[embed-3d]

Wires the PrimitiveAccess defaults that were still UNPORTED stubs, although their targets are ported, and makes the difftest fixtures reach them.

Changes

• B-rep face services: planar_faces / planar_face / surface_faces / surface_face (topology.brep_p8: CreatePlanarFace(s)With3DCurves, CreateFace(s)FromSurface with SM_CT_G1 as at every call site) and delete_primitive_face (topology.brep_p3: DeleteFace(face, FALSE, TRUE)) are now real trait defaults over SweepAccess::with_brep_mut. PrimitiveModel's identical overrides are removed.
• Axis classification: axis_inside_face is a trait default over the classifier SweepStore already used (SmFace::CurveOnClassify). That removes SweepStore's override and PrimitiveModel's Unported(query.face_p2).
• Consumed curve slots: planar_faces now empties each consumed slot instead of clearing the vector, as the trait documents. No current caller depended on the old behaviour (skin.rs sets its transfer flags in the Ok branch).
• Difftest fixtures (p2, p3):
  • A tag store and untagged curve lookups, as in PrimitiveModel.
  • The p3 fixture deletes faces through KernelManifoldIntersector::delete_faces_with_keys, as in SweepStore. It used to fail there silently, which the ignored upstream status hid.
  • The p3 entry-point records (shim and runner) now also compare the B-rep's face, edge and vertex counts, so the construction tails compare real topology. For example, the repeated linear wire sweep gives 2/7/6 and the full rotation of a line 1/3/2, both identical to C++. Face geometry is still not compared.
• Tests:
  • store.rs: the test that pinned the brep_p8 stub now checks that the real constructor runs.
  • api_primitives: partial solids now get past the axis check. Sphere and cylinder stop at SmFace::FindVertexSector (topology.face_p4), the torus at SmFace::Point3DClassify (query.face_p1).
• Plan: primitive.primitive_creation_p3 → green; p2 notes updated.

Difftest (macOS arm64, Apple clang, pinned c9c5979; Rust side merged with main 10a92977e)

PASS primitive.primitive_creation_p3 cases=296 passed=295 failed=0 deferred=0 max_rel_err=0e0 with_values=295 rejected=1
PARTIAL primitive.primitive_creation_p2 cases=328 passed=196 failed=0 deferred=112 max_rel_err=0e0 with_values=196 rejected=20
PARTIAL sweep.api_brep cases=573 passed=502 failed=0 deferred=71 max_rel_err=0e0
PARTIAL primitive.primitive_creation_p1 cases=773 passed=738 failed=0 deferred=26 rejected=9
PARTIAL primitive.api_primitives cases=1981 passed=1613 failed=0 deferred=368
PARTIAL sweep.api_surfaces cases=882 passed=862 failed=0 deferred=20
PASS sweep.topology_sweep cases=568 passed=568 failed=0 deferred=0
PARTIAL boolean.api_brep cases=287 passed=151 failed=0 deferred=136
PARTIAL topology.brep_p3 cases=897 passed=812 failed=0 deferred=85

Compared with main's CI reports:

• p3: 286 → 295 passes, 9 → 0 deferred.
• p2: 146 → 196 passes, 162 → 112 deferred (topology.brep_p8 71 → 0).
• sweep.api_brep: 501 → 502 passes.
• The other touched units are unchanged.

Full --green --jobs 4 on this Mac: every failing unit fails identically on unmodified main, from known macOS/arm64 issues (#448, #449, #450). I checked heal.heal_data_p1 case by case against main. blend.fillet_solver_p5 and geom.local_solve1d also fail identically on main locally. No absolute floors and no loosened tolerances. Run on macOS only, so CI is the Linux gate.

Also run: cargo fmt --all --check, cargo clippy --workspace --all-targets -- -D warnings, cargo test -p solidcore (2053 passed) and cargo check -p solidcore --target wasm32-unknown-unknown.

Related

• wire: primitive and sweep APIs onto ported primitive_creation #536 introduced PrimitiveModel, whose implementations this lifts into the trait.
• fix: prepare face geometry for shell ray queries #613 (draft, shell ray queries) edits the same api_primitives test. The scopes are independent; whichever lands second has a small text conflict in the partial-solids assertion block.

🤖 Generated with Claude Code

https://claude.ai/code/session_01TYEpC4tUKxEP3MCaFL5Kz2