v2.1.0

Minor Changes

• 9264ebb: Unify the coordinate frame: the Three.js scene is now Rhino's frame (Z-up), end to end.

  Previously the display pipeline rotated Rhino Z-up geometry into Three's native Y-up
  ((x, y, z) → (x, z, −y)) during mesh decompression and display-item parsing. That hidden
  rotation meant every feature producing or consuming positions — measurements, mesh metadata,
  label anchors, picking, the new camera presets/grid — had to round-trip through it or silently
  land in the wrong frame.

  The rotation is removed everywhere. A Rhino point (x, y, z) is now the Three point (x, y, z):

  • rhinoToThree is the identity (kept, deprecated, for call-site compatibility).
  • The int16/float32 vertex paths in webdisplay/batch-parser.ts pass vertices through unrotated.
  • initThree orients the camera, default iso position, sunlight, floor, and reference grid to the
    scene up axis (Z-up by default); the camera controller's presets are likewise up-derived.

  Breaking: any consumer that assumed viewer geometry was Y-up (e.g. reading mesh vertex
  positions, placing objects, or computing directions in Three space) must drop the
  (x, z, −y) conversion — Three space now equals Rhino space. The applyTransforms option is
  retained but no longer rotates; it will be removed in a future release.

• 9264ebb: Release v2.1.0: Z-up coordinate frame, measurement tool, and CAD viewer tooling.

  Coordinate frame unification (breaking)

  The Three.js scene is now Z-up end-to-end, matching Rhino's native frame. The hidden (x, y, z) → (x, z, −y) rotation that was applied during mesh decompression and display-item parsing is removed. rhinoToThree is kept as an identity for call-site compatibility but deprecated. Consumers that read mesh vertex positions or place objects in Three space must drop any (x, z, −y) conversion — Three space now equals Rhino space.

  CAD viewer tooling — public exports

  The camera controller, reference grid, view gizmo, edge overlays, label layer, and measurement tool are now exported from the public visualization entry point. Previously their factories and types were only available internally. New exports include createCameraController, createGrid, createViewGizmo, addEdges/removeEdges, createLabelLayer, createMeasureTool/snapToVertex, and all associated config/option types (GridConfig, GizmoConfig, EdgesConfig, MeasureConfig, ViewPreset, CameraProjection, …). Additive only — no existing export changed.

  Measurement tool improvements

  • Snapping extended to lines and points: snapToVertex snaps to the nearer endpoint of a struck segment and to the struck vertex on point objects. Raycast thresholds for lines and points are scaled by view distance so thin geometry is actually clickable at any zoom.
  • Cursor preview: a ghost marker follows the cursor and jumps to the snap vertex before a click commits it.
  • Drag-to-orbit no longer disturbs a measurement — clicks that follow a drag past the slop threshold are ignored.
  • Labels now show per-axis deltas (Δx/Δy/Δz) alongside the total distance. The format callback is widened to (distance, delta) => string; existing (distance) => string callbacks remain valid.
  • Labels ship with a default dark-pill style so they are legible on any background. Pass labelClassName to opt out.

  Camera, grid, and label layer fixes

  • initThree sets camera.up before constructing OrbitControls so preset views (top/front/…) and orbit behavior are correct for Z-up scenes.
  • The grid's default plane is derived from the scene up axis (Z-up → plane: 'z'); an explicit plane still takes precedence.
  • clearScene preserves the persistent floor, grid, and label-layer groups across content updates, fixing measurement labels disappearing in streaming viewers (e.g. per Grasshopper solve).
  • The CSS2D label overlay gets an explicit z-index so labels stack above container scrims while staying below menus and popovers.

• 9264ebb: Export the CAD viewer tooling from the public visualization entry point.

  The camera controller, reference grid, view gizmo, edge overlays, label layer, and
  measurement tool shipped in 2.1.0-beta.1 were wired through initThree at runtime, but
  their factories and types were only re-exported from the internal
  features/visualization/index.ts barrel — not from src/visualization.ts, the actual
  published entry. Consumers could enable the tools via options and read them off the
  initThree return, but could not import the supporting type names
  (CameraController, MeasureTool, ViewPreset, CameraProjection, …) or the
  config types (GridConfig, GizmoConfig, EdgesConfig, MeasureConfig).

  • Re-export createCameraController, createGrid, createViewGizmo, addEdges/removeEdges/isEdgeOverlay, createRenderPipeline, createLabelLayer, createMeasureTool/snapToVertex and their types from visualization.
  • Re-export the GridConfig/GizmoConfig/EdgesConfig/MeasureConfig option types.
  • Also surface parseColor, applyOffset, and computeCombinedBoundingBox from three-helpers.

  Additive only — no existing export changed.

Patch Changes

• 9264ebb: Fix measurement/dimension labels never appearing in viewers that stream new content (e.g. per
  Grasshopper solve).

  • updateScene/clearScene removed every top-level scene child except the floor on each update,
    which detached the persistent CSS2D label-layer group. Labels added afterwards were parented to
    an orphaned group, so the CSS2D renderer (which walks the live scene) never injected their DOM.
    clearScene now preserves persistent infrastructure — floor, grid, and label-layer — across
    content updates. Demos that add geometry directly (never calling updateScene) were unaffected,
    which is why the label only went missing in consumer apps.
  • The CSS2D label overlay also gets an explicit z-index so it stacks above container scrims (e.g.
    blur/loading overlays) that previously painted over it, while staying below menu/popover layers.

• 9264ebb: Extend the measurement tool to lines and points, not just meshes.

  • snapToVertex now snaps line hits to the nearer endpoint of the struck segment and point hits to
    the struck vertex, in addition to the existing mesh triangle-vertex snapping. Hits without usable
    vertex indices (e.g. fat Line2) still fall back to the raw point.
  • Line and Points raycast thresholds are raised per-pick, scaled by the view distance, so thin lines
    and points are actually clickable at any zoom instead of being nearly impossible to hit with the
    default ~1-unit threshold.

• 9264ebb: Make the measurement tool easier to read and aim, and report per-axis deltas.

  • Distance labels now carry a default style (dark translucent pill, light text) so they stay
    legible on any background instead of inheriting the page color (previously invisible white-on-white).
    Passing labelClassName still opts out of all default styling.
  • The tool previews the snap point: a ghost marker follows the cursor and jumps to the vertex a
    click would lock onto, so you can aim before committing. MeasureTool gains handleMove(event),
    which initThree wires to canvas mousemove.
  • Orbiting/panning no longer disturbs a measurement: the click a drag fires on release is ignored
    (pointer moved past a small slop threshold), so in-progress points and finished measurements survive
    rotation instead of being cleared or mis-placed.
  • The default label now shows the per-axis breakdown (Δx/Δy/Δz) under the total distance. The
    format callback signature widens to (distance, delta) => string; existing (distance) => string
    callbacks remain valid.

• 9264ebb: Make the viewer's camera controller, presets, and grid respect the scene up axis.

  The CAD tooling assumed Three's native Y-up, but Selva scenes are Z-up. initThree set
  scene.up to Z yet never set camera.up, so OrbitControls orbited as if Y-up and the
  preset views (top/front/…) framed the wrong faces; the grid also defaulted to the
  horizontal Y-up plane.

  • initThree now sets camera.up to the configured sceneUp before constructing
    OrbitControls and the camera controller (both capture the orbit/preset basis from up).
  • The camera controller derives its preset view directions, iso angle, and orthographic
    camera up from the up axis instead of a hardcoded Y-up table, via a new optional up
    dependency (defaults to the perspective camera's up).
  • The grid's default plane is derived from the up axis (Z-up → plane: 'z'), so the grid
    lies under the model without callers passing plane explicitly. An explicit plane
    still wins.

  No API changes; behavior is corrected for non-Y-up scenes and unchanged for Y-up.