Two follow-up fixes after the FloorElevationSystem landed:
- `viewer/src/components/viewer/index.tsx`: the auto-format hook
stripped the `FloorElevationSystem` import a second time, leaving
the JSX mount unresolved.
- `packages/nodes/src/column/definition.ts`: column stores Y rotation
as a scalar `number`, but `floorPlaced.footprint` types the rotation
field as the full Euler tuple. Wrap as `[0, column.rotation, 0]`.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
The previous commit's `floorPlaced` capability used `getScaledDimensions`
in `item/definition.ts` and `ColumnNodeType` in `column/definition.ts`,
but the editor's auto-format hook ran between the import edit and the
body edit and removed both as "unused" — breaking `bun dev`.
Re-add the imports.
Symptom: a shelf placed on a level with a raised slab underneath
visually clipped through it — `ItemSystem` lifted items onto slabs,
but shelves (and other floor-placed registry kinds) had no equivalent
path.
Fix: lift the slab-elevation logic out of `ItemSystem` into a generic
`<FloorElevationSystem>` keyed off a new `capabilities.floorPlaced`
config. Any kind that opts in declares a `footprint(node)` (dimensions
+ rotation used to query overlapping slabs) and an optional `applies`
predicate (skips items whose `asset.attachTo` is wall / ceiling).
The new system runs at frame priority 1 so its `mesh.position.y`
override lands before `ItemSystem` / `GeometrySystem` (priority 2)
clear the dirty mark. The spatial-grid sync's `markNodesOverlappingSlab`
also dropped its hardcoded `item` branch in favour of an iteration over
every registered kind that declares `floorPlaced` — so any new
floor-placed kind picks up slab-driven re-elevation automatically.
Tagged kinds:
- `item` — `footprint = getScaledDimensions`, `applies = !asset.attachTo`
- `shelf` — `footprint = (w, h, d)`
- `column` — `footprint = (w, h, d)`
- `spawn` — `footprint = (0.6, 1.8, 0.6)` (marker)
`ItemSystem` retains only the wall-side z-offset block (`mesh.position.z =
wallThickness / 2`). The elevation block + its `getScaledDimensions` /
`resolveLevelId` / `spatialGridManager` imports moved to the generic
system.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Final sweep after the layers, handlers, and state were dismantled —
pruning the remaining orphan types and constants that only the dead
code referenced.
Removed:
- 27 dead constants: `FLOORPLAN_*_HOVER_*_STROKE_WIDTH` for wall /
item / endpoint (consumed by the deleted legacy SVG layers),
`FLOORPLAN_WALL_OUTER_MEASUREMENT_*` + `FLOORPLAN_WALL_INNER_
MEASUREMENT_*` + `FLOORPLAN_OPENING_MEASUREMENT_*` palette
constants (no live measurement consumer), `FLOORPLAN_ITEM_
CLEARANCE_*` thresholds, `FLOORPLAN_MEASUREMENT_LABEL_*` / `LINE_
OUTLINE_*` + `FLOORPLAN_ACTION_MENU_OFFSET_Y` / `FLOORPLAN_NODE_
FOOTPRINT_*` / `FLOORPLAN_SPAWN_*` / `FLOORPLAN_TRACE_*_FILL_
OPACITY` + several `FLOORPLAN_WALL_*_STROKE_WIDTH` variants.
- Dead drag-state types: `SlabBoundaryDraft`, `SlabHoleBoundaryDraft`,
`SlabVertexDragState`, `SlabHoleVertexDragState`, `SlabHoleMoveDraft`,
`CeilingBoundaryDraft`, `CeilingVertexDragState`,
`CeilingHoleBoundaryDraft`, `CeilingHoleVertexDragState`,
`CeilingHoleMoveDraft`, `ZoneBoundaryDraft`, `ZoneVertexDragState`,
`WallFaceLine`.
- Dead helpers: `getWallMeasurementOverlay` + `getLinearMeasurementOverlay`
(both defined but never called now that the measurement layers are
gone).
Floor-plan panel: 9,006 → 8,651 lines (-355). Cumulative reduction
vs `main`: 17,913 → 8,651 (-9,262 lines, ~52%).
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
After the legacy SVG layers and handler callbacks were dismantled, the
backing drag-state useStates, mirror useRefs, and watch-effects became
orphans, along with a handful of top-level math helpers whose callers
were already gone.
Removed:
- 12 dead `useState` declarations: `slabBoundaryDraft` /
`slabVertexDragState` / `slabHoleBoundaryDraft` /
`slabHoleVertexDragState` / `slabHoleMoveDraft` / `ceilingBoundary
Draft` / `ceilingVertexDragState` / `ceilingHoleBoundaryDraft` /
`ceilingHoleVertexDragState` / `ceilingHoleMoveDraft` /
`zoneBoundaryDraft` / `zoneVertexDragState`. All written only by
the dead vertex-drag handlers and read only by the dead
`clear*BoundaryInteraction` callbacks and `transientFloorplanFit`
checks (which collapse to `false` once their inputs are gone).
- 5 dead `useRef` mirrors of the above (`slabBoundaryDraftRef` etc.)
plus their mirror `useEffect` writes.
- 5 dead `clear*BoundaryInteraction` callbacks (slab, slabHole,
ceiling, ceilingHole, zone) — only called by deleted useEffects
and by `clearDraft`, where the call became a no-op.
- 7 dead `useEffect` watchers for `slabVertexDragState`,
`ceilingVertexDragState`, `slabHoleVertexDragState`,
`slabHoleMoveDraft`, `ceilingHoleVertexDragState`,
`ceilingHoleMoveDraft`, `zoneVertexDragState` (each early-returns
because state is always null, so all listener wiring + commit /
cancel paths inside were unreachable).
- 5 dead `shouldShow*BoundaryHandles` flags + the 5 `useEffect`s
that called `clear*BoundaryInteraction` when they flipped.
- `selectedSlabEditingHoleIndex` / `selectedSlabEditingHole` /
`selectedCeilingEditingHoleIndex` / `selectedCeilingEditingHole`
— derived from now-`null` selectedX entries.
- Dead drag-state guards in `handlePointerMove`, `handleSvgPointer
Move`, and the floor-plan fit `useMemo`.
- Dead top-level helpers: `getRaySegmentIntersection`,
`getSlabHandlePolygon`, `getSlabVisualOffsets`,
`getDraftSlabVisualPolygon`, `WallMeasurementFaceContext` type,
`getWallMeasurementFaceContext`, `getAdjacentOpeningBounds`,
`getSelectedWallMeasurementOverlays`,
`getItemDimensionMeasurementOverlays`, `polygonCentroid`.
Floor-plan panel: 10,466 → 9,006 lines (-1,460). Cumulative reduction
vs `main`: 17,913 → 9,006 (-8,907 lines, **~50%**).
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Continuing the post-migration cleanup. After registry-driven kinds
absorbed selection chrome, action menus, and boundary editing through
`def.floorplanAffordances`, the legacy plumbing in `floorplan-panel.tsx`
was left mounted with empty inputs — another ~4k LoC of dead code.
Removed:
- `Editor2dFloorplanActionMenuLayer` mount + its 10 `selectedX
ActionMenuPosition` useMemos. All ten action menus computed positions
from empty entry arrays / null `selectedXEntry`, so the layer never
rendered anything. `FloorplanRegistryActionMenu` is the only mount left.
- Legacy handle layers: `FloorplanWallEndpointLayer`,
`FloorplanFenceEndpointLayer`, `FloorplanWallCurveHandleLayer`, and
the four `FloorplanPolygonHandleLayer` mounts for slab / slab-hole /
ceiling / ceiling-hole — all rendered from `wallEndpointHandles` /
`slabVertexHandles` / etc. which were empty after the registry took
over endpoint, curve, and polygon affordances.
- `FloorplanZoneLabelLayer` mount + component. Zone labels are now
emitted as `kind: 'text'` from `def.floorplan` on the zone kind.
- Dead handlers: every `handle*Select`, `handle*PointerDown`,
`handle*DoubleClick`, `handle*HoverChange`, `handleFloorplan*Hover
Enter`, `handleSelected*` (Move/Delete/Duplicate/AddHole/HoleMove/
HoleDelete/Curve for all 10 legacy kinds), `duplicateSelected*`,
`handleSelectedWallCurve`, `handleSlabVertex*`/`Midpoint*`/`Edge*`/
`Hole*` (same for ceiling and zone), `handleWallEndpointPointerDown`,
`handleFenceEndpointPointerDown`, `handleWallCurvePointerDown`,
`emitFloorplanNodeClick`, `syncDeleteHoveredId`,
`handleZoneLabelClick`, `hasDuplicatableFloorplanSelection`,
`handleDuplicateFloorplanSelection`, `FloorplanDuplicateHotkey` mount.
- Dead handle data: `wallEndpointHandles`, `fenceEndpointHandles`,
`wallCurveHandles`, `canCurveSelectedWall`, and all the
slab/slab-hole/ceiling/ceiling-hole/zone vertex+midpoint+edge handle
useMemos.
- Dead selection / measurement useMemos: `selectedItemEntry`,
`selectedOpeningEntry`, `selectedSpawnEntry`, `selectedFenceEntry`,
`selectedStairEntry`, `selectedRoofEntry`, `selectedElevatorEntry`,
`selectedWallEntry`, plus `selectedItemClearanceMeasurements` and
`movingOpeningPlacementMeasurements` (both 100-200 lines of math
that consumed the empty entries).
- Two dead `<FloorplanMeasurementsLayer>` mounts (clearance / opening
placement) — both fed by useMemos that returned `[]`.
- `<FloorplanZoneLayer>` mount — `visibleZonePolygons` is always empty.
Kept:
- `siteVertexHandles` / `siteMidpointHandles` + `<FloorplanPolygonHandle
Layer>` mount for site — site is the only kind not registry-driven.
- `FloorplanStairLayer` mount (preview-only, hover/click props swapped
for noop helpers since the preview isn't interactive).
- Stub `selectedSlabEntry` / `selectedCeilingEntry` / `selectedZoneEntry`
as `null` typed values so the remaining hole-editing fallback code
compiles. Those fallbacks themselves run as no-ops now and can be
torn down in a follow-up.
Floor-plan panel: 14,395 → 10,466 lines (-3,929). Cumulative reduction
vs `main`: 17,913 → 10,466 (-7,447 lines, ~42%).
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
The Phase 5 / 6 migration moved 7 kinds (fence, column, spawn, item,
elevator, stair, roof) to the registry's `def.floorplan` path, but the
legacy SVG layers in `floorplan-panel.tsx` were left mounted with empty
entry arrays — dead code carrying ~3k lines of useless cost in the diff.
Removed:
- `FloorplanGeometryLayer` (~1.7k lines): inline component rendering
walls / slabs / ceilings / openings from `wallPolygons` /
`slabPolygons` / `ceilingPolygons` / `openingsPolygons` — all
permanently empty stubs after the migration. Wall / slab / ceiling
/ door / window now render via `FloorplanRegistryLayer`.
- `FloorplanFenceLayer` (~265 lines): fence entries always empty
post-migration; fence renders via the registry.
- `FloorplanElevatorLayer` (~420 lines): elevator entries always
empty post-migration.
- `FloorplanNodeLayer` (~415 lines): rendered items / spawns / stairs.
Items + spawns are registry-driven; stair only needed the in-flight
preview, which is now mounted directly via `FloorplanStairLayer`
(preserved as a sibling of the registry layer).
- `FloorplanItemImage` (~40 lines): internal helper used only by
`FloorplanNodeLayer`.
- `floorplan-roof-layer.tsx` (113 lines): roof / roof-segment now
registry-driven.
Net: -2,989 lines from `floorplan-panel.tsx`, plus the deleted roof
layer file. The remaining 14k-line monolith still owns the orchestration
state (selection lookups, marquee, helper lifecycles, hit-test plumbing)
— that's a separate teardown.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Three bugs surfaced after the Stage E node-registry migration:
1. Ceiling intercepts 3D hover/click selection
Selecting via the floor-plan helper or the boundary-editor handles
is the intended flow; a direct 3D click on the ceiling should fall
through to whatever's underneath. `SelectionManager` now early-returns
on `ceiling` in onEnter/onLeave/onClick, so `event.stopPropagation`
is skipped and the ray reaches the item/wall/floor below.
2. Ceiling item placement: final click does nothing
When a ceiling-attached draft hangs in front of the ceiling-grid
mesh, the click ray hits the draft first and fires `item:click`,
not `ceiling:click`. `onItemClick` already forwards self-clicks to
shelf-surface / item-surface hosts; this PR adds the matching
ceiling branch so the commit lands on the ceiling under the cursor.
3. Floor-plan item move drift after the commit click
Two contributing causes, both fixed:
- `usePlacementCoordinator`'s `useFrame` lerped the draft mesh
toward `gridPosition.current` (the item's pre-move spot) every
frame, fighting React's render from `scene.position` while the
2D `FloorplanRegistryMoveOverlay` drove the move. Gated the lerp
on a `has3DPointerDrivenMoveRef` flag set on first 3D pointer
event — pure 3D drags are unchanged.
- The overlay's pointer-up handler skipped a final `session.apply`
and committed at the last pointermove position. Browsers don't
guarantee a pointermove right before pointerup, so a quick click
after a drag could land a few pixels off. Re-apply at pointer-up
coords so commit matches where the user actually released.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Wholesale move of every remaining kind into its own subdirectory under
`packages/nodes/src/`, finishing the registry-driven migration. Each
kind now ships its definition, schema (re-exported from core), and any
of `geometry` / `renderer` / `system` / `floorplan` / `tool` /
`move-tool` / `panel` / `floorplan-move` / `floorplan-affordances` /
`parametrics` / `preview` it needs — no per-kind code remains under
`packages/editor/src/components/tools/` or
`packages/viewer/src/components/renderers/`.
Deleted (replaced by registry-driven equivalents):
- `tools/{ceiling,column,door,fence,item,slab,spawn,wall,window}/...`
(boundary editors, hole editors, placement tools, move tools,
endpoint movers, curve tools, helpers, math libs)
- `ui/helpers/{ceiling,slab,wall}-helper.tsx`
- `ui/panels/{column,door,elevator,item,roof,roof-segment,spawn,
stair,stair-segment,wall,window}-panel.tsx`
- `viewer/src/components/renderers/{building,ceiling,column,door,
elevator,fence,guide,item,level,roof,roof-segment,scan,site,slab,
spawn,stair,stair-segment,wall,window,zone}-renderer.tsx`
- `viewer/src/components/viewer/legacy-system.tsx`
Added under `packages/nodes/src/`:
- `building/`, `column/`, `elevator/`, `guide/`, `level/`, `roof/`,
`roof-segment/`, `scan/`, `shared/`, `site/`, `stair/`,
`stair-segment/` packages with definition + schema + renderer / system
/ floorplan / panel as appropriate.
- New `floorplan-move.ts` for every kind that supports 2D moves
(ceiling, door, item, shelf, slab, window) — single registry-driven
dispatch path via `def.floorplanMoveTarget`.
- New `floorplan-affordances.ts` for kinds with polygon / endpoint
drags (ceiling, fence, slab, wall) — using the shared
`polygon-vertex-affordance` factories.
- New per-kind `panel.tsx` for kinds with custom inspector content
(door, item, shelf, spawn, wall, window).
- New per-kind `tool.tsx` for placement (door, item, shelf, window).
- New per-kind `move-tool.tsx` for kinds with custom 3D move flows
(door, item, slab, window).
Coordinator + manager updates in `packages/editor/`:
- `tool-manager.tsx` resolves tools from the registry only — no
hardcoded type→component map.
- `panel-manager.tsx` resolves inspector panels the same way.
- `placement-{coordinator,strategies,types}.ts` extended with
shelf-surface placement.
- `selection-manager.tsx` adds the registry-selectable fallback.
- `floorplan-panel.tsx`, `floorplan-background-placement.ts`,
`floorplan-render-context.tsx` updated for the registry layer's new
contract (props, affordance dispatch, render context).
Viewer updates:
- `viewer/index.tsx` drops legacy renderer mounts.
- `node-renderer.tsx` resolves by registry only.
- `scene-bvh.tsx`, `use-node-events.ts`, `level-system.tsx`,
`wall-cutout.tsx`, `zone-system.tsx`, `materials.ts` adjusted for
the registry-only world.
Sidebar tree nodes for ceiling / fence / slab / shelf / tree-node
updated to read from the registered nodes instead of the deleted
legacy renderer trees.
Wiki: new `plugin-authoring.md` page, README index updated.
Tests in `packages/nodes/src/index.test.ts` validate every registered
kind has the required shape.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Captures the gotchas surfaced while building the shelf so the next
contributor adding a registry-driven kind doesn't rediscover them.
`node-definitions.md` — new Pitfalls section + a rule that builders
must emit local-space children. Covers:
- `<GeometrySystem>` must NOT mutate `group.position` / `group.rotation`
after rebuild (the renderer binds them via JSX prop).
- Tag geometry-built children with `userData.__fromGeometry` so
rebuilds don't dispose React-mounted hosted children (the
item-disappears-on-shelf bug).
- Previews must clone materials before mutating them when the kind's
builder caches at module scope.
- Host kinds need a `children: z.array(...).default([])` field on their
schema (and a migration patch for older scenes).
`tools.md` — three new move/placement pitfalls:
- Disable raycast on the moved mesh during drag, otherwise it captures
the ray and starves `grid:move` → commits land at the stale start.
- Commit handlers listen to `grid:click` AND every `${kind}:click` to
catch clicks that land on neighbouring 3D geometry first.
- Move tools must preserve the node's actual `rotation[1]` in
`useLiveTransforms` — hardcoding 0 makes the node un-rotate mid-drag.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Two related bugs caused floor items to disappear mid-2D-drag:
1. `startLevelId` did `parent.parentId` unconditionally for non-wall
parents — so for a floor item parented directly to the level (the
canonical convention from `use-placement-coordinator`), it returned
`level.parentId = building.id`. `findContainingSurface` then iterated
the building's children (levels, not slabs) and the fallback
`parentId: startLevelId` reparented the item to the building. Now we
walk the parent chain until we hit a `level` and short-circuit on
`parent.type === 'level'`.
2. `buildSurfaceItemSession` reparented floor items to a slab when the
cursor was over one. Slabs don't carry a `children` field on their
schema (only ceilings + level do), so `updateNodesAction`'s reparent
logic operated on `undefined.children` — the item dropped out of the
level→children DFS the floor-plan layer walks, and the polygon
stopped rendering mid-drag. Split into `buildFloorItemSession`
(always parents to the level, just updates position) and a
ceiling-only `buildSurfaceItemSession`. `findContainingSurface`
narrows to `'ceiling'` as the only valid target.
Matches the 3D `detachItemSurfaceToFloor` convention: floor items live
as level children, not slab children.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
`buildZoneFloorplan` now emits a centered name label at the polygon's
area-weighted centroid (Shoelace formula, with bbox-center fallback for
degenerate rings). Label uses the legacy `FloorplanZoneLabel` styling:
`fontSize: 0.2`, white fill, zone-color stroke, `paintOrder: 'stroke'`
for the "outlined text" look that stays legible above any fill.
When the zone is selected the builder also emits the polygon editor —
edge-handle per edge, midpoint-handle per midpoint, endpoint-handle per
vertex — driven by the shared `createPolygonVertexAffordance` /
`createPolygonAddVertexAffordance` / `createPolygonMoveEdgeAffordance`
factories slabs and ceilings already use. Zones have no `holes` field
so the factory's optional `holeIndex` stays undefined and the operations
target `node.polygon` directly.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
The slab body covered the underlying zone fill almost completely —
`opacity: 0.85` (gray) and `opacity: 0.95` (white-on-select) drowned out
any zone color sitting beneath. Switched to independent `fillOpacity`
(0.6 unselected / 0.45 selected) and `strokeOpacity` (0.85 / 0.96) so
the outline stays crisp while the zone color reads through the fill.
On the selected state the hatch overlay carries the visual weight; no
need for an opaque white background underneath.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Three layered improvements on the 2D registry layer:
- Z-order buckets so the SVG document order reflects intent: zones (0)
paint first, slabs/ceilings (1) next, every other kind (walls / items /
shelves / columns / stairs / …) on top. Stable sort preserves DFS order
within a bucket.
- Base / overlay split: each entry's `FloorplanGeometry` tree is walked
through `splitFloorplanOverlay`, partitioning into a base group
(polygons, paths, fills, hatches) and an overlay group (interactive
handles + labels — `text` / `endpoint-handle` / `midpoint-handle` /
`edge-handle` / `move-handle` / `dimension` / `dimension-label`). Base
renders rank-sorted; overlays paint after every base entry so polygon-
editor chrome on a selected slab and zone name labels stay legible
above the structural fills sitting on top of them.
- Click guard on the outer layer `<g>`. The base/overlay split means
pointer-down lands on base and pointer-up lands on overlay (selection
mounts the overlay on top mid-gesture). The browser then dispatches
`click` to the lowest common ancestor, ABOVE the entry's
`onClick={handleClickStop}`. Without a higher-level stop, the click
reached the SVG's `handleBackgroundClick` → `clear-elements` and the
selection set on pointer-down vanished a frame later. Scoped to
`onClick` only so pointer / hover / drag still propagate inside the
registry tree.
Also extends the `text` FloorplanGeometry with stroke / strokeWidth /
paintOrder / fontFamily so kinds can match the legacy "white fill +
colored outline" label look (used by the new zone name label).
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Two bugs caused dragged shelves to snap to a weird position on commit:
1. `MoveRegistryNodeTool` wrote `useLiveTransforms.set(id, { ..., rotation: 0 })`,
so during the drag `ParametricNodeRenderer` applied `<group rotation={[0,0,0]}>`
and the shelf visually un-rotated. On commit the live transform cleared
and the renderer re-read the node's true rotation — the snap-back read
as "reverts to a weird position." Now we capture `originalRotationY`
from the node at mount time and forward it on every set.
2. `<GeometrySystem>` reset `group.position.set(0,0,0)` /
`group.rotation.set(0,0,0)` after every rebuild. That was carry-over
from legacy per-kind systems that didn't bind `position` on the group.
`ParametricNodeRenderer` now drives the transform via JSX prop, and
the reset clobbered it — React doesn't necessarily re-render on a
rebuild tick, so R3F never re-applied the prop and the registered
`<group>` stayed at the origin. Removed the reset; builders are
expected to emit local-space children.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Schema v2 adds style/rows/columns/withBack/withSides/withBottom/bracketStyle
and a `children: ItemNode[]` field for item hosting. Schema-level defaults
preserve the v1 wall-shelf visual so existing scenes load unchanged; the
placement tool spreads `shelfDefinition.defaults()` for fresh shelves
(cubby 3x2 at 1m × 0.5m × 1.8m, thickness 0.05m, back/sides/bottom on).
Four style geometries (wall-shelf / bookshelf / open-rack / cubby) share
the dimensional schema. `shelfRowSurfaceYs` exposes one host surface per
row, plus the bottom-board top when `withBottom` is on for cubby /
bookshelf.
Material is a single paintable surface (same shape walls / slabs / stairs
use); `DEFAULT_SHELF_MATERIAL` aligned with `DEFAULT_WALL_MATERIAL` so
unpainted shelves read as the canonical off-white.
Preview clones each cached material before mutating `transparent / opacity`
on the ghost — without the clone the mutation leaked into the cached
`getShelfMaterial` instance every committed shelf was using, rendering
them all see-through after the first placement preview rendered.
Store hardening: `migrateNodes` patches missing `children: []` on v1
shelves, and `updateNodesAction` reparenting tolerates a missing children
array on the new parent. `MaterialTarget` enum adds `'shelf'` so paint
mode picks up the kind.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Three remaining wall affordances ported into nodes/src/wall/ as
direct copies of the legacy implementations. Wall D is now complete.
- move-endpoint-tool.tsx (426 LoC) — linked-wall corner cascade +
Alt-detach + angle label. Mounted via `affordanceTools.move-endpoint`.
- move-tool.tsx (804 LoC, the most complex tool in the editor) —
center-drag with axis lock, linked-wall corner cascade via
`planWallMoveJunctions`, bridge wall ghost previews, auto-slab
live preview via `planAutoSlabsForLevel`, R/T rotation in 45°
steps, Shift to bypass grid snap, isNew metadata strip on first
commit. Mounted via `affordanceTools.move`.
- tool.tsx (332 LoC) — two-click placement with length/angle HUD,
Shift to bypass angle snap. Mounted via `def.tool`.
Editor public surface gains:
- createWallOnCurrentLevel, snapWallDraftPoint, WallPlanPoint
- MovingWallEndpoint type
ToolManager dispatch for `movingWallEndpoint` routes through the
registry with the legacy fallback (same shape as the fence
move-endpoint dispatch).
Wall is now A ✅ C ✅ D ✅. Stage B still pending (geometry depends on
level-batch miter data, blocked on `ctx.levelData` design decision).
Stage E pending (drop WallPanel — has slider drags + actions).
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
User pointed at the palette's PNG icons in the bottom toolbar — they
expected the inspector title to use the same artwork, not the iconify
lucide glyphs. Different visual style.
- IconRef gains `{ kind: 'url'; src: string }`. Plain `<img>` render
in ParametricInspector (no next/image — the inspector is
`'use client'`).
- All currently-registered kinds switched to URL refs matching their
palette `iconSrc`:
fence → /icons/fence.png
slab → /icons/floor.png
ceiling → /icons/ceiling.png
wall → /icons/wall.png
spawn → /icons/site.png
shelf → /icons/column.png (placeholder, same as palette)
- Kind-owned panels (slab/ceiling) already pass URL strings to their
own PanelWrapper; unchanged.
Door/window/item will get URL refs when they register at Stage A.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
User pointed out the auto-derived panel was title-only; the legacy
panels rendered a small icon in front. Legacy used a URL path (next
Image). The registry-driven path has structured `IconRef` values
(iconify / svg / lazy component) declared on `def.presentation.icon`.
- PanelWrapper's `icon` prop now accepts `string | React.ReactNode`.
String → next/image (legacy URL behavior). Node → rendered as-is.
- ParametricInspector resolves `def.presentation.icon` to a node:
iconify → `<Icon icon="lucide:fence" />`, svg → inline svg, component
→ Suspense + lazy.
Kind-owned custom panels (slab/ceiling) keep their existing legacy
URL icons since they pass `<PanelWrapper>` themselves — no change
needed there.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
User noticed the auto-derived fence inspector was missing the legacy
panel's Length + Curve sliders, and the section labelling was wrong
(Posts → Structure). Both Length and Curve are awkward for the
parametrics field model:
- **Length** doesn't map to a single node key — it's derived from
`start`/`end`, and editing it moves `end` along the existing
direction.
- **Curve** maps to `curveOffset` but the slider's min/max are
bounded per-node by the chord length, plus updates need
`normalizeWallCurveOffset`.
Adds a `kind: 'custom'` field with a kind-supplied
`component: ComponentType<{ node, onUpdate }>`. The inspector mounts
it and lets the kind own rendering + update logic. `key` becomes a
free-form React key/label since it no longer needs to map to a node
property.
Fence parametrics now mirrors the legacy layout 1:1:
- Style (segmented controls + showInfill toggle).
- Dimensions (Length, Curve, Height, Thickness).
- Structure (Base Height, Top Rail, Post Spacing, Post Size, Ground
Clear, Edge Inset).
Length + Curve live in fence/inspector-editors.tsx.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
User pointed out two regressions in the auto-derived fence inspector:
- A `color` field rendered for fence — but the legacy `FencePanel`
hid it (fence's color is a leftover schema field that isn't part of
the inspector UX). Dropped from `fenceParametrics`.
- Style + base-style enums rendered as a dropdown, but the legacy
used the inline segmented switcher (Slat/Rail/Privacy +
Grounded/Floating). Added a `display?: 'select' | 'segmented'` hint
to the enum field kind. ParametricInspector renders SegmentedControl
when set; defaults to dropdown otherwise.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
The Stage E kind-owned panels (slab/ceiling) import lucide icons.
Without peerDependencies declaring lucide-react, consumers bundling
nodes/dist/<kind>/panel.js fail with "Module not found: Can't resolve
'lucide-react'" because the bundler has no signal that nodes needs it.
Matches the same pattern @react-three/drei + @react-three/fiber
already use here.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Three flavors of Stage E in one PR:
- **Fence** — fully auto-derived. Adds `kind: 'boolean'` to ParamField
(rendered as ToggleControl), wires `showInfill` through
`def.parametrics`. Legacy `FencePanel` deleted; the auto-derived
`<ParametricInspector>` now drives fence editing entirely.
- **Slab / Ceiling** — kind-owned via `parametrics.customPanel`. The
legacy panels have shape-specific bits (elevation/height presets,
area display, holes list with auto-vs-manual provenance) that don't
fit the auto-derived field model yet. `<ParametricInspector>` learns
to lazy-load and mount `parametrics.customPanel` when present;
legacy `SlabPanel` + `CeilingPanel` files relocate to
`nodes/src/<kind>/panel.tsx` and the legacy copies delete.
When `list` / `computed` / `action` field kinds eventually graduate
to auto-derived support, these custom panels collapse back into
`parametrics.groups`. The plan calls this out under "Custom-behavior
escape hatch" and Foot-gun 5 of the recipe.
Public-surface additions in `@pascal-app/editor`:
- `ActionButton`, `ActionGroup`, `PanelSection`, `SegmentedControl`,
`ToggleControl`, `PanelWrapper` — needed by the kind-owned panels.
Per-kind progress table: fence/slab/ceiling all flip to E ✅.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Two regressions in the Stage D port surfaced the same gap: the wiki
already said "the mesh offset must mirror the useLiveTransforms entry"
but didn't spell out that for polygon-based kinds (slab / fence /
ceiling / wall), the value to mirror is the DELTA (the group's local
position) — not the world location of the polygon's center.
Expanded the live-drag rule with a position-based vs polygon-based
split and an explicit note that ParametricNodeRenderer reconciles the
group's position via React on every Zustand notification, so the two
writes must agree to the exact value or they fight every frame.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
User reported the slab/fence mesh jittered/teleported continuously
while dragging (not on commit — during the drag). Root cause: the
move tool wrote TWO conflicting values per grid:move tick:
1. `mesh.position.set(deltaX, 0, deltaZ)` — relative offset, direct
Three.js mutation.
2. `useLiveTransforms.set(id, { position: [originalCenter + deltaX,
0, originalCenter + deltaZ], rotation: 0 })` — absolute world
position of the translated polygon center.
`ParametricNodeRenderer` reads `useLiveTransforms` and binds it via
React: `<group position={liveTransform.position}>`. So every Zustand
notification re-rendered the renderer and reconciled the group's
position back to "originalCenter + delta" (the absolute), overriding
the "delta" the direct mutation had just written. The two systems
fought every frame → visible jitter.
Fix: `useLiveTransforms.position` now holds the SAME delta the direct
mutation uses (`[deltaX, 0, deltaZ]`). React reconciles to the same
value the direct mutation already set — no conflict.
The cursor sphere position stays as the translated polygon center
(it's tracked separately via React state, not `useLiveTransforms`).
Ceiling aligned for consistency, though CeilingRenderer doesn't read
`useLiveTransforms` so the value there has no rendering effect.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
User-reported regressions on the 1:1 legacy ports: slab/ceiling moves
were slow (polygon CSG rebuilds per scene.update tick), fence moves
teleported briefly on commit (residual mesh.position offset survived
the geometry rebuild). All three now use the same live-drag pattern
the legacy fence move was designed for:
- During drag, write only to `sceneRegistry.nodes.get(id).position`
+ `useLiveTransforms`. No `scene.update`, no polygon rebuild, no
React re-render of geometry.
- History stays UNPAUSED — scene state isn't changing.
- On commit, a single `scene.update` writes the translated
polygon (or fence start/end + linked-fence cascade). Recorded as one
natural undo step.
- Tools leave `mesh.position` at the drag delta on commit;
GeometrySystem / CeilingSystem reset it to (0,0,0) when they
rebuild the geometry on the next frame. By the time position
clears, the new geometry is in place — no teleport.
Two framework changes enable this:
- `GeometrySystem` (viewer/systems/geometry) now resets
`group.position` + `group.rotation` after every rebuild, matching
the legacy `FenceSystem.updateFenceGeometry` behavior. Tools that
translate the group during live-drag can rely on the reset.
- Legacy `CeilingSystem.updateCeilingGeometry` extends its existing
`position.y` reset to cover X/Z too — previously it left X/Z at the
drag delta after rebuild, double-translating the visual.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Each tool is now a direct copy of the legacy implementation, relocated
under @pascal-app/nodes/<kind>/ and dispatched via the registry's
def.affordanceTools. No DragAction abstraction, no clever live-drag
exception, no novel snap pipeline — same code, same UX, same
performance, same history dance.
Ports:
- fence/curve-tool.tsx (legacy CurveFenceTool, 1:1)
- fence/move-tool.tsx (legacy MoveFenceTool, 1:1 — including
the mesh.position + useLiveTransforms
exception that the legacy uses for fence
specifically)
- wall/curve-tool.tsx (legacy CurveWallTool, 1:1)
- slab/move-tool.tsx (legacy MoveSlabTool, 1:1)
- ceiling/move-tool.tsx (legacy MoveCeilingTool, 1:1 — preview
fill + outline overlay preserved)
Drops the obsolete DragAction-based action files
(packages/nodes/src/{fence,wall,slab,ceiling}/actions/{curve,move}.ts)
and their now-empty actions/ directories where applicable. Fence
keeps actions/move-endpoint.ts since that port works.
Editor public surface gains `getWallGridStep` + `snapScalarToGrid`
(transitional exports — Stage F moves them into @pascal-app/nodes).
ToolManager + MoveTool dispatch unchanged: the same legacy-fallback
branches now mount the registry component because the affordances are
declared, but the rendered behavior matches the legacy because the
implementations are copies.
Per-kind progress: fence D ✅ (curve / move-endpoint / move / placement
all kind-owned), slab D ✅, ceiling D ✅, wall D 🟡 (curve only,
endpoint/move/placement still legacy).
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
User-reported regressions made it clear that my D-ports of the curve
and whole-item move tools introduced more friction than they removed.
The legacy CurveFenceTool / CurveWallTool / MoveFenceTool /
MoveSlabTool / MoveCeilingTool ship more polish than the new ports do
right now:
- Legacy curve tools pre-snap the pointer position to the 0.5m grid
BEFORE projecting onto the chord normal, and they Shift-toggle to
a free-place mode. The ports skipped both — finer math, but the
user-visible UX regressed (laggy because the cascade resolver fires
per move, history feels broken near the no-op threshold).
- Legacy whole-item moves use scene.update per tick which keeps
hosted children visually aligned. The live-drag mesh.position port
cleared the offset before the GeometrySystem could rebuild,
producing a one-frame teleport on commit.
Drop the affordance registrations for those tools — the ToolManager /
MoveTool dispatch falls back to the legacy per-kind tools when the
registry doesn't declare the affordance. Stage D progress preserved
for: fence move-endpoint (linked cascade + alt-detach), slab/ceiling
boundary + hole editors, fence placement, slab placement, ceiling
placement.
The kind-owned files (curve-tool.tsx, move-tool.tsx, actions/curve.ts,
actions/move.ts) stay on disk for the next iteration — when they
reach parity with legacy UX, re-add the affordance entries.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
User-reported regressions after the previous Stage D move ports:
1. Slab/ceiling moves were sluggish because the actions wrote
scene.update(polygon) every grid:move tick → React re-render +
CSG-with-holes geometry rebuild per frame.
Apply the live-drag exception (same recipe fence move already used):
visual translation via `sceneRegistry.nodes.get(id).position` +
`useLiveTransforms`; scene.polygon is only written on commit. Polygon
center precomputed in ctx, mesh-offset clears on commit/cancel.
2. Cursor sphere sat at the polygon center (offset from the user's
actual cursor by `originalCenter - first_cursor`). Move wrappers
now subscribe to `grid:move` and set `cursorRef.current.position`
directly — no React state, no per-tick reconcile. Cursor lands on
the user's pointer.
3. Fence move had the same React-reconcile-per-tick cost via its
`useLiveTransforms` subscription. Switched to the same direct
ref-mutation pattern.
Adds a "REAL bend" test pinning that one Ctrl-Z after a real curve
drag undoes only the bend, not the create. The previously reported
"first undo does nothing" outcome reproduces only for no-op bends
(drags within `normalizeWallCurveOffset`'s straight-snap threshold,
≈1.5cm on a 3m fence). For visible bends the dance pushes a real
pastState entry and one undo step rolls back the bend.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Three user-reported regressions, all in the Stage D move/curve ports:
1. **Fence/wall bend cancelled the fence creation on Ctrl-Z.** The
action.commit's `return false` shortcut on "no offset change"
bypassed the dance entirely — pastStates wasn't touched, so the
next Ctrl-Z fell through to whatever preceded activation
(typically the create step). Fix: always run the dance, even on
no-op commits. First Ctrl-Z then absorbs a silent no-op entry,
subsequent presses roll back real prior actions. Same fix applied
to fence move-endpoint, fence move, slab move, ceiling move.
2. **Slab/ceiling move 'maximum update depth exceeded' loop.** The
`useScene` selector in `SlabMoveTool` returned a freshly-allocated
`[sx, sz]` tuple on every call. Zustand's `Object.is` equality
failed each comparison → re-subscribe → re-render → loop. Fix:
subscribe to the stable live-node reference and derive the center
via `useMemo`. Same recipe for fence/ceiling move-tools, including
memoizing the `originalCenter` fallback that was getting a new
array per render.
3. **No grid-snap sfx during move drag.** Action `preview` now tracks
the last snapped pointer on a mutable `lastSnapped` ctx field and
emits `sfx:grid-snap` when it changes between ticks. Matches the
legacy MoveFenceTool's per-tick sound.
Locking test for foot-gun 1 lives in
`packages/core/src/services/single-undo-dance.test.ts`.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Includes a regression test for the "no-op bend" case that reproduces
the user-reported 'fence bend cancels fence creation' bug: when
action.commit returns false (draft === original), no pastState push
happens during the drag's commit path. The very next Ctrl-Z falls
through to whatever was on the stack before activation — typically
the fence creation itself.
The actual dance with real changes (delta !=0) works correctly — the
test exercising the full createDragSession flow with action.commit
returning true pins one undo step covering only the drag.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Direct copy of the fence curve recipe — pure
`curveWallDragAction` (chord-perpendicular projection + clamp +
normalize + single-undo dance) plus a thin wrapper feeding
`useDragAction`. Mounted via `def.affordanceTools.curve`.
Slight precision difference vs legacy: the legacy CurveWallTool snapped
the pointer position to `getWallGridStep()` before projecting onto the
chord normal; the ported action skips that pre-snap and relies on
`normalizeWallCurveOffset` to settle the final value. The user-visible
result is the same magnitude of step, just with the snap applied at
the offset level instead of the position level.
Remaining wall D affordances (endpoint move, whole-wall move,
placement) are larger and queued for future sessions — wall's move
tool alone is 804 LoC with the linked-wall corner-cascade logic.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Replicates the slab Stage D recipe for ceiling: four affordances
routed through the registry — def.tool for the placement flow,
def.affordanceTools for boundary edit / hole edit / whole-ceiling move.
Ceiling-specific bits preserved:
- Placement tool keeps the dual-cursor + vertical TSL-gradient
connector + ground-shadow lines (1:1 with legacy).
- Move tool wrapper renders the translucent preview fill + outline
overlay so the user sees the destination before clicking.
ToolManager mount sites for CeilingBoundaryEditor / CeilingHoleEditor
now route through `getRegistryAffordanceTool` with legacy fallback.
Per-kind progress: ceiling A ✅ B (n/a, def.renderer escape hatch
preserved) C ✅ D ✅; E + F pending.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Replicates the fence Stage D recipe for slab — three drag affordances
+ one placement tool, all routed through the registry:
affordanceTools:
'boundary-edit' → thin <PolygonEditor> wrapper (vertex/edge drag)
'hole-edit' → same for a single hole polygon
move → DragAction with single-undo dance
tool: () => placement (multi-click polygon with axis/45° snap)
`PolygonEditor` + `PolygonEditorProps` exported from
`@pascal-app/editor` as Stage D transitional surface (Stage F cleanup
moves them into `@pascal-app/nodes`).
ToolManager mount sites for SlabBoundaryEditor / SlabHoleEditor now
route through `getRegistryAffordanceTool` with legacy fallback.
Slab move action does not use the live-drag exception (polygon CSG
rebuild every tick) — matches legacy behavior; optimization is a
separate task once we measure.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Fourth and final Stage D fence affordance — the placement tool itself.
Unlike the drag affordances (curve / move / endpoint), placement is a
two-click flow with state across grid events, not a single drag-down →
drag-up lifecycle. `DragAction` doesn't fit; the component owns its
own emitter subscriptions directly. The kind exposes it via
`def.tool: () => import('./tool')` and ToolManager's existing
`getRegistryTool()` lookup mounts it (legacy `tools.structure.fence =
FenceTool` falls through when the registry entry is missing).
Adds transitional exports from `@pascal-app/editor` for the helpers
the kind-owned tool needs at module scope: `createFenceOnCurrentLevel`,
`markToolCancelConsumed`, `EDITOR_LAYER`. Stage F cleanup moves these
into `@pascal-app/nodes` once every consumer is registry-driven.
Fence Stage D is now complete. Per-kind progress: A ✅ B ✅ C ✅ D ✅
(four affordances ported — curve, move-endpoint, move, placement).
E (drop legacy panel) and F (cleanup) pending across all kinds.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Third Stage D affordance port (302 LoC legacy → action + thin wrapper).
The action (`actions/move.ts`) uses the live-drag exception documented
in editor/wiki/architecture/tools.md: visual-only updates via
`sceneRegistry.nodes.get(id).position` + `useLiveTransforms` during
the drag, no scene mutations. Avoids re-rebuilding the fence geometry
(many posts + infill panels) every pointer tick. Commit performs the
single-undo dance — writes final start/end to scene, geometry rebuilds
once, Ctrl-Z reverses the whole drag.
Linked-fence cascade follows the same shape as MoveFenceEndpoint —
any fence in the same parent that shared an endpoint at activation
moves with the drag.
`getRegistryAffordanceTool` extracted to
`tools/shared/affordance-dispatch.ts` so move-tool.tsx and
tool-manager.tsx share the lazy-load helper (no duplicate caches).
MoveTool dispatch gains a generic `affordanceTools.move` check after
the capability-driven `movable` shortcut — fence routes through here;
wall / slab / etc. fall through to their legacy per-kind chain until
their D ports land.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
These files had been showing as modified in every dev session — biome's
canonical formatting (line-length collapses, import sort, type-modifier
placement) didn't match the committed state. No semantic changes.
Committing now so the working tree stays clean across sessions.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
The previous commit() just returned true, relying on
createDragSession.terminate() to resumeHistory. But that alone never
captures the drag in zundo's pastStates — the pause window's mutations
are skipped entirely. After commit, Ctrl-Z jumped past the curve *and*
past the prior fence creation.
Adds the same dance now used by move-endpoint: restoreAll →
resumeHistory → re-apply the final draft. Zundo records one undo step
for the whole drag. cancel() becomes a no-op (orchestrator's
restoreAll covers it).
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Second Stage D affordance port (425 LoC legacy → ~430 split across
action + wrapper). All math (snap, linked-fence cascade, alt-detach,
min-length gate, single-undo dance) lives in the pure
`moveFenceEndpointDragAction`. The React wrapper handles the UI
overlays (cursor sphere, Drag/Detach badge, angle label) and the live
state subscriptions.
The action introduces the single-undo dance pattern for multi-write
commits: `commit()` calls `scene.restoreAll()` → `resumeHistory()` →
re-applies the final draft so zundo records the entire drag as one
undo step. Reusable shape for slab/wall/door endpoint ports.
Transitional exports added to `@pascal-app/editor`'s public surface
(`snapFenceDraftPoint`, `isWallLongEnough`, the segment-angle helpers,
`MovingFenceEndpoint`). Stage F cleanup moves these into
`@pascal-app/nodes` once every consumer is registry-driven.
`getRegistryAffordanceTool` is now generic (`ComponentType<any>`) so
affordances with different prop shapes (`{ node }`, `{ target }`, …)
all dispatch through the same helper.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
The previous dispose() was documented as "equivalent to cancel()" and
fired onCancel. That breaks React StrictMode's mount → cleanup → mount
cycle for useDragAction consumers: the first cleanup's onCancel resets
the parent state machine (e.g. setCurvingFence(null)), which unmounts
the component before the second mount runs. Net result: the tool blinks
in and out instantly.
dispose() now restores scene state + resumes history but skips onCancel.
Explicit cancel() still fires onCancel (Esc / external aborts). New
test locks this in.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
The legacy CurveFenceTool ignored grid:click for 150ms after mount —
otherwise the very click that activates a tool (e.g. the floating menu
"curve" button) cascades through the R3F drei <Html> portal into the
grid, fires grid:click on the just-mounted tool, and commits the drag
before any preview move runs. The new useDragAction was missing this
guard, so the Stage D fence curve port "click → place sfx → exit"
without ever letting the user adjust.
Adds `activationGraceMs` (default 150) on useDragAction; ports the
sfx:item-place commit emission into FenceCurveTool so the kind-owned
tool matches legacy UX. Same guard will cover the upcoming Stage D
ports (endpoint move, whole-fence move, placement, plus
slab/ceiling/wall D).
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
First Stage D port — `CurveFenceTool` (178 LoC legacy) split into a pure
`DragAction` primitive (`packages/nodes/src/fence/actions/curve.ts`) plus
a thin React wrapper (`packages/nodes/src/fence/curve-tool.tsx`) that
feeds it through `useDragAction`. The kind declares the affordance via
`def.affordanceTools.curve`; ToolManager lazy-loads it at runtime when
`useEditor.curvingFence` activates. Falls back to the legacy
`CurveFenceTool` for any kind that hasn't been ported.
The lazy-load dispatch dodges the editor→nodes circular dep (nodes
already depends on editor for `useDragAction` + `CursorSphere`).
Establishes the pattern for the remaining fence affordances
(`MoveFenceEndpoint`, `MoveFence`, placement) and for slab/ceiling/wall
D ports.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Shelf clicks in 3D weren't selecting: getSelectionTarget routed shelf
(category='furnish') to the furnish phase, but furnish.isValid hard-
coded `node.type !== 'item'` and rejected shelf. Click switched phase,
nothing selected.
Fix: extend furnish.isValid to also accept registry-driven kinds whose
def.category === 'furnish' AND def.capabilities.selectable. Item's
asset.category door/window special-case stays first.
Future furnish-category kinds (tables, lamps, etc.) are selectable in
furnish phase without further changes.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Item is the last Stage A-only kind to gain def.floorplan. Closes Stage C
across every registered kind.
Files added:
- nodes/src/item/floorplan.ts: buildItemFloorplan inlines a self-
contained parent-chain transform walker using `ctx.resolve`. Mirrors
the legacy `getItemFloorplanTransform` math from editor/lib/floorplan/
items.ts:
* Wall parent: rotate item.position by wall's angle, anchor at
wall.start, handle wall-side attachTo via wall.thickness offset.
* Item parent (nested): recurse for parent's transform.
* Level / slab / ceiling parent: item.position is level-local.
Returns a rotated width × depth rectangle. asset.floorPlanUrl image
overlay deferred for Phase 5 follow-up.
Files changed:
- nodes/src/item/definition.ts: wires `floorplan: buildItemFloorplan`.
- floorplan-panel.tsx: floorplanItemEntries useMemo short-circuits to
[] when nodeRegistry.has('item'). Phase 6 deletes the entire useMemo.
Stage C coverage (all 9 registered kinds):
shelf ✅ spawn ✅ fence ✅ slab ✅ ceiling ✅ wall ✅ door ✅ window ✅ item ✅
Next sessions: Stage B for door / window / wall (each large geometry
extraction), Stage D per kind (DragAction affordance ports), Stage E
(drop legacy panels), Phase 6 Stage F cleanup.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Three remaining kinds at Stage C this session:
wall → C
- buildWallFloorplan: uses ctx.siblings to gather other walls in the
level, runs calculateLevelMiters, computes plan footprint via
getWallPlanFootprint. Same visual output as legacy.
- getFloorplanWall thickness exaggeration inlined (~25 lines from
editor/lib/floorplan/walls.ts) to keep nodes/wall self-contained.
- floorplan-panel.tsx's wallPolygons short-circuits to [] when wall
is registered.
- Performance note: recomputes level miter data per wall (O(N²) for N
walls in a level). Acceptable for typical scenes; ctx.levelData?.
miters optimization deferred to Stage B's wall design pass.
door → C
- buildDoorFloorplan: inlines getOpeningFootprint math from
floorplan-panel.tsx (40 lines, pure math). Uses ctx.parent as the
wall to compute direction + perpendicular for the cutout footprint.
- Returns null when parent isn't a wall (orphaned doors during
placement).
window → C
- buildWindowFloorplan: same shape as door, glass-blue tint to
distinguish visually.
Both share the legacy openingsPolygons gating:
- floorplan-panel.tsx's openingsPolygons useMemo filters per kind so
a partial migration still works (e.g., if only door registers, only
doors get skipped). When both registered, returns [] entirely.
Item C intentionally deferred — needs parent-chain transform helpers
(buildFloorplanItemEntry / getItemFloorplanTransform from editor/lib/
floorplan/items.ts) exposed publicly or moved into core. A focused
session is the right place to design that boundary.
Stage B for door / window / wall still pending — each is a focused
session per kind (large geometry math extractions, wall needs ctx.
levelData design).
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Depth-first session: drive registered kinds through Stage B (pure
def.geometry, drop system re-export) and Stage C (def.floorplan,
short-circuit legacy inline rendering in floorplan-panel.tsx).
spawn → C
- buildSpawnFloorplan wired on definition (was written but deferred
to avoid double-render).
- floorplan-panel.tsx's floorplanSpawnEntries useMemo short-circuits
to [] when nodeRegistry.has('spawn').
fence → B
- generateFenceGeometry exported from viewer; buildFenceGeometry
wraps it in a Group+Mesh with DEFAULT_STAIR_MATERIAL.
- def.geometry set; renderer + system fields dropped.
- Deleted nodes/src/fence/{renderer.tsx,system.tsx}.
fence → C
- buildFenceFloorplan: polyline along centerline (sampled for curved
fences via sampleWallCenterline from core). Stroke width = node.thickness.
- floorplan-panel.tsx's floorplanFenceEntries short-circuits.
slab → B
- generateSlabGeometry exported from viewer; buildSlabGeometry wraps
it in a Group+Mesh + cached material (preset / custom / default
pattern preserved from legacy renderer).
- def.geometry set; renderer + system fields dropped.
- Deleted nodes/src/slab/{renderer.tsx,system.tsx}.
slab → C
- buildSlabFloorplan: SVG path with outer polygon + hole subpaths
(uses getRenderableSlabPolygon from core for wall-clipping parity).
- floorplan-panel.tsx's slabPolygons short-circuits.
ceiling → B INTENTIONALLY SKIPPED
- Ceiling renderer renders React children (hosted items) + uses TSL
shader materials + named meshes that other systems poke
(getObjectByName('ceiling-grid')). Pure def.geometry can't preserve
that. Ceiling keeps def.renderer (the custom escape hatch) — same
pattern item uses. Documented in ceiling/definition.ts.
ceiling → C
- buildCeilingFloorplan: dashed-outline path with hole subpaths
(visually distinct from slab since ceilings are above).
- floorplan-panel.tsx's ceilingPolygons short-circuits.
Per-kind progress after this session:
- shelf: B ✅ C ✅ (Stage E since brand-new)
- spawn: A ✅ C ✅
- wall: A ✅ (B blocked on ctx.levelData design)
- fence: A ✅ B ✅ C ✅
- slab: A ✅ B ✅ C ✅
- ceiling: A ✅ C ✅ (B intentionally not applicable)
- door / window / item: A ✅ (B+C pending in future sessions)
Known test issue: `bun test` in packages/nodes fails to load
`three-bvh-csg` through the viewer's transitive imports (UMD/ESM
mismatch in Bun's test runner). The Next.js editor build works fine
because it bundles differently. Fix requires either dynamic imports
(breaks sync def.geometry contract) or test env config — deferred.
Other tests (schema, geometry, parity) pass.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
User report: items needed a double click to select after the item kind
registered (Phase 5). Root cause: `getSelectionTarget` checked
`isRegistrySelectable(node.type)` as part of the FIRST branch (which
routes to structure phase), matching `item` before the item-specific
branch below could route door/window-category items to structure +
everything else to furnish.
Effect: clicking an item triggered phase switch (structure ← furnish),
then the next click selected. Hence the double click.
Fix:
1. Item-specific case moved to the TOP of getSelectionTarget. Its
asset.category-driven routing (door/window items → structure;
everything else → furnish) beats any generic registry fallback.
2. Generic registry fallback at the bottom now reads `def.category`
to pick the phase — `category: 'furnish'` → furnish phase,
everything else → structure/elements. Future furnish-category
kinds (only shelf right now) route correctly without a special
case.
3. `isRegistrySelectable(node.type)` clause removed from the
structure branch — replaced by the def.category check at the
bottom.
Net: single-click selection works for items again, and the routing
logic is now cleanly capability/category-driven instead of "all
registered kinds → structure" which was a Stage A simplification
that broke as soon as a furnish-category kind registered.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Item is the first kind to use the `def.renderer` escape hatch (custom
React component with `useGLTF` + drei + interactive widgets) — not
expressible as a pure `def.geometry`. Catalog-backed + multi-host
(free / wall / wall-side / ceiling).
Files added (packages/nodes/src/item/):
- schema.ts: re-exports ItemNode from core.
- parametrics.ts: empty groups[]. Item parametrics come from the
asset's catalog-defined interactive controls (toggles / sliders /
temperature) — too dynamic for the auto-inspector at Stage A.
Legacy ItemPanel renders the catalog-driven controls; Phase 5
Stage E will likely use parametrics.customPanel.
- definition.ts: capabilities (no `movable` — item move is bespoke
MoveItemContent that handles attachTo transitions floor↔wall↔
ceiling mid-drag; capability-driven dispatch keeps legacy mover),
parametrics, renderer (wrap-export of ItemRenderer), system
bundling ItemSystem + ItemLightSystem, toolHints matching the
user's screenshot (Place item / R rotate ccw / T rotate cw /
Shift free place / Esc cancel). defaults() casts an object
literal with a stub asset since asset is required by the schema;
createNode re-parses through ItemNode at runtime.
- renderer.tsx: wrap-export of legacy ItemRenderer (~280 lines with
useGLTF + interactive widgets — too much to duplicate at Stage A).
- system.tsx: bundles ItemSystem + ItemLightSystem.
- index.ts: barrel.
Files changed:
- packages/viewer/src/index.ts: new public exports for ItemRenderer,
ItemSystem, ItemLightSystem.
- packages/nodes/src/index.ts: appends itemDefinition.
- packages/editor/src/components/ui/panels/item-panel.tsx: panel
slider-drag fix recipe applied (nodeRef pattern, drop subscribed
updateNode dep, drop node from useCallback deps). Item panel has
scale + position + rotation sliders all subject to the cascade.
Item is the registry's stress test for `def.renderer` escape hatch.
GLB loading via useGLTF + drei works as-is; nothing in the registry
forces a pure-geometry shape on kinds that don't fit.
Phase 5 progress: shelf ✅ spawn ✅ wall ✅ fence ✅ slab ✅ ceiling ✅
door ✅ window ✅ item ✅. Nine kinds on the registry. Stair / roof /
zone / containers remain.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Both kinds share traits — hosted on walls, cuttable, animated open/
close state via a geometry system + animation system. Stage A
migration: register + wrap-export the legacy renderer + bundle both
per-kind systems. Pure geometry + floor-plan ports are later
milestones.
Files added (packages/nodes/src/door/, packages/nodes/src/window/):
- schema.ts: re-export from core.
- parametrics.ts: minimal — dimensions only. Door has 29 sliders +
segmented controls + presets in its legacy panel; window has 15+
sliders. Auto-inspector can't cover them at Stage A — legacy
panel keeps rendering via panel-manager.tsx case fall-through.
Stage E may extend parametrics or use parametrics.customPanel
escape hatch.
- definition.ts: capabilities (no `movable` — wall-bound drag is
bespoke; capability-driven dispatch keeps legacy MoveDoorTool /
MoveWindowTool), parametrics, renderer, system. defaults() uses
`DoorNode.parse({...stub})` to leverage zod's schema-level
`.default()` annotations — door has 40+ fields, window has 20+;
listing them inline duplicates the schema.
- renderer.tsx: wrap-export of legacy DoorRenderer / WindowRenderer
(thin 33-36 lines each).
- system.tsx: bundles each kind's TWO systems — DoorSystem +
DoorAnimationSystem, WindowSystem + WindowAnimationSystem. Both
per-kind systems mount via RegisteredSystems when the kind is
registry-driven; `<LegacySystem kind="door|window">` wrappers
around each individual system short-circuit.
- index.ts: barrel.
Files changed:
- packages/viewer/src/index.ts: new public exports for DoorRenderer,
DoorSystem, DoorAnimationSystem, WindowRenderer, WindowSystem,
WindowAnimationSystem.
- packages/nodes/src/index.ts: appends doorDefinition + windowDefinition.
- packages/editor/src/components/ui/panels/door-panel.tsx + window-
panel.tsx: panel slider-drag fix recipe applied. Drop the
subscribed `updateNode` action, drop the `node` dep from
handleUpdate / previewDoorUpdate / commitDoorPreview useCallbacks.
Use useScene.getState() inside. Door panel has 29 SliderControls,
window 15+ — both at high risk of the Maximum update depth
cascade without the fix.
Phase 5 progress: shelf ✅, spawn ✅, wall ✅, fence ✅, slab ✅, ceiling ✅,
door ✅, window ✅. Eight kinds on the registry. Item / stair / roof /
zone / containers remain.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
