Dynamic Data samples: Custom metadata providers

6/25 Update: This sample has become part of the Dynamic Data Futures project on codeplex. You can still use this code but you should check out the Futures project as it contains a lot more. The key conecpts are the same, though some names or APIs might have changed.

Note: This post is part of a series, see the list of other Dynamic Data samples.

A while back I posted some sample code for adding custom metadata providers. Since then much has changed in the Dynamic Data runtime and it is time to post an updated version. This sample illustrates how to write a custom metadata provider that allows you to programmatically add metadata attributes to an in-memory store before you register your model with the Dynamic Data runtime in Global.asax.

This sample should work with the latest preview release of Dynamic Data.

Running the sample

Download the metadata provider sample solution, extract to your preferred location, and open in Visual Studio 2008 SP1. The solution contains two projects: a sample website and a small library project that contains the metadata provider code.

Run the included website and go the Products list page. Once there hit Edit on any of the rows. You will be taken to a details view for the row you are editing. Change the UnitsInStock field to a negative value and move out of the field: a range validation error appears. Go back to the same field, clear it, and move out again: a required validation error appears.

How it works

The website scaffolds a small subset of the Northwind sample database. The UnitsInStock column on the Product table is decorated with two attributes: RequiredAttribute and RangeAttribute. However, the unique thing here is that each attribute is coming from a different source.

The RequiredAttribute is declared in the default Dynamic Data way using a metadata proxy class:

 [MetadataType(typeof(Product_MD))]
public partial class Product {

    private class Product_MD {
        [Required(ErrorMessage="This field is required [from MetadataType]")]
        public object UnitsInStock { get; set; }
    }
}

The RangeAttribute is added to the InMemoryMetadataManager class in Global.asax:

 InMemoryMetadataManager.AddColumnAttributes<Product>(p => p.UnitsInStock,
    new RangeAttribute(0, 1000) { 
        ErrorMessage = "This field must be between {1} and {2} [from InMemeroyMetadataManager]." }
);

The AddColumnAttributes function shown here has a way to strongly type the property references (as opposed to writing something like AddColumnAttributes("UnitsInStock", ...)) using a simple lambda expression. This provides for some nice IntelliSense support that would not be available if you were referring to properties using simple strings.

In order to have the Dynamic Data runtime pick up the metadata attributes added to InMemoryMetadataManager you need to modify your model registration call in Global.asax:

 model.RegisterContext(typeof(NorthwindDataContext), new ContextConfiguration() {
    ScaffoldAllTables = true,
    MetadataProviderFactory =
        (type => new InMemoryMetadataTypeDescriptionProvider(type, new AssociatedMetadataTypeTypeDescriptionProvider(type)))
});

The MetadataProviderFactory property of ContextConfiguration lets you specify a metadata provider factory method that is used by the Dynamic Data runtime to obtain an instance of a TypeDescriptionProvider that acts as the source of metadata attributes for a given table (type). The sample above also illustrates TypeDescriptionProvider chaining, which is what supports the fact that the table receives metadata from two sources: the InMemory provider adds its metadata to the results returned by the AssociatedMetadataType provider and returns a combined collection.

Truly dynamic metadata

At this point it is tempting to consider a dynamic metadata provider that would let you add, remove, and modify attributes throughout the lifetime of the application and not just during the application's startup phase. For example, you could imagine having some sort of administrative interface that would let you turn on and off whether a field is required or modify its description, even after the model has already been registered.

While it would seem that such a scenario would require minimal changes to the InMemoryMetadataManager class there is one detail about Dynamic Data that makes this a lot more complicated: the current Dynamic Data runtime fetches the metadata upon model registration and then caches it inernally for the lifetime of the app domain. This means that any changes that you make to the metadata through InMemoryMetadataManager after the model has been registered will be ignored.

This is a limitation of Dynamic Data that we will address in future versions. For the time being the only way to get around this would be to unload the app domain. However, this is means that you will need to find a place (such as a database) to store the new metadata while the app restarts.

What about the XML metadata?

My old custom metadata provider sample had an implementation of an XML-based metadata provider but the current one does not. The reason for this is that since the December CTP we have added support for even more attributes and some of the attributes have become a lot more complex. Because much of the old sample had to do with deserializing CLR attributes from an XML representation instead of anything specific to Dynamic Data I decided not to develop the old XML provider any further. However, if somebody ever wrote the appropriate XML parsing code it could easily be used to populate the InMemoryMetadataManager with the right values, essentially resulting in an XML metadata provider. Just remember about the domain unloading mentioned earlier.

MetadataProviderSample.zip