Module parquet_schema 

The geometry Parquet WIRE SCHEMA — the three tables the standard transport emits, in one place.

Split out of parquet.rs so the contract each client decodes against is readable on its own (and so the serializer stays under its module-size budget). Any column added or removed here is a wire-format change: keep it additive and backward-compatible, and pin it with a test in parquet_tests.rs — silently dropping a column is exactly how the per-mesh origin went missing in issue #1841.

Structs

MeshRow 🔒

One mesh’s row across the three tables, in the order mesh_schema() lists.

RowPlacement 🔒

Where one mesh row’s geometry lives and how it is placed. Separate from the mesh because on the shared-shapes layout the two come apart: the ranges belong to a shape the occurrence may only borrow, possibly one an earlier stream batch wrote (#5407), which is why the counts travel here rather than being read off a MeshData the row may no longer have.

Constants

ABSENT_SOURCE_ID 🔒

Absent marker for the two source ids. Mirrors ABSENT_SOURCE_ID in packages/cache/src/sections/geometry.ts.

Functions

index_schema 🔒

instance_schema 🔒

The optimized transport’s instance table.

mesh_schema 🔒

The flat transport’s mesh table.

rotation_fields 🔒

The nine row-major rotation columns appended to instance_schema() and, as of -parquet-v6 (#3888), to mesh_schema() too (see either doc comment for the coordinate-frame contract).

shared_trailing_fields 🔒

The trailing per-mesh columns both transports carry, in order.

vertex_schema 🔒