Saturday, January 30, 2010

Entity Framework v4 Object-Graph Deleting Quirks

Update: Danny was quick to respond with a great explanation about how EF4 deals with object-graphs and deletes.

I’ve been getting up to speed on Entity Framework V4 (EF4) lately, for use in a WPF application that I’m working on.   One of the issues I’m coming up against is removal of objects in one-to-many and many-to-many relationships.  It’s still not clear to me why things work the way they do in EF4, but I’ll lay out a few scenarios and talk about what works and what doesn’t work and hope that someone can later comment as to why things may not be working as I would expect.

Scenario 1: A many-to-many relationship where the join table has its own primary key (an identity column in this case):

image

Given the following code, I’d expect to be able to remove an object from the Employee.ProjectAssignments collection and see that object removed from the database when I call SaveChanges on my ObjectContext:

using (var context = new EFTestDBEntities()){
                var employeeGraph = (from e in context.Employees.Include("ProjectAssignments").Include("ProjectAssignments.Project") where e.ID == employeeID select e).FirstOrDefault();
                if (employeeGraph != null){
                    var assignment = employeeGraph.ProjectAssignments.First();
                    employeeGraph.ProjectAssignments.Remove(assignment);

                    context.SaveChanges();
                }
            }

When I hit the “SaveChanges()” call, I get the following exception:

System.InvalidOperationException: The operation failed: The relationship could not be changed because one or more of the foreign-key properties is non-nullable. When a change is made to a relationship, the related foreign-key property is set to a null value. If the foreign-key does not support null values, a new relationship must be defined, the foreign-key property must be assigned another non-null value, or the unrelated object must be deleted.

This strikes me as strange – I assume that I’ve indicated that the employee is no longer related to this project-assignment, so the object is essentially “worthless” and I’d expect it to be removed.

In order to successfully remove this project-assignment, I had to do the following:

using (var context = new EFTestDBEntities()){
                var employeeGraph = (from e in context.Employees.Include("ProjectAssignments").Include("ProjectAssignments.Project") where e.ID == employeeID select e).FirstOrDefault();
                if (employeeGraph != null){
                    ProjectAssignment assignment = employeeGraph.ProjectAssignments.First();
                    context.ProjectAssignments.DeleteObject(assignment);

                    context.SaveChanges();
                    Assert.AreEqual(0, employeeGraph.ProjectAssignments.Count);
                }
            }

Okay, so if I “manually” delete the object by the explicit call to DeleteObject, things work.  So far so good, I can work with this.

Scenario 2: A many-to-many relationship where the join table has a primary key that is composed by foreign-keys to the tables on either side of the relationship:

image 

In this case, the code I’d expect to work, actually worked and allowed me to delete a class-schedule from a teacher:

using (var context = new EFTestDBEntities()){
                var teacherGraph = (from t in context.Teachers.Include("ClassSchedules").Include("ClassSchedules.Class") where t.ID == teacherID select t).FirstOrDefault();
                if(teacherGraph != null){
                    var classSchedule = teacherGraph.ClassSchedules.First();
                    teacherGraph.ClassSchedules.Remove(classSchedule);

                    context.SaveChanges();
                }
            }

Why does EF4 make a distinction in behavior between these two scenarios that are essentially the same operation yet one object-graph allows me to write what I naturally want to write (scenario 2), but not the other?

Scenario 3: A one-to-many relationship:

image

Once again, I tried to write the code in a way I’d expect to naturally work:

using (var context = new EFTestDBEntities()){
                var orderGraph = (from o in context.Orders.Include("OrderDetails") where o.ID == orderID select o).FirstOrDefault();
                if(orderGraph != null){
                    var detail = orderGraph.OrderDetails.First();
                    orderGraph.OrderDetails.Remove(detail);

                    context.SaveChanges();
                }
            }

And once again, I got the following exception:

System.InvalidOperationException: The operation failed: The relationship could not be changed because one or more of the foreign-key properties is non-nullable. When a change is made to a relationship, the related foreign-key property is set to a null value. If the foreign-key does not support null values, a new relationship must be defined, the foreign-key property must be assigned another non-null value, or the unrelated object must be deleted.

As with scenario 1, I had to write the following code to successfully remove the OrderDetail record:

using (var context = new EFTestDBEntities()){
                var orderGraph = (from o in context.Orders.Include("OrderDetails") where o.ID == orderID select o).FirstOrDefault();
                if (orderGraph != null){
                    var detail = orderGraph.OrderDetails.First();
                    context.OrderDetails.DeleteObject(detail);

                    context.SaveChanges();
                }
            }

Alright, so I have managed to work around some of the quirks with EF4, but I really don’t understand why removal of objects in the object-graph don’t cause a proper remove/delete in some cases, yet work just as I expect in others. 

Saturday, August 29, 2009

LINQ to SQL SubmitChanges Order of Operations

I did not realize the following is true in LINQ to SQL: Calling SubmitChanges fires ChangeSet changes in the following order:

  • Inserts
  • Updates
  • Deletes

Normally, this might not be a big deal, and you might never notice it, but it was the source of some pain on my current project.

Consider an order-entry screen that allows the user to add/edit/delete line items on an order.  The order-detail records are keyed by order-number/line-number, and this is the source of the problem.  If the user deletes line #1 in the order, what was originally line #2 gets re-ordered as line #1.  When the user decides they are done editing the order, they will then hit the save button and the queued up add/edit/delete operations are committed via the LINQ to SQL SubmitChanges command.

Fortunately, we were able to change the primary-key on our order-detail table, but we effectively lost some automatic data-integrity since we were previously able to ensure uniqueness on order-number/line-number.  For teams that don’t have the luxury of changing their database keys, this design decision by the LINQ to SQL team may be cause to move away from LINQ to SQL. 

Why the LINQ to SQL team decided to not submit changes in the order of operations, or allow the user to specify a non-default order to submit changes is beyond me.  I’m sure they had valid reasons for this decision, but it’s a bit short-sighted in that this interaction pattern is fairly common in order-entry system. 

This could perhaps been avoided by stamping various LINQ class members as virtual and letting us developers (consumers) override behaviors as we need – and assume the risks that come with overriding the framework code.

Tuesday, July 14, 2009

Linq to SQL Object Equality Issue

I started having some issues where Linq to SQL objects are "different” instances between calls to db.GetTable(type) and comparing objects returned.  In the majority of my types, object equality was working just fine, but in a few types, the hash-codes for otherwise “equal” objects was different, while the remainder of the object data was as expected.

Deleted the designer-objects and re-added to my DBML and the issue is resolved. 

A compare of my DBML file shows that the “IsPrimaryKey” attribute on my Identity columns was missing.  I had probably added the table to the DBML and realized later that I hadn’t specified a primary key and later added said key, without re-synchronizing my DBML.

Saturday, May 16, 2009

Mocking IQueryable

On my current project, we’ve been pretty successful in mocking out all of our LINQ to SQL repository operations, with one exception. One of our new methods returns an IQueryable object. No problem I think, I’ll just set an expectation on that call and return a list of the property entity type. But that gives me a build-error that I can’t convert a List<T> to IQueryable<T>.

A quick google yields the simplest of solutions to mocking IQueryable: the “AsQueryable” extension method. So now, my mock expectation returns something like “List<T>.AsQueryable” and all is well. This finding also clued me into a whole slew of interesting looking extension methods in the System.Linq namespace…..

Monday, April 06, 2009

LINQ to SQL and Overriding .Equals = not a good idea.

Short version: don’t override .Equals in your LINQ TO SQL entities, unless you really know what “equals” means to your objects!

Long version: on my current project, we’re using LINQ to SQL for our ORM needs, and it’s been working pretty well for us so far.  Pretty well, except for late last week when we realized that we weren’t properly persisting a number of records in our object graph.

We have a fairly complex object model, with some deep graphs of related objects, and have been using the .Add method to add child and grandchild objects into our “Order” object’s graph.  There was one particular “Child” record that was only being created one time per order, when we expected several variations of this child, yet only the first was being saved.

Our object graphs were build correctly, SubmitChanges was doing its job, we weren’t getting any errors, and yet these records just weren’t making it to the database.  After a few conversations with other people that have some LINQ to SQL experience and me trying some of the ugliest hacks I could think of, it dawned on me to try testing another theory I had – create another order-detail that looked similar to an existing order-detail.

Lo and behold, this similar order-detail also did not persist, and it was behaving like the other “exceptional” order-details – they were all in the object-graph, but not in the Insert Changes of my data-context!  Once I was able to see this, I took a closer look at the order-detail partial class we had written, and there I saw an overridden .Equals implementation.  Unfortunately, this .Equals was testing for property values within the order-detail to determine if a detail might match another detail instance, not testing for reference equality, which is evidently important to LINQ to SQL.

So there you have it – when working with LINQ to SQL entities, take care in how you decide to utilize the .Equals method and how you decide to override it.