Map a parsed glTF/GLB Document onto automovie's core node graph and animation clips, with no three.js and no humanoid assumptions.
Current repository callers exercise this parsed-document mapper through tests. The production asset inventory uses the separate byte inspector and host loaders; this export does not establish a wired parsed-document runtime import flow.
@gltf-transform/core parses the bytes headlessly, and this mapper rewrites its structures structurally so props, cameras and characters import the same way. Every node keeps its local TRS, its parent, and the mesh, camera and skin it carries; a skin joint is a bone even when it also carries a mesh, then a mesh node is mesh, a camera node camera, and anything else group. Lights are not read. Every animation channel becomes a node track with the glTF sampler's keyframes, its interpolation mapped LINEAR, STEP and CUBICSPLINE to linear, step and cubicspline with an absent one read as LINEAR as glTF specifies, and its values decoded from normalized integers where the accessor declares them. Naming-based humanoid skeleton mapping is the separate humanoidSkeleton stage.
Channel checks refuse no target node, no or unknown target path, no sampler, no input or output data, an unknown interpolation, keyframe times that are not finite, non-negative and strictly increasing, non-finite output values, or an output length that disagrees with the keyframe count, the path's component count and the cubic-spline tangent triple. This is not complete model admission: node TRS, graph/resource membership, duplicate target channels, rotation magnitude and weights against the target morph count are caller premises rather than checks performed here.
Map a parsed glTF/GLB Document onto automovie's core node graph and animation clips, with no three.js and no humanoid assumptions.
Current repository callers exercise this parsed-document mapper through tests. The production asset inventory uses the separate byte inspector and host loaders; this export does not establish a wired parsed-document runtime import flow.
@gltf-transform/coreparses the bytes headlessly, and this mapper rewrites its structures structurally so props, cameras and characters import the same way. Every node keeps its local TRS, its parent, and the mesh, camera and skin it carries; a skin joint is aboneeven when it also carries a mesh, then a mesh node ismesh, a camera nodecamera, and anything elsegroup. Lights are not read. Every animation channel becomes a node track with the glTF sampler's keyframes, its interpolation mapped LINEAR, STEP and CUBICSPLINE tolinear,stepandcubicsplinewith an absent one read as LINEAR as glTF specifies, and its values decoded from normalized integers where the accessor declares them. Naming-based humanoid skeleton mapping is the separate humanoidSkeleton stage.Channel checks refuse no target node, no or unknown target path, no sampler, no input or output data, an unknown interpolation, keyframe times that are not finite, non-negative and strictly increasing, non-finite output values, or an output length that disagrees with the keyframe count, the path's component count and the cubic-spline tangent triple. This is not complete model admission: node TRS, graph/resource membership, duplicate target channels, rotation magnitude and weights against the target morph count are caller premises rather than checks performed here.