[erlis]

Claim for site 4 of #681 (sweep/topology_sweep/store/classification.rs): SmPrimitiveCreation::CreateRotationalSweep's axis check runs SmFace::CurveOnClassify on a private copy of the B-rep. The UV trims and tolerance growth that the interval point classification makes are dropped there; upstream keeps them on the B-rep. Plan: run it on a shared CurveOnClassifyCache and copy back with apply_uv_trim_updates, as the fillet session's classification does.

Work in progress; the description will carry the difftest numbers.

🤖 Generated with Claude Code

[erlis]

Closing: measured first, and this site has no observable effect in difftest today.

• Geometry::vertex_face_gap's per-query copy is never reached here: axis_inside_face classifies with has_big_gaps = false, and the adapter's gap is only read when it is true.
• What the private copy could drop is the interval point classification's UV trims, tolerance growth and gap caches. With the fix applied (classification on a CurveOnClassifyCache, copied back with apply_uv_trim_updates, as the fillet session does) plus a probe, the five units that reach CreateRotationalSweep gave 23 axis checks, all in sweep.api_brep, and none of them changed the B-rep: no UV trims, no tolerances, no gap caches. All five units pass with and without it.

With no case to prove it on main, I'm not proposing the change. The patch is kept locally in case a fixture ever reaches this path with edits.

🤖 Generated with Claude Code