Thursday, November 15, 2007
« Primary motivations to move from 2- to 3... | Main | Barcelona User Group & Live Meeting »

I'm busy getting my work and home life in order, as I head off to Barcelona with my oldest son for a week. I'm speaking at the Barcelona .NET user group on Wednesday, 21 November, so if you are in the area stop by and say hello! Otherwise, we'll be around the Barcelona area, enjoying Spain - I really love Spain!

Some of what I'm busy trying to get organized are some of the upcoming changes to CSLA .NET in version 3.5. Thematically of course, the focus is on support for LINQ, primarily LINQ for Objects (like CSLA objects) and LINQ for SQL. As usual however, I'm also tackling some items off the wish list, and making other changes that I think will make things better/easier/more productive - and some that are just fun for me :)

When talking about LINQ for Objects, there are a couple things going on. First, Aaron Erickson, a fellow Magenicon, is incorporating his i4o concepts into CSLA .NET to give CSLA collections the ability to support indexed queries. Second, there are some basic plumbing things that need to happen for LINQ to play well with CSLA objects, most notably ensuring that a result (other than a projection) from a query against a CSLA collection results in a live, updatable view of the original collection.

By default LINQ just creates a new IEnumerable list with the selected items. But this can be horribly unproductive, because a user might add a new item or remove an existing item from that list - which would have no impact at all on the original list. Aaron has devised an approach by which a query against a BusinessListBase or ReadOnlyListBase will result in a richer resulting list, that is still connected to the original - much like SortedBindingList and FilteredBindingList are today. This way, when an item is removed from the view, it is also removed from the real list.

Back to the i4o stuff, allowing indexing of properties on child objects contained in a list can make for much faster queries. If you only do a single query on a list it might not be worth building an index. But if you do multiple queries to get different views of a list (projections or not), having an index on one or more child object properties can make a very big performance difference!

When talking about LINQ for SQL, the focus is really on data access. LINQ for SQL, in the CSLA world-view, is merely a replacement for ADO.NET. The same is true for the ADO.NET Entity Framework. Architecturally, both of these technologies simply replace the use of a raw DataReader or a DataSet/DataTable like you might use today. Alternately, you can say that they compete with existing tools like nHibernate or LLBLgen - tools that people already use instead of ADO.NET.

In short, this means that your DataPortal_XYZ methods may make LINQ queries instead of using an ADO.NET SqlCommand. They may load the business object's fields from a resulting LINQ object rather than from a DataReader. In my testing thus far, I find that some of my DataPortal_Fetch() methods shrink by nearly 50% by using LINQ. That's pretty cool!

Of course nothing is faster than using a raw DataReader. LINQ loads its objects using a DataReader too - so using LINQ does incur overhead that you don't suffer when using raw ADO.NET. However, many applications aren't performance-bound as much as they are maintenance-bound. That is to say that it is often more important to improve maintainability and reduce coding than it is to eek out that last bit of performance.

To that end, I'm working on enhancing and extending DataMapper to be more powerful and more efficient. Since nHibernate, LINQ and ADO.NET EF all return data transfer objects or entity objects, it becomes important to be able to easily copy the data from those objects into the fields of your business objects. And that has to happen without incurring too much overhead. You can do the copies by hand with code that copies each DTO property into a field, which is fast, but is a lot of code to write (and debug, and test). If DataMapper can do all that in a single line of code that's awesome, but the trick is to avoid incurring too much overhead from reflection or other dynamic mapping schemes.

One cool side-effect of this DataMapper work, is that it will work for LINQ and EF, but also for XML services and anything else where data comes back in some sort of DTO/entity object construct.

There are also some issues with LINQ for SQL and context. Unfortunately, LINQ's context object wasn't designed to support n-tier deployments, and doesn't work well with any n-tier model, including CSLA. This means that LINQ doesn't/can't keep track of things like whether an object is new/old or marked for deletion. On the upside, CSLA already does all that. On the downside, it makes doing inserts/updates/deletes with LINQ for SQL more complex than if you used LINQ in the simpler 2-tier world for which it was designed. I hope to come up with some ways to bridge this gap a little so as to preserve some of the simpler LINQ syntax for database updates, but I'm not overly confident...

I'm also doing a bunch of work to minimize and standardize the coding required to build a business object. Each release of CSLA has tended to standardize the code within a business object more and more. Sometimes this also shrinks the code, though not always. At the moment I'm incorporating some methods and techniques to shrink and standardize how properties are implemented, and how child objects (collections or single objects) are managed by the parent.

By borrowing some ideas from Microsoft's DependencyProperty concept, I've been able to shrink the typical property implementation by about 35% - from 14 lines to 9:

Private Shared NameProperty As PropertyInfo(Of String) = RegisterProperty(Of String, Customer)("Name")
Private _name As String = NameProperty.DefaultValue
Public Property Name() As String
  Get
    Return GetProperty(Of String)(NameProperty, _name)
  End Get
  Set(ByVal value As String)
    SetProperty(Of String)(NameProperty, _name, value)
  End Set
End Property

Unlike a DependencyProperty, this technique continues to use a strongly typed local field, which I think is a better approach (faster, though slightly less abstract).

Child objects are similar:

Private Shared LineItemsProperty As PropertyInfo(Of LineItems) = RegisterProperty(Of LineItems, Order)("LineItems")
Public ReadOnly Property LineItems() As LineItems
  Get
    If Not ChildExists(LineItemsProperty) Then
      SetChild(Of LineItems)(LineItemsProperty, LineItems.NewList())
    End If
    Return GetChild(Of LineItems)(LineItemsProperty)
  End Get
End Property

In this case the implementation is more similar to a DependencyProperty, in that BusinessBase really does contain and manage the child object. There's value to that though, because it means that BusinessBase now automatically handles IsValid and IsDirty so you don't need to override them. The next step is to make BusinessBase automatically handle PropertyChanged/ListChanged/CollectionChanged events from the child objects and echo those as a PropertyChanged event from the parent. Along with that I'll also make sure that the child's Parent property is automatically set. This should eliminate virtually all the easy-to-forget code related to child objects - resulting in a lot less code and much more maintainable parent objects.

As I noted earlier, there are a lot of things on the wish list, and I'll hit some of those as time permits.

One item that I've done in VB and need to port to C# is changing authorization to call a pluggable IsInRole() provider method, allowing people to alter the algorithm used to determine if the current user is in a specific role. The default continues to do what it has always done by calling principal.IsInRole(), but for some advanced scenarios this extensibility point is very valuable.

Beyond that I'll hit items on the wish list given time and energy.

However, I need to have all major work done on 3.5 by the end of January 2008, because I'll start working on Expert 2008 Business Objects, the third edition of the Business Objects book for .NET, in January. The tentative plan is to have the book out around the end of June, which will be a tall order given that it will probably expand by 300 pages or so by the time I'm done covering .NET 3.0 and 3.5...

Now I must do some packing, and get into the mindset required to survive 11 hours in the air - they just don't make airplane seats for people who are 6'5" tall.