How to sort depended objects by dependency

Question

I have a collection:

List> dependencyHierarchy;

The first item in pair is some object (item) and the

Answer 1

There's a nuget for that.

For those of us who prefer not to re-invent the wheel: use nuget to install the QuickGraph .NET library, which includes multiple graph algorithms including topological sort.

To use it, you need to create an instance of AdjacencyGraph<,> such as AdjacencyGraph<String, SEdge<String>>. Then, if you include the appropriate extensions:

using QuickGraph.Algorithms;

You can call:

var sorted = myGraph.TopologicalSort();

To get a list of sorted nodes.

Answer 2

I would make this easier on myself by storing the dependencies of an Item within the Item itself:

public class Item
{
    private List<Item> m_Dependencies = new List<Item>();

    protected AddDependency(Item _item) { m_Dependencies.Add(_item); }

    public Item()
    {
    }; // eo ctor

    public List<Item> Dependencies {get{return(m_Dependencies);};}
} // eo class Item

Then, given this you can implement a custom Sort delegate for List that sorts based on whether the given Item is contained within the other's list of dependencies:

int CompareItem(Item _1, Item _2)
{
    if(_2.Dependencies.Contains(_1))
        return(-1);
    else if(_1.Dependencies.Contains(_2))
        return(1);
    else
        return(0);
}

Answer 3

A different idea, for cases with only one "parent" depending:

Instead of deps, you'd store the parents.
So you can tell very easily whether an issue is a dependency of some other.
And then use Comparable<T>, which would claim the dependencies "lesser" and the dependency "greater".
And then simply call Collections.sort( List<T>, ParentComparator<T>);

For multi-parent scenario, a tree search would be needed which would result in slow execution. But that could be solved by a cache in a form of A* sort matrix.

Answer 4

Having struggled with this for a while, here's my attempt at a Linq style TSort extension method:

public static IEnumerable<T> TSort<T>( this IEnumerable<T> source, Func<T, IEnumerable<T>> dependencies, bool throwOnCycle = false )
{
    var sorted = new List<T>();
    var visited = new HashSet<T>();

    foreach( var item in source )
        Visit( item, visited, sorted, dependencies, throwOnCycle );

    return sorted;
}

private static void Visit<T>( T item, HashSet<T> visited, List<T> sorted, Func<T, IEnumerable<T>> dependencies, bool throwOnCycle )
{
    if( !visited.Contains( item ) )
    {
        visited.Add( item );

        foreach( var dep in dependencies( item ) )
            Visit( dep, visited, sorted, dependencies, throwOnCycle );

        sorted.Add( item );
    }
    else
    {
        if( throwOnCycle && !sorted.Contains( item ) )
            throw new Exception( "Cyclic dependency found" );
    }
}

Answer 5

I don't like recursive methods so DMM is out. Krumelur looks good but seems to use a lot of memory? Made an alternative stack based method that seems to work. Uses same DFS logic as DMM 's and I used this solutions as comparison when testing.

    public static IEnumerable<T> TopogicalSequenceDFS<T>(this IEnumerable<T> source, Func<T, IEnumerable<T>> deps)
    {
        HashSet<T> yielded = new HashSet<T>();
        HashSet<T> visited = new HashSet<T>();
        Stack<Tuple<T, IEnumerator<T>>> stack = new Stack<Tuple<T, IEnumerator<T>>>();

        foreach (T t in source)
        {
            stack.Clear();
            if (visited.Add(t))
                stack.Push(new Tuple<T, IEnumerator<T>>(t, deps(t).GetEnumerator()));

            while (stack.Count > 0)
            {
                var p = stack.Peek();
                bool depPushed = false;
                while (p.Item2.MoveNext())
                {
                    var curr = p.Item2.Current;
                    if (visited.Add(curr))
                    {
                        stack.Push(new Tuple<T, IEnumerator<T>>(curr, deps(curr).GetEnumerator()));
                        depPushed = true;
                        break;
                    }
                    else if (!yielded.Contains(curr))
                        throw new Exception("cycle");
                }

                if (!depPushed)
                {
                    p = stack.Pop();
                    if (!yielded.Add(p.Item1))
                        throw new Exception("bug");
                    yield return p.Item1;
                }
            }
        }
    }

Here is also a simpler stack based BFS variant. It will produce different result than the above, but still valid. I'm not sure if there is any advantage to using the DFS variant above, but it was fun creating it.

    public static IEnumerable<T> TopologicalSequenceBFS<T>(this IEnumerable<T> source, Func<T, IEnumerable<T>> dependencies)
    {
        var yielded = new HashSet<T>();
        var visited = new HashSet<T>();
        var stack = new Stack<Tuple<T, bool>>(source.Select(s => new Tuple<T, bool>(s, false))); // bool signals Add to sorted

        while (stack.Count > 0)
        {
            var item = stack.Pop();
            if (!item.Item2)
            {
                if (visited.Add(item.Item1))
                {
                    stack.Push(new Tuple<T, bool>(item.Item1, true)); // To be added after processing the dependencies
                    foreach (var dep in dependencies(item.Item1))
                        stack.Push(new Tuple<T, bool>(dep, false));
                }
                else if (!yielded.Contains(item.Item1))
                    throw new Exception("cyclic");
            }
            else
            {
                if (!yielded.Add(item.Item1))
                    throw new Exception("bug");
                yield return item.Item1;
            }
        }
    }

For .NET 4.7+ I suggest replacing Tuple with ValueTuple for lower memory use. In older .NET versions Tuple can be replaced with KeyValuePair.

Answer 6

I merged DMM's idea with the depth-first-search algorithm on Wikipedia. It works perfect for what I needed.

public static class TopologicalSorter
{
public static List<string> LastCyclicOrder = new List<string>(); //used to see what caused the cycle

sealed class ItemTag
{
  public enum SortTag
  {
    NotMarked,
    TempMarked,
    Marked
  }

  public string Item { get; set; }
  public SortTag Tag { get; set; }

  public ItemTag(string item)
  {
    Item = item;
    Tag = SortTag.NotMarked;
  }
}

public static IEnumerable<string> TSort(this IEnumerable<string> source, Func<string, IEnumerable<string>> dependencies)
{
  TopologicalSorter.LastCyclicOrder.Clear();

  List<ItemTag> allNodes = new List<ItemTag>();
  HashSet<string> sorted = new HashSet<string>(StringComparer.OrdinalIgnoreCase);

  foreach (string item in source)
  {
    if (!allNodes.Where(n => string.Equals(n.Item, item, StringComparison.OrdinalIgnoreCase)).Any())
    {
      allNodes.Add(new ItemTag(item)); //don't insert duplicates
    }
    foreach (string dep in dependencies(item))
    {
      if (allNodes.Where(n => string.Equals(n.Item, dep, StringComparison.OrdinalIgnoreCase)).Any()) continue; //don't insert duplicates
      allNodes.Add(new ItemTag(dep));
    }
  }

  foreach (ItemTag tag in allNodes)
  {
    Visit(tag, allNodes, dependencies, sorted);
  }

  return sorted;
}

static void Visit(ItemTag tag, List<ItemTag> allNodes, Func<string, IEnumerable<string>> dependencies, HashSet<string> sorted)
{
  if (tag.Tag == ItemTag.SortTag.TempMarked)
  {
    throw new GraphIsCyclicException();
  }
  else if (tag.Tag == ItemTag.SortTag.NotMarked)
  {
    tag.Tag = ItemTag.SortTag.TempMarked;
    LastCyclicOrder.Add(tag.Item);

    foreach (ItemTag dep in dependencies(tag.Item).Select(s => allNodes.Where(t => string.Equals(s, t.Item, StringComparison.OrdinalIgnoreCase)).First())) //get item tag which falls with used string
      Visit(dep, allNodes, dependencies, sorted);

    LastCyclicOrder.Remove(tag.Item);
    tag.Tag = ItemTag.SortTag.Marked;
    sorted.Add(tag.Item);
  }
}
}