Manifold STLs for unions of stacked rings; loon prisms instead of a lying segment count

[ecto]

Two defects found while 3D-printing real parts from loon models. One is fixed; the second is diagnosed, bisected, and captured as an #[ignore]d test rather than left as folklore.

1. Non-manifold exports from unions of stacked rings

Exported STLs carried hundreds to thousands of edges not shared by exactly two triangles. Bambu Studio reports them ("3701 non-manifold edges") and its auto-repair resolves them by filling — a printed rotor came out with no shaft hole at all. Welding vertices did not move the count (flat from 1e-4 to 1e-1 tolerance), so it is topology, not float precision.

The reported repro never reached the mesh fallback. It came out of the analytic B-rep path, from two bugs in the planar circle splitter, both about a face's holes:

1. Hole routing by centroid. When a planar face was split by a circle, its existing hole loops were assigned to the disk or ring sub-face by testing the hole's centroid. Every concentric loop — whatever its radius — has its centroid at the shared center, so a hole larger than the splitting circle went to the disk: a face whose hole is bigger than its own outer loop (an r24 disk carrying an r27.5 hole), which the tessellator draws as the full disk — a membrane over the bore. Replaced with a real containment test on the loop's vertices (loop_vs_circle: Inside / Outside / Coincident / Straddles).
2. A circle lying inside one of the face's holes counted as "inside the face" — only the outer loop was consulted — splitting off a disk over the hole plus a redundant nested hole on the ring.

Nested annular caps are exactly what a union-of-differences produces, which is why this only showed up on stacked rings.

Measured

case	before	after
reported repro pipe(ann(8,46,2.5), union(ann(24,27.5,1.5)))	1902 bad edges	0, volume 16965.4 vs 16965.8 analytic truth
backplate.loon	2063	22 — and all 22 are unpaired hairline seams, no doubled surface
carrier.loon	manifold	manifold

Measurement, not repair. MeshReport / BooleanReport gain overused_edges. The existing open_edges is a net directed count, so it cancels to zero on a doubled surface — it could not see this defect at all.

I first wired that up as a fallback trigger (swap in the mesh boolean whenever the B-rep result had over-used edges) and reverted it after CI caught the cost. The claim it rested on — that an over-used edge is never legitimate geometry — is false: instrumenting this crate's own catalogue shows known-good results scoring up to 13 over-used edges, b1's blade cut scoring 1, against 11 and 14 for the defects that motivated the measurement. The populations overlap exactly the way they do for open edges, and the swap did what the comment above it warns about: traded b1's analytic r45 wall for coarse soup, −505 mm³. So the count is reported and documented as advisory, and the repair stays where this PR actually puts it — at the root, in the splitters.

The mesh CSG also cancels coincident triangle pairs (zero-thickness flaps, which add no volume and so slip past the volume oracle).

2. cylinder-n r h 6 silently rendering a circle

[cylinder-n 7.5 24.0 6] looks like a hex boss in source review and is not one: the third argument is a fidelity hint for the boolean/seam machinery, the surface stays an analytic cylinder, and the tessellator draws a circle. That shipped a "hex drive" that was a plain round bore and cost a print.

Worse, for a bare primitive the count never reaches tessellation at all — a cylinder exports at 32 segments whether the source says 8, 16, 48 or 96 — while lib.loon told authors to reach for these forms when bore fidelity is load-bearing.

The -n forms now fail closed under 8 and name the primitive that does make facets:

CylinderN: segments must be at least 8, got 6. The count is a fidelity hint for
booleans and seams — the surface stays a true circle, so a low count does NOT make
a faceted prism. For real facets use [prism sides radius height] (or [hex-prism
across-flats height]); to let the kernel choose the fidelity, drop the -n form.

The kernel already had a true faceted Prism, so this adds polygon-prism sides radius height and hex-prism across-flats height (across-flats — what a wrench or hex key measures; [hex-prism 13.0 24.0] verified at 13.0 mm flats, 12 verts / 20 tris), and rewrites the misleading lib.loon docstring.

Nothing in the repo or in the reporting model used a count below 8 (real usages are 16/20/24/48/64/96), so the floor breaks no existing source.

Known open defect, not fixed here

planet-72t (2133), ring-168t (1711), stator-proto (635) are unchanged — a second, distinct class: a boss whose top and bottom faces are flush with the body's goes non-manifold when its footprint straddles a line an earlier boolean split clear across that face. Gear teeth are the canonical case.

Bisection: one tooth on a fresh blank is manifold at every angle; a second tooth is manifold everywhere except a ~3° band around 180° from the first — the diameter opposite it, because the first union's side-plane lines cut the cap all the way across. The doubled patch has zero net volume (a flap), which is why the volume oracle passes it.

Captured in coplanar_boss_straddling_a_split.rs with the full diagnosis: one #[ignore]d failing test, and one passing test proving the arrangement is representable (0.5 mm of overhang takes a 72-tooth blank from 1286 to 31 bad edges). Workaround for models today: give bosses a small overhang past the body's faces. On the full planet-72t that only gets 2133 → 922, so that part has further sources I did not chase.

The root fix is in the pipeline, not the splitters this PR touches: either stop the line splitter cutting a face clear across where the other solid's face does not reach, or classify coplanar contact per-fragment across sub-faces.

Reviewer notes

• DegradeReason gains a NonManifoldSwap variant and BooleanReport an overused_edges field — additive, but they are public API.
• New regression tests: nested_annulus_union_manifold.rs (3 cases: coplanar contact, interpenetrating, boss over the bore — each asserting manifoldness and volume, since a membrane over a bore is invisible to a volume check alone until it is filled).
• cargo test green across vcad-kernel-booleans, vcad-eval, vcad-kernel, vcad-kernel-tessellate, vcad-loon (38 test binaries, 0 failures); clippy and cargo fmt --check clean.

🤖 Generated with Claude Code

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

4 Skipped Deployments

Project	Deployment	Actions	Updated (UTC)
mecheval	Ignored		Aug 28, 2026 12:21pm
vcad	Ignored		Aug 28, 2026 12:21pm
vcad-docs	Ignored		Aug 28, 2026 12:21pm
vcad-mcp	Ignored		Aug 28, 2026 12:21pm