Struct StreamShapePlanner 

pub struct StreamShapePlanner {
    layout: ParquetLayout,
    prior: Option<PriorShapes>,
    retained_bytes: usize,
    template_budget: usize,
    vertex_offset: u32,
    index_offset: u32,
}

The running state behind a streamed model’s whole-stream offsets.

Fields

layout: ParquetLayout

prior: Option<PriorShapes>

None: every mesh emits its own block, as the stream always did. Some: cross-batch sharing, against every shape emitted so far.

retained_bytes: usize

Raw buffer bytes held by prior’s retained templates, and their cap.

template_budget: usize

vertex_offset: u32

index_offset: u32

Implementations

impl StreamShapePlanner

pub fn new(layout: ParquetLayout, stream_shapes: StreamShapes) -> Self

A planner for layout. Sharing needs BOTH the shared-shapes layout and the client’s cross-batch opt-in; the route rejects the opt-in on the flat layout, so ignoring it here is never what a caller sees.

pub fn plan( &mut self, meshes: &[MeshData], baked_basis: Option<&[f64; 16]>, ) -> Result<PlannedBatch, ParquetError>

Plan the next batch and advance the whole-stream offsets past the shapes it emits. Fails loud, advancing nothing, when an offset would overflow the mesh table’s u32 columns: wrapping would decode as garbage, not as an error.

baked_basis is the row-major native_to_baked of the frame the batch’s vertices are in (StreamEvent::Batch::baked_basis); ignored unless the planner shares.