[bluecmd]

Ports unit topology.brep_p2: the 23 SmBrep functions it owns in source/SMLib/src/SmBrep.cpp, as solidcore::topology::brep::p2.

What is in it

• Queries: GetEdgeuses, GetShellVertices, IsManifoldSolid, IsSheet, IsOrientedSheet, GetMinEdgeBBoxSize.
• Tolerance and trim curves: SetTolerance (SM_USE_OLDTOL), RebuildUVTrimCurves, sm_CheckAndFixUVTrimCurves, SwapUV, DeleteStaleCaches.
• Notifications: Notify, sm_AddToSpatialTree, sm_RmFromSpatialTree. The SM_OC_BREP cache is held by the caller and passed to notify as a BrepCacheSlot.
• Topology construction: MakeShellVertexTopology, MakeEdgeVertexTopology, MakeEdgeTopology.
• Geometry and classification: ReplaceSurface, LocalCurveSector, RayFire, sm_GetPointFromSol, MergeCurvesOnSurface, MateLaminaEdgeToSurface.
• Back-edges wired: SmBrep::Notify in topology::edge (topology.edge's deferrals on brep_p2 5 → 0) and in topology.brep_p6's TestClosureSplitRegion (PRE_EDIT, SPLIT_IN_BREP(old, old, new), statuses ignored, SmBrep.cpp:27954-5; the split API and heal_data_p1's fix_bad_sheets take the caller's NotifyEnv, see below), SetTolerance in heal.heal_data_p2 (and through it heal.heal_data_p1 / heal.heal_data_p3) and topology.trimming_tools, IsManifoldSolid in mass.integrator and query.api_queries. Passes against main: query.api_queries 648 → 773, heal.heal_data_p2 347 → 364, heal.heal_data_p3 267 → 314, heal.heal_data_p1 265 → 266; none of them defers on topology.brep_p2 any more.
• Also wired: query::edge's stale EvaluateBinormalStepOff stub now calls topology::edgeuse::p2::evaluate_binormal_step_off_into (an in-place variant that keeps upstream's partially written out-params on failure). AreCoincidentMaybe in MergeCurvesOnSurface now calls intersect.curve_class_p7 (port: intersect.curve_class_p7 #65). ProtoTopologyManager::new's tolerance fallback (SmProtoTopology.cpp:3548-3556) now calls Brep::tolerance and brep::p2::min_edge_bbox_size: new() takes the kernel and each B-rep as (handle, &Brep), the proto_topology_p1 and query.face_p2 runners pass two empty Breps as their shims build two empty SmBreps, and 5 new proto cases (approx_tol 0, -1e-3, -1, 1e-13 and SM_EFF_ZERO) compare the branch. With the fallback disabled, the four proto cases below SM_EFF_ZERO fail. Only the tolerance half is compared against C++ in proto_topology_p1 (two empty B-reps never let the edge-size half win); since the main merge with port: boolean.topology_intersector_p1 #146, boolean.topology_intersector_p1's MgrNew op passes its two real runner B-reps too, and MgrNew now outputs the manager's GetTolerance, so the fallback value is compared: with tol = 3*min(..) mutated, 20 intersector p1 cases fail.
• topology.face_p1 (port: topology.face_p1 #148, merged) wired: ReplaceSurface's SmFace::SetSurface (default bDeleteCurrFace FALSE) and RemoveUVTrimCurves call topology::face::p1::set_surface / remove_uv_trim_curves, and face_p1's MoveSeam calls brep::p2::replace_surface exactly as SmFace.cpp:17602 (ReplaceSurface(this, pJoinedSurface, FALSE, NULL, FALSE)). Every ReplaceSurface case still defers first at SmFace::Notify (topology.face_p3).
• topology.shell (port: topology.shell #154) and topology.edgeuse_p3 (port: topology.edgeuse_p3 #117) wired: the 8 own shell sites of the Make*Topology functions (region shell-list PostInsert / Remove with its status ignored, IsVertexShell, the vertex shell's vertex use, the vertex / wire shell type and head, SER(TransferChildrenTo)), as SmBrep.cpp:18714 / 13799 / 13641; and the 3 MateLaminaEdgeToSurface sites (GetOrCreateUVTrimCurve copy and trim, NormalizedEvaluate(t, FALSE), header defaults). make_edge_topology takes the kernel; local_curve_sector / ray_fire take the B-rep mutably (port: topology.edgeuse_p3 #117's sectoring may make UV trim curves).
• Make ops compare topology: the shim and runner output, after the status, the B-rep read back through fresh list queries (per region its shells as kind and vertex / first wire edge, per vertex its vertex-use count, per edge the vertices and shells of its primary edge use and that use's mate): 17 to 86 values per case instead of a status alone.
• Back-edges into this unit after the merges (overseer rulings): topology.brep_p6's split takes env: &mut NotifyEnv<'_> (as do heal_data_p1's fix_bad_sheets); heal_data_p3's FixBadsheets step and the runners pass brep::p2::RegionNotifyState (an empty AttributeStore and default region records). The heal path carries no attribute state at any level, so SmSAGObject attribute propagation during healing is not modelled; this predates the PR and is recorded in plan/findings.md. topology.shell's ShellRayWorld::shell_ray_sector default is retagged UNPORTED(query.brep) (the B-rep-cache world overrides it with brep::p2::local_curve_sector).
• Shared FFI reader (ffi: share the validate Reader through solidcore_ffi::reader #186, lesson 30): topology_vertexuse's FFI Reader is unit_reader!(MAX_COORD) and private again; the brep_p2 wrapper uses its own and reads blocks through vu::surface_at / curve_at / brep_block_at.

Builds on #145 (topology.brep_p1, head 5b1b28ae, which includes main through e1ae2640), merged in as-is; main is merged up to ca3f8c29. Calls into units that are not ported yet go through UNPORTED stubs: topology.face_p3 (#139), query.brep, query.face_p1 and intersect.curve_class_p2/p4. MergeIntervals is wired to intersect.curve_class_p5. Nothing owned by another open PR is co-ported.

SmBrep::Notify end to end (#210)

brep::p2::notify ports every arm of SmBrep::Notify (SmBrep.cpp) with all its arguments: PreEdit / Destruction delete the SM_OC_BREP cache, MergeInBrep removes d2 from the spatial tree, RmFromBrep removes d1, SplitInBrep adds d2 and d3 (when they differ from d1), AddToBrep adds d1, every other type is a no-op, and then SmSAGObject::Notify runs. With no cache slot the spatial-tree helpers return before any stub, so the no-cache path (every current Rust caller) has none. With a cache, the face branch stops at UNPORTED(query.face_p1) SmFace::CalculateBoundingBox and the cache rebuild at UNPORTED(query.brep); the code after those calls is ported. The 7 remaining UNPORTED stub sites in brep/p2.rs target todo units (topology.face_p3 3, query.brep 1, query.face_p1 1, intersect.curve_class_p2 1, intersect.curve_class_p4 1).

Review 5578349117 round (head 167dd352, main merged through 39ceb995)

• All 33 UNPORTED(topology.brep_p2) sites main had are wired (685608d2), each with upstream's status handling.
• p2.rs's own stubs into topology.face_p3 and the intersect.curve_class_p4 constructor are wired (02b28cd1).
• The mismatches this exposed are fixed (660b3270, da779c51).
• Every touched unit was re-run (863c5810): 229 units, 0 fail. The per-unit lines and the mapping to each must-fix are in the review reply. The other unit numbers in this description below predate this round; the review reply has the current ones.
• Deviation (net): 3 offset.offset_geometry_creation cases that pass on main now defer to ANALYTIC_EXPORT: gen0119 and gen0707 on an SmSurfOfExtrusion, gen0376 on an SmSphere. On main they passed only because a brep_p2 stub stopped them before LiftCurve; past it, the C++ runs the subclass's iso-curve override, and the Rust SurfaceKind has no extrusion or sphere variant to import those faces as (issue Coordinate Kernel-aware Surface dispatch for analytic surface kinds #76 tracks that gap).
  • ed40f92c imports SmSurfOfRevolution faces exactly: the shim exports their binary WriteToDB record and the runner reads it back with SurfOfRevolution::read_from_db. The other 3 of the original 6 (gen0009, gen0148, gen0520) pass again, as does gen0683.
  • offset.offset_geometry_creation is now 861 cases, 552 pass (was 542), 0 fail, 309 deferred: 10 harness deferrals pass, 9 move to intersect.surface_drop_curve. --perturb SENSITIVE 552/552.
• Known fuzz timeout (not a defect in this unit; no guard): offset.offset_geometry_creation --fuzz 2000 has 0 differs, crashes or panics and 1 timeout, fuzz00504. Its trim_extended_edge takes C++ about 2 s and Rust 18-24 s, against a 20 s budget. The time goes into rebuilding curve.curve_p5's curve cache on every call under global_curve_intersect, a documented deviation of that unit; this PR's wiring made the path reachable. It is recorded in plan/findings.md under curve.curve_p5 (3a6ea1df) and in COORD.md, and left to that unit's owners.
• Fuzz differs (a note, not a failure): topology.brep_p2 --fuzz 2000 is FUZZ-PASS with differs=2. fuzz00488 and fuzz00815 are Mate with the plane's origin at ±9.2e18: status 1001 in C++ vs 1006 in Rust. The cause is in the plane's global_point_solve (surface_p4 / plane_p1), not brep_p2 logic. Both inputs are in plan/findings.md.
• batch: land 5 reviewed-clean contributor PRs #283 merge round (a37776b1..167dd352): main's 5 new UNPORTED(topology.brep_p2) sites are wired, and merge_p4's ArenaPair implements brep_arena / BrepGlue. merge_p3's arena-only boolean_lists notifies its component copy through an arena-only store; this reverses port: boolean.merge_p3 #245's stated design (see the reply). Pre-review fixes are in 63997139 (region notification keys, p5 Deviations). The gate on 167dd352 has all 232 units at 0 fail; query.api_queries gains 42 passes and topology.brep_p7 gains 40.
• Main's own breakages: offset_offset_executive_p1::each_highlevel_probe_calls_its_real_method and intersect_curve_class_p4::undefined_scripts_are_rejected fail on clean main too. They are not touched here and are listed in COORD.md.

Difftest

PARTIAL topology.brep_p2 cases=1709 passed=1437 failed=0 deferred=272 max_rel_err=0e0 with_values=1437 no_values=0 blocked_by=intersect.curve_class_p2(1),intersect.surface_drop_lib_p3(85),query.brep(93),surface.surface_cache_p1(20),topology.face_p2(19),topology.face_p3(16),topology.face_p4(38)
SENSITIVE topology.brep_p2 perturbed=1367 detected=1367 unperturbed=0 deferred=342
FUZZ-PASS topology.brep_p2 cases=2000 rejected=442 malformed=767 ran=791 differs=0 crashes=0 panics=0 timeouts=0

PARTIAL topology.brep_p7 cases=316 passed=197 failed=0 deferred=119 (the 51 topology.brep_p2 deferrals: face_p3 43, brep_p1 7, brep_p4 1)
PARTIAL topology.brep_p10 cases=1404 passed=851 failed=0 deferred=553 (12 IsManifoldSolid deferrals pass, 2 -> topology.face_p2)
PARTIAL heal.stitch_p1 cases=300 passed=206 failed=0 deferred=94 (unchanged)

PARTIAL topology.brep_p6 cases=498 passed=450 failed=0 deferred=48 (the 6 former SmBrep::Notify deferrals relabel to query.brep); SENSITIVE 450/450
PARTIAL heal.heal_data_p1 cases=493 passed=271 failed=0 deferred=222
PARTIAL heal.heal_data_p2 cases=738 passed=469 failed=0 deferred=269; SENSITIVE 469/469
PARTIAL heal.heal_data_p3 cases=474 passed=329 failed=0 deferred=145; SENSITIVE 329/329
PARTIAL boolean.topology_intersector_p1 cases=827 passed=710 failed=0 deferred=117 with_values=707 no_values=3; SENSITIVE 710/710
PARTIAL topology.edge cases=1787 passed=1781 failed=0 deferred=6; SENSITIVE 1781/1781
PARTIAL mass.integrator cases=1528 passed=1521 failed=0 deferred=7
PARTIAL topology.shell cases=911 passed=892 failed=0 deferred=19 (unchanged)
PARTIAL topology.vertexuse cases=2686 passed=2250 failed=0 deferred=436 (unchanged)
PARTIAL query.edge cases=1368 passed=1316 failed=0 deferred=52
PARTIAL topology.face_p1 cases=2312 passed=807 failed=0 deferred=1505
PARTIAL topology.proto_topology_p1 cases=1862 passed=1770 failed=0 deferred=92
PARTIAL query.api_queries cases=1203 passed=773 failed=0 deferred=430

sc-remote green: 208 units, failed=0
valgrind (--error-exitcode=99, harness/valgrind/ub11.supp): 74 sampled cases on the new paths (Make, IsSheet, Sector, Mate, SwapUV, Rebuild, the brep_p6 split, heal_data_p2, intersector MgrNew, topology.edge notify): 0 errors

Flips against the pre-merge round (7311346): 228 EvaluateBinormal → pass (Sector), 66 GetShells → pass (IsSheet / IsOrientedSheet), 76 shell stubs → pass (Make, compared topology), 74 → intersect.surface_drop_lib_p3, 38 SwapUV → topology.face_p4, 7 → topology.face_p2; 0 new fails or hangs. Of the 52 orientation-0 (SM_OT_UNKNOWN) sector fixtures 46 pass now.

The shim (harness/cpp/shim/topology_brep_p2.cpp) builds real SmBreps with the shared topology fixture: 15 upstream .smb files, a wire edge, a 10x10 plane face, and the my_test_topology wire network. It replaces analytic geometry by plain B-splines so the Rust import is exact, and registers the B-rep face, edge and vertex lists first, so the import order is upstream's list order. 1309 fixture cases (upstream test calls: SetTolerance smTopoTest.cpp:337 and :696, smMergeTest.cpp:1728; IsManifoldSolid smTopoTest.cpp:449; MergeCurvesOnSurface smTopoTest.cpp:1664) plus 400 generated cases with tiny and huge scales. The two file statics, which the shim cannot call, are covered by Rust unit tests.

Review of 0258c83: back-edges wired (at 8fdfa2c8)

No UNPORTED(topology.brep_p2) remains in the crate.

• topology.brep_p7 (port: topology.brep_p7 #155): brep_notify → brep_p2::notify through glue_edges_general's env (PRE_EDIT with this, MERGE_IN_BREP (surv, del, this)), brep_set_tolerance → set_tolerance, check_and_fix_uv_trim_curves and rebuild_uv_trim_curves → p2; statuses ignored as SmBrep.cpp:9840/9854/2014.
• heal.stitch_p1 (port: heal.stitch_p1 #196): brep_notify_pre_edit → brep_p2::notify(.., PreEdit, [this, None, None]) on an empty notify state (no B-rep cache, no attribute state on the heal path).
• topology.brep_p10 (port: topology.brep_p10 #130): is_manifold_solid (m_bManifold = IsManifoldSolid()), point_from_solution, shell_vertices(b, None), and merge_curves_on_surface, which now calls brep_p2::merge_curves_on_surface as SmBrep.cpp:25074. Its classification world over the B-rep cache is uninhabited (CacheMergeWorld), and its constructor is the UNPORTED(query.brep) stub: query.brep (todo) is the real blocker, as for topology.shell's shell_ray_sector.
• brep_p2::connect_intervals → CurveClassification::connect_intervals(index, pos, false) (SmBrep.cpp:22072, SER); dropped from merge_curves_on_surface's # Errors doc.

Unit tests changed in other units' code (the back-edges make new paths reachable):

• topology/brep/p10.rs (port: topology.brep_p10 #130): healing_runs_and_manifold_is_unported is now healing_runs_and_manifold_is_computed; an UNSURE manifold flag is computed (0 for a sheet) instead of deferring.
• topology/brep/p7.rs (port: topology.brep_p7 #155): glue_no_ops gluing two bare edges (no edge uses) now reaches InvalidInput("edge without primary edge use") past the notifications; read_errors_and_deferral reads an empty file successfully (sm_CheckAndFixUVTrimCurves has nothing to fix).

Also from the follow-up branch (your nits on 87c57c02): the runner's Notify env (type-distinct keys, a B-rep record, 68 fixtures for all 17 operations), the ReplaceSurface pTopFace Deviations line, and the Sector guard refusing only t0 == t1 (12 inverted-interval fixtures).

Coverage, stated plainly

with_values=1367 no_values=0: every pass compares at least one value; the 330 below compare only a status code or constant output, which the summary counts as a value.

• Passes with numeric content: 1037 of 1367. The queries, IsSheet / IsOrientedSheet, GetMinEdgeBBoxSize, SetTolerance, LocalCurveSector, DeleteStaleCaches and the Make*Topology topology summaries.
• Passes without numeric content: 330. Notify (230, including the 68 that cover all 17 operations with the B-rep as a data object: the constant 0, neither side has a B-rep cache), MateLaminaEdgeToSurface (91: the early SM_ERR, including 3 noedit cases with the mating plane at ±1e300 / 1e160), MergeCurvesOnSurface (3: a status), SwapUV (2: face-less B-reps), RebuildUVTrimCurves (4: no UV curves).
• No compared values yet: RayFire (query.brep's B-rep box), ReplaceSurface (face_p3), and MateLaminaEdgeToSurface past its early returns (it defers at SmFace::Notify before its wired edgeuse_p3 sites).
• Normalised geometry: analytic face and edge geometry other than SmPlane / SmLine is turned into plain B-splines before each op; none of this unit's own branches depend on the kinds.
• B-rep cache: no Rust caller can build one until query.brep lands, so the spatial-tree add/remove paths of Notify are not compared.

All of this is also in plan/findings.md ("topology.brep_p2 (PR #149): coverage limits of the difftest").

Upstream issues

• HANG-18 (new): MateLaminaEdgeToSurface with a negative extension factor never returns (N_SrfExtendByDist). The FFI rejects a negative factor; the Rust returns InvalidInput.
• HANG-9 (main's entry, extended): a mating surface so far from the edge that the squared distance overflows makes dMaxDist infinite and CreateExtendedSurface never stops halving. A MateProbe shim op replays Mate's early returns and then computes upstream's dMaxDist; the FFI refuses only a Mate that would extend by a non-finite distance.
• HANG-28 (new): the shared fixture's SmVertexuse::GetVertexFaceGap (under init_gaps) slows down and then stops returning on a degree ≥ 2 B-spline face with a corner far out. Measured with timeouts on x, y and z: non-rational corners are fast up to 1.34e154 (≈ √DBL_MAX), slow from 1e155, not returning around 1e156; for a rational net (homogeneous, vertex = stored / w) a small weight moves the vertex out and a large one the stored value. The FFI refuses a non-rational corner coordinate above 1e154 and a rational one above 1e151 on the stored or the Cartesian value (pod probes and two re-checks).

UNPORTED tags and deferral labels changed

• Removed (wired): the 8 UNPORTED(topology.shell) and 3 UNPORTED(topology.edgeuse_p3) sites in topology/brep/p2.rs; topology.brep_p6's UNPORTED(topology.brep_p2) SmBrep::Notify; UNPORTED(topology.edgeuse_p2) in query/edge.rs (superseded by port: topology.edgeuse_p3 #117's wiring in the merge); UNPORTED(intersect.curve_class_p7) in brep/p2.rs; port: topology.brep_p1 #145's UNPORTED(topology.brep_p2) SmBrep::GetMinEdgeBBoxSize in proto.rs.
• Retagged: topology.shell's ShellRayWorld::shell_ray_sector default UNPORTED(topology.brep_p2) → UNPORTED(query.brep) (overseer ruling). topology.brep_p6's split now stops at UNPORTED(query.brep) SmBrep::CalculateBoundingBox, upstream's next call.
• No UNPORTED(topology.brep_p2) remains in the crate.
• Deferral labels (runner, not tags): heal.heal_data_p1's 7 SmApiHealBrep box cases defer to primitive.api_primitives (the Rust side cannot build the box); mass.integrator's former brep_p2 deferral is port: topology.brep_p1 #145's harness:mass.integrator-face-construction.

AGENTS.md §4 step 5 checklist

1. Default arguments match upstream: SetTolerance(TRUE, TRUE) and the cascades (FALSE), SmEdge::CalculateBoundingBox bTight = FALSE, curve boxes bExpandBox = TRUE, SetUVTrimCurve(NULL, 0.0, TRUE, NULL), CreateExtendedCurve bPrecise = TRUE, ReplaceSurface(pFace, pExt, FALSE, NULL) with bOptCreateUVTrimCurves = TRUE, AreCoincidentMaybe(.., FALSE), SmFace::SetSurface bDeleteCurrFace = FALSE, GetOrCreateUVTrimCurve(pUV) (NULL, NULL, TRUE), NormalizedEvaluate(t, FALSE, pnt) (NULL, TRUE), TransferChildrenTo(pTo, NULL).
2. No dropped parameters: attribute filters are passed through; SmZoneTol3d values are converted with zone_tol3d (fabs).
3. Special branches all ported: the plane mating-surface extension, the wire and shell-vertex shortcuts of LocalCurveSector, the vertex vote, [B111], [B674], the vertex-shell swap and removal of MakeEdgeTopology.
4. Statuses honoured as upstream: ignored statuses stay ignored (region Remove, the first GetOrCreateUVTrimCurve calls, Trim, the p6 notifications); only Unported and representation failures propagate.
5. Real geometry kinds: the shim builds real SmPlane faces and mating surfaces next to B-splines.
6. FFI guards reject only what is undefined or hangs upstream: bad indices; LocalCurveSector with an empty or non-finite interval, a point class outside 100..=110 or an orientation outside 0..=2; a null region without an infinite region; a negative mate extension factor (HANG-18); a mate whose probed extension distance is not finite (HANG-9); a degree ≥ 2 face under init_gaps with a corner past the HANG-28 bounds; non-finite values.
7. Generated coverage: degenerate inputs, boundaries, extrapolated sector intervals, unknown orientation, tiny and huge scales.
8. No panics or recursion: nothing recurses; every index goes through .get().
9. Stubs: UNPORTED stubs exist only for units that are neither green nor partial.

🤖 Generated with Claude Code

https://claude.ai/code/session_01Xbnf2bFY8cTu4zUgAv9s85

Status (updated 2026-10-20 19:18 UTC)

• Phase: batch: land 5 reviewed-clean contributor PRs #283 merge round pushed at 167dd352, on top of the review round for bherbruck's review 5578349117 (reply, updated with this round: port: topology.brep_p2 #149 (comment)). Main is merged in through 39ceb995 (batch: land 5 reviewed-clean contributor PRs #283, no rebase). No UNPORTED(topology.brep_p2) remains in any crate.
• Gates: fmt, clippy (both configurations), wasm, cargo test -p solidcore (1507) pass; all 232 units have 0 failures, and cargo test -p difftest --features cpp fails only main's two known tests. Known deviation: 3 offset.offset_geometry_creation cases that pass on main now defer (ANALYTIC_EXPORT; SmSurfOfExtrusion / SmSphere, which have no Rust SurfaceKind, issue Coordinate Kernel-aware Surface dispatch for analytic surface kinds #76). offset --perturb is SENSITIVE 552/552 and its --fuzz 2000 is FUZZ-FAIL on 1 timeout (fuzz00504, the curve.curve_p5 cache-rebuild cost, not this unit's defect). brep_p2 --fuzz 2000 is FUZZ-PASS with differs=2 (the plane point solve at ±9.2e18).
• Dependencies: builds on port: topology.brep_p1 #145 (merged). port: topology.brep_p3 #204 builds on this head.
• Next: maintainer re-review 5589603096 at 167dd352 confirmed all three earlier must-fixes (6,839 extra cases, 0 mismatches). The only must-fix is a merge of current main (modules.toml and two reports); once that is pushed and CI is green it is good to merge. That merge round is in progress.

[bluecmd]

First full difftest run: topology.brep_p2 is PARTIAL with 523 pass, 0 fail and 483 deferred (--perturb sensitive 523/523). Most deferrals wait on edgeuse_p2 rebuild/UV paths, face_p3 and topology.shell. Builds on #145. Still a draft.

🤖 Generated with Claude Code

[tangency-monitor]

Landed in batch #315; thanks @bluecmd!

🤖 Generated with Claude Code