Pattern to get objects structured difference and statistics or how to create diff tool

Question

Let's say i have such structure

public class Form
{
    #region Public Properties

    public List<Field> Fields { get; set; }

    public string Id { get; set; }

    public string Name { get; set; }

    public string Version { get; set; }

    public int Revision { get; set; }

    #endregion
}

So the Form class contains the list of fields, let's say that the Field itself is represented with such structure

public class Field
{
    #region Public Properties

    public string DisplayName { get; set; }

    public List<Field> Fields { get; set; }

    public string Id { get; set; }

    public FieldKind Type { get; set; }

    public FieldType FieldType { get; set; }

    #endregion
}

The Field is basically composite structure, each Field contains the list of child fields. So the structure is hierarchical. Also Field has reference to the FieldType class.

public class FieldType
{
    #region Public Properties

    public DataType DataType { get; set; }

    public string DisplayName { get; set; }

    public string Id { get; set; }

    #endregion
}

And at the end we have reference to the DataType

public class DataType
{
    #region Public Properties

    public string BaseType { get; set; }

    public string Id { get; set; }

    public string Name { get; set; }

    public List<Restriction> Restrictions { get; set; }

    #endregion
}

What I want to achieve is to get the difference of such complex structure, let's say if I have some sort of Comparer it will give me structured difference for the whole form as one class, let's say DifferenceResult. When I said structured difference I mean that it should be something like that.

1. Difference for the Form fields (output Form has difference in Version field (different color)
2. Differences for the Fields collection, including the hierarchy of the fields to the edited field
3. Same behavior for the FieldType and DataType
4. Detect removing and adding the Field into the Form (so probably each Difference will have a Type)

What I have right now. I started with generic approach and tried to use ReflectionComparer implementing IEqualityComparer interface

public bool Equals(T x, T y)
{
    var type = typeof(T);

    if (typeof(IEquatable<T>).IsAssignableFrom(type))
    {
        return EqualityComparer<T>.Default.Equals(x, y);
    }

    var enumerableType = type.GetInterface(typeof(IEnumerable<>).FullName);

    if (enumerableType != null)
    {
        var elementType = enumerableType.GetGenericArguments()[0];
        var elementComparerType = typeof(DifferenceComparer<>).MakeGenericType(elementType);

        var elementComparer = Activator.CreateInstance(elementComparerType, new object[] { _foundDifferenceCallback, _existedDifference });

        return (bool)typeof(Enumerable).GetGenericMethod("SequenceEqual", new[] { typeof(IEnumerable<>), typeof(IEnumerable<>), typeof(IEqualityComparer<>) }).MakeGenericMethod(elementType).Invoke(null, new[] { x, y, elementComparer });
    }

    foreach (var propertyInfo in type.GetProperties())
    {
        var leftValue = propertyInfo.GetValue(x);
        var rightValue = propertyInfo.GetValue(y);

        if (leftValue != null)
        {
            var propertyComparerType = typeof(DifferenceComparer<>).MakeGenericType(propertyInfo.PropertyType);

            var propertyComparer = Activator.CreateInstance(propertyComparerType, new object[] {_foundDifferenceCallback, _existedDifference});

            if (!((bool)typeof(IEqualityComparer<>).MakeGenericType(propertyInfo.PropertyType)
                .GetMethod("Equals")
                .Invoke(propertyComparer, new object[] { leftValue, rightValue })))
            {
                // Create and publish the difference with its Type
            }
        }
        else
        {
            if (!Equals(leftValue, rightValue))
            {
                // Create and publish the difference with its Type
            }
        }
    }

    //return true if no differences are found
}

And Difference class

public struct Difference
{
    public readonly string Property;

    public readonly DifferenceType Type;

    public readonly IExtractionable Expected;

    public readonly IExtractionable Actual;
}

But probably it's not the way I want to go, because I should compare Field more precisely taking in consideration that each Field has Id, which can be different for different forms as well and I want to take more control of the comparing process. For me it sounds more like a diff tool.

So I am looking for good pattern and pretty good structure to publish the Difference to the client code, which can easily visualize it?

Answer 1

Given your classes are very well known already, I'd use the following things :

• A IDiffable interface, that makes a class have a DiffWith(T other) class, returning an enumeration of differences. It make then sense to just implement your DiffWith in each of your classes, which should be trivial given they are pretty much fixed.

• A separate ReflectionDiff (possibly static) class that would compare, the way you do, using reflection. This one could use attributes to whitelist/ignore any property/field. This way, you can have a default diff engine for quick implementation of differences, and use exactly the kind of reflection code you are using. You could simply have an implementation of diff with like :

  public override IEnumerable<Difference> DiffWith(Field other)
  {
    return ReflectionDiff.Diff(this, other);
  }
  

You get one single class with all the dirty reflection code (which isn't really needed), and can have a simple implementation of DiffWith in each of your diffable classes.

Answer 2

After investigating a bit and comparing different resources I implemented such algorithm.

1. Take your object and construct the graph (using .GetType().GetProperties(), adding each property as a child node of you root node. The key of the node will be the property name.
2. Go recursively if you meet properties of any custom type or IEnumerable.
3. Create graph comparer and simple compare two graphs

Later I will publish my code and update the answer.