ReadonlyadapterReadonlydeviceReadonlyfallbackResolves the object under a client-space pointer position together with
the world-space surface point it hit, or point: null over the
background. The position is rendered camera-relative into a transient
rgba32float attachment that lives only for this call, so the persistent
attachment pair reported by resourceStats() does not change, and the
float64 camera origin is added back on the CPU so the result keeps
millimetre precision at any distance from the world origin.
Reconciles application-keyed batches without re-uploading untouched GPU resources. This is the residency boundary used by progressive loaders.
A batch is validated when it is first uploaded, not on every reconcile: retained batches are treated as immutable, so admission cost scales with the batches that actually change rather than the whole resident set. All validation runs before any GPU resource is destroyed or created.
Byte census of renderer-owned resources. Vertex pools are walked from the
pool map rather than from the batches that reference them: several material
groups hold the same GPUBuffer, so a per-batch sum reports allocations
that were never made.
It does not include caller-owned decoded typed arrays; those are charged by the residency set that retains them.
Enables one world-space section plane. Passing undefined disables clipping.
Optionalplane: SectionPlaneSelects an occurrence for surface and explicit-edge highlighting. Zero clears selection.
Re-packs visible occurrences from dense per-prototype index tables. When
changedBatchIndexes is provided, only those batches are re-packed and
re-uploaded; every other batch keeps its current GPU packing and count.
OptionalchangedBatchIndexes: readonly number[]Staticcreate
Estimated buffer allocation currently retained by this renderer.