Stable, Hash Based Symmetrical Difference

Question

A lot of thought went into the implementation of the set algorithms in LINQ: Distinct, Except, Intersect and Union. They guarantee that the items are returned in the same order that they appear in before calling the method. So, these two expressions return the items in the same order:

var authors1 = books.Select(b => b.Author).OrderBy(a => a).Distinct();
var authors2 = books.Select(b => b.Author).Distinct().OrderBy(a => a);

The implementation of Distinct would look something like this:

public IEnumerable<T> Distinct<T>(this IEnumerable<T> source)
{
    HashSet<T> set = new HashSet<T>();
    foreach (T item in source)
    {
        if (set.Add(item))
        {
            yield return item;
        }
    }
}

The implementation of Except would look like this:

public static IEnumerable<T> Except<T>(this IEnumerable<T> source1, IEnumerable<T> source2)
{
    HashSet<T> set = new HashSet<T>(source2);
    foreach (T item in source1)
    {
        if (set.Add(item))
        {
            yield return item;
        }
    }
}

The implementation of Intersect would look something like this:

public static IEnumerable<T> Intersect<T>(this IEnumerable<T> source1, IEnumerable<T> source2)
{
    HashSet<T> set = new HashSet<T>(source2);
    HashSet<T> found = new HashSet<T>();
    foreach (T item in source1)
    {
        if (set.Contains(item) && found.Add(item))
        {
            yield return item;
        }
    }
}

I added a second set to remove duplicates from the list.

Finally, Union is a little more complex, but basically the same thing:

public static IEnumerable<T> Union<T>(this IEnumerable<T> source1, IEnumerable<T> source2)
{
    HashSet<T> set = new HashSet<T>();
    foreach (T item in source1)
    {
        if (set.Add(item))
        {
            yield return item;
        }
    }
    foreach (T item in source2)
    {
        if (set.Add(item))
        {
            yield return item;
        }
    }
}

So, the only operation not supported by LINQ is symmetrical difference or SymmetricExcept. I have been playing around with creating a "stable" version of this algorithm and out of pure curiosity am wondering if there is a better implementation. The most straight-forward implementation is to just call Except twice:

public static IEnumerable<T> SymmetricExcept<T>(this IEnumerable<T> source1, IEnumerable<T> source2)
{
    var except1 = source1.Except(source2);
    var except2 = source2.Except(source1);
    return except1.Concat(except2);
}

Although, this requires going through both lists twice. So I wrote a version that only requires going through the second list twice:

public static IEnumerable<T> SymmetricExcept<T>(this IEnumerable<T> source1, IEnumerable<T> source2)
{
    HashSet<T> set = new HashSet<T>(source2);
    HashSet<T> found = new HashSet<T>();
    foreach (T item in source1)
    {
        if (!set.Contains(item) && found.Add(item))
        {
            yield return item;
        }
    }
    foreach (T item in source2)
    {
        if (found.Add(item))
        {
            yield return item;
        }
    }
}

I haven't verified the correctness of these algorithms. There is some ambiguity about how LINQ handles duplicates. I am curious if there is a more efficient way to do SymmetricExcept, something better than O(m + 2n).

Stable, Hash Based Symmetrical Difference

0 Answers0