Multiple linked 1-1 relations don't sync PKs as expected - entity-framework

Question

I tried searching the SO, but all results I've found seem to deal with updating PK of entities that have already been persisted to the DB. My case is different.
I have 3 tables in a database with 1-0..1 relationships. The relationships look like this:
A <- B <- C
where '<-' represents a relationship and points to the principal end. I.e. each B always has a related A, but an A can have no B. In other words, A's cardinality is 1 and B's is 0..1.
Each relationship is represented by a FK that goes from the child entity's PK to the parent entity's PK. Each PK is a uniqueidentifier Id column with client-generated value. I've generated an EF 4 model from the database which has the same relationships with same cardinality.
I'm trying to add child entites B and C to an existing A entity. For design reasons the couple of new instances is created in one peace of code, and the A entity is linked to the B entity in another one. Furthermore, I don't want the latter one to know that C exists.
Here's how the B and C creation code looks like:
public B CreateB()
{
return new B
{
Id = Guid.NewGuid(),
C = new C(),
};
}
And now the link and save code:
// a is an instance of A that has been loaded from DB
// and hence has a persistent Id value.
// b is a just-created instance of B
// that has a non-persistent Id value and null reference to A.
void SaveBlahBlahBlah(A a, B b)
{
// At this point b and c have the same Id value.
// It differs from a's Id, but that's expected, they haven't been linked yet.
b.A = a;
// At this point b receives a's Id value, but c keeps the original one,
// therefore the existing b-c link gets broken!
using(var ctx = new MyContext())
{
ctx.As.Attach(a); // This throws exception saying
// I've violated referential integrity.
// It doesn't say which relationship is broken,
// but I guess it's the B-C one since
// the debugger shows them to have different values if PKs
ctx.Bs.AddObject(b);
ctx.SaveChanges();
}
}
I've tried this both with the default EF's code generator (the one that uses EF's Entity class as base class for generated entities) and with the Self-Tracking Entities code generator. The result is the same.
So, the code crashes. The reason is likely to be that after A and B have been linked, B and C get different PK values which is illegal for entities with 1-1 relationship.
What I expected was C to automatically get it's PK synchronized to the value B got from A instance. That seems reasonable because I work with an object graph, I have an existing B-C relation which is OK and I expect it to remain OK after linking B with A. Why would it break? I would understand it if either B or C existed in DB and I wasn't able to change their PKs. But it's not the case, both entites have been just created.
I cannot break the chain of keys by using separate from PKs columns for FKs because EF requires both sides of a 1-1 relationship to be the PKs.
I don't want to sync keys manually because in fact there are more 1-1 related tables and that would require the sync code to appear in many places.
I believe I will be able to update the T4 template of the STE generator to cascade PK updates down 1-1 relationships. But I'm not too familiar with T4 and not too happy to do that.
I have 2 questions:
Is my expectation of cascaded PK updates in my case wrong for some reasons? (Seems bizarre though) I.e., is it a bug or a feature?
Are there any other and preferably simpler ways to fix the issue than modifying the STE template? Maybe some magic options in EF mappings or context?
Thanks in advance.

Answer 1

The problem is that the service which handles the assignment of IDs from one referenced object to another is the context. But at the time when you actually make the association, neither object is in the context. That wouldn't normally be a problem because the relationship will be fixed up when you add B to the context.
Unfortunately, you don't do this. Instead, you create an additional relationship with A, but then lie to the context and claim that everything is already fixed up. More precisely, you call EntitySet.Attach, which is really only intended for already fixed-up objects.
On the other hand, code like this should work just fine:
public B CreateB()
{
return new B
{
Id = Guid.NewGuid(),
C = new C(),
};
}
void SaveBlahBlahBlah(A a, B b)
{
using(var ctx = new MyContext())
{
ctx.Bs.AddObject(b);
ctx.SaveChanges();
}
}
Note that all I've done here is delete the problematic code, which has nothing to do with the relationship between B and C.
In short, beware of Attach. You need to know what you're doing when you call it.
UPDATE
A version that handles existing instances of A:
void SaveBlahBlahBlah(A a, B b)
{
Debug.Assert(a.B != b);
using(var ctx = new MyContext())
{
ctx.As.Attach(a);
a.B = b; // it's crucial that this link is set after attaching a to context!
ctx.Bs.AddObject(b);
ctx.SaveChanges();
}
}