In a blog posting about a few weeks ago, I'd written about noticing a DataSet being created in an Entity Framework program by using the ADO.NET client trace facility. Entity Framework programs do (indirectly) cause a DataSet to be created, but its only for resolving the ProviderInvariantName of the underlying data provider. NOT for anything related to the functioning of the EntityClient or ObjectContext in any way. I've revised the original blog posting to remove the DataSet reference, but just in case you already read it….
The connection string for the EntityClient data provider contains a parameter that refers to the ProviderInvariantName of the underlying data provider (e.g. System.Data.SqlClient). Entity Framework validates the underlying data provider name by using the ADO.NET provider enumerator class DbProviderFactories. DbProviderFactories creates the DataSet to store the provider information obtained from the machine.config file.
I might have guessed this…as I'd written a whitepaper on the functionality of the ADO.NET provider model, "Generic Coding with the ADO.NET 2.0 Base Classes and Factories" not more than a few years ago. Using the DbProviderFactories class is the only way to enumerate the ADO.NET data providers installed on your machine.
Come to my "End-To-End Tracing with SQL Server 2008 and ADO.NET" talk at TechEd next week and I'll show you the trace.