digraph-js

Make Directed Graphs traversal and construction effortless, also includes deep circular dependency detection.

digraph-js is a lightweight library allowing you to create a Directed Acyclic Graph data structure with embedded features such as deep cycle dependency detection and graph introspection (find deeply ancestors and successors for any given vertex). It can be used to model complex dependencies systems based on graphs.

✅ Create a graph structure including edges and vertices seamlessly

✅ Traverse graph, using Depth-first or Breadth-first searchs

✅ Deeply find direct/indirect children and parent dependencies of each vertex in the graph (top-to-bottom or bottom-to-top traversals)

✅ Ensure that a given graph is Acyclic by deeply detecting circular dependencies while having the possibility to limit the search depth

✅ Find precisely all vertices involved in cycles and sub-cycles

Installation

$ npm install digraph-js

How to use it

import { DiGraph } from "digraph-js";
import assert from "node:assert";

const myGraph = new DiGraph();

const myDependencyA = { id: "dependencyA", adjacentTo: [], body: {} };
const myDependencyB = { id: "dependencyB", adjacentTo: [], body: {} };
const myDependencyC = { id: "dependencyC", adjacentTo: [], body: {} };

// Add vertices to the graph
myGraph.addVertices(myDependencyA, myDependencyB, myDependencyC);

// Link graph vertices: A ---> B link created
myGraph.addEdge({ from: myDependencyA.id, to: myDependencyB.id });

// Graph traversels
myGraph.addEdge({ from: myDependencyB.id, to: myDependencyC.id });
// getDeepChildren traverses the graph in a Depth-First Search fashion
const deepDependenciesOfA = myGraph.getDeepChildren("dependencyA");
// deepDependenciesOfA is an iterable structure that can be lazily consumed
assert.deepEqual([...deepDependenciesOfA], ["dependencyB", "dependencyC"]);

// Here we voluntarily create a cyclic dependency
myGraph.addEdge({ from: myDependencyB.id, to: myDependencyA.id });
// Detect if the Directed Graph is acyclic (Directed Acyclic Graph)
assert.equal(myGraph.isAcyclic, false);
assert.equal(myGraph.hasCycles(), true);
assert.deepEqual(myGraph.findCycles().cycles, [["dependencyA", "dependencyB"]]);

// Limit cycles search or dependency depth
// Imagine a case where the cycle is created at depth 6
assert.equal(myGraph.hasCycles({ maxDepth: 5 }), false);
// Or that you want to get all children of a vertex but with a max depth of 5
// meaning that you don't want dependencies going over 5 generations
assert.equal(myGraph.getDeepChildren("dependencyA"), 5);


// Traversals

// Lazily pull vertices from the graph 
for(const vertex of myGraph.traverse({ traversal: "dfs" })) {
  console.log(vertex.id);
}

// Eagerly pull all the graph vertices at once
const graphVertices = myGraph.traverseEager({ traversal: "dfs" });
console.log(graphVertices.length);

You already manipulate Directed Graphs without knowing it

Take for instance the image above with four Vertices each representing a JavaScript file.

Now the question is: what are the relationships between these files? In all programming languages, one file might import one or multiple files. Whenever a file imports another one, an implicit relationship is created.

hello.js

export function sayHello() {}

main.js

import { sayHello } from "hello.js";

As you can see above, main.js imports hello.js to use the sayHello function. The static import creates an implicit relationship between both files. In the fields of graphs, this relationship can be modeled as a directed edge from main.js to hello.js (can be written as main.js ---> hello.js) We can also say that main.js depends on hello.js.

We can update the graph with our edges represented:

Basically this graph says that:

• FileA directly depends on FileD and FileB
• FileA indirectly depends on FileC (through both FileD and FileB)
• FileD directly depends on FileC
• FileB directly depends on FileC
• FileC directly depends on nothing

This structure may seem simple but can in fact be used to model very complex schemas such as:

• Static dependencies analysis such as cycle dependencies detection (e.g: ESLint no-cycle plugin)
• Incremental/Affected tasks Bundlers/Monorepos tools make extensive use of it (e.g: NX's affected build/test/lint...)
• Task orchestration using a directed acyclic graph, parallel vs sequential computations can be modeled (e.g: Continuous Integration schemas with stages, jobs, tasks)

Further exploring with examples which recreate common features:

• affected builds (NX alike)
• detect cyclic imports (eslint-plugin-import alike)
• Sequential vs Parallel execution => Work in progress