[embed-3d]

Follow-up to #617. The primitive.primitive_creation_p2 runner used a one-B-rep fixture adapter with two problems. First, it could not make the scratch and piece B-reps that skin, taper and planar path need: an UNPORTED(topology.brep_p3) SmBrep::SmBrep stub in the runner, although the constructor is ported. Second, it keyed new faces in its own namespace. The runner now runs every entry point through the production adapter primitive::primitive_creation::model::PrimitiveModel. It also removes ~230 lines of fixture adapter. Library code is unchanged.

Why some former passes now defer

The fixture's face keys (FACE_TYPE, 100000 + 97·i) cannot be resolved to a face by the B-rep caches of the shell ray fire (StitchAndOrient → region split → FindInfiniteRegion → SmShell::RayFire). Those caches therefore solved the faces' surfaces as untrimmed, which is the shortcut SurfaceCache::get_tree refuses for a face-owned surface.

12 pipe, skin and planar-path cases matched C++ only because the record compares status and list sizes, not geometry. To confirm, I changed nothing in the old fixture except its face keys, using the production notify_ref formula. Exactly those 12 then defer at TrimSrfCache::build_tree_with_face for a face-owned surface (passes 196 → 184, surface.trim_srf_cache 6 → 18). With the production adapter they reach the trimmed cache, as C++ does, and defer honestly. Details are in plan/findings.md.

Difftest (macOS arm64, pinned c9c5979, on main 7935b4c09)

before: PARTIAL primitive.primitive_creation_p2 cases=328 passed=196 failed=0 deferred=112 rejected=20 blocked_by=...topology.brep_p3(81)...
after:  PARTIAL primitive.primitive_creation_p2 cases=328 passed=194 failed=0 deferred=114 max_rel_err=0e0 with_values=194 no_values=0 vacuous=0 rejected=20 blocked_by=boolean.merge_p2(34),curve.curve_p2(7),query.face_p1(3),surface.trim_srf_cache(51),topology.face_p2(5),topology.face_p4(14)

Case-by-case against main's CI report:

• 10 cases that waited for SmBrep::SmBrep now pass.
• The 12 cases above now defer.
• No case fails.

The other 71 former SmBrep::SmBrep cases now reach deeper unported calls. The fixture README's deferred table is updated.

Checks: cargo fmt --all --check, cargo clippy --workspace --all-targets -- -D warnings, the same with -p difftest --features cpp, cargo test -p solidcore (2054 passed), cargo test -p difftest and wasm check.

Follow-up, not in this PR

PrimitiveModel::new_merge still returns Unported(SmMerge::SmMerge, boolean.merge_p2) although the unit is ported; that accounts for the 34 taper deferrals. I left it alone because #615 is reworking model.rs.

Related

• fix: preserve rotational surface geometry and production cleanup #615 (draft) moves the p3 runner to the same production adapter. No overlapping files.
• wire: resolve edge/curve/vertex handles in solver objects #616 (draft) changes solver handles and moves some p2 deferrals from curve.curve_p2 to their next blocker. No overlapping files.

🤖 Generated with Claude Code

https://claude.ai/code/session_01TYEpC4tUKxEP3MCaFL5Kz2