[bluecmd]

Ports unit topology.brep_p7 (source/SMLib/src/SmBrep.cpp, 6 functions): ReadFromFile, MakeFace, MakeFaceWithCurves, GlueEdgesGeneral, CopyFaces, MergeSelfIntersectionCurves → crates/solidcore/src/topology/brep/p7.rs (new topology::brep module; brep_p1/p2/p10 add their pub mod lines next to p7), with shim harness/cpp/shim/topology_brep_p7.cpp, FFI wrapper and difftest runner.

Status: partial. Calls into units not ported yet are UNPORTED stubs (no co-porting): GetInfiniteRegion (topology.brep_p1); SmBrep::Notify, SetTolerance, sm_CheckAndFixUVTrimCurves, RebuildUVTrimCurves (topology.brep_p2); MakeWireEdge(Vertex) (brep_p3); DeleteEdge (brep_p4); GlueVertices (brep_p5); MakeShellVertex (brep_p9); MakeTopologyFromData, TestClosureSplitRegion(edge) (brep_p10, #130); SmFace::Notify/SetTolerance/CreateUVTrimCurves (face_p3); SmFace::GetBrep/GetEdges (face_p4); SmMerge (boolean.merge_p2); the curve constructor of SmCurveClassification (curve_class_p4). ReadFromFile takes the file's bytes (the library reads no files); the reader is BrepData::read_from_file, which bounds every count and nesting level.

Difftest

PARTIAL topology.brep_p7 cases=316 passed=197 failed=0 deferred=119 max_rel_err=0e0 with_values=197 no_values=0 blocked_by=cpp-guard:ub-165(3),curve.crv_on_surf(1),curve.curve_p6(2),topology.brep_p1(12),topology.brep_p10(30),topology.brep_p2(51),topology.brep_p9(19),topology.face_p3(1)
SENSITIVE topology.brep_p7 perturbed=197 detected=197 unperturbed=0 deferred=119
FUZZ-PASS topology.brep_p7 cases=2000 rejected=233 malformed=296 ran=1471 differs=0 crashes=0 panics=0 timeouts=0
cases_both_errored=0

What the passes cover: all 197 passes are no-ops, empty inputs or error statuses. No pass completes a path that builds topology (a successful read, a face made, a real glue, a copy, a non-empty self-intersection merge); every such path defers, mostly at topology.brep_p2's SmBrep::Notify.

• 110 glue no-ops: null, identical, or wrapped-to-identical edges. Each compares the B-rep's counts and its sorted edge and vertex tolerances after the op, 10 to 52 values per case.
• 64 MergeSelfIntersectionCurves calls with no curves.
• 23 error statuses: 12 read errors, 5 face-from-curves input errors, 4 MakeFace null surfaces, 1 square too small for its lines on both sides, and 1 generated case.

no_values=0; the shim's -2 guards are deferrals, not passes.

Coverage, by op (from the report):

op	pass	deferred	what the passes compare
ReadFromFile	12	30	read errors (status 1001) of damaged streams the topology.brep_data guard accepts, pinned by a deterministic search (an overwritten digit of an integer token); every other read defers at MakeTopologyFromData (brep_p10)
MakeFaceWithCurves	5	27	input errors (no curves, no loops, degenerate surfaces), the last two now also with curve on surface and no UV curves; 3 such calls past those checks defer to cpp-guard:ub-165; a curve on surface over a plane defers to curve.crv_on_surf; an open UV-only curve-on-surface loop on a B-spline patch defers to curve.curve_p6
MakeFace	5	5	SM_ERR_NULL_POINTER for a null surface; a square too small for its lines (-3 on both sides); square loops defer at MakeShellVertex (brep_p9)
GlueEdgesGeneral	107	43	no-ops only (null, identical, or wrapped-to-identical edges), compared as the B-rep's counts and tolerances after the op. Everything past that early return defers at SmBrep::Notify (brep_p2): non-adjacent pairs (SM_ERR), wires, real merges, and every shared-vertex pair of the files
MergeSelfIntersectionCurves	60	0	empty input (no curves): no edges
CopyFaces	0	6	- (SmBrep::SetTolerance)
generated	8	8	glue no-ops (3), empty merge-self (4), 1 status; the generator draws all six ops (face-from-curves and copies had been left out by oversight), and every orientation code including SM_OT_UNKNOWN

Every path that builds topology is deferred. partial is the right status until brep_p1/p2/p3/p6/p9/p10 land.

C++ guards.

• Damaged streams SmBrepData accepts are not run in C++ (cpp-guard:unvalidated-topology). Both sides decide this with the C++ reader. The guard is needed: MakeTopologyFromData uses the file's cross-reference indices unchecked (UB-164, [PSIRT]). With the guard off, 11 of 600 single-byte damages crashed, and a batch of 100 crashed in 13 of the 15 files. Revisit when brep_p10 lands. The 45 overwrites of the leading comment line were dropped: C++ reads them exactly as the plain file.
• -2 guards (null dereferences) are deferred. These cover:
  • a curve on surface without UV curves (UB-165);
  • a glue with SM_OT_SAME, equal intervals and exactly one edge without a curve (upstream's exact dereference condition; no fixture edge lacks a curve);
  • MergeSelfIntersectionCurves on a face without a surface.
• No other guards. Shared-vertex glue runs in C++. The native read skips the stream check above only for bytes identical to one of the shipped test files; there is no caller-settable trust flag.

Upstream issues found:

• UB-165: SmBrep::MakeFaceWithCurves with pbOptCrvOnSurf TRUE and no UV curves dereferences a null array (SmBrep.cpp:16600). Confirmed by a C++ run. Renumbered from UB-163, which port: heal.heal_data_p3 #164 took on main.
• UB-164 [PSIRT]: SmBrep::MakeTopologyFromData indexes its section arrays with unchecked indices from the file. Confirmed by C++ runs.

Gaps in other units' ports: the port of SmCurve::DeformToEndPointTargets (curve.curve_p6) has no pOptSurfCrvOnSurf. A curve on surface with a gap target therefore defers (UNPORTED(curve.curve_p6)) instead of being approximated (case face_curves_cos_open_loop); other curves still use the port. The gap is recorded in curve.curve_p6's notes. CrvOnSurf holds only B-spline surface kinds, so other kinds defer to curve.crv_on_surf, with the same label and tag as that gap's other sites. See plan/findings.md.

Back-edges (§4a)

Wired:

• topology.trimming_tools (TrimSurfaceWithModelSpaceCurves → make_face_with_curves).
• boolean.topology_intersector_p1 (FaceFaceIntersect → GlueEdgesGeneral): a new IntersectorBrep::brep_glue_edges_general trait method with all upstream arguments (including the default-NULL face-use list), called with upstream's exact arguments; the runner's adapter calls p7::glue_edges_general. No case reaches the glue yet (earlier stubs defer first), so its report is byte-identical. port: topology.brep_p1 #145, port: topology.brep_p3 #204 and boolean.topology_intersector_p1: follow-ups to #146 #227 touch the same trait (noted in COORD.md).
• This unit now calls topology.shell's SmShell::IsInnerShell (port: topology.shell #154) in MergeShells.
• topology.brep_p6 (port: topology.brep_p6 #169): MakeFace calls MakeFaceTopology with upstream's arguments, and GlueEdgesGeneral's region nesting calls TestClosureSplitRegion(faceuse) (status ignored; each call resets rpNewRegion, as upstream). The one case that deferred at brep_p6 (face_curves_cos_no_uv_1_1) now defers inside it at SmFace::SetTolerance (topology.face_p3).

Relabelled, not wired: offset.offset_executive_p1 (#175) had an UNPORTED(topology.brep_p7) MakeFaceWithCurves stub. It still does not call this unit. Its first argument, m_pOffsetBrep->GetInfiniteRegion(), is unported (topology.brep_p1), so the stub now names that real first blocker. No case reaches it. For whoever wires it later: upstream ignores a failed MakeFaceWithCurves there (if (... == SM_SUCCESS) {...} and then falls through), so that error must not be passed up with ?.

PARTIAL topology.trimming_tools cases=3263 passed=2842 failed=0 deferred=421
PARTIAL boolean.topology_intersector_p1 cases=827 passed=653 failed=0 deferred=174
PARTIAL topology.shell cases=911 passed=892 failed=0 deferred=19
PARTIAL topology.brep_p6 cases=498 passed=450 failed=0 deferred=48
PARTIAL offset.offset_executive_p1 cases=693 passed=675 failed=0 deferred=17

Verification (at f787d107; merged with origin/main at e6414686)

• fmt, plus rustfmt --edition 2021 --check on the include!-ed runner and FFI unit files (lesson 25: they had never been formatted; the reformat is in its own commit).
• clippy --workspace --all-targets [--features difftest/cpp] -D warnings, cargo test -p solidcore (1369 passed), cargo check --workspace --all-targets and -p difftest --features cpp, wasm32 check.
• --fuzz 2000: FUZZ-PASS, 0 crashes/panics/timeouts, 0 differs.
• Full green check (sc-remote green, 207 units, failed=0).
• valgrind (C++ and Rust in one process, unmodified generated programs, compared with a baseline run): 0 errors. 90 generated programs (face-from-curves 25, glue 15, merge-self 15, reads 15, make-face 10, copies 10), plus the 32 built-in face-from-curves and 6 copy cases.
• The 300 built-in cases keep their outcomes; the one deferral that changed is intended: face_curves_4 (a curve on surface over a plane) now defers to curve.crv_on_surf instead of topology.brep_p9.
• The topology.edgeuse_p1, topology.edgeuse_p2 and topology.loopuse runners now name topology.brep_p10 MakeTopologyFromData for their inexact-import deferrals (200, 225 and 87 cases); their reports differ only in those labels. topology.vertex had its own copy of the label (relabelled, 224 cases), and topology.edgeuse_p3 shares edgeuse_p2's (381 cases).
• gen0014 changed from pass to deferred, by an input change from main. Since topology.shell (port: topology.shell #154) the shared fixture walk registers wire edges, so this generated glue now draws a wire-network edge pair instead of a pair that took the no-op early return. The C++ reaches SmBrep::Notify on it and the Rust defers there (topology.brep_p2), which is correct; the port did not regress. Hence 197 passes / 119 deferred (was 198 / 118).
• The FFI wrapper uses ffi: share the validate Reader through solidcore_ffi::reader #186's shared reader (unit_reader!(MAX_COORD)); only the unit's signed and orient checks stay local.
• Merging main (ffi: share the validate Reader through solidcore_ffi::reader #186, port: offset.offset_executive_p1 #175, port: topology.edgeuse_p3 #117, fix main build: offset executive passes the kernel and a mutable B-rep #230, port: topology.brep_p6 #169) and wiring changed no other case: brep_p6, offset_executive_p1, intersector_p1, trimming_tools and shell are byte-identical.

🤖 Generated with Claude Code

https://claude.ai/code/session_01EA8bchjV6Km7PFkmnvYBTA

Status (updated 2026-10-19 04:19 UTC)

• Phase: ready for re-review at f787d107: the 17c43c2a round (every point of the review of a5359b48) refreshed against main up to e6414686 (port: topology.shell #154 shell, port: topology.brep_p6 #169 brep_p6, port: offset.offset_executive_p1 #175/port: topology.edgeuse_p3 #117/fix main build: offset executive passes the kernel and a mutable B-rep #230, ffi: share the validate Reader through solidcore_ffi::reader #186 shared reader), with the intersector_p1 and offset_executive_p1 back-edges.
• Difftest: PARTIAL, 316 cases: 197 pass / 0 fail / 119 deferred (with_values=197, no_values=0), --perturb 197/197, FUZZ-PASS 2000, green 207 units / 0 failed. All 197 passes are no-ops (110), empty inputs (64) or error statuses (23). gen0014 changed from pass to deferred because its input changed (explained under Verification).
• Next: approved by the maintainer at f787d107 (CI green). The optional GetInfiniteRegion follow-up will be a separate PR after merge.

[bluecmd]

First full difftest run: topology.brep_p7 is PARTIAL with 310 cases: 169 pass, 0 fail and 141 deferred, --perturb sensitive 169/169. The topology-building paths wait on #130 and #154, so the passes so far are error paths and no-ops. Still a draft; an independent pre-review is running before it's marked ready.

🤖 Generated with Claude Code

[bluecmd]

Ready for review. topology.brep_p7 is PARTIAL with 316 cases: 197 pass, 0 fail and 119 deferred, --perturb sensitive 197/197, and 0 both-errored. Of the passes, 173 compare values and 24 compare only a status. The rest defer at brep_p2/p10/p9/p1/p6 or at C++ crash guards. The trimming_tools back-edge (MakeFaceWithCurves) is wired: 3263 cases, 2842 pass, 0 fail. Every green/partial unit's difftest passes (198 units, 0 failed), and fmt, clippy, tests and wasm are clean at bbdb8bf1.

Upstream issues: UB-165 (MakeFaceWithCurves with curves on surface and no UV curves) and a [PSIRT]-class entry UB-164 (MakeTopologyFromData). Details are in plan/upstream-issues.md.

Three independent pre-review rounds went before this. They found a faked curve-on-surface deform path, which now defers, and they measured every C++ guard for over-rejection.

CI needs a maintainer's approval to run on this fork PR.

🤖 Generated with Claude Code

[bluecmd]

Merged origin/main (#148, #152, #176) and refreshed the numbers in the description. topology.brep_p7 is now 316 cases: 198 pass, 0 fail, 118 deferred, with with_values=198 and no_values=0. --fuzz 2000 gives FUZZ-PASS, valgrind is clean, and green passes on 200 units.

Two other changes since the last push. The generator now also draws MakeFaceWithCurves and CopyFaces. The include!-ed runner and FFI files have been rustfmt-ed, in a commit of their own.

🤖 Generated with Claude Code

[bluecmd]

@bherbruck, this is refreshed against main at f787d107: main merged up to e6414686, so it merges cleanly again. It adds no change to this unit's review points; it only follows main:

• topology.shell (port: topology.shell #154): MergeShells calls SmShell::IsInnerShell.
• topology.brep_p6 (port: topology.brep_p6 #169):
  • MakeFace calls MakeFaceTopology with upstream's arguments.
  • GlueEdgesGeneral's region nesting calls TestClosureSplitRegion and ignores its status, as upstream does. Each call resets rpNewRegion.
• Back-edge: boolean.topology_intersector_p1 reaches GlueEdgesGeneral through a new IntersectorBrep::brep_glue_edges_general method with upstream's arguments. No case reaches it yet, and its report is byte-identical.
• Relabelled, not wired: offset.offset_executive_p1's MakeFaceWithCurves stub now names its real first blocker, GetInfiniteRegion (topology.brep_p1, its first argument). It still does not call this unit. Upstream ignores that call's failure, so the eventual wiring must not pass the error up.
• ffi: share the validate Reader through solidcore_ffi::reader #186: the FFI wrapper uses the shared reader.
• Relabel: topology.vertex and topology.edgeuse_p3 (from port: topology.edgeuse_p3 #117) get the same topology.brep_p10 MakeTopologyFromData label as the other three runners. Their reports differ only in labels.

Two deferrals changed, both expected:

• gen0014 now defers at SmBrep::Notify instead of passing. Since port: topology.shell #154 the shared fixture walk registers wire edges, so this generated glue draws a wire-network pair. The C++ also goes past the early return to Notify, so this is a changed input, not a regression.
• face_curves_cos_no_uv_1_1 moved from brep_p6 to SmFace::SetTolerance inside it (topology.face_p3).

PARTIAL topology.brep_p7 cases=316 passed=197 failed=0 deferred=119
SENSITIVE topology.brep_p7 perturbed=197 detected=197
FUZZ-PASS topology.brep_p7 cases=2000 differs=0 crashes=0 panics=0 timeouts=0

Workspace check, clippy, 1369 tests, wasm and the full green check (207 units, 0 failed) pass. The description has the details.

🤖 Generated with Claude Code

[bherbruck]

Thanks for the thorough follow-up, @bluecmd. I reviewed head f787d107 and it's good to merge once CI is green (CI has already passed on this head).

Earlier review (a5359b4): every item is fixed

• Must-fix: stand-in CrvOnSurf base (p7.rs:676). Fixed. Only a plain SurfaceKind::BSpline builds a CrvOnSurf now (p7.rs:694-705). Planes and surfaces of revolution return Unported(curve.crv_on_surf), as main does.
• MergeSelfIntersectionCurves loop shape. Fixed. Checked against SmBrep.cpp:22701-22790: skip when size ≠ 1 before the merge, SM_ERR after it, SE + continue for a middle not on an edge, then the GlueVertices/GlueEdgesGeneral order. MergeIntervals resets pOptNewFaces (SmCurveClass.cpp:18610), so returning fresh faces per call is faithful.
• Deferral labels in the edgeuse_p1/p2, loopuse, vertex and edgeuse_p3 runners. Fixed. They now point at topology.brep_p10 MakeTopologyFromData.

I also checked the new wiring since then. MakeFace → p6::make_face_topology restores the owner whether the build fails or succeeds (SmBrep.cpp:18870). GlueEdgesGeneral's region nesting matches :10443-10471. The trimming_tools and topology_intersector_p1 back-edges pass upstream's arguments (SmTrimmingTools.cpp:156, pInsideFaceuses NULL).

Gates (all pass)

• cargo fmt --check, plus rustfmt --edition 2021 --check on the include!d unit files. The PR's own files are clean. The edgeuse_p1/p2, loopuse and vertex runners have the same diffs as main, so nothing new.
• clippy -D warnings, both plain and with solidcore-ffi/cpp,difftest/cpp.
• cargo test -p solidcore: 1369 tests pass.
• The wasm32 check.
• git merge-tree against origin/main is clean.

Extra adversarial cases

• I added 497 hand-written cases (not pushed):
  • MakeFaceWithCurves at scales from 1e-7 to 1e6;
  • corner gaps from 0 to 0.5, bracketing the tolerance at 1e-6, 1e-5 and 1e-4;
  • zero-length and 1e-9 curves, including loops made only of degenerate curves;
  • odd loop-count shapes, empty and inverted domains, near-zero and tilted plane normals, loop points;
  • MakeFace sides from 1e-11 to 1e9;
  • glue with gaps of -1, 0 and 1e3 in every orientation;
  • MergeSelf and Copy with extreme tolerances and counts.
• I also ran 1,800 generated cases (--seeds 1500) and --fuzz 3000.
• Results: 0 real failures, 0 crashes, 0 panics.
  • The 4 "failures" are inputs I wrote whose loop counts exceed the curves. The FFI guard rejects them, which is correct, because upstream reads past cp3DCurves.

PARTIAL topology.brep_p7 cases=316 passed=197 failed=0 deferred=119 max_rel_err=0e0 with_values=197 no_values=0
extra (497 hand + 1800 generated): passed=852 failed=0 real (4 guard-rejected invalid inputs) deferred=1757
SENSITIVE topology.brep_p7 perturbed=403 detected=403
FUZZ-PASS topology.brep_p7 cases=3000 rejected=557 malformed=573 ran=1870 differs=0 crashes=0 panics=0 timeouts=0
partial topology.trimming_tools cases=3263 passed=2842 failed=0 deferred=421
partial boolean.topology_intersector_p1 cases=827 passed=653 failed=0 deferred=174
partial offset.offset_executive_p1 cases=693 passed=675 failed=0 deferred=17
partial topology.edgeuse_p1/p2/p3, loopuse, vertex: 0 failed
--green: 207 unit(s), 113 passed, 94 partial, 0 failed

Should-fix (optional, can be a follow-up)

• p7.rs:127-133: the GetInfiniteRegion stub. Upstream's getter is an inline field read (SmBrep.h:736), and main already exposes it as Brep::infinite_region(); p6.rs:123 and shell.rs:359 use it. The stub is allowed, since topology.brep_p1 is todo. But returning b.infinite_region().ok_or_else(..) would stop 12 cases (197 in the seeded run) deferring at it. The relabelled offset stub (offset_executive/p1.rs:815) could use the same accessor and call make_face_with_curves directly.

🤖 Generated with Claude Code

[bluecmd]

Thanks for the re-review and the extra adversarial runs, @bherbruck.

I'll leave f787d107 as it is. Your GetInfiniteRegion should-fix will come as a small follow-up PR once this merges. It reads Brep::infinite_region() (the inline m_pInfiniteRegion read) in MakeFace, MakeFaceWithCurves, CopyFaces and the glue's shell merge, and keeps a NULL region as None.

The reset after a glue removes an empty region (m_pInfiniteRegion = m_pList) stays a stub, because the B-rep's region list order isn't modelled. The offset site needs that unit's surface ownership and face identity settled, and upstream ignores the call's status there, so I've left it for its porter.

🤖 Generated with Claude Code

[bluecmd]

merged main (#145/#196) to resolve conflicts, no other change

🤖 Generated with Claude Code

[bherbruck]

Merged, thanks @bluecmd!

🤖 Generated with Claude Code