vehicles: /vehicles list+sidebar details (cumulative, supersedes #67)#74
Merged
Conversation
Single source of truth for the chip filter set used on vehicle-domain list views. Adds the NeTEx TransportModeEnumeration union plus the matching FilterDefinition[] keyed under transportMode.* i18n keys. Consumed by the new /vehicle list (#26) from day one; #24 retrofits VehicleType and DeckPlan ViewConfigs separately. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Conventional Inanna feature directory replacing the packages/vehicle-details PoC approach. Mirrors src/data/vehicle-types/. - vehicleTypes.ts: flat Vehicle row + VehicleColumnKey - fetchVehicles.ts: thin flatten over ../vehicle-types/fetchVehicleTypes (no duplicated GraphQL query; enriches each Vehicle with parent VehicleType context for display + client-side TransportMode chip filter) - useVehicles.ts: data hook mirroring useVehicleTypes (single fetch, client-side sort/pagination) - vehicleSortValue.ts: comparator via compareWithEmptyLast - vehicleViewConfig.tsx: ViewConfig consuming shared transportModeFilters - VehicleView.tsx: thin GenericDataViewPage wrapper - VehicleDetails.tsx: sidebar editor mounted via EditingContext; read-only by default, Edit chip toggles to form view, no Save button (gated on layout approval per #26 iter 2) - cells/RowClickCell.tsx: row trigger calling setEditingItem(...) packages/vehicle-details/ is intentionally left untouched as the prior PoC. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
ProtectedRoute mounting VehicleView, placed beside /vehicle-type and /deck-plan inside the existing EditingProvider so the sidebar pattern works out of the box. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Both en and nb gain: - transportMode.* (13 NeTEx-aligned modes; rail/metro/tram/bus/coach/ trolleyBus/water/air/taxi/cableway/funicular/lift/snowAndIce) - vehicles.* (title, detailsTitle, edit/view chip labels, notFound, field.* labels) transportMode.* is shared infrastructure for #24's chip-filter retrofit across VehicleType/DeckPlan as well. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
- isTransportMode: runtime guard for narrowing unknown strings (e.g. GraphQL responses, URL params) to the TransportMode union. Keeps unknown backend values out of the typed surface and surfaces drift via undefined rather than crashing or silently mis-rendering. - transportModeLabelKey: maps a TransportMode to its transportMode.* i18n key for use in column cells and detail labels. Pure additions; existing callers unchanged. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Multi-agent review surfaced 5 must-fix items + a missing test file.
- Rename Vehicle → VehicleRow to avoid collision with vehicle-types'
existing Vehicle type (the wire shape nested under VehicleType). The
new VehicleRow is a flat view model with parent* enrichment fields.
- Drop the unchecked `as TransportMode` cast in fetchVehicles; use the
new isTransportMode guard so unknown backend values become undefined
rather than silently polluting the typed surface.
- Pass the full row via closure from RowClickCell to VehicleDetails.
VehicleDetails no longer calls useVehicles() — removes a full-dataset
re-fetch on every row click (high-severity efficiency finding).
- Add TransportModeChip cell that renders the localized label
(t('transportMode.*')) rather than the raw NeTEx enum string.
Same chip is used in the column cell and in the details fieldValue.
- Drop the redundant Cancel button in VehicleDetails — the Chip
already toggles edit→view; Cancel was a second affordance for the
same direction. Close remains for dismissing the panel entirely.
- Drop a future-rot "iter 1" comment; the JSDoc on the component
already states the no-save contract.
- Add vehicleSortValue.test.ts (8 cases) mirroring the convention
established by vehicleTypeSortValue.test.ts and deckPlanSortValue.test.ts.
Deferred (consistent with existing template; bigger blast radius):
shared GraphQL error helper across the 3 list hooks, useMemo on
sort/pagination, AbortController in doFetch, i18n for column headers
and page title.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
`undefined` was a leaky state — every caller had to handle the missing case. Adding `'unknown'` as a first-class member of the TransportMode union keeps the typed surface total and encapsulates the mapping inside `netex/transportMode.ts`: - TransportMode union gains `'unknown'`. - New `toTransportMode(s)`: total mapping. Unknown / missing / null / unrecognised backend strings all collapse to `'unknown'`. - `isTransportMode` now also accepts `'unknown'` (it is a valid mode). - New `transportModeSortValue(m)`: returns `''` for `'unknown'` so rows still sink via `compareWithEmptyLast`; known modes sort by id. - `transportModeFilters` intentionally omits `'unknown'` — chip filters are for specific known modes; rows with `'unknown'` simply don't match any chip. Downstream: - VehicleRow.parentTransportMode is now non-optional `TransportMode`. - fetchVehicles uses toTransportMode at the boundary; the conditional is gone. - TransportModeChip drops the em-dash fallback; renders the localized label (en: "Unknown", nb: "Ukjent"). - VehicleDetails fieldValue handles parentTransportMode explicitly; em-dash branch only fires for other genuinely-missing fields. - vehicleSortValue uses transportModeSortValue. - i18n: added `transportMode.unknown` (en + nb). - Tests: new transportMode.test.ts (8 cases); vehicleSortValue.test.ts updated to use `'unknown'` instead of `undefined`. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Wires a query-param URL contract for the Vehicle list — chips on /vehicle-type rows and the per-row icon on /vehicle both navigate to /vehicle?selected=<id>, which the new useVehicleUrlSelection hook reconciles with EditingContext, sidebar visibility, and pagination (auto-jumps to the page containing the selected row). GenericDataViewPage gains an optional useUrlEffect ViewConfig field (stable no-op fallback preserves hook order on pages that don't opt in) and a symmetric sidebar collapse so the editor's Close button also drops ?selected= from the URL and collapses the drawer. Archives the design rationale in a new FORK_DECISIONS.md — ADR-style log for resolved fork-stage decisions; companion to OPEN_QUESTIONS.md. Backfill of earlier decisions tracked in #66. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Self-review pass on PR #67. Iter 1 ships only data-independent no-auth tests (`.app-content` visible); /vehicle's deep-link slider, chip filter, pagination jump, and edit-chip behaviour are only verifiable end-to-end. None of that is reachable without a stub for the `vehicleTypes` GraphQL query. `autosys-helpers.ts` already wires this pattern for the /vehicle-type tests but its fixture has 3 vehicles with `transportMode: null` — too lean for pagination (need 10+ rows) and chip-filter (need known + unknown modes). Add a parallel fixture rather than mutate the existing one, since /vehicle-type specs depend on the leaner shape. - `e2e-tests/fixtures/vehicles-list-mock.json` — 3 VehicleTypes with TransportMode variety (`rail`, `bus`, `someUnknownNetexValue` to exercise the `toTransportMode → 'unknown'` boundary), 15 total vehicles (12 + 2 + 1), deterministic ids (`NMR:Vehicle:rail-11` lands on page 2 at `rowsPerPage=10`). - `e2e-tests/no-auth/vehicle-list-helpers.ts` — exports `interceptVehicleListQuery(page)` mirroring the inline pattern in `autosys-helpers.ts` (match `postData.query` substring, fulfill matching requests, continue() everything else). No spec changes yet — commit 2 wires this into a new vehicle.spec.ts. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Iter 1 self-review on PR #67 — pin the integration contracts that live behind `useUrlEffect` so future indirection refactors and iter-2 Save wiring have a regression net. Six cases in `e2e-tests/no-auth/vehicle.spec.ts`, exercising the no-auth path with the GraphQL stub from the previous commit: 1. list renders 15 rows, default sort = registrationNumber asc, page 0 contains 2 bus + 8 rail chips (localised via `transportMode.*`). 2. clicking the search filter button + 'Rail' checkbox narrows the total to 12 and hides BUS rows. 3. row action navigates to `/vehicle?selected=NMR:Vehicle:bus-1` and opens the sidebar. 4. **Edit chip toggles to TextFields, TransportMode stays localised.** Asserts `getByLabel('Transport Mode').toHaveValue('Bus')` — **this case fails today** (received `'bus'`, the raw NeTEx enum, instead of the localised `'Bus'` shown in view mode). The next commit routes view + edit modes through one display helper to turn this green. 5. clicking Close drops `?selected=` from the URL and collapses the sidebar. 6. deep-linking `/vehicle?selected=NMR:Vehicle:rail-9` auto-paginates to page 2 (`pagination-displayed-rows` contains '11') and opens the sidebar — exercises the `useVehicleUrlSelection` page-jump path and the load-guard that prevents committing `vehicle=null` while `loading: true`. Tests are independent (each `goto`s fresh) — dropped the `mode: 'serial'` configure so a failure in any case doesn't mask the others. The no-auth project still uses `workers: 1` for shared dev-server safety, so ordering is preserved. Also adds `data-testid="search-filter-button"` on `SearchFilterControl`'s IconButton so the chip-filter popover is addressable without depending on MUI internals. Pure addition; no behaviour change. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Edit mode wrote `vehicle[key]` straight into the TextField, so for
`parentTransportMode` the field showed the raw NeTEx enum (`'bus'`,
`'rail'`, `'unknown'`) instead of the localised label that view mode
rendered via `transportModeLabelKey + t()`. Toggling the Edit chip
flipped 'Bus' → 'bus' — visible inconsistency.
Rename `fieldValue` → `fieldDisplay` and parametrise the missing-value
fallback so both branches share one display path:
- view mode: `fieldDisplay(key, '—')` — em-dash for empties (unchanged).
- edit mode: `fieldDisplay(key, '')` — empty input for empties.
TransportMode mapping lives in the helper, so the two modes now agree
by construction. Turns the failing case in `vehicle.spec.ts` green
(`getByLabel('Transport Mode').toHaveValue('Bus')`).
No save-path involvement — fields stay `readOnly` per iter 1.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Iter 1 self-review on PR #67. `fetchVehicles` is the only place that delegates to `fetchVehicleTypes`, flattens `vehicleType.vehicles[]`, and routes `vt.transportMode` through `toTransportMode()` — each load-bearing for the /vehicle list, the TransportMode chip filter (#24), and the no-duplicated-GraphQL claim in the PR description. Yet the function had zero direct coverage. Add `fetchVehicles.test.ts` — pure vitest, mocks `../vehicle-types/fetchVehicleTypes.ts` via `vi.mock`. Mirrors the template of `vehicleSortValue.test.ts` (table-driven via a small `mkVT` factory). Four cases: 1. empty `vehicleTypes` → `[]`. 2. VehicleType with `vehicles: undefined` → zero rows (pins the `?? []` fallback). 3. parent context carries through — `id`, `name.value`, `transportMode` land on every flattened row. 4. unrecognised parent `transportMode` collapses to `'unknown'` via `toTransportMode` (pins the boundary mapping at the call site — complements `transportMode.test.ts` which pins it at the helper). Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Iter 1 self-review on PR #67. The null/undefined/empty boundary cases the v1 plan called out for `toTransportMode` were already covered (lines 16-20 of the existing test). The genuine gap is `transportModeFilters` — exported but completely untested, despite the JSDoc stating a deliberate UX contract that `'unknown'` is omitted from the chip filter set (rows with `'unknown'` simply don't match any chip). Add a `transportModeFilters` describe block with two cases: 1. `'unknown'` is not in the filter set (pins the JSDoc'd omission — prevents accidental re-inclusion if someone treats the union and the filter list as the same data). 2. Every known `TransportMode` has a filter entry, and the list length matches — regression net for new NeTEx modes added to the union but forgotten in the filter array. The known-modes array here is the source of truth; if `TransportMode` grows it must be updated, which is exactly the prompt we want a reviewer to see. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
…onDelivery (#69) Type-blind, slot-based envelope builder in src/data/netex/. Caller passes each slot's items pre-shaped via the matching *ToXmlShape(...) function; builder composes the PublicationDelivery → ResourceFrame envelope as one object tree and emits via a single XMLBuilder.build pass. Migrates VehicleTypeDetails save path to the new helper and deletes the legacy string-template wrapInPublicationDelivery. Envelope defaults (namespace, version, RF:1, Europe/Oslo) are carried forward verbatim so the VehicleType save output stays compatible. Vitest covers slot composition, sibling emission, omitted-empty-slot behaviour, and envelope defaults. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
#69) Produced with `netex-ts-gen --schema base.schema.json --dest-dir src/data/vehicles --collapse-refs --collapse-collections Vehicle VehicleModel`. Post-generation hand-patches required to satisfy strict tsc + lint: - rename unused `reshapeComplex` Reshape params in leaf *ToXmlShape helpers to `_reshapeComplex` (TS6133 under noUnusedParameters); - delete unused `childWrapped` helpers from both *-mapping.ts; - eslint override for `src/data/vehicles/Vehicle*.ts` patterns: turn off `@typescript-eslint/no-unused-vars` and `no-explicit-any` since both fire on machine-generated emission shapes. Also: standardise `argsIgnorePattern: '^_'` for the unused-vars rule globally — common convention for intentionally-unused names. Upstream fix in netex-typescript-model is the proper resolution; the hand-patch will need re-applying on every regen until then. The Vehicle interface exposes `VehicleModelRef?: Ref<'VehicleModel'>`, the 1-to-1 link the new form populates with `INLINE:VehicleModel:1`. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
New shared VehicleEditForm component (`src/data/vehicles/`) takes
`{ value, onChange, mode }` and renders Vehicle fields + the
VehicleModel-unique fields (Manufacturer / Range / FullCharge) in one
column. Mode toggles between view (disabled inputs) and edit. Form
holds typed `{ vehicle: Vehicle, model: VehicleModel }` state per
issue #69. Overlap fields (Name, Description, TransportTypeRef,
BrandingRef, ValidBetween) live on Vehicle only; VehicleModel emits
only its unique fields on save.
New VehicleCreatePage hosts the form via GenericDetailsPage chrome at
`/vehicle/new` with `model.$id = 'INLINE:VehicleModel:1'` and
`vehicle.VehicleModelRef = 'INLINE:VehicleModel:1'` preset; server
collapses the inline model on import.
New NewVehicleFab mounts via `GenericDataViewPage`'s `floatingAction`
slot on `/vehicle` and navigates to `/vehicle/new`. Same pattern as
AutosysImportFloatingMenu on `/vehicle-type`.
Save handler is stubbed (throws) in this commit. Commit 5 wires
buildPublicationDeliveryXml + importAsNetexToBackend, after-save
navigation, toast on error, and useBlocker nav guard for dirty form.
NOTE comment in VehicleCreatePage points at issue #68 (pending
`modification="new|revise"` client policy decision).
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Sidebar swaps its inline 5-field display for the shared VehicleEditForm introduced in 9808ea0, hydrating `{ vehicle, model }` from the partial VehicleRow. Only 4 fields persist from the list row today ($id, $version, RegistrationNumber, TransportTypeRef = parent VehicleType ref); the rest start blank until the deferred /netex/vehicles/{id} read lands. VehicleRow-only enrichment (id, version, parent VehicleType name, localised transport mode) moves into a read-only context panel above the form so the underlying Vehicle entity (which doesn't carry those fields) doesn't have to. VehicleModel side always starts at `{ $id: 'INLINE:VehicleModel:1' }` per issue #69 — server collapses the inlined model on import. (Edit-vs-create semantics are uniform under this rule; an existing Vehicle's prior VehicleModelRef is overwritten with the INLINE placeholder, and server handles identity.) E2E: the existing 'Edit chip toggles to TextFields' test was tracking a regression in the now-removed editable Transport Mode field. Replaced with a positive test that pins the new contract: Transport Mode renders localised in the read-only context panel regardless of mode. Save is still stubbed; commit 5 adds the Save button and wires the buildPublicationDeliveryXml + importAsNetexToBackend pipeline. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
The desktop Sidebar wraps its content in a fixed-height Box with `overflow: 'hidden'`. With the old 5-field inline display this was fine; after commit 5fcfac2 swapped in the full VehicleEditForm (~12 inputs + context panel + chrome), the bottom fields fell off the viewport with no way to reach them. Constrain the VehicleDetails outer Box to `height: 100%` and let it scroll vertically (`overflowY: 'auto'`, `boxSizing: 'border-box'`). The fix is local to VehicleDetails so it doesn't change the behaviour of the other consumer of Sidebar (the nav drawer fallback in SidebarContent → WorkAreaContent). Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
End-to-end save for both Vehicle + VehicleModel flows:
- `buildVehiclePairXml(form)` composes the PublicationDelivery via
`buildPublicationDeliveryXml`, embedding `vehicleToXmlShape(form.vehicle)`
and `vehicleModelToXmlShape(form.model)` into their slots. One
XMLBuilder.build over the full tree.
- `parseVehicleImportResponse(body)` extracts `Vehicle/@id` from a Sobek
`/import` response in either CompositeFrame-wrapped or flat-ResourceFrame
layouts. Returns null on empty/non-XML/no-Vehicle bodies. Six vitests.
- `useVehiclePairSave()` wraps build → POST → parse: takes the form, returns
`{ newId: string | null }` on success, sets `error: string | null` on
failure. Surfaces `saving` for button-disabled state.
- `useDirtyFormBlock(isDirty)` is a thin wrapper around
`react-router-dom`'s `useBlocker` with a `window.confirm` step; blocks
cross-pathname navigation only (intra-route `?selected=` toggles are
allowed without a prompt).
Wiring:
- `VehicleCreatePage` calls `save(form)` then `navigate('/vehicle?selected=
<newId>')` (D2a, issue #69). When the response doesn't echo a Vehicle id
the fallback navigates to `/vehicle` (plain list) — the proper D3
refetch+find-by-RegistrationNumber path will land once a refetch handle
is available to the create page. MUI Snackbar shows save errors;
`useDirtyFormBlock` guards nav when the form differs from blank.
- `VehicleDetails` (sidebar) adds an edit-mode-only Save button; success
closes the sidebar (drops `?selected=`). Save is disabled when the form
is pristine (dirty check via `JSON.stringify` against the latest
hydration from `VehicleRow`). MUI Snackbar shows errors; dirty-form
blocker covers cross-pathname nav. List refetch on edit success is
deferred — `useVehicles().refetch` is not yet plumbed through
`EditingContext` to the sidebar.
NOTE comment on each save handler points at issue #68 — the
`modification="new|revise"` client policy is unresolved; handler does not
set the attribute today.
TODO comment on create-flow navigation flags the future `?search=` filter
interaction: a deep-linked new row may be hidden behind an active filter
once filter UX lands (issue #24).
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
…outer (#69) `react-router-dom`'s `useBlocker` invariant-throws when not under a data router. Hathor still mounts the legacy `<BrowserRouter>` component, so calling `useBlocker` in either the create page or the sidebar crashed the render — surfacing as 8 failed sidebar e2e tests (chromium + firefox × 4 specs) where the "Vehicle Details" heading was never rendered. Rewrite the hook to use a `beforeunload` listener only when dirty. Modern browsers ignore the custom message and show their own native dialog; we just need to call preventDefault + set returnValue. Limitation acknowledged in the JSDoc: in-app React Router navigation is no longer guarded — only page unloads (tab close, refresh, URL bar). A proper in-app guard requires migrating from `<BrowserRouter>` to `createBrowserRouter` so `useBlocker` becomes usable. That's a router migration, out of scope here. Verified: e2e no-auth suite 36 passed (was 28 + 8 failed); 113 vitests green; tsc -b clean. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
This commit's diff is the /simplify cleanup pass (+93 / -243); the message below documents the full phase-2 branch since this is the head commit and is the most visible entry point when reviewing the PR. ## Goal Single combined Vehicle + VehicleModel details form for Sobek's national vehicle registry. Save POSTs both entities to /import as a NeTEx ResourceFrame; the server collapses the 1-to-1 relationship. ## Branch commit ladder 8 commits, each independently reviewable: 1. 2ce3e37 netex: introduce buildPublicationDeliveryXml, replace wrapInPublicationDelivery Type-blind, slot-based envelope builder at src/data/netex/. Callers pass slot values pre-shaped via the matching *ToXmlShape function. One XMLBuilder.build per emission. Replaces the legacy string-template wrapInPublicationDelivery. VehicleType save migrated. 7 vitests cover slot composition, sibling emission, omitted-empty-slot behaviour, and envelope defaults. 2. 3e3e079 vehicles: generate Vehicle + VehicleModel NeTEx types via netex-ts-gen Produced via: netex-ts-gen --schema base.schema.json --dest-dir src/data/vehicles --collapse-refs --collapse-collections Vehicle VehicleModel Post-gen hand-patches needed: _ prefix on unused Reshape params (TS6133 under tsconfig.app.json's noUnusedParameters); delete unused childWrapped helpers; eslint override on src/data/vehicles/Vehicle*.ts for the generator's intrinsic 'Extensions?: any' emission. Upstream fix in entur/netex-typescript-model tracked separately. 3. 9808ea0 vehicles: form shell + /vehicle/new create route New shared VehicleEditForm (12 TextField rows; view/edit mode), VehicleCreatePage hosted on /vehicle/new, NewVehicleFab on /vehicle. Save stubbed. 4. 5fcfac2 vehicles: sidebar VehicleDetails adopts shared VehicleEditForm Sidebar swaps its inline 5-field display for VehicleEditForm. Hydrates from VehicleRow (4 fields persist; rest start blank). Read-only context panel above the form for VehicleRow-only enrichment (id, version, parent VehicleType name, localised TransportMode). E2E: 1 regression test rewritten (TransportMode is no longer editable; test now pins the read-only context contract). 5. 7a910ed vehicles: sidebar scrolls when VehicleDetails form overflows Outer Box constrained to height: 100% + overflowY: auto. The desktop Sidebar shell hides overflow; the old 5-field form fit, the 12-field form doesn't. 6. a189a99 vehicles: wire save → /import for create + sidebar edit flows End-to-end save: buildVehiclePairXml(form) → buildPublicationDeliveryXml({ vehicles: [vehicleToXmlShape(form.vehicle)], vehicleModels: [vehicleModelToXmlShape(form.model)], }) → importAsNetexToBackend(applicationImportBaseUrl, xml, token) → parseVehicleImportResponse(body) → navigate('/vehicle?selected=<newId>') (create) or closeSlider() (edit) New: useVehiclePairSave hook, useDirtyFormBlock hook, parse helper, 6 vitests for parseVehicleImportResponse. 7. ff4a73a vehicles: useDirtyFormBlock falls back to beforeunload under BrowserRouter react-router-dom's useBlocker requires a data router; the app uses the legacy <BrowserRouter>. The invariant-throw crashed the sidebar on render — caught by 8 e2e failures. Reverted to a beforeunload listener; in-app nav guard requires the router migration. 8. f32256b vehicles: simplify — dedup constants, extract Snackbar, tighten save contract ← this commit Code-review pass: extracted INLINE_MODEL_ID / BLANK_FORM / hydrateFromRow to vehicleFormDefaults.ts; extracted <SaveErrorSnackbar> (was byte-identical in 2 hosts); useVehiclePairSave.save() no longer throws but returns { newId, error } (removes the latent double error display + try/catch ceremony); parseVehicleImportResponse reuses the existing xmlParser singleton; stripped narrative / issue-number / commit-number comments. +93 / -243 net. ## Key design decisions (locked in the /grill-me session) - Object-level XML builder; one XMLBuilder.build per emission. - Form scope: both create and edit; partial hydration accepted on edit (full hydration awaits /netex/vehicles/{id}). - State shape: { vehicle: Vehicle, model: VehicleModel } — typed netex-ts-gen objects, no flat-union splitter. - VehicleModel emits only its unique fields (Manufacturer, Range, FullCharge, CustomerServiceContactDetails). Overlap fields (Name, Description, TransportTypeRef, BrandingRef, ValidBetween) live on Vehicle only. - VehicleModel.\$id = constant 'INLINE:VehicleModel:1'. Vehicle's VehicleModelRef always points at that constant; the server collapses the inline model under the 1-to-1 relationship. - Routing: edit in sidebar /vehicle?selected=<id>; create at /vehicle/new full-page; shared form component for both. - Envelope helper is type-blind; callers pass slot values pre-shaped via *ToXmlShape. Adding a new entity type doesn't change the builder. - Envelope defaults (namespace, version, RF:1, Europe/Oslo TimeZone) carried forward verbatim from the legacy wrapInPublicationDelivery to preserve output compatibility on the VehicleType save path. - After-save: navigate to /vehicle?selected=<newId>. Parse Vehicle/@id from the response body; fall back to /vehicle plain when parse returns null. ## Vehicle name collision (resolved by directory scope) Three Vehicle-named types coexist; none clash because no file imports more than one: - src/data/vehicles/Vehicle.ts Full NeTEx entity, netex-ts-gen output. Used by the form, mappings, and save path. - src/data/vehicle-types/vehicleTypeTypes.ts::Vehicle GraphQL wire projection (id, registrationNumber, version) returned as VehicleType.vehicles[]. Used by the VehicleType details page's nested list. - src/data/vehicles/vehicleTypes.ts::VehicleRow Flat enriched list-row used by /vehicle list code (9 files). VehicleRow is still load-bearing today: - The GraphQL list still returns the flat projection. Sobek's vehicles(filter) GraphQL pushdown (GH #24) hasn't landed; until it does, list code cannot fetch full Vehicle entities. - hydrateFromRow(row: VehicleRow): VehicleEditFormValue bridges list-row → form state with a 4-field mapping (id, version, registrationNumber, parentVehicleTypeId → TransportTypeRef). The remaining ~10 Vehicle fields start blank in edit mode until /netex/vehicles/{id} lands. - The sidebar's read-only context panel reads VehicleRow-only enrichment fields (parentVehicleTypeName, parentTransportMode) that aren't on the Vehicle entity itself. VehicleRow retires when Sobek exposes a per-id Vehicle read endpoint that returns the full NeTEx entity. Out of scope for this branch. ## Open follow-ups (not blockers for this PR) - #68 — modification=\"new|revise\" client policy. NOTE comment in useVehiclePairSave; the handler omits the attribute today. - #24 — Sobek vehicles(filter) GraphQL pushdown. When list filtering lands, ?search= filter UX may hide the deep-linked new row after create save; TODO comment in VehicleCreatePage flags the interaction. - useBlocker upgrade — requires <BrowserRouter> → createBrowserRouter migration. Today's useDirtyFormBlock is beforeunload-only; documented in the hook's JSDoc. - List refetch after sidebar-edit save — useVehicles().refetch is not yet plumbed through EditingContext to the sidebar editor closure; today the list shows cached data until the user navigates away. - /netex/vehicles/{id} read endpoint — unblocks full edit hydration (memory: project_vehicle_details_route_hathor). - netex-ts-gen post-gen hand-patches (Reshape param naming, dead childWrapped emission, Extensions?: any) — file upstream issue on entur/netex-typescript-model so future regens don't reintroduce them. ## Verification - tsc -b clean - vitest 113 / 113 green (13 new across two test files) - e2e no-auth 36 passed, 2 skipped - lint 10 warnings (all pre-existing) ## Files changed across the branch 21 files, +1,449 / -98 from base (fix/issue26-list-form). Generated NeTEx files: 4 (Vehicle.ts, Vehicle-mapping.ts, VehicleModel.ts, VehicleModel-mapping.ts — ~630 LOC). Hand-written code: ~890 LOC across 17 files. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Inanna's initial scaffold shipped two near-twin list-chrome pages. The Edit variant (inline-cell-edit mode) has never been imported in this fork's git history and forced bubbled-up tunables (e.g. the initial details-pane width) to live in both files. Delete src/pages/GenericDataEditPage.tsx and record the decision in FORK_DECISIONS.md. The surviving GenericDataViewPage carries the DETAILS_PANE_INIT_WIDTH const (1/4 of viewport on first paint). Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Sidebar/ToggleButton/useResizableSidebar gain an optional side: 'left' | 'right' (default 'left'). GenericDataViewPage pins it to 'right' via a bubbled-up DETAILS_PANE_SIDE const and computes the content-area margin key from it. Mechanism lifted from theme/extends commit 9b1028e, with theme.sidebarSide reads replaced by hardcoded props. We deliberately skip the theme key augmentation, settings selector, and other layout knobs from that branch — only the side-flip is wanted here. The resizer math anchors to the right edge (window.innerWidth - clientX) when side='right', so dragging the handle inward narrows the pane. ToggleButton's border-radius and chevrons mirror the side. Mobile Drawer anchors to the chosen side too. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
- Share `Side = 'left' | 'right'` from Sidebar.tsx; drop the inline
union duplicated across 4 files.
- Bubble up more tunables:
- GenericDataViewPage: DETAILS_PANE_WIDTH_FACTOR (0.25),
DETAILS_PANE_GUTTER_PX (replaces inline `+ 4`), MARGIN_TRANSITION
(narrowed from broad `margin` to the actually-animated side).
- useResizableSidebar: MIN_W, MAX_W_RATIO (was inline 100 / 0.8).
- Fix module-load staleness: compute initWidth from
window.innerWidth inside the component so a viewport resize
before first list-page mount is honored.
- Extract closeIcon const in Sidebar to mirror ToggleButton's
collapsedIcon/expandedIcon pattern.
- Drop the WHAT-narrating "Change is here: use editingItem" comment.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
VehicleEditForm originally stacked 12 floating-label TextFields,
burning ~56px per row and forcing the sidebar pane to scroll. Switch
to a label-on-left / input-on-right layout using a single CSS Grid
container with two columns (`minmax(8rem, 12rem) 1fr`) and a tiny
local <FieldRow id label alignTop?> helper that uses
`display: contents` so its <InputLabel> and <TextField> become direct
grid items of the parent.
Column alignment across all rows is guaranteed by the shared grid,
labels associate to inputs via htmlFor/id (native a11y), and
`size="small"` drops each row ~56 -> 40px vertical. Section dividers
and overline headers span both columns via `gridColumn: '1 / -1'`.
Responsive fallback: `xs: '1fr'` collapses to single-column on
narrow viewports.
Also in this pass:
- Rename the second section header from "Vehicle Model" to "Model".
- Drop the ShortName field (not authored separately in this UI).
- Reorder Vehicle fields: Name -> Reg.no -> Op.no -> Chassis ->
BuildDate -> Reg.date -> Description -> TransportTypeRef.
- Description is now a regular single-row input (no multiline).
Document the decision in FORK_DECISIONS.md with the alternatives
considered (Stack+Typography, Grid v2, FormControlLabel) and why each
was rejected.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Add src/data/netex/NetexId.tsx — a small chip component for NeTEx ids
rendered as CODESPACE:TYPE:**VALUE** (bold VALUE, monospace, outlined),
with an optional nested vN chip for version. Missing or empty
CODESPACE/TYPE segments render as ???; the rightmost segment is always
treated as VALUE so partial ids like `abc-123` show as ???:???:abc-123
and colons inside VALUE are preserved.
VehicleDetails read-only head:
- Replace the separate ID + Version ContextRow pair with a single
<NetexId id={vehicle.id} version={vehicle.version} /> chip.
- Vehicle Type row: render the parent name AND a <NetexId> chip of
parentVehicleTypeId on the same row, so the canonical NeTEx ref is
visible without an extra labelled row.
- ContextRow rewritten with `display: contents` so its label and value
are direct grid items of a wrapping CSS Grid (same column template
as the edit form below). value prop widened to ReactNode for the
Vehicle Type row.
- Label typography aligned with the form's <InputLabel> styling
(fontSize 0.875rem, color text.primary).
VehicleEditForm:
- Export LABEL_COL_MIN / LABEL_COL_MAX / COL_GAP so the read-only head
can share the same column template.
- Remove the TransportTypeRef FieldRow — that data is now in the head.
- Date labels drop the "(YYYY-MM-DD)" suffix; the format is shown as
ghost placeholder text inside the input via DATE_PLACEHOLDER.
Document the new NetexId component in FORK_DECISIONS.md.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
VehicleDetails header was a static "Vehicle Details" h6 with the
NetexId chip sitting in the context block below. It now reads as a
live mirror of the form's Vehicle Name field — typing in the Name
input updates the title in place — with the NetexId chip flowing
on the same line, right of the title text. The Edit/View chip stays
at the far right; the title gets noWrap + minWidth: 0 + flex: 1
so long names truncate via ellipsis without pushing the chip out.
Blank-name fallback: when the trimmed Name is empty, the title
renders `[ unnamed ]` with the brackets in the default h6 weight
and only the word `unnamed` styled as a placeholder (italic,
text.disabled, fontWeight 400, slight letter-spacing). Bubbled-up
const: BLANK_NAME = 'unnamed'.
The read-only context block below loses its outer Stack wrapper
and the NetexId entry — it's now just the two-row CSS Grid
(Vehicle Type, Transport Mode) with data-testid="vehicle-context"
moved onto the grid Box itself.
Extract `firstText(arr)` to src/data/netex/multilingualString.ts —
the helper reads NeTEx MultilingualString[] shape and is now used
by both VehicleEditForm and VehicleDetails. Sharing it via the
edit-form export tripped react-refresh/only-export-components;
a dedicated module is the right home for a netex shape helper.
E2E: five `getByRole('heading', { name: 'Vehicle Details' })`
assertions swapped to `getByTestId('vehicle-details-title')`,
since the heading text now varies with the form's Name field.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Vertical rail anchored at the sidebar's content-facing edge. View mode shows collapse + pen; edit mode shows collapse + cancel + save. Cancel reverts the form to the hydrated row and returns to view mode, confirming via the existing discard dialog when dirty. Replaces the standalone ToggleButton + per-feature Edit chip + Save + Close trio in VehicleDetails. GenericDataViewPage exposes --sidebar-width and --app-header-height so the rail can position itself with position:fixed. e2e selector updated to editor-rail-collapse. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Addresses Gunnar's review on PR #67 — REST URL convention is plural for collections (`/vehicles/:id` reads naturally, `/vehicle/:id` reads as a typo). Renamed across: - src/App.tsx route definitions (/vehicles, /vehicles/new, /vehicle-types, /vehicle-types/new, /vehicle-types/:id, /deck-plans, /deck-plans/:id). - Navigation callsites: components/Menu, NewVehicleFab, VehicleCreatePage, AutosysImportFloatingMenu, deckPlanViewConfig, DeckPlanDetailsView, vehicleTypeViewConfig, vehicleUrlParams.vehicleSelectedHref, Home menu paths, vehicle-type quick-create link. - JSDoc, comments, theme/createThemeFromConfig notes. - e2e specs: every page.goto in auth + no-auth suites. Verified: tsc -b clean, npm run lint 0 errors, npm test 127/127 green, prettier clean. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
PR #74's plural rename made the bug visible: the "vehicles" column on /vehicle-types builds chips that deep-link to /vehicles/:id, and the nested vehicles[].id from the vehicleTypes(...) GraphQL query is the bare DB row id (e.g. "42") instead of the full NeTEx id — same default- fetcher fallback as transportType.id (sobek#125, secondary mention). The slider on the destination page can't match the bare id against the fully-formed ids in allData and renders the not-found body. Workaround mirrors the established transportType.id one in fetchVehicles.ts: project the nested vehicles[] in fetchVehicleTypes.ts through restructNetexId(parent.id, 'Vehicle', v.id). The pass-through guard means call sites stay correct after the Sobek fix lands. - fetchVehicleTypes.ts: project nested vehicles[].id via restructNetexId. - restructNetexId.ts: JSDoc enumerates both affected nested fields. - restructNetexId.test.ts: reverse-direction case (VehicleType → Vehicle). Refs sobek#125 (confirmed parallel via comment). Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
dynnamitt
changed the title
dynnamitt
force-pushed
the
phase2/issue72-vehicles-gql-cutover
branch
from
4028e71 to
b38438d
Compare
…to form grid - NetexId chips revert to MUI's plain outlined Chip surface; the CHIP_GRADIENT_LIGHT/DARK + CHIP_SHADOW_LIGHT/DARK constants and theme-mode branching go away. Outlined variant is still the default so the chip stays distinct from inline text. - VehicleDetails.tsx — both callsites bumped to size="small". Title header now lays out as a 2-col grid (Name | NetexId) matching the context strip + EditForm grid template, so all three blocks share the same column edges. Name truncates with ellipsis + native title tooltip when wider than the label column. VehicleDetails is the only NetexId consumer in the tree today, so the gradient went away wholesale rather than behind a flag. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
dynnamitt
force-pushed
the
phase2/issue72-vehicles-gql-cutover
branch
from
b38438d to
bd0585b
Compare
Contributor
Author
Code review — PR #745,308-line diff. Cumulative replacement for the iter-1 PR #67. Carries OverviewThe PR is well-structured with strong ADR documentation (FORK_DECISIONS.md), thoughtful directory split ( BLOCKERB1.
Comm: Acceptable! ; its a speculative TEMP test on backend MAJORM1. M2. M3. M4. M5. M6. EditorRail discard-then-save unhappy path. MINORN1. N2. N3. N4. N5. N6. N7. e2e selector scoping. N8. N9. N10. The two N11. N12. N13. FORK_DECISIONS rationale for the z-index bump. Claims the bump lifts the rail "with the sidebar's stacking context" — but the rail is PRAISE
Suggested gating
🤖 Generated with Claude Code |
…tuck spinner Early-return in doFetch left loading=true forever when applicationImportBaseUrl was undefined (initial useState(!!id) starts true; no setLoading(false) on the guard path). Split the guard: !id stays benign; missing baseUrl sets error + loading=false synchronously, mirroring useVehiclePairSave's message. Adds e2e regression in vehicle-loading-bug.spec.ts using a baseUrl-less config fixture; self-contained GraphQL intercept matches the new vehicles() query directly (skips the broken vehicleTypes-keyed helper). Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
…ehicle pair - Parser fix (M1 from PR #74 review): Vehicle-mapping emits domain TransportOrganisationRef as <AuthorityRef> and VehicleModelProfileRef as <CarModelProfileRef>; parser now translates XML element name → domain key via a RefMap, so the round-trip preserves both. Adds round-trip alias cases to Vehicle-parser.test.ts and VehicleModel-parser.test.ts. - Adds VehicleModel-parser.ts (+ tests) so the Model side has read parity with Vehicle whenever Sobek's streamOneVehicle starts shipping the Model. - Renames parseVehicleXml.{ts,test.ts} → Vehicle-parser.{ts,test.ts} for parity with Vehicle-mapping.ts (serializer <-> parser pair). - Extracts shared XML helpers into netex/ to drop ~140 lines of duplication between the Vehicle and VehicleModel mapping/parser files: netex/mapping-helpers.ts — attr/elem/refAttr/... + validBetween/keyValue/privateCode/text shapes + baseReshape dispatcher netex/parser-helpers.ts — attr/scalar/ref/textArr/projectValidBetween/... - Promotes generic NeTEx XML utils out of vehicle-imports/ into netex/: netex/xmlUtils.ts — ParsedXml, findResourceFrame, toArray, xmlParser vehicle-imports/xmlUtils.ts now keeps only import-specific orchestration (mergeResourceFrames, pubDeliverySingleRcFrame, extractVehicleTypeIds). Dependency direction is one-way now: netex/ → consumed by vehicles/xml/ and vehicle-imports/, never the reverse. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
…lpers
VehicleDetails previously split dirty tracking across three non-atomic
state slices: `useState(form)`, `useMemo(initialFormKey from xmlVehicle)`,
and `useEffect(setForm(hydrate(xmlVehicle)))`. The memo and the effect
both fire on xmlVehicle change but commit on different frames, so for one
render `form !== initialFormKey` even though no user edit has happened —
arming `beforeunload` on a freshly-loaded vehicle.
Extract a pure `vehicleFormState` reducer so `hydrate` advances `form`
AND `hydratedKey` in a single returned state. Divergence between the two
is now the only signal for `isDirty`, and it can only originate from a
user `edit` action.
Refactor VehicleDetails to drive the form via `useReducer(formReducer,
initialFormState)`. Cancel-edit dispatches `hydrate` instead of mirroring
the hydrate logic inline.
Tests (TDD red→green): 7 cases pinning the M2 invariant and the
edit/revert/cancel cycles. No jsdom needed — the reducer is pure node
TypeScript.
Also park NOTE comments on netex/{parser,mapping}-helpers.ts pointing at
entur/netex-typescript-model#49: once the codegen tool emits
bidirectional `Codec<T>` per entity (v2), this hand-written mirror pair
is deleted and consumers swap to `fooCodec.encode/decode`.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
… edit/post Codifies the boundary established by the projection/ + xml/ split: - projection types (VehicleGQLShaped) power search, list, filter, sort. - Detail / edit / POST flows must round-trip through xml/Vehicle + xml/VehicleModel via useVehicle → Vehicle-parser (hydrate) and Vehicle-mapping → /import (save). - Using a projection row as a write source silently drops NeTEx fields and carries display-flattened data the import path doesn't recognise. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
…n there FORK_DECISIONS.md header now identifies the file explicitly as Architecture Decision Records (ADR) and declares itself the single home for design rationale. CLAUDE.md's pointer is tightened correspondingly: do not restate rationale in CLAUDE.md (it describes what the patterns are; FORK_DECISIONS describes why). No content shuffled between the two files — a duplication audit didn't find any actual repetition. The change is purely about making the contract between the two documents explicit so future updates land in the right place. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
M3 — useVehicles.doFetch wrapped `fetchVehicles(...).then(...).catch(...).finally(...)`
in an `async` function with no return — the async returned an
already-resolved promise while the chain ran detached. Callers awaiting
`refetch()` couldn't observe the new data.
Extract `fetchVehiclesAndApply` (pure async orchestration) so the await
covers the full chain. `useVehicles.doFetch` is now a thin
`try { await fetchVehiclesAndApply(...) } finally { setLoading(false) }`.
Bonus: `translateVehiclesFetchError` is now an exported pure function
with its own test coverage.
M4 — VehicleDetails.handleSave fired the success snackbar and flipped
out of edit mode without refreshing the list — the `/vehicles` row
stayed stale until manual reload after the save.
Extract `commitSave(save, form, onSaved?)` — pure async helper that
awaits onSaved on success (skips on error) and returns the save result.
Plumb a refetch channel end-to-end:
UseDataReturn.refetch? (viewConfigTypes)
→ GenericDataViewPage destructures from useData()
→ useUrlEffect params gain optional refetch
→ useVehicleUrlSelection forwards into VehicleDetails closure
→ VehicleDetails.onSaved prop is wired into commitSave
After a successful save, the success snackbar appears only once the list
refetch promise has resolved — so the user sees the refreshed row, not
a stale one.
TDD: red→green on both. 10 cases pin fetchVehiclesAndApply (M3 await
semantics + error translation), 5 cases pin commitSave (M4 chain
ordering + error skip).
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Sorting while the sidebar editor held a selection interacted with
useVehicleUrlSelection's URL→editor→setPage cascade, producing a
flicker-then-revert glitch on the column header. UX call: lock sort
on all sortable columns while a row is being edited rather than
chasing the cascade.
When GenericDataViewPage sees an active editingItem, it forwards
sortLocked=true to PageContentComponent. DataTableHeader then
wraps each sortable label in a MUI Tooltip carrying a
LockOutlinedIcon + i18n("data.table.sortLocked"), dims to opacity
0.55, sets cursor:not-allowed, and drops the onClick — the
active-column sort arrow stays visible so the user retains
ordering context.
Generic by design: every projected list (/vehicles, /vehicle-types,
/deck-plans, future ones) inherits the behaviour through
GenericDataViewPage; no per-view wiring.
E2E coverage in vehicle-registration-sort.spec.ts:
parameterized harness runSortStabilityTests({ locked }) emits one
test per sortable column. Sidebar-closed run asserts toggle works
and the order persists across a settle window (anti-flicker).
Sidebar-open run asserts the click is a no-op and the hover
surfaces the lock tooltip. 10/10 green on Chromium and Firefox.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Drops the 16px CircularProgress + "Loading vehicle…" caption while
xmlVehicle hydrates from /netex/vehicles/{id}, in favour of a
layout-matching wave-animated skeleton: title row + NetexId pill,
two context rows, seven vehicle field rows, model section divider
and heading, three model rows. Reuses the same grid template
(LABEL_COL_MIN/MAX, COL_GAP, ROW_GAP) so the swap from skeleton
to real form is zero layout shift.
Screen-reader semantics preserved: skeleton container carries
role="status" + aria-label={t('vehicles.loading', 'Loading vehicle…')},
so the previously-visible caption text is now the announced label.
VehicleDetails xmlLoading branch short-circuits to skeleton +
EditorRail (collapse stays live during load); the bottom path
simplifies from a triple xmlLoading/xmlError/form switch to a
plain xmlError-or-form ternary.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
The Playwright runs have been red on this branch since the issue-#72 GQL cutover (21885b7, May 16) — pre-existing rot, not caused by the sidebar/sort/visual commits above. Two stale assumptions, three small fixes: 1. vehicle-list-helpers.ts matched the legacy `vehicleTypes` query token. Post-cutover the production query is `Vehicles { vehicles(...) }` — no `vehicleTypes` substring. The intercept fell through, the dev server has no backend, and the table never rendered. Matcher now anchors on `vehicles(` (field name, stable across operation renames). 2. vehicles-list-mock.json reshaped from the legacy nested `data.vehicleTypes.content[].vehicles[]` envelope to the flat `data.vehicles.content[]` shape with `transportType` inlined per row — matches `fetchVehicles.ts`'s wire contract. 15 rows preserved (12 rail + 2 bus + 1 unknown-mode). 3. Singular `/vehicle` URL regexes in vehicle.spec.ts:103 and vehicle-type-filter.spec.ts:121 pluralised to `/vehicles$` and `/vehicle-types$` to match the route rename (66de9d7). Also: vehicle.spec.ts row-action test used `row.getByRole('button')` to find a per-row action button that no longer exists — clicking now goes via the whole-row `useVehicleRowClick`. Test clicks the row directly. Comment paragraph in vehicle-registration-sort.spec.ts about the shared helper being "still keyed on the legacy vehicleTypes query" dropped — drift is fixed. Validation: vehicle.spec.ts + vehicle-type-filter.spec.ts + vehicle-registration-sort.spec.ts + vehicle-loading-bug.spec.ts now 44/44 pass across Chromium and Firefox. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
GunnarAtEgde
approved these changes
May 18, 2026
Contributor
GunnarAtEgde
left a comment
GunnarAtEgde
left a comment
There was a problem hiding this comment.
This contains a LOT of AI generated code that is difficult to understand.
I say OK for now and will register some bugs that I have observed
Contributor
Author
|
4 new issues added catching review "minors" left here |
This was referenced May 18, 2026
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.
Summary
Cumulative replacement for the iter-1 PR #67. Carries the
/vehiclelist + sidebar details all the way to a working save round-trip against Sobek's real GraphQL + NeTEx XML endpoints.vehicles(...)GraphQL query (no more flatten overvehicle-types). Projection split intosrc/data/vehicles/projection/with TransportMode chip filter + OperationalNumber column + whole-row click.GET /services/vehicles/netex/vehicles/{id}(NeTEx PublicationDelivery) via a newparseVehicleXmlthat mirrors the existingvehicleToXmlShapewrite path — read and write are inverse mappers over the samexml/Vehicle.tsdomain shape.useVehiclePairSave→buildPublicationDeliveryXml→ import, with a positiveSaveSuccessSnackbar(4s timeout) confirming success. The sidebar stays open after save so the toast gets its full timeout.EditorRail(collapse · edit → cancel/save); CSS-grid form withFieldRow.Closes / refs
vehicles()GQL + restructure dir #72 — Vehicle list: cut over to Sobekvehicles()GQL + restructure dir.restructNetexId); entur/sobek#126 (Vehicle.name GQL field — required for the deferred Name column on the list).What's in here
Projection (list side, #72):
src/data/vehicles/projection/—vehicleGqlShaped.ts,fetchVehicles.ts(Sobekvehicles()GQL, nested transportType),useVehicles.ts,vehicleSortValue.ts,vehicleViewConfig.tsx,useVehicleUrlSelection.tsx.src/graphql/vehicles/queries/fetchVehicles.ts— paged GQL request.src/data/netex/restructNetexId.ts— bridge for nestedtransportType.id(DB row id → reconstructed NeTEx id; sunsets when sobek#125 ships).Form/details (xml side, #26/#69/#73):
src/data/vehicles/xml/— generatedVehicle.ts+VehicleModel.ts(via netex-ts-gen),Vehicle-mapping.ts+VehicleModel-mapping.ts(write-sidevehicleToXmlShape),buildVehiclePairXml.ts,parseVehicleImportResponse.ts(+ test),useVehiclePairSave.ts, and the newparseVehicleXml.ts(+ test) +fetchVehicleNetexXml.ts.src/data/vehicles/useVehicle.ts—{data, loading, error, refetch}, fetch+parse pipeline.src/data/vehicles/VehicleDetails.tsx,VehicleEditForm.tsx,vehicleFormDefaults.ts— readonly head with NetexId chip, CSS-grid label/input form, EditorRail with collapse/edit/cancel/save, dirty-formbeforeunloadguard.MISSING_MODELplaceholder — Sobek'sstreamOneVehicleemits only the Vehicle, not its Model; the form's model slot showsManufacturer: [missing]until Sobek adds a single-Model export or sobek#126'snamefield unblocks a richer projection.Chrome / shared:
src/components/sidebar/EditorRail.tsx— collapse + edit → cancel/save segments; right-side flip on list pages.src/data/netex/NetexId.tsx— copy button, 3D gradient, size scale.src/data/netex/publicationDeliveryXml.ts— builder replacingwrapInPublicationDelivery; tested viapublicationDeliveryXml.test.ts.src/data/netex/transportMode.ts—TransportModeunion with'unknown',toTransportMode()boundary mapping,transportModeFilters.i18n:
transportMode.*,vehicles.*(incl.vehicles.loading,vehicles.saveSuccess),vehicles.field.*in en + nb.Docs:
FORK_DECISIONS.md— ADR-style log seeded with the deep-link rationale and theprojection/+xml/split decision.Out of scope (follow-ups)
TransportTypeRefandVehicleModelRef— generic Ref rendering today; pick once cross-form Ref UX lands.[missing]placeholders for model fields; needs a Sobek endpoint mirror ofstreamOneVehicleto populate.Test plan
npm run lintclean (0 errors; pre-existing react-refresh / exhaustive-deps warnings on unrelated files only).npm run checkclean (Prettier).npm run buildclean (tsc -b + Vite).npm test— 127 vitest cases pass across 17 files, including the newparseVehicleXml.test.ts(seed-fixture, CompositeFrame, empty RF, multi-Name, round-trip viavehicleToXmlShape).npm run localagainst Sobek :37999):/vehiclelist renders; TransportMode chip filters narrow rows; OperationalNumber column populated; whole-row click opens sidebar.GET /services/vehicles/netex/vehicles/{id}per slider open./vehicle?selected=does-not-existshows the existing not-found body.vehicles.saveSuccessshows in nb when locale is switched ("Kjøretøy lagret").🤖 Generated with Claude Code