Repository navigation
Tablet layouts, AGP 9, and the fixes from a full QA pass - #8
Merged
Merged
Conversation
upcomingTrips.firstOrNull() meant the second and third trips the server sent were fetched and thrown away; the web lists all of them. Recently updated had a bare count with no way to reach the other 19 of 22 places. Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
The Collection section was built but never showed: collections came from
_locationState.map reading allCollections.value once, and the location
usually finishes loading before that flow's first emission lands, so the
filter ran against an empty list forever. Switched to combine().
That uncovered a second bug once the chip became reachable: tapping it
called navigate() on the app-level NavController, but the collection
detail route only exists on MainShell's own nested one - an
IllegalArgumentException that killed the app on every tap. Registered
the collection screens on the outer graph too, reusing the existing
"detail/{locationId}" route for the read path; the rarer write actions
(edit collection, add transportation, edit an unrelated adventure) fall
back to going home rather than duplicating those destinations again.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
…y completed ones The percentage used completeCount, on the assumption that's what the web does. Checked live: this account has 6 visited / 0 complete out of 250, and the web's own bar reads "2%" — that's 6/250, not 0/250. The web counts any visit as progress; only the per-country row distinguishes complete from partial. Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
…ogged The list rendered days in chronological order with no scroll-to-today, so an account with a year of history landed a "See all" tap on September 2025 instead of anywhere near the present. The web opens its calendar on the current month for the same reason this needed fixing here. Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
…t just All The add button existed in the All tab's Transportations section header but was never copied into the dedicated Transportations tab - its empty state invited you to "start adding transportation methods" with no way to actually do it. This is likely why it was never verified end to end. Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
…d say what got deleted
The form defaulted both timezone fields to the literal string "UTC" - the server's
choices don't include it ('"UTC" is not a valid choice'), so every transportation
creation failed. Default to the device's own IANA zone instead, which the server
always accepts.
Also: deleting a transportation reported "Adventure deleted successfully" - the
delete-confirmation snackbar was shared with place deletion and never distinguished
which one happened. DeleteState.Success now carries its own message.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Archiving succeeded on the server but the collection stayed on screen until the process died: setArchived was the one mutation that neither updated the cached list nor bumped the version counter paging invalidates on, unlike create, delete and update. Restoring appeared to work only because the Archived tab isn't paged. The header count was wrong in the other direction - it came from /api/collections/all/, which includes archived collections, while the list under it comes from /api/collections/, which doesn't. Six collections, five cards. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Everything was already wired - the clustering marker calls onAdventureClick with the id, and MainShell already takes an onOpenLocationById that reaches goToDetailById. The two were just never connected: the map's callback was an empty TODO, so tapping a pin showed its name and went nowhere. The web's pin popup offers "Ver detalles" for the same journey. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
…d on grey The "Route Map" on the transportation form was a placeholder that shipped: a 300dp grey box printing "Map View (No coordinates set)", or the raw lat/lng pair as text once you had searched for one. The signature was already right, so this makes it an expect/actual like every other map in the app - Google Maps on Android with a marker per end and a line between them, framed to hold both; MapKit on iOS. With neither end set it now says what to do instead of naming a state nobody asked about. Also drops the 4dp elevation on that card, which no other card in the app has. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Last of the TODO stubs sitting at the end of a finished chain: the button was
drawn, the callback was declared, and the body was a comment. The web has the
same button ("Abrir mapa") above its country list.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Starting the app with no network logged the user out for good: the token check is a plain network call, and its catch treated every exception as a bad token - so an unreachable server cleared the stored session, and reconnecting did not bring it back. Found by turning the emulator's network off and cold-starting. Now only a 401 or 403 clears it. Anything else keeps the session and carries on with the stored profile; the screens that need the network still fail on their own and say so, which is recoverable in a way a logout is not. ApiResponse already drew this distinction - this use case just never used it. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Home stated the problem and stopped there. The commonest reason to reach it is no signal - which is over the moment it is over - so the only way back was to restart the app or switch tabs and return. The calendar has offered "Try again" all along; this is the same, calling loadDashboard(). Worth having now that a network failure at startup no longer logs people out: they land here instead, which is the point where they need it. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Same shape as the archive bug, found by looking for its siblings rather than by tripping over them: - duplicateLocation and duplicateCollection announced a copy the list did not show until the screen was left and re-entered. Duplicating a collection was the visible one: refresh() refills the cached flow the header counts from but not the counter paging invalidates on, so the header said seven above six. - refreshCollections now bumps that counter, which is what accepting an invitation relies on too. - updateLocation left selectedLocation holding the pre-edit copy, and getLocation answers from it whenever the ids match - so opening a place you had just edited showed what it used to say. deleteLocation left the deleted one there. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
loadInitialData runs at construction, before anything has asked the server anything, and its catch deleted the stored token. So a field this build no longer understands - or a half-written value - logged the user out permanently, over a parse the next build might well manage. Now it leaves the token alone and says so in the log. The user starts at the login screen either way; this way there is still a session there to come back to. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Editing your name with no connection reverted the field and told you to try
again - which meant typing it again first. The revert is right when the server
refuses ("that username is taken"): that is a verdict on the value, and the
form should show what the server holds. Being unable to reach the server is not
a verdict on anything.
The account calls returned the server's own message and nothing else, so the
screen could not tell those apart. They now return an AccountError carrying that
same message plus whether the server actually answered, and only a refusal
reverts. The message stays the server's words wherever there are any - that part
of the design was deliberate and is unchanged.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Kotlin 2.2.21 → 2.4.10, Compose Multiplatform 1.8.2 → 1.12.0, AGP 8.13 → 9.3.2, Gradle 8.13 → 9.7.1, compileSdk 35 → 37. Measured a week ago as one atomic piece of work - every library bump failed on the same wall - and that held. Two things the notes did not predict: - AGP 9 refuses to apply com.android.application/library alongside the Kotlin Multiplatform plugin, and points at com.android.kotlin.multiplatform.library instead, which is library-only and so cannot replace the application plugin here. android.builtInKotlin=false and android.newDsl=false keep the previous behaviour; migrating the library modules to the new plugin is its own job. - Compose 1.12 wants compileSdk 37, not the 36 the notes said. targetSdk deliberately stays at 35: it changes runtime behaviour, and mixing that into a toolchain upgrade makes any breakage twice as hard to attribute. Android and iosSimulatorArm64 both compile. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
… screen First runtime casualty of the Compose 1.12 jump, and it took every screen with it: the app bar's content sat mid-screen and the content below it was pushed off entirely. The cause was one modifier - the title slot wrapped its content in Box(Modifier.fillMaxHeight()), which under the looser slot constraints of the new version grew the bar to fill half the display. The app bar centres its own title, so the fill was never needed. Compiling was not evidence of anything here; this only showed up on the device. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Coil 3.3.0 → 3.6.1, Koin 4.1.1 → 4.2.2, Ktor 3.3.2 → 3.5.2. All three failed a week ago against the old toolchain - Coil and Koin wanted a newer compileSdk, Ktor's klibs wanted a newer Kotlin - and all three go through untouched now that the wall is gone. No source changes were needed. Android and iOS compile; the app runs, loads its data through Ktor, resolves its graph through Koin and draws its photographs through Coil. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The five destinations move from a NavigationBar the shell drew itself to items handed to NavigationSuiteScaffold, which picks the shape the window deserves: a bottom bar on a phone, a rail down the left on a tablet. A bar pinned to the bottom edge of a 1600px display is a long way from where the hands are. Verified both ways on the resizable emulator - rail at 1600x2560, bar back at 1080x2400. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Places, Collections, World and Home's recent places were LazyColumns, which on a tablet means one photograph stretched across 1600px. They are adaptive grids now: a minimum card width, and the window works out how many fit - one on a phone, two on a tablet, three on something wider, without anything naming a device. The rows that would read worse cut in half keep the full line: Home's featured trip, its stats card and every section header, World's search, progress and filters, and both lists' paging spinners. Verified at 1600x2560 and back at 1080x2400, where nothing changed. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
…forms Adds navigation3-runtime (androidx, which does publish native klibs), navigation3-ui (JetBrains, 1.1.1) and material3 adaptive-navigation3 (1.3.0-beta02 - a different version line from the navigation-suite artifact, which tracks Compose and is 1.12.0-alpha03). The probe is not wired to anything. It exists because the two-pane migration should start from a shape known to compile for Android *and* iOS, and three things had to be found out the hard way first: - entry is a member of EntryProviderScope, so importing it fails to resolve. - NavDisplay takes sceneStrategies, a list, not a single sceneStrategy. - rememberNavBackStack's no-configuration overload is Android-only. Shared code must pass a SavedStateConfiguration or it builds on Android and breaks on iOS - which is exactly the kind of thing that would have surfaced late. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The first screen on Navigation 3. A NavDisplay with a ListDetailSceneStrategy decides from the window whether the place opens beside the list or on top of it: two panes on a tablet, the same full-screen page as before on a phone. Choosing between places means comparing them, and the old behaviour threw the list away to show you one. Nested rather than a rewrite: a NavDisplay is only a composable, so it sits inside the existing NavHost and the other tabs carry on unmigrated. That keeps this to one screen instead of every module at once. Two things worth writing down, both found by running it: - SavedStateConfiguration.DEFAULT cannot serialise a NavKey. The back stack is persisted by serialising its keys and NavKey is open, so the subclasses have to be declared in a polymorphic module - it compiles either way and throws the moment the screen composes. - The detail pane hides its back arrow when both panes are visible. There is nothing to go back to; the list never left. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The second Navigation 3 screen, the same shape as Places: the list on the left, the collection's contents on the right, and the old full-screen page on a phone. A trip is chosen by looking between them, which is hard when opening one hides the rest. Both list screens now register their own route and pass registerListRoute=false to the feature's graph, which still owns the add and edit destinations. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
360dp per column meant a foldable (674-841dp) and the list side of a two-pane screen both fell one pixel short of a second column and showed a single card in the middle of an empty page. 300dp fits two. The dashboard was worse, because its rows spanned every column: a progress bar drawn 780dp long stops reading as a proportion. The four stats now pair off into two columns on a foldable and four on a tablet, matching the web, and the trip and event rows flow into the grid instead of each owning a full line. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Half a tablet holding the sentence "Pick a place to see it here." is a worse answer than showing a place. Both two-pane screens now open their first item once the list loads. Phones have no second pane to fill, so nothing opens by itself there. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
A search field, a region dropdown and a hero card, each drawn the full width of a landscape tablet. Capped at 720dp they are the size of the thing they are for again. The cap has to go through a Box. A spanned lazy-grid item is measured at an exact width, and widthIn coerces its own maximum back up into that fixed range - so the modifier applied directly to the content compiles, runs, and changes nothing. The Box takes the exact width and hands its child loose constraints. Also names the collection in the detail pane. The name used to come from the shell's breadcrumb, which reads the outer route; the two-pane screen never sets one, so the pane opened on a description with nothing to attach it to. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Capping elements one at a time was the wrong fix: it left a screen holding a full-bleed progress card, a 720dp search field and a three-column grid, each finding its own edge. Three measurements on one page. There is now a single content column - 1040dp, centred - and everything lives inside it: the app-bar title, the search field, the filters, the stats, the cards. They share a left and a right edge at every window size, and the space left over on a tablet becomes margin instead of a fourth width. The app bar is inset to the same column. It is transparent, so this only moves the title and the actions; without it the greeting starts at the window edge while the page under it starts 120dp further in. Two filter rows also carried their own 16dp padding on top of the grid's, which put them on an edge of their own. Below 1040dp none of this does anything, so phones and foldables are unchanged. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
…olour Two changes that turned out to be the same finding. Material You was on by default, so on Android 12+ the app took its palette from the phone's wallpaper. Every screenshot in this branch was lavender because the emulator's wallpaper is lavender; #006973 - the colour in Color.kt, the colour of the launcher icon - never appeared on a modern device at all. That is what "it feels generic" was: the app had no colour of its own. Now it does, and the switch stays in Settings for anyone who prefers the wallpaper. The redesign needed no new palette. Its twenty-two colours are the light scheme already in Color.kt, tone for tone. Past 1000dp the dashboard becomes two columns - the trip and the photographs on the left, a 348dp rail of numbers and dates on the right - which is the redesign's own `minmax(0,1fr) 348px`, and what the empty third of a tablet was for. Below that everything stacks, unchanged. The hero carries the trip's photograph: beside the text when there is room for both, above it when there is not. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
…screenshots The first attempt was transcribed by eye from screenshots of the mock, and it showed: a grey hero instead of the cyan one, icons the design does not have, two separate lists it had deliberately merged. The mock's own HTML was readable the whole time. It is now in design/, and this is built against it. - The hero is the primary container. It used the secondary one unless the trip had already started, which is where the washed-out grey-green came from. It carries the dates, the two actions, and the photograph beside the text. - The stats lose their icons: four of them down the left edge were a column of decoration pushing the numbers into the middle of the card. Counts are monospaced so four rows line up as a column of figures. - "Upcoming trips" and "Coming up" become one dated list. A trip starting on the 5th and an event on the 5th are the same kind of fact, and two lists made the reader merge them by eye. - Recent places get their own small card - photograph on top, facts underneath on white - because a name laid over a bright sky disappears at that size. Two layout traps, both the same shape: an Image inside height(IntrinsicSize.Min) reports its natural height and grew the card to nine hundred dp, and a Card with enabled=false is not "not clickable", it is greyed out. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
…, round actions The bar was a MediumTopAppBar: a collapsing two-row thing with the greeting on its own line under the icons. The redesign has one row - name, one line saying where you stand, and the search and the avatar as two circles beside it. Home is the only screen with the subtitle for now. The others print their own count inside the page, and "22 places" twice, one line above the other, reads as a mistake rather than as emphasis. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The map's error state had no control at all and nothing reloaded it: with no network it showed "Error loading map data" and stayed there for the rest of the session — leaving the tab and coming back did not help, only restarting the app did (`refresh()` had no caller). It now offers Try again, and coming back to a map that is in that state retries by itself. Test: MapViewModelTest.aMapThatFailedCanBeTriedAgain, seen red on two breaks. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
On a wide window Places and Collections are two panes. Deleting the selected one left its detail open on the right, Edit and Share and all, until something else was chosen — the server had already returned 404. The delete result now carries the id of what left, and each pane drops any detail key pointing at it. Covered by PaneBackStackTest (3 tests: the shown one goes, another stays, and the same on the collections side), each seen red. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
… sides Found with three real accounts - one owning, one invited, one granted nothing - and fixed together because they are one experience: - The share sheet's messages were drawn by the screen behind it. A modal sheet is its own window, so nothing it raised was ever seen: not the server's "Invite already sent to this user", not an offline failure. The sheet now says what happened itself. - A guest was offered the owner's whole menu, and the server refused every owner-only entry. Collections now carry their owner, and the collection and place sheets offer a guest only what the server allows (open, the share card, the PDF; edit and duplicate a place). - Every repository answered 401 and 403 alike, so a refusal of one record read "Session expired. Please log in again." A 403 is now its own case with its own words. - An invitation that arrived while the app was open could not be reached: invites were read once, and the empty list could not even be pulled, since a plain Box passes no drag to PullToRefreshBox. A refresh now asks again, and the empty states are pullable. - Home shows pending invitations and opens Collections on Invites, and a person in People opens their profile numbers. Tests: CollectionsSharingViewModelTest (5), RefusalIsNotASessionTest (2), OwnershipTest (3), UsersViewModelTest (+2), GuestActionsTest (4), DashboardScreenTest (+2), SharedComponentsTest (+1) - each seen red. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
…screens looked like The queued P2s: - A checklist saved unchanged is no longer sent: the server recreates every item on any save, ignoring the ids it is given (CO-04). - The transport date and time buttons open their pickers; dates are stored as the web stores them - local time in the chosen zone, as UTC (CO-06). - Deleting a note, checklist or stay asks first (CO-07). - A collection save that fails keeps the form and what was typed (CO-08). - Duplicate and Share externally work on a place inside a collection, and it can be removed from just that collection (CO-09, CO-10). - Tapping the words of a Public row toggles it (CO-11). Found while verifying, by looking at the screenshots: - A link typed while creating a collection was dropped (CO-12). - Transport dates are stacked and labelled - the arrival was pushed out of its card - an arrival before the departure is refused, and "Constrain to Collection Dates" limits the pickers instead of doing nothing. - Note, checklist and stay dates use a calendar field with a label that stays. - The place sheet opens fully, so Delete is not cut off; "Rating" appears once; multi-line fields have no icon floating halfway down; the empty Places tab no longer talks about adventures. Tests seen red: AddEditChecklistViewModelTest (+2), TransportDatesTest (5), AddEditCollectionViewModelTest (3), RemoveLocationFromCollectionUseCaseTest, TransportDatePickersTest (5), CollectionTabsTest (+2), PublicRowsTest (2), FormFieldsTest (3). Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
…ed like - A pull to refresh shows that it is working: the screen set its own flag on and off in the same call, before the request started (WO-01). - A region ticked on a country's page reaches the world list: the cached country's count follows the tick and the list follows the cache (WO-02). - Region rows are checkboxes with their state, not buttons with an icon described "Visited" (WO-03). The filter chips already exposed theirs - the campaign had read the wrong attribute. - "Can't reach the server. Check your connection." on the World paths, as everywhere else (WO-04). Found by looking at the screenshots: - The country page had no back button: onNavigateBack arrived and nothing drew it. - The tick on the selected "All" chip was dark teal on teal. - Mostly white flags (Japan) had no edge against the white card. - "3 / 47" sat on two baselines. Tests seen red: RegionTickCountsTest (2), WorldViewModelTest (3), CountryDetailScreenTest (2). SettingsViewModelTest pinned the old wording and now asserts the new one. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
- Chosen filters survive process death (SavedStateHandle). - The stats card sits above the map instead of covering its top edge. - City and region markers say what they are to a screen reader. - Counts show a dash while unknown (loading, failed, regions not yet in) instead of zeros; the region count is the visited regions themselves. - Filter sheet: one drag handle, layer rows are switches, opaque card. - 403 read as an ended session in seven places the 09 round missed. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
… looked like - Rows open what they name: a visit its place, a trip or anything dated inside one the trip; the event's resource id is now read. - Today is marked where it falls, and the list opens on it. - "Show everything before" loads the rest of the journal on request, says when there is nothing, and can be retried offline. - Filter chips say which one is on (MetaChip selected), here and in the Collections filter; Completed gets its own emoji so the tick is not doubled. - The date column no longer hangs below one-line cards. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
…over The photo's slot stays and is filled on purpose: a tonal surface with the category's emoji (a line pin without one), from placeholderColors() so dark theme follows. Replaces the gradient and the Static Maps thumbnail that never loaded, so feature/ui no longer needs a Maps key. - Places card: ink on a tonal band instead of white on a scrim, and dashed "+ Add photo" / "+ Category" chips that open the editor. - Home strip: every line always drawn, so cards side by side keep one height. - Detail page: a compact bar (back, edit, share) and a "No photo yet" card instead of a 300dp stand-in cover. The name stays in the page heading only. - "Add photo" (card and detail) opens the editor on its images section, expanded and scrolled to, via createEditRoute(openImages = true). - Detail page provides LocalContentColor (titles were black on dark theme when opened from Home); the shell consumes the bar insets it pads for. Verified on the Pixel 6a, light and dark, with a test place created and deleted. Not reachable on this server: a place with no category (the API requires one) and a photo-less place in the Home strip (the dashboard endpoint leaves it out). Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The convention plugin added consumer-proguard-rules.pro to every library module and only core/network has one. AGP 8 ignored the missing file, AGP 9 fails :core:model:mergeReleaseConsumerProguardFiles, so assembleRelease could not run at all. Add the file only where it exists. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…er one Every keystroke started its own geocoding request and whichever answered last won: "Valencia" listed places called "Valen" in Norway, and "Madrid Atocha" said "No results found" because "Madrid A" took five seconds. Both the place and the transport forms now cancel the previous search and wait 300 ms before sending (QA RL-05). Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The editor's checkboxes had no label, so a screen reader said "checkbox, not checked" without saying which item. The whole row is now the toggle, read with the item's text (and tappable on the text too), and Remove says which item it removes (QA RL-06). Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The header read "0 collections · Invites" above two pending invitations. On that tab it now says "2 invitations" (QA RL-08). Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Cancel (and Back) left the checklist editor at once, losing the items typed so far. Like the place form, it now asks "Discard this checklist?" (or "Discard changes?" when editing) whenever something would be lost, including an item typed but not added yet (QA RL-09). Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The note, lodging, transport and collection forms left on Cancel or Back at once, losing whatever had been typed. They now ask the same question as the place and checklist forms, through one shared rememberDiscardGuard in feature/ui; an untouched form still leaves straight away (QA RL-09). Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
versionCode 8 is the bundle already uploaded to the Play internal-testing draft, and Play refuses an upload that reuses a code (QA #21). Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Release runs tasks debug never does (R8, consumer ProGuard rules). On AGP 9 assembleRelease stopped working and only the main-branch release job would have noticed. The build job now also runs an unsigned assembleRelease (QA RL-03). Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
People was pushed with a plain navigate, so it landed inside whichever section was open. Each section's stack is saved and restored, so after People had been opened from Calendar, Calendar kept reopening People until the app restarted. People now opens as a section of its own, like Calendar and Settings (QA RL-10). Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Three of Home's open product decisions (QA HM-09), as the owner chose: - the featured trip's "Add a place" opens the form for that trip, and the new place is created inside it (the form took no trip before); - Coming up rows for visits and stays open what the calendar opens for them, the place or the trip, instead of nothing; - "Recently updated" no longer shows the library total over a handful of cards. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The owner's choices for the look-and-feel items left from QA 04:
- a filled field keeps its label ("Reservation", "Price") instead of
losing it with the placeholder; this is the shared StyledTextField;
- the stay, note and checklist forms put their card at the width of
everything else, and Save/Cancel as on the collection form;
- the stay's price shows its currency, not a "$" over euros;
- the transport form has a title, and its card says when and where;
- Places inside a collection gets a "+" that adds a place to it, in
place of a count bubble the chip already showed.
The view chips cut at the edge stay as they are: the cut says it scrolls.
SectionCard also applies the modifier it was always given.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The decisions behind the CO-19 round: labels that stay, cards at the form's width, Save across and Cancel under, a title on every item form, a question before discarding, currencies by code, and a "+" in every section that lists something. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
NEARBY_WIFI_DEVICES had no code behind it. READ_PHONE_STATE and WRITE_EXTERNAL_STORAGE were never declared: the merger implied them for the Compose Hot Reload runtime, which reaches the release build without a targetSdkVersion. Play shows all three on the store page and in Data safety. The release APK now lists INTERNET, CAMERA, READ_EXTERNAL_STORAGE (up to Android 12) and ACCESS_NETWORK_STATE. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The scope decision from the QA pass (queue #5): MFA/SSO sign-in, Spanish, leaving a collection and attachment upload come next; the rest of what the web client has is out of scope for now. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The terms were another app's ("Shiori", under Apache 2.0) and the privacy
policy still read "Effective as of [Insert Date Here]" while claiming the
app reached no third party. Both are rewritten, approved by the owner,
from what the code does on this date: an independent MIT-licensed client,
no data collected, the session token (not the password) kept on the
phone, and Google Maps, Wikipedia and the user's own server as the only
places anything goes (QA #4).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The app said the same thing four ways - "No internet connection.", "Network unavailable", "Network error", "Can't reach the server..." - and three of them blamed the phone's network, usually fine when the server is asleep or the VPN is off. Every network failure now reads "Can't reach the server. Check your connection." (CANT_REACH_SERVER in core/domain); sign-in keeps its own, which also names the address. Offline, the collections list printed the raw "Unable to resolve host ..." with the server's address; it now says the same as every screen. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
android-actions/setup-android@v3 asks for "tools platform-tools" by default, and the SDK repository no longer has "tools", so the iOS job failed at its first step on PR #8. Gradle fetches the rest itself. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Design 3a from the Claude Design file "App Icon.dc.html", chosen by the owner: the AdventureLog project's folded map, redrawn flat - a white map in three panels with light-cyan continents on the app's teal. Every asset comes from one definition in scripts/icon/generate_icon.py: the Android adaptive layers (with a monochrome layer for themed icons on Android 13+), the legacy PNGs for Android 7, the 1024 px iOS icon and the 512 px Play icon. On Android the map is drawn at 0.66 rather than the design's 0.72, which put two corners past the safe circle where a round launcher clipped them. The iOS and Play PNGs no longer carry an alpha channel, which the App Store refuses. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
AGP 9.2, Navigation 3, the tests that exist now, signing that is wired but waits on its secrets, the Releases page's newest build, and the iOS notes as they stand (the discard prompt works on Cancel; the date-picker fix is in but untested on a device). The banner points at the scope in "What's next" instead of repeating an older list. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Everything since
develop(ff6f8bf, the build in the Play internal-testing draft): the adaptive/tablet work and the toolchain jump that came with it, then a month-long QA pass over the whole app with every fix verified on a device against a real server.Toolchain and layout (the
feat/adaptive-tabletwork, never pushed as its own branch)QA campaign fixes — 3 P0, 16 P1, ~70 P2, each with a test seen failing first where a test could catch it
assembleReleasedid not build at all on AGP 9.docs/design.md→ "One shape for every form").Release
scripts/icon/generate_icon.py; iOS/Play PNGs without alpha.NEARBY_WIFI_DEVICESwas unused, and READ_PHONE_STATE / WRITE_EXTERNAL_STORAGE were implied by the Compose Hot Reload runtime.assembleReleaseon every PR.Tests run
./gradlew allTests testDebugUnitTest connectedDebugAndroidTeston an API 37 emulator — green on the last full run (476 tests ran, 64 instrumented; the rest up to date)b4a73cd: create/edit/delete of every item kind, photos, camera, attachments, share, PDF, invitations, settingsNot in this PR: screenshots — the QA reports and screen captures live in a local, unversioned folder.
🤖 Generated with Claude Code