Edit

Share via

CLR scalar-valued functions

  • Article

Applies to: SQL Server

A scalar-valued function (SVF) returns a single value, such as a string, integer, or bit value. You can create scalar-valued user-defined functions in managed code using any .NET Framework programming language. These functions are accessible to Transact-SQL or other managed code. For information about the advantages of common language runtime (CLR) integration and choosing between managed code and Transact-SQL, see CLR integration overview.

Requirements for CLR scalar-valued functions

.NET Framework SVFs are implemented as methods on a class in a .NET Framework assembly. The input parameters and the type returned from an SVF can be any of the scalar data types supported by SQL Server, except varchar, char, rowversion, text, ntext, image, timestamp, table, or cursor. SVFs must ensure a match between the SQL Server data type and the return data type of the implementation method. For more information about type conversions, see Map CLR parameter data.

When you implement a .NET Framework SVF in a .NET Framework language, you can specify the SqlFunction custom attribute to include additional information about the function. The SqlFunction attribute indicates whether or not the function accesses or modifies data, if it's deterministic, and if the function involves floating point operations.

Scalar-valued user-defined functions might be deterministic or non-deterministic. A deterministic function always returns the same result when it's called with a specific set of input parameters. A non-deterministic function might return different results when it's called with a specific set of input parameters.

Note

Don't mark a function as deterministic if the function doesn't always produces the same output values, given the same input values and the same database state. Marking a function as deterministic, when the function isn't truly deterministic can result in corrupted indexed views and computed columns. You mark a function as deterministic by setting the IsDeterministic property to true.

Table-valued parameters

Table-valued parameters (TVPs), user-defined table types that are passed into a procedure or function, provide an efficient way to pass multiple rows of data to the server. TVPs provide similar functionality to parameter arrays, but offer greater flexibility and closer integration with Transact-SQL. They also provide the potential for better performance.

TVPs also help reduce the number of round trips to the server. Instead of sending multiple requests to the server, such as with a list of scalar parameters, data can be sent to the server as a TVP. A user-defined table type can't be passed as a table-valued parameter to, or be returned from, a managed stored procedure or function executing in the SQL Server process. For more information about TVPs, see Use table-valued parameters (Database Engine).

Example of a CLR scalar-valued function

Here's a simple SVF that accesses data and returns an integer value:

using Microsoft.SqlServer.Server;
using System.Data.SqlClient;

public class T
{
    [SqlFunction(DataAccess = DataAccessKind.Read)]
    public static int ReturnOrderCount()
    {
        using (SqlConnection conn
            = new SqlConnection("context connection=true"))
        {
            conn.Open();
            SqlCommand cmd = new SqlCommand(
                "SELECT COUNT(*) AS 'Order Count' FROM SalesOrderHeader", conn);
            return (int)cmd.ExecuteScalar();
        }
    }
}

The first line of code references Microsoft.SqlServer.Server to access attributes and System.Data.SqlClient to access the ADO.NET namespace. (This namespace contains SqlClient, the .NET Framework Data Provider for SQL Server.)

Next, the function receives the SqlFunction custom attribute, which is found in the Microsoft.SqlServer.Server namespace. The custom attribute indicates whether or not the user-defined function (UDF) uses the in-process provider to read data in the server. SQL Server doesn't allow UDFs to update, insert, or delete data. SQL Server can optimize execution of a UDF that doesn't use the in-process provider. This is indicated by setting DataAccessKind to DataAccessKind.None. On the next line, the target method is a public static (shared in Visual Basic .NET).

The SqlContext class, located in the Microsoft.SqlServer.Server namespace, can then access a SqlCommand object with a connection to the SQL Server instance that is already set up. Although not used here, the current transaction context is also available through the System.Transactions application programming interface (API).

Most of the lines of code in the function body should look familiar to developers who write client applications that use the types found in the System.Data.SqlClient namespace.

using(SqlConnection conn = new SqlConnection("context connection=true"))
{
   conn.Open();
   SqlCommand cmd = new SqlCommand(
        "SELECT COUNT(*) AS 'Order Count' FROM SalesOrderHeader", conn);
   return (int) cmd.ExecuteScalar();
}

The appropriate command text is specified by initializing the SqlCommand object. The previous example counts the number of rows in table SalesOrderHeader. Next, the ExecuteScalar method of the cmd object is called. This returns a value of type int based on the query. Finally, the order count is returned to the caller.

If this code is saved in a file called FirstUdf.cs, it could be compiled into as assembly as follows:

csc.exe /t:library /out:FirstUdf.dll FirstUdf.cs

/t:library indicates that a library, rather than an executable, should be produced. Executables can't be registered in SQL Server.

Visual C++ database objects compiled with /clr:pure aren't supported for execution on SQL Server. For example, such database objects include scalar-valued functions.

The Transact-SQL query and a sample invocation to register the assembly and UDF are:

CREATE ASSEMBLY FirstUdf
    FROM 'FirstUdf.dll';
GO

CREATE FUNCTION CountSalesOrderHeader()
RETURNS INT
AS EXTERNAL NAME FirstUdf.T.ReturnOrderCount;
GO

SELECT dbo.CountSalesOrderHeader();
GO

The function name as exposed in Transact-SQL doesn't need to match the name of the target public static method.

Related content