masi/solvecraft
1
0
Fork 0
mirror of https://github.com/bherbruck/solidcraft.git synced 2026-11-08 13:01:52 +00:00
Code Issues Packages Projects Releases 1 Wiki Activity
Files
c39c1fd388334becc6af40a4ec075094de9f37c5
solvecraft/plan/STATUS.md
T
lillith 6838bb1f76 Merge latest main and validate revolution surface dispatch 
Use the newly ported revolution surface's virtual iso constructor and B-spline
base. Add 18 direct canonical-cylinder comparisons against the real C++ class.
Refresh cache reports after the upstream dispatch changes and retain explicit
CSI and classification deferrals.

All required local gates pass; full differential validation covers 133 units
with 85 green, 48 partial and zero failed. Part 9 compares 727/778 cases exactly.

Co-Authored-By: OpenAI Codex <noreply@openai.com>
2026-10-14 20:10:40 +02:00

83 KiB
Raw Blame History

solidcore status

Last updated: 2026-10-13, PR workflow (surface.trim_srf_cache verified partial in PR #4). Update a summary line here when your unit changes it; per-unit state lives in plan/modules.toml.

Handoff (2026-10-13): from one local swarm to pull requests

The original local swarm (one machine, up to 10 agents) was stopped here so that others can contribute. From now on all work goes through PRs; see AGENTS.md section 4.

  • Where things get ticked off: plan/PROGRESS.md (checklist, regenerated on main by .github/workflows/progress.yml from plan/modules.toml and open port: <id> PRs). The source of truth stays plan/modules.toml; evidence per unit is reports/difftest/<unit>.json.
  • Ready now: curve.curve_p9, surface.plane_p1, surface.surf_of_revolution_p1, surface.surface_p9. surface.trim_srf_cache is being contributed in PR #4, resuming the interrupted origin/unit/surface.trim_srf_cache WIP.
  • Throughput limit: the dependency chain, not people. Often only 2 to 5 units are ready at once. A useful non-port task: find depends_on links that are only 1 or 2 calls and could become back-edges (section 4a), which would widen the ready set.
  • CI was red on main before this handoff (newer clippy on GitHub; a test-only FFI buffer that needed ~37 GB on GitHub's runner). Fixed here: toolchain pinned to 1.98.0, and topology.database_io's FFI wrapper now sizes buffers exactly.
  • Last unit, after everything else: audit.recursion (bounds input-driven recursion; wasm has no stack guard page).

Progress

Counts are from plan/modules.toml. Line counts are the upstream function-body lines (est_lines) assigned to each unit.
Units planned 261 (65 layers), 387,903 upstream lines
Green 83 (104,137 lines): geom.extent3d, geom.vector2d, nlib.knots, nlib.math, topology.database_io, geom.tol, nlib.geom, geom.assert_array, geom.context, geom.extent1d, geom.matrix, geom.periodic_extent1d, geom.vector3d, nlib.basis_basic, nlib.srf_basic, geom.axis2_placement, geom.extent2d, geom.geom_utility_p2, geom.hash_table, geom.local_solve1d, geom.local_solve_nd, geom.polynomial, nlib.basis_adv, nlib.bezier, nlib.funcs_basic, geom.geom_utility_p1, geom.polar_box, nlib.crv_basic, geom.pseudo_box, nlib.crv_fit_p1, nlib.crv_tool_p1, nlib.iges, nlib.srf_tool_p1, curve.hermite_curve, geom.point_set, nlib.crv_fit_p2, nlib.crv_nurb, nlib.funcs_adv, nlib.srf_fit, nlib.crv_fit_p3, nlib.crv_geom, nlib.crv_tool_p2, nurbs.nurbs_crv, topology.topology, nlib.crv_adv_p1, nlib.crv_common, nlib.spiral, nlib.crv_adv_p2, nlib.srf_common, nlib.crv_approx, nlib.crv_offset, nlib.srf_nurb, nlib.srf_tool_p2, nurbs.b_spline_curve_n_lib, nurbs.nurbs_srf, nlib.srf_adv_p2, nlib.srf_adv_p3, nlib.srf_tool_p3, nlib.srf_tool_p4, curve.conic, curve.projected_curve, nlib.srf_adv_p4, nlib.srf_adv_p1, nlib.srf_adv_p5, nlib.srf_adv_p6, nlib.srf_offset, curve.offset_curve, surface.b_spline_volume, nurbs.b_spline_curve_p1, curve.parabola, primitive.api_curves, nurbs.b_spline_curve_p2, curve.curve_p2, curve.curve_cache, curve.hyperbola, curve.line, curve.ellipse_p2, curve.curve_p6, curve.curve_p7, curve.ellipse_p1, curve.curve_p8, curve.composite_curve_p1, curve.composite_curve_p2
Partial 31 (41,289 lines): geom.attribute, geom.tree, geom.object, geom.grid, geom.solution_array, geom.global_solver, geom.tools, geom.transform, geom.a_object, nurbs.b_spline_surface_p1, nurbs.b_spline_curve_p3, curve.crv_in_volume, curve.tangent_field, nurbs.b_spline_surface_n_lib_p1, curve.iso_curve, nurbs.b_spline_surface_n_lib_p2, nurbs.b_spline_surface_p2, surface.volume_p1, curve.curve_p4, surface.twist_volume, curve.curve_p3, curve.curve_p1, curve.crv_on_surf, surface.surface_p2, surface.curve_bounded_surface, surface.surface_cache_p1, curve.curve_p5, surface.trim_srf_cache, surface.plane_p1, surface.unbend_volume, surface.bend_volume
Ported (green + partial) 114 units, 145,426 upstream lines (37.5%); about 150,000 Rust lines in crates/solidcore/src
Claimed 0 (claims now use draft PRs)
Blocked 0
Todo 147 (242,477 lines)

What Wave 0 produced

  • C++ reference kernel (agent A). harness/cpp/build.sh builds upstream SMLib and SM_API at c9c5979 with CMake and Ninja into bin/cpp/ (static libs in bin/cpp/lib, plus compile_commands.json and ffi.env). It uses no TBB and makes no upstream patches. harness/cpp/run_tests.sh runs the upstream suites. Details are in plan/baseline.md.
  • Port plan (agent B). The plan is in plan/modules.toml (260 units), plan/order.md and plan/closure.md. The call-graph tooling is in harness/callgraph/ (run.sh, units.json, why.py).
  • Scaffolding (agent C). This covers LICENSE, NOTICE, ATTRIBUTION, README, AGENTS.md, workspace lints, CI and the crates/solidcore skeleton (math, error, kernel, topology arenas, and module stubs).
  • difftest harness (agent D). It has three parts: crates/solidcore-ffi (shims built from harness/cpp/shim/*.cpp), crates/difftest (runner, comparison and reports) and the harness.selftest unit. To add a unit you create 4 new files and edit no shared file. See crates/difftest/README.md.
  • Wave 0 close (this pass):
    • Integration check, all green: build.sh, selftest difftest, workspace fmt/clippy/tests and the wasm32 check.
    • Fixed the docs to use difftest --unit <id>.
    • Added SolidError::Nlib / NlibCode.
    • Ported the first unit and wrote plan/findings.md.

Baseline (C++ reference, plan/baseline.md)

  • Build: a clean build with -j6 under nice takes 111 s (271 Ninja steps). An incremental build takes about 7 s, and a no-op build under 1 s.
  • SM_API_test_app: PASS. It reports 115 calls; that is 112 real tests passing, 0 failing and 2 skipped. The two skipped tests are stubbed because they compile upstream/Examples. Runtime is 50 s.
  • prog_test_app: PROG_TEST SUCCESS (16 default suites, 46 s). All 17 suites pass when run one at a time, including stitch.

Green units

Unit Rust module difftest Notes
harness.selftest (harness only) 28 cases, max_rel_err 0 not a plan unit
nlib.knots nurbs::knots 2953 cases (1753 fixture + 1200 generated), max_rel_err 0 bit-identical; a 51,753-case run was also clean
geom.vector2d geom::vector2d 4060 cases (60 fixture + 4000 generated), max_rel_err 0 bit-identical; a 200,060-case run was also clean
topology.database_io topology::database_io 2059 cases (59 fixture + 2000 generated), max_rel_err 0 stream state and position match per step; a 200,059-case run was also clean
nlib.math nurbs::math 3076 cases (76 fixture + 3000 generated), max_rel_err 0 bit-identical; a 100,076-case run was also clean
geom.periodic_extent1d geom::periodic_extent1d 4054 cases (54 fixture + 4000 generated), max_rel_err 0 bit-identical; upstream quirks in findings.md
geom.assert_array geom::assert_array 3734 cases (14 fixture + 720 label-table + 3000 generated), max_rel_err 0 bit-identical; all 86 label tables ported here, lengths checked against libSMLib.a symbol sizes
geom.matrix geom::matrix 4044 cases (44 fixture + 4000 generated), max_rel_err 0 bit-identical; a 100,044-case run was also clean; upstream quirks in findings.md
geom.extent1d geom::extent1d 3091 cases (91 fixture + 3000 generated), max_rel_err 0 bit-identical; wired extent3d intervals and tol MapTo1d / GetScaledZero(SmExtent1d)
geom.polynomial geom::polynomial 6038 cases (38 fixture + 6000 generated), max_rel_err 0 bit-identical; quartic UB with tol > 1 noted in findings.md
nlib.srf_basic nurbs::srf_basic 3220 cases (220 fixture incl. Surface.sms / offset_nurb.sms / DropCurveSurface.sms + 3000 generated), max_rel_err 0 bit-identical; a 100,220-case run was also clean; NL_SURFACE as Surface (storage/index split), quirks in findings.md
geom.extent2d geom::extent2d 3069 cases (69 fixture + 3000 generated), max_rel_err 0 bit-identical; a 30,069-case run was also clean; wired tol GetScaledZero(SmExtent2d); PeriodicWrap hang note in findings.md
nlib.basis_basic nurbs::basis_basic 2596 cases (1096 fixture incl. every pt_TestFiles .smc/.sms knot vector + 1500 generated), max_rel_err 0 bit-identical; a 201,096-case run was also clean; upstream endless loops / OOB reads noted in findings.md
geom.local_solve_nd geom::local_solve_nd 3043 cases (43 fixture + 3000 generated), max_rel_err 0 bit-identical (full Evaluate trace of a shared test equation set compared); a 30,043-case run was also clean; wired tol GetScaledZero(SmEvalNFunctionsObject); quirks in findings.md
geom.hash_table geom::hash_table 619 cases (19 fixture incl. SmHealData coincident-vertex replays + 600 generated), max_rel_err 0 bit-identical; a 10,019-case run was also clean; upstream use-after-free and quirks in findings.md
geom.local_solve1d geom::local_solve1d 3316 cases (316 fixture + 3000 generated), max_rel_err 0 bit-identical (full Evaluate trace of a shared test function, outputs and solver state compared); a 20,246-case run was also clean; interval-less constructor and other quirks in findings.md
geom.polar_box geom::polar_box 3030 cases (30 fixture programs incl. pt_TestFiles control-polygon edges + 3000 generated programs on 4 box registers), max_rel_err 0 bit-identical; Union UB case (stale basis after ReSet), aliased Union and region-to-arc quirk in findings.md
nlib.bezier nurbs::bezier 2223 cases (223 fixture incl. SM_API_test rectangle/polygon segments, FitCrvPoints.pts, Points.pts, Surface.sms + 2000 generated), max_rel_err 0 bit-identical; a 50,223-case run was also clean; dangling-matrix, NL_PascalTri bounds and recursion-limit quirks in findings.md
nlib.funcs_basic nurbs::funcs_basic 3130 cases (130 fixture + 3000 generated), max_rel_err 0 bit-identical; a 50k-seed run was also clean; notes in findings.md
nlib.basis_adv nurbs::basis_adv 1744 cases (244 fixture + 1500 generated), max_rel_err 0 bit-identical; a 100,244-case run was also clean; N_BasisDerivsArray unbounded span walk and N_BasisInsertKnots X[mx+1] read in findings.md
geom.geom_utility_p1 geom::geom_utility::p1 4185 cases (185 fixture incl. SM_API_test rectangle/polygon, FitCrvPoints.pts, Points.pts, PointSet sections + 4000 generated), max_rel_err 0 bit-identical; file-static helpers tested via a namespaced compile of SmGeomUtility.cpp; best-fit-line eigenvector, silhouette and FALSE-status quirks in findings.md
geom.axis2_placement geom::axis2_placement 4048 cases (48 fixture + 4000 generated), max_rel_err 0 bit-identical; promoted from partial when AreCoaxial was wired to geom.geom_utility_p1
geom.geom_utility_p2 geom::geom_utility::p2 3270 cases (70 fixture + 3200 generated), max_rel_err 0 bit-identical; promoted from partial when its p1 back-edges were wired; PointSetType is p1's
geom.extent3d geom::extent3d 3086 cases (86 fixture + 3000 generated), max_rel_err 0 bit-identical; promoted from partial when geom.pseudo_box wired Circumscribe (axis-map quirk in findings.md)
geom.pseudo_box geom::pseudo_box 3011 cases (11 fixture incl. my_test_pseudobox + 3000 generated programs on 4 box registers), max_rel_err 0 bit-identical; wired extent3d Circumscribe, tol GetScaledZero(SmPseudoBox), transform EvaluateBoundingBoxSimple (aligned branch)
nlib.funcs_adv nurbs::funcs_adv 2080 cases (80 fixture incl. 6 pt_TestFiles .smc curves + 2000 generated), max_rel_err 0 bit-identical; the three upstream products share one generic body; inputs rescaled in place as upstream
nlib.crv_tool_p1 nurbs::crv_tool::p1 2143 cases (143 fixture on the six pt_TestFiles .smc curves, in-place and out-of-place C++ forms + 2000 generated), max_rel_err 0 bit-identical; promoted from partial when nlib.crv_nurb wired N_CrvGetMinMaxWeightsAndPts (rational N_CrvRemoveKnot)
nlib.srf_fit nurbs::srf_fit 1556 cases (56 fixture on the TestSmCreateSurfaceFromPoints 5x7 grid and the 11 PointSet sections + 1500 generated), max_rel_err 0 bit-identical; a 30,056-case run was also clean; N_FitSrfInterpParams always returns NL_NO, LstSqApprox ruling overwrites the caller grid (copy in Rust)
geom.tol geom::tol 3440 cases, max_rel_err 0 bit-identical; SmRelation pointer-hash order sorted on both sides
geom.context geom::context 2013 cases (13 fixture scripts + 2000 generated), max_rel_err 0 bit-identical; wired the Option<&Context> overloads of geom.tol
nlib.geom nurbs::geom 3173 cases (173 fixture + 3000 generated), max_rel_err 0 bit-identical
geom.vector3d geom::vector3d 6114 cases (114 fixture from my_test_vector3d + edge cases, 6000 generated), max_rel_err 0 bit-identical; Vec3 lives here
nlib.iges nurbs::iges 2029 cases (29 fixture + 2000 generated), max_rel_err 0 bit-identical; entity data passed as Entity126 / Entity128
nlib.srf_tool_p1 nurbs::srf_tool::p1 3132 cases (132 fixture + 3000 generated), max_rel_err 0 bit-identical; extract_patch takes a Rectangle
nlib.crv_fit_p1 nurbs::crv_fit::p1 2084 cases (84 fixture + 2000 generated), max_rel_err 0 bit-identical; M[2i+1][-1] write ported as is (findings.md)
nlib.crv_fit_p3 nurbs::crv_fit::p3 2242 cases (242 fixture on FitCrvPoints.pts / Points.pts and the upstream driver call shapes, edge cases + 2000 generated), max_rel_err 0 bit-identical; errors upstream ignores in the end-derivative estimate are returned, NL_INHERITED reads the caller curve's knots (findings.md)
nlib.spiral nurbs::spiral 2032 cases (32 fixture from the helix tests and edge cases + 2000 generated), max_rel_err 0 bit-identical; upstream crashes on its early input errors (ffi rejects them), span-limit *NumPoints quirk ported (findings.md)
geom.point_set geom::point_set 2029 cases (29 fixture + 2000 generated), max_rel_err 0 bit-identical; unprojected tight-box angles and centroid block quirk kept (findings.md)
topology.topology topology::topology 657 cases (57 fixture + 600 generated scripts), max_rel_err 0 bit-identical; SmTopologyChangeCallback::Execute unused upstream and not ported; dangling-callback UB on tracker destruction skipped (findings.md)
nlib.crv_nurb nurbs::crv_nurb 2871 cases (871 fixture on the six pt_TestFiles .smc curves, SM_API_test rectangle/polygon/circle and reversed-line curves, edge cases + 2000 generated), max_rel_err 0 bit-identical; promoted from partial when nlib.crv_geom wired N_CrvArcLength (N_CrvGetDegenSegs)
nlib.crv_geom nurbs::crv_geom 1723 cases (223 fixture on the six pt_TestFiles .smc curves + edge cases, 1500 generated incl. circles, lines, degenerate and closed curves), max_rel_err 0 bit-identical; closest-point best values on failure and polyline non-termination quirks in findings.md
nlib.crv_tool_p2 nurbs::crv_tool::p2 2124 cases (124 fixture on the six pt_TestFiles .smc curves and curve pairs, in-place and out-of-place C++ forms, degenerate-segment edge cases + 2000 generated), max_rel_err 0 bit-identical; RemoveKnots / ReduceDegree(Once) / MakeCompatible* / RemoveDegenSegs; wired N_CrvsMakeCompatible into crv_basic
nurbs.nurbs_crv nurbs::nurbs_crv 4089 cases (1089 fixture on the six pt_TestFiles .smc curves, SM_API_test rectangle/polygon edges and edge cases + 3000 generated), max_rel_err 0 bit-identical; sm_BasisEval flags the domain ends, sm_CrvTrim returns NL_INP_ERR (findings.md)
nlib.crv_basic nurbs::crv_basic 3417 cases (417 fixture incl. the six pt_TestFiles .smc curves, .sms cross-boundary derivatives, SM_API_test rectangle/polygon segments, my_test_reversed_line_join lines + 3000 generated), max_rel_err 0 bit-identical; promoted from partial when nlib.crv_tool_p2 wired N_CrvsMakeCompatible (N_CrvsAreCoincident maps its NLib errors to false, as upstream)
nlib.crv_adv_p1 nurbs::crv_adv::p1 2670 cases (670 fixture on the six pt_TestFiles .smc curves, SM_API_test rectangle/hexagon edges and edge cases + 2000 generated), max_rel_err 0; a 100,670-case run was also clean bit-identical; ST_curxcdr / ST_curxcrr as Curve::extend_by_*_at_depth; in-place ST_curxcrr G2 recursion frees garbage upstream (findings.md)
nlib.crv_adv_p2 nurbs::crv_adv::p2 2815 cases (815 fixture on the six pt_TestFiles .smc curves, SM_API_test rectangle/hexagon edges and edge cases + 2000 generated), max_rel_err 0; a 300,815-case run was also clean bit-identical; remove_knots shares its body with N_CrvRemoveKnotsDerivConstraints; rational N_CrvSumDiffCrv over different ranges corrupts the heap upstream (findings.md)
nlib.crv_approx nurbs::crv_approx 1646 cases (146 fixture: sm_ApproxCircArc-style counts and arc fits, ApproximateRationalCurve-style approximations of the six pt_TestFiles .smc curves and exact NURBS circles, edge cases + 1500 generated), max_rel_err 0 bit-identical; degree 2 parabolic arcs with parallel tangents and failed tangent projections use uninitialised points upstream (ffi rejects them; the port errors)
curve.hermite_curve curve::hermite_curve 2131 cases (131 fixture + 2000 generated), max_rel_err 0 green once curve.curve_p2 wired SmCurve::IsKindOf (93 deferred before)
nurbs.b_spline_curve_n_lib nurbs::b_spline_curve_n_lib 2282 cases (282 fixture + 2000 generated), max_rel_err 0; a 100,282-case run was also clean green once curve.curve_p2 wired SmCurve::ApproximateLength (57 deferred before)
curve.conic curve::conic 3240 cases (240 fixture: all 15 smCurveTest.cpp IntersectConic scenarios with their solution-count asserts, MakeDupin indicatrix conics, edge cases + 3000 generated), max_rel_err 0; --perturb SENSITIVE bit-identical; first CurveKind added after Hermite; wired SmCurve::ReadFromDB(SmConic_TYPE) in curve.curve_p2; Evaluate zero-tangent fix reads an unwritten slot upstream (findings.md)
curve.projected_curve curve::projected_curve 2727 cases (227 fixture: the SmTransSweepGeometry parallel and SmRotationalSweepGeometry rotation projections of a FitCrvPoints cubic, perspective views, ownership-gated edits, MakeOrthoNormal edge cases, ReadFromDB streams v34/v36 ASCII and binary + 2500 generated), max_rel_err 0; --perturb SENSITIVE bit-identical; CurveKind::Projected; wired SmCurve::ReadFromDB(SmProjectedCurve_TYPE) in curve.curve_p2; upstream MakeOrthoNormal only unitizes zero vectors and Transform needs UNSURE ownership (findings.md)
nurbs.b_spline_curve_p1 nurbs::b_spline_curve::p1 2216 cases (216 fixture on the six pt_TestFiles .smc curves, smSweepTest lines/circles, edge cases + 2000 generated), max_rel_err 0; a 30,216-case run was also clean green once curve.curve_p1 wired SmCurve::TessellateByBisection (4 deferred before)
curve.parabola curve::parabola 3044 cases, max_rel_err 0, 0 deferred bit-identical; CurveKind::Parabola runs the SmBSplineCurve methods on its base (as Hyperbola); GlobalPointSolveSTEP wired to curve.curve_p1 at merge (206 deferred before)
nurbs.b_spline_curve_p2 nurbs::b_spline_curve::p2 2328 cases (328 fixture: the six pt_TestFiles .smc curves through every op and via ReadFromDB, my_test_reversed_line_join lines, SM_API_test rectangle/polygon joins, arc/line joins, Bezier box cases, edge cases + 2000 generated), 2239 bit-identical; a 20,328-case run was also clean promoted to green by curve.curve_p5 (0 deferred after its stubs were wired)
curve.curve_p2 curve::curve::p2 3021 cases (21 fixture: prog_test line and circle curves, ApproximateLength(ivl, 20), circle seam continuity, Hermite base defaults, ReadFromDB dispatch incl. conics + 3000 generated over B-spline, Hermite and arc curves, static helpers, solver box tests), 3016 bit-identical; --perturb SENSITIVE promoted to green by curve.curve_p5 (0 deferred after its stubs were wired)
curve.curve_cache curve::curve_cache 638 cases (38 fixture: lines, cubics, C0 quadratic, closed, degenerate, 2D, rational arc and Hermite curves through Tessellate with every option, sub/beyond/disjoint/reversed intervals, GetAnalyticCurve, type queries + 600 generated), 636 bit-identical promoted to green by curve.curve_p5 (0 deferred after its stubs were wired)
curve.hyperbola curve::hyperbola 2062 cases (62 fixture: smCurveTest.cpp my_test_analytic_curves hyperbola (CreateCanonical, AdjustSTEPInterval(-1,2), evaluation, GlobalPointSolveSTEP, ReverseParameterization), the smSurfaceTest.cpp extrusion-generator hyperbola, edge cases, ReadFromDB records + 2000 generated), 2030 bit-identical; --perturb SENSITIVE promoted to green by curve.curve_p5 (0 deferred after its stubs were wired)
curve.line curve::line 2103 cases (103 fixture: my_test_reversed_line_join, smTopoTest canonical lines, my_test_analytic_curves / my_test_bsplinecurve_nlib lines, SM_API lines, edge cases + 2000 generated op programs), 2102 bit-identical; a 40,103-case run was also clean; --perturb SENSITIVE promoted to green by curve.curve_p5 (0 deferred after its stubs were wired)
curve.ellipse_p2 curve::ellipse::p2 2102 cases (102 fixture: my_test_analytic_curves circle/ellipse, my_test_drop_curve_near_seam circles, smCurveTest circle/circle and line/circle configurations as SmEllipse, edge cases + 2000 generated op programs), 2005 bit-identical; a 30,102-case run was also clean; --perturb SENSITIVE; constructor co-ported from curve.ellipse_p1 promoted to green by curve.curve_p5 (0 deferred after its stubs were wired)
curve.curve_p6 curve::curve::p6 3136 cases (136 fixture: SM_API_test rectangle / hexagon edges (boxes, planarity, coincidence, gaps, drops), my_test_reversed_line_join line pairs, smCurveTest DeformToEndPointTargets B-spline and line steps, circle boxes, edge cases + 3000 generated), 3048 bit-identical; 30k-seed run clean; --perturb SENSITIVE promoted to green by curve.curve_p5 (0 deferred after its stubs were wired)
curve.curve_p7 curve::curve::p7 3027 cases (27 fixture: SM_API_test rectangle / hexagon edge pairs, my_test_reversed_line_join segments, arc plane intersection / classification with comb, edge cases: coincident, tolerance-limit gaps, degenerate Hermite, self-intersecting loop, projections + 3000 generated: whole SolveIt and single leaf-pair LocalSolve, GlobalPropertyAnalysis over all 29 properties), 1716 bit-identical; --perturb SENSITIVE; co-ported sm_FindAdditionalCcTangentGuess promoted to green by curve.curve_p5 (0 deferred after its stubs were wired)
curve.ellipse_p1 curve::ellipse::p1 2028 cases (28 fixture: my_test_analytic_curves circle/ellipse, smSurfaceTest / smTopoTest / smCurveTest circles (CreateCanonical, AdjustSTEPInterval, EditParameterization, copy), edge cases + 2000 generated programs over the constructors, CreateCanonical and ReadFromDB records), 1896 bit-identical; --perturb SENSITIVE; CurveKind::Ellipse (ellipses and circles); &Kernel threaded through the Curve methods whose ellipse override rebuilds the polar converter promoted to green by curve.curve_p5 (0 deferred after its stubs were wired)
curve.curve_p8 curve::curve::p8 3058 cases (58 fixture: SM_API_test rectangle / polygon edge pairs, my_test_reversed_line_join line pairs, with and without the coincidence check; edge cases: line and B-spline line crossings / overlaps / self intersections, vertex touch, parabola tangency, parallel gaps at the tolerance limit, coincident cubics, degenerate Hermite, looping self intersection, disjoint boxes + 3000 generated GlobalCurveIntersect calls), 1968 bit-identical; --perturb SENSITIVE; wired the TrimGap stub in nurbs.b_spline_curve_p3 promoted to green by curve.curve_p5 (0 deferred after its stubs were wired)
curve.composite_curve_p1 curve::composite_curve::p1 3093 cases, all bit-identical promoted to green by curve.curve_p5 (0 deferred after its stubs were wired)
curve.composite_curve_p2 curve::composite_curve::p2 661 cases, all bit-identical promoted to green by curve.curve_p5 (0 deferred after its stubs were wired)

Partial units

Every non-deferred difftest case passes; deferred cases wait on the listed unit (never reported as green).

Unit Rust module difftest Deferred on
geom.tree geom::tree 618 cases (18 fixture + 600 generated), 611 bit-identical; a 20,018-case run was also clean 7 (surface.surface_cache_p3: GetBezierPtr and GetGwNurbPointer read the cache's patch data, which a bare tree reaches once the tree construction hands it over; GetOwnerObject of surface trees wired by surface.surface_cache_p1)
geom.attribute geom::attribute 2029 cases (29 fixture + 2000 generated), 1991 bit-identical; a 200,029-case run was also clean 38 (topology.shell: merge-region Merge)
geom.object geom::object 624 cases (24 fixture + 600 generated), 593 bit-identical 31 (every non-empty SmShape: topology.face_p2 11, topology.brep_p1 10, topology.vertex 6, topology.edge 4)
geom.grid geom::grid 615 cases (15 fixture + 600 generated), 613 bit-identical; a 30,015-case run was also clean 2 (surface.surface_cache_p3 1: SmSurfaceGridElement::Init needs a bare tree node's patch data; surface.surface_p1 1: ExpandHierarchy via smsurf_GetSurfaceCache / CacheMakeOrValidate)
geom.solution_array geom::solution_array 810 cases (10 fixture + 800 generated), 754 bit-identical 56 (object geometry: curve.curve_p2 16, topology.face_p2 14, topology.edge 12, surface.surface_p2 9, topology.vertex 5)
geom.a_object geom::a_object 2040 cases (40 fixture + 2000 generated), 1872 bit-identical; a 50,040-case run was also clean 168 (topology.brep_cache 131: new SmBrepCache; topology.shell 37: SmRegion::IsVoid)
geom.tools geom::tools 913 cases, 721 bit-identical 192 (curve.curve_p2 68, topology.brep_p1 64, topology.vertex 60)
geom.transform geom::transform 3042 cases, 3039 bit-identical 3 (surface.volume_p2: SmVolume::ReadFromDB / EvaluatePoint); surface.volume_p1 wired
geom.global_solver geom::global_solver 620 cases, 611 bit-identical 9 (curve.curve_p2: opaque solution objects not yet resolvable to a Curve, see findings)
nurbs.b_spline_surface_p1 nurbs::b_spline_surface::p1 2120 cases (120 fixture on the three pt_TestFiles .sms surfaces incl. ReadFromDB, prog_test cone/bilinear patches, canonical, sweep, edge cases + 2000 generated), 2058 bit-identical 62 (surface.surface_p1 43: offset, degenerate-curve test, STEP inversion solves; surface.surface_p7 19: EvaluateNormal in polar boxes; IsKindOf wired by surface.surface_p2)
nurbs.b_spline_curve_p3 nurbs::b_spline_curve::p3 2195 cases (195 fixture: the six pt_TestFiles .smc curves through every op, my_test_curve_interface JoinWith / GetControlPoint, my_test_reversed_line_join lines, SM_API_test rectangle/polygon joins and gaps, edge cases + 2000 generated), 2036 bit-identical 159 (curve.curve_p5 159; GlobalCurveIntersect wired by curve.curve_p8, GlobalPropertyAnalysis by curve.curve_p7, SmCurve::AssertValid by curve.curve_p4, SmCurve::IsLine co-ported by curve.curve_cache)
nurbs.b_spline_surface_n_lib_p1 nurbs::b_spline_surface_n_lib::p1 665 cases (65 fixture: my_test_bsplinesurface_nlib offset/sweep/swung, my_test_surface_interface ops on the surf1 net, surf1 Coons patches, the three pt_TestFiles .sms surfaces and 3D .smc curves through every op, edge cases + 600 generated), 638 bit-identical; --perturb SENSITIVE 27 (surface.surface_p5: SmSurface::IsSingularity in CreateExtendedSurface)
nurbs.b_spline_surface_n_lib_p2 nurbs::b_spline_surface_n_lib::p2 654 cases (54 fixture: my_test_surface_interface skinning of surf1 iso curves, smAdvancedTest TrimWithDomain fractions, the three pt_TestFiles .sms surfaces through Evaluate / TrimWithDomain / skinning / bilinear corner blends, degenerate-side zero-tangent fixes, edge cases + 600 generated), 628 bit-identical; --perturb SENSITIVE; 20k-seed run clean 26 (curve.crv_on_surf 17: ApproximateCurve in G1 corner blends; surface.surface_p7 5: EvaluateNormal in n-sided blends; curve.curve_p5 4)
nurbs.b_spline_surface_p2 nurbs::b_spline_surface::p2 763 cases (63 fixture: my_test_surface_interface surf1 through SwapUV / iso curves / SplitAt / CopySubPatch / AdjustSTEPUVDomain / control point ops, my_test_sp_analy CreateSurfOfRevolution, my_create_nurb_surf SetCanonical, the three pt_TestFiles .sms surfaces, revolutions of three .smc curves, edge cases + 700 generated), 718 bit-identical; --perturb SENSITIVE 45 (surface.surface_p1 20: Evaluate1stDerivatives in non-rectangular trimmed planes, SmSurface::IsDegeneratePoint; curve.curve_p5 11: IsArc in spheres from three arcs; surface.surface_p7 9: IsPlanar in GetMaxAnalyticDomain; surface.plane_p2 5: IsNurbSurfacePlane in CopyAndAddAnalytics)
curve.tangent_field curve::tangent_field 655 cases, 511 bit-identical 144 (surface.surface_p7 132, surface.surface_p1 12; GetNaturalUVDomain wired by surface.surface_p2, the runner now builds the shim bilinear surface); SmCrvOnSurf::ReadFromDB wired by curve.crv_on_surf
curve.iso_curve curve::iso_curve 2874 cases (874 fixture: iso curves of Surface.sms / offset_nurb.sms / DropCurveSurface.sms, bilinear and NURB-less edge cases, ReadFromDB streams + 2000 generated), 2754 bit-identical; --perturb SENSITIVE 120 (surface.surface_p6 120; SmSurface::ReadFromDB wired by surface.surface_p2); the SmBSplineSurface::Evaluate back-edge was wired in style review batch #13, SmLine by curve.line
curve.crv_in_volume curve::crv_in_volume 2030 cases, 1822 bit-identical 208 (intersect.curve_class_p4 175: uncached breaks; surface.volume_p2 33: EvaluatePoint, ReadFromDB); the real Volume replaced the stand-in (surface.volume_p1)
surface.volume_p1 surface::volume::p1 2886 cases (886 fixture: plain / transform / oriented / compound chains through every op, TestSmTransform rotations and translations, lifts of every form and order, ConcatEvals, edge cases + 2000 generated), 2847 bit-identical; a 20,886-case run was also clean; --perturb SENSITIVE 39 (surface.unbend_volume: its simple maps; twist wired)
surface.twist_volume surface::twist_volume 1572 cases (72 fixture: constructor repairs, evaluations n 0..5 with / without a next map, bounding boxes, inverse and solves, predicates, copies, equality, ASCII reads + 1500 generated), 1569 bit-identical; a 20,072-case run was also clean; --perturb SENSITIVE 3 (surface.volume_p2: SmVolume::ReadFromDB)
curve.curve_p4 curve::curve::p4 3259 cases (259 fixture: SM_API_test rectangle / polygon edges (also the far 1e-9 / 1.5e-6 ones), my_test_reversed_line_join line pairs, the my_test_drop_point unit circle through every curve property + 3000 generated: chains, closed / degenerate tests, polar boxes, AssertValid, solution filter, projected distance, Newton evaluators, analytic intersections), 3210 bit-identical; a 30,259-case run was also clean; --perturb SENSITIVE 49 (surface.surface_p8 49: SmSurface::GlobalCurveIntersect in HasPlaneCrossings); ellipse intersections wired by curve.ellipse_p1
curve.curve_p3 curve::curve::p3 3076 cases (76 fixture: SM_API_test rectangle / hexagon loops (normals, gap fixing, linearity, plane intersections), smCurveTest line at-distance and circle drop-point / extrema solves + 3000 generated), 2334 bit-identical; --perturb SENSITIVE 742 (curve.curve_p5: LocalPropertyAnalysis 385, SmCCIntersectENFO::Evaluate 136, base SmCurve::Length 120, SmFindCCNormalENFO::Evaluate 101)
curve.curve_p1 curve::curve::p1 3083 cases (83 fixture: smCurveTest my_test_point_extrema / my_test_drop_point line and circle solves, the circle centre, SM_API_test rectangle / hexagon edges through every op, edge cases + 3000 generated), 2493 bit-identical; a 12,083-case run was also clean; --perturb SENSITIVE 590 (curve.curve_p5: LocalPropertyAnalysis 310, SmFindCCNormalENFO::Evaluate 163, SmCCIntersectENFO::Evaluate 73, IsArc 44)
curve.crv_on_surf curve::crv_on_surf 2094 cases (94 fixture: smTopoTest loop UV-gap edges on the bulge patch, smMergeTest plane boundary via the UV-line constructor, face trim curves on Surface.sms / DropCurveSurface.sms + 2000 generated op programs incl. upstream WriteToDB records), 1598 bit-identical; --perturb SENSITIVE; CurveKind::CrvOnSurf; ApproximateCurve / DropPoint / CalculateBoundingBox dispatch to the override 496 (intersect.curve_class_p4 294: uncached breaks; surface.surface_p7 147: EvaluateNormal; surface.surface_p1 50: Evaluate2ndDerivatives; curve.curve_p5 5; SmSurface::ReadFromDB wired by surface.surface_p2)
surface.surface_p2 surface::surface::p2 869 cases (69 fixture: the three pt_TestFiles .sms surfaces through queries / Evaluate / knots / edits / copies / SmSurfaceEval / ReadFromDB, base SmSurface() defaults, TestSmPrincipalDirectionPairing + 800 generated), 756 bit-identical; --perturb SENSITIVE; introduces Surface / SurfaceKind (B-spline only) and SurfaceEvaluate 113 (surface.surface_p5 36: extensions; curve.curve_p5 33: CC local solves and plane splits; surface.surface_p7 18: GetMaxAnalyticDomain; surface.surface_p1 13: STEP inversion solves; surface.plane_p2 11: CopyAndAddAnalytics; intersect.adv_surface_intersector 1; topology.trimming_tools 1)
surface.curve_bounded_surface surface::curve_bounded_surface 535 cases (135 fixture: Surface.sms / DropCurveSurface.sms / offset_nurb.sms bases with the pt_TestFiles .smc boundaries through state / queries / Evaluate / SmSurface defaults / operator== / SetContext, WriteToDB-layout ReadFromDB records incl. the dropped implicit flag and the UV-read-with-3D-type bug + 400 generated), 359 bit-identical; --perturb SENSITIVE; SurfaceKind::CurveBounded 176 (surface.surface_p1 85: EvaluateSimple / EvaluateSTEP / EvaluateSTEPPoint / operator==; surface.surface_p3 56: EvaluateGeometric / EvaluateNormalSection; surface.surface_p7 25: EvaluateNormal; surface.surface_p5 10: EvaluateNormalAtSingularity)
surface.surface_cache_p1 surface::surface_cache::p1 872 cases (72 fixture: the three pt_TestFiles .sms surfaces through boundary bounding planes, Bezier approximation, size tolerances, corners, iso boundary curves, knot continuity, BuildTreeTop over their knot spans and polar boxes, hand-built decompositions through FindUnmarkedLeafNode / GetConnectedRegion / MarkFaceContainment / GetFirstBezierPatch, alternate cache matching, static helper edge cases + 800 generated), 709 bit-identical; a 20,072-case run was also clean; --perturb SENSITIVE; SurfaceCache keeps the patch / aux arenas, tree nodes carry the arena slot (SurfaceAux::slot) 163 (surface.surface_p7 107: EvaluateNormal in every polar box and at collinear boundary samples; surface.surface_p5 37: IsSingularity in every Bezier approximation; surface.surface_p1 19: GetFace)
surface.surface_cache_p2 surface::surface_cache::p2 905 cases (105 fixture: the three pt_TestFiles .sms surfaces (and curve-bounded surfaces over them) through Bezier patch blocks, n-th bounding planes, TestAgainstTolerances on hand-built patch / aux-data / head nodes at the tolerance limits, SubdivideByBlock without surface, SM_API_test rectangle / hexagon planes, alternate cache eviction + 800 generated incl. 1e-4 / 1e4 scaled surfaces), 721 bit-identical; --perturb SENSITIVE; wired the SmPlane::IsSingularity override into the cache's Bezier approximation 184 (surface.plane_p1 53: plane iso curves; surface.surface_p1 47: GetFace in the B-spline bounding box; surface.surface_p5 43: IsSingularity, curve-bounded iso curves; surface.surface_p7 28: EvaluateNormal; surface.surface_p9 13: FindTessellationSplits in every SubdivideByBlock with a surface)
curve.curve_p5 curve::curve::p5 3097 cases (97 fixture: SM_API_test rectangle / hexagon edges through clean-up, duplicates, IsLine / IsLinear / IsArc / Length / IsDomainBoundary, my_test_reversed_line_join lines, closest-point inversions, Newton Jacobians, edge cases + 3000 generated over 16 ops), 3067 bit-identical; a 20,097-case run was also clean; --perturb SENSITIVE; wiring its 13 stubs promoted 12 units to green 30 (mass.integrator 28: SmIntegrator::IntegrateIt in Length of curved spans; curve.curve_p2 2: GetNaturalInterval through a solution handle in TrimGap)

Ready set

2026-10-13 (after surface.surface_cache_p1): surface.trim_srf_cache (surface.surface_cache_p2 also waits for curve.curve_p5 and surface.surface_p9, surface.surface_cache_p3 for both). curve.curve_p5 has landed (partial), so surface.surface_cache_p2 and surface.surface_cache_p3 now wait only on surface.surface_p9. New curve classes become CurveKind variants of curve::curve::Curve (see plan/findings.md, curve.curve_p2); a merge that adds a variant must also extend the exhaustive matches in curve::curve_cache.

Recent findings (details in plan/findings.md)

  • curve.curve_p5 (LocalPropertyAnalysis, IsArc, Length, the curve/curve Newton equations, ...) wired its 13 stubs and promoted 12 units to green. Upstream CrvsCleanup crashes in TrimGap next to a closed curve, so those cases are not generated. The wiring also exposed gaps in dependent runners: solution handles without the curve's interval, the ellipse conversion's out-parameter, and an uninitialised branch slot of SmGlobalSolver::SolveIt. Each was fixed, or the case was excluded as undefined behaviour.

  • No Rust-vs-C++ disagreement so far: every non-deferred case of the 110 ported units is bit-identical (max_rel_err = 0).

  • curve.curve_p7 ports the curve/curve intersection local solve and GlobalPropertyAnalysis; CciGlobalSolver now carries its curves and kernel (with_curves), and a pruned curve tree (aux data without full tree) is rejected (its leaves are not spans; upstream reads them as spans).

  • curve.curve_p6 bounds GlobalCoincidenceChecker's recursion (MAX_COINCIDENCE_DEPTH) and retags the ellipse stub to curve.ellipse_p1; curve.curve_p1 co-ports IsLine dispatch and CacheMakeOrValidate's cache build.

  • curve.iso_curve: SmIsoCurve::Evaluate writes the surface's [nu + 1][nv + 1] derivative array (one order zero) straight into the curve's array; the port flattens BSplineSurface::evaluate's rows. The 355 cases deferred on nurbs.b_spline_surface_n_lib_p2 now pass bit-identical (that unit's porter had found no stub because curve.iso_curve merged later).

  • curve.curve_cache co-ported SmFindCurveDistanceExtremaEFO::Evaluate (curve.curve_p4) and SmCurve::IsLine (curve::curve::p5::is_line_base); porters of curve.curve_p4 / _p5 must not port them twice. sm_ComputeConstants stops at the first non-zero polygon angle with angle tolerance 0 (kept).

  • curve.hyperbola: a NURB-less hyperbola (failed MakeNurb) segfaults upstream in GlobalPointSolveSTEP; curve.curve_p1's porter decides the port (SM_ERR).

  • Nested polymorphic reads (SmCurve::ReadFromDB -> projected/offset/iso curve -> SmCurve::ReadFromDB) have no depth limit upstream; the port bounds them at MAX_READ_NESTING = 64.

  • curve.tangent_field: SmTangentField derives from SmCrvOnSurf; SetAll, GetNaturalInterval, IsKindOf, operator== are CO-PORT(curve.crv_on_surf).

  • nlib.srf_adv_p5: with rails and rational input the derivative curves share the derivative surfaces' knot arrays, so in-place knot writes reach der (ported with make_compatible_knot_tol_traced). nlib.srf_adv_p1: the periodic wrap of N_SrfEvalPtDerivsUnbounded can loop forever (port: NotConverged); N_CreateSkinSpine reads pSpineParams[-1] for one profile with end derivatives.

  • nurbs.b_spline_curve_p3: IsSegmentLine returns a failed status as a "true" SmBoolean; AssertValid test 4 checks the start pair, and its radius-of-curvature tests never fire. JoinWith may degree-elevate the other curve in place.

  • nlib.srf_adv_p4: N_ApproxCrossBoundaryDerivs ignores N_VectorsAngle's status and can refine with an uninitialised knot; N_MergeKnotVectors loops forever on NaN knots; the Coons / skin-boundary routines modify their input curves (port returns errors for the UB paths, compares curves after the call).

  • nurbs.b_spline_curve_n_lib: DampenKnotSpacing can loop forever (port stops after DAMPEN_SPLIT_LIMIT splits); CreateHelixSegment lets through inputs that crash N_CrvApproxSpiral. nurbs.b_spline_surface_p1: the data constructor drops the v knots when s > r (kept); file readers bound header-declared sizes.

  • nurbs.b_spline_curve_p1: SmBSplineCurve.h binds Evaluate's default bNonZeroTangents = FALSE (SmCurve.h says TRUE), so the zero-tangent fix only acts when two or more derivatives are asked; curves built with SmBSplineCurve(lDim) have a NULL NURB that nearly every query dereferences (port: SM_ERR).

  • nlib.crv_nurb: N_CrvGetMinMaxWeightsAndPts always returns pmax = 0; the two N_CrvReverse codings (in place / new curve) agree only when m = n + p + 1 (both kept as reverse / reversed); N_CrvGetType reports degree-1 non-line polylines as free-form.

  • geom.point_set: SetUpTightPseudoBox discards its plane projections (angles use raw points); SetUpCentroid drops the last 50 points when the size is a multiple of 50, so GetCentroid depends on call order (both kept).

  • curve.hermite_curve: Evaluate leaves entry num_derivs unset for num_derivs >= 4 (port and shim return zero there).

  • nlib.crv_tool_p1: green since nlib.crv_nurb landed (N_CrvGetMinMaxWeightsAndPts always returns pmax = 0); N_CrvRemoveKnot can finish a removal and still return its error flag; the insert/split/refine routines do no input checks (port returns InvalidInput). Knot-valued fixture parameters use {"knot": i} because serde_json can parse decimals 1 ulp off.

  • geom.transform: EvaluateBoundingBoxSimple dereferences a null pOptParamPseudoBox when an output is requested (port: InvalidInput).

  • nlib.basis_adv: N_BasisDerivsArray walks spans forward without a bound (reads past U[m] for an inner parameter at or past the domain end) and has no zero-denominator guard; N_BasisInsertKnots reads X[mx+1] (port skips the no-op comparison).

  • nlib.bezier: elevation/product matrices are built on a stack freed before return (dangling storage upstream); NL_PascalTri is [18][18] (degree sums above 17 read past it; port returns InvalidInput).

  • geom.polar_box: Union loops over an unset count after ReSet (UB; port returns InvalidInput), and p->Union(t, *p) drops p's own extent (kept).

  • geom.local_solve1d: the interval-less constructor makes every Newton step a huge step against the Newton direction (kept); always pass an interval.

  • geom.hash_table: SmObjsInVoxels::IncMaxItemCount leaves every bucket item pointing into the freed point-item array when a model with items grows (port uses indices; shim refuses it); FindCoincidentPairs only pairs items with the bucket head and overwrites stored points through a reference (both kept).

  • geom.attribute: several upstream undefined-behaviour paths (double delete, NULL dereference in SmMergeRegionAttribute::Merge, writes past "other" lists) return InvalidInput in Rust and are skipped on both sides; shim liveness is predicted.

  • geom.polynomial: SolveQuarticEqn reads an uninitialised cubic root when the tolerance is above 1 (port uses 0.0; FFI rejects such inputs).

  • geom.tol: SmRelation returns pointer-hash order upstream; the port sorts, and the shim sorts the C++ results before comparing. The small-topology zone tolerance is 1000x larger, not smaller.

  • geom.extent3d: upstream ContainsLineSeg3d ORs its two end-point tests (kept).

  • geom.tree: a point tree (SmTree(point, ...)) divides by zero in AddToSpatialTree upstream (the ffi wrapper refuses it); SmObjectList::FindObject returns the last item as "previous" when not found (kept).

  • topology.database_io: short binary reads are undefined behaviour upstream; the port returns InvalidInput with the same stream state.

  • nlib.math: the N_SrfPowerBasisEvalPt back-edge into nlib.srf_basic was removed; the Rust function takes the control net directly.

Style review batch #1 (2026-10-12)

Reviewed geom.vector2d, geom.extent3d, topology.database_io, nlib.math, geom.tol and geom.tree against AGENTS.md. Numerics and operation order were not changed, and every unit re-ran difftest with unchanged reports.

  • geom.tree: the neighbour searches (left/right/bot/top_neighbors, neighbors) now return Result<Vec<NodeId>> instead of filling a caller out-param. The four GetXNeighbors recursions and the four GetXSide bodies are one direction-generic loop and one recursion. Cyclic parent/child links now return InvalidInput instead of looping forever or overflowing the stack. This applies to find_node, propagate_to_parents, rm_from_spatial_tree, first_bezier_patch and the neighbour searches.
  • geom.extent3d: approx_axis names its near-centre tests (same comparisons).
  • geom.tol: removed a let _ = small;.
  • topology.database_io: error operation paths now include DatabaseReader::.
  • No change was needed in geom.vector2d or nlib.math. nlib.math keeps a justified module-level allow(clippy::if_not_else) (a pedantic lint, not a denied one).
  • Left for geom.vector3d (claimed): sm_sqrt, sm_asin, sm_acos, SmVector3d::GetMaxDimension/DistanceBetween and the smos_Max/Min family (SmMath.h) are duplicated as private CO-PORT(geom.vector3d) helpers in geom/vector2d.rs, geom/extent3d.rs, geom/tol.rs and geom/tree.rs. When that unit lands, replace them with the shared versions. grep -rn "CO-PORT(geom.vector3d)" finds every site.

Style review batch #2 (2026-10-12)

Reviewed geom.periodic_extent1d, geom.matrix, geom.assert_array, nlib.geom, geom.vector3d and geom.context against AGENTS.md. Numerics and operation order were not changed; every touched unit re-ran difftest with the same counts and max_rel_err = 0 (also at --seeds 4000), plus fmt, clippy (with and without cpp), tests and the wasm32 check.

  • geom.vector3d: the last CO-PORT(geom.vector3d) copies (sm_sqrt in geom/matrix.rs; sm_sqrt, DistanceBetween, GetMaxDimension in geom/tree.rs) now call vector3d::sqrt, Vec3::distance and Vec3::max_dimension. sqrt lost a branch that duplicated its result; length/closer_than reuse length_squared/distance_squared.
  • geom.matrix: is_symmetric is one short-circuiting iterator chain.
  • geom.periodic_extent1d: add_interval's integer case_id with a dead _ arm is a private SeamCase enum; complement/set_min_max reuse set_full; error paths name PeriodicExtent1d::.
  • geom.context: Context::new builds its fields directly; MarkLock errors name the failing method; the header lists MarkLock::context2.
  • geom.assert_array: AssertReport equality is one && chain.
  • nlib.geom: API change transform_matrix_from_axes(from: Frame, to: Frame) replaces eight positional Vec3s (new nurbs::geom::Frame); the clamps in vectors_cos_angle/vectors_angle are single expressions; ColinearDim::flag is const.

Style review batch #3 (2026-10-12)

Reviewed geom.extent1d, geom.attribute, geom.polynomial, geom.axis2_placement, nlib.srf_basic and geom.extent2d against AGENTS.md. Numerics and operation order were not changed. Every touched unit (also geom.extent3d, geom.tol, geom.geom_utility_p2, geom.matrix, geom.vector2d, geom.vector3d) re-ran difftest with unchanged reports and max_rel_err = 0 (also at --seeds 4000), plus fmt, clippy (with and without cpp), tests and the wasm32 check.

  • Shared SmMath.h helpers. math::{max2, min2, max3, min3, max4} (smos_Max family) and geom::tol::{is_zero, is_zero_squared, is_zero_to_tol, are_same, are_same_to_tol} (SM_IS_ZERO*, SM_ARE_SAME*) replace 18 private copies in extent1d, extent2d, extent3d, tol, geom_utility/p2, matrix, vector2d, vector3d and axis2_placement. New code should import these instead of adding private smax/is_zero helpers.
  • geom.extent2d/geom.extent3d: the private lerp_exact/classify_coord copies call Extent1d::evaluate/classify_point; extent2d error paths name Extent2d::.
  • geom.axis2_placement: API change Axis2Placement::from_to(from: &Self, to: &Self) replaces six positional vectors (build each with Axis2Placement::new).
  • nlib.srf_basic: API change Surface::with_indices(net, u, v, idx), set_size_indices(idx) and size_arrays(slot, idx) take a new SurfaceIndices { n, m, p, q, r, s } (getter Surface::indices), removing an allow(clippy::too_many_arguments); is_flat_cheap uses combine2_pts/Vec3::length instead of private N_VectorCombine/N_VectorMagnitude copies; Direction::flag / from_flag are const.
  • geom.attribute: region_type::{SOLID, VOID, INFINITE} replace the 0/1/2 literals; the two accumulation loops of SmMergeRegionAttribute::Merge are one append_new_regions; set_regions bounds-checks once.
  • No change was needed in geom.polynomial.

Style review batch #4 (2026-10-12)

Reviewed geom.geom_utility_p2, nlib.basis_basic, geom.hash_table, geom.object, geom.local_solve_nd and geom.local_solve1d against AGENTS.md. Numerics and operation order were not changed; every unit re-ran difftest with unchanged counts and max_rel_err = 0 (geom.local_solve1d and geom.geom_utility_p2 also at --seeds 4000, geom.local_solve1d at --seeds 200000), plus fmt, clippy (with and without cpp), tests and the wasm32 check.

  • geom.geom_utility_p2: API change segment_segment_perspective_intersect(.., view: &PerspectivePlane { point, unit_normal, eye }, tolerance) and line_coplanar_ellipse_intersect(line_pnt, line_vec, ellipse: &CoplanarEllipse { center, x_axis, x_radius, y_axis, y_radius }, tolerance). inverse_4x4 builds each cofactor minor in a loop (minor_3x3) instead of 16 hand-unrolled nine-argument calls; the private determinant_3x3 stub now takes &[[f64; 3]; 3]. No allow(clippy::too_many_arguments) is left in the file.
  • geom.local_solve1d: eval_into evaluates into an IterationValue (no t/f/f' out-params); SolveByBrent has one bisection fallback instead of three copies; the Brent min/max checks of SolveIt and the bracket-end choice of SolveOscillating are one branch each (same comparisons); new SolverAlgorithm::is_brent; a discarded contains_value call was removed; solve_by_bisection takes mut parameters.
  • nlib.basis_basic: operation paths inlined (no OP_* indirection); Errors: prose became # Errors sections.
  • geom.hash_table, geom.local_solve_nd: # Errors sections on every fallible public fn. No other change was needed there or in geom.object.
  • Outside the batch: nurbs::geom's header used a redundant explicit intra-doc link that failed cargo doc with -D warnings (nlib.geom difftest re-run, unchanged).

Style review batch #5 (2026-10-12)

Reviewed geom.polar_box, nlib.bezier, nlib.funcs_basic, nlib.basis_adv, geom.grid and geom.geom_utility_p1 against AGENTS.md. Numerics and operation order were not changed; every unit re-ran difftest with unchanged counts and max_rel_err = 0 (geom.grid still 613 pass + 2 deferred), plus fmt, clippy, tests and the wasm32 check. No public API changed.

  • Shared status codes. error::{SM_ERR, SM_ERR_WARNING} and SolidError::sm_err(operation) replace the private copies in polar_box, grid and geom_utility/p1. Twelve other files still carry const SM_ERR: i32 = 1001 and/or a private fn sm_err (grep -rn "const SM_ERR\|fn sm_err" crates/solidcore/src); new code should import the shared ones, and a later pass can migrate the rest.
  • nlib.bezier: N_BezGetDegreeElevationMatrix and N_BezFuncMultiplyBezCrvMatrix run identical loops; both call one private band_matrix_into. given_or_local replaces the hand-rolled null-matrix fallbacks; reduce_degree shares its left/right loops between odd and even degrees.
  • geom.grid: Setup3DDDA's three per-axis blocks are one private AxisStep::new; GetVoxelAddress's outside test is one expression.
  • geom.polar_box: PolarBoxState::is_unknown (uninitialised or unbounded) serves Cross's early exits and compute_polar_coord; redundant resets dropped.
  • geom.geom_utility_p1: checks that return the same error are one || chain; PointSetType::code reads the discriminant.
  • nlib.funcs_basic, nlib.basis_adv: removed allow(clippy::too_many_arguments / too_many_lines) (allowed workspace-wide); CurveFunction::compact truncates in place.

Style review batch #6 (2026-10-12)

Reviewed geom.solution_array, nlib.crv_basic, geom.tools, geom.transform, geom.global_solver and geom.pseudo_box against AGENTS.md. Numerics and operation order were not changed; every unit (and nlib.crv_tool_p1, merged meanwhile) re-ran difftest with unchanged counts and max_rel_err = 0 (geom.solution_array, nlib.crv_basic, geom.global_solver, geom.tools also at --seeds 4000), plus fmt, clippy (with and without cpp), tests, cargo doc -D warnings and the wasm32 check.

  • geom.solution_array: API change Solution::classify_solution_end returns a new EndClass enum (BelowStart, AtStart, Within, AtEnd, AboveEnd, Inconsistent; code() gives upstream's -2..3) instead of an i64. # Errors sections on every fallible public fn; remove_at uses saturating_sub.
  • geom.global_solver: API change GlobalSolve::solve_it(&SolveRequest { operation, solution_requested, tolerance_3d, best_answer_so_far_sq, opt_vectors }, solutions) replaces five positional arguments (SolveRequest::default() is Intersect / All / 0 / SM_BIG_DOUBLE / no vectors). sm_CombineMinMax matches on EndClass; FindBestSubdivisionNode returns its swap flag directly; the perspective radius loop iterates the four offsets; # Errors sections.
  • geom.transform, geom.pseudo_box, geom.global_solver: the private SM_ERR consts / sm_err fns are replaced by error::SM_ERR / SolidError::sm_err. Removed transform's unused VOLUME_TYPE const and its let _ placeholders. Twelve files still carry a private copy (grep -rn "const SM_ERR\|fn sm_err" crates/solidcore/src).
  • geom.tools: the CurveMaxDistanceBetween stub takes the sample count as a parameter (no let _ = COINCIDENCE_SAMPLES); the edge range is one expression.
  • nlib.crv_basic: OP_* operation paths inlined; "Errors with" prose became # Errors sections; equals is one short-circuiting chain; is_planar skips non-normalisable corners with let ... else (same sums and comparisons).

Style review batch #7 (2026-10-12)

Reviewed nlib.iges, nlib.srf_tool_p1, nlib.crv_tool_p1, nlib.crv_fit_p1, curve.hermite_curve and geom.a_object against AGENTS.md. Numerics and operation order were not changed; every unit re-ran difftest at --seeds 2000 before and after merging main with unchanged counts and max_rel_err = 0 (nlib.crv_tool_p1 is now green with 2143 cases since nlib.crv_nurb landed), plus fmt, clippy (with cpp), tests and the wasm32 check.

  • nlib.iges: API change iges_126_curve(&Entity126 { k, m, t, w, pts, v }) and iges_128_surface(&Entity128 { k1, k2, m1, m2, s, t, w, pts, u, v }) replace six / ten positional arguments (the IGES entity data as named fields).
  • nlib.srf_tool_p1: API change extract_patch(sur, rect: Rectangle) replaces four corner floats (the rectangle is also what the result is reparametrised to); OP_* paths inlined; KnotVector imported; span_ul/span_ur/span_vl/span_vr instead of sul/sur_/...
  • nlib.crv_tool_p1, nlib.crv_fit_p1: OP_* operation paths inlined into each routine's local OP.
  • curve.hermite_curve: mut pseudo_box parameter instead of a rebinding; direct dimension test in approx_eq.
  • geom.a_object: API change the boolean_op const module became the BooleanOperation enum (all SmBooleanOperationType values, from_code); region propagation matches on it, the DIFFERENCE/SLICE/... comment now sits on the default arm. SolidError::sm_err instead of a hand-built Upstream; assert_valid is one expression; Attribute imported.
  • Not changed (judged faithful structure, not transliteration): the u/v twin loops of the surface tools, notify's seven positional arguments (same shape as geom::object::notify), the rational weight check of remove_knot.

Style review batch #8 (2026-10-12)

Reviewed geom.point_set, nlib.funcs_adv, nlib.crv_nurb, nlib.srf_fit, nlib.crv_fit_p2 and topology.topology against AGENTS.md. Numerics and operation order were not changed; every touched unit (also nlib.crv_fit_p1, and nlib.crv_geom / nlib.crv_tool_p2 after merging main) re-ran difftest with unchanged reports, counts and max_rel_err = 0, plus fmt, clippy (with and without cpp), tests and the wasm32 check. No public signature changed.

  • Shared NLib error constructor. SolidError::nlib(operation, code) replaces the private fn nlib in the four NLib units of the batch (and crv_fit::p1). Thirteen older files still carry their own copy (grep -rn "^fn nlib(" crates/solidcore/src); new code should call SolidError::nlib.
  • Shared signed-index helpers. Two crate-private modules hold the upstream NL_INDEX array access the ports keep: nurbs::index (isize: get, set, get2, set2, storage, sx, ux, oob; used by funcs_adv, crv_nurb) and nurbs::fit_arrays (i64: at, put, at2, put2, alloc, alloc2, Block, mget/mset for RM[i][j], banded, full, basis, copy_knots, curve, ...; used by crv_fit::p1, crv_fit::p2, srf_fit), removing about 450 lines of copies. crv_tool/p1.rs, srf_tool/p1.rs and iges.rs still have private isize copies that can move to nurbs::index; new NLib ports should import these modules instead of copying helpers.
  • nlib.crv_nurb: pascal returns the triangle (no &mut out-param), get_tri / set_tri folded into the generic get2 / set2; greville_abscissa reads one checked knot window; OP_* paths inlined (eval_tangent and reversed now name themselves).
  • nlib.funcs_adv, nlib.crv_fit_p2, nlib.srf_fit: OP_* paths inlined; crv_fit_p2 dropped two allow(clippy::too_many_lines) the workspace grants.
  • nlib.srf_fit: the index / start / end bookkeeping shared by the u and v passes of lst_sq_approx is one private Band value.
  • geom.point_set: unit_ortho returns the new [x, y, z] axes instead of writing three &mut out-params; the containment count matches on ranges.
  • topology.topology: Notify's two per-target propagation blocks are one propagate helper (the copy mark reset still applies only to pOther1); the record lookups share one error; stop_tracking lost a dead store.
  • Not changed (judged faithful structure): topology::notify's eight positional arguments (same shape as a_object::notify), the per-direction loops of crv_fit::p2::approx_lst_sq (upstream's second copy of the band bookkeeping differs from srf_fit's in its zeroing), point_set's eight min/max angle updates.

Style review batch #9 (2026-10-12)

Reviewed nurbs.b_spline_curve_p1, nlib.srf_nurb, nlib.crv_adv_p2, nlib.srf_tool_p2, nlib.crv_offset and nurbs.nurbs_srf against AGENTS.md (earlier in the day nlib.crv_geom and nlib.crv_tool_p2 moved to nurbs::index / SolidError::nlib, commits ad631e4, 3c02e52). Numerics and operation order were not changed; every touched unit (also nurbs.nurbs_crv, and nlib.crv_approx after merging main) re-ran difftest with unchanged reports at the default seeds and a 2000-seed run, all max_rel_err = 0, plus fmt, clippy, tests and the wasm32 check.

  • Shared TO_EUCLID. nurbs::nurbs_crv::to_euclid (crate-private; divides by w, unlike CPoint::to_euclid, which multiplies by the reciprocal as N_CPtToPtEuclid) replaces the copies in b_spline_curve/p1.rs, nurbs_srf.rs and the inline one in crv_eval.
  • Public API change: BSplineCurve::edit_parameterization(&Extent1d) lost its unused notify flag (notifications are omitted per the module deviations). The difftest runner still passes notify to the C++ side.
  • SolidError::nlib / SolidError::sm_err replace the private forwarding fns in srf_nurb, crv_adv/p2, srf_tool/p2 and b_spline_curve/p1; nurbs_srf uses nurbs::index::oob. NL_MAX is math::max2 in srf_tool/p2 and crv_offset.
  • crv_adv/p2: OP_* paths inlined; the end-constraint test of remove_knots_deriv_constraints is one boolean expression. crv_offset: one sample-count error constructor. nurbs_srf dropped an if_not_else allow that no longer applies.
  • Not changed (judged faithful or out of scope): srf_tool/p2 keeps importing srf_tool/p1's private Vec<Vec<_>> index helpers (at2 / put2 have no nurbs::index counterpart for ragged nets); b_spline_curve/p1 keeps its module-level OP_* constants (several are shared by more than one method); TypeFinder's labelled-section state machine in srf_nurb.

Style review batch #10 (2026-10-12)

Reviewed nlib.crv_approx, nlib.srf_tool_p3, nlib.srf_adv_p3, nlib.srf_tool_p4, nurbs.b_spline_surface_p1 and nurbs.b_spline_curve_n_lib against AGENTS.md (merge b458c36). Numerics and operation order were not changed; every touched unit (also nurbs.b_spline_curve_p1, nurbs.nurbs_crv, nlib.crv_tool_p2, nlib.crv_nurb and, after merging main, nlib.srf_adv_p2) re-ran difftest with unchanged reports and status, plus fmt, clippy, tests and the wasm32 check.

  • SolidError::nl_ser(op) (crate-private): upstream SER(SM_ERR) after a failed NLib call, letting Unported through. Replaces the private nl_err copies in b_spline_curve/p1.rs, b_spline_surface/p1.rs and b_spline_curve_n_lib.rs; use it as .map_err(SolidError::nl_ser(OP)) in new Sm* ports.
  • Shared NL_BIGD: nurbs::math::BIGD is pub(crate); the private copies in nurbs_crv, crv_tool/p2, crv_nurb and srf_tool/p3 are gone.
  • b_spline_surface/p1 uses nurbs_crv::to_euclid (same branches and divisions) and lists knot_type / is_nurb_borrowed in its header table; it and b_spline_curve_n_lib call SolidError::sm_err directly.
  • b_spline_curve_n_lib: one extension_continuity mapping (SM_G1/G1G2/ CINFINITY -> NL_*) for both CreateExtendedCurve overloads.
  • srf_adv/p3: SolidError::nlib instead of a private wrapper; MAX_EXTEND_DEPTH in the header table. crv_approx: math::max2 for NL_MAX, f64 * Vec3 for N_VectorScale (same products).
  • Not changed (judged faithful or out of scope): srf_adv/p3's local g2/s2/net ragged-net helpers (same shape as srf_tool/p1's at2/put2, which are pub(super)); n_lib's knot_at/cpt_at/vec_at (different error reasons); BSplineCurve::approximate_arc's seven arguments (upstream signature, all required); make_unit_ortho_into's out-params (upstream's partial writes on failure).

Style review batch #11 (2026-10-12)

Reviewed nlib.srf_adv_p1, nlib.srf_adv_p2, nlib.srf_adv_p4, nlib.srf_adv_p5, nurbs.b_spline_curve_p2 and nurbs.b_spline_curve_p3 against AGENTS.md (merge 6472fe3). Numerics and operation order were not changed; every touched unit (also nlib.srf_adv_p3, nurbs.b_spline_curve_p1 and, after merging main, nlib.srf_offset) re-ran difftest with unchanged counts and status, plus fmt, clippy, tests and the wasm32 check.

  • nurbs::index::{at, put, at2, put2}: bounds-checked usize access (a[i], a[i][j] of Vec rows) next to the signed get/set. srf_adv/p1 and p4 use them and index::oob instead of private copies; p1, p2, p4 call SolidError::nlib directly; p5's oob delegates to index::oob.
  • One KN_TOL: srf_adv/p2's duplicate KNOT_SNAP_TOL is gone; it uses srf_adv::p3::KNOT_TOL (same value, same upstream constant).
  • srf_adv/p5: the colinear alias is gone (calls cpts_are_colinear); srf_adv/p4 lists Mat3 in its header table; srf_adv/p1 drops a redundant #[allow(clippy::too_many_lines)] (allowed workspace-wide).
  • b_spline_curve/p2: errors from create_analytic_curve, has_repeated_end_control_points, is_line, number_of_unique_knots and match_weights carry their own operation path instead of the shared OP_NURB; the private nurb_p2/nurb_p2_mut copies are replaced by p1's nurb_ref / nurb_ref_mut (pub(super)).
  • Not changed (judged faithful or out of scope): p5's op-less at/put wrappers (one operation per file); the upstream-named fields of SkinBoundaryOptions (der, cmp, cnt, ...; FFI/difftest API); the per-file invalid helpers of b_spline_curve/p1..p3; the AnalyticCurve empty enum (until curve.line / curve.ellipse_p1 land).

Style review batch #12 (2026-10-12)

Workflow batch #4: reviewed nlib.srf_offset, nlib.srf_adv_p6, curve.curve_p2, curve.conic and nurbs.b_spline_surface_n_lib_p1 against AGENTS.md (merge bd22019). curve.parabola was still claimed (not merged) and is left for a later batch. Numerics and operation order were not changed; every touched unit (also nurbs.b_spline_surface_p1, topology.database_io and, after merging main, curve.hyperbola and curve.projected_curve) re-ran difftest with unchanged counts and status, plus fmt, clippy, tests and the wasm32 check.

  • error::SM_ERR_NULL_POINTER + SolidError::null_pointer(op) (upstream NER on a NULL pointer): replaces four private copies of the 1006 constant and the ner / no_stream / inline Upstream { code: SM_ERR_NULL_POINTER, .. } builders in b_spline_surface_n_lib/p1, b_spline_surface/p1, curve/curve/p2 and topology/database_io (whose pub const is gone: import it from crate::error).
  • srf_adv/p6: nurbs::index::{at, put, oob} and SolidError::nlib instead of private oob / nlib / at copies; *mt(v, i, op)? = x writes are put(v, i, x, op)?, and the rescaled derivative curves are rescaled through one slot borrow.
  • srf_offset: offset_patch_net is a Surface method (was an associated fn taking the patch); UNIT_SQUARE listed in the header table.
  • conic: Conic::TYPE is curve::curve::p2::CONIC_TYPE (no private CURV_BASE_TYPE); error operation paths name the methods (curve::conic::Conic::evaluate, ...).
  • curve/curve/p2: stack_look destructures last_chunk::<7>() instead of indexing with an unwrap_or(0.0) fallback.
  • Not changed (judged faithful or out of scope): b_spline_surface_n_lib/p1's identical build_nlib_curve / create_nlib_curve (two upstream statics) and its invalid helper; srf_adv/p6's rf / mt reference accessors (no shared equivalent yet); curve/curve/p2's u64 as usize multiplicities of Hermite/conic knots; private CURV_BASE_TYPE copies in hermite_curve, hyperbola and b_spline_curve/p1 (use curve::curve::p2::*_TYPE next time those files are touched).

Style review batch #13 (2026-10-12)

Workflow batch #5: reviewed curve.tangent_field, curve.hyperbola, curve.projected_curve, curve.curve_cache, curve.iso_curve and curve.offset_curve against AGENTS.md (merges 16559c3, a0b635d). Numerics and operation order were not changed. Every touched unit (also curve.hermite_curve, nurbs.b_spline_curve_p1, curve.curve_p2, nurbs.b_spline_surface_n_lib_p2) re-ran difftest with unchanged counts and status, except curve.iso_curve (see below), plus fmt, clippy, tests and the wasm32 check.

  • Back-edge wired: curve.iso_curve's UNPORTED(nurbs.b_spline_surface_n_lib_p2) stub for SmBSplineSurface::Evaluate now calls BSplineSurface::evaluate (EvaluateFlags, rows flattened). Deferred dropped from 640 to 285, still partial.
  • Type codes: Hyperbola::TYPE, HermiteCurve::TYPE and BSplineCurve::TYPE use curve::curve::p2::{HYPERBOLA,HERMITE_CURVE,BSPLINE_CURVE}_TYPE; the hyperbola's owner-surface codes come from geom::assert_array::labels. No private CURV_BASE_TYPE remains outside curve::curve::p2.
  • Shared helpers: curve::curve::p2::at (an evaluation entry or UNDEF) is pub(crate) and replaces copies in curve_cache, hyperbola and tangent_field (crv_in_volume still has one); curve_cache uses SolidError::nl_ser instead of its private gw_ser; offset_curve builds its null-base error with SolidError::null_pointer.
  • Enum codes: SurfParamType::from_code and ProjectionType::from_code (returning Option) replace the ad hoc matches in iso_curve and projected_curve.
  • Smaller fixes: curve_cache gets a max_step helper (was computed twice) and an OP_PARENT operation constant; redundant #[allow(too_many_arguments / too_many_lines)] attributes were removed (both lints are allowed workspace-wide); fully qualified paths became imports.
  • Not changed (judged faithful or out of scope): offset_curve's raw SmBoolean i32 flags and no-op set_context; tangent_field's generic surface S; hyperbola's global_point_solve_step unused arguments (consumed once curve.curve_p1 lands); the private ignored / ignore_status copies in curve_cache, hyperbola, curve::curve::p2, b_spline_curve/p3 and geom_utility/p2. A shared error helper would remove them, but that touches units outside this batch.

Style review batch #14 (2026-10-12)

Workflow batch #6: reviewed nurbs.b_spline_surface_n_lib_p2, curve.crv_in_volume, curve.line, nurbs.b_spline_surface_p2, curve.ellipse_p2 and surface.volume_p1 against AGENTS.md (merges 4be3702 and its follow-up). Numerics and operation order were not changed. Every touched unit re-ran difftest with unchanged counts and status: the six above plus geom.solution_array, curve.curve_p2, curve.hyperbola, curve.curve_cache, nurbs.b_spline_curve_p3 (partial) and geom.geom_utility_p2, geom.local_solve1d (green). fmt, clippy, tests and the wasm32 check also passed.

  • Shared error helpers in crate::error: SolidError::invalid_input, SolidError::unported, SolidError::is_unported, and the crate-level ignored (Result<Option<T>>, failure -> None) / ignore_status for calls whose status upstream ignores (Unported still propagates). These replace the private bad / invalid / unported / is_unported / keep_unported / or_ignored copies in this batch's units and the ignored / ignore_status / ignore_failure copies in curve_cache, hyperbola, curve::curve::p2, b_spline_curve/p3, geom_utility/p2 and local_solve1d. That was the follow-up batch #13 asked for.
  • nurbs.b_spline_surface_n_lib_p2: the private SM_ERR_NULL_POINTER / ner became SolidError::null_pointer, and its get became nurbs::index::at.
  • curve.crv_in_volume: dropped the leftover pass-through wrappers from the stub era (volume_natural_param_domain, volume_evaluate, volume_map, volume_map_point, volume_transform, volume_eq). It now calls Volume directly; only the surface.volume_p2 and intersect.curve_class_* stubs remain. It also uses curve::curve::p2::at, and the stale "unported SmVolume::operator==" docs were fixed.
  • Type codes: surface::volume::p1's *_VOLUME_TYPE constants re-export geom::assert_array::labels. curve.line and curve.ellipse_p2 take SURF_OF_*_TYPE / POLAR_CONVERSION_TYPE from labels, and nurbs.b_spline_surface_p2 takes CRV_ON_SURF_TYPE from curve::curve::p2. No private CURV_BASE_TYPE remains.
  • Solution::new / zeroed / Default now need only O: Copy, so curve.line and curve.ellipse_p2 no longer keep their own new_solution.
  • Smaller fixes: Line::join_with's other_is_line no longer makes a dead is_line() call, and Volume::copy_from uses ignored instead of a three-arm match.
  • Not changed (judged faithful or out of scope): Volume::trim_param_bounding_box_simple and has_discontinuities_simple keep upstream's in/out argument, because a disjoint box leaves the caller's value and Transform shares that signature. The per-file OP_* operation-path constants and the Err(SolidError::Unported { .. }) struct literals in other units still work and were left alone. The many other private fn invalid(op, reason) copies under nurbs/ can switch to SolidError::invalid_input in a later batch.

Style review batch #15 (2026-10-13)

Workflow batch #7: reviewed curve.curve_p4, surface.twist_volume, surface.b_spline_volume, curve.curve_p3, curve.curve_p1 and curve.curve_p6 against AGENTS.md (commits 90d5180, 38d50f3). Numerics and operation order were not changed. difftest counts and status are unchanged: p1 3083 (590 deferred), p3 3076 (742), p4 3259 (74), p6 3136 (88), twist_volume 1572 (3), volume_p1 2886 (39) partial; b_spline_volume 2333 green; all max_rel_err 0. fmt, clippy, tests and wasm32 pass.

  • Shared helpers: curve::curve::{p1, p6} dropped their keep aliases of error::ignored; p1 / p3 dropped private invalid for SolidError::invalid_input; p3 / p4 dropped private is_unported for the method; p4's point_or_undef, eval_into and reverse_ignored use ignored / ignore_status instead of hand-written three-arm matches.
  • One eval_n: curve::curve::p3::eval_n (pub(super)) replaces p1's eval_pv and p4's own eval_n (which was p3's with from_left = true).
  • Type codes: p4 takes EDGE_TYPE, EDGEUSE_TYPE, FACE_TYPE, SURF_OF_*_TYPE from geom::assert_array::labels (its local copies are gone).
  • Lints: removed #[allow(clippy::too_many_arguments / too_many_lines)] in p4 / p6; both are allowed workspace-wide. p4's CurvePropertyEfo::evaluate takes _signal_errors instead of let _ =.
  • surface.twist_volume: members_eq no longer computes the dead period comparison (upstream overwrites it; documented); read_from_db takes _version.
  • Judged fine and left alone: surface.b_spline_volume (the USE_VOLUMES #else branches are ported as always-SM_ERR stubs with documented signatures), twist_volume's Option<&mut _> box outputs (documented deviation), the per-file OP_* constants. curve::curve::p2 still has local EDGE_TYPE / EDGEUSE_TYPE copies (outside this batch), and the private fn invalid helpers under nurbs/ remain.

Style review batch #16 (2026-10-13)

Workflow batch #8: reviewed curve.curve_p7, curve.crv_on_surf, curve.ellipse_p1, curve.curve_p8, primitive.api_curves and curve.composite_curve_p1 against AGENTS.md (commits bff8a69, e06dbd7, merged in 1d80f47). Numerics and operation order were not changed. difftest counts and status are unchanged: curve_p7 3027 (1311 deferred), curve_p8 3058 (1090), ellipse_p1 2028 (132), composite_curve_p1 3093 (312), crv_on_surf 2094 (541) partial; api_curves 1629, parabola 3044, geom.tol 3440, geom_utility_p1 4185, geom_utility_p2 3270 green; all max_rel_err 0. fmt, clippy, tests and wasm32 pass.

  • Shared SmMath.h helpers: math::rad_to_deg (SM_RAD2DEG) and math::min4 (smos_4Min) replace private copies in curve::curve::{p7, p8}, geom::tol and geom::geom_utility::{p1, p2}.
  • Shared crate helpers: curve::curve::p3::set_param, curve::curve::p4::point_or_undef and geom::vector3d::is_valid_double are pub(crate); the copies in p7, p8 and primitive::api_curves are gone. curve::curve::p2::at replaces .get(i).copied().unwrap_or(Vec3::UNDEF) in p7, p8 and composite_curve::p1; ellipse::p1 reads its polar tables with nurbs::index::at (private table gone).
  • Type codes: curve::curve::p2::CURV_BASE_TYPE is pub(crate); composite_curve::p1 and parabola dropped their private copies (Parabola::TYPE is PARABOLA_TYPE).
  • Unported propagation: ellipse::p1::read_from_db and p7's line/line guess point use error::ignored instead of hand-written three-arm matches.
  • Dead code: Ellipse::convert_to_2d no longer computes the unused theta_y_rad (upstream computes it and uses ThetaX for both vectors; noted in a comment); join_with picks the common/end points directly instead of swapping and discarding.
  • Lints: removed #[allow(clippy::too_many_arguments / too_many_lines)] in p7, p8 and composite_curve::p1 (allowed workspace-wide).
  • Judged fine and left alone: crv_on_surf (back-edge stubs carry UNPORTED tags, lift formulas keep upstream order), api_curves' #[allow(clippy::neg_cmp_op_on_partial_ord)] (the negated test rejects NaN, as upstream), the remaining three-arm matches whose failure arm returns or continues, and let _ = extent.set_min_max(..) (never Unported). curve::hyperbola and ellipse::p1 still each have a private extract (is >> double); a shared InputStream method would remove both.

Resume note (2026-10-13, after style review batch #16)

  • main is pushed to origin (github.com/bherbruck/solidcore). Counts: 71 green, 35 partial, 3 claimed (curve.curve_p5, curve.composite_curve_p2, surface.curve_bounded_surface), 152 todo, 0 blocked. Ready set: surface.surface_cache_p1.
  • Worktrees branched before batch #16: private rad_to_deg, min4, set_param, point_or_undef, is_valid_double and CURV_BASE_TYPE copies are gone; use crate::math::{rad_to_deg, min4}, curve::curve::p3::set_param, curve::curve::p4::point_or_undef, geom::vector3d::is_valid_double and curve::curve::p2::CURV_BASE_TYPE.
  • Next style review batch: surface.surface_p2 and curve.parabola; also curve::curve::p2's local EDGE_TYPE / EDGEUSE_TYPE, the nurbs/ fn invalid copies and the two private extract helpers.

Known problems

  • Long dependency chain. There are 65 layers, a 116,850-line critical path, and one 220-unit cycle broken by 800 dropped back-edges. A unit with back-edges can only be fully difftested once their targets are green (plan/order.md).
  • Nondeterministic upstream paths. srand(time(NULL)) in SmFace::GetPointsInFace (more than one point) and in SmShell.cpp. Pointer-ordered maps at 4 iteration sites. Process-wide tolerances, and statics set by the first call. See plan/closure.md and plan/findings.md.
  • Large debug library. libSMLib.a is about 296 MB because it is built with -g. Use build.sh --no-tests, or drop -g, if disk gets tight.
  • prog_test granularity. prog_test reports pass/fail per suite only. Its log names and timing lines depend on the clock, so strip them before diffing.
  • Approximate call graph. Override matching (by signature) and template matching (by name) are approximate, so unit closures may over- or under-include a few functions. query.poly and query.tess are included only through over-broad virtual-call resolution; they are low priority.
  • NaN/inf in difftest inputs. JSON can't represent NaN or inf, so non-finite inputs are covered by Rust unit tests only.
  • Misleading seeds field. The difftest report's seeds field shows the CLI value, not the number of cases a runner actually generated.
  • Self-tagged iso curve stubs. surface/surface_cache/p1.rs keeps UNPORTED(surface.surface_cache_p1) stubs for SmPlane::CreateIsoParametricCurve and SmSurfOfRevolution::CreateIsoParametricCurve, although that unit and the units that ported both overrides (surface.plane_p1, surface.surf_of_revolution_p1) are partial. The cache stores iso curves as BSplineCurve, so the SmLine / SmCircle results cannot be stored. Whoever changes SurfaceCache iso curve storage to Curve must wire them (grep the tag). No difftest case reaches them today.

Wave 0 verification (independent check, 2026-10-12)

Check Result
Licence PASS. LICENSE is byte-identical to upstream; NOTICE and ATTRIBUTION.md are present and correct. No tracked file comes from OCCT, USD, Hydra, SmPy or proposals. The CMake build and the call-graph extractor read only source/SMLib, source/SM_API and tests/*. Fixtures cite only tests/prog_test. Crate names carry no NVIDIA or SMLib branding. The README now says "a Rust port of NVIDIA's SMLib".
Gate PASS. A clean --green run in a fresh target dir gave nlib.knots 2953/2953, max_rel_err 0. The C++ side really is upstream: N_BasisEval and the other upstream symbols are linked into the binary. REL_TOL is 1e-9 and there are no abs floors. Seeds are fixed (fnv1a(unit)+i). Two runs produced byte-identical reports, also identical to the committed report. --perturb caught 861/861 perturbed cases.
Rules PASS. fmt, clippy -D warnings (with and without cpp), the wasm32 check and the tests are all clean. The library code has no unsafe (forbid), and its only unwraps are in tests. It has no statics, clock or HashMap. Kernel: Send is asserted.
Plan PASS. 260 unique ids, all fields present, no dangling or self deps, no cycles. Layers equal the longest path, and every upstream path exists under SMLib/SM_API. The ready set (extent3d, vector2d, nlib.math, database_io) matches the table above.

Open issues

  • geom.vector2d / geom.vector3d vs math.rs. The plan puts these units at geom::vector2d / geom::vector3d, but crates/solidcore/src/math.rs already defines Vec2/Vec3, mapped to SmVector2d/SmVector3d. The agent porting those units should extend math.rs, or re-export from it, rather than create duplicate types, and update rust_module to match. Resolved for Vec2: geom.vector2d moved Vec2/Point2 into geom::vector2d and math re-exports them; geom.vector3d can do the same for Vec3.

Surface tessellation and iso boundaries (2026-10-14, PR #6 review)

surface.surface_p9 is partial: 778 cases (178 fixtures and 600 generated), 727 bit-identical, zero failed and 51 deferred (intersect.curve_class_p4: 50; surface.surface_p6: 1), with no absolute floors. All 96 review regressions and all 16 explicit analytic-plane cases without diagonals pass. The regression corpus reproduced 32 failures before the degeneracy/max-side fixes.

The solver uses the existing surface_p1 CsiGlobalSolver and constructor; surface/node/point helpers call the ported p1 implementations. The disjoint CSI fixture now passes; crossing reaches surface.surface_p6. End-to-end CSI is still unverified. FindTessellationSplits is wired into block subdivision: cache p2 compares 808/905 cases (97 deferred), cache p3 450/600 (150 deferred). The latest-main revolution surface also has 18 direct C++ comparisons (canonical cylinder, virtual iso curves), all bit-identical. The committed reports give the current deferral owners; earlier tables above are historical snapshots.

Reference in New Issue View Git Blame Copy Permalink