The relationship could not be changed because one or more of the foreign-key properties is non-nullable

Question

You should delete old child items thisParent.ChildItems one by one manually. Entity Framework doesn't do that for you. It finally cannot decide what you want to do with the old child items - if you want to throw them away or if you want to keep and assign them to other parent entities. You must tell Entity Framework your decision. But one of these two decisions you HAVE to make since the child entities cannot live alone without a reference to any parent in the database (due to the foreign key constraint). That's basically what the exception says.

Edit

What I would do if child items could be added, updated and deleted:

public void UpdateEntity(ParentItem parent)
{
    // Load original parent including the child item collection
    var originalParent = _dbContext.ParentItems
        .Where(p => p.ID == parent.ID)
        .Include(p => p.ChildItems)
        .SingleOrDefault();
    // We assume that the parent is still in the DB and don't check for null

    // Update scalar properties of parent,
    // can be omitted if we don't expect changes of the scalar properties
    var parentEntry = _dbContext.Entry(originalParent);
    parentEntry.CurrentValues.SetValues(parent);

    foreach (var childItem in parent.ChildItems)
    {
        var originalChildItem = originalParent.ChildItems
            .Where(c => c.ID == childItem.ID && c.ID != 0)
            .SingleOrDefault();
        // Is original child item with same ID in DB?
        if (originalChildItem != null)
        {
            // Yes -> Update scalar properties of child item
            var childEntry = _dbContext.Entry(originalChildItem);
            childEntry.CurrentValues.SetValues(childItem);
        }
        else
        {
            // No -> It's a new child item -> Insert
            childItem.ID = 0;
            originalParent.ChildItems.Add(childItem);
        }
    }

    // Don't consider the child items we have just added above.
    // (We need to make a copy of the list by using .ToList() because
    // _dbContext.ChildItems.Remove in this loop does not only delete
    // from the context but also from the child collection. Without making
    // the copy we would modify the collection we are just interating
    // through - which is forbidden and would lead to an exception.)
    foreach (var originalChildItem in
                 originalParent.ChildItems.Where(c => c.ID != 0).ToList())
    {
        // Are there child items in the DB which are NOT in the
        // new child item collection anymore?
        if (!parent.ChildItems.Any(c => c.ID == originalChildItem.ID))
            // Yes -> It's a deleted child item -> Delete
            _dbContext.ChildItems.Remove(originalChildItem);
    }

    _dbContext.SaveChanges();
}

Note: This is not tested. It's assuming that the child item collection is of type ICollection. (I usually have IList and then the code looks a bit different.) I've also stripped away all repository abstractions to keep it simple.

I don't know if that is a good solution, but I believe that some kind of hard work along these lines must be done to take care of all kinds of changes in the navigation collection. I would also be happy to see an easier way of doing it.

The reason you're facing this is due to the difference between composition and aggregation.

In composition, the child object is created when the parent is created and is destroyed when its parent is destroyed. So its lifetime is controlled by its parent. e.g. A blog post and its comments. If a post is deleted, its comments should be deleted. It doesn't make sense to have comments for a post that doesn't exist. Same for orders and order items.

In aggregation, the child object can exist irrespective of its parent. If the parent is destroyed, the child object can still exist, as it may be added to a different parent later. e.g.: the relationship between a playlist and the songs in that playlist. If the playlist is deleted, the songs shouldn't be deleted. They may be added to a different playlist.

The way Entity Framework differentiates aggregation and composition relationships is as follows:

• For composition: it expects the child object to a have a composite primary key (ParentID, ChildID). This is by design as the IDs of the children should be within the scope of their parents.

• For aggregation: it expects the foreign key property in the child object to be nullable.

So, the reason you're having this issue is because of how you've set your primary key in your child table. It should be composite, but it's not. So, Entity Framework sees this association as aggregation, which means, when you remove or clear the child objects, it's not going to delete the child records. It'll simply remove the association and sets the corresponding foreign key column to NULL (so those child records can later be associated with a different parent). Since your column does not allow NULL, you get the exception you mentioned.

Solutions:

1- If you have a strong reason for not wanting to use a composite key, you need to delete the child objects explicitly. And this can be done simpler than the solutions suggested earlier:

context.Children.RemoveRange(parent.Children);

2- Otherwise, by setting the proper primary key on your child table, your code will look more meaningful:

parent.Children.Clear();

This is a very big problem. What actually happens in your code is this:

• You load Parent from the database and get an attached entity
• You replace its child collection with new collection of detached children
• You save changes but during this operation all children are considered as added becasue EF didn't know about them till this time. So EF tries to set null to foreign key of old children and insert all new children => duplicate rows.

Now the solution really depends on what you want to do and how would you like to do it?

If you are using ASP.NET MVC you can try to use UpdateModel or TryUpdateModel.

If you want just update existing children manually, you can simply do something like:

foreach (var child in modifiedParent.ChildItems)
{
    context.Childs.Attach(child); 
    context.Entry(child).State = EntityState.Modified;
}

context.SaveChanges();

Attaching is actually not needed (setting the state to Modified will also attach the entity) but I like it because it makes the process more obvious.

If you want to modify existing, delete existing and insert new childs you must do something like:

var parent = context.Parents.GetById(1); // Make sure that childs are loaded as well
foreach(var child in modifiedParent.ChildItems)
{
    var attachedChild = FindChild(parent, child.Id);
    if (attachedChild != null)
    {
        // Existing child - apply new values
        context.Entry(attachedChild).CurrentValues.SetValues(child);
    }
    else
    {
        // New child
        // Don't insert original object. It will attach whole detached graph
        parent.ChildItems.Add(child.Clone());
    }
}

// Now you must delete all entities present in parent.ChildItems but missing
// in modifiedParent.ChildItems
// ToList should make copy of the collection because we can't modify collection
// iterated by foreach
foreach(var child in parent.ChildItems.ToList())
{
    var detachedChild = FindChild(modifiedParent, child.Id);
    if (detachedChild == null)
    {
        parent.ChildItems.Remove(child);
        context.Childs.Remove(child); 
    }
}

context.SaveChanges();

I found this answer much more helpful for the same error. It seems that EF does not like it when you Remove, it prefers Delete.

You can delete a collection of records attached to a record like this.

order.OrderDetails.ToList().ForEach(s => db.Entry(s).State = EntityState.Deleted);

In the example, all of the Detail records attached to an Order have their State set to Delete. (In preparation to Add back updated Details, as part of an Order update)

I've no idea why the other two answers are so popular!

I believe you were right in assuming the ORM framework should handle it - after all, that is what it promises to deliver. Otherwise your domain model gets corrupted by persistence concerns. NHibernate manages this happily if you setup the cascade settings correctly. In Entity Framework it is also possible, they just expect you to follow better standards when setting up your database model, especially when they have to infer what cascading should be done:

You have to define the parent - child relationship correctly by using an "identifying relationship".

If you do this, Entity Framework knows the child object is identified by the parent, and therefore it must be a "cascade-delete-orphans" situation.

Other than the above, you might need to (from NHibernate experience)

thisParent.ChildItems.Clear();
thisParent.ChildItems.AddRange(modifiedParent.ChildItems);

instead of replacing the list entirely.

UPDATE

@Slauma's comment reminded me that detached entities are another part of the overall problem. To solve that, you can take the approach of using a custom model binder that constructs your models by attempting to load it from the context. This blog post shows an example of what I mean.

If you are using AutoMapper with Entity Framework on the same class, you might hit this problem. For instance if your class is

class A
{
    public ClassB ClassB { get; set; }
    public int ClassBId { get; set; }
}

AutoMapper.Map<A, A>(input, destination);

This will try to copy both properties. In this case, ClassBId is non Nullable. Since AutoMapper will copy destination.ClassB = input.ClassB; this will cause a problem.

Set your AutoMapper to Ignore ClassB property.

 cfg.CreateMap<A, A>()
     .ForMember(m => m.ClassB, opt => opt.Ignore()); // We use the ClassBId