[brett19]

Replaces the stale UNPORTED(intersect.curve_class_p4): SmCurveClassification::GetLoopCrossings stub in query/face/p4.rs (SmFace::ClassifyLoops, bisector branch) with the real call. It passes what upstream passes: the target loops, the face props and the bisector classification. The status stays ignored, and InvalidInput / NotFound (upstream undefined behaviour) and Unported propagate.

The query.face_p4 runner now exports the face props the bisector and GetLoopCrossings read (zone tolerance, no-vertex poles, pole points, missing seam direction). Without them, 32 of the newly compared cases disagreed.

On main this stub blocked about 410 deferred cases (query.face_p4 90, topology.brep_p10 153, heal.heal_data_p1 109, topology.brep_p7 26, heal.heal_data_p3 19, topology.brep_p1 12, heal.face_props 1).

Difftest, main (CI run 40061276652) → this branch (local)

query.face_p4      500 cases: 395 pass, 105 deferred                → 481 pass, 0 failed, 17 deferred, 2 known mismatches (454 compare values, 27 both-errored as before)
heal.heal_data_p3  478 cases: 458 pass, 20 deferred                 → 477 pass, 0 failed, 1 deferred
topology.brep_p7   348 cases: 289 pass, 59 deferred                 → 315 pass, 0 failed, 33 deferred
heal.heal_data_p1  495 cases: 367 pass, 128 deferred                → 450 pass, 0 failed, 19 deferred, 26 known mismatches
topology.brep_p1   2627 cases: 2461 pass, 150 deferred              → 2465 pass, 0 failed, 138 deferred, 8 known mismatches
topology.brep_p10  1404 cases: 1156 pass, 231 deferred              → 1282 pass, 0 failed, 99 deferred, 6 known mismatches
heal.face_props    1528 cases: 1439 pass, 89 deferred               → 1440 pass, 0 failed, 88 deferred
query.face_p3      unchanged (1087 pass, 112 deferred)

Cases still deferred now mostly stop at SmFace::FindVertexSector (topology.face_p4). The others stop at SmEdge::GetTolerance / GetCurve (topology.edge), SmFace::PointClassify (query.face_p2) and the trimmed-surface-cache tolerance growth (surface.surface_p2).

Parked so this can land

• 40 heal-path cases (heal.heal_data_p1 26, topology.brep_p1 8, topology.brep_p10 6) are known mismatches. Upstream's loop bisector edits the real B-rep: CurveOnClassify → IntersectTSBound → SmVertexuse::ComputeUVPoint(uv) with bOKToMakeUVTrimCurve=TRUE builds missing UV trim curves. SetUVTrimCurve then resets the vertex/face gaps (SmEdgeuse.cpp:4537). The port runs the bisector on the b.clone() copy in query/face/p3.rs, so those edits are lost.
• A follow-up that threads &mut Brep through ClassifyLoops comes right after this PR. It is coordinated separately because it overlaps the planned bisector work and the heal/face_props.rs changes in Propagate remaining topology and healing failures (follow-up #521) #700.
• 2 query.face_p4 cases (gen0157, gen0300, cylinders at scale 1e-3) are known mismatches. The importer's approximation of the analytic edge curves moves the bisector's UV line by 1.6e-8 relative, across a loop. An exactness-based exclusion could not separate these two from about 85 cases that agree, so those cases are compared.
• Both causes are recorded in plan/findings.md.

🤖 Generated with Claude Code

[brett19]

@erlis, a heads-up, since this lands on site 2 from your #681 list (find_and_classify_loop_bisector_into, with LoopObjectCurves<'b, O> holding &'b Brep).

Wiring GetLoopCrossings here lets about 410 cases get past the stub: query.face_p4 goes green, topology.brep_p7 gains 26 and heal.heal_data_p3 gains 19. But 40 newly compared heal-path cases disagree with C++, and we're parking them as known mismatches in this PR so the rest can land:

• heal.heal_data_p1: 26
• topology.brep_p1: 8
• topology.brep_p10: 6

Cause: upstream's bisector runs sms_FindAndClassifyLoopBisector → CurveOnClassify → IntersectTSBound → SmVertexuse::ComputeUVPoint (default bOKToMakeUVTrimCurve=TRUE) on the real B-rep. That builds missing UV trims, and SetUVTrimCurve then resets the vertex-use/face gap caches (SmEdgeuse.cpp:4537). The port runs the bisector on the b.clone() copy in query/face/p3.rs, so those edits are lost. Example: on plane_square, C++ ends with one vertex use's gap set to UNDEF, while Rust keeps 0.

The fix: &mut Brep through ClassifyLoops → the bisector / CurveOnClassify boundary views → set_props_stage3 → HealData::cache_face_props_stage3. That's your site 2 plus the heal chain you described as mechanical. It also touches heal/face_props.rs, which #700 edits.

So these 40 entries are waiting on that conversion. Are you still planning to take site 2? If so, the 40 entries can come off in your PR. If you'd rather we did it, tell us and we'll pick it up straight after this lands.

🤖 Generated with Claude Code

[erlis]

@brett19 thanks for the heads-up and the plane_square trace. Yes, I'll take site 2 and the heal chain, and take the 40 entries off in that PR.

One overlap: your chain goes CurveOnClassify → IntersectTSBound → ComputeUVPoint. On our side that reaches the intersect_ts_bound snapshot in query::face::p2::curve_on_classify_cached, which is site 1 and goes with #697. To get those 40 bit-identical, the bisector has to reach the real B-rep through that path too. Which do you prefer?

• (a) I convert that one curve_on_classify_cached snapshot along with site 2, leaving the other three to port: face-bound trimmed surface cache and IntersectTSBound's cached branch #697.
• (b) port: face-bound trimmed surface cache and IntersectTSBound's cached branch #697 lands its intersect_ts_bound change first, and I build on it.

Either way, the first step is making LoopObjectCurves stop holding &'b Brep, so the bisector can take &mut. I'll check with @fkirill on #700 before touching heal/face_props.rs.

🤖 Generated with Claude Code

[brett19]

@erlis thanks, and sorry for the slow reply. (b) is already done: #697 merged earlier today (merge d2944e9), with the intersect_ts_bound cached branch. curve_on_classify_cached's snapshot can now take a caller's &mut Brep through CurveOnClassifyCache (described in #697's body). So please build on main as it is, and take whatever you need from site 1 along with site 2.

One update on the count: #718 (FindVertexSector) parked 9 more cases under the same reason, so it's now 49 in known_mismatches.toml (heal_data_p1 28, brep_p1 9, brep_p10 12). For three of the brep_p10 ones (read_Fillet_cyl_cyl_shell, read_Fillet_bottle, read_Fillet_fillet0227), #718's findings note a possible second source: HasSeamProblem → EvaluateNormalAtSingularity returns InvalidInput on the Rust side, and the port ignores it as upstream does. So those three may not all clear with the &mut change alone.

We won't touch LoopObjectCurves, ClassifyLoops, the bisector or the heal chain while you're on it.

🤖 Generated with Claude Code