries the other
end's date.
CAL-025b covers the drop, -025c the one-sided stay, -025d the second room.
The third preview with the same shape, and the one that matters most while the
crash is live: the phone layout is where someone would go to change the basemap
when the desktop tab will not open.
Same shape as the user-facing map tab, and reported alongside it: the field holds
what the admin is editing, so the preview resolved a keyless CARTO template and
drew the app default instead of the basemap being set.
The map settings fields hold what the user is editing rather than what
useTileUrl already resolved, so the preview received a bare template with no key
on it. resolveTileUrl then read it as a keyless CARTO url and did what it is
meant to do with one: fall back to the app default. The preview drew
OpenFreeMap while the fields said CARTO, whether or not a key was saved.
Same reason the tab crashed for anyone on a CARTO basemap even after entering a
key, which is the loop that made this hard to escape: the key is entered on the
page the fallback had taken down.
FE-COMP-MAP-033b pins the template the preview is handed, with and without a key.
Opening a trip on 4.1.0 could land on the error boundary with "Map has no maxZoom
specified", and so could the map settings tab, which left no screen from which to
pick a different basemap.
Leaflet answers getMaxZoom() from the map options first and otherwise from a
layer that brought one, and only a GridLayer ever contributes via its beforeAdd
hook. A vector basemap is a GL canvas on a plain L.Layer, so it contributes
nothing, and MapContainer set no ceiling of its own. MarkerClusterGroup.onAdd
refuses an infinite one by throwing, before any clustering, so an empty trip hit
it too.
That path opened when the default basemap became a MapLibre style. It reaches
further than an operator choosing OpenFreeMap: resolveTileUrl sends a keyless
CARTO template to the app default, which is now a vector style, so every
instance still carrying the old CARTO basemap fell into it on upgrade.
The ceiling belongs on the map rather than on the base layer, because it has to
hold whichever of the three branches renders. 19 matches what the raster and
satellite layers already carried, so nothing changes for the maps that worked.
SharedTripPage gets the same treatment: it has no cluster to throw, but it draws
the same vector basemap and its fitBounds asks for a ceiling.
mapZoomCeiling.test.ts pins Leaflet's own semantics against the real library
rather than a mock, since a mocked react-leaflet would have kept passing through
all of this.
savePlace was not the only caller of findDuplicateCollectionPlace. The bulk copy
already carries google_place_id/google_ftid/osm_id into the row it writes but
asked without them, and the import picker never selected them at all, so a
renamed place was recognised on one path and not on the other two. The picker
answering differently from the import is the drift the method exists to prevent:
a row shown as new would come back refused.
COLLECTIONS-SVC-102 and -103 cover the bulk copy and the picker.
Three places answered "is this a place the trip already has?", not two:
PlacesService.findDuplicatePlace and isPlaceDuplicate got the shared strategy
list in #2130; CollectionsService.findDuplicateCollectionPlace was missed and
kept its own hand-rolled copy of the pre-#2130 order.
Unlike findDuplicatePlace, this one had no isPlaceDuplicate() guard in front of
it — savePlace calls it directly — so the coordinate fallback for a NAMED
candidate was live, not latent: two distinct places sharing an address (a
restaurant and the bar upstairs) would report each other as duplicates the
moment their coordinates round to the same ~11 m box, regardless of name.
It also never read google_place_id/google_ftid/osm_id at all, even though
every collection_places row stores them, so a place renamed since it was
saved could be saved again instead of being recognised by its provider id.
Now walks placeMatchStrategies() from @trek/shared, same as
findMatchingPlaceId: provider id, then name, then coordinates only when there
is no name. savePlace passes the provider id fields through since they are
already on the request body; the two bulk call sites (importablePlaces,
saveFromTripPlaces) are unchanged, since their source rows don't carry
provider ids to pass.
Closes#2137
The fourth preview with the same shape, missed when the other three were fixed.
Sonar pointed at it: the desktop and phone admin pages are near-identical, so the
duplication report named the file the change had not reached.
Closes#2136
A stay was always exported as an all-day block across its whole range, and the
check-in and check-out markers came on top of it. When the stay records both
ends of its clock the markers already say everything the block does, and the
block is the half nobody can act on: it takes a week of calendar to repeat what
two one-hour events state precisely. Same shape as the parking and rental
windows in #2068.
Only where the markers actually stand in for the booking, though. They are
emitted once per stay and titled from its lowest-id reservation, so a second
room on the same stay would otherwise be left with nothing at all. A stay that
knows only one end keeps its block too, since nothing else carries the other
end's date.
CAL-025b covers the drop, -025c the one-sided stay, -025d the second room.
The third preview with the same shape, and the one that matters most while the
crash is live: the phone layout is where someone would go to change the basemap
when the desktop tab will not open.
Same shape as the user-facing map tab, and reported alongside it: the field holds
what the admin is editing, so the preview resolved a keyless CARTO template and
drew the app default instead of the basemap being set.
The map settings fields hold what the user is editing rather than what
useTileUrl already resolved, so the preview received a bare template with no key
on it. resolveTileUrl then read it as a keyless CARTO url and did what it is
meant to do with one: fall back to the app default. The preview drew
OpenFreeMap while the fields said CARTO, whether or not a key was saved.
Same reason the tab crashed for anyone on a CARTO basemap even after entering a
key, which is the loop that made this hard to escape: the key is entered on the
page the fallback had taken down.
FE-COMP-MAP-033b pins the template the preview is handed, with and without a key.
Opening a trip on 4.1.0 could land on the error boundary with "Map has no maxZoom
specified", and so could the map settings tab, which left no screen from which to
pick a different basemap.
Leaflet answers getMaxZoom() from the map options first and otherwise from a
layer that brought one, and only a GridLayer ever contributes via its beforeAdd
hook. A vector basemap is a GL canvas on a plain L.Layer, so it contributes
nothing, and MapContainer set no ceiling of its own. MarkerClusterGroup.onAdd
refuses an infinite one by throwing, before any clustering, so an empty trip hit
it too.
That path opened when the default basemap became a MapLibre style. It reaches
further than an operator choosing OpenFreeMap: resolveTileUrl sends a keyless
CARTO template to the app default, which is now a vector style, so every
instance still carrying the old CARTO basemap fell into it on upgrade.
The ceiling belongs on the map rather than on the base layer, because it has to
hold whichever of the three branches renders. 19 matches what the raster and
satellite layers already carried, so nothing changes for the maps that worked.
SharedTripPage gets the same treatment: it has no cluster to throw, but it draws
the same vector basemap and its fitBounds asks for a ceiling.
mapZoomCeiling.test.ts pins Leaflet's own semantics against the real library
rather than a mock, since a mocked react-leaflet would have kept passing through
all of this.
savePlace was not the only caller of findDuplicateCollectionPlace. The bulk copy
already carries google_place_id/google_ftid/osm_id into the row it writes but
asked without them, and the import picker never selected them at all, so a
renamed place was recognised on one path and not on the other two. The picker
answering differently from the import is the drift the method exists to prevent:
a row shown as new would come back refused.
COLLECTIONS-SVC-102 and -103 cover the bulk copy and the picker.
Three places answered "is this a place the trip already has?", not two:
PlacesService.findDuplicatePlace and isPlaceDuplicate got the shared strategy
list in #2130; CollectionsService.findDuplicateCollectionPlace was missed and
kept its own hand-rolled copy of the pre-#2130 order.
Unlike findDuplicatePlace, this one had no isPlaceDuplicate() guard in front of
it — savePlace calls it directly — so the coordinate fallback for a NAMED
candidate was live, not latent: two distinct places sharing an address (a
restaurant and the bar upstairs) would report each other as duplicates the
moment their coordinates round to the same ~11 m box, regardless of name.
It also never read google_place_id/google_ftid/osm_id at all, even though
every collection_places row stores them, so a place renamed since it was
saved could be saved again instead of being recognised by its provider id.
Now walks placeMatchStrategies() from @trek/shared, same as
findMatchingPlaceId: provider id, then name, then coordinates only when there
is no name. savePlace passes the provider id fields through since they are
already on the request body; the two bulk call sites (importablePlaces,
saveFromTripPlaces) are unchanged, since their source rows don't carry
provider ids to pass.
Closes#2137
The 4.0.0 notice was scoped to a 4.0.x window on the assumption that each
release would bring its own entry. 4.1.0 shipped without one, so the modal
switched itself off and nothing took its place: the thank-you notice it
replaced is retired below 4.0.0, leaving a 4.1 install with no notice at all.
Hold it open to the 5.0.0 boundary instead. What it introduces is the 4.x
line, which is still what somebody arriving on 4.1 or 4.2 is meeting for the
first time, and a later major gets its own entry the way this one did.
A flight leaving overnight contributes only its departure airport on the day
it leaves, and the evening hotel bookend was attaching to whatever point sat
at the day's edge. So the map drew a straight line from the airport you took
off from to the hotel you were checking into.
The same blind spot on the run builder produced the long rays: a day's order
is a bare wall clock, so an inbound long-haul whose departure clock reads late
sorts after the day's local stops, and its far departure airport got appended
to that run. OSRM answers such a pair with NoRoute and the fallback straightens
the whole chunk, taking the genuine local drive with it.
Both now ask whether the pair could be driven at all. A flight, train, ferry,
cruise or coach carries you out of the day's geography, so its departure point
has no drive back to tonight's hotel and its arrival point had no drive from
this morning's. A hire car is the opposite and stays out of that set: you keep
driving it, so a multi-day rental's pickup and drop-off points keep their hotel
legs. Alongside it, any leg touching a booking endpoint has to be within
MAX_DRIVE_KM, which catches the mis-sorted edges the role alone cannot see.
The sidebar gets both guards so its hotel connectors stay in step with the
drawn route; the mobile map and timeline read the same hook.
The scope that turns on third-party tool execution was the one missing from the
scope reference, and the count beside it was stale. It says what the scope does
not do, since the boundary is the admin's mcp:tools grant rather than anything on
the token.
`.meta()` advertises every node of the declaration; `nodeToZod` only walks the
ones it can type. Everything below follows from that gap, so the keyword
allowlist moves into `sanitiseNode`, which is the one walker that sees every
node.
A `pattern` was compiled straight into a RegExp and run on the main thread
against caller input, with no timeout and nothing to abort it: `^(a+)+$` against
33 characters stops the whole server, not just MCP. That is an escape from the
supervised child into the host event loop, and it does not take malice, since a
hand-written email pattern is the classic accident. Quantified groups that repeat
or alternate inside are refused, and the source is capped.
`mcp:tools` counted as a side effect, so `readOnlyHint` was structurally
unreachable: holding that grant is what lets a plugin advertise a tool at all.
Every plugin tool was therefore advertised as potentially destructive and every
client's auto-approval was off. It is a host entry point like `hook:*`, and is
classified as one.
Property names, `default` and `examples` were the strings that never met the
sanitiser. A key may carry newlines, so `\n## SYSTEM\n...` in one defeated the
newline collapse everything else goes through. Names are identifiers now; sample
values keep their own keys but have their strings bounded.
Also refused at parse time, which is what the rest of the file already does: an
object `enum` member carrying `$ref` past the deep scan, a non-boolean
`additionalProperties` advertised and checked by nothing, `enum` beside
`properties` skipping the children, a root `enum`/`const`/`nullable` producing a
non-object schema the SDK cannot read, and a draft-4 `exclusiveMinimum: true`
dropped from the check while staying in the advertisement.
`required` used `in`, so a field named `toString` read as present on every call.
Nodes are built with a null prototype, and the three names Zod cannot round-trip
are refused rather than silently lost.
A result block that had to be shortened got no marker and no log line, which
reads to the model like the whole answer. It is counted now, like a dropped one,
and `isError` survives every envelope path.
The security test file carried two raw NUL bytes, so git, GitHub and ripgrep all
treated it as binary: no diff, no blame, no review. Escaped, same runtime value.
Found in review of the branch. Each one is a case where the host told the
model or the admin something it did not enforce.
Advertised schemas are now enforced. buildToolInputSchema built a loose
object that checked only which keys were present, so a tool advertising
an integer, an enum and additionalProperties: false accepted a string, an
unlisted value and an extra property. .meta() still advertises the
declaration verbatim, and the schema underneath it is now built from that
same declaration, so the two cannot disagree. Anything the builder cannot
enforce (oneOf, unknown formats, non-primitive enum members) is refused by
parseManifest rather than advertised, and the failure lands at install
time where an admin sees it.
Writing the test for that turned up a live bug: z.custom() cannot be
represented in JSON Schema, so building enums that way would have thrown
inside tools/list for the whole session. Enums and consts are real literal
types now.
readOnlyHint classification fails closed. The side-effect list spelled out
db:write: and missed db:create:trips entirely, letting a plugin that can
create trips advertise a tool as read-only and idempotent, which is what
clients gate auto-approval on. It is an allowlist of read-only grants now,
and mcp-tool-annotations.test.ts walks the whole catalog so a new
permission fails a test rather than producing a wrong hint.
The result cap covers the whole result. It bounded each text block, so a
hundred blocks cost a hundred times the limit: 6.5M characters in a probe,
now 65K. Large objects are serialised with a budget instead of stringified
whole and then sliced, and the block count is capped.
The admin tool list was dead code. The registry preview carries only
widget and tripPage, so the section added for it never rendered, and the
UI hid the description behind the title when both existed. The preview
carries sanitised tools now (previewManifest is exported and tested, which
is how the gap survived), and both fields render, description always.
plugins:use is out of the DCR default. Adding it to ALL_SCOPES meant a
registration that names no scopes got it and had it pre-selected at
consent. Defaults come from DEFAULT_CLIENT_SCOPES now; asking for it by
name still works.
The public ManifestCapabilities gained mcpTools, so a typed manifest can
express what the runtime already validates.
MCPANN-001..005, MCPENF-001..010, MCPPREV-001..005, MCPTOOLS-020..022,
SDKP-015b/c.
The manifest declaration only becomes an admin-visible control if the
admin can actually read it. Nothing rendered capabilities.mcpTools, so
approving mcp:tools meant approving text you would first meet inside a
chat.
The pre-install dialog now lists the tool names and descriptions, beside
the existing access and egress sections, and the grant gets an
at-a-glance chip like every other capability.
Desktop and mobile both, since the panels are parallel copies.
Integration rather than unit, because what is being proved is that
nothing reaches the global exception filters. registerTools is called
outside the transport's try block, and mcp-transport.service.ts states
that every response there is written by hand and nothing may be thrown,
so a registry-level unit test cannot show it.
MCP-P15: a contributed tool that takes a built-in's name is skipped, the
built-in is the one still listed, and the session serves normally.
MCP-P16: a source that throws costs its tools and nothing else, where it
would previously have been a 500 on every /mcp initialize.
MCP-P17: a contributed tool call writes the same mcp.tool_call audit row
as a built-in, which is the whole reason the source hangs off
McpAttachOptions rather than the host calling registerTool itself.
MCP-P18 came out of writing the others: the first draft of MCP-P17 minted
a trips:read token and got no audit row, because the declarative access
marker had correctly hidden the tool. That is worth pinning.
mintOauthToken takes scopes now, and rpcResult parses the SSE frame the
transport answers with.
The advertised tool surface is frozen at initialize, so anything that
changes which plugin tools exist has to end the sessions built before it.
tools/list_changed is not an alternative: no event store is configured,
and the surface would not be re-read anyway.
Unconditional, no diffing. The surface is per-session, since it depends
on the caller's scopes and the plugin's grants, so there is no ctx-free
set to compare against.
Five routes, mirroring what admin.controller.ts already does for addons:
activate, deactivate, update, retrust and uninstall. activate(id,
consent), update and retrust are also the grant-change paths, so "on an
mcp:tools grant change" needs nothing extra. A failed call invalidates
nothing, because nothing changed.
The addon cascade needed its own call. Disabling an addon can stop a
plugin that publishes tools, and the addon doing the cascading is not
necessarily one of MCP_RELEVANT_ADDONS, so mcpAffected does not cover it.
Still gated on a plugin actually stopping, which keeps the #1414 rule
that a config-only save kills nothing.
Imported from sessionManager rather than the src/mcp barrel, which
evaluates readEnv().mcp at module scope.
MCPINV-001 to MCPINV-007.
The advertise path, shaped like trip-warnings.mcp.ts: gate on the kill
switch, fan out over providersOf, let a failing provider contribute
nothing rather than fail the surface.
A tool is advertised only if the signed manifest declares it AND the
running build reported implementing it, the same two-sided check
callPlugin makes for exports. The manifest half is what makes the text
auditable: the child composes its loaded payload after onLoad, so without
it a sha256-pinned artifact could change what it publishes on every
restart with no re-consent.
The result envelope is built here, always. The registry returns a
handler's value verbatim and skips output validation without an
outputSchema, so a plugin returning {ok: true, data} would emit a result
with no content key and every conforming client would error. A failed or
timed-out invoke becomes isError rather than a JSON-RPC fault, so the
model can see it and try something else.
Gates in order: kill switch, then demo mode. The plugins domain had no
demo call site before, and the child has none of its own.
The service owns the process-level source rather than PluginRuntimeService
beside the other sinks, because it needs PluginHooks and PluginHooks
injects PluginRuntimeService. It clears the source on destroy, which the
existing setPluginChannelSource forgets to do: a torn-down buildApp()
would otherwise leave a live source closing over a dead runtime and bleed
into the next suite.
mcpToolCapabilities is a parallel reader rather than a capabilityList
generalisation, since that one returns string[]. It re-runs the manifest
parser instead of trusting the stored blob, which was written by whatever
parser was current at install time and whose caps are the security
control.
Two caps not in the brief: MCP_TOOLS_TOTAL_MAX across all plugins, since
8 each times twenty installed plugins would put 160 tools in front of a
model already choosing from ~200. Truncation is logged, because a
truncated surface reads exactly like a plugin that published nothing.
MCPTOOLS-001 to MCPTOOLS-019.
The child adds the tool names its mcpToolProvider hook implements to the
loaded payload, beside routes/jobs/exports, and the supervisor caches
them on Supervised like the rest. The hook itself already rides the
existing Object.keys(def.hooks) line, so the "did the child report the
hook too" half was free.
Declarative and synchronous: the host never asks the child what tools it
has. It intersects this list with the signed manifest, so a build that
reports a tool the manifest never declared is ignored rather than
trusted.
grantsOf comes along because the advertise path has to clamp what a
tool's annotations may claim against what the plugin was actually
granted, and sup.granted is already in memory. A per-tool DB read would
be the alternative.
plugin-host-entry.ts is excluded from coverage (it runs in a forked
subprocess), so the two supervisor integration cases are the only thing
proving any of this: the loaded payload, the acting-user-bound invoke,
and that both stop the moment the plugin is no longer active.
Tools are declared in the signed manifest, not only reported at load.
plugin-host-entry composes its loaded payload AFTER onLoad runs, so a
signed, sha256-pinned artifact could otherwise emit different tool text on
every restart with no version bump and no re-consent. The host will
advertise the intersection of the two.
It also gives the admin something to read. What mcp:tools will put in
front of every user is a manifest field now, so it can be shown on the
consent screen before the grant is given.
Rejects rather than drops, unlike most capability parsing: a malformed
tool should fail the install visibly, not become a tool that silently
never appears once the plugin is running. Caps and sanitising come from
mcp-tool-schema.ts, so the install path and the advertise path cannot
disagree about them.
Declaring tools without mcp:tools is an error on both sides. Otherwise
the consent screen lists tools that can never run.
No migration: install/discovery.ts already persists capabilities as JSON.
The SDK validator mirrors the checks it can make without the server's
normaliser, so `trek-plugin validate` catches a bad declaration before
publish rather than at install.
The security control for plugin MCP tools, landed on its own so it is
reviewable as one diff. Pure and Nest-free: no container, no DB, no
running plugin needed to test it.
The threat is not a plugin crashing. A tool name, title and description
are a persistent, admin-invisible write into every user's assistant
context, and .meta() puts the JSON Schema there verbatim. TREK's own code
proves the channel is obeyed: the static-token deprecation notice is
delivered in a tool result precisely because that is what an assistant
reliably surfaces.
sanitiseAssistantText is a sibling of stripEmoji, not a change to it.
stripEmoji preserves newlines on purpose (PDF paragraphs) and ~15 render
call sites depend on that. The new one collapses every whitespace run to
a single space, so a description cannot forge "## System" as a heading
that reads to the model like host text, and strips more than the brief
asked for: the bidi overrides and isolates and U+2028/9 are neither C0
controls nor \n, so they defeat both halves of the stated rule.
The schema is advertised rather than converted. z.looseObject({}).meta()
passes the plugin's own schema through byte for byte, keeping pattern,
oneOf and format, which a hand-rolled converter flattens to unknown. The
.check() keeps required enforced, so the handler cannot receive a call
missing arguments its own schema declared mandatory.
That only works with a normaliser in front, since .meta() is a
passthrough. It rejects a root type that is not object (which would
advertise a non-object inputSchema while the runtime still passes an
object), rejects $ref/$defs/definitions at any depth, and caps bytes,
properties and nesting. It strips a top-level id, which vanishes from the
advertisement and lands in z.globalRegistry._idmap, a strong Map unlike
the WeakMap holding the rest.
Sanitising recurses into every nested description, title and string enum
member. That is the part a top-level cap misses, and where a property
named trek_session_token with a plausible description would live.
Annotations are clamped asymmetrically. readOnlyHint gates client
auto-approval, so a plugin may lower its danger level and never raise it;
openWorldHint goes the other way, since egress makes a tool open-world
whether it admits it or not. db:own is deliberately not a side effect:
counting the plugin's private SQLite would make nearly every plugin
non-read-only and kill auto-approval for legitimate read tools.
MCPSCHEMA-001 to MCPSCHEMA-030.
Modelled on the hook:* family rather than on geolocation:read. The host
dispatches INTO the child for every tool call, so it needs what
HOOK_PERMISSION buys: providersOf()'s active-and-holds-the-grant check,
the second gate in invokeHook, and the acting-user binding. That last one
is the point. The user is bound host-side from the supervisor's
invocation map, so a tool called by user A reaches A's data whatever the
model puts in args.
The permission keeps no hook: prefix because the surface it opens is MCP,
not a TREK render slot, and nothing keys off that prefix. It is the first
HOOK_PERMISSION value shaped that way; `satisfies` still holds.
One commit, because it cannot be split. requireTotalCoverage makes a
HOOK_PERMISSION key with no @PluginHook consumer a boot failure, not a
test failure, and PluginRpcRegistryService.onModuleInit runs in every
buildApp() — so a half-landed version breaks every integration test in the
repo. The generator is chained to it too: it refuses to run without both
the en admin label and the CLI hint.
15s is the longest budget of any read hook. The others back a render
someone is waiting on and fall back when they time out; this one backs a
chat turn, and the plugin is likely calling a third-party API of its own.
Still under the supervisor's unused 30s default, which is a hung client
rather than a budget.
One callTool fn for every tool, since tool names are runtime data and
hookContracts() is keyed by (hook, fn).
PluginDefinition exists in two copies with no generator between them, so
both get McpToolProvider. grantGaps and the dev mock host need no change:
both read HOOK_PERMISSION, which is generated.
PLUGHOOK-010, plus the CONTRACTS row.
One coarse scope rather than per-plugin granularity. Which plugins are
installed is runtime data, so a per-plugin scope cannot be statically
enumerated: it would break the ${group}:${mode} derivation and the
AssertExact guards in nest-mcp-policy.ts, and its consent copy could not
be written ahead of time for a third-party plugin name. Tokens issued
before a plugin existed could never carry its scope either. Per-plugin
granularity already exists one layer down as the admin's mcp:tools grant.
'use' is a fourth mode, which widens ScopeMode. Nothing else moves:
McpAccessGroup and McpAccessMode derive from Scope, so both AssertExact
guards self-heal, and the policy's fallback branch already covers a third
mode the way it covers files:content.
The presets needed an explicit exclusion. Five of the six are written as
"everything except deletes", so plugins:use would have joined Claude.ai,
Cursor, Windsurf and Zed silently and handed out third-party tool
execution without anyone choosing it. That filter was duplicated across
the desktop and mobile files with nothing pinning it, so both now use one
shared constant, and FE-OAUTH-SCOPES-012 holds the line for the next
scope like this one.
The consent copy says what the scope does not do: the RPC host authorises
from grants frozen at construction, so this scope plus weather:read still
reaches whatever the plugin itself was granted.
Two i18n gates, not one. The dot-only group key is tracked by
i18n-parity.mjs, whose regex has no colon in its character class; the
colon-bearing label/description pair is invisible to it but is compared by
client/tests/unit/i18n/parity.test.ts, which diffs merged bundles. All 23
locales get all three keys, including the three in neither gate.
The consent glyph gets a ':use' branch. Falling through to the eye would
tell the user a tool-running scope only looks at things.
Sibling of plugin-event-sink.ts: the plugins layer registers a source, and
src/mcp/tools.ts reads it as a defaulted argument. mcp-transport/ is not
touched at all, so it never learns that plugins exist and the tool-call
audit row for a plugin tool comes out of code that did not change.
Pull rather than push, unlike the event sink, because the answer depends
on the session's ctx: there is nothing a push could carry ahead of time,
and reading it per attach means activating or uninstalling a plugin takes
effect on the next session with nothing to invalidate. Same argument
channel-registry.ts makes for setPluginChannelSource.
An unset source means "no plugin tools" rather than an error, the right
answer for any app built without the plugins runtime. MCPSINK-003 pins
that case, which none of the sibling sinks covers today.
Not try/caught here: attach() already contains a throwing source and logs
which one failed, and catching again would swallow that.
The harness takes dynamicTools explicitly so suites exercise the path
without installing a process-level source.
MCPSINK-001 to MCPSINK-005.
A host can now contribute tools for one session through
McpAttachOptions.dynamicTools, consulted once per attach() with that
session's ctx.
It belongs here rather than in the host calling server.registerTool()
itself, because attach() is where the onInvoke seam fires. Anything
registered outside it is invisible to the tool-call audit trail, and
handlers that run as the calling user are the last thing that should sit
in an audit blind spot.
The source is synchronous: attach() runs inside the initialize request,
so an awaited source would put a third party's round trip in front of
every session creation.
Two properties carry the weight, both security ones:
- Names are reserved against every registered entry, not against what
this session was granted. Reserving against the session would mean a
token scoped to trips:read has no create_trip attached, so a
contributed tool could take that name for exactly the callers holding
no scope for it.
- Nothing a contributor does can make attach() throw. Hosts call it
outside their request try block, so an escape is a 500 on every
initialize and one bad contributor takes MCP down instance-wide. The
source is contained once, each entry again. Both live throw sites are
covered: the SDK duplicate-name throw, and the policy-less declarative
access throw that registered entries only avoid because validate()
pre-checks them at boot.
ToolOptions.inputSchema widens to ZodRawShape | McpZodSchema. The SDK has
accepted either since it normalises both to one object schema, and a
whole schema is what lets a host advertise one it did not author (via
.meta()) without converting it to Zod first. The arms are disjoint, so
nothing else moves; one test cast in decorators.test.ts records which arm
that decorator declared.
DYNTOOL-001 to DYNTOOL-014.
The body was a hand-inlined duplicate of sessionManager's own closeSession
helper, so it moves next to revokeUserSessions and reuses it. The barrel
re-exports it, leaving admin.service.ts and admin.controller.ts untouched.
The reason to move it is that callers outside src/mcp/ are coming. index.ts
evaluates readEnv().mcp at module scope and installs the sweep interval, so
importing the barrel from a domain module drags both into every test file
that mocks app-config with a partial export set, which is the hazard kill-switch.ts
documents. sessionManager.ts has no module-scope side effects.
The log line loses "due to addon state change"; addons stop being the only
thing that invalidates a session surface.
SESS-017 to SESS-020 cover it.
The importer's parsed item spells the provider id `googleFtid`; the match rule
reads `google_ftid`. `isPlaceDuplicate` was handed a mapped object, but
`findDuplicatePlace` got the raw one, so its id strategy was always empty. The
name then decided the match, and a second place sharing that name could take the
ftid belonging to the renamed one. One candidate now serves all three calls.
Sharing the strategy list shares the ORDER, not the comparison, and two
differences survive it. SQLite `lower()` is ASCII-only where JavaScript's is not,
so a row stored as CAFÉ CENTRAL does not match the candidate Café Central in SQL
while it does in memory. The coordinate fallback used to paper over that for a
named candidate, sometimes with the wrong row. And `buildDedupSet` collects
coordinates only for unnamed rows while the coordinate query considers every row,
which is the answer `findMatchingPlaceId` wants and the opposite of what the
in-memory half says. Both are now written down where someone will read them, and
pinned by tests, rather than implied away.
PLACE-SVC-028e covers the backfill landing on the right row. PLACES-SVC-014 to
-017 cover the id beating a misleading name, both remaining divergences, and the
tolerance as a number rather than as a reference to itself.
Two consumers decide whether a candidate place is one the trip already has,
and they spelled the rule out separately: the bulk importers ask an in-memory
set (isPlaceDuplicate), while findDuplicatePlace asks the database because its
caller needs the matched row's id. The two had drifted — the SQL copy falls
back to a coordinate match for a NAMED candidate, which the in-memory copy
deliberately never does, since that would merge the restaurant and the bar at
the same address.
The order now lives once in @trek/shared as placeMatchStrategies(), and each
consumer is a mechanical interpreter of it: provider id (the only part that
survives a rename, #1550), then name, then coordinates and only when there is
no name.
Adds PlacesService.findMatchingPlaceId as the public door for an importer that
has to link to the existing place rather than merely skip the candidate.
findDuplicatePlace stays private behind it — it also returns google_ftid for
the bulk importer's backfill, which is that caller's business and not part of
"which place is this?".
No behaviour change for either existing caller. The coordinate branch is
currently unreachable from the one call site, which sits inside an
isPlaceDuplicate() guard that has already matched on name and returned; the
drift only bites once something calls it standalone.
COORD_DEDUP_TOLERANCE and externalIdsOf move to shared and are re-exported from
places.helpers, so collections.service and places.service keep their imports.
Two reports, one line: the automatic bounce to the provider in useLogin,
which fires whenever password login is off. Both reporters are on an
OIDC-only install, which is what arms it.
#2126, the redirect loop. The re-entry guard sat inside the oidc_code
branch, but the exchange strips oidc_code from the URL before it
navigates away. Any re-run of the effect during that window - and t is a
dependency, so a language change is enough, which every non-English
install does on the login page - read an empty search, missed the guard,
fell through to the config probe and bounced to the provider. The
provider answered silently, landed back on /login?oidc_code=, and round
it went. Each pass mints a valid session, which is why closing the tab
and reopening lands the reporter logged in. The guard is now the first
statement of the effect, so no re-run can reach the redirect.
#2123, logout that logs you back in. The brake for the bounce was
noRedirect in router location state, and it does not survive the trip:
logout flips isAuthenticated on a sync lane while the router defers its
location update, so ProtectedRoute renders once on the old route and
emits <Navigate to=/login replace> with no state, overwriting the entry
the logout handler had just pushed. Location state cannot cross a full
document load either, and api/client.ts performs one on any 401. The
fact now lives in sessionStorage, set in authStore.logout() so all seven
sign-out call sites are covered, and cleared on any successful auth and
on the manual SSO links. Per tab, so a fresh tab still single-signs-on -
that being the point of single sign-on.
Both tests were confirmed red without their fix and green with it. The
first needed rewriting to do that: the obvious version held the page
mounted past the window the race lives in and passed either way.
Not included: RP-initiated logout at the provider. TREK clears its own
session correctly, and ending the provider's is a separate decision -
it needs an end_session_endpoint the provider may not advertise, a
post_logout_redirect_uri registered on their side, and an answer to
whether signing out of TREK should sign you out everywhere.
FE-ADMIN-STOR-036 went red on a loaded runner while every assertion in it
passed on the run before. The two storage tests that capture the migration
POST body read it immediately after awaiting the "Storage configuration
saved" toast - but that toast comes from the PUT, and AdminStoragePanel fires
the migration POST from a later effect reacting to the refreshed admin.state.
The two are not the same tick, so reading the body straight after the toast
only held while that effect happened to flush first.
Both now await the body through waitFor. A genuine regression still fails,
with the same assertion error, once the wait times out.
jsdom has no WebGL, so the real vector basemap layer can never work in the
client tests. More immediately: several map tests hand react-leaflet a partial
useMap() stub with no addLayer on it, which was harmless while the basemap was
a stubbed <TileLayer> and is not now that it is a real layer - addTo() reaches
for map.addLayer and rejects into the void. 79 unhandled rejections, every
assertion still green, and vitest fails the run on those.
Stubbed in the setup file for the same reason ResizeObserver is: the capability
does not exist in this environment. The tests that assert on the layer register
their own mock, which takes precedence.
CARTO started stamping "API KEY REQUIRED" across every keyless basemap tile
on 26.08.2026. #2060 added a key field as the immediate answer, but it leaves
a self-hosted instance needing an email exchange with CARTO before its maps
look right, and every default in the client still pointed at CARTO. Default
to OpenFreeMap Positron instead: no key, no account, no request limit,
commercial use allowed, attribution required.
OpenFreeMap serves vector tiles only, so the Leaflet basemap is a MapLibre
style document rather than an XYZ template. maplibre-gl-leaflet hangs a GL
canvas into Leaflet's own tile pane, which leaves everything above it alone -
markers, GeoJSON, clusters, plugin layers, the custom panes, and the click
handling, since Leaflet's CSS already puts pointer-events:none on that layer.
maplibre-gl still arrives only through a dynamic import, so no chunk that
draws a map grows by a megabyte.
Switched: the trip planner, the MapView fallback every preview goes through,
the journey map, collections, the atlas, the shared trip page and the offline
tile prefetch. OpenFreeMap Positron and Bright join the tile presets in
settings, the admin defaults and both mobile screens.
A CARTO template that is already saved keeps working when a key backs it -
that stays a supported choice. Without a key it is not drawn at all:
resolveTileUrl falls back to the default rather than to a watermark, so an
instance that upgrades with one in its settings sees the new basemap
immediately and needs no migration. The stored value is left as it is, so the
Map settings tab still shows what was chosen and the warning underneath it
explains why - one sentence appended, in all 23 locales.
OpenFreeMap asks for a credit of its own; OpenStreetMap alone under its tiles
is wrong and a licence problem. attributionForTile() derives it from the URL
rather than from a flag at each call site, and the shared trip map, which
carried no attribution at all, now gets one.
Offline maps keep working and get cheaper. A vector style is four things
instead of one, so sync/glPrefetcher.ts walks the style document, its sprite,
the Latin glyph ranges of each font stack it uses, the TileJSON - the planet
path is versioned, so guessing it answers 200 with an empty body - and then
the .pbf tiles. Over the Berlin bounding box TREK computes, raster z10-16 is
2237 requests and 34.6 MB and gives up above z16; vector z10-14 is 177
requests and 32.1 MB and stays sharp above that, because MapLibre scales the
z14 tile rather than asking for another. The Workbox rule for
tiles.openfreemap.org moves to CacheFirst and shares the prefetcher's cache,
so browsing tops a downloaded region up instead of filling a second copy
beside it - a versioned tile path can never be stale. Clearing offline maps
clears it too.
The atlas loses its second raster layer. It existed to warm the HTTP cache of
neighbouring zoom levels, which a vector basemap does not need; a second one
would only cost a second WebGL context. OpenFreeMap has no label-free style,
so its labels are switched off on the live map instead - they are all symbol
layers, while borders and coastlines are line and fill - re-armed on
style.load so a theme switch keeps them off.
Theme changes call setStyle in place rather than rebuilding the layer: a swap
per toggle would leak a WebGL context, and browsers cap those in the low
teens.
The place pins visibly swam back and forth while the map was dragged and only
settled onto their real position when it stopped. The clustered bubbles never did,
and that is the clue: those are drawn inside the canvas, the pins are HTML.
A library Marker repositions itself on the map's 'move' event, which fires once per
pointer sample, while the canvas is redrawn once per animation frame. When the style
is heavy or tiles are still streaming, the pins therefore run ahead of the map they
are supposed to be pinned to.
They now live in their own layer inside the canvas container and are positioned from
the map's own 'render' event, so pin and map move on the same clock. Mapbox keeps the
library marker: its terrain path needs the altitude handling that comes with it, and
the drift was never reported there. Clicks, hover and dragging a pin onto a day are
untouched — they hang off the element, which is the same element as before.
Auto layout wrote journey data straight into the document without holding
it to the caps the contract sets, and the save path sends the document
unvalidated - so the route refused the whole body with a 400 and the
editor reported "couldn't save" for the rest of the session, because
autosave kept offering the same rejected book.
The everyday way in was the itinerary list: thirteen stops a page over
two pages is eighty elements on a spread capped at sixty, so a run of
twenty bare stops was unsavable and a longer one put two spreads over.
Sixty was below what this app's own layouts draw, which makes it a bug
with a constant in front of it rather than a limit, so the cap goes to
ninety and the page keeps the design somebody chose. It stays a
serialisation limit: the largest document the contract now admits is
under four megabytes against the book route's eight.
A long journal entry, a long place name, a journey past eighty countries
and a long title each did the same thing on their own, and a bound story
did it on open without anyone touching Auto layout. The caps become named
exports on the contract and every producer builds against them, so the
two cannot drift apart again - which is also why the stations chunk is
derived from the cap and the salvage fixture counts off it.
Fixes#2085
The dashboard rendered its finished empty state while it was still fetching: no
hero, no cards, and a stats row reading zero everywhere. On a slow connection
that reads as "my trips are gone" rather than "not here yet", which is what the
report describes.
The stats row was the worse half. It comes from a separate call that carries no
loading flag at all, so every tile fell back to its zero until the response
landed. A dash placeholder says "not yet" where a 0 says "none".
The skeletons already existed. SpotlightSkeleton and TripCardSkeleton were
written for this page, described as matching its layout, and never wired up:
JourneyDetailPage was the only consumer of that module.
Fixes#2115
Guests are trip-scoped accounts. Trip Members excludes them from its picker and
refuses them on the write path, and Collections does the same, but Vacay's two
directories did not: a guest stayed selectable in a plan's Invite User dialog
and in the shared-calendar picker, including after being removed from the trip
it was created for.
The queries are server-side, so the filter belongs there rather than in the
client, which never received the flag to filter on. Both write paths get it too:
the picker only hands out a list, the id comes back from the client, and the
Vacay MCP tools reach the same two service methods.
Fixes#2112
The Atlas hero counted more countries than the continent bars below it drew,
and more than the dashboard widget. The extra ones were invented: a place
without coordinates fell back to parsing its address, and any two-letter
uppercase segment at the end became a country code. "New York, NY" produced NY,
"Toronto, ON" produced ON.
Those phantoms only ever showed up in the one number. The continent strip files
an unknown code under 'Other' and never draws it, the map skips codes it cannot
resolve, and the dashboard builds its set from stored regions instead, which is
why it stayed right.
Filtering against a list of valid codes would not have been enough, and would
have hidden the worse half of this: 26 US state abbreviations are real ISO
country codes. CA is California far more often than Canada, DE is Delaware, IN
is Indiana. The existing comment above resolveCountryCode already describes that
collision and solves it by trusting coordinates first and sanity-checking the
address against the country's own bounding box. The gap it names in its last
line is this bug: with no coordinates there is nothing to check against.
So the bare-code branch is now only offered to callers that have coordinates. A
spelled-out country name still resolves either way. A lone "NY" or "CA" resolves
to nothing, which is the honest answer.
Fixes#2111
Pointing the instance at a different OIDC provider left two things from the old
one on every account. The avatar kept a URL on a host this instance no longer
talks to, so it rendered as a broken image and stayed that way: the sync only
ran when the new provider sent a picture claim, and having no picture for a user
is a perfectly normal answer.
The account also stayed pinned to the old provider. Login still worked, because
the verified-email lookup finds the row, but the relink sat behind a check for a
missing sub, so oidc_sub and oidc_issuer never moved. The SSO note in settings
went on naming a provider that was gone.
An uploaded avatar is a bare filename rather than a URL and is left alone on
both paths, exactly as before.
Fixes#2110
A permit valid for one day has no clock on either end. isEndBeforeStart filled
both missing times with midnight and then compared strictly, so start and end
came out equal and the form refused it: inline error, dead save button, toast
on submit.
It also made existing rows uneditable. A booking whose end is a bare date on the
start day is exactly what the booking import writes, and what the mobile sheet
has been allowed to save since 72a82b3c. Opening one of those on the desktop
left the save button dead with no way to clear it.
With a time missing on either side the booking is all-day, so only the dates are
compared. An end date before the start is still refused, with or without times,
and two explicit clocks still have to be in order.
The hotel guard comes along for the same reason mobile carries it: the date
panel is hidden for hotels, so switching type after typing dates left the button
disabled with its explanation inside the hidden block.
Fixes#2107
Every row of an expanded search result is anchored at its leg's start: the time,
the stop name, the platform. A leg's arrival appears as the next row's heading,
and the last one gets the closing row. The walking legs broke that and printed
their DESTINATION, which is the stop the next row already names.
MOTIS brackets every journey in walking legs, so the last leg of a real
connection is the walk from the arrival station to wherever you were headed.
Its origin is the stop you get off at, and because that row printed the
destination instead, the station had no row anywhere in the card and vanished.
The same branch blanked the time cell, so the train's arrival time went with it.
At the other end the access walk swallowed the origin the user had picked.
Rows now read from.name for every mode, and the walk says it is a walk in its
badge row instead. The platform tags along, which gives the exit row the arrival
platform it was already carrying. The time falls back to scheduledTime the way
the save path does, since a schedule-only feed would otherwise leave the row
that matters without a clock.
Fixes#2106
The handler had eight cases, its two doubles had none, which Sonar counted as
new code without coverage. The mock is the more useful of the two to pin
anyway: an importer developing against it should hit the same refusals it will
hit in production, so the SVG, the missing extension, the empty payload and the
size cap are all asserted against the double rather than only against the host.
A journal importer can create the journey, create its entries and write their
stories, and then it stops: there was no way to give an entry a picture. The
seven journal.* methods cover text and nothing else, so an importer reading an
export archive could only upload its photos as trip files, which live on a trip
and never reach the journey's gallery.
journal.addEntryPhoto takes the bytes. Deliberately not a link to an existing
gallery photo and not a provider asset: an importer holding an archive has
neither, it has a file it just unpacked.
The access decision stays where it already was. addPhoto answers null unless the
acting user may edit the journey the entry belongs to, and null becomes
RESOURCE_FORBIDDEN here, so a plugin reaches exactly the entries its acting user
does. Everything the REST upload refuses, this refuses too: images only and no
SVG, the operator's allowed-file-types setting, 10MB decoded, demo mode. The
stored filename is the host's, never the plugin's, because the gallery dedupes
on file_path. The object is written before any row points at it and removed
again when the link is refused.
JournalRpc moves into its own module for this. It now needs StorageService and
the EXIF backfill out of MemoriesModule, and JourneyDomainModule exists
precisely so places, assignments and the plugin host can reach a few journey
functions without dragging both photo providers into their graphs.
Discussion #1365
useTileUrl returns a new string once carto_api_key reaches the settings
store, which happens after the first render. That string was a dependency
of the effect that *builds* the Leaflet map, so the key arriving tore the
map down and built a new one.
Two things break in that window: Leaflet finishes a queued redraw against
the canvas renderer whose map has gone ("Cannot read properties of
undefined (reading 'save')"), and the country layer never comes back —
its effect only re-runs when its own data changes, so the rebuilt map
shows tiles and nothing else.
The template belongs to the tile layer, not to the map. Hold the layers
in a ref, read the template through a ref inside the build effect, and
setUrl them from an effect of their own. Same fix in JourneyMap, where
cartoKey and mapTileUrl sat in the dep arrays of the effects that draw
every marker and track.
Fixes#2097
Homepage's customapi widget and everything shaped like it want a handful
of numbers from one request; /api/v1/trips hands out the rows, but a
widget maps fields, it cannot aggregate a list.
The figures come from AtlasService.getTravelStats, the same call behind
the dashboard's passport card, rather than from a second count. Visited
countries carries years of rules — countries reached only by a flight,
layovers that do not count, countries hidden by hand — and a second
implementation would disagree with the card next to it within a release.
That is also why the controller sits in atlas/ and not in public-api/:
PublicApiModule stays a leaf, so the route comes to the data instead.
TravelStatsController next door already does this for
/api/auth/travel-stats. ApiTokenGuard is provided locally rather than
imported with PublicApiModule — @UseGuards instantiates a guard in the
declaring controller's module, so an export would still leave its
TokenService unresolvable here.
last_trip means the most recently started trip, not the most recently
created one, and is null while every trip is still ahead.
Closes#1367
The three api-tokens endpoints landed without tests, which took
src/nest/auth/** under its coverage floor. Same four paths the
mcp-tokens block above already pins, plus the rate limit: unlike MCP
tokens these are not refused on a managed instance, so the limiter is
the only thing bounding a scripted caller.
Reading the surface back against real data turned up three things it dropped
without saying so.
Places with no day were invisible, because the payload only reached places
through day_assignments. On a live instance that is half of them, and they
are the ones carrying a coordinate and no claim about when — exactly what an
integration can act on. They come back under `unplanned_places`, minus the
hotel, which is already reported as an accommodation.
Bookings with no day were dropped by the day filter. reservations.day_id is
nullable and deleting a day detaches its bookings rather than removing them,
so this is not a hypothetical.
The bucket list had no endpoint at all. It hangs off the user rather than a
trip, so it gets its own, and it answers whether or not the Atlas addon is
switched on: the addon decides what TREK shows, not what a key can read.
Also fixes ?include=notes returning the bare trip: places, notes and
reservations are reported on days, so asking for one now implies days
instead of quietly answering with nothing.
The module stays a leaf. Reaching for AtlasService pulled AuthModule and the
storage registry in behind it and broke the surface's own boot, so the one
SELECT lives here like every other query in the file.
Minting a key for other software meant finding it under MCP, behind a
heading about AI assistants and gated on an addon it does not need. It now
sits on its own in Settings, above the MCP block, and works on an instance
with MCP switched off.
Deliberately separate state rather than another branch of the tab's shared
hook: the section has to stand on its own, and that hook is already carrying
forty fields through three levels of props.
Translated into all 23 languages.
An MCP token drives every assistant tool; an API key reads trips over HTTP.
Handing the same credential to a third-party integration gives it the whole
MCP surface, so the two are now separate kinds on `mcp_tokens` and the kind
sits in the WHERE clause — a token of the wrong kind is indistinguishable
from one that does not exist.
Also adds `travellers` to the trip payload: display name and an owner flag,
no ids and no email addresses. An integration key reads its owner's
itinerary; it is not a way to enumerate the people around them.
Documents the surface in the wiki, since integrators need something to read
that is not the source.
Two gaps from the first pass, both about a client that polls.
Rate limiting is keyed by user rather than by IP. A self-hosted integration and
its owner's browser routinely leave the same address, and an IP bucket would let
one starve the other; 120 requests a minute is generous for a sync surface and
still bounded, so a runaway poll degrades its own integration instead of the
instance.
updated_at now ships with every trip, and its docblock says plainly what it is
not: a content timestamp. It moves when the trip's own fields change — title,
dates, currency, cover — and stays put when a place, a day or a note is edited,
because the child tables carry no updated_at of their own. That also rules out
the ?since= filter this was meant to enable: it would silently hide every trip
whose itinerary changed, which is worse than no filter at all. Making it honest
needs a schema change, not an API field, so the field is documented for what it
is and the filter is left out.
TREK's REST surface is internal: it answers a session cookie, carries no
version, and changes with the client. That is fine for our own frontend and
useless for anyone building against us — the Dawarich integration people have
been asking for since April needs something it can rely on not moving.
So this adds a small one rather than opening the existing 549 endpoints:
GET /api/v1/trips the trips the caller can reach
GET /api/v1/trips/:id one trip, ?include= picks the sections
Auth reuses the trek_ tokens users already manage under Settings, via Bearer
or X-API-Key. No new credential type, no new table — the guard calls the same
hash lookup the MCP transport does. It deliberately refuses a session JWT and,
for now, OAuth bearers: a credential whose scopes this surface does not yet
interpret is worse than one it rejects outright.
Access is decided per trip against canAccessTrip, the same owner-or-member
predicate as everywhere else, and a trip the caller may not read answers 404
identically to one that does not exist — a 403 would confirm the id is real and
turn the endpoint into a way to count someone else's trips.
The payload is not the rows. Ids, foreign keys and order_index stay inside;
accommodation day ids resolve to ISO dates, places arrive in planned order, and
every dated thing carries its date so a consumer can join on days rather than
on ids it has no way to look up.
fixArrivalDate normalized the arrival and then looked for the time behind a
leading date. The model usually sends the arrival as a bare clock, "13:00" with
no date on it, and slicing ten characters off that leaves nothing to match, so
the function returned early and stopped deriving the arrival date at all.
Try the bare 12-hour form first, then fall back to the normalized string, which
covers both shapes: "13:00", "01:11 pm", and a full date-time.
shared compiles with noUncheckedIndexedAccess, so a capture group is
string | undefined there even when the pattern guarantees it. Pull the groups
through optional access with explicit fallbacks instead of indexing into the
match, which also drops the two throwaway locals.
No behaviour change: same inputs, same outputs, the spec is untouched.
When neither the keyword table nor the model produced a type, the router wrote
"hotel" and cast whatever string it got to FlatType without checking it. A hotel
is a booking type, so a ferry voucher imported from the Transport tab opened the
booking form, which offers six chips and not one transport among them, and
"other" was the only honest pick left. An unrecognised string got through the
cast too and then vanished silently in nuExtractToKiReservations.
The type is validated against FLAT_TYPES now. When it cannot be read the item is
still extracted with the lodging shape so it survives mapping, but it says so,
and the tie-breaker #2080 added finally has something to fire on: isUnplaceableItem
only ever matched a type outside both form's vocabularies, and the server can
only emit its own eight, so on a real document it never matched at all.
The tab travels with the job instead of sitting in component state. The parse
deliberately outlives navigation and reload and the review is triggered by the
global widget on any page, so by the time it ran the state had remounted back to
its default and the context was gone.
splitIso reached the time by taking characters 11..16, so "2026-06-11T01:11 PM"
became 01:11. The hour survived and the half of the day did not, which is why
AM and 24-hour times in the same document imported correctly. The arrival then
sorted before the departure and the router rolled it onto the next day, so a
three-hour hop was rendered as an overnight.
The client has done this properly since #1725; this is the server half, which
was never pulled across. normalizeLocalDateTime rewrites a value only when it
actually finds a meridiem, so a date-only value, a value carrying an offset and
anything the parser does not know come back byte for byte.
The mapper has only two exits for a date, toIsoString and splitIso, so this
covers all eight types and every emitted field at once. It also stops
sameConnection comparing a NaN timestamp, which silently merged unrelated legs
into one booking.
The ICS feed is public with a token that never expires and prints the
confirmation number into the DESCRIPTION, and the shared-trip payload does a
SELECT * over reservations with share_bookings on by default. Neither filters
anything, so whatever lands on a trip is published the moment it is written.
That is fine while a person types every booking in, and it stops being fine as
soon as something writes on their behalf.
reservations.ingest_state is that gate. It defaults to live, SQLite backfills
every existing row on the ALTER, and no current writer names the column, so
every booking visible today stays visible. Nothing writes staged yet.
The predicate takes its table alias because share.service.ts queried
reservations without one. A stay is withheld only when bookings point at it and
all of them are staged: a hand-added hotel has no booking at all and has to keep
showing up.
Files, settings, calendar feeds, invite links, the AirTrail import, provider
photos, help and the addon list. Also says why accepting an invite and
reading photo bytes have no tools.
Ten new controllers in the test registry, and one line in the session
instructions: without it nothing tells an assistant to read the display
settings before rendering a temperature, which is what makes that tool worth
having.
A plugin warning is a plugin telling the user something is wrong with their
trip, and get_trip_summary omitted them, so an assistant reviewed a trip
from a strictly smaller picture than the user was looking at.
The whole memories domain had no tools, so the Immich and Synology
libraries were invisible even though the attach body is plain JSON and the
photos never pass through the call. Search, album listing, album contents,
and the attach.
The routes that stream photo bytes stay out: those are img targets.
TREK's own bulk travel-history importer, and the closest thing in the
product to what this whole branch is about. Both halves are server-side
JSON, so neither was ever browser-bound; they simply had no tools. The list
tool's output carries what the import tool needs, so a model can run the
pair without a detour through the UI.
export_trip_ics is a snapshot of one trip, so subscribing was out of reach
and the all-trips feed had no counterpart at all. Both feed tokens can now
be read, minted, rotated and revoked.
The invite link is the one that grants membership, which create_share_link
is not, and the descriptions say so in both directions so a model does not
reach for the wrong one. The two consumer routes stay out: the ics URLs are
what gets subscribed to, and accepting an invite is a human act.
test-db's reset list was missing trip_invite_tokens, so a leftover invite
re-attached itself to the next test's trip once ids restarted.
131 of 201 tools carry an addon gate, and the registry drops an entry when
either the addon gate or the scope gate fails: one filter, one outcome. The
session instructions tell the model to check what is available before
assuming a tool exists while giving it nothing to check with. list_addons is
that something, and it is the difference between a confused apology and
'ask your admin to enable Budget'.
The help routes were already public, so the two reader tools add no scope
surface at all. The asset route stays out: it serves bytes to an img tag.
An assistant rendering a temperature, a distance or a clock time was
guessing: the session instructions named only the trip currency, and nothing
could read what the user had picked.
The write works off an allow-list rather than off the settings row, because
that row also holds the webhook URL, the ntfy token and the Mapbox, Carto
and LLM keys, and getUserSettings hands some of those back in cleartext.
Everything outside the list is refused, including a key nobody has heard of
yet. The guard test asserts against the service's own encrypted and masked
sets rather than a copy of them, so a sixth key added there fails it.
Often the travel history is the uploaded PDFs, and an assistant could not
learn that a trip had any: there was no files.mcp.ts, no files scope, and no
files section in the trip summary.
Six tools: list, read contents, set description and primary link, add a
link, remove one, list them. Reading contents rides files:content rather
than files:read, because listing what a trip carries and reading what is
inside a booking PDF are different privileges.
The content read is not a second copy of the plugin path. readContent moved
out of files.rpc.ts into FilesService and both callers use it, so the 10 MB
cap, the storage-layer read and the off-thread decode stay in one place. The
RPC keeps its own error wording, which its tests still pin.
The admin gate on the three new category tools reached for AuthService, which
drags the storage registry in behind it, and the categories e2e builds a
container around one module against a two-table database. It stopped booting.
budget/ already solved this: isDemoUserId reads RuntimeEnvService and the
users table directly, and its module comments say why AuthModule is absent.
Categories now does the same, which also takes eight hand-built collaborators
out of its unit test.
The e2e gained RealtimeModule, which is global in the app graph but not in a
partial container, the way days.e2e.test.ts already imports it.
Two domains an assistant cannot reach at all today need somewhere to hang,
so this is the credential half on its own, ahead of the tools.
files:content is a third mode rather than a second group. Listing the
documents on a trip and reading the bytes inside a booking PDF are different
privileges, and the plugin host already draws that line between
db:read:files and db:read:files:content. It is not implied by files:write,
the same way journey:share is not implied by journey:write; the policy's
fallback branch covers both.
settings:write is scoped to display preferences. Settings also holds the
webhook URL, the ntfy token and the Mapbox, Carto and LLM keys, so the tools
that ride this scope work off an allow-list rather than off the settings
row.
34 scopes across 16 groups, translated into all 23 languages.
A dateless trip made by a tool silently got seven days, and reminder_days
was stuck on the default of three with no way to turn it off. Cover images
had no search at all.
add_trip_member and remove_trip_member hardcoded an owner check while REST
consults the configurable permission matrix and the plugin RPC does too:
three surfaces, two behaviours. They now use the same permission call
update_trip already made, so an instance that lowered member_manage behaves
consistently. The guest tools stay owner-only, because their REST routes are
owner-only by design and document it.
Asking whether a place is already in one of your lists took a walk over
every collection and lost the coordinate tolerance the server matches on.
Marking a trip's places visited across every list holding them was one call
per place.
The dashboard's next-booking view is a three way union that folds in
synthetic check-in and check-out rows, so rebuilding it from per-trip
summaries silently drops the accommodations.
A day note could not be coloured or positioned, so the service default of
9999 pinned every note an assistant wrote to the bottom of the timeline. A
collab note could not carry the website the card renders.
Categories could be assigned by bulk_update_places but never created: the
whole file was one read tool and a resource. The three new tools are
admin-only, matching the REST routes exactly.
image_url was in no places tool although the service persists it and the
picker in the UI sets a URL, which is not a file transport question.
search_place passed two of the service's four arguments, so an unbiased
query resolved to the wrong continent.
get_place_details gained expand for reviews and the editorial summary. It
defaults to off because that field mask bills as a Google Enterprise SKU,
and it now honours the admin kill switch the REST route checks: an instance
owner who turns Place Details off to stop the billing was still billable
through this one surface.
location_name was settable when creating an entry and never afterwards,
which is the sharpest create-versus-update asymmetry in the surface.
Coordinates, weather, tags and the pros and cons verdict were persisted by
the insert but had no way in, so an entry written by an assistant never
placed on the journey map.
mark_region_visited wanted a code like US-CA that nothing in the tool
surface could produce. The visited-region list also read the raw manual
marks while the UI reads the list derived from trip places, so the two
disagreed about where somebody had been. Travel stats gained the cities and
the distance the dashboard shows.
The toggle took a date and nothing else, then hardcoded a full vacation day
for the calling user, so a half day, a comp day or a day logged for another
plan member had no expression.
School holidays needed the plan flags that switch them on, and the region
list that makes them usable. That list needs care: a group region is stored
as COUNTRY|group:CODE and the reader treats a bare code as a subdivision, so
the tool now returns the string add_holiday_calendar actually wants. Belgium
and the Netherlands are the two countries this affects.
Bag assignment was reachable only while bulk-importing a list, and by bag
name, never for an item that already exists. Quantity, weight and the
private flag were not reachable at all, and the whole three-tier sharing
model was invisible: an item made by a tool was always Common.
Historical costs are rarely split evenly, and every budget tool took only a
list of member ids, which the service stores as a null amount, meaning an
equal share. members[] carries what each person owes instead.
A custom split has to reconcile before it is persisted, so the tool refuses
one whose shares do not add up in whole cents. Two ways round that check
needed closing with it: the write path drops a payer who is not on the trip
and then derives the total from what survived, and an empty payer list means
a total of zero rather than the previous one. Both would have certified a
split the stored row does not satisfy.
Also on update, which had none of them: currency (so the FX freeze is no
longer a no-op), expense_date, and a settlement currency, plus place_id on
create so a museum ticket can point at the museum.
An imported itinerary landed in whatever order the assistant appended it,
with no way to fix it afterwards: create_day only ever appended, and
days.service.insert was reachable from exactly one place in the server, the
REST controller. Because both halves were missing, inserting a day into the
middle of a trip could not be expressed at all.
url is persisted on create and update and rendered as a link by the
planner, and reservation_end_time is what makes a dinner run 19:00-21:30
rather than just starting. Neither was in a tool schema, so a booking
written by an assistant lost both - which for an imported itinerary is
every confirmation link it came with.
url validates through reservationUrlSchema from @trek/shared, the same
refinement the REST route applies, rather than a second hand-written one.
The transport form's picker has nine entries, not ten: a transit booking
carries a provider itinerary in metadata.transit and create_transit_journey
is what writes one, so letting a caller hand-make a bare transit row
produces a shape the transit UI does not expect.
Splits the one list in two - what may be asked for, and what counts as a
transport for the update gate - so a stored transit booking stays editable
through update_transport, which is where it was already reachable.
Also reports the ids set_reservation_travelers dropped instead of
returning a plain success with a shorter list.
Reads have hydrated a booking's travellers since #1517; there was no way
to write them, so an imported booking could not record who was on it -
and creating a guest is only half an answer if the guest cannot then be
attached to the flight they were on.
Mirrors PUT /api/trips/:tripId/reservations/:id/travelers. The roster
filter stays in the service, so an off-trip id is dropped rather than
attached and the tool cannot be used to enumerate users.
add_trip_member resolves an existing account by username or email, so a
travelling companion who never signed up could be put on a trip in the
planner and not through an assistant. A guest is assignable to budget
splits, packing, to-dos and day participants like any member, so leaving
them out of an import loses who was actually there.
Mirrors the owner-only /api/trips/:id/guests routes, including the
trip-scoped guard that keeps one trip's owner away from another trip's
guest and off a real member's user row.
create_transport and update_transport took flight, train, car and cruise.
The picker in TransportModal has ten modes and ReservationsPanel renders
all of them, so a bus, a ferry or a taxi could be planned in the UI and
not through an assistant - which is where importing a travel history from
another service runs aground.
Both lists now come from one constant in the picker's order rather than
four literals repeated across two schemas, which is how they drifted.
legs[] stays on flight and train: the form writes metadata.legs for those
two and for nothing else.
Every push to dev has been red since the 4.0.0 bump, always on the same test:
FE-MOB-MSTOR-015, where the second queued migration POSTs while the first is
still running. It passes on the PR runner and fails on the slower push one.
Instrumenting the queue effect logs the whole story:
DEQUEUE files stateMigrations []
APPLY refresh ["running"]
START RESOLVED files error null
DEQUEUE journey stateMigrations []
The refresh had already landed the running migration before the single-flight
lock was released, exactly as intended. What fires afterwards is the effect of
a commit scheduled BEFORE the POST, so its captured admin.state is the pre-POST
world, and it runs late enough that the lock is gone. The lock is a ref and is
read live; the "is one already running" test was read off that dead closure.
Under load the two drift apart and a second migration starts on top of a
running one.
So the hook answers the question instead. It keeps its world in a ref as well,
written where the response is applied rather than when the render lands, and
storageBusy() reads that. The queue effect still WAKES on admin.state, it just
no longer DECIDES on it, and the server's one-job-at-a-time rule now lives with
the hook that talks to the server rather than twice in two shells. Same shape as
the dirtyRef/savingRef the hook already keeps for the same reason.
An earlier attempt made the panels refresh again after startMigration, which was
both redundant (startMigration has awaited its own refresh since #1949) and no
fix at all: the stale read is on the next effect run, not on that promise.
The manifest list was assembled from a glob over the downloaded digest files,
which are named after the digest itself. The order was therefore alphabetical
by hash, and it changed from one release to the next: 3.4.1 lists amd64 first,
4.0.0 lists arm64 first.
That is not supposed to matter — a client is meant to pick the entry matching
its own platform — but several people updating from 3.4.1 to 4.0.0 on amd64
reported that `:latest` still resolved to 3.4.1 and only a pinned `:4` or
`:4.0.0` pulled the new image. Unraid came up twice. Content, format and digest
of the three tags are identical, so the flipped order is the only thing that
changed between the release that worked and the one that did not.
The digests are now collected per platform in a fixed order, amd64 first. That
needs the artifacts unmerged, since the per-platform folder name is the only
thing that says which digest belongs to which architecture.
Second thing the loop fixes: the glob happily built a single-platform manifest
if one of the two builds had failed to upload. It now fails loudly instead of
publishing a `:latest` that half the users cannot pull.
fix(import): stop reading tickets as lodging, and stop scroll escaping a dropdown
Two v4.0.0 reports.
The file importer classified nearly every transport document as a hotel. Its
keyword table tested lodging second, on a group that included "check-in" and
"check-out" — words printed on almost every ferry, train and bus ticket there
is — so a ferry ticket came back a hotel, and the review then handed the user
the booking form, whose six chips contain no transport type at all. That is
where "everything ends up as other" comes from: it was the only honest pick
left. Transport is tested first now, the lodging pattern asks for a word that
names lodging, and the rental pattern asks for a rental rather than for a
pick-up time that ferries and shuttles also print.
Two more in the same function. The flight path forces its type on everything it
returns, and it claimed any document with two letters followed by digits — which
is what a German rail number looks like, so an IC ticket was imported as a
flight without the keyword table ever running. It now stands down when the
document already names another kind of transport. And a type neither form can
express opened the booking form regardless; it now follows the tab the import
was started from, while a type either form knows still wins over the tab, since
one PDF routinely holds a flight and a hotel. An unreadable type reaching the
transport form arrives as transport_other rather than as a flight: a wrong
flight looks right enough to be saved unnoticed.
The other report: a dropdown inside a sheet scrolled the page instead of its own
list. The panel is portaled to document.body and positioned fixed, so its scroll
chain runs to the viewport rather than to the sheet it looks like it belongs to,
and nothing stopped it. Three portaled lists carry overscroll-behavior: contain
now — the select, the navigation menu (which reaches the phone through the place
sheet, not only the desktop inspector) and the phone day picker.
@
fix(planner): route a layover flight per leg, and count linked places as planned
Two v4.0.0 reports, both a surface answering a question with less than it had.
The map built its own waypoint order. It read the BOOKING position and never
expanded metadata.legs, while a multi-leg booking stores its position per leg —
so a Frankfurt to Vancouver flight via Reykjavik was dropped from the list and
the airport before it was joined to the airport after it in one driving run
across the Atlantic. It now takes the order from getMergedItems, the same one
the day plan renders, which brings the span filter and the leg expansion with
it. getMergedItems in turn had never read day_plan_position, the one slot the
map did read, which is how the two could disagree at all.
getTransportRouteEndpoints handed back the booking outermost airports for every
leg. A stop between two legs was therefore routed from the arrival airport back
to the departure one — a drive from Vancouver to Frankfurt with a cafe in
Reykjavik on the way. A leg now spans its own pair, when the endpoints line up
one per stop with the legs; anything ambiguous keeps the role lookup.
Being planned meant one thing: a day_assignments row. A hotel is linked through
its stay and a venue through its booking, and neither is ever dragged onto a
day, so both sat under Unplanned while the day header printed their names. The
three places that asked the question — desktop sidebar, map markers, phone
browser — now ask one helper that also counts a stay and a day-anchored
booking. A booking that names no day still counts as unplanned, because nothing
says when it happens, and the day plan renders nothing for it either.
@
test(calendar): cover the window paths the split added
The domain branch ratchet went from 93 to 92.3: the new hand-over code brought
branches with no case behind them. Three cases and one simplification, no
threshold touched.
Covered: a split rental whose endpoints carry no IANA zone and fall back to
their coordinates; a bare end clock, which resolves against the start date and
therefore does NOT split; and an unsplit rental with a day but no clock, whose
hand-over has nothing to place and correctly emits nothing.
The metadata parse now happens once, inside describeReservation, rather than
three times. The two extra copies were only there to hand `meta` to it, and one
of their arms is unreachable from SQLite, where the column is always text.
@
fix(export): print the reader clock format, and split window bookings in the feed
Three v4.0.0 reports that all come down to a surface printing what is stored
instead of what the reader asked for.
The PDF (#2066) is assembled as an HTML string outside React, so no hook could
reach it and no time format was ever passed in. It printed the column: four
clock sites, all 24h whatever the personal setting or the admin default said,
and a place time typed by a 12h user came out as "3:00 PM" for a 24h reader.
Both export entry points now hand the format over the way they already hand
over the locale, and the four sites go through formatClockTime.
The journey editors (#2067) used a native <input type="time">, which paints
12h or 24h from the browser locale and cannot be told otherwise by any
attribute. Every other time field in TREK has used CustomTimePicker for exactly
that reason; the journey editor, its phone sheet and the phone place sheet were
never migrated, so the same place form obeyed the setting on desktop and
ignored it on a phone. The picker gained a disabled prop so the read-only
journey sheet keeps the affordance the native input gave it.
In the subscribed calendar (#2068), a booking that is a window rather than an
activity read as one block across its whole range: a week of airport parking
sat on all seven days. Both hand-overs now stand on their own — an hour each,
carrying what the block used to say — and the block goes. Only when the two
ends fall on different days: a same-day booking is one sitting, and a rental
with only its pickup geocoded keeps the block, which is the only thing holding
the return time and its zone at all. Rental cars were missing outright, because
the planner writes reservation_time = NULL when its optional time pickers are
left blank while still setting day_id, and nothing in the exporter ever read
the linked days. It reads them now, as the last resort, for every type — so a
hand-typed bus or train shows up as well.
@
test(server): unpin three fixtures from the 3.x version window
The 4.0.0 bump turned three suites red on dev, all for the same reason: they
described a world where the host was still 3.x.
The plugin install fixtures declared trek ">=3.2.0 <4.0.0" and ">=3.0.0
<4.0.0", so every default artifact became incompatible with its own host and
23 tests failed on the version gate instead of on what they were testing. The
upper bound bought nothing — the cases that exercise the gate pass their own
range — so it is gone rather than moved to 5.0.0.
The system-notice suite named thank-you-support twice. It is capped at 4.0.0
and handed over to release-4-0-0 there, so the established-user case saw a
notice it had never been told about, and the per-version recurrence case rode
on a notice that no longer applies at all. The first now reads the always-on
set out of the registry instead of naming it, and the second drives the
mechanism through a notice it pushes itself. Neither ages out at the next
release.
@
fix(dashboard): fix the upcoming widget dropping bookings, and give it the stays back
Three things, all in the same widget.
reservation_time holds two shapes. The booking form writes the datetime-local
YYYY-MM-DDTHH:MM; day-anchored rows, the demo seed among them, hold a bare
HH:MM. The query compared the column against an ISO timestamp as a string,
where ":" sorts above "2": every booking from 20:00 on passed the filter and
everything before it silently vanished, and the order ran off the same column.
On the demo that is why the Broadway show shows and the ferry does not. The
query now separates the date from the time before it filters, so a value with
no date falls back to the day it hangs off instead of being read as one, and
the order is by day and clock. The demo seed writes full timestamps, which is
the shape every real write path already produces.
Hotels came out of this list because a stay covers a range and a week in one
would hold a slot for the whole week. Arriving and leaving do not: they are
moments, and they are the two worth reminding someone of. Check-in and
check-out now appear as their own entries, named after the stay place and
falling back to the linked booking. The stay itself stays out. The two carry
the accommodation id, so the list key had to take the type with it.
An unconfirmed booking looked exactly like a confirmed one. The status was
already in the payload and is now a badge, on both the desktop tool and the
phone widget.
@
The day-swipe move handler mixed classifying the gesture with following the
finger, which put it at cognitive complexity 20. The classifying half is a
pure question about a gesture, so it moves out as classifyLock: null keeps
waiting for the next move, dead and h end the same way they did before.
Also reads the reduce-motion flag through dataset, seeds the closest-year
reduce with years[0] so it cannot throw if the empty guard above it ever
loosens, and compares the in-flight preview entry against undefined instead
of leaning on a promise being truthy.
fix(plugin-sdk): stop the README heading match backtracking
`missingSections()` matched headings with `(.+?)\s*(?:<!--.*)?$`, where the
lazy body and the trailing `\s*` compete over the same run of spaces. That
backtracks quadratically: a heading line padded with 80k spaces takes about
12 seconds, 200k about 75. The gate runs over READMEs anyone can submit to
the registry, so the padding is not the author own input.
The trailing anchor comment is now cut with indexOf, and the body is matched
with its edges pinned to non-space, so nothing competes with the trailing
`\s*`. Same shape the wiki sidebar parser was rewritten to for the same
reason. Behaviour is unchanged on every heading that is not the attack:
trailing comment, unterminated comment, tab separator, padded title.
scripts/check-readme.mjs in the TREK-Plugins registry carries the same
regex and needs the same change.
@
CARTO stamps an "API KEY REQUIRED" watermark onto every basemap tile
fetched without a key since 26.08.2026, and both CartoDB presets plus the
built-in fallback are CARTO. The key is free and needs no account, so add
it as a setting rather than dropping the provider.
carto_api_key sits next to mapbox_access_token: per user in Settings/Map,
instance-wide under Admin/User Defaults, encrypted at rest, injected from
CARTO_API_KEY on managed instances. Unlike Mapbox it never switches the
map provider, since a raster key says nothing about the renderer.
The key is appended as ?key= at render time instead of being stored in the
template, so a saved URL stays portable and survives a rotation. That is
also what reaches the four maps that pinned CARTO and ignored the user's
tile setting: atlas, collections, the public trip share and the public
journey. The two share pages read it from their token-gated payload, the
same way baseCurrency already travels.
Also lifts the release notice above page chrome. Its overlay sat at
z-index 50, under the Vacay mode toolbar at 61, so the dialog was neither
on top nor modal. Adds a six-step layer scale to index.css and moves the
three notice shells onto it.
Closes#2054Closes#2052
The preview endpoint takes a URL from whoever is typing in a trip chat and
fetches it from the server. Trip access was added to it earlier; this is the
rest of what that route needed.
The SSRF guard had a hole under it. `isAlwaysBlocked` knew `::1` and `0.0.0.0`
but not the IPv6 unspecified address: `::` starts with a colon, so the `0.`
check never saw it, and connecting to it lands on loopback. `http://[::]:9200/`
therefore passed the guard and came back with the target's title and og: tags
scraped out of the body — an exfiltrating read of anything listening locally,
from any account. `[::0]`, `[0:0:0:0:0:0:0:0]` and `[::ffff:0:0]` are the same
address in other spellings and behaved the same way. The two `::ffff:`-prefix
regexes had the same shape of gap: they matched the dotted spelling of loopback
and link-local, so `::ffff:7f00:1` walked past them. Both checks now decide on
the expanded hextets, which covers every spelling at once, and fe80::/10 is
read as ten bits rather than as the four characters 'fe80'. This is in
ssrfGuard, so every caller gets it, not just collab.
The scrape could be turned into a denial of service. Each og: regex had two
unbounded `[^>]*` runs separated by a literal, so a document with no `>` in it
made the engine rescan the remainder from every `<meta` it passed — quadratic,
and the byte cap admits half a megabyte. Measured on this build: 240KB of
'<meta ' took 58 seconds, and Node has one thread, so that is the entire
server, health check included, on one request. Bounding the runs at 512
characters makes it linear (540KB now takes 1.4s); no real attribute list comes
close to that.
The five-second timeout only covered the headers — `clearTimeout` ran before
the body was read — so a server that answered fast and then dripped one byte
every few seconds held the handler, the socket and the per-request undici agent
open indefinitely. The byte cap counts bytes and at that rate would never reach
one. `AbortSignal.timeout` covers both halves, the agent is closed in a
`finally`, and the three bail-outs that used to return without touching the
body now discard it, which is what cappedFetch's helper is there for.
The rest is smaller. Nothing metered the fetches, so an account with a trip of
its own could drive arbitrary outbound requests in a loop; there is a budget of
sixty new fetches a minute per user now, charged per outbound fetch rather than
per request. Previews are cached for ten minutes and simultaneous askers for
one URL share a single fetch, so the client's habit of requesting a preview per
rendered message — up to a hundred at once, with no debounce — costs one fetch
per new link rather than one per render. Refusals answer with one constant
reason instead of the guard's own: it distinguishes "could not resolve" from
"private address", which together map out the server's internal DNS for anyone
willing to guess hostnames. A body whose declared type is not markup is dropped
rather than combed for og: tags, and an og:image that is not an http(s) URL is
not passed to the client, which puts it straight into an <img src>.
Deliberately no `@RequirePermission` on the route: the client asks for a
preview while rendering a message or a note, never while composing one, so it
belongs with the GET siblings. Requiring collab_edit would blank every preview
for members of a trip whose owner narrowed that right while leaving them able
to read the chat, and it would not stop the abuse it looks like it addresses —
the permission check passes for a trip's owner at every level, and anyone can
create a trip.
* @
feat(mobile): swipe the day details to step through days
The phone shell hangs the days off a chip rail pinned to the top of the
screen, which on a large phone is a thin target a thumb cannot reach one
handed. The day details fill the rest of the screen and did nothing.
They now take a horizontal swipe: left for the next day, right for the
previous one, with the panel following the finger, rubber banding at the
first and last day and springing back when the swipe falls short. The
chip rail is untouched and still the accessible path; a committed swipe
announces the day it reached through a live region, and the active chip
scrolls itself back into view once the rail overflows.
The gesture never calls preventDefault or stops propagation on a touch
event and sets no touch-action, so the browser keeps sole ownership of
the card scroll and of the header chip strip, and the long press reorder
bridge keeps its own 10px self cancel. In edit mode the swipe also
refuses to claim a press held past the bridge window.
@
* @
fix(mobile): line the map controls up and widen the POI bar
The compass sat on its own under the POI bar, centre top, while the
locate button floated 42px above the dock on the other end of the
screen. Both are one handed controls, so they now share the band just
above the dock: compass bottom left, locate bottom right, off the same
--bottom-nav-h the map already reads, a dock gap lower than before.
That frees the row under the chip rail, so the nearby places bar takes
the full width between the screen margins and its segments end up the
size of everything else the thumb aims at there. Desktop keeps the
floating, content width bar.
@
The UP NEXT card kept pointing at this morning's first stop hours after it had
started, and at day one of a trip that ended last month. A header promising what
comes next, sitting above something that is over. It now stays away on a day that
has passed and once today's timed plan has run out — a day whose stops carry no
times at all still shows its first stop, there is nothing stale about that. The
timeline card already collapses the reserved space when there is no up-next, so
the layout closes cleanly.
Seeding the plan day also lands on the next day still ahead when today falls into
a gap in the dates, instead of dropping back to day one. A trip that is already
over keeps opening on its first day.
And a day that was selected before the shell mounted now counts as seeded, so
clearing the selection is left alone instead of being re-seeded on the next
render — which is what the guard was there for in the first place.
Based on #2055.
The year list comes back sorted, so picking its last element meant Vacay always
landed on the highest year — add 2030 to plan ahead and every visit from then on
opened 2030. It now opens the leave-year period today falls into, falls back to
the closest year the plan has when that period has none, and keeps the year the
viewer is on when a live update reloads the list.
* Revert "chore(deps): bump dompurify from 3.4.11 to 3.4.12 (#1681)" (#1694)
This reverts commit 94736112cf.
* feat: add PhotoGrid and PhotoImg components for journey photo gallery display
* feat: implement JourneyPublicPage to display shared journey data with timeline, map, and gallery views
* feat: create MobileEntryView component for displaying detailed journey entries on mobile devices
* feat: add docker-compose configuration for video testing environment
* feat: add gallery view components with support for local photo/video uploads and external provider integration
* fix: green up the video layout branch
Four things stood between this and a merge.
index.css carried committed merge conflict markers. A format-on-save had run
over the whole file and collapsed `=======` and `>>>>>>>` onto the line that
starts the popup-tip rule, which makes .trek-map-popup .mapboxgl-popup-tip part
of a malformed selector list - so the parser drops the whole rule and the map
popup's tail comes back as the stray notch its comment describes. Nothing in CI
catches that; invalid CSS just disappears. The file is back to the dev version
with only the video block added, 1301 -> 1323 lines instead of 2057, and the
same for the lockfile, which had only picked up "peer": true annotations from a
different npm run.
PhotoImg referenced an onClick it never took as a prop - the click already sits
on the button wrapping it.
PublicPhoto did not declare media_type even though journey-share.service.ts
selects it in both queries, which is what the `as any` casts around it were
working around. Declared once, the way JourneyPhoto does, and the casts are gone.
The single-photo hero had moved to a thumbnail. Right for a video, whose
/original is the video file and leaves an <img> blank - but a photo hero is
deliberately full size, which FE-JRN-PUBWIRE-008 pins. It now picks per media
type.
---------
Co-authored-by: jubnl <66769052+jubnl@users.noreply.github.com>
Co-authored-by: Maurice <mauriceboe@icloud.com>
uploadQueue called crypto.randomUUID() bare, which is undefined on
non-secure origins (plain-http localhost and LAN installs) and threw
instead of uploading. Route it through randomId() like the other two
idempotency-key producers so it falls back to getRandomValues, and
update randomId's header comment to name all three producers.
The `| 0` truncation in colorForUserId/colorFor existed to coerce an
undefined user_id to 0, but the settlement API always returns user_id —
only the two BudgetPanel test fixtures omitted it. Add user_id to the
fixtures, switch both helpers to Math.trunc (with a Number.isFinite
guard in colorForUserId preserving the old undefined->0 behaviour), and
drop nothing else: TravelerPicker only ever receives required numbers.
Instance-specific findings (S2245 singles, S4790, S5332, S2819, S7767,
S4158 sessionManager, S8786, S6825) are resolved server-side as false
positive / accepted / reviewed-safe instead, so future violations of the
same rule in the same file still surface. Narrow the i18n secrets scope
to the locale directories so top-level code files stay fully scanned,
and drop the comment claims the code contradicts.
iOS paints its own background inside a text control, and under color-scheme:
dark that fill is darker than the card the field sits on. The mobile fields
are deliberately bg-transparent because the row supplies the surface - which
is why the native fill has nothing covering it and shows through as a dark
rectangle, while a desktop field hides it behind its own background.
Closes#2028
Admin-Storage pointed at the README's Storage section for the seed file
example and the compose mount line. That section is gone, so the link went
nowhere and the example existed in no copy at all - the wiki page described
storage-config.json without ever showing one. Both are on the page now.
The commented mount in docker-compose.yml pointed at the same removed
section.
Three headings sat directly on the paragraph above them while every other one
is preceded by a <br />. Sub-headings keep following their prose directly, as
they did before.
Three more sections that only restated what is maintained elsewhere, and had
started to drift: the embedded compose example still pinned tmpfs at 64m where
docker-compose.yml has said 128m for a while, and it was missing the proxy,
language and kitinerary options the real file documents inline.
The headings stay, so the in-page nav link to #docker-compose-production and
any external link into these anchors still land.
The scan that replaced the parenthetical-stripping regex went into the Leaflet
overlay and its Mapbox GL counterpart as two byte-identical copies, which is
what the duplication gate caught. trackColors.ts next door exists for exactly
this reason - both renderers read a track's colour through one function so it
cannot come out different on one map than the other - so the cleaner follows it.
Behaviour is unchanged. The test brute-forces the new function against the regex
it replaced over all 55,987 names up to six characters of "ab() \t", and pins the
32k-space input that used to take a second.
The move to SonarCloud brought a wider rule set with it, and 23 more regexes
that can be made to hang on a crafted near-match. 21 are fixed here, each proven
equivalent to the pattern it replaces by brute force rather than by reading.
Two of them are reachable across users, not just by yourself. A reservation
endpoint name is free text any trip member types, stored and then read back in
every other member's browser while the map draws its labels, and the pattern
over it takes 2.3 seconds on a 32k paste - so one member could freeze the map
for the whole trip. The mobile transport sheet reads the same field. Both now
scan instead of backtrack, and both were matched against the desktop twin the
earlier sweep had already fixed rather than given a third spelling.
The markdown link stripper needed the same treatment as the tag stripper before
it: the label class cannot be disambiguated in the pattern, because it matches
an opening bracket as well, and narrowing it changes what "[a[b](c)" produces.
It is a scan now, with all four of the regex's quirks intact.
Also here: the two encoding warnings the scanner reported. A decorative comment
line in JourneyDetailPage.test.tsx had picked up a replacement character
somewhere along the way, and calendar.service.test.ts asserts on one
deliberately - that one is now written as a U+FFFD escape, so the file no longer
carries the raw bytes while the test says exactly what it said before.
Seven findings were deliberately not changed and are flagged in SonarCloud with
the reasoning. Five are false positives - patterns with no overlapping
quantifier pair, where the only cure would change how nested brackets parse.
Two are marked accepted rather than dismissed, because they are real: the readme
check's heading pattern has a proven linear replacement that cannot land until
the registry's copy of the same script can move with it, and
`export let JWT_SECRET` is what makes an admin's key rotation take effect
in-process instead of silently waiting for a restart.
The coverage ratchet caught something worth keeping: an inline
`(a, b) => (a < b ? -1 : a > b ? 1 : 0)` is a new function with four branches at
every call site, and two of the three in src/nest/common sit on error paths that
a passing test never walks. Three restatements of the default sort order were
enough to pull the domain under its pinned thresholds (functions 97.8% against
99, branches 87.92% against 89).
Writing tests to walk those error paths would have been solving the wrong
problem. byCodeUnit says the same thing once, is covered once, and the callers
read better for it.
Its test asserts equality with a bare `.sort()` rather than hand-written
expectations, since "identical to the default order" is the whole contract. It
also pins what byCodeUnit is NOT: `['Zelt', 'Übernachtung']` keeps that order,
where localeCompare swaps it — so the next person reaching for a comparator on a
list a human reads picks the other one.
Two passes over the ReDoS findings arrived at the same scanner independently —
one replacing `/<[^>]+>/g` in the KML importer and the Commons metadata helper,
the other in the mail/HTML text extractor. The algorithm is the same down to the
edge cases, so the extractor now imports common/stripHtmlTags rather than
carrying its own.
Checked before folding them together rather than assumed: the two
implementations were compared over 2,441,406 generated strings and agree on
every one. The shared helper takes the replacement string as a parameter, which
is the only thing that differed — the extractor substitutes a space, the KML
importer nothing.
Two cases the shared helper's own test covers only in the abstract get a
concrete one here, since this is where the input is a file somebody uploaded: an
empty `<>` stays literal, and a `<` inside a tag is content, so
`3 < 4 and <b>5</b>` strips to `3 5` rather than keeping the first bracket. Both
are the old regex's behaviour, which is the point.
The bulk of what SonarQube still had open on the reliability side, read site by
site rather than swept. Two clusters are worth calling out because they are
defects, not lint.
**Four regexes could be made to hang the process.** The worst is
`/.+@.+\..+/` in the forgot-password path: three unbounded `.+` that all
overlap, unanchored, on an endpoint that needs no login and whose schema is a
bare z.string() under a 100kb body cap. A thousand '@' characters had not
finished after 100 seconds — one request pins the event loop, and Node has one.
EMAIL_REGEX in auth.helpers is the same shape on register and profile update
(310ms for 8kb). The two tag strippers over an uploaded KML description and over
Commons metadata took 0.52s on a 16k run of '<'. Each rewrite was proven
equivalent by brute force against the original — millions of strings for the
coordinate patterns, 671,876 for the tag stripper — because a subtly different
regex on a parsing path is worse than the finding.
`<[^>]+>` could not be fixed in the pattern: `[^>]` matches `<` too, so the only
unambiguous form changes what happens to malformed markup. It is now a scan in
common/stripHtmlTags.ts, quirks included — `<>` is left alone, `<a<b>` is one
tag — with the equivalence pinned by a test that compares against the old regex
over every string up to length six.
**And roughly 500 places could only be reached with a mouse.** Clickable divs
became real buttons where the markup allowed (52 of them), took
role + tabIndex + Enter/Space where they had to stay divs, and modal backdrops
took role="presentation" — the honest role, since a backdrop is a mouse shortcut
for a Cancel control that is already in the tab order, and making it a button
plants a nameless full-screen tab stop around its own contents.
Alongside those: `.sort()` calls that now say the order they always had, `{n &&
<X/>}` guards that rendered a literal "0" for a zero-byte file and for a place at
latitude 0, an optimistic budget write that stored a boolean where the contract
and both transports use 0/1, and the Number statics.
sonar-project.properties documents the six that were deliberately left: two wiki
patterns with no overlapping quantifier pair whose "fix" would stop parsing
`[see [1]](url)`, four tar header fields bounded at 12-155 bytes by the format,
two djb2 hashes where the 32-bit wrap is the algorithm, two `| 0` that are also
coercing undefined to 0 on a path where the API sends no user_id, and the MCP
session maps Sonar reads as empty because they are filled from another module.
Sonar's S2871 flags a bare .sort(), which orders by UTF-16 code unit rather than
by value. All 30 remaining sites turned out to sort ASCII strings — ISO dates,
currency and country codes, setting keys, route names, OAuth scopes, zip entry
names — where that order is already the intended one. Each says so explicitly
now, with a comparator that reproduces it exactly rather than a localeCompare
that would quietly reorder things.
That mattered more than it sounds. Several of these feed something other than a
display: useAtlas joins the sorted codes into the signature that decides whether
to rebuild the region layer, daysSlice re-pins sorted dates positionally onto
reordered days, and three timeline views pick a day's colour by indexOf into the
sorted array. Switching any of them to a locale collator would have been a silent
regression.
Two smaller ones came with it. The .map(placeBtn) calls in FileManagerAssignModal
pass the index as a second argument, which placeBtn declares and uses for its
React key — correct already, now written out as .map((p, i) => placeBtn(p, i)).
And budgetSlice's optimistic paid-toggle stored the raw boolean behind
`paid as unknown as number`, while the schema declares z.number() and both the
REST response and the WebSocket event carry 0/1. So the optimistic state and the
confirmed state disagreed on type until a refresh. It normalises now, and the two
tests that asserted `true` assert the contract.
Four `| 0` truncations are listed in sonar-project.properties instead of changed.
Two are a djb2 string hash where the 32-bit wrap is the algorithm. The other two
look like plain truncation of a row id, but `| 0` is also coercing undefined to
0 there — the settlement flows the API returns carry no user_id, so that is the
normal path, and Math.trunc returns NaN and crashes the lookup.
Adds a version picker to the plugin admin panel, on desktop and mobile: every
published version with its own server-computed compatibility verdict, so the
client never re-derives semver range logic.
A deliberate non-latest pick sets an update hold on the row, which keeps the
update banner and "Update all" from immediately nagging the rollback away. The
hold releases when an install lands back on the newest compatible version by any
path, or when the admin resumes updates explicitly. An unresolvable registry
never sets one: refusing to pin on missing information beats silently muting
future updates.
With a day selected, the 'planned' filter now shows only that day's stops,
matching how the other filters follow the selection; with no day selected
it keeps showing the whole plan. 'unplanned' still uses the whole-trip
assigned set. Discussed in #2024.
Two separate reasons v4.0.0 (#1719) is red.
The SonarQube workflow killed all 616 client test files with "Cannot find module
'@testing-library/dom'". That package is a peer dependency of
@testing-library/react and not a direct dependency of the client, and the step
that installs @swc/core's Linux binary passes --legacy-peer-deps, which tells npm
not to put peer dependencies back when it re-resolves the tree. In test.yml that
step lives in the server job, which never runs client tests, so folding the jobs
into one (#2039) is what brought the two together. It now installs alongside the
swc binary, with its version read from the peer range rather than pinned, and a
one-line resolve check right after fails fast with the module name instead of six
hundred identical failures twenty minutes later.
MAdminStoragePanel's FE-MOB-MSTOR-015 was a genuine race in the test, not in the
panel. Its POST handler incremented postCount, then awaited request.json(), and
only then pushed to migrationPosts — so between the two there was a window where
one POST had been counted but migrationPosts was still empty. The GET handler
reports `migrations: []` for an empty migrationPosts, i.e. "nothing running", so
a poll landing in that window told the queue it was free to start the second
migration, and the waitFor(postCount === 1) above could resolve inside it too.
Parsing before counting closes it. Verified by running the whole mobile suite at
--maxWorkers=8, where it used to surface.
Review feedback on #2038: the root already has a script that does all of this,
so the five jobs and the artifact shuffling between them were doing by hand what
`npm run test:cov` does on its own. It ends with `--prefix plugin-sdk`, so all
four packages are covered, and each writes its lcov to <pkg>/coverage/ — exactly
where sonar.javascript.lcov.reportPaths looks. 131 lines become 28.
Three things the plain root sequence still needs on a runner, kept as their own
steps so the reason stays visible:
- plugin-sdk is not a root workspace (workspaces = client, server, shared), so
the root npm ci never reaches it and it needs its own.
- The lockfile was generated on Windows and omits @swc/core's Linux optional
binary, so the server's SWC transform cannot load. test.yml carries the same
step for the same reason.
- Only shared is built, not the root `build`. server and client run their suites
from source through vitest, so building them adds a tsc pass and a full vite
bundle that nothing here reads.
The cost is wall-clock. The first run of the parallel version took 16 minutes,
with client (11) and server (9) overlapping; sequentially that is closer to 25.
Nobody waits on this job, so the simpler file is the better trade — but it is a
trade, not a free win.
The scan has been a manual step on somebody's machine, which is why the
dashboard could sit on a stale commit without anyone noticing. This runs it on
every push to dev.
Push to dev only, and that is the point: the instance is Community Edition, so
it has no branch or pull-request analysis and the project holds exactly one
view. Analysing a feature branch would overwrite that view with the branch's
numbers until somebody scanned dev again, so the one branch that may publish is
the one everybody reads.
The four coverage jobs mirror the ones in test.yml because Sonar needs the lcov
each package emits — without it every line reports as uncovered and the
dashboard shows a collapse that did not happen. They run in parallel and hand
their reports to the scan job, which needs a full clone: Sonar reads git blame
to date each issue and to decide what counts as new code.
sonar.projectVersion stays pinned here rather than read from package.json.
package.json is still on 3.4.1 while the project has been analysed as 4.0.0 all
along, and since New Code is "previous version", taking the lower number would
silently re-base every new-code metric.
Needs two repository secrets before it can run: SONAR_TOKEN and SONAR_HOST_URL.
The last pass reviewed diffs, so it never saw a new paragraph contradicting an
older one on the same page, or a long-standing claim nobody had checked.
Plugin authors were told ctx.config carries their per-user secrets and that
definePlugin takes a permissions key — neither is true, and both fail silently.
FORCE_HTTPS was said to need TRUST_PROXY, which it does not.
* fix(security): mint idempotency keys properly, and resolve the extractor by path
Three findings out of a sweep of every open Sonar security issue. The other 49
turned out to be false positives and are documented in sonar-project.properties
rather than changed — each entry scoped to a file, never a whole rule, so a
future Math.random() on an auth path still fails.
The idempotency key is the one that can lose data. Both producers fell back to
Math.random when crypto.randomUUID was missing, and randomUUID needs a secure
context — TREK's own quickstart is http://localhost:3000 and LAN installs run on
http://192.168.x.x, so on the deployments the README documents the fallback is
what actually runs. The axios one was
`Math.random().toString(36).slice(2)`, which is not length-stable: 0.5 yields a
single character and 0 yields the empty string, which the interceptor reads as
"no key" and skips deduplication for entirely. Two writes sharing a key collapse
into one — the second gets the first one's cached response and the edit is gone
with no error anywhere. Both now come from utils/randomId, which has the
getRandomValues rung that works in insecure contexts too.
The extractor probe stored the bare binary name, so every extraction re-resolved
it through whatever PATH held at that moment. The official image sets
KITINERARY_EXTRACTOR_PATH and never reaches that branch, but a bare-metal install
without it ran an unqualified name from a PATH it does not control. It now walks
the search path itself and keeps the absolute path of the file it verified;
execFileSync also drops the /bin/sh hop the old probe needed. PATH comes through
app-config because process.env is walled off in src/.
* fix: four tests and sorts that were not doing what they claimed
Sonar reported 43 high-impact reliability issues. Reading all of them, four are
real; the rest are cosmetic or would be made worse by the suggested change.
budget.test.ts hand-built `paid: true`, but both emitters send `paid: body.paid
? 1 : 0` and the handler stores the value verbatim — the test asserted a payload
the server never sends. The line below it already used a number.
admin.controller.test.ts had a `.resolves` assertion with no await. The promise
was dropped, so that line could not fail the test.
The two to-do category lists sorted with a bare .sort(), which is byte order:
every accented name files behind the whole ASCII range, so in a German trip
"Übernachtung" lands after "Zelt". TREK ships 23 locales, so this is real data.
One side effect worth knowing: katColor picks a category's dot colour by its
index, so re-sorting can move which category shows which colour — it already
moves whenever a category is added, and the colour carries no meaning.
Deliberately not changed: the two `| 0` truncations (Math.trunc is not
equivalent — `| 0` wraps at 32 bits and these carry millisecond timestamps), the
two .map() callbacks whose second parameter is declared and used, and the sha1
in thumbnail.service (its own comment says a swap would orphan every existing
thumbnail).
* refactor: give every button a type, and prefer the Number statics
Two mechanical sweeps, no behaviour change intended in either.
A <button> with no type attribute is type="submit" per the HTML spec. Outside a
form that does nothing today, but it turns into a surprise the moment the markup
is wrapped in one, which is what Sonar's S9011 is warning about. 870 buttons get
an explicit type="button". Two things were left alone on purpose: buttons that
take a {...spread}, because the spread may carry a type and the merge order
decides which wins, and anything inside a <form> — there the answer is a
judgement call, and checking all eleven form files found every real submit button
already carrying type="submit".
The Number sweep covers only the three globals that are exactly equivalent to
their Number twin: parseInt, parseFloat and NaN. isNaN is NOT in it — the global
coerces its argument, so isNaN('abc') is true where Number.isNaN('abc') is false,
and renaming it would change what the code does. It reaches server and plugin-sdk
as well as the client; both copies of egress-policy.ts got the identical edit, so
the mirror the SDK's parity test guards stays intact.
* docs(wiki): correct the install and operations pages for 4.0.0
Quick Start passed docker an empty ENCRYPTION_KEY: an assignment prefix is not
expanded on the same command line, so the server silently generated its own key
instead of taking the one the page told you to set. The tag examples pinned to
the 3.x line, the documented tmpfs size did not match the shipped compose file,
and several .env values were described as reaching the container when compose
never interpolates them.
* docs(wiki): match the MCP reference to the registered surface
The scope table was missing the whole Collections group and the totals said 27
scopes across 13 groups where the code registers 29 across 14. The Cloudflare
link on the setup page pointed at an anchor on its own page instead of the
heading on Troubleshooting.
* docs(wiki): correct the plugin and admin pages
Three links to Plugin-Development carried a .md suffix, which does not resolve on
the wiki. The admin pages described tabs, permissions and storage behaviour that
had moved on.
* docs(wiki): bring the feature pages back in line with the app
Costs still documented the old panel, where categories could be added, renamed,
reordered and deleted; they are fixed now and the panel is a day-grouped ledger
with settlements. Trip sharing, journeys, maps, packing, reservations and the
account pages had all drifted in the same way.
* docs(wiki): cover the pages the first pass did not reach
The MCP catalog listed 168 of the 201 registered tools; the settlement tools and
rate_place were missing here and the Collections family from the addon page.
Several resource rows claimed a trips:* scope where the handler asks for
trips:read.
* docs(wiki): tighten the corrections a review picked apart
The audit-log resource is the row id, not the credential id; an admin password
reset does not bump password_version the way the page implied; and a few of the
new paragraphs described behaviour more confidently than the code supports.
* docs(wiki): fix what a last read-through caught
Costs shows every figure in your display currency, not the trip's, and an
expense with no payer still counts against the members it is split between.
Studio's cutoff is a 1024px viewport rather than phones alone, and the
Integrations tab is gated on the providers an admin has configured.
* docs(wiki): the last few the review caught
A passkey satisfies the MFA policy everywhere except the WebSocket handshake,
which still asks for TOTP, so a passkey-only account under that policy gets no
live updates. Mobile optimizes a day from two places, not three. The ctx.db row
carried an unescaped pipe that split the table, and the Backups table linked
with a wiki link whose pipe did the same.
All three sit in code whose whole job is to survive a plugin that misbehaves.
readTarGz trusted the tar size field, which parseInt reads with a leading minus.
At -512 the cursor advance cancels out and below it goes backwards, so the same
header is read forever while every pass appends a member — a 70-byte upload pins
the main thread and exhausts the heap, synchronously, taking HTTP, the socket
and the database with it. The entry budgets never get a say: they run on the
list readTarGz has to finish building first. Sideload is the reachable path.
The supervisor handed onMessage's promise to an EventEmitter listener, which
discards it. A rejection there is an unhandledRejection, the host has no net for
it, and Node ends the process — the supervisor killed by the plugin it contains.
dispatch() read `'_inv' in req.params` before handle()'s try/catch, so params
that are not an object threw a TypeError into exactly that discarded promise.
`POST /api/admin/plugins/:id/uninstall` walks its route parameter into
pluginCodeDir() and hands the result to a recursive remove. Express decodes
`%2F` after routing, so `..%2F..%2Fuploads` matches the route and reaches the
helper as `../../uploads` — path.join then lands in the uploads tree, and
`deleteData` gets the data root the same way. Verified against the express
version in the lockfile.
pluginCodeDir and pluginDataDir now reject an id that could leave its parent.
Narrower than the manifest's ID_RE on purpose: that rule wants 3-40 lowercase
characters and belongs at install time, while this one only has to make the id
unable to steer a path, so it stays permissive about length and case. Every
other filesystem sink in the backend already has its guard — storage keys,
extraction, the wiki, backup filenames. This one did not.
Both functions had two exits handing back the identical value, which reads as a
decision and is none — Sonar's S3516, and the last two Blockers on the backend.
colorizeImportedTracks returned the very object it was given, coloured in place.
Saying `void` and letting the two importers return their own result is what the
function was doing all along. PLACES-SVC-005 checked the pass-through and now
checks the thing worth checking: an empty import writes no colours.
acceptInvite bailed out early on a refusal only to return the same result the
line below returns. The notification is what the early exit was really skipping,
so it is the notification that gets the condition.
SMTP_SKIP_TLS_VERIFY (and the matching admin setting) turns off certificate
checking for outgoing mail. That escape hatch stays — an install behind an
internal relay with a self-signed certificate depends on it — but until now it
downgraded the connection without a word, and a transport is built per send, so
whoever flipped it had nothing to go on afterwards.
It now warns once per process, naming the host it applies to. Once, not per
mail: a line on every notification is a line nobody reads.
It was written out twice — once in the seeder that creates the account, once
in the config payload the login screen reads it from. The two have to match
exactly or the published demo credentials stop working, which is a poor thing
to leave to copy-paste. It moves next to DEMO_EMAIL_PRIMARY, which the same
comment block already argues for.
Also clears a Sonar S2068 "hard-coded password" on the auth service: the value
is public on purpose, but a literal next to a `demo_password` key is exactly
what the rule looks for.
SonarQube's S2699 flagged 196 test cases as assertion-less. Almost all of
them end on a bare `await screen.findByText(...)`, which does throw when the
text never shows up — the test is real, the analyser just cannot see it.
Spelling the check out keeps the behaviour identical and makes it readable.
Writing them out turned three tests red, each for a good reason:
- FE-COMP-NOTES-008 read the heading out of the loading panel, which renders
its own copy and swaps the whole node out once the fetch settles. It now
waits for the loaded state.
- FE-COMP-PHOTOPROVIDERS-014 matched /connected/i, which also matches the
"Not connected" the badge starts as — the test passed whether or not the
button did anything. Exact text now.
- FE-FILES-002 compared an empty list against the empty list it already was.
It seeds a file first, so the 500 has something to leave alone.
JOURNEY-SVC-066 asserted nothing at all behind a "should not throw" comment;
it now says that out loud and checks no journey entry is left orphaned.
PluginRuntimeService booted enabled plugins from onModuleInit, but Nest fires
same-module onModuleInit hooks in providers-array order, where the runtime
precedes PluginRpcRegistryService — so every boot-activated plugin's rpc host
was bound from a still-empty registry and its first RPC (typically db.migrate
in onLoad) came back PERMISSION_DENIED on every cold boot, until an admin
deactivated and reactivated it by hand. Move the boot reconcile to
onApplicationBootstrap, which is guaranteed to run after every module's
onModuleInit, registry scan included.
Also fix the denial message, which interpolated the plugin id where the
permission belonged ('db.migrate requires a permission "featured-guides" was
not granted'); it now names both the missing permission and the plugin.
* fix(plugins): enforce manifest apiVersion at install and activation
* test(plugins): exercise the absent-apiVersion default
* feat(plugins): generate the core-event catalog into host-facts
* docs(plugins): explain the packing.listBags write-permission binding
* test(plugins): cover apiVersion upload refusal; tidy docblocks and generator header
* fix(plugin-sdk): validate settings options and oauth like the host
* fix(plugin-sdk): model the full manifest shape and stop dropping fields
* fix(plugin-sdk): drop non-string routeProfiles icons like the host
* docs(plugin-sdk): document host limits, oauth contract, and ignored asUserId
* feat(plugin-sdk): enforce daily budgets and meta quotas in the mock host
* feat(plugin-sdk): expose the generated core-event catalog
* feat(plugin-sdk): scaffold trek range >=3.4.0 <5.0.0 ahead of TREK 4.0
* feat(plugin-sdk): scaffold the oauth broker settings for oauth:client plugins
* chore(plugin-sdk): fix stale mirror docs and the colliding permissions export
* docs(plugin-sdk): add the rate-limit string; restore trek docblock; match host meta sizing
* feat(plugin-sdk): raise the scaffold trek floor to 4.0.0
* docs(wiki): bring the plugin pages in line with the current runtime
* docs: bring the workspace CLAUDE.md guides up to date and make them committable
* feat(plugin-sdk): roll back failed publishes, add unrelease, sync the fork, retro-sign updates
* docs(wiki): document the publish rollback, unrelease, fork sync, and retro-signing
* chore(plugin-sdk): 1.6.0
Closes#2019
The route toggle belongs to one day; the reservation list it filtered is the
whole trip. So turning on a day's routing let every transit reservation in the
trip draw its geometry at once, including the ones belonging to other days —
on a day that has no transport of its own, which is what the reporter saw.
Only automated transports rode it, and that is the distinction they drew: a
manually entered train or flight is type 'train'/'flight' and reaches the map
only through its own per-booking toggle. Only the transit search stores
type 'transit'.
Gate the transit branch on the day, and gate only that one — the per-booking
toggle stays trip-wide, because switching a booking's route on is a deliberate
act about that booking. The predicate is the day-order range, not the departure
day: an overnight journey carries end_day_id and has to stay on both of its
days, the same rule the sidebar timeline already applies. Day order rather than
day id, since ids stop being monotonic once days are reordered.
The day context is optional throughout, so a map without one — collections —
keeps behaving exactly as it did. Not a 4.0.0 regression: it reproduces on
3.4.1 too.
Closes#2021
A country is one path but not one landmass. France's is five rings, three of
them Kerguelen, French Guiana and New Caledonia, so fitting a map to its
bounding box asked for 221 degrees of longitude and printed a Paris-Lyon-Nice
route as a single dot. Britain does it through South Georgia, Norway through
Svalbard, Ecuador through the Galapagos — which is why a Netherlands journey
looks perfect and a France one does not.
Satellite only escaped by accident: imagery leaves the shape list empty, so the
fit fell through to the stops. Same control, silently doing nothing.
Fit to the rings the journey is actually near, and keep the whole shape as the
fallback for a map with no route yet. The margin is the route's own size with a
two-degree floor, which is the judgement in this fix: at two degrees a
Paris-Nice route still keeps Corsica, which is the picture that trip wants,
while Kerguelen stays out.
The outlines themselves are untouched — the overseas rings are still drawn,
they just run off the frame the way a coastline does.
Closes#2022
The story preview carried both `block` and `line-clamp-2`. Tailwind emits the
display plugin after line-clamp, same specificity, so `display:block` won and
`-webkit-line-clamp` — which only does anything on a -webkit-box — was inert.
The whole entry text rendered, the card grew with it, and the timeline it sits
in is bottom-anchored with no height of its own, so it climbed up the screen.
Measured at 430x932: three words put its top edge at y=744, four hundred words
at y=24, two thousand at y=-2916. That is the reporter's "at the top in one
journey, in the middle in another" — it is not the install, it is the longest
comment in the journey being opened.
The same growth explains the other half. The timeline sits above the map and is
a horizontal scroller, so once it covers the screen it takes every drag Leaflet
would have got, and only the gap between it and the dock still reaches the map.
Dropping the display utility fixes today's cards; capping the card height means
no future one can do this again. 140px is what the older journey timeline pins
its active card to, so the two look like the same product.
Also lift the map attribution over the dock, which the reporter's screenshots
show sitting on top of it. Both engines, since a phone meets either.
The redirect grew three paths nothing exercised: a malformed APP_URL falling
back to the request host, a request the proxy already terminated TLS for, and
the health probe, which talks plain HTTP on the loopback and must not be sent
to a hostname it cannot resolve. The per-domain coverage ratchet caught the
gap - src/middleware sat at 93.44% against its 94% pin.
safeExternalHref prefixed https:// to anything without a scheme, which turns
a note that reads "call the hotel" into a link to a host that does not exist.
A dot and no whitespace is what separates www.hotel.com from free text.
The rollbacks this branch added were snapshot-shaped: capture the whole
collection before the write, restore it on any failure. That is wrong in two
ways at once. It undoes work the server already accepted when a later step is
the one that fails, and it throws away whatever a collaborator's socket event
applied while the requests were in flight.
The day-plan reorder is the clearest case: five awaits, and a failing day-note
write reverted transport positions the server had already stored. It now
records only the fields it is about to overwrite, per booking, and drops each
one as its request lands - so a note failure reaches the catch with nothing to
undo, and a booking that arrived mid-write survives.
The notification toggle had the same shape underneath: the rollback was made
surgical but the request still shipped the whole matrix built from the
pre-toggle render, so two overlapping toggles overwrote each other anyway. It
sends the one cell that changed. The test pinned the disagreement as correct;
it asserts the payload now.
Photo reorder awaits allSettled rather than all, so the UI is not snapped back
while the remaining writes are still landing - a partial run really does leave
the server holding part of the new order, and only a total refusal justifies
restoring the old one. Both the desktop editor and the mobile sheet, which was
still swallowing every rejection.
Also: the geolocation start guard is a run generation now, because a start
issued after a stop re-sets a plain flag while the first run is still parked in
the iOS permission prompt; a category rename reports once for the batch and
rethrows, so the settings dialog stops closing as if it had worked; and a 404'd
import job stops telling the user their file was invalid when it had simply
expired.
The clipboard fallback for plain-HTTP installs was pasted into three files by
this branch. It is one helper now, and the two copy buttons in the members
modal that still called navigator.clipboard unguarded use it too. Same for the
fourth local-date helper written next to the three that already import the
existing one.
Four of the fixes already on this branch were wrong, and an adversarial read
of the diff is what turned them up.
The shared merge helper claimed to implement the server's
COALESCE(assignment_time, place_time) and did not: it pinned the embedded
times to whatever the assignment already showed, ignoring the override
column entirely. Sharing that with the WS applier traded one bug for
another - a collaborator moving a place from 08:00 to 10:00 stopped reaching
the day card for every assignment WITHOUT an override, which is most of them.
It merges the way the server projects now, and the test that was supposed to
pin this seeded a fixture the server could never produce (embedded times that
differ from the pool row with no override to explain the difference).
The per-iteration auth gate went into every loop in tripSyncManager except
the one that matters most: the mutation-queue replay, which writes server
responses straight into Dexie. After a logout that is the shared anonymous
database. The fire-and-forget tags/categories writes two lines below a gate
added in the same commit were ungated for the same reason - they resolve
later, and the check has to happen then.
safeHttpUrl is backed by placeWebsiteSchema: http(s) only, capped at 500
characters. A booking url is deliberately free-form - shared's own contract
says so in as many words, because people paste bare hosts and long provider
deep links - so guarding it with the stricter schema turned a link somebody
saved last year into dead text. safeExternalHref refuses the schemes that
execute in this origin and gives a scheme-less value https://, which is what
an address bar does with the same input.
And the memories timeout was 20s while the server's own budget for a Synology
call is 30s, doubled again when the session has expired and it re-authenticates:
a slow NAS that answered before this branch would have started reading as
disconnected. Those calls had no client timeout at all before; they have none
again, and the ones that need to give up early carry an AbortSignal.
The notes thumbnail also loses its NUL byte cache separator - it made git treat
the whole file as binary, so a security fix landed in it unreviewable - and
gains the same cache cap its chat sibling got.
Same behaviour, and the root lint job stops failing on no-control-regex. The
stripping is the point and has to stay: browsers drop everything at or below
U+0020 before they resolve a scheme, so a tab spliced into 'javascript:' is
still a javascript: URL to them.
An adversarial review of what had already landed here turned up three that
were wrong, and they are the reason for this commit.
The CSP change removed 'unsafe-eval', which is not leftover: heic-to's
libheif build initialises embind through new Function(), and without it every
iPhone .heic upload fails in the browser. 'wasm-unsafe-eval' alone was tried
once already and was not enough (93b51a0b). Put back, with the reason written
down this time and the test pinning it rather than its absence.
The idempotency in-flight guard never fired. finalize() runs when the
handler's observable completes, but Nest writes the response - and therefore
stores the row - several microtasks later, so the waiting request always woke
up before the row existed and ran the handler a second time. It releases from
the point the row is written now; finalize is the backstop for handlers that
answer another way. The test drove the two in the opposite order from Nest,
which is why it passed either way.
And the trip-scoping check treated an id that resolves to nothing like one
belonging to someone else. accommodation_id has no foreign key, so shortening
a trip's dates leaves a booking pointing at a gap - which update() has always
healed by clearing the link, and which would have become a permanent 400.
The rest: a reply to a soft-deleted chat message is refused and the create
path blanks the quote like the list path does; the ws-token buckets are keyed
by account rather than by IP, where every user behind one NAT shared a
ceiling; the weather and vacay caps read through the stream, since a chunked
response has no content-length to check; the Google host gate takes country
domains again; abandoned response bodies are cancelled instead of left to GC;
the boot failure prints its errno; and a share link that is not yours to read
now says so instead of reporting that the journey has none.
TICKETJSON gets a real repair: the in-place GLOB fix only helps installs that
have not run it yet, so an appended migration puts back the note text the
case-insensitive LIKE cut off, for rows where what it stored is not valid
JSON. A genuine receipt always parses, so it cannot be touched.
Wiki: the audit log records the IP Express resolves after TRUST_PROXY, not
the leftmost X-Forwarded-For entry a caller can write. That is the right way
round, but it means the hop count has to be accurate - said plainly in both
places now.
Several of the fixes in this branch landed on a mobile screen while the
desktop component with the identical bug sat in another package and was left
alone, which leaves the two disagreeing about the same handler.
So: the appearance tab flushes its debounced write on unmount, the journey
share copy guards navigator.clipboard (undefined on a plain-HTTP self-hosted
install), the admin collab toggle rolls back from a scoped snapshot instead
of a stale one, the reservation modal parses metadata through the guarded
helper rather than a bare JSON.parse that takes the route down, and the
notification toggle surfaces its failures and stops racing concurrent
toggles.
The Journey components reached the memories provider-status endpoint with a
raw fetch(), past every apiClient interceptor; they go through a small
memoriesApi now. And the server half of the OSM menu_url finding: the client
renders it through safeHttpUrl, so place-enrichment should not be handing a
javascript: value out in the first place.
The share section hides itself when the server says the link is not this
user's to manage, rather than offering a create button that 403s.
Three surfaces still had English written straight into JSX in a codebase that
gates on key parity: the reworked expense modal, the Studio quickbar and
rotation handles, and the leg-mode menu on the mobile day plan. Keyed, and
translated into all 23 locales rather than pasted.
Also reworded the plugin geolocation permission: it told the user their
browser would ask, and it never does - TREK answers the plugin from the
page's own location permission.
Left out of the commits they belong to by hand, not by intent: the mobile
screen suites, the tripRepo cache-first read behind the startup-destination
fix, and the MSW handlers the stricter unhandled-request setting needs.
The SDK's parity suite had an unescaped apostrophe in an it() title, so the
file did not parse and vitest reported it as "no tests". Both guards in it
were dead - including the byte-identical egress-policy check the package's own
notes call the worst possible bug in that module to lose. No CI job runs the
SDK suite, so it landed quietly; the next npm publish would have failed on it,
since prepublishOnly runs the tests.
Also in the SDK: the dev preview answered trek:geolocation without checking
the geolocation:read grant, so a plugin whose manifest is missing the
permission works perfectly in dev and is dead once installed. That is the
lenient direction, which is the one that ships broken plugins.
Client side: the background-task poll never dismissed a 404'd import job and
spun forever, "Update All" collapsed its busy flag after the first plugin
settled, the lightbox had no cancellation so a slow signed URL rendered the
wrong image, the plugin geolocation watch was torn down by any unrelated
settings write, manual date entry accepted 31 February, cover previews were
never revoked, and check-all issued one request per packing item.
Four screens derived "today" from toISOString(), which is the UTC calendar
day: an expense filed at 17:00 in Los Angeles was dated tomorrow, a to-do went
overdue eight hours early in Tokyo, and a new journal entry got the wrong day
either side of midnight. All four take the viewer's own calendar day now,
through the helper the planner already had.
Three sheets built preview object URLs in render and never revoked them, so
scrolling a picker leaked one blob per pass. Photo reorder fired its PATCHes
from inside a setState updater, which StrictMode double-invokes.
And several writes failed in silence: the notification toggle, the collab
feature switch rolling back to a stale snapshot, the appearance change whose
debounced write was dropped on unmount so it reverted on the next reload. The
backup-code print page skipped the escaping its desktop twin does, the
clipboard copy threw on a non-HTTPS install, and reservation metadata was
JSON.parsed unguarded in render, which takes the whole route down.
After a save conflict, Keep mine set the version and unblocked writes but
never re-queued the local document, and write() had already cleared the
queue before the request. So if the user picked Keep mine and then simply
closed the editor without typing anything else, saveNow() found nothing
pending and did nothing: their afternoon of edits existed only in memory,
the indicator said idle, and the server kept the other version. The exact
outcome the conflict dialog promises to prevent.
Ctrl+Z was intercepted while typing in a text field too. Studio commits text
on blur, so mid-edit the typing is not in the undo history yet: the browser's
own text undo was suppressed and a previous layout change was undone
underneath the open editor instead. Reads as random data loss.
Also: delete no longer removes locked elements, the auto-layout no longer
freezes an empty GPS path when tracks resolve after the one-shot build, the
back-cover tally cannot print "0 places", the map fetch aborts when Studio
closes, and a double Enter in the list-name field stops creating two lists.
The page-pattern gate had quietly stopped covering anything: the Desktop-shim
refactor moved every real container out of the default export the checker
reads, and its regex missed useRef<T>(null) anyway. Widened on both counts,
and the state it immediately found moved into the hooks.
The notes and polls WebSocket handlers applied every event they saw without
checking which trip it belonged to, so an event for another trip the user is
a member of landed in the open one. The poll load effect had no cancellation
either, so switching trips fast let the old response win.
What's Next compared trip-local calendar dates against a UTC one while
building the time from the local clock. West of UTC that means the rest of
today disappears from the widget every evening and tomorrow's items are
labelled Today; east of UTC it happens after local midnight instead.
A note website was rendered as a raw href with no scheme check, the same
stored-javascript: shape the planner had. Category rename walked the notes in
a loop with no handling for a partial failure. And the school-holiday region
labels resolved German first for every locale.
A booking's url is rendered as a plain href to every other member of the trip
and nothing ever checked its scheme, so a collaborator could store
javascript: there. The place website next to it has been going through
safeHttpUrl for exactly this reason, with a comment saying stored values
predate the server check - this is the surface that was left. Both ends now:
safeHttpUrl at the render site, and the scheme in the shared contract so
nothing new gets stored, additively so existing rows still load.
The per-leg routing rewrite turned one multi-waypoint request per run into
one serial request per leg. A twelve-stop day is eleven round trips to a
public OSRM in sequence before a single connector appears, and toggling km/mi
does it all again. Group consecutive legs that resolve to the same mode into
one request and fire the groups concurrently, keeping leg order on the join.
Transport reordering wrote to the store and persisted with a bare catch, so
a failed PUT left the sidebar showing an order the server never got; it
reconciles now. Plus the smaller ones: the OSM menu_url rendered unchecked,
expand/collapse-all persisting to a different storage than the day list reads,
the debounced address search with no ordering guard, the accommodation save
chain with no error path, and window.__trek_map holding a destroyed map.
logout() unregistered every sync trigger, and the only place that registers
them is App's mount effect - which never runs again, because logout navigates
through the router rather than reloading. Log out, log back in, go offline,
edit something: it queues and nothing ever flushes it. No online listener, no
visibilitychange, no 30s heartbeat, no WS reconnect hook. Register on auth
success instead; registerSyncTriggers is already idempotent.
The background sync loops only checked the auth gate on entry. Logout swaps
the Dexie handle to the shared anonymous database, so a trip loop or a blob
download still in flight afterwards wrote the logged-out account's itinerary
and documents into it - the database that survives logout and greets the next
person on a shared device. Check the gate per iteration.
And the remote place:updated applier replaced the whole embedded place, while
the local reducer right next to it preserves the assignment's own times. Same
event, two answers, and the remote one wipes a scheduled 09:00-10:30 off the
itinerary when someone else renames the place. One merge helper now, used by
both, so they cannot drift again. The test that should have caught it seeded
null times and asserted only the name.
Also: moving a day note deleted before it created, and restored the note only
locally when the create failed; t() interpolation was corrupted by a '$' in a
parameter value; a 429 upload was treated as permanently failed; and the
geolocation watch was started inside a state updater.
Book page setup and element geometry were unbounded numbers, so a shared
book could be saved with zero, negative or astronomical dimensions and every
consumer downstream - the canvas, the PDF export - had to cope. Day-note caps
had drifted from the limits their own doc comment states, and icon had none.
On the test side: the path filter meant a PR touching only the root lockfile
ran no jobs at all, which is exactly the change class most likely to break a
native binding; the canAccessTrip mock is hand-copied into 72 files and had
already drifted from the real query; the path-traversal test asserted only
status !== 200, so a 500 from a broken guard passed it; and one unit test
shelled out to a real kitinerary binary if the host happened to have one.
An unreadable or empty encryption-key file was treated the same as a missing
one: regenerate and overwrite. That is the one case where regenerating is
the worst possible move - every secret already at rest becomes undecryptable,
permanently. Both now refuse to start and say which file to fix.
The key-rotation script advertised re-encrypting all stored secrets while its
hardcoded column lists had drifted: the AirTrail key, the per-user LLM key
and the plugin OAuth tokens were never migrated, so they silently stopped
decrypting after a rotation. The parity test now pins the full set against
what the code actually encrypts, rather than two tables of it.
The schema_version bump ran outside the migration's transaction, so a crash
in between left the schema advanced and the version stale - the next boot
replayed a non-idempotent step and exited, turning one crash into a boot
loop. And the TICKETJSON receipt migration matched with LIKE, which SQLite
treats case-insensitively, so a note somebody had typed starting 'ticketjson:'
was cut up and its text dropped.
Also: APP_VERSION was interpolated into migration SQL instead of bound, the
capture-metadata backfill passed (photoId, userId) to a (userId, photoId)
signature and its mock keyed on the wrong argument, TRUST_PROXY=0 was
accepted and silently coerced to 1, demo-reset hardcoded the db path instead
of honouring TREK_DB_FILE, and a demo instance with no DEMO_ADMIN_PASS set
now says out loud that its admin password is the published default.
Permissions cached an empty map before it had loaded, so a DB error during
the load cached permissive defaults and every later check answered from that
cache - fail-open, and it never expired. The rate limiter grew its per-IP map
forever; the GC the legacy version had was dropped in the fold. Idempotency
did a check-then-insert, so two concurrent duplicates both ran the handler.
Audit rows recorded whatever X-Forwarded-For said, which the client writes,
and the request-body debug logger redacted by exact key match - it missed
refresh tokens, api keys and PKCE verifiers. FORCE_HTTPS built its redirect
from the client's Host header. CSP asked for unsafe-eval on top of the
wasm-unsafe-eval it actually needs.
The OAuth deny path redirected to an unvalidated URI, ws-token minting had
no rate limit at all, and PKCE was skipped when code_verifier was absent
rather than refused. An MCP session resumed with a token whose scopes had
since been reduced kept the capability set it was created with, and the
session cap was checked before the async registration that increments it.
Plugin frame containment was lexical, so a symlink out of the directory was
never noticed, and instance config accepted keys the manifest never
declared.
Files: two :id handlers took the file id from the URL and never resolved it
against :tripId. Membership of any trip was enough to read another trip's
file links (with the joined reservation titles) or to delete one. Both now
resolve the file the way every sibling handler already does.
Journey: a contributor with role 'viewer' - blocked from every other write
in the domain - could overwrite or delete the Studio book, and could read
the share token, which is an anonymous credential that outlives their
contributor status. The book writes take canEdit now and the token read
takes the owner check its own create and delete already use.
Also: the share-link MCP tool reset every flag to permissive defaults when
it was called to refresh a link, so a journey deliberately shared without
its gallery or map quietly started publishing both; setShareLink dropped
newest_first entirely; the public routes ignored the Journey addon switch;
and photo uploads committed to storage before the access check, leaking the
objects on the 403 path.
Trips: DELETE answered 403 to someone with no access at all, which turns
sequential ids into an existence oracle. A member who simply lacks
trip_delete still gets 403.
A deleted message keeps its row so the client can draw the placeholder off
the flag, but listMessages still returned the original text - in the list
itself and in the reply preview joined onto whatever quoted it. Both REST
and the MCP resource read through that one method, so blanking it there
covers every surface.
Also cap the link-preview read, and delete the spooled upload when
addNoteFile refuses the request instead of leaving it on disk.
The rule is a timeout, a size cap and boundary validation on anything that
leaves the process, and several call sites had only the timeout. A link
preview, a maps URL resolve, a wiki fallback and a holiday lookup all read
the whole body into memory, so a caller who picks the URL picks how much
memory the server spends - a few concurrent multi-hundred-megabyte responses
is the whole heap.
Two of them also fetched more than their name suggests. resolve-url fell
through to a full page GET for any host once it found no coordinates, which
makes the server a general-purpose outbound proxy wearing its own IP. The
weather URLs took lat/lng straight from the caller into the query string.
readCapped/exceedsDeclaredLength collect the pattern transit.service.ts and
the plugin installer had each grown separately, so there is one of it now.
Callers that only scan the head of a document (og: tags, a coordinate regex)
keep the truncated bytes and find fewer fields; callers that must have the
whole thing reject it.
The body-contract boot gate only looked at POST/PUT/PATCH, so a DELETE that
reads a body slipped past it. Two do; one was already typed, the other read
photo_id straight off an untyped Record. Give it a contract from shared like
everything else, then add DELETE to the gate so the next one cannot repeat it.
The ICS feed routes are documented as public - the secret token in the URL
is the credential - but nothing in the code said so, and the guard this
release makes global is default-deny. Every calendar client subscribed to
/api/feed/trip/<token>.ics would have started getting 401 on upgrade, and
a calendar app that stops fetching just goes quiet.
The e2e suite asserted anonymous access the whole time. It booted
[DatabaseModule, RealtimeModule, FeedsModule], and the APP_GUARDs live in
AppModule, so it was asserting against an app that had no opinion on
anonymity. It boots the real guard stack now, and fails without the fix.
The boot ratchet could not have caught this either: it only looked at routes
carrying @Public(), and GlobalAuthGuard stands down for ANY declared chain -
so a chain of guards that refuse a caller but never resolve one answers
strangers, invisibly. Split the inventory in two and write that half down as
well. It turns up two routes nobody had listed: the passkey login ceremony,
which genuinely runs before a session exists.
reservation_edit on the trip in the URL says the caller may write here. It
says nothing about the ids they put in the body, and the write contracts are
open records, so day_id, place_id, assignment_id and accommodation_id all
arrived unchecked. A booking created in one trip could carry another trip's
accommodation_id, and deleting it ran the cascade against that foreign row.
The MCP tools have refused foreign ids since they were written. Put the same
check where every writer reaches it, and scope the two day_accommodations
statements to the trip so a row stored before this still cannot be followed
out of it.
A flush marks each row 'syncing' before it sends, and only ever picks up
rows that are 'pending'. Kill the tab mid-request - the OS evicting a PWA
does exactly this - and the row stays 'syncing' forever: it keeps counting
towards the pending badge, but nothing ever retries it, fails it or shows
it as a conflict. Anything queued behind it that depends on it later fails
with 'unresolved temp id'.
Stamp the row when it goes in flight and reclaim it on the next flush. The
axios instance times out at 8s, so a minute is well past anything that
could still be running, and the mutation id doubles as the idempotency key,
so replaying a write the server already applied is answered from its cache.
Without an upper bound it would greet somebody who jumped 3.x straight to
4.2 with a summary of a release they skipped, and it would run alongside
the next release's own notice.
maxVersion is exclusive, so 4.1.0 covers every 4.0.x patch. Pinned in the
registry test, together with the handover from the thank-you notice: at no
version may both be active.
The thank-you modal was one column of markdown for every notice. A release
needs more room than that: what shipped, and why it took what it took.
Adds a `release` block to the notice contract. A notice carrying it renders
through ReleaseNoticeModal instead of the generic body — the release on the
left, the note from the maintainer on the right, both halves driven entirely
by translation keys so the next release only edits the registry entry.
Sizing lives on one CSS variable (--rn-s), so the panel scales as a piece and
the design's proportions hold at every breakpoint.
The thank-you notice gets maxVersion 4.0.0: from there on the release modal
carries the same thank-you and the same two support links, and installs still
on 3.x keep the old one.
All 23 locales.
This file provides guidance to Claude Code (claude.ai/code) when working with code in this repository.
Guidance for Claude Code in this repository. This file holds the commands, the invariants, the gates and the philosophy. Inventories (file lists, counts, history) deliberately live in the code and its READMEs, not here — when this file disagrees with the code, the code wins; fix the doc.
## What TREK is
A self-hosted, real-time collaborative travel planner. npm-workspaces monorepo with four workspaces:
A self-hosted, real-time collaborative travel planner. npm-workspaces monorepo:
- **`shared/`** (`@trek/shared`) — Zod schemas that are the **single source of truth** for API contracts, consumed by both server and client. Also owns all i18n locale files. Must be built before server/client typecheck or run.
- **`nest-mcp/`** (`@trek/nest-mcp`) — decorator-driven MCP tool/resource/prompt registration for NestJS, consumed by the server. Deliberately **extraction-clean**: no imports from `@trek/server`/`@trek/shared`, no TREK semantics.
- **`shared/`** (`@trek/shared`) — Zod schemas, the **single source of truth** for API contracts, plus all i18n locales. Must be built before server/client typecheck or run.
- **`server/`** (`@trek/server`) — NestJS API (Express adapter), SQLite via `better-sqlite3`, WebSocket sync, built-in MCP server, sandboxed plugin runtime.
- **`client/`** (`@trek/client`) — React 19 + Vite + Zustand + Tailwind PWA, offline-first via Dexie/IndexedDB.
- **`plugin-sdk/`** (`trek-plugin-sdk`) — **not a root workspace**: own lockfile, published to npm independently, must ship standalone. Run its commands from `plugin-sdk/`.
**`plugin-sdk/`** (`trek-plugin-sdk`) also lives in the repo but is **not a root workspace** — it has its own lockfile, is published to npm independently, and must ship standalone. Run its commands from `plugin-sdk/`, not the root.
Each package has its own `CLAUDE.md` with internals; this file covers the monorepo picture.
Each package has its own `CLAUDE.md`; this file covers the monorepo picture.
## Commands
Run from the repo root unless noted. Per the user's preference,**use `npm run ... --workspace=<pkg>`, not `npm -w <pkg>`**.
Run from the repo root unless noted. **Use `npm run ... --workspace=<pkg>`, not `npm -w <pkg>`**.
```bash
npm run dev # build shared + nest-mcp, then watch shared + server + client concurrently
npm run build # build shared → nest-mcp → server → client (order matters)
npm run lint # lint all four workspaces
npm run format # prettier --write all four
npm run dev # build shared, then watch shared + server + client concurrently
npm run build # build shared → server → client (order matters)
npm run lint # lint all three workspaces (server and shared run eslint --fix and rewrite files; client is check-only)
npm run format # prettier --write all three
```
Tests (the user runs client tests as `cd client && npm run test`):
Tests:
```bash
npm run test# all workspaces
npm run test --workspace=server # vitest run (server)
npm run test:unit --workspace=server # tests/unit only
npm run test:integration --workspace=server
npm run test:ws --workspace=server # websocket tests
npm run test:e2e # server e2e (boots Nest against temp SQLite)
npm run test:cov # coverage, server + client
npm run test --workspace=server # vitest run (server); also test:unit / test:integration / test:ws
npm run test:e2e # server e2e (boots Nest against a temp SQLite)
npm run test:cov # coverage (lcov) for all four packages incl. plugin-sdk
cd client && npm run test# client vitest run
cd client && npm run e2e # Playwright
cd client && npm run lint:pages # enforce the Page pattern (see below)
cd client && npm run theme:lint # flag styling that bypasses appearance tokens (--strict in CI mindset)
npm run typecheck --workspace=server # tsc --noEmit (also client/shared)
cd client && npm run e2e # Playwright (local only)
cd client && npm run lint:pages # enforce the Page pattern (CI gate)
cd client && npm run theme:lint # flag styling that bypasses appearance tokens (local only)
npm run typecheck --workspace=server # tsc --noEmit (also client/shared; server also has typecheck:tests)
```
Run a single test file or test:
@@ -50,65 +45,84 @@ npx vitest run tests/unit/nest/weather.controller.test.ts # (from server/ or c
npx vitest run -t "name of the test"# by test-name pattern
```
i18n parity (enforced in CI — every non-`en` locale must have identical file set + top-level keys):
i18n parity (CI gate — every non-`en` locale must have the identical file set and top-level keys):
```bash
npm run i18n:parity --workspace=shared # audit (exit 0)
npm run i18n:parity:strict --workspace=shared # CI gate (exit 1 on drift)
```
Dev ports: server listens on **3001**, Vite dev server proxies `/api`, `/ws`, `/uploads`, `/oauth`, etc. to `http://localhost:3001` (see `client/vite.config.js`).
Dev ports: the server listens on **3001**; the Vite dev server proxies the API, websocket, uploads, MCP and OAuth paths to it (see `client/vite.config.js`). There is no Node version pin — the CI workflows and the Dockerfile define what is tested. `server/.env.example` is the env-var reference.
## Server architecture
## Server invariants
Nest owns everything: the strangler migration **completed 2026-08-10**, and the last seam — the pre-`app.init()` MCP/OAuth mount — moved behind the container 2026-08-11: **zero `*.bridge.ts` files remain**. Legacy `src/services/` is deleted (ESLint errors on any import of it) and all business logic lives in DI modules under `server/src/nest/<domain>/`. `server/src/nest/README.md` is the module blueprint. Every cron is Nest-owned (`server/src/nest/scheduling/`, `@nestjs/schedule`).
Nest owns everything: every domain is a DI module under `server/src/nest/<domain>/` (`controller` + `service` + `module`, registered in `app.module.ts`), and ESLint refuses imports of the deleted legacy `services/` layer. `server/src/nest/README.md` is the module blueprint; **`weather/` is the reference implementation** — copy its shape. The domain's Zod contract lives in `shared/src/<domain>/`.
- **`bootstrap.ts` — `buildApp()`** is the single shared builder used by both production (`index.ts`) and the integration/e2e test harness. **Composition order is load-bearing**: global middleware → platform uploads → the discovery metadata middleware → static assets → the named body-parser wrappers are all registered on the raw Express instance **before `app.init()`**, because Nest's router throws `NotFoundException` on unmatched routes and won't fall through to anything registered after init. The transport surface (`/api/health`, `/mcp`, `/.well-known/*` documents, `/oauth/*`) is Nest controllers/middleware. The SPA `index.html` fallback and the global error envelope are `APP_FILTER`s (`SpaFallbackFilter`, `TrekExceptionFilter`).
- **`/mcp` bodies are raw by design** — bootstrap's parser wrappers exempt `/mcp` so the MCP SDK reads the untouched stream; never `@Body()` a `/mcp` route.
- **Per-domain module shape** (one folder under `server/src/nest/<domain>/`): `<domain>.controller.ts` + `<domain>.service.ts` + `<domain>.module.ts`, registered in `app.module.ts`. **`weather/` is the reference implementation** — copy its shape. The Zod contract for the domain lives in `shared/src/<domain>/<domain>.schema.ts`.
- **Auth is default-deny**: `GlobalAuthGuard` + `MfaPolicyGuard`run as `APP_GUARD`s; public routes opt out with `@Public()`, and a boot-time ratchet (`validate-route-guards.ts`) refuses unknown public routes. New domain logic goes in the module's `.service.ts` — never resurrect a `services/*`-style plain module.
- **Trip-scoped routes** must verify trip access (404 on no access) and the relevant permission (403) — the standard shape is `@UseGuards(JwtAuthGuard, TripAccessGuard)` + `@RequirePermission('<action>')` per handler (a few domains still check inline; see `roadmap.md`) — and forward `X-Socket-Id` to the WebSocket broadcast so the originating client doesn't echo its own change. Never gate a multipart upload with a guard (ECONNRESET instead of 403 — check in the handler).
- **DB** (`server/src/db/`): `better-sqlite3`, WAL mode, FK on. `database.ts` initializes; `schema.ts` creates tables; `migrations.ts` + `seeds.ts` run on boot. In `NODE_ENV=test` each vitest worker gets an isolated `:memory:` DB; the Playwright harness uses `TREK_DB_FILE` for a throwaway file DB.
- **WebSocket** (`server/src/websocket.ts`): JWT-authed, room-per-trip (`tripId → Set<WebSocket>`), heartbeat keep-alive, per-connection rate limiting, served on `/ws`.
- **MCP** (`server/src/mcp/`): OAuth 2.1-authenticated MCP server exposing TREK to AI assistants (150+ tools in `tools/`, resources, scopes, session manager). See `MCP.md`.
- **Addons** are admin-toggleable feature modules keyed by `ADDON_IDS` in `server/src/addons.ts` (mcp, packing, budget, documents, vacay, atlas, collab, journey, airtrail, llm_parsing, collections).
- **Plugins** (`server/src/nest/plugins/`): sandboxed third-party plugin runtime — one forked child process per plugin (`supervisor/`), permission-gated RPC (`host/rpc-host.ts` registers a handler only if the plugin holds the unlocking permission), install-time manifest/signature/egress checks (`install/`, `runtime/egress-policy.ts`). Author tooling lives in `plugin-sdk/`; the `trek-plugin-dev` skill covers plugin authoring.
- **`bootstrap.ts` — `buildApp()`** is the single builder for production and the test harness. **Composition order is load-bearing**: everything registered on the raw Express instance must come **before `app.init()`**, because Nest's router throws `NotFoundException` on unmatched routes and nothing registered after init is reachable. Don't add routes to that pre-init shell — it bypasses every global guard, interceptor and filter.
- **`/mcp` bodies are raw by design** — the parser wrappers exempt `/mcp` so the MCP SDK reads the untouched stream; never `@Body()` a `/mcp` route.
- **Auth is default-deny**: global guards run as `APP_GUARD`s; public routes opt out with `@Public()`, and a boot-time ratchet (`validate-route-guards.ts`) refuses any public route not on its allow-list.
- **Trip-scoped routes** verify trip access (404) and the permission (403) — `@UseGuards(JwtAuthGuard, TripAccessGuard)`+ `@RequirePermission('<action>')` per handler — and forward `X-Socket-Id` to the broadcast so the originating client doesn't echo its own change. **Never gate a multipart upload with a guard** (the client sees ECONNRESET instead of 403 — check in the handler).
- **DB** (`server/src/db/`): `better-sqlite3`, WAL, FK on. Migrations are positional and append-only. `NODE_ENV=test` gives each vitest worker an isolated `:memory:` DB.
- **Realtime**: `/ws` is a Nest gateway in `server/src/nest/realtime/` — inject `RealtimeService`; `server/src/websocket.ts` is a legacy re-export stub.
- **MCP**: tools are declared in `<domain>.mcp.ts` files beside each module, on the `server/src/nest-mcp/` decorator layer. A validation or permission change on a REST route must land in the parallel MCP tool in the same change. See `MCP.md`.
- **Addons** are admin-toggleable feature modules keyed by `ADDON_IDS` in `server/src/addons.ts`.
- **Plugins** (`server/src/nest/plugins/`): one forked child process per plugin, permission-gated RPC (a handler is registered only if the plugin holds the unlocking permission), install-time manifest/signature/egress/version-range gates. The protocol tables are **generated** into `plugin-sdk/` and `shared/` by `gen:plugin-facts`; `check:plugin-facts` is the CI gate. Author tooling is `plugin-sdk/`; the `trek-plugin-dev` skill covers authoring.
- **Storage**: uploads go through the driver abstraction in `server/src/nest/storage/` (local / S3 / mirror). Never touch the filesystem or the S3 SDK from a domain service.
- **Crons** go through `server/src/nest/scheduling/` only — never a `@Cron` decorator (it bypasses the test-time gate).
## Client architecture
## Client invariants
Offline-first, with a layered data flow. A Page never owns state directly:
- **Page pattern** (enforced by `lint:pages`, see`client/src/pages/PATTERN.md`): a `*Page.tsx` default export is a **wiring container** — it composes a co-located `use<Page>()` hook into JSX and must **not** call `useState/useReducer/useEffect/useLayoutEffect/useMemo/useCallback/useRef` in its own body. Stateful logic lives in the hook.
- **State**: Zustand stores in `client/src/store/`. The big one is `tripStore.ts`, assembled from **slices** in `store/slices/` (places, packing, todo, budget, reservations, files, …). `slices/remoteEventHandler.ts` applies incoming WebSocket events to local state.
- **Data flow (offline-first)**: `store → repo (client/src/repo/) → api (client/src/api/) | Dexie (client/src/db/offlineDb.ts)`. Repos check `navigator.onLine`: online they hit the REST API and upsert into Dexie; offline they read from Dexie. Writes go through `sync/mutationQueue.ts` — optimistic Dexie write, then on reconnect `flush()` replays REST with an `X-Idempotency-Key` header so retried writes don't double-apply. `sync/tripSyncManager.ts` and `sync/syncTriggers.ts` drive reconnection sync; `sync/tilePrefetcher.ts` caches map tiles.
- **Styling**: use the appearance tokens (`var(--token)` inline or `bg-[var(--...)]` Tailwind classes), never hardcoded color literals or numeric `fontSize` — `theme:lint` flags bypasses so user-chosen scheme/transparency/text-size keep working. Suppress intentional exceptions (map/PDF/brand colors) with a `theme-lint-disable` line comment.
- **Page pattern** (enforced by `lint:pages`, spec in`client/src/pages/PATTERN.md`): a `*Page.tsx` is a **wiring container** composing a co-located `use<Page>()` hook; it must not call React state/effect/memo hooks itself.
- **Data flow**: `store (Zustand, client/src/store/) → repo (client/src/repo/) → api (client/src/api/) | Dexie (client/src/db/offlineDb.ts)`. Writes go through `sync/mutationQueue.ts`: optimistic Dexie write, then a replay with an `X-Idempotency-Key` header on reconnect. `store/slices/remoteEventHandler.ts` applies incoming WebSocket events. Components never call the API directly.
- **Styling**: semantic Tailwind tokens (`bg-surface*`, `text-content*`, `border-edge*`, `bg-accent*`) or the underlying `var(--token)` variables — never color literals, arbitrary-value color classes or numeric inline `fontSize`, so user-chosen scheme/transparency/text-size keep working. `theme:lint` flags bypasses; suppress intentional exceptions (map/PDF/brand) with a `theme-lint-disable` line comment.
- **PWA**: `vite-plugin-pwa` + Workbox cache tiles, API and uploads; `prebuild` generates icons.
## Shared contracts & i18n
- A route is "done" only once its contract lives in `shared/` and both sides import the inferred types. Edit the Zod schema in `shared/src/<domain>/`, rebuild shared, then both server (validation + DTO types) and client (typed requests) pick it up.
-**i18n locale files live in `shared/src/i18n/<locale>/`**, one file per domain.`en/` is canonical. When you add/change a translation key, **add it to every locale** or `i18n:parity:strict` fails CI. The client resolves keys via `client/src/i18n/TranslationContext.tsx` (`t(key)`), which only reads top-level string keys.
- A route is "done" only once its contract lives in `shared/` and both sides import the inferred types. Edit the Zod schema, rebuild shared, then server (validation + DTO types) and client (typed requests) pick it up.
-Locale files live in `shared/src/i18n/<locale>/`, one file per domain;`en/` is canonical. When you add or change a key, **add a real translation to every locale** — parity fails CI on missing keys, and an English placeholder is not acceptable in a non-`en` locale.
## Project direction (from the 2026 full-repo audit)
## Project direction
A verified audit of the whole repo lives in `audit_report/TREK_audit_reports/` (`00_MASTER_SUMMARY.md` is the entrypoint, `01_STEERING_ROADMAP.md` the plan). Its verdict shapes how all new code should be written:
These principles come out of a full-repo audit and shape all new code:
- **Protect the crown jewels.** The client's offline-first data core (repos + mutation queue + Dexie), the out-of-process plugin sandbox, the Zod contract layer, and the auth primitives are production-grade. Extend them; never route around them or rewrite them in anger.
- **No "modern shell over legacy core."** The debt pattern is a modern frame delegating to the old approach underneath (historically Nest controllers over legacy `services/*` — that migration is done; on the client it's feature components calling the raw API past the offline core, still ~190 files). New code goes all-in on the target architecture; when touching a legacy seam, migrate it rather than adding another wrapper.
- **No "modern shell over legacy core."** The debt pattern is a modern frame delegating to the old approach underneath (on the client: feature components calling the raw API past the offline core). New code goes all-in on the target architecture; when touching a legacy seam, migrate it rather than adding another wrapper.
- **Single source of truth over manual synchronization.** Never add a hand-mirrored copy of state, schema, or contract (server↔Dexie↔Zustand, schema↔SQL, the plugin contract). Derive from one source, or guard the copy with a parity test that cannot silently skip.
- **DI over global mutable module state.** The `db` proxy, `ws` singleton, and event-sink globals are what force `require()`-to-dodge-cycles and load-bearing boot ordering. New server code injects its dependencies; new client code avoids `window`-global buses and mutable module state.
- **Use the runtime's native idioms.** Prefer React 19 features (`useOptimistic`, Actions, `use()`/Suspense) and Nest subsystems (DI providers, pipes, guards, `@nestjs/schedule`) over hand-rolled equivalents of the same machinery.
- **Fail closed, gates stay on.** Security switches default to safe; misconfiguration must refuse, not silently degrade. Never lower a quality gate (strict TS, lint severity, coverage, parity checks) to land a change — no new `any`, no new `eslint-disable`, no downgrading rules to `warn`.
- **DI over global mutable module state.** New server code injects its dependencies; new client code avoids `window`-global buses and mutable module state.
- **Use the runtime's native idioms.** Prefer React 19 features (`useOptimistic`, Actions, `use()`/Suspense) and Nest subsystems (providers, pipes, guards, `@nestjs/schedule`) over hand-rolled equivalents.
- **Fail closed, gates stay on.** Security switches default to safe; misconfiguration must refuse, not degrade. Never lower a quality gate to land a change — no new `any`, no new `eslint-disable`, no downgrading rules to `warn`, no lowered coverage thresholds.
## Conventions (from CONTRIBUTING.md)
- **Target the `dev` branch** for PRs, not `main` (exception: `wiki/`-only changes).
- **Discuss first**: outside contributions must be pitched in the `#github-pr` Discord channel before any code is written — undiscussed PRs are closed.
- **PRs follow `.github/PULL_REQUEST_TEMPLATE.md`** and require a linked issue (`Closes #N`) for bug fixes or an approved feature-request discussion for features — no issue/discussion, no PR.
- **Conventional commits** (`fix(maps): ...`, `feat(budget): ...`). Per user preference,**do not add Co-Authored-By Claude attribution**.
- One focused change per PR; no breaking changes; no unrelated reformatting/refactors. Tests required — the project holds **80%+ coverage** (the `src/nest/**` vitest coverage gate enforces ≥80%).
-When migrating/adding a route, **parity is law**: same URL, method, query/body, HTTP status, `Set-Cookie`, and JSON body — including bespoke error strings (e.g. reproduce `{ error: 'Admin only' }` exactly rather than relying on a generic guard message). Note Nest defaults POST to 201; add `@HttpCode(200)` where the legacy contract returned 200. Declare static sub-routes (`/reorder`, `/in-app/all`) **before**`:id` param routes.
- **Discuss first**: outside contributions are pitched in the `#github-pr` Discord channel before any code is written.
- **PRs follow `.github/PULL_REQUEST_TEMPLATE.md`** and need a linked issue (`Closes #N`) for fixes or an approved feature discussion for features.
- **Conventional commits** (`fix(maps): ...`, `feat(budget): ...`). **No Co-Authored-By or other tool-attribution trailers.**
- One focused change per PR; no breaking changes; no unrelated reformatting. Tests ship in the same change — the project holds 80%+ coverage.
-**Parity is law** when touching a route: same URL, method, query/body, HTTP status, `Set-Cookie` and JSON body, including bespoke error strings. Nest defaults POST to 201 — add `@HttpCode(200)` where the contract returns 200. Declare static sub-routes (`/reorder`) **before**`:id` param routes.
## Quality gates in CI
`.github/workflows/test.yml` runs, per package: typecheck (server also `typecheck:tests`), `lint:check`, `lint:pages`, `check:plugin-facts`, `i18n:parity:strict`, the S3 contract suite, coverage for all four packages, then a SonarCloud scan. `lint-prettier.yml` additionally runs `lint` + `format:check` on `shared/`. `plugin-sdk` has no lint script and is outside the eslint gates. **Not in CI**: `theme:lint`, Playwright, `check:gl-split` — run them locally. There are no git hooks; nothing runs before a commit except you. Other workflows build Docker images, scan them, and publish `trek-plugin-sdk` on `plugin-sdk-v*` tags.
**SonarCloud** (project `liketrek_TREK`, the built-in Sonar way gate; the run takes 20–25 min, so get it right before pushing) measures **new code only** on a PR; any failing condition blocks it:
| Condition (new code) | Must be |
|---|---|
| Coverage | ≥ 80% |
| Duplicated lines | ≤ 3% |
| Security / reliability / maintainability rating | A |
| Security hotspots reviewed | 100% |
- **Duplication is the one that bites.** Desktop and phone shells (`client/src/components/<X>` ↔ `client/src/mobile/screens/.../M<X>`) are deliberate mirrors that are already heavily duplicated. Every line added identically to both counts against the 3% budget, so a feature touching both shells fails on its own. Put the logic in ONE hook/module (precedents: `useInstanceSettings`, `useRangeBypass` under `client/src/components/Admin/`) and leave only markup in each shell. Locale files are excluded from duplication, so i18n additions are free.
- **Coverage is per new line across all four packages.** A new file without tests drags the PR under 80%. Files in `sonar.coverage.exclusions` don't count either way.
- **Ratings are about new issues.** One new bug drops reliability; one new vulnerability drops security. Don't introduce hotspot patterns (`Math.random` in anything security-adjacent, `child_process` through PATH, backtracking regexes, secret-looking strings). Path-wide suppressions live only in `sonar-project.properties` with a written justification; one-off findings are resolved on SonarCloud, never with an inline `// NOSONAR`.
- **Check it yourself.** The `sonarqube` MCP is connected: `list_pull_requests` → `get_project_quality_gate_status` shows each condition's value; `search_duplicated_files` + `get_duplications` show what tripped duplication; `search_sonar_issues_in_projects` lists new issues. Test files are excluded from analysis — when duplication fires, the source is the problem.
## Reference docs
-`MCP.md`— MCP server/tools/scopes.`README.md`— deployment, env vars, reverse-proxy setup.`server/src/nest/README.md`— per-module blueprint and test layout (unit / parity / e2e).
The repository ships a ready-to-use [`docker-compose.yml`](docker-compose.yml)
with secure defaults and every option documented inline. Download it, then:
```bash
docker compose up -d
```
**HTTPS notes:**`FORCE_HTTPS=true` is optional — it adds a 301 redirect, HSTS, CSP upgrade-insecure-requests, and forces the `secure` cookie flag. Only use it behind a TLS-terminating reverse proxy. `TRUST_PROXY=1` tells the server how many proxies sit in front so real client IPs and `X-Forwarded-Proto` work.
</details>
See [Install with Docker Compose](https://github.com/liketrek/TREK/wiki/Install-Docker-Compose)
for the full walkthrough.
<br />
@@ -309,6 +263,8 @@ helm install trek trek/trek
See [`charts/README.md`](https://github.com/liketrek/TREK/blob/main/charts/README.md) for values.
<br />
<h2 id="install-as-app-pwa">Install as App (PWA)</h2>
TREK works as a Progressive Web App — no App Store needed.
@@ -323,36 +279,10 @@ TREK then launches fullscreen with its own icon, just like a native app.
## Updating
**Docker Compose:**
See [Updating](https://github.com/liketrek/TREK/wiki/Updating) — Docker Compose,
Docker run, Helm, Portainer, Unraid and Proxmox, plus the encryption-key note.
> Not sure which paths you used? `docker inspect trek --format '{{json .Mounts}}'` before removing the container.
Your data stays in the mounted `data` and `uploads` volumes — updates never touch it.
> [!IMPORTANT]
> Mount **only** the data and uploads directories — `-v ./data:/app/data -v ./uploads:/app/uploads`. **Never mount a volume at `/app`.** Doing so hides the application code shipped in the image and the container fails to start with `Cannot find module 'tsconfig-paths/register'`. If you previously mounted `/app`, switch to the two mounts above; your data in `data/` and `uploads/` is preserved.
<h3>Rotating the Encryption Key</h3>
If you need to rotate `ENCRYPTION_KEY` (e.g. upgrading from a version that derived encryption from `JWT_SECRET`):
| `ENCRYPTION_KEY` | At-rest encryption key for stored secrets (API keys, MFA, SMTP, OIDC). Recommended: generate with `openssl rand -hex 32`. If unset, falls back to `data/.jwt_secret` (existing installs) or auto-generates a key (fresh installs). | Auto |
| `TZ` | Timezone for logs, reminders and cron jobs (e.g. `Europe/Berlin`) | `UTC` |
| `TREK_WIKI_DIR` | Where the in-app Help pages (`/help`) read their content from. TREK ships its wiki and serves it from disk, so Help always matches the version you are running — you should not need to set this. Point it at your own directory to serve custom docs. If the path does not exist, Help falls back to fetching the public GitHub wiki (needs outbound network, and tracks the latest release). | bundled `wiki/` |
| `DEFAULT_LANGUAGE` | Default language on the login page for users with no saved preference. Browser/OS language is auto-detected first; this is the fallback. Supported: `de`, `en`, `es`, `fr`, `hu`, `nl`, `br`, `cs`, `pl`, `ru`, `zh`, `zh-TW`, `it`, `ar`, `id`, `tr`, `ja`, `ko`, `uk`, `gr` | `en` |
| `ALLOWED_ORIGINS` | Comma-separated origins for CORS and email links | same-origin |
| `FORCE_HTTPS` | Optional. When `true`: 301-redirects HTTP to HTTPS, sends HSTS, adds CSP `upgrade-insecure-requests`, forces the session cookie `secure` flag. Useful behind a TLS-terminating reverse proxy. Requires `TRUST_PROXY`. | `false` |
| `HSTS_INCLUDE_SUBDOMAINS` | When `true`: adds the `includeSubDomains` directive to the HSTS header, extending HTTPS enforcement to all subdomains. Only effective when HSTS is active (`FORCE_HTTPS=true` or `NODE_ENV=production`). Leave `false` if you run other services on sibling subdomains over plain HTTP. | `false` |
| `COOKIE_SECURE` | Controls the `secure` flag on the `trek_session` cookie. Auto-derived: on when `NODE_ENV=production` or `FORCE_HTTPS=true`. Escape hatch: set `false` to allow session cookies over plain HTTP. Not recommended in production. | auto |
| `SESSION_DURATION` | How long a login session stays valid when **"Remember me" is unchecked** (the default): sets the `trek_session` JWT `exp` and issues a browser-session cookie (cleared when the browser closes). Accepts `ms`-style strings: `1h`, `12h`, `7d`, `30d`, `90d`. Invalid values warn at startup and fall back to the default. | `24h` |
| `SESSION_DURATION_REMEMBER` | Session length when **"Remember me" is ticked** at login: a longer-lived JWT plus a persistent `trek_session` cookie that survives browser restarts. Same format and startup-fallback behaviour as `SESSION_DURATION`. | `30d` |
| `TRUST_PROXY` | Number of trusted reverse proxies. Tells the server to read client IP from `X-Forwarded-For` and protocol from `X-Forwarded-Proto`. Defaults to `1` in production; off in dev unless set. | `1` |
| `ALLOW_INTERNAL_NETWORK` | Allow outbound requests to private/RFC-1918 IPs (e.g. Immich on your LAN). Loopback and link-local addresses remain blocked. | `false` |
| `APP_URL` | Public base URL of this instance (e.g. `https://trek.example.com`). Required when OIDC is enabled; used as base for email notification links. | — |
| `ADMIN_EMAIL` | Email for the first admin on initial boot. Must be set together with `ADMIN_PASSWORD`. If either is omitted a random password is printed to the server log. No effect once a user exists. | `admin@trek.local` |
| `ADMIN_PASSWORD` | Password for the first admin on initial boot. Pairs with `ADMIN_EMAIL`. | random |
| **Other** | | |
| `TREK_MANAGED` | Marks the install as centrally administered: the operator owns the configuration, credentials and upgrades, so the instance admin is not offered settings the operator sets. Leave unset when you run TREK yourself. | `false` |
| `PLACES_API_KEY` | Places credential supplied by whoever operates the install. When set, it is used for every lookup and the per-user keys in Admin > Settings are ignored. Leave unset to keep the stored keys. | — |
| `PLACES_API_BASE` | Send the Places calls to this origin instead of `https://places.googleapis.com` (an egress proxy, a cache, a gateway holding the key). Path and query are unchanged, so it has to speak the same API. | — |
| `UNSPLASH_ACCESS_KEY` | Optional Unsplash Access Key for trip-cover and place-image search. Without one, TREK uses Unsplash's unauthenticated endpoint, which some datacenter/VPS IPs are blocked from. Get a free key at [unsplash.com/developers](https://unsplash.com/developers). Overrides any per-admin key set in Admin > Settings (where it can also be configured instead). | — |
| `MCP_RATE_LIMIT` | Max MCP API requests per user per minute | `300` |
| `MCP_MAX_SESSION_PER_USER` | Max concurrent MCP sessions per user. At the cap, the least-recently-active session is closed to make room | `20` |
<img alt="Star History Chart" src="https://api.star-history.com/chart?repos=liketrek/TREK&type=date&legend=bottom-right&sealed_token=tW9agibKugOyPk0yPURgSdnizM7K72zbJENGD9iyk24ZcwYHtCUn-hk9jh1gdacO0O7xJ3Io7LaNrwYD3mChzcpzZIqDC8_HevYG1qnT0XNTmN6Ez4UZrQ" />
</picture>
</a>
## Data sources
@@ -609,6 +382,8 @@ The Atlas map's country and sub-national (province/county) boundaries come from
[CC BY 4.0](https://creativecommons.org/licenses/by/4.0/). See [NOTICE.md](NOTICE.md)
for full third-party attributions.
<br />
## License
TREK is [AGPL v3](LICENSE). Self-host freely for personal or internal company use. If you modify and offer TREK as a network service to third parties, your modifications must be open-sourced under the same licence.
@@ -8,7 +8,7 @@ This is a minimal Helm chart for deploying the TREK app.
- Optional persistent storage for `/app/data` and `/app/uploads`
- Configurable environment variables and secrets
- Optional generic Ingress support
-Health checks on `/api/health`
-Configurable health checks (liveness/readiness probes) on `/api/health`
## Helm Repository
@@ -46,6 +46,7 @@ See `values.yaml` for more options.
- With `persistence.enabled: false`, the data and uploads volumes use an `emptyDir` — storage is ephemeral and lost on pod restart. Intended for testing only.
-`JWT_SECRET` is managed entirely by the server — auto-generated into the data PVC on first start and rotatable via the admin panel (Settings → Danger Zone). No Helm configuration needed.
-`ENCRYPTION_KEY` encrypts stored secrets (API keys, MFA, SMTP, OIDC) at rest. Recommended: set via `secretEnv.ENCRYPTION_KEY` or `existingSecret`. If left empty, the server falls back automatically: existing installs use `data/.jwt_secret` (no action needed on upgrade); fresh installs auto-generate a key persisted to the data PVC.
- Liveness and readiness probes are configurable under `probes`. Overrides are merged over the defaults, so `--set probes.liveness.initialDelaySeconds=60` keeps the `httpGet` handler. When switching to `exec` or `tcpSocket`, clear `httpGet` in the same override. Set `probes.liveness: null` to remove a probe.
- If using ingress, you must manually keep `env.ALLOWED_ORIGINS` and `ingress.hosts` in sync to ensure CORS works correctly. The chart does not sync these automatically.
- Set `env.ALLOW_INTERNAL_NETWORK: "true"` if Immich or other integrated services are hosted on a private/RFC-1918 address (e.g. a pod on the same cluster or a NAS on your LAN). Loopback (`127.x`) and link-local/metadata addresses (`169.254.x`) remain blocked regardless.
-`FORCE_HTTPS` is optional. Set `env.FORCE_HTTPS: "true"` only when ingress (or another proxy) terminates TLS. It enables HTTPS redirects, HSTS, CSP `upgrade-insecure-requests`, and forces the session cookie `secure` flag. Requires `TRUST_PROXY` to be set.
Scope: the **`@trek/client`** workspace (React 19 + Vite + Zustand + Tailwind PWA). See the repo-root `CLAUDE.md` for the monorepo picture and the philosophy, `server/CLAUDE.md` for the API. This file holds the client's commands, invariants and pitfalls; `ls` the directories for the inventory.
## Commands (run from `client/`)
```bash
npm run dev # Vite dev server; proxies the API/ws/uploads/MCP/OAuth paths → http://localhost:3001
npm run build # prebuild generates PWA icons, then vite build
npm run typecheck # tsc --noEmit (CI)
npm run lint # eslint . (lint:check in CI)
npm run lint:pages # enforce the Page pattern (CI gate)
npm run test# vitest run (tests/** + co-located src/**/*.test.{ts,tsx}); also test:unit / test:integration / test:coverage
npm run e2e # Playwright (local only; e2e:report opens the last report)
npm run shots # Playwright screenshot project (shots:promote to accept)
npm run theme:lint # theme conformance audit (theme:lint:strict exits 1; local only)
npm run check:gl-split # after build: fails if one chunk bundles both mapbox-gl and maplibre-gl (local only)
```
Single test: `npx vitest run src/store/slices/budgetSlice.test.ts`, or `npx vitest run -t "optimistically adds the place"`.
The dev server needs the API on **:3001** — run `npm run dev` at the repo root to start both.
Every `src/pages/*Page.tsx` is a thin **wiring container**; all state/effects/handlers live in a co-located **`use<Page>()` hook** under `src/pages/<page>/`. The page body must **not** call React state/effect/memo/ref hooks — only context hooks like `useTranslation()`. An optional `<page>Model.ts` holds pure, React-free types and helpers. Full spec in `src/pages/PATTERN.md`. When extracting, keep the rendered JSX byte-identical — it is a refactor of where logic lives.
## Data flow — offline-first
The layering is **component → feature hook → store/slice → `repo/` → `api/` | Dexie**, and writes go through a mutation queue:
- **`src/store/`** — Zustand. `tripStore.ts` is composed from slices in `store/slices/`; `slices/remoteEventHandler.ts` applies inbound WebSocket events to local state.
- **`src/repo/`** — per-entity repositories mediating between the network and the offline cache: online, call REST and upsert into Dexie; offline, read Dexie. Always go through a repo for trip data.
- **`src/api/client.ts`** — the single Axios instance, typed from `@trek/shared`. New API surface goes in a per-domain `api/<domain>.ts`, not the god module.
- **`src/db/offlineDb.ts`** — the Dexie schema. Database and table names are the on-device contract — renaming one orphans every user's cache.
- **`src/sync/`** — `mutationQueue.ts`: offline writes do an optimistic Dexie write (temporary **negative id** for creates), then `flush()` replays REST on reconnect with the mutation's UUID as the **`X-Idempotency-Key`** header (matching the server interceptor) and reconciles the row. The rest of the directory drives reconnection sync, the auth gate that stops background loops before logout swaps the DB, offline preferences, and map pre-download (raster and vector prefetchers).
## Rules for new code
The offline core is flagship work surrounded by a periphery that ignores it. New code extends the core's discipline, not the periphery's shortcuts:
- **One data path, no layer skips.** Never import `api/client` (or raw fetch/axios) from a component or modal — especially never *write* from one. A long tail of existing files bypasses the layering and no lint ratchet holds the line yet; that is debt, not precedent.
- **Offline-first is the product promise.** New domains get a repo with read-through and offline writes (temp id → optimistic Dexie write → idempotent queue entry). If a domain is deliberately online-only, say so explicitly.
- **No god components or god hooks.** The page pattern is a floor, not a ceiling. Decompose hooks by concern, split renders into memoizable sections, and don't pass huge prop bags with inline-arrow callbacks.
- **Prefer React 19 idioms**: `useOptimistic`, `useActionState`/`useFormStatus`, `use()` + Suspense, `useSyncExternalStore` for external subscriptions (`hooks/useNetworkMode` is the reference).
- **Optimistic writes must reconcile** — rollback + user-visible toast on failure. No `.catch(() => {})`, no bare `catch {}`.
- **WS remote-event handling must match local slice reducers field-for-field.** Divergence is silent, cross-user data loss.
- **Subscribe with selectors** (`useShallow` for multi-field reads); never whole-store subscriptions; `getState()` for imperative access in effects.
- **Connectivity**: `isEffectivelyOffline()` is the single source of truth — never read `navigator.onLine` in feature code.
- **Rendering security**: never interpolate user content into HTML strings (map popups included — build via DOM + `textContent`, or `escapeHtml` from `@trek/shared`); untrusted markdown gets `rehype-sanitize`; never `rehype-raw` near it.
- **No `window` event buses or global mutable `window` state.** No `any` at boundaries: WS payloads and map renderer props get real types.
- **Hygiene**: error boundaries around new shells/routes (`components/shared/ErrorBoundary.tsx`); every async `.then(setState)` needs a cancelled flag or `AbortController`; no routine `eslint-disable exhaustive-deps` (a missing dep has already shipped wrong money on screen); no sequential-await N+1 fetch loops; search for an existing utility before writing a duplicate.
- **Theming**: use the semantic Tailwind tokens defined in `tailwind.config.js` (`bg-surface*`, `text-content*`, `border-edge*`, `bg-accent*`, status colors) — no palette classes, hex literals, arbitrary-value color classes, or invented CSS vars. Only `applyAppearance()` in `src/theme/` mutates `<html>` styling (see `src/theme/README.md`). `theme:lint` catches literals and `bg-[#...]`-style classes but not named palette classes — review those by hand.
## Big-picture pieces
- **Maps** (`src/components/Map/`): two interchangeable renderers — Leaflet and Mapbox/MapLibre GL — chosen at runtime by `MapViewAuto.tsx`. Keep both in sync when changing map features, and keep `mapbox-gl` and `maplibre-gl` in separate chunks (`check:gl-split`). The raster prefetcher fetches tiles `no-cors` so custom tile providers without CORS headers keep working; the vector prefetcher deliberately uses `cors` because it has to read the responses — don't "align" one with the other.
- **i18n** (`src/i18n/TranslationContext.tsx`): `en` is bundled; every other locale is a dynamic `import('@trek/shared/i18n/<locale>')` so Vite emits one chunk per locale. Strings live in `shared/`, never here.
- **Mobile shell** (`src/mobile/`): below the phone breakpoint (`useIsPhone`) `App.tsx` wraps routes in `MobileShell` and the `M*` screens under `mobile/screens/` take over. A UI change to a domain with an `M*` twin usually needs both — and the shared-hook rule below.
- **Plugins UI** (`src/components/Plugins/`): third-party surfaces render only inside the sandboxed `PluginFrame` iframe (postMessage bridge, live theme-token sync). Host-rendered contribution points (widgets, schedule rows, map layers, planner columns/actions) render server-normalized primitives only — plugin markup never runs inline. Authoring is covered by the `trek-plugin-dev` skill.
- **Managed installs** (`src/managed/index.tsx`): the attachment point for screens that only exist on a centrally administered install. Empty here by design — an operator replaces it at build time. Don't put features there.
## Desktop ↔ mobile shells and the SonarQube duplication gate
Every desktop component with a phone twin is a deliberate copy that is already heavily duplicated. CI's SonarCloud gate allows **≤ 3% duplicated lines on the PR's new code**, so adding the same lines to both shells fails the PR by itself. Put logic in one shared hook/module (`useInstanceSettings`, `useRangeBypass` under `components/Admin/` are the pattern) and keep only markup in each shell. Full gate rules in the root `CLAUDE.md`.
## Tests
vitest with `@vitejs/plugin-react`, a custom jsdom environment (`tests/environment/`), `forks` pool. Tests live in `tests/{unit,integration}/` and co-located as `src/**/*.test.{ts,tsx}`. `msw` mocks HTTP, `fake-indexeddb` backs Dexie. Page tests render JSX against a mocked hook; hook/slice logic is tested in isolation (see `store/slices/budgetSlice.test.ts`).
<inputtype="password"className={fieldCls}value={apiKey}onChange={e=>setApiKey(e.target.value)}placeholder={apiKey===MASKED?MASKED : provider==='local'?'(often not required)':'sk-…'}/>
<inputautoComplete="off"className={fieldCls}value={model}onChange={e=>setModel(e.target.value)}placeholder={provider==='anthropic'?'claude-opus-4-8':provider==='openai'?'gpt-4o':'select or pull below'}/>
<inputtype="password"className={fieldCls}value={apiKey}onChange={e=>setApiKey(e.target.value)}placeholder={apiKey===MASKED?MASKED : provider==='local'?'(often not required)':'sk-…'}/>
<inputautoComplete="off"className={fieldCls}value={model}onChange={e=>setModel(e.target.value)}placeholder={provider==='anthropic'?'claude-opus-4-8':provider==='openai'?'gpt-4o':'select or pull below'}/>
it('FE-COMP-PLUGINS-ACT-009: a 404 "Plugin is not active" from the server (DB active, no live child) shows the inactive message and flips active off',async()=>{
vi.mocked(adminApi.runPluginAction).mockRejectedValue({response:{status: 404,data:{error:'Plugin is not active'}}})
:`Sum of splits: ${sym(currency)}${splitSum.toFixed(2)} of ${sym(currency)}${totalNum.toFixed(2)} (${(totalNum-splitSum)>0?'under by':'over by'}${sym(currency)}${Math.abs(totalNum-splitSum).toFixed(2)})`}
Some files were not shown because too many files have changed in this diff
Show More
Reference in New Issue
Block a user
Blocking a user prevents them from interacting with repositories, such as opening or commenting on pull requests or issues. Learn more about blocking a user.
Maurice