[tdamsma]

Machine-generated (Claude Fable 5.1 coordinator, Claude Opus 5.5 agent) on behalf of @tdamsma.

Found

• The query.face_p1 deferrals came from four DISPATCH GAP stubs. There was no runner-side producer:
  • intersect::curve_class::p7 (ClassifyFaceInterval: EdgeCurveOnClassify and Point3DClassify)
  • intersect::topology_solver::p1 (SmTLSPointFace::SolveIt: Point3DClassify)
  • topology::brep::p9 (inner-loop transfer: Point3DClassify)
  • topology::face::p3 (FaceIntersect touching points). Left alone: it is a FaceIntersect bridge.
• Every remaining query.face_p1 deferral goes through query::face::p2::curve_on_classify. That code wraps the caller's objects in WithMadeTrimCurves, which does not forward topology_entity, so the face handle cannot be resolved. p2.rs is Codex-owned and not edited here. The forwarding already exists on codex/primitive-p2-production (4526f4aa0).
• Once that forwarding is in, newly reachable code shows disagreements outside query.face_p1. Details are in plan/findings.md:
  • heal.face_props has 8 failures and intersect.curve_class_p4 has 2. Rust classifies a seam crossing at an edge end as EDGE (t=1.0); C++ classifies it as VERTEX.
  • topology.brep_p3 hit an empty BrepCache. This is fixed here.

Done

• p7: face_edge_curve_on_classify / face_point_3d_classify resolve the face (and edge) through ClassGeometry::topology_entity.
  • They call query::face::p1 on a copy of that B-rep, because the upstream const members may add UV trim curves. The C++ argument list is kept.
  • The classification is mapped back with topology_handle_in. An edge from another B-rep gives "not inside", as upstream.
  • An unresolved handle keeps the query.face_p1 deferral.
• topology_solver p1: the Point3DClassify retry pair (tol, then 100*tol, boundary intersections on) runs on a copy of the B-rep that TopologyObjects::face_entity returns.
• brep p9: point_3d_classify takes the caller's &mut Brep (guess UV, outer loop only).
• brep p3: brep_curve_solve builds its cache with make_brep_cache (CacheMakeOrValidate), so the solve has trees.
• plan: dated notes for query.brep, topology.brep_p8 and sweep.api_surfaces. A findings entry covers the blockers and the latent disagreements.

Result (Mac Studio runs; Linux CI is the gate). Before = main reports (CI 39939270570); after = this branch.

• query.brep: before partial 1172/1089 pass, 83 deferred (query.face_p1 29) -> PARTIAL query.brep cases=1172 passed=1115 failed=0 deferred=57 max_rel_err=0e0 with_values=1115 no_values=0 vacuous=0 blocked_by=curve.curve_p2(9),topology.edgeuse_p3(2),topology.face_p2(46)
• topology.brep_p8: before partial 339/336 pass, 3 deferred (query.face_p1 2) -> PARTIAL topology.brep_p8 cases=339 passed=338 failed=0 deferred=1 max_rel_err=0e0 with_values=338 no_values=0 vacuous=0 blocked_by=curve.curve_p2(1)
• sweep.api_surfaces: before 882/862 pass, 20 deferred (query.face_p1 1) -> PARTIAL sweep.api_surfaces cases=882 passed=862 failed=0 deferred=20 max_rel_err=0e0 with_values=862 no_values=0 vacuous=0 blocked_by=boolean.topology_intersector_p2(17),surface.trim_srf_cache(3) (the deferral moves on)
• Unchanged (blocked by WithMadeTrimCurves):
  • PARTIAL intersect.curve_class_p4 cases=3201 passed=2918 failed=0 deferred=283 max_rel_err=0e0 with_values=2918 no_values=0 vacuous=0 blocked_by=query.face_p1(97),topology.face_p4(186)
  • PARTIAL topology.brep_p3 cases=897 passed=812 failed=0 deferred=85 max_rel_err=0e0 with_values=812 no_values=0 vacuous=0 blocked_by=query.face_p1(38),topology.face_p3(47)
  • PARTIAL heal.face_props cases=1528 passed=1431 failed=0 deferred=97 max_rel_err=0e0 with_values=1422 no_values=9 vacuous=0 blocked_by=query.face_p1(27),topology.edge(1),topology.face_p4(69)
  • PARTIAL topology.face_p2 cases=2692 passed=2532 failed=0 deferred=160 max_rel_err=0e0 with_values=2179 no_values=353 vacuous=0 blocked_by=query.face_p1(19),query.face_p2(30),topology.edge(106),topology.face_p4(3),topology.vertex(2)
  • PARTIAL query.face_p2 cases=1344 passed=1326 failed=0 deferred=18 max_rel_err=0e0 with_values=1299 no_values=27 vacuous=0 blocked_by=query.face_p1(4),topology.edge(14)
  • PARTIAL topology.brep_p1 cases=2627 passed=2461 failed=0 deferred=150 max_rel_err=0e0 with_values=2416 no_values=45 vacuous=0 rejected=16 blocked_by=nurbs.b_spline_surface_n_lib_p2(8),query.face_p1(1),topology.edge(14),topology.face_p2(109),topology.face_p3(18)
  • PARTIAL query.face_p4 cases=500 passed=395 failed=0 deferred=105 max_rel_err=0e0 with_values=368 no_values=27 vacuous=0 blocked_by=query.face_p1(1),topology.edge(104)
• Touched-source units:
  • PASS query.face_p1 cases=2266 passed=2266 failed=0 deferred=0 max_rel_err=0e0 with_values=2266 no_values=0 vacuous=0
  • PARTIAL intersect.topology_solver_p1 cases=3532 passed=3434 failed=0 deferred=98 max_rel_err=0e0 with_values=3434 no_values=0 vacuous=0 blocked_by=topology.edge(98)
  • PARTIAL intersect.topology_solver_p2 cases=2429 passed=2396 failed=0 deferred=33 max_rel_err=0e0 with_values=2396 no_values=0 vacuous=0 blocked_by=geom.tol(2),query.face_p2(8),surface.surface_p2(12),topology.edge(3),topology.face_p2(8)
  • PARTIAL topology.brep_p9 cases=684 passed=683 failed=0 deferred=1 max_rel_err=0e0 with_values=683 no_values=0 vacuous=0 blocked_by=topology.face_p2(1)
  • PARTIAL intersect.curve_class_p7 cases=1516 passed=1515 failed=0 deferred=1 max_rel_err=0e0 with_values=1515 no_values=0 vacuous=0 blocked_by=topology.vertex(1)
• Checks (Mac, after merging origin/main): fmt, clippy (with and without difftest/cpp), cargo test --workspace and the wasm32 check all pass.
• Skipped:
  • blend.fillet_solver_p5 (3 query.face_p1 deferrals) was not verified. It FAILs 100/100 on the Mac at main too (cpp_exception on both sides), so Linux CI decides.
  • No unit reaches green.

🤖 Generated with Claude Code

https://claude.ai/code/session_01RvuXmr7S5RoUXxVohC1giE