IChartDecomposition

Remarks

docs/plan/42 § D3's plug point, and the only one this library has. The charter is PartUV's shape — decompose, flatten, accept or recurse, merge back — and the one part of PartUV that cannot be taken is the top of it, because PartField is a learned decomposition. So the recursion is kept and the decomposition is made replaceable, with a classical concavity-driven default that owes nobody anything.

⚠ The default path never calls an implementation of this. Leaving Decomposition null selects the built-in decomposition, which is an approximate convex split over the dual graph weighted by dihedral concavity and surface occlusion. This interface exists so that a learned part field can be dropped in behind it under § D14's rules — it proposes and never decides, because whatever comes back is still flattened, still measured, and still has to pass DistortionThreshold before it is kept. A bad decomposition costs chart quality and can never cost validity.

⚠ An implementation may decline, and declining is not a failure. Returning null, an empty list, or anything that does not describe at least two non-empty parts falls back to the built-in decomposition for that region alone. A proposer that only understands some shapes is a useful proposer.

Methods (1)

• IReadOnlyList<int>? Decompose(EditMesh mesh, IReadOnlyList<int> faces, int parts)

  Breaks a region of faces into parts.

Used by (4)

• AlternatingVixen.Geometry.Uv.Tests
• CharterVixen.Geometry.Uv
• DecliningVixen.Geometry.Uv.Tests
• UvSettingsVixen.Geometry.Uv