Dependency Graph

Introduction

This library will help navigate a directed acyclic graph (DAG).

The goal of this library is to make it easy to:

• Find a specific node within a graph.
• Find all nodes on a certain level of the graph.
• Find all nodes between two levels of the graph.
• Find all nodes related to a given node. (i.e. find its decedents and/or its precedents on any level of the graph.)
• Sort the nodes in topological order

Naming Convention

• Descendants. i.e. What descends from or comes after: Child
• Precedents. i.e What precedes, or comes before: Parent
• Level. We consider level as topological level of the node. I.e. Level 1 consists of nodes whose Precedents are of Level 0. In general level is the longest path from the node to the root of the graph.

Interface

Graph

    public interface IGraph<T>
        where T : IEquatable<T>
    {
        INode<T> Find(T value); 

        void AddSequence(IEnumerable<T> sequence);
        void AddSequences(IEnumerable<IEnumerable<T>> sequences);

        IEnumerable<INode<T>> Nodes { get; }

        bool CanSort();
        bool CanSort(IEnumerable<T> sequence);

        int CountNodes { get; }
        int CountLevels { get; }
        IOrderedEnumerable<INode<T>> GetNodes(int level);
        IOrderedEnumerable<INode<T>> GetNodesBetween(int levelFrom, int levelTo);
        IOrderedEnumerable<INode<T>> Sort();
    }

Note:

• AddSequence(IEnumerable<T> sequence): the sequence must contain at least 2 items. The relationship between the items is automatically assumed as item1 -> item2 -> item3 etc...

Node

    public interface INode<T>
        where T: IEquatable<T>
    {
        T Value { get; }
        IGraph<T> Graph { get; }
        int Level { get; }

        // relativeLevel >= relativeLevelFrom && relativeLevel <= relativeLevelTo
        IOrderedEnumerable<INode<T>> GetNeighbours(int relativeLevelFrom, int relativeLevelTo);
        // relativeLevel < 0
        IOrderedEnumerable<INode<T>> Precedents { get; }
        // relativeLevel > 0
        IOrderedEnumerable<INode<T>> Descendants { get; }
        // relativeLevel == relativeLevel - 1
        IOrderedEnumerable<INode<T>> ImmediatePrecedents { get; }
        // relativeLevel == relativeLevel + 1
        IOrderedEnumerable<INode<T>> ImmediateDescendants { get; }
        // Precedents of the node without precedents (roots)
        IOrderedEnumerable<INode<T>> TerminatingPrecedents { get; }
        // Descendants of the node without descendants (leafs)
        IOrderedEnumerable<INode<T>> TerminatingDescendants { get; }
    }

Note:

• All the methods return an ordered enumerable of INode. The ordering is based on the "level" of the node. (Within a level the ordering is not important.)
• If possible the methods returns all the INodes lazily.
• A Node object has a reference to the Graph object.

Algorithms, Time Complexity

The Dependency Graph is a static data structure. All the nodes and their relationships should be known ahead of time.

Method Names	Algorithms	Time Complexity
AddSequence()	-	O(1)
AddSequences()	-	O(1)
Sort()	Topological Sort	O(V+E)
CanSort()	Topological Sort	O(V+E)
ComputeLevels()	Critical Path, DFS	O(E+V)
CountNodes()	-	O(1)
CountLevels()	-	O(1)
GetNodesWithLevel()	DFS	O(V+E)
GetNodesWithLevelBetween()	DFS	O(V+E)
Precedents()	DFS	O(V+E)
Descendants()	DFS	O(V+E)
ImmediatePrecedents()	DFS	O(V+E)
ImmediateDescendants()	DFS	O(V+E)
TerminatingPrecedents()	DFS	O(V+E)
TerminatingDescendants()	DFS	O(V+E)

ComputeLevels private method

ComputeLevels method performs initial pre-calculation (e.g. pre-calculate levels for nodes) Graph will be rebuild automatically on first call of any method related to node levels after a graph structure change.

1. Find the longest path. Critical path method O(V+E)
2. DFS from the source of the longest path, decrementing the level value for every child DFS - O(V+E)

Example

#####NOTE:

• The root nodes are 11 and 12.
• The leaf nodes are 61 and 62
• This graph has 6 levels.
• The root nodes have a level value equal to 0

Create Graph Structure

	new Graph(new []
	{
		new[] {11, 27, 32},
		new[] {12, 27},
		// etc....
	});

Or

    var graph = new Graph();
    graph.AddRange(new []
    {
        new[] {11, 27, 32},
        new[] {12, 27},
		// etc....
    });

Interaction

	[Test]
	public void BasicOperationsTest()
	{
		var graph = CreateExampleGraph();

		Assert.IsTrue(graph.CanSort());

		Assert.AreEqual(20, graph.Count);

		Assert.IsTrue(graph.CanSort());
			
		Assert.AreEqual(6, graph.CountLevels);
            
		AssertCollectionsConsistsOfNodes(new[] {31}, graph.GetNodes(4));

		AssertCollectionsConsistsOfNodes(new[] {51, 61, 62}, graph.GetNodesBetween(4, 5));

		AssertCollectionsConsistsOfNodes(new[] {11, 12, 25, 26, 27}, graph.Find(32).Precedents);

		AssertCollectionsConsistsOfNodes(new[] {51, 61, 62}, graph.Find(43).Descendants);

		AssertCollectionsConsistsOfNodes(new[] {25, 26, 27}, graph.Find(32).ImmediatePrecedents);

		AssertCollectionsConsistsOfNodes(new[] {51}, graph.Find(43).ImmediateDescendants);

		AssertCollectionsConsistsOfNodes(new[] {11, 12}, graph.Find(32).TerminatingPrecedents);

		AssertCollectionsConsistsOfNodes(new[] {61, 62}, graph.Find(43).TerminatingDescendants);
	}

Things To Think About

• How to return all nodes between two levels that relate to a certain node.

	GetNodesRelatedTo(T value, int minLevel == 0, int maxLevel == max)

	graph.GetNodesRelatedTo(11, 1, 3) => new[]{27, 32, 46}
	graph.GetNodesRelatedTo(32, 0, 3) => new[]{11, 12, 25, 26, 27, 32, 46}

• Node.GetNext()
• Node.GetPrevious()

POIs

• There are some ways how we can improve CanSort(sequence) method:
• We can copy graph much faster if we will find relations using temporary array and node.Key.
• We can track changes, which were made to graph and UnDo them after the sorting.

Links

• http://msdn.microsoft.com/en-us/library/ms379574(v=vs.80).aspx
• http://en.wikipedia.org/wiki/Critical_path_method