Depiction backends

Structural SVG has two independent halves: where atoms sit and how they are drawn. AsciiChem owns the drawing (bonds, labels, stereo marks) and delegates placement to a depiction backend, so a stronger layout engine can be plugged in behind the model without touching the renderer.

AsciiChem::Layout.backend = MyBackend.new
formula.to_structural_svg   # drawn from MyBackend's positions

The backend contract

A backend responds to layout(molecule, algorithm:) and returns a Layout::Result:

class MyBackend
  def layout(molecule, algorithm: 'layered')
    AsciiChem::Layout::Result.new(
      atoms: [AsciiChem::Layout::PositionedAtom.new(
        id: 1, element: 'C', x: 10.0, y: 20.0
      )],
      bonds: [],
      width: 100.0, height: 100.0
    )
  end
end

Positions are plain numbers in the drawing’s coordinate space; the renderer scales to the SVG viewport. Atom ids must match the ids the walker assigns, exactly as the built-in pipeline does.

The built-in backend

AsciiChem::Layout::BuiltinBackend is the default: authored coordinates when the model has them (CML x2/y2), otherwise an elkrb layered layout that is deterministic across runs — the property visual regression goldens depend on.

When to plug in another

• A heavier depiction engine (RDKit, CDK, or a Rust layout core) produces better coordinates for crowded structures.
• A domain-specific arrangement (crystal packing, polymer repeat units) needs placement rules elkrb does not know.

Assigning Layout.backend is process-global; tests and servers that swap it should restore the default (Layout.backend = nil).

The model, the formatters, and the wire form are unchanged by a backend swap — which is the point: depictions are views.