Viewport

2D only. Despite the name, Viewport is not a 3D viewport — it owns a 2D pan/zoom transform, not a camera. For 3D, host your own WebGL/WebGPU canvas (for example inside an AppShell.Dock) and reach for Viewport only for 2D editor surfaces.

Viewport is the foundation primitive for editor-style 2D surfaces. It combines pan/zoom transform, perf-isolated canvas layers, world-anchored HTML, screen-space overlays, and marquee selection into a single composable primitive. Built to be the base for NodeGraph, Timeline, and similar surfaces.

Live Preview

Import

import {
  Viewport,
  ViewportLayer,
  ViewportWorld,
  ViewportOverlay,
  useViewportContext,
  worldToScreen,
  screenToWorld,
} from 'entangle-ui';

Quick start

<Viewport responsive>
  <ViewportLayer
    name="grid"
    draw={(ctx, { size, transform, theme }) => {
      // canvas drawing — runs only when transform/size/invalidateOn change
    }}
  />
  <ViewportWorld>
    <div style={{ position: 'absolute', left: 100, top: 200 }}>
      Node at world (100, 200)
    </div>
  </ViewportWorld>
  <ViewportOverlay>
    <Toolbar />
  </ViewportOverlay>
</Viewport>

Transform model

The viewport transform is a screen-space translation + uniform scale:

// screenPos = worldPos * zoom + (x, y)
type ViewportTransform = { x: number; y: number; zoom: number };

Use worldToScreen / screenToWorld (re-exported pure functions) to convert between spaces in your own code.

Controlled vs uncontrolled

// Controlled
const [transform, setTransform] = useState({ x: 0, y: 0, zoom: 1 });
<Viewport transform={transform} onTransformChange={setTransform} />

// Uncontrolled
<Viewport defaultTransform={{ x: 0, y: 0, zoom: 1 }} onTransformChange={log} />

Zoom is automatically clamped to [minZoom, maxZoom] (defaults 0.1–8).

Gestures

Pan — middle-mouse drag by default. Hold Space + left-drag for laptop-friendly pan.
Zoom — wheel scroll (zooms toward cursor). Trackpad pinch is supported via the browser’s ctrl+wheel convention.
Marquee — opt-in via selectionRect={{ enabled: true }}. Shift-drag for additive selection.

<Viewport
  pan={{ button: 'middle', spaceKey: true }}
  zoom={{ wheel: true, pinch: true, speed: 0.0015 }}
  selectionRect={{ enabled: true, button: 'left', additiveModifier: 'shift' }}
  onSelectionChange={info => {
    // info.rect is in world coordinates, normalized
    // info.additive is true when the modifier was held
    // info.inProgress is false on the final pointerup
  }}
/>

Set pan={false} or zoom={false} to disable a gesture family entirely.

Layers (perf isolation)

Each <ViewportLayer> is an independent <canvas> with its own draw cycle. A layer redraws when:

The viewport transform or size changes,
Any value in its invalidateOn array changes, or
handle.invalidate(layerName) is called.

Stack layers in JSX order — earlier = behind:

<Viewport>
  <ViewportLayer
    name="grid"
    draw={drawGrid}
    invalidateOn={[gridStep]} // only redraws when gridStep changes (+ transform/size)
  />
  <ViewportLayer
    name="content"
    draw={drawNodes}
    invalidateOn={[nodes]} // unaffected by gridStep changes
  />
</Viewport>

The draw callback receives:

draw(ctx, {
  size, // { width, height } in CSS px
  transform, // current transform
  theme, // resolved theme colors (CanvasThemeColors)
  worldToScreen, // helper bound to current transform
  screenToWorld, // inverse helper
});

DPR scaling is applied automatically — draw using CSS-pixel coordinates.

World vs Overlay

<ViewportWorld> — HTML children positioned in world coordinates. The container is transformed by translate(x, y) scale(zoom), so children with position: absolute and world-space left/top follow pan/zoom for free.
<ViewportOverlay> — HTML children in screen coordinates, rendered above all canvas layers and world children. Use for toolbars, minimap, status text.

Both wrappers default to pointer-events: none; child elements opt in with pointerEvents: 'auto' so the Viewport’s pan/zoom/marquee gestures keep working on the background.

Imperative API

<Viewport> accepts a ref to ViewportHandle:

const ref = useRef<ViewportHandle>(null);
ref.current?.fitToContent({ x: 0, y: 0, width: 400, height: 300 }, 32);
ref.current?.zoomToRect({ x: 100, y: 100, width: 50, height: 50 });
ref.current?.centerOn({ x: 200, y: 150 }, 2);
ref.current?.invalidate('grid'); // force redraw of one layer
ref.current?.getTransform();
ref.current?.getSize();

Reading state from children — `useViewportContext`

Any child rendered inside <Viewport> can read live state:

function CursorReadout() {
  const { transform, size, isPanning } = useViewportContext();
  return (
    <span>
      zoom: {transform.zoom.toFixed(2)}x · {size.width}×{size.height}
      {isPanning ? ' · panning' : ''}
    </span>
  );
}

Use this in custom Overlay components (minimap, HUD, etc.) instead of prop-drilling transform.

Recipes — building features we intentionally left out of v1

Viewport ships mechanisms, not styles. The following features were deliberately excluded from the v1 surface because they’re either opinionated (inertia, snap) or scene-specific (minimap). Each can be implemented in user code with the primitives Viewport does provide.

Snap-to-zoom

Snap the zoom level to discrete stops as the user scrolls. All you need is controlled transform:

const ZOOM_STOPS = [0.25, 0.5, 1, 2, 4];

function snapZoom(z: number) {
  return ZOOM_STOPS.reduce((a, b) =>
    Math.abs(b - z) < Math.abs(a - z) ? b : a
  );
}

const [t, setT] = useState({ x: 0, y: 0, zoom: 1 });

<Viewport
  transform={t}
  onTransformChange={next => setT({ ...next, zoom: snapZoom(next.zoom) })}
/>;

The same shape works for snapping x/y translation to a grid step.

Minimap

A minimap is just an <ViewportOverlay> child that subscribes to viewport state via useViewportContext and renders its own small <canvas>. handle.centerOn makes the minimap clickable:

function Minimap({ worldBounds }: { worldBounds: WorldRect }) {
  const { transform, size, handle } = useViewportContext();
  const ref = useRef<HTMLCanvasElement>(null);

  useEffect(() => {
    const canvas = ref.current;
    if (!canvas) return;
    const ctx = canvas.getContext('2d');
    if (!ctx) return;

    // Scale world bounds into the minimap's box (e.g. 120x80)
    const scale = Math.min(120 / worldBounds.width, 80 / worldBounds.height);
    ctx.clearRect(0, 0, 120, 80);

    // Draw your scene at this scale (e.g. nodes as rectangles)
    // drawSceneScaled(ctx, scale)

    // Draw current viewport rectangle
    const viewportWorld = {
      x: -transform.x / transform.zoom,
      y: -transform.y / transform.zoom,
      width: size.width / transform.zoom,
      height: size.height / transform.zoom,
    };
    ctx.strokeStyle = 'cyan';
    ctx.strokeRect(
      (viewportWorld.x - worldBounds.x) * scale,
      (viewportWorld.y - worldBounds.y) * scale,
      viewportWorld.width * scale,
      viewportWorld.height * scale
    );
  }, [transform, size, worldBounds]);

  return (
    <canvas
      ref={ref}
      width={120}
      height={80}
      style={{ pointerEvents: 'auto', cursor: 'crosshair' }}
      onPointerDown={e => {
        const rect = e.currentTarget.getBoundingClientRect();
        const localX = e.clientX - rect.left;
        const localY = e.clientY - rect.top;
        const scale = Math.min(
          120 / worldBounds.width,
          80 / worldBounds.height
        );
        handle.centerOn({
          x: worldBounds.x + localX / scale,
          y: worldBounds.y + localY / scale,
        });
      }}
    />
  );
}

<Viewport>
  {/* layers + world... */}
  <ViewportOverlay>
    <div
      style={{
        position: 'absolute',
        bottom: 8,
        right: 8,
        pointerEvents: 'auto',
      }}
    >
      <Minimap worldBounds={{ x: 0, y: 0, width: 1000, height: 800 }} />
    </div>
  </ViewportOverlay>
</Viewport>;

Inertia (momentum scrolling)

onPanEnd reports the gesture’s end velocity in screen px/ms. Animate a decaying translation on top of the controlled transform:

function useInertia(handleRef: React.RefObject<ViewportHandle>) {
  const rafRef = useRef(0);

  const onPanStart = () => cancelAnimationFrame(rafRef.current);

  const onPanEnd = ({ velocity }: ViewportPanEndInfo) => {
    cancelAnimationFrame(rafRef.current);
    let { x: vx, y: vy } = velocity;
    let last = performance.now();
    const step = (): void => {
      const now = performance.now();
      const dt = now - last;
      last = now;
      vx *= 0.92;
      vy *= 0.92;
      if (Math.hypot(vx, vy) < 0.01) return;
      const t = handleRef.current?.getTransform();
      if (!t) return;
      handleRef.current!.centerOn(
        {
          x:
            -(t.x + vx * dt) / t.zoom +
            handleRef.current!.getSize().width / 2 / t.zoom,
          y:
            -(t.y + vy * dt) / t.zoom +
            handleRef.current!.getSize().height / 2 / t.zoom,
        },
        t.zoom
      );
      rafRef.current = requestAnimationFrame(step);
    };
    rafRef.current = requestAnimationFrame(step);
  };

  return { onPanStart, onPanEnd };
}

const ref = useRef<ViewportHandle>(null);
const inertia = useInertia(ref);
<Viewport ref={ref} {...inertia} />;

Replace the 0.92 constant with whatever deceleration curve fits your editor feel.

Why these aren’t built in

| Feature | Why it lives in user code | | ----------- | ---------------------------------------------------------------------------------------------------------------------------------------------------------------- | | Snap | Snap rules vary per editor (frame-rate stops in Timeline vs zoom presets in NodeGraph). One built-in rule would be opinionated; the recipe above is a few lines. | | Minimap | A minimap needs to know what to draw at small scale — that’s scene-specific. The Overlay slot + useViewportContext give you everything else. | | Inertia | Deceleration feel is a per-product decision (Figma vs Blender vs custom). onPanEnd exposes the velocity so any curve works. |

If you find yourself copy-pasting the same recipe across three components, that’s the signal to ask for it as a first-class prop.

API reference

`<Viewport>` props

interface ViewportProps {
  transform?: ViewportTransform;
  defaultTransform?: ViewportTransform;
  onTransformChange?: (t: ViewportTransform) => void;
  minZoom?: number; // default 0.1
  maxZoom?: number; // default 8
  pan?: ViewportPanConfig | false;
  zoom?: ViewportZoomConfig | false;
  selectionRect?: ViewportSelectionConfig;
  onSelectionChange?: (info: ViewportSelectionEvent) => void;
  onPanStart?: () => void;
  onPanEnd?: (info: ViewportPanEndInfo) => void;
  onZoomStart?: () => void;
  onZoomEnd?: () => void;
  responsive?: boolean; // default false
  height?: number; // default 300, ignored when responsive
  disabled?: boolean;
  role?: string;
  ariaLabel?: string;
  ariaRoledescription?: string;
  ref?: React.Ref<ViewportHandle>;
}

`<ViewportLayer>` props

interface ViewportLayerProps {
  name: string;
  draw: (ctx: CanvasRenderingContext2D, info: ViewportLayerDrawInfo) => void;
  invalidateOn?: ReadonlyArray<unknown>;
  paused?: boolean;
  className?: string;
}

`ViewportHandle`

interface ViewportHandle {
  fitToContent(bounds: WorldRect, padding?: number): void;
  zoomToRect(rect: WorldRect, padding?: number): void;
  centerOn(point: Point2D, zoom?: number): void;
  getTransform(): ViewportTransform;
  getSize(): ViewportSize;
  invalidate(layerName?: string): void;
}