Can I force the auto-generated Linq-to-SQL classes to use an OUTER JOIN?

Question

Let's say I have an Order table which has a FirstSalesPersonId field and a SecondSalesPersonId field. Both of these are foreign keys that reference the SalesPerson table. For any given order, either one or two salespersons may be credited with the order. In other words, FirstSalesPersonId can never be NULL, but SecondSalesPersonId can be NULL.

When I drop my Order and SalesPerson tables onto the "Linq to SQL Classes" design surface, the class builder spots the two FK relationships from the Order table to the SalesPerson table, and so the generated Order class has a SalesPerson field and a SalesPerson1 field (which I can rename to SalesPerson1 and SalesPerson2 to avoid confusion).

Because I always want to have the salesperson data available whenever I process an order, I am using DataLoadOptions.LoadWith to specify that the two salesperson fields are populated when the order instance is populated, as follows:

dataLoadOptions.LoadWith<Order>(o => o.SalesPerson1);
dataLoadOptions.LoadWith<Order>(o => o.SalesPerson2);

The problem I'm having is that Linq to SQL is using something like the following SQL to load an order:

SELECT ...
FROM Order O
INNER JOIN SalesPerson SP1 ON SP1.salesPersonId = O.firstSalesPersonId
INNER JOIN SalesPerson SP2 ON SP2.salesPersonId = O.secondSalesPersonId

This would make sense if there were always two salesperson records, but because there is sometimes no second salesperson (secondSalesPersonId is NULL), the INNER JOIN causes the query to return no records in that case.

What I effectively want here is to change the second INNER JOIN into a LEFT OUTER JOIN. Is there a way to do that through the UI for the class generator? If not, how else can I achieve this?

(Note that because I'm using the generated classes almost exclusively, I'd rather not have something tacked on the side for this one case if I can avoid it).

---

Edit: per my comment reply, the SecondSalesPersonId field is nullable (in the DB, and in the generated classes).

Answer 1

If you make the secondSalesPersonId field in the database table nullable, LINQ-to-SQL should properly construct the Association object so that the resulting SQL statement will do the LEFT OUTER JOIN.

UPDATE: Since the field is nullable, your problem may be in explicitly declaring dataLoadOptions.LoadWith<>(). I'm running a similar situation in my current project where I have an Order, but the order goes through multiple stages. Each stage corresponds to a separate table with data related to that stage. I simply retrieve the Order, and the appropriate data follows along, if it exists. I don't use the dataLoadOptions at all, and it does what I need it to do. For example, if the Order has a purchase order record, but no invoice record, Order.PurchaseOrder will contain the purchase order data and Order.Invoice will be null. My query looks something like this:

DC.Orders.Where(a => a.Order_ID == id).SingleOrDefault();

I try not to micromanage LINQ-to-SQL...it does 95% of what I need straight out of the box.

UPDATE 2: I found this post that discusses the use of DefaultIfEmpty() in order to populated child entities with null if they don't exist. I tried it out with LINQPad on my database and converted that example to lambda syntax (since that's what I use):

ParentTable.GroupJoin
(
    ChildTable, 
    p => p.ParentTable_ID, 
    c => c.ChildTable_ID, 
    (p, aggregate) => new { p = p, aggregate = aggregate }
)
.SelectMany (a => a.aggregate.DefaultIfEmpty (), 
    (a, c) => new 
    { 
        ParentTableEntity = a.p, 
        ChildTableEntity = c
    }
)

From what I can figure out from this statement, the GroupJoin expression relates the parent and child tables, while the SelectMany expression aggregates the related child records. The key appears to be the use of the DefaultIfEmpty, which forces the inclusion of the parent entity record even if there are no related child records. (Thanks for compelling me to dig into this further...I think I may have found some useful stuff to help with a pretty huge report I've got on my pipeline...)

UPDATE 3: If the goal is to keep it simple, then it looks like you're going to have to reference those salesperson fields directly in your Select() expression. The reason you're having to use LoadWith<>() in the first place is because the tables are not being referenced anywhere in your query statement, so the LINQ engine won't automatically pull that information in.

As an example, given this structure:

MailingList               ListCompany
===========               ===========
List_ID (PK)              ListCompany_ID (PK)
ListCompany_ID (FK)       FullName (string)

I want to get the name of the company associated with a particular mailing list:

MailingLists.Where(a => a.List_ID == 2).Select(a => a.ListCompany.FullName)

If that association has NOT been made, meaning that the ListCompany_ID field in the MailingList table for that record is equal to null, this is the resulting SQL generated by the LINQ engine:

SELECT [t1].[FullName]
FROM [MailingLists] AS [t0]
LEFT OUTER JOIN [ListCompanies] AS [t1] ON [t1].[ListCompany_ID] = [t0].[ListCompany_ID]
WHERE [t0].[List_ID] = @p0

Answer 2

The default behaviour actually is a LEFT JOIN, assuming you've set up the model correctly.

Here's a slightly anonymized example that I just tested on one of my own databases:

class Program
{
    static void Main(string[] args)
    {
        using (TestDataContext context = new TestDataContext())
        {
            DataLoadOptions dlo = new DataLoadOptions();
            dlo.LoadWith<Place>(p => p.Address);
            context.LoadOptions = dlo;

            var places = context.Places.Where(p => p.ID >= 100 && p.ID <= 200);
            foreach (var place in places)
            {
                Console.WriteLine(p.ID, p.AddressID);
            }
        }
    }
}

This is just a simple test that prints out a list of places and their address IDs. Here is the query text that appears in the profiler:

SELECT [t0].[ID], [t0].[Name], [t0].[AddressID], ...
FROM [dbo].[Places] AS [t0]
LEFT OUTER JOIN (
    SELECT 1 AS [test], [t1].[AddressID],
        [t1].[StreetLine1], [t1].[StreetLine2],
        [t1].[City], [t1].[Region], [t1].[Country], [t1].[PostalCode]
    FROM [dbo].[Addresses] AS [t1]
) AS [t2] ON [t2].[AddressID] = [t0].[AddressID]
WHERE ([t0].[PlaceID] >= @p0) AND ([t0].[PlaceID] <= @p1)

This isn't exactly a very pretty query (your guess is as good as mine as to what that 1 as [test] is all about), but it's definitively a LEFT JOIN and doesn't exhibit the problem you seem to be having. And this is just using the generated classes, I haven't made any changes.

Note that I also tested this on a dual relationship (i.e. a single Place having two Address references, one nullable, one not), and I get the exact same results. The first (non-nullable) gets turned into an INNER JOIN, and the second gets turned into a LEFT JOIN.

It has to be something in your model, like changing the nullability of the second reference. I know you say it's configured as nullable, but maybe you need to double-check? If it's definitely nullable then I suggest you post your full schema and DBML so somebody can try to reproduce the behaviour that you're seeing.