User-Defined Type Requirements
You must make several important design decisions when creating a user-defined type (UDT) to be installed in Microsoft SQL Server. For most UDTs, creating the UDT as a structure is recommended, although creating it as a class is also an option. The UDT definition must conform to the specifications for creating UDTs in order for it to be registered with SQL Server.
To run in SQL Server, your UDT must implement the following requirements in the UDT definition:
The UDT must specify the Microsoft.SqlServer.Server.SqlUserDefinedTypeAttribute. The use of the System.SerializableAttribute is optional, but recommended.
The UDT must implement the System.Data.SqlTypes.INullable interface in the class or structure by creating a public static (Shared in Microsoft Visual Basic) Null method. SQL Server is null-aware by default. This is necessary for code executing in the UDT to be able to recognize a null value.
The UDT must contain a public static (or Shared) Parse method that supports parsing from, and a public ToString method for converting to a string representation of the object.
A UDT with a user-defined serialization format must implement the System.Data.IBinarySerialize interface and provide a Read and a Write method.
The UDT must implement System.Xml.Serialization.IXmlSerializable, or all public fields and properties must be of types that are XML serializable or decorated with the XmlIgnore attribute if overriding standard serialization is required.
There must be only one serialization of a UDT object. Validation fails if the serialize or deserialize routines recognize more than one representation of a particular object.
SqlUserDefinedTypeAttribute.IsByteOrdered must be true to compare data in byte order. If the IComparable interface is not implemented and SqlUserDefinedTypeAttribute.IsByteOrdered is false, byte order comparisons will fail.
A UDT defined in a class must have a public constructor that takes no arguments. You can optionally create additional overloaded class constructors.
The UDT must expose data elements as public fields or property procedures.
Public names cannot be longer than 128 characters, and must conform to the SQL Server naming rules for identifiers as defined in Identifiers.
sql_variant columns cannot contain instances of a UDT.
Inherited members are not accessible from Transact-SQL because the SQL Server type system is not aware of the inheritance hierarchy among UDTs. However, you can use inheritance when you structure your classes and you can call such methods in the managed code implementation of the type.
Members cannot be overloaded, except for the class constructor. If you do create an overloaded method, no error is raised when you register the assembly or create the type in SQL Server. Detection of the overloaded method occurs at run time, not when the type is created. Overloaded methods can exist in the class as long as they are never invoked. Once you invoke the overloaded method, an error is raised.
Any static (or Shared) members must be declared as constants or as read-only. Static members cannot be mutable.
If the SqlUserDefinedTypeAttribute.MaxByteSize field is set to -1, the serialized UDT can be as large as the large object (LOB) size limit (currently 2 GB). The size of the UDT cannot exceed the value specified in the MaxByteSized field.
Although it is not used by the server for performing comparisons, you can optionally implement the System.IComparable interface, which exposes a single method, CompareTo. This is used on the client side in situations in which it is desirable to accurately compare or order UDT values.
Choosing the right serialization attributes for your UDT depends on the type of UDT you are trying to create. The Native serialization format utilizes a very simple structure that enables SQL Server to store an efficient native representation of the UDT on disk. The Native format is recommended if the UDT is simple and only contains fields of the following types:
bool, byte, sbyte, short, ushort, int, uint, long, ulong, float, double, SqlByte, SqlInt16, SqlInt32, SqlInt64, SqlDateTime, SqlSingle, SqlDouble, SqlMoney, SqlBoolean
Value types that that are composed of fields of the above types are good candidates for Native format, such as structs in Visual C#, (or Structures as they are known in Visual Basic). For example, a UDT specified with the Native serialization format may contain a field of another UDT that was also specified with the Native format. If the UDT definition is more complex and contains data types not on the above list, you must specify the UserDefined serialization format instead.
The Native format has the following requirements:
The type must not specify a value for Microsoft.SqlServer.Server.SqlUserDefinedTypeAttribute.MaxByteSize.
All fields must be serializable.
The System.Runtime.InteropServices.StructLayoutAttribute must be specified as StructLayout.LayoutKindSequential if the UDT is defined in a class and not a structure. This attribute controls the physical layout of the data fields and is used to force the members to be laid out in the order in which they appear. SQL Server uses this attribute to determine the field order for UDTs with multiple values.
For an example of a UDT defined with Native serialization, see the Point UDT in Coding User-Defined Types.
The UserDefined format setting for the Microsoft.SqlServer.Server.SqlUserDefinedTypeAttribute attribute gives the developer full control over the binary format. When specifying the Format attribute property as UserDefined, you must do the following in your code:
Specify the optional IsByteOrdered attribute property. The default value is false.
Specify the MaxByteSize property of the Microsoft.SqlServer.Server.SqlUserDefinedTypeAttribute.
Write code to implement Read and Write methods for the UDT by implementing the System.Data.Sql.IBinarySerialize interface.
For an example of a UDT defined with UserDefined serialization, see the Currency UDT in Coding User-Defined Types.
Beginning with the SQL Server 2005 RTM version, CLR UDTs with user-defined serialization were allowed to have their fields indexed as part of non-persisted computed columns or views. In such situations, non-deterministic UDT serialization/deserialization could lead to index corruption, and therefore has been removed from SQL Server 2005 SP1. In SQL Server 2005 SP1, UDT fields must use native serialization or be persisted in order to be indexed. Any existing indexes on UDT fields should continue to function as before.
Attributes determine how serialization is used to construct the storage representation of UDTs and to transmit UDTs by value to the client. You are required to specify the Microsoft.SqlServer.Server.SqlUserDefinedTypeAttribute when creating the UDT. The Microsoft.SqlServer.Server.SqlUserDefinedTypeAttribute attribute indicates that the class is a UDT and specifies the storage for the UDT. You can optionally specify the Serializable attribute, although SQL Server does not require this.
The Microsoft.SqlServer.Server.SqlUserDefinedTypeAttribute has the following properties.
When the Microsoft.SqlServer.Server.SqlUserDefinedTypeAttribute.IsByteOrdered property is set to true, you are in effect guaranteeing that the serialized binary data can be used for semantic ordering of the information. Thus, each instance of a byte-ordered UDT object can only have one serialized representation. When a comparison operation is performed in SQL Server on the serialized bytes, its results should be the same as if the same comparison operation had taken place in managed code. The following features are also supported when IsByteOrdered is set to true:
The ability to create indexes on columns of this type.
The ability to create primary and foreign keys as well as CHECK and UNIQUE constraints on columns of this type.
The ability to use Transact-SQL ORDER BY, GROUP BY, and PARTITION BY clauses. In these cases, the binary representation of the type is used to determine the order.
The ability to use comparison operators in Transact-SQL statements.
The ability to persist computed columns of this type.
Note that both the Native and UserDefined serialization formats support the following comparison operators when IsByteOrdered is set to true:
Equal to (=)
Not equal to (!=)
Greater than (>)
Less than (<)
Greater than or equal to (>=)
Less than or equal to (<=)
In addition to specifying the attributes for your assemblies correctly, your class must also support nullability. UDTs loaded into SQL Server are null-aware, but in order for the UDT to recognize a null value, the class must implement the INullable interface. For more information and an example of how to implement nullability in a UDT, see Coding User-Defined Types.
To support string conversion to and from the UDT, you must provide a Parse method and a ToString method in your class. The Parse method allows a string to be converted into a UDT. It must be declared as static (or Shared in Visual Basic), and take a parameter of type System.Data.SqlTypes.SqlString. For more information and an example of how to implement the Parse and ToString methods, see Coding User-Defined Types.
UDTs must support conversion to and from the xml data type by conforming to the contract for XML serialization. The System.Xml.Serialization namespace contains classes that are used to serialize objects into XML format documents or streams. You can choose to implement xml serialization by using the IXmlSerializable interface, which provides custom formatting for XML serialization and deserialization.
In addition to performing explicit conversions from UDT to xml, XML serialization enables you to:
Use Xquery over values of UDT instances after conversion to the xml data type.
Use UDTs in parameterized queries and Web methods with Native XML Web Services in SQL Server. For more information, see Handling the xml Data Type and CLR User-defined Types.
Use UDTs to receive a bulk load of XML data.
Serialize DataSets that contain tables with UDT columns.
UDTs are not serialized in FOR XML queries. To execute a FOR XML query that displays the XML serialization of UDTs, explicitly convert each UDT column to the xml data type in the SELECT statement. You can also explicitly convert the columns to varbinary, varchar, or nvarchar.