d3-dag

Lightweight, TypeScript-first DAG layout for the web. d3-dag provides layered graph layout algorithms for directed acyclic graphs.

Why d3-dag?

• Different layouts - optimal edge crossing minimization, multiple coordinate assignment strategies, and unique layout algorithms (Zherebko, Grid) not available elsewhere in JavaScript
• Small bundle - a fraction of elkjs's ~500KB transpiled Java
• TypeScript-first - full type safety with generic operators and immutable builders; dagre v3 added TypeScript support, but d3-dag's generic operator pattern provides deeper type safety
• Works with React Flow - drop-in replacement for dagre as a layout engine (see below)

Quick Start with React Flow

If you're using React Flow and want different layouts from dagre, d3-dag is a drop-in replacement:

import { dagre } from "d3-dag";

function getLayoutedElements(nodes, edges, direction = "TB") {
  const grf = new dagre.graphlib.Graph();
  grf.setGraph({ rankdir: direction });
  grf.setDefaultEdgeLabel(() => ({}));
  for (const node of nodes) {
    grf.setNode(node.id, {
      width: node.measured?.width ?? 172,
      height: node.measured?.height ?? 36,
    });
  }
  for (const edge of edges) grf.setEdge(edge.source, edge.target);
  dagre.layout(grf);
  return {
    nodes: nodes.map((node) => {
      const pos = grf.node(node.id);
      return {
        ...node,
        position: { x: pos.x - pos.width / 2, y: pos.y - pos.height / 2 },
      };
    }),
    edges,
  };
}
Copy

The dagre-compatible API supports familiar options (rankdir, nodesep, ranksep) with progressive disclosure into d3-dag's advanced algorithms:

import { dagre, sugiyama, decrossOpt, coordQuad } from "d3-dag";

const grf = new dagre.graphlib.Graph();
grf.setGraph({ rankdir: "LR", nodesep: 50, ranksep: 100 });
// ... add nodes/edges ...
dagre.layout(grf);

// Or upgrade to better algorithms
dagre.layout(grf, sugiyama().decross(decrossOpt()).coord(coordQuad()));
Copy

Migrating from dagre

Replace import dagre from "dagre" with import { dagre } from "d3-dag". Most graph construction and layout methods work the same.

Supported dagre API

Category	Methods
Setup	setGraph, graph, setDefaultNodeLabel, setDefaultEdgeLabel, isDirected, isCompound, isMultigraph
Nodes	setNode, setNodes, removeNode, hasNode, node, nodes, nodeCount, filterNodes
Edges	setEdge, removeEdge, hasEdge, edge, edges, edgeCount, setPath
Traversal	predecessors, successors, neighbors, inEdges, outEdges, nodeEdges, sources, sinks
Layout	dagre.layout(grf) with optional Operator (e.grf. sugiyama() or zherebko())

Unsupported dagre API

• Compound graphs: setParent, parent, children
• Serialization: json.write, json.read
• Algorithms: alg.* (use d3-dag's native operators instead)

Notable differences from dagre

• setGraph accepts additional quality, ranker, and algorithm options
• dagre.layout accepts an optional Operator for fine-grained algorithm control
• graph() returns a shallow copy of the config (mutations require setGraph)
• Default nodesep and ranksep are 50 (same as dagre)

Quality Presets

The dagre-compatible API accepts a quality option that controls the layout speed/quality trade-off:

grf.setGraph({ quality: "fast" }); // faster layout, lower quality
grf.setGraph();                  // default: "medium", better layout, slower
Copy

preset	description	genealogy (184 nodes)	vs dagre v3
"fast"	simplex layering, DFS decrossing, greedy coordinates	5.1 ms	~4x faster
"medium"	simplex layering, two-layer decrossing, simplex coordinates	49 ms	~2x slower
"slow"	simplex layering, optimal decrossing, simplex coordinates	small graphs only	—

The quality and ranker options only apply when using the sugiyama algorithm (the default). You can also set algorithm to "zherebko" or "grid" for alternative layouts:

grf.setGraph({ algorithm: "zherebko" }); // linear topological layout
grf.setGraph({ algorithm: "grid" });     // grid-based topological layout
Copy

The default ("medium") produces better layouts at the cost of being roughly 2x slower on large graphs. Use "fast" when layout speed matters more than visual quality, e.grf. for interactive or animated graphs.

dagre v3 added a customOrder callback and ordering constraints, allowing users to bring their own node ordering algorithm. d3-dag provides multiple built-in strategies (including ILP-optimal crossing minimization) out of the box, along with multiple coordinate assignment and layering algorithms not available in dagre.

Examples

• Sugiyama [examples] [api] - a layered layout
• Zherebko [examples] [api] - a linear topological layout
• Grid [examples] [api] - a grid based topological layout
• Dagre Comparison [docs] [api] - side-by-side comparison of d3-dag vs dagre

Installing

If you use node, npm i d3-dag or bun add d3-dag. Otherwise you can load it using unpkg:

<script src="https://unpkg.com/d3-dag@1.1.0"></script>
<script>
const dag = d3.graphStratify(...);
const layout = d3.sugiyama();
layout(dag);

// ... actually render here ...
for (const node of dag.nodes()) {
  console.log(node.x, node.y);
}
for (const { points } of dag.links()) {
  console.log(points);
}

</script>
Copy

General Usage Notes

This library is built around the concept of operators. Operators are functions with a fluent interface to modify their behavior. Every function that modifies behavior returns a copy, and will not modify the original operator. For example, the stratify operator creates dags from id-based parent data, can be used like so:

// note initial function call with no arguments to create default operator
const stratify = graphStratify();
const dag = stratify([{ id: "parent" }, { id: "child", parentIds: ["parent"] }]);

stratify.id(({ myid }: { myid: string }) => myid);
// doesn't work, stratify was not modified
const dag = stratify([{ myid: "parent" }, { myid: "child", parentIds: ["parent"] }]);

const myStratify = stratify.id(({ myid }: { myid: string }) => myid);
// works!
const dag = myStratify([{ myid: "parent" }, { myid: "child", parentIds: ["parent"] }]);
Copy

Updating

For information about changes between releases see the changelog.

Contributing

Contributions, issues, and PRs are all welcome!