Recursive List Flattening

Question

I could probably write this myself, but the specific way I'm trying to accomplish it is throwing me off. I'm trying to write a generic extension method similar to the others introduced in .NET 3.5 that will take a nested IEnumerable of IEnumerables (and so on) and flatten it into one IEnumerable. Anyone have any ideas?

Specifically, I'm having trouble with the syntax of the extension method itself so that I can work on a flattening algorithm.

Answer 1

Function:

public static class MyExtentions
{
    public static IEnumerable<T> RecursiveSelector<T>(this IEnumerable<T> nodes, Func<T, IEnumerable<T>> selector)
    {
        if(nodes.Any() == false)
        {
            return nodes; 
        }

        var descendants = nodes
                            .SelectMany(selector)
                            .RecursiveSelector(selector);

        return nodes.Concat(descendants);
    } 
}

Usage:

var ar = new[]
{
    new Node
    {
        Name = "1",
        Chilren = new[]
        {
            new Node
            {
                Name = "11",
                Children = new[]
                {
                    new Node
                    {
                        Name = "111",
                        
                    }
                }
            }
        }
    }
};

var flattened = ar.RecursiveSelector(x => x.Children).ToList();

Answer 2

Since yield is not available in VB and LINQ provides both deferred execution and a concise syntax, you can also use.

<Extension()>
Public Function Flatten(Of T)(ByVal objects As Generic.IEnumerable(Of T), ByVal selector As Func(Of T, Generic.IEnumerable(Of T))) As Generic.IEnumerable(Of T)
   If(objects.Any()) Then
      Return objects.Union(objects.Select(selector).Where(e => e != null).SelectMany(e => e)).Flatten(selector))
   Else
      Return objects 
   End If
End Function

public static class Extensions{
  public static IEnumerable<T> Flatten<T>(this IEnumerable<T> objects, Func<T, IEnumerable<T>> selector) where T:Component{
    if(objects.Any()){
        return objects.Union(objects.Select(selector).Where(e => e != null).SelectMany(e => e).Flatten(selector));
    }
    return objects;
  }
}

edited to include:

• empty enumerable per https://stackoverflow.com/a/30325216/107683,
• null enumerable per https://stackoverflow.com/a/39338919/107683
• C# implementation.

Answer 3

class PageViewModel { 
    public IEnumerable<PageViewModel> ChildrenPages { get; set; } 
}

Func<IEnumerable<PageViewModel>, IEnumerable<PageViewModel>> concatAll = null;
concatAll = list => list.SelectMany(l => l.ChildrenPages.Any() ? 
    concatAll(l.ChildrenPages).Union(new[] { l }) : new[] { l });

var allPages = concatAll(source).ToArray();

Answer 4

Okay here's another version which is combined from about 3 answers above.

Recursive. Uses yield. Generic. Optional filter predicate. Optional selection function. About as concise as I could make it.

    public static IEnumerable<TNode> Flatten<TNode>(
        this IEnumerable<TNode> nodes, 
        Func<TNode, bool> filterBy = null,
        Func<TNode, IEnumerable<TNode>> selectChildren = null
        )
    {
        if (nodes == null) yield break;
        if (filterBy != null) nodes = nodes.Where(filterBy);

        foreach (var node in nodes)
        {
            yield return node;

            var children = (selectChildren == null)
                ? node as IEnumerable<TNode>
                : selectChildren(node);

            if (children == null) continue;

            foreach (var child in children.Flatten(filterBy, selectChildren))
            {
                yield return child;
            }
        }
    }

Usage:

// With filter predicate, with selection function
var flatList = nodes.Flatten(n => n.IsDeleted == false, n => n.Children);

Answer 5

I had to implement mine from scratch because all of the provided solutions would break in case there is a loop i.e. a child that points to its ancestor. If you have the same requirements as mine please take a look at this (also let me know if my solution would break in any special circumstances):

How to use:

var flattenlist = rootItem.Flatten(obj => obj.ChildItems, obj => obj.Id)

Code:

public static class Extensions
    {
        /// <summary>
        /// This would flatten out a recursive data structure ignoring the loops. The end result would be an enumerable which enumerates all the
        /// items in the data structure regardless of the level of nesting.
        /// </summary>
        /// <typeparam name="T">Type of the recursive data structure</typeparam>
        /// <param name="source">Source element</param>
        /// <param name="childrenSelector">a function that returns the children of a given data element of type T</param>
        /// <param name="keySelector">a function that returns a key value for each element</param>
        /// <returns>a faltten list of all the items within recursive data structure of T</returns>
        public static IEnumerable<T> Flatten<T>(this IEnumerable<T> source,
            Func<T, IEnumerable<T>> childrenSelector,
            Func<T, object> keySelector) where T : class
        {
            if (source == null)
                throw new ArgumentNullException("source");
            if (childrenSelector == null)
                throw new ArgumentNullException("childrenSelector");
            if (keySelector == null)
                throw new ArgumentNullException("keySelector");
            var stack = new Stack<T>( source);
            var dictionary = new Dictionary<object, T>();
            while (stack.Any())
            {
                var currentItem = stack.Pop();
                var currentkey = keySelector(currentItem);
                if (dictionary.ContainsKey(currentkey) == false)
                {
                    dictionary.Add(currentkey, currentItem);
                    var children = childrenSelector(currentItem);
                    if (children != null)
                    {
                        foreach (var child in children)
                        {
                            stack.Push(child);
                        }
                    }
                }
                yield return currentItem;
            }
        }

        /// <summary>
        /// This would flatten out a recursive data structure ignoring the loops. The     end result would be an enumerable which enumerates all the
        /// items in the data structure regardless of the level of nesting.
        /// </summary>
        /// <typeparam name="T">Type of the recursive data structure</typeparam>
        /// <param name="source">Source element</param>
        /// <param name="childrenSelector">a function that returns the children of a     given data element of type T</param>
        /// <param name="keySelector">a function that returns a key value for each   element</param>
        /// <returns>a faltten list of all the items within recursive data structure of T</returns>
        public static IEnumerable<T> Flatten<T>(this T source, 
            Func<T, IEnumerable<T>> childrenSelector,
            Func<T, object> keySelector) where T: class
        {
            return Flatten(new [] {source}, childrenSelector, keySelector);
        }
    }

Answer 6

Here's an extension that might help. It will traverse all nodes in your hierarchy of objects and pick out the ones that match a criteria. It assumes that each object in your hierarchy has a collection property that holds its child objects.

Here's the extension:

/// Traverses an object hierarchy and return a flattened list of elements
/// based on a predicate.
/// 
/// TSource: The type of object in your collection.</typeparam>
/// source: The collection of your topmost TSource objects.</param>
/// selectorFunction: A predicate for choosing the objects you want.
/// getChildrenFunction: A function that fetches the child collection from an object.
/// returns: A flattened list of objects which meet the criteria in selectorFunction.
public static IEnumerable<TSource> Map<TSource>(
  this IEnumerable<TSource> source,
  Func<TSource, bool> selectorFunction,
  Func<TSource, IEnumerable<TSource>> getChildrenFunction)
{
  // Add what we have to the stack
  var flattenedList = source.Where(selectorFunction);

  // Go through the input enumerable looking for children,
  // and add those if we have them
  foreach (TSource element in source)
  {
    flattenedList = flattenedList.Concat(
      getChildrenFunction(element).Map(selectorFunction,
                                       getChildrenFunction)
    );
  }
  return flattenedList;
}

Examples (Unit Tests):

First we need an object and a nested object hierarchy.

A simple node class

class Node
{
  public int NodeId { get; set; }
  public int LevelId { get; set; }
  public IEnumerable<Node> Children { get; set; }

  public override string ToString()
  {
    return String.Format("Node {0}, Level {1}", this.NodeId, this.LevelId);
  }
}

And a method to get a 3-level deep hierarchy of nodes

private IEnumerable<Node> GetNodes()
{
  // Create a 3-level deep hierarchy of nodes
  Node[] nodes = new Node[]
    {
      new Node 
      { 
        NodeId = 1, 
        LevelId = 1, 
        Children = new Node[]
        {
          new Node { NodeId = 2, LevelId = 2, Children = new Node[] {} },
          new Node
          {
            NodeId = 3,
            LevelId = 2,
            Children = new Node[]
            {
              new Node { NodeId = 4, LevelId = 3, Children = new Node[] {} },
              new Node { NodeId = 5, LevelId = 3, Children = new Node[] {} }
            }
          }
        }
      },
      new Node { NodeId = 6, LevelId = 1, Children = new Node[] {} }
    };
  return nodes;
}

First Test: flatten the hierarchy, no filtering

[Test]
public void Flatten_Nested_Heirachy()
{
  IEnumerable<Node> nodes = GetNodes();
  var flattenedNodes = nodes.Map(
    p => true, 
    (Node n) => { return n.Children; }
  );
  foreach (Node flatNode in flattenedNodes)
  {
    Console.WriteLine(flatNode.ToString());
  }

  // Make sure we only end up with 6 nodes
  Assert.AreEqual(6, flattenedNodes.Count());
}

This will show:

Node 1, Level 1
Node 6, Level 1
Node 2, Level 2
Node 3, Level 2
Node 4, Level 3
Node 5, Level 3

Second Test: Get a list of nodes that have an even-numbered NodeId

[Test]
public void Only_Return_Nodes_With_Even_Numbered_Node_IDs()
{
  IEnumerable<Node> nodes = GetNodes();
  var flattenedNodes = nodes.Map(
    p => (p.NodeId % 2) == 0, 
    (Node n) => { return n.Children; }
  );
  foreach (Node flatNode in flattenedNodes)
  {
    Console.WriteLine(flatNode.ToString());
  }
  // Make sure we only end up with 3 nodes
  Assert.AreEqual(3, flattenedNodes.Count());
}

This will show:

Node 6, Level 1
Node 2, Level 2
Node 4, Level 3

Answer 7

I thought I'd share a complete example with error handling and a single-logic apporoach.

Recursive flattening is as simple as:

LINQ version

public static class IEnumerableExtensions
{
    public static IEnumerable<T> SelectManyRecursive<T>(this IEnumerable<T> source, Func<T, IEnumerable<T>> selector)
    {
        if (source == null) throw new ArgumentNullException("source");
        if (selector == null) throw new ArgumentNullException("selector");

        return !source.Any() ? source :
            source.Concat(
                source
                .SelectMany(i => selector(i).EmptyIfNull())
                .SelectManyRecursive(selector)
            );
    }

    public static IEnumerable<T> EmptyIfNull<T>(this IEnumerable<T> source)
    {
        return source ?? Enumerable.Empty<T>();
    }
}

Non-LINQ version

public static class IEnumerableExtensions
{
    public static IEnumerable<T> SelectManyRecursive<T>(this IEnumerable<T> source, Func<T, IEnumerable<T>> selector)
    {
        if (source == null) throw new ArgumentNullException("source");
        if (selector == null) throw new ArgumentNullException("selector");

        foreach (T item in source)
        {
            yield return item;

            var children = selector(item);
            if (children == null)
                continue;

            foreach (T descendant in children.SelectManyRecursive(selector))
            {
                yield return descendant;
            }
        }
    }
}

Design decisions

I decided to:

• disallow flattening of a null IEnumerable, this can be changed by removing exception throwing and:
  • adding source = source.EmptyIfNull(); before return in the 1st version
  • adding if (source != null) before foreach in the 2nd version
• allow returning of a null collection by the selector - this way I'm removing responsibility from the caller to assure the children list isn't empty, this can be changed by:
  • removing .EmptyIfNull() in the first version - note that SelectMany will fail if null is returned by selector
  • removing if (children == null) continue; in the second version - note that foreach will fail on a null IEnumerable parameter
• allow filtering children with .Where clause on the caller side or within the children selector rather than passing a children filter selector parameter:
  • it won't impact the efficiency because in both versions it is a deferred call
  • it would be mixing another logic with the method and I prefer to keep the logic separated

Sample use

I'm using this extension method in LightSwitch to obtain all controls on the screen:

public static class ScreenObjectExtensions
{
    public static IEnumerable<IContentItemProxy> FindControls(this IScreenObject screen)
    {
        var model = screen.Details.GetModel();

        return model.GetChildItems()
            .SelectManyRecursive(c => c.GetChildItems())
            .OfType<IContentItemDefinition>()
            .Select(c => screen.FindControl(c.Name));
    }
}

Answer 8

Here is a modified Jon Skeet's answer to allow more than "one level":

static IEnumerable Flatten(IEnumerable enumerable)
{
    foreach (object element in enumerable)
    {
        IEnumerable candidate = element as IEnumerable;
        if (candidate != null)
        {
            foreach (object nested in Flatten(candidate))
            {
                yield return nested;
            }
        }
        else
        {
            yield return element;
        }
    }
}

disclaimer: I don't know C#.

The same in Python:

#!/usr/bin/env python

def flatten(iterable):
    for item in iterable:
        if hasattr(item, '__iter__'):
            for nested in flatten(item):
                yield nested
        else:
            yield item

if __name__ == '__main__':
    for item in flatten([1,[2, 3, [[4], 5]], 6, [[[7]]], [8]]):
        print(item, end=" ")

It prints:

1 2 3 4 5 6 7 8 

Answer 9

The SelectMany extension method does this already.

Projects each element of a sequence to an IEnumerable<(Of <(T>)>) and flattens the resulting sequences into one sequence.

Answer 10

Isn't that what [SelectMany][1] is for?

enum1.SelectMany(
    a => a.SelectMany(
        b => b.SelectMany(
            c => c.Select(
                d => d.Name
            )
        )
    )
);

Answer 11

static class EnumerableExtensions
{
    public static IEnumerable<T> Flatten<T>(this IEnumerable<IEnumerable<T>> sequence)
    {
        foreach(var child in sequence)
            foreach(var item in child)
                yield return item;
    }
}

Maybe like this? Or do you mean that it could potentially be infintly deep?

Answer 12

Hmm... I'm not sure exactly what you want here, but here's a "one level" option:

public static IEnumerable<TElement> Flatten<TElement,TSequence> (this IEnumerable<TSequence> sequences)
    where TSequence : IEnumerable<TElement> 
{
    foreach (TSequence sequence in sequences)
    {
        foreach(TElement element in sequence)
        {
            yield return element;
        }
    }
}

If that's not what you want, could you provide the signature of what you do want? If you don't need a generic form, and you just want to do the kind of thing that LINQ to XML constructors do, that's reasonably simple - although the recursive use of iterator blocks is relatively inefficient. Something like:

static IEnumerable Flatten(params object[] objects)
{
    // Can't easily get varargs behaviour with IEnumerable
    return Flatten((IEnumerable) objects);
}

static IEnumerable Flatten(IEnumerable enumerable)
{
    foreach (object element in enumerable)
    {
        IEnumerable candidate = element as IEnumerable;
        if (candidate != null)
        {
            foreach (object nested in candidate)
            {
                yield return nested;
            }
        }
        else
        {
            yield return element;
        }
    }
}

Note that that will treat a string as a sequence of chars, however - you may want to special-case strings to be individual elements instead of flattening them, depending on your use case.

Does that help?

Answer 13

Basicly, you need to have a master IENumerable that is outside of your recursive function, then in your recursive function (Psuedo-code)

private void flattenList(IEnumerable<T> list)
{
    foreach (T item in list)
    {
        masterList.Add(item);

        if (item.Count > 0)
        {
            this.flattenList(item);
        }
    }
}

Though I'm really not sure what you mean by IEnumerable nested in an IEnumerable...whats within that? How many levels of nesting? Whats the final type? obviously my code isn't correct, but I hope it gets you thinking.