Entity Framework Detached Object Merging

Question

I have a scenario where I am using Entity Framework in a WCF service, and changes happen on a non-tracked instance of a type that is mapped back to the database via code-first (non-trivial updates and deletes throughout the instance's object tree). When I try to attach the non-tracked instance into the context, EF is only recognizing changes to the simple value types on the root object.

Does anyone know of an elegant solution for this scenario? I am looking for a way to do this by using a generic repository, and avoiding having to run through the instance's entire object tree managing the "attach/detach" state of every object. I have considered possibly using ValueInjecter or AutoMapper to run the changes on a fully hydrated and tracked instance of the "old" state in order for the context to pickup the changes. Also, how would Nhibernate handle this situation?

Thanks in advance for your input!

UPDATE (7/31/2012): I have updated the code to handle genericly-typed keys, and some typing issues with EF Proxies. Also added some helper extensions when dealing with IEntity types. This implementation isn't perfect, but it is very functional.

UPDATE (3/13/2012): I have added a feature request for cleaner merging in EF. The request is located here: http://data.uservoice.com/forums/72025-ado-net-entity-framework-ef-feature-suggestions/suggestions/2679160-better-merging-change-tracking

UPDATE (3/12/2012): I have posted my solution below. It uses FubuCore, ValueInjecter, and requires entities to be marked with one of two interfaces, either IEntity, or IRecursiveEntity for recursive classes. The solution will handle recursive, self-linked entities.

Also, I am referencing a generic repository (Repository) that allows me to get a reference to the IDbSet that EF exposes. This could be substituded with any other generic or specific repository. Lastly, the IEntity interface uses an int? id, however you could define that however you want (Guid/Guid?). The solution itself isn't quite as elegant as I would like, however it allows for much more elegant data access code when behind a physical WCF service boundary.

public class DomainMergeInjection : ConventionInjection
{
    private readonly Repository _repository;
    private readonly Dictionary<string, object> _potentialParentObjectDump;
    private readonly Cache<Type, Type> _entityTypesAndKeysCache;

    public DomainMergeInjection(Repository repository)
    {
        _repository = repository;
        _potentialParentObjectDump = new Dictionary<string, object>();
        _entityTypesAndKeysCache = new Cache<Type, Type>();
    }

    protected override bool Match(ConventionInfo c)
    {
        return c.SourceProp.Name == c.TargetProp.Name;
    }

    protected override object SetValue(ConventionInfo c)
    {
        if(c.SourceProp.Value == null)
            return null;

        //for value types and string just return the value as is 
        if(c.SourceProp.Type.IsSimple())
            return c.SourceProp.Value;

        //TODO: Expand on this to handle IList/IEnumerable (i.e. the non-generic collections and arrays).
        //handle arrays
        if(c.SourceProp.Type.IsArray)
        {
            var sourceArray = c.SourceProp.Value as Array;
            // ReSharper disable PossibleNullReferenceException
            var clonedArray = sourceArray.Clone() as Array;
            // ReSharper restore PossibleNullReferenceException

            for(int index = 0; index < sourceArray.Length; index++)
            {
                var sourceValueAtIndex = sourceArray.GetValue(index);

                //Skip null and simple values that would have already been moved in the clone.
                if(sourceValueAtIndex == null || sourceValueAtIndex.GetType().IsSimple())
                    continue;

                // ReSharper disable PossibleNullReferenceException
                clonedArray.SetValue(RetrieveComplexSourceValue(sourceValueAtIndex), index);
                // ReSharper restore PossibleNullReferenceException
            }

            return clonedArray;
        }

        //handle IEnumerable<> also ICollection<> IList<> List<>
        if(c.SourceProp.Type.IsGenericEnumerable())
        {
            var t = c.SourceProp.Type.GetGenericArguments()[0];

            if(t.IsSimple())
                return c.SourceProp.Value;

            var tlist = typeof(List<>).MakeGenericType(t);
            dynamic list = Activator.CreateInstance(tlist);

            var addMethod = tlist.GetMethod("Add");

            foreach(var sourceItem in (IEnumerable)c.SourceProp.Value)
            {
                addMethod.Invoke(list, new[] { RetrieveComplexSourceValue(sourceItem) });
            }

            return list;
        }

        //Get a source value that is in the right state and is tracked if needed.
        var itemStateToInject = RetrieveComplexSourceValue(c.SourceProp.Value);
        return itemStateToInject;
    }

    private object RetrieveComplexSourceValue(object source)
    {
        //If the source is a non-tracked type, or the source is a new value, then return its value.
        if(!source.ImplementsIEntity(_entityTypesAndKeysCache) || source.IsEntityIdNull(_entityTypesAndKeysCache))
            return source;

        object sourceItemFromContext;

        //Handle recursive entities, this could probably be cleaned up.
        if(source.ImplementsIRecursiveEntity())
        {
            var itemKey = source.GetEntityIdString(_entityTypesAndKeysCache) + " " + ObjectContext.GetObjectType(source.GetType());

            //If we have a context item for this key already, just return it.  This solves a recursion problem with self-linking items.
            if(_potentialParentObjectDump.ContainsKey(itemKey))
                return _potentialParentObjectDump[itemKey];

            //Get the source from the context to ensure it is tracked.
            sourceItemFromContext = GetSourceItemFromContext(source);

            //Add the class into the object dump in order to avoid any infinite recursion issues with self-linked objects
            _potentialParentObjectDump.Add(itemKey, sourceItemFromContext);
        }
        else
            //Get the source from the context to ensure it is tracked.
            sourceItemFromContext = GetSourceItemFromContext(source);

        //Recursively use this injection class instance to inject the source state on to the context source state.
        var itemStateToInject = sourceItemFromContext.InjectFrom(this, source);

        return itemStateToInject;
    }

    private object GetSourceItemFromContext(object source)
    {
        if(source == null)
            return null;

        //Using dynamic here to "AutoCast" to an IEntity<>.  We should have one, but it's important to note just in case.
        dynamic sourceEntityValue = source;
        var sourceEntityType = ObjectContext.GetObjectType(source.GetType());
        var sourceKeyType = sourceEntityType.GetEntityKeyType();

        var method = typeof(DomainMergeInjection).GetMethod("GetFromContext", BindingFlags.Instance | BindingFlags.NonPublic);
        var generic = method.MakeGenericMethod(sourceEntityType, sourceKeyType);

        var sourceItemFromContext = generic.Invoke(this, new object[] { new object[] { sourceEntityValue.Id } });
        return sourceItemFromContext;
    }

    // ReSharper disable UnusedMember.Local
    private TItem GetFromContext<TItem, TKey>(object[] keys) where TItem : class, IEntity<TKey>
    // ReSharper restore UnusedMember.Local
    {
        var foundItem = _repository.GetDbSet<TItem>().Find(keys);

        return foundItem;
    }
}

public static class EntityTypeExtensions
{
    /// <summary>
    /// Determines if an object instance implements IEntity.
    /// </summary>
    /// <param name="entity"></param>
    /// <param name="entityCache">A cache to hold types that do implement IEntity.  If the cache does not have the Type and the Type does implement IEntity, it will add the type to the cache along with the </param>
    /// <returns></returns>
    public static bool ImplementsIEntity(this object entity, Cache<Type, Type> entityCache = null)
    {
        //We need to handle getting the proxy type if this is an EF Code-First proxy.
        //Please see for more info: http://msdn.microsoft.com/en-us/library/dd456853.aspx
        var entityType = ObjectContext.GetObjectType(entity.GetType());

        if(entityCache != null && entityCache.Has(entityType))
            return true;

        var implementationOfIEntity = entityType.GetInterfaces().FirstOrDefault(x => x.IsGenericType && x.GetGenericTypeDefinition() == typeof (IEntity<>));

        if(implementationOfIEntity == null)
            return false;

        if(entityCache != null)
        {
            var keyType = implementationOfIEntity.GetGenericArguments()[0];
            entityCache.Fill(entityType, keyType);
        }

        return true;
    }

    /// <summary>
    /// Determines if an object instances implements IRecurisveEntity
    /// </summary>
    /// <param name="entity"></param>
    /// <returns></returns>
    public static bool ImplementsIRecursiveEntity(this object entity)
    {
        //We need to handle getting the proxy type if this is an EF Code-First proxy.
        //Please see for more info: http://msdn.microsoft.com/en-us/library/dd456853.aspx
        var entityType = ObjectContext.GetObjectType(entity.GetType());

        var implementsIRecursiveEntity = entityType.GetInterfaces().Any(x => x.IsGenericType && x.GetGenericTypeDefinition() == typeof(IRecursiveEntity<>));

        return implementsIRecursiveEntity;
    }

    /// <summary>
    /// Determines whether or not an Entity's Id is null.  Will throw an exception if a type that does not implement IEntity is passed through.
    /// </summary>
    /// <param name="entity"></param>
    /// <param name="entityCache"></param>
    /// <returns></returns>
    public static bool IsEntityIdNull(this object entity, Cache<Type, Type> entityCache = null)
    {
        bool isEntityIdNull = ExecuteEntityIdMethod<bool>("IsEntityIdNull", entity, entityCache);

        return isEntityIdNull;
    }

    /// <summary>
    /// Determines whether or not an Entity's Id is null.  Will throw an exception if a type that does not implement IEntity is passed through.
    /// </summary>
    /// <param name="entity"></param>
    /// <param name="entityCache"></param>
    /// <returns></returns>
    public static string GetEntityIdString(this object entity, Cache<Type, Type> entityCache = null)
    {
        string entityIdString = ExecuteEntityIdMethod<string>("GetEntityIdString", entity, entityCache);

        return entityIdString;
    }

    private static T ExecuteEntityIdMethod<T>(string methodName, object entityInstance, Cache<Type, Type> entityCache = null)
    {
        if(!entityInstance.ImplementsIEntity(entityCache))
            throw new ArgumentException(string.Format("Parameter entity of type {0} does not implement IEntity<>, and so ist not executable for {1}!", entityInstance.GetType(), methodName));

        //We need to handle getting the proxy type if this is an EF Code-First proxy.
        //Please see for more info: http://msdn.microsoft.com/en-us/library/dd456853.aspx
        var entityType = ObjectContext.GetObjectType(entityInstance.GetType());
        var keyType = entityCache != null ? entityCache[entityType] : entityType.GetEntityKeyType();

        var method = typeof(EntityTypeExtensions).GetMethod(methodName, BindingFlags.Static | BindingFlags.NonPublic);
        var generic = method.MakeGenericMethod(keyType);

        T returnValue = (T)generic.Invoke(null, new[] { entityInstance });

        return returnValue;
    }

    private static string GetEntityIdString<TKey>(IEntity<TKey> entity)
    {
        var entityIdString = entity.Id.ToString();

        return entityIdString;
    }

    private static bool IsEntityIdNull<TKey>(IEntity<TKey> entity)
    {
        //We need to handle getting the proxy type if this is an EF Code-First proxy.
        //Please see for more info: http://msdn.microsoft.com/en-us/library/dd456853.aspx
        var entityType = ObjectContext.GetObjectType(entity.GetType());

        if(entityType.IsPrimitive)
            return false;

        //NOTE:  We know that this entity's type is NOT primitive, therefore we can cleanly test for null, and return properly.
        // ReSharper disable CompareNonConstrainedGenericWithNull
        var entityIdIsNull = entity.Id == null;
        // ReSharper restore CompareNonConstrainedGenericWithNull

        return entityIdIsNull;
    }

    public static Type GetEntityKeyType(this Type typeImplementingIEntity)
    {
        var implementationOfIEntity = typeImplementingIEntity.GetInterfaces().FirstOrDefault(x => x.IsGenericType && x.GetGenericTypeDefinition() == typeof(IEntity<>));

        if(implementationOfIEntity == null)
            throw new ArgumentException(string.Format("Type {0} does not implement IEntity<>", typeImplementingIEntity));

        var keyType = implementationOfIEntity.GetGenericArguments()[0];
        return keyType;
    }
}

public interface IEntity<TKey>
{
    TKey Id { get; set; }
}


public interface IRecursiveEntity<TKey> : IEntity<TKey>
{
    IRecursiveEntity<TKey> Parent { get; }
    IEnumerable<IRecursiveEntity<TKey>> Children { get; }
}

Answer 1

you could use the detached object only as a DTO,

and after refill the object from context with values from the DTO

with ValueInjecter this would be:

//manually
conObj.InjectFrom(dto);
conObj.RefTypeProp.InjectFrom(dto.RefTypeProp);
...

//or by writing a custom injection:
conObj.InjectFrom<ApplyChangesInjection>(dto);

here's the Injection that will do that automatically, (I did it by modifying a bit the DeepClone Injection from VI's home page)

the trick here is that the Injection uses itself in the SetValue method

public class ApplyChangesInjection : ConventionInjection
{
    protected override bool Match(ConventionInfo c)
    {
        return c.SourceProp.Name == c.TargetProp.Name;
    }

    protected override object SetValue(ConventionInfo c)
    {
        if (c.SourceProp.Value == null) return null;

        //for value types and string just return the value as is
        if (c.SourceProp.Type.IsValueType || c.SourceProp.Type == typeof(string))
            return c.SourceProp.Value;

        //handle arrays - not impl
        //handle IEnumerable<> also ICollection<> IList<> List<> - not impl

        //for simple object types apply the inject using the corresponding source

        return c.TargetProp.Value
            .InjectFrom<ApplyChangesInjection>(c.SourceProp.Value);
    }
}

//Note: I'm not handling collections in this injection, I just wanted you to understand the principle, you can look at the original http://valueinjecter.codeplex.com/wikipage?title=Deep%20Cloning&referringTitle=Home