[brett19]

Stacked on #693 by @outlookhazy, which ports topology.face_p4 and wires SmFace::FindVertexSector on the mutable curve-classification path.

This PR carries #693's commits unchanged (3ef352f1c, 1ba9727ff, c27cab26d; no rebase or squash). On top of them it adds merges and one revert commit (below):

• merges of the latest main, which port: topology.face_p4 #693 conflicts with;
• the merge of c27cab26d. That commit was written before IntersectTSBound's cached branch moved into ts_bound_cached_visits, so in this branch its vertex pruning sits in the loop-use fallback of intersect_ts_bound, and the curve cache domain is computed there.

Upstream-faithful reverts (our commit d1a8c32, on top of #693's commits)

Two parts of c27cab26d do not match upstream, so they are reverted in a separate commit. #693's commits themselves are left unchanged.

• intersect_ts_bound fallback: SmCurveClassification::IntersectTSBound's loop-use branch (face trimmed cache not up to date, SmCurveClass.cpp ~15880-16000) visits every vertex use with no curve-domain check and builds no curve cache. The pruning of non-singular vertices outside the curve's UV box, and the cache build for it, are removed.
• find_vertex_sector: upstream evaluates cpUVCurve only when the caller passes one, so a 2D cr3DCurve is no longer substituted for it.

Neither revert changes any outcome in intersect.curve_class_p4/p7/p9, topology.face_p2/face_p4/brep_p5/brep_p9 or query.face_p2. The counts match the head before the reverts.

Nine cases now reach the loop bisector on a copy (known mismatches)

Main now has #717, which wired GetLoopCrossings in query/face/p4.rs. With that in place, nine heal-path cases that used to defer at FindVertexSector run on into SetProps_Stage3 -> ClassifyLoops -> the loop bisector, which the port runs on a copy of the B-rep. They disagree the same way as #717's cases (UV trim curves and vertex/face gap resets missing from the real B-rep):

• heal.heal_data_p1: file_plane_circ, file_plane_circ_rebuild_reverse
• topology.brep_p1: plane_circ_heal
• topology.brep_p10: read_Fillet_bottle, read_Fillet_cyl_cyl_shell, read_Fillet_fillet0227, read_Solids_extrusion2, gen0040, gen0165

They are listed in known_mismatches.toml under #717's existing reason and findings entry. A new findings entry records the evidence and a caveat: in read_Fillet_cyl_cyl_shell, HasSeamProblem's seam classification is a second possible source. After the change:

PARTIAL heal.heal_data_p1 cases=495 passed=456 failed=0 deferred=11 known_mismatch=28
PARTIAL topology.brep_p1 cases=2627 passed=2466 failed=0 deferred=136 known_mismatch=9
PARTIAL topology.brep_p10 cases=1404 passed=1380 failed=0 deferred=5 known_mismatch=12

All credit for the port and the fixes goes to @outlookhazy. This PR closes nothing by itself. If #693 merges first, this PR will be closed as superseded.

Difftest before, at #693's 1ba9727ff merged with main:

FAIL intersect.curve_class_p4 cases=3201 passed=3156 failed=43 deferred=2 max_rel_err=1e0 blocked_by=surface.surface_p2(2)
FAIL topology.face_p2 cases=2692 passed=2611 failed=1 deferred=80 (#693's CI)

Difftest after, with c27cab26d, latest main and the two reverts:

PARTIAL intersect.curve_class_p4 cases=3201 passed=3032 failed=0 deferred=169 max_rel_err=0e0 blocked_by=surface.surface_p2(169)
PASS topology.face_p4 cases=5506 passed=5506 failed=0 deferred=0 max_rel_err=0e0
PARTIAL topology.face_p2 cases=2692 passed=2611 failed=0 deferred=81 max_rel_err=0e0 blocked_by=query.face_p2(34),topology.edge(47)

For reference, main has curve_class_p4 at 2921 passed / 280 deferred, face_p4 at 5484 / 22, and face_p2 at 2596 / 96.

🤖 Generated with Claude Code