feat(procedural-items): revolve shapes (AK-03c)

[Aymericr]

AK-03c: revolve shapes

Stacked on #951 (AK-03b) → … → #945. Review the top commit only. This is part 3 of AK-03; sweep is deferred to B-05 (see Seam).

What. A v2 revolve shape: { primitive: 'revolve', profile: [[radius, height], …], segments?, arc?, position, rotation?, slot }.

• Profile. 2–64 points, and each coordinate may be an expression. The profile turns about local Y, starting from +Z toward +X like the cylinder. segments (default 24) and arc (default 2π) are the cylinder fields.
• Closed or open.
  • The shape is a closed solid when the profile meets the axis at both ends over a full turn. The builder orients the profile so the solid's normals face outward.
  • Otherwise it is an open surface, such as a lamp shade, and it also draws its back face. Materials are front-sided because WebGPU MRT forces DoubleSide to FrontSide.
  • The same rule is added to feat(procedural-items): cylinder segments, open, inner and arc (AK-03a) #950: an open solid cylinder now draws both wall faces (separate commit on that branch).
• Evaluation.
  • Points that collapse onto the previous one are dropped, so a select-driven feature can shrink to zero length.
  • The profile is re-centred vertically in its box, which keeps the oriented-box invariant.
  • revolveIsClosed() is exported for the builder and the tests.
• Triangles are analytic: 2 · segments · (points − 1), doubled for open surfaces, which matches LatheGeometry's index.

Why. Audit limit #1: +3.5 % of /next (256 elements) needs lathe: pool details, openings, devices, pendants, balusters. It is also trial blocker 6: an ellipsoid only approximates E7's bell shade, and E8's turned balusters had to be boxes.

Failing case first. Before the fix, core/src/procedural-items/revolve.test.ts failed 5 of 5 tests (scratch/AK-CORE/03c-failing-before.txt).

Proof. The trial's own recipes are built with this builder and scored with the trial's compare.py against the /next source meshes (scratch/AK-CORE/proofs/e7-e8.out, e8-balusters.out).

Element	Candidate	R@2 mm	R@5 mm	R@2 cm	P@2 cm	Triangles
E7 five-drop pendant (whole element)	ellipsoid shades (before)	0.444	0.689	0.895	0.914	8,172
	revolved measured bell	0.795	1.000	1.000	1.000	7,932
E8 turned balusters (baluster region)	boxes (before)	0.190	0.287	1.00	1.00	192
	revolve, 12 segments	0.434	0.775	1.00	1.00	4,932

• E8 authoring. Each baluster is one revolve: a foot, two beads and the shaft. Each baluster's foot, bead offsets and length come from select(index, …) over values read from the source occurrences (e8-baluster-params.json), the way a converter would read them. At 2 cm both E8 candidates are already perfect, so the gain shows only at 2–5 mm.
• Tests.
  • nodes geometry.test.ts: closed baluster, reversed baluster, open shade, half-turn shade, quarter-turn solid. Built triangles equal the evaluated count, the height matches the box, and every side-wall normal on a closed solid points outward.
  • core: box and centring, open versus closed counts, expression points, collapse, validation.
• bun test src (core): 3017 pass. bun test src/procedural-item (nodes): 53 pass. turbo check-types for core, nodes and editor: pass. bun run check: clean.

Seam for B-05 (sweep, deferred). core/procedural-items/section.ts (#951) turns any resolved F1 section into rings. A future sweep shape takes { section, path: [Vec3], ends?: [SweepEndSpec, SweepEndSpec] }, evaluates the section exactly as extrude does, and emits the same rings along the path. Its triangle count follows from rings × path edges. Nothing in this PR constrains that shape.

House effect: no change. The /next scenes have no procedural-item nodes; the E7 and E8 gains land when converters write v2.

Rule	Check
R1	Additive primitive. segments and arc now also apply to revolve
R2	LatheGeometry runs once per (recipe, params) and is cached (0.25 ms at 64×40, CHALLENGE lpb)
R5	Oriented-box invariant, so handles, bounds, paint and surfaces are unchanged
R3, R4, R6–R9	Not touched

Local preview

Shared setup for the whole stack: previews/AK.md (preview.sh start AK-PV previews/AK/ak-after.json serves on :3301).
Setup is in previews/AK.md.

What you will notice

• The balusters are turned, with a foot, two beads and a shaft. Before, they were square boxes.
• The pendant shades are bell shapes revolved from the measured /next profile. Before, they were ellipsoids.
• The counter's steel basin is revolved.

Where to look

Pose 5-balusters; poses 1-overview and 2-ceiling for the pendant.

Before / after

previews/AK/before-5-balusters.png / after-5-balusters.png, and before-1-overview.png / after-1-overview.png.

Checked by me: the static renders at the listed poses (captured from :3100 and :3301). I did not click through the interactive steps in a browser; unit tests cover them.

Known limits

• A revolve that does not close on its axis draws both faces; materials are front-sided.

For Wassim to review:

1. The closed-versus-open rule and back faces.
2. Dropping coincident points rather than refusing them.
3. [radius, height] point order: radius first, like three's Vector2(x = r, y).

🤖 Generated with Claude Code

---

Note

Medium Risk
New v2 geometry path touches recipe validation, bounds/footprints, triangle budgets, and mesh winding; behavior is covered by tests but affects procedural item authoring and rendering.

Overview
Adds a recipe version 2 revolve shape: authors supply a 2–64 point profile of [radius, height] pairs (expressions allowed), optional segments and arc like cylinders, spun about local Y.

Core (recipe.ts). The primitive joins the shape union with validation (v2-only, no support surfaces, segments/arc on revolve). Evaluation resolves the profile, drops coincident points, centres it vertically in the oriented box, and exports revolveIsClosed() for closed solids (axis at both ends + full turn). Triangle counts, footprints (for ceiling cuts / motion), and v2 feature detection are extended accordingly.

Rendering (geometry.ts). Builds meshes with LatheGeometry, fixes profile winding for outward normals on closed solids, and duplicates flipped geometry for open or partial surfaces so front-sided materials still show both sides. Revolves use v2 world-scale UVs; flipped() is adjusted for already non-indexed geometry.

Tests. New revolve.test.ts covers evaluation, validation, and review fixes; geometry.test.ts checks built triangles vs budget, normals, and profile fidelity.

^{Reviewed by Cursor Bugbot for commit 0dddbbe. Bugbot is set up for automated code reviews on this repo. Configure here.}

Codex review round (rev-952-a)

Failing test first: test(... AK-03c, failing), then the fix.

• Revolve footprints follow their rings, so they fit a circle cut.
• A revolve rejects a stray radius (strict per-primitive schema).
• There is one LatheGeometry per revolve. Open surfaces reuse it flipped for back faces.
• UVs are face-frame and world-scale. Vertices match the profile exactly.

[pascal]

I hit an error while handling your request (Model unavailable on AI Gateway free tier: Free tier users do not have access to this model. Upgrade to paid credits at https://vercel.com/d?to=%2F%5Bteam%5D%2F%7E%2Fai%3Fmodal%3Dtop-up for unrestricted…).

Please try again, rephrase, or reach out if it keeps failing.

Error id: 0872f387-67a0-4dc1-8a2a-4748540b9be0

[Aymericr]

Landed on main via #971 (merge commit cf874a1), which contains this branch's commits and history.