Entity Framework 4.1+ many-to-many relationships change tracking

Question

How can I detect changes of ICollection<> properties (many-to-many relationships)?

public class Company
{
    ...

    public virtual ICollection<Employee> Employees { get; set; }
}

using (DataContext context = new DataContext(Properties.Settings.Default.ConnectionString))
{
    Company company = context.Companies.First();
    company.Employees.Add(context.Employees.First());

    context.SaveChanges();
}

public class DataContext : DbContext
{
    public override int SaveChanges()
    {
        return base.SaveChanges();

        // Company's entity state is "Unchanged" in this.ChangeTracker
    }
}

Answer 1

Here is how to find all the changed many-to-many relationships. I've implemented the code as extension methods:

public static class IaExtensions
{
    public static IEnumerable<Tuple<object, object>> GetAddedRelationships(
        this DbContext context)
    {
        return GetRelationships(context, EntityState.Added, (e, i) => e.CurrentValues[i]);
    }

    public static IEnumerable<Tuple<object, object>> GetDeletedRelationships(
        this DbContext context)
    {
        return GetRelationships(context, EntityState.Deleted, (e, i) => e.OriginalValues[i]);
    }

    private static IEnumerable<Tuple<object, object>> GetRelationships(
        this DbContext context,
        EntityState relationshipState,
        Func<ObjectStateEntry, int, object> getValue)
    {
        context.ChangeTracker.DetectChanges();
        var objectContext = ((IObjectContextAdapter)context).ObjectContext;

        return objectContext
            .ObjectStateManager
            .GetObjectStateEntries(relationshipState)
            .Where(e => e.IsRelationship)
            .Select(
                e => Tuple.Create(
                    objectContext.GetObjectByKey((EntityKey)getValue(e, 0)),
                    objectContext.GetObjectByKey((EntityKey)getValue(e, 1))));
    }
}

Some explanation. Many-to-many relationships are represented in EF as Independent Associations, or IAs. This is because the foreign keys for the relationship are not exposed anywhere in the object model. In the database the FKs are in a join table, and this join table is hidden from the object model.

IAs are tracked in EF using "relationship entries". These are similar to the DbEntityEntry objects you get from the DbContext.Entry except that they represent a relationship between two entities rather than an entity itself. Relationship entries are not exposed in the DbContext API, so you need to drop down to ObjectContext to access them.

A new relationship entry is created when a new relationship between two entities is created, for example by adding an Employee to the Company.Employees collection. This relationship is in the Added state.

Likewise, when a relationship between two entities is removed, then the relationship entry is put into the Deleted state.

This means that to find changed many-to-many relationships (or actually any changed IA) we need to find added and deleted relationship entries. This is what the GetAddedRelationships and GetDeletedRelationships do.

Once we have relationship entries, we need to make sense of them. For this you need to know a piece of insider knowledge. The CurrentValues property of an Added (or Unchanged) relationship entry contains two values which are the EntityKey objects of the entities at either end of the relationship. Likewise, but annoyingly slightly different, the OriginalValues property of a Deleted relationship entry contains the EntityKey objects for the entities at either end of the deleted relationship.

(And, yes, this is horrible. Please don’t blame me—it is from well before my time.)

The CurrentValues/OriginalValues difference is why we pass a delegate into the GetRelationships private method.

Once we have the EntityKey objects we can use GetObjectByKey to get the actual entity instances. We return these as tuples and there you have it.

Here’s some entities, a context, and an initializer, I used to test this. (Note—testing was not extensive.)

public class Company
{
    public int Id { get; set; }
    public string Name { get; set; }
    public virtual ICollection<Employee> Employees { get; set; }

    public override string ToString()
    {
        return "Company " + Name;
    }
}

public class Employee
{
    public int Id { get; set; }
    public string Name { get; set; }
    public virtual ICollection<Company> Companies { get; set; }

    public override string ToString()
    {
        return "Employee " + Name;
    }
}

public class DataContext : DbContext
{
    static DataContext()
    {
        Database.SetInitializer(new DataContextInitializer());
    }

    public DbSet<Company> Companies { get; set; }
    public DbSet<Employee> Employees { get; set; }

    public override int SaveChanges()
    {
        foreach (var relationship in this.GetAddedRelationships())
        {
            Console.WriteLine(
                "Relationship added between {0} and {1}",
                relationship.Item1,
                relationship.Item2);
        }

        foreach (var relationship in this.GetDeletedRelationships())
        {
            Console.WriteLine(
                "Relationship removed between {0} and {1}",
                relationship.Item1,
                relationship.Item2);
        }

        return base.SaveChanges();
    }

}

public class DataContextInitializer : DropCreateDatabaseAlways<DataContext>
{
    protected override void Seed(DataContext context)
    {
        var newMonics = new Company { Name = "NewMonics", Employees = new List<Employee>() };
        var microsoft = new Company { Name = "Microsoft", Employees = new List<Employee>() };

        var jim = new Employee { Name = "Jim" };
        var arthur = new Employee { Name = "Arthur" };
        var rowan = new Employee { Name = "Rowan" };

        newMonics.Employees.Add(jim);
        newMonics.Employees.Add(arthur);
        microsoft.Employees.Add(arthur);
        microsoft.Employees.Add(rowan);

        context.Companies.Add(newMonics);
        context.Companies.Add(microsoft);
    }
}

Here’s an example of using it:

using (var context = new DataContext())
{
    var microsoft = context.Companies.Single(c => c.Name == "Microsoft");
    microsoft.Employees.Add(context.Employees.Single(e => e.Name == "Jim"));

    var newMonics = context.Companies.Single(c => c.Name == "NewMonics");
    newMonics.Employees.Remove(context.Employees.Single(e => e.Name == "Arthur"));

    context.SaveChanges();
}