Struct ProducedElementMeshes 

pub struct ProducedElementMeshes {
    pub meshes: Vec<MeshData>,
    pub instance_occurrences: Vec<RawInstanceOccurrence>,
    pub geometry_hash: Option<u64>,
    pub geometry_aabb: Option<[f64; 6]>,
    pub geometry_volume: Option<f64>,
    pub geometry_closure: Option<GeometryClosure>,
    pub csg_failures: FxHashMap<u32, Vec<BoolFailure>>,
    pub degenerate_triangles_dropped: u64,
}

Everything one element produced.

Fields

meshes: Vec<MeshData>

instance_occurrences: Vec<RawInstanceOccurrence>

#1623 Phase 2 don’t-bake output: this element’s occurrences of a repeated IfcRepresentationMap that skipped the per-occurrence materialize. Empty unless the router was armed with an instancing plan (GeometryRouter::enable_output_instancing); the streaming finalize resolves each into a crate::InstanceRecord against the shared template MeshData.

geometry_hash: Option<u64>

Per-ELEMENT fingerprint, accumulated across all of the element’s meshes in the native IFC frame (pre-split, pre-site-rotation). None when hashing is off, nothing was produced, or the job is a TypeProduct.

geometry_aabb: Option<[f64; 6]>

The same pass’s world-space AABB, [minx, miny, minz, maxx, maxy, maxz] in unquantized f64 world coordinates (the file’s RTC folded back in), over every triangle corner the hasher saw. Some exactly when Self::geometry_hash is Some, so the two stay index-parallel at the FFI boundary.

Why the diff engine needs it: the hash conflates moved / reshaped / re-tessellated into one “different” bit. The box separates them — same extent at a new centre is a MOVE, a different extent is a reshape, an identical box with a different hash is retriangulation.

geometry_volume: Option<f64>

The element’s enclosed volume in m³ from the SAME pass — Some ONLY when the produced geometry was provably a single closed orientable solid, None otherwise (#1891). None is the common case for material-layered walls, open SurfaceModel geometry, and any element assembled from more than one representation item.

Read ifc_lite_geometry::GeometryHasher::volume before widening any clause of that gate: the alternative is not a slightly-off volume, it is a confidently wrong one with nothing about it that looks wrong.

geometry_closure: Option<GeometryClosure>

The folded per-segment topology verdict behind Self::geometry_volume — which clause held and which refused. Some exactly when Self::geometry_hash is. A model checker wants it: “open shell” and “multi-item assembly” are different findings with different fixes.

csg_failures: FxHashMap<u32, Vec<BoolFailure>>

CSG diagnostics recorded while producing THIS element, attributed by product id. The router is fully drained on return, so a warm router reused across a batch never leaks one element’s failures into the next. Failures from a superseded strategy (a fallback re-attempting the same cuts) are discarded — only the path that produced the returned meshes contributes.

degenerate_triangles_dropped: u64

Triangles dropped by the f32-collapse degenerate-triangle backstop (see the degenerate child module) across ALL of this element’s meshes. Zero when the backstop is disabled or nothing was degenerate. Request-local (scoped per produce_element_meshes call) so concurrent passes never cross-contaminate. Non-zero also RETRACTS Self::geometry_closure and Self::geometry_volume — the drop happens after the verdict was taken and can open a certified shell.