[raykholo]

Fixes #3951

A closed B-spline edge is split at its vertex parameter. The reader treated a vertex within span * 1e-7 of the curve's start as "at the start", but truck's knot-range check is absolute (1e-6), so a vertex between those two (2e-7 or 8e-7 on a 0 … 1 curve) produced a piece with a zero knot range. Evaluating that piece panicked with ZeroRange and the whole body was left out.

The reader now uses max(TOLERANCE, span * 1e-7) for recognising a vertex at either end, for trimming and as the minimum piece range. Ranges are checked before cutting and again on each returned piece; a range that still cannot be evaluated is reported as a degenerate spline edge instead of panicking. The change stays in the STEP reader (crates/kernel/src/step_in/geom.rs).

Before and after

The 43-line STEP from the issue (one planar face bounded by one closed degree-1 B-spline edge round a 20 mm square, vertex at parameter 2e-7), solidcraft-cli eval closed-edge-vertex-after-start.step:

Before (current main), nothing is imported, so there is nothing to render:

thread 'main' panicked at vendor/truck-geometry/src/nurbs/bspcurve.rs:301:14:
called `Result::unwrap()` on an `Err` value: ZeroRange
error: closed-edge-vertex-after-start.step: the operation failed: STEP import: no bodies (solid 'square': not imported: kernel internal error: step brep: called `Result::unwrap()` on an `Err` value: ZeroRange)

After (this branch): one body square, 1 face (plane), 1 edge, area 400.0 mm², bounding box (0, 0, 0) … (20, 20, 0).

solidcraft-cli snapshot closed-edge-vertex-after-start.step on this branch: the 20 × 20 mm face is read, iso view.

Tests

Two regression tests in the existing STEP reader test module:

• the issue's square with the vertex at t = 2e-7, 8e-7, 0 and 1 − 8e-7, both edge orientations: one face, 400 mm² each;
• tiny trims at several parameter scales, plus a closed curve too short to split, which now reports an error instead of panicking.

Both fail on main and pass here.

Validation (Linux x86_64)

• cargo fmt --all -- --check: clean.
• cargo clippy -p solidcraft-kernel --all-targets -- -D warnings: clean.
• cargo test --profile ci -p solidcraft-kernel --lib -- step_in: 64 passed, 2 manual tests ignored.

[raykholo]

Checked against the public model in tangency-labs/models#22 (five surface bodies, each bounded by one closed spline edge whose vertex sits 1e-9, 3e-8, 2e-7, 8e-7 or 5e-6 mm past the curve's start):

	bodies kept	area
0.18.1	3 of 5 (2e-7 and 8e-7 lost to ZeroRange)	—
this PR	5 of 5	400 mm² each

The file attached to #3951 also gives 1 body, 400 mm², no panic. An open edge that crosses a closed curve's seam (starting 3e-8 … 5e-6 before the curve's end) already reads correctly on 0.18.1, so the problem is limited to closed edges.

[jyviko]

Thank you for this work! We're evaluating a new geometry kernel (solidcore), so kernel pull requests are on hold for now and I've set this one to draft: see #3992.

Nothing is lost: your branch, commits and review history stay exactly as they are. The case you fixed here goes into the new kernel's regression suite, so it stays fixed whatever kernel we end up on. And if you'd like to keep going on this kind of problem, we'd love your help in solidcore, starting with porting this case over as a test.