Writing reflexive extension functions for .NET

On the .NET platform extension functions may be implemented in any .NET language (the examples here assume C#).

An extension function is invoked using a name such as prefix:localname(). The prefix must be the prefix associated with a namespace declaration that is in scope. The namespace URI is used to identify a .NET class, and the local name is used to identify a method, property, or constructor within the class.

The command line option -TJ is useful for debugging the loading of .NET extensions. It gives detailed information about the methods that are examined for a possible match.

The basic form of the namespace URI is clitype: followed by the fully-qualified type name (for example xmlns:env="clitype:System.Environment"). This form works for system classes and classes in a loaded assembly. If an assembly needs to be loaded, extra information can be given in the form of URI query parameters. For example xmlns:env="clitype:Acme.Payroll.Employee?asm=payroll;ver=4.12.0.0".

The parameters that are recognized are:

If the from keyword is present, the other parameters are ignored. The value of from must be the URI of the DLL to be loaded: if it is relative, it is relative to the base URI of the expression containing the call to the extension function (regardless of where the namespace is actually declared).

If the partialname keyword is present, the assembly is loaded (if possible) using Assembly.LoadWithPartialName() and the other parameters are ignored.

If the assembly is a library DLL in the global assembly cache, use the gacutil /l command to list the assemblies present in the GAC, and to extract the required version, culture, and strong name attributes. For example, suppose you want to call a static method disappear() in class Conjurer in namespace Magic.Circle, and this class is contained in an assembly Magic listed as:

Magic, Version=7.2.2200.0, Culture=neutral, PublicKeyToken=b03f5f7f11d50a4b, Custom=null

Then the URI you would use to identify this class is clitype:Magic.Circle.Conjurer?asm=Magic;ver=7.2.2200.0;sn=b03f5f7f11d50a4b, and an actual call of the function might take the form:

Tips for Dynamic Loading in .NET

Here are some hints and tips that might help you to get dynamic loading working under .NET.

First, do you actually need to use dynamic loading? An easier approach is to link the code containing the extension functions into your C# or VB.NET code and pass Saxon a reference to the containing type. This can be done with:

If you do this, then extension functions using the namespace URI "http://example.com" will be found in the class/type Extension.Functions without any need for dynamic loading.

The command line option -TJ is useful for debugging the loading of .NET extensions. It gives detailed information about the methods that are examined for a possible match. The equivalent in the API is to set the Processor property "http://saxon.sf.net/feature/trace-external-functions" .

First decide whether you want to load the assembly (DLL) containing the extension functions from local filestore or from the Global Assembly Cache.

If you want to load it from the GAC you must compile the assembly with a strong name, and deploy it to the GAC. For advice on doing this, see http://support.microsoft.com/kb/324168. (In .NET framework 2.0 there was a handy Control Panel tool for this, but it is no longer available for security reasons.)

For local loading, the following techniques work:

• If the query or transformation is controlled from a user-written application in C# or another .NET language, include the extension function implementation in the same assembly as the controlling application, or in another assembly which is statically referenced from the controlling application. In this case it is only necessary to name the assembly, for example <e att="{my:ext()}" xmlns:my="clitype:Namespace.ClassName?asm=Samples"/>.

• Another approach is to copy the assembly into the same directory as the controlling application or, if using the Saxon Query or Transform command from the command line, the directory containing the Saxon Query.exe and Transform.exe executables. Again in this case it should simply be necessary to give the assembly name as above.

• As an alternative, an assembly held in local filestore can be loaded by reference to the file, for example xmlns:my="clitype:Namespace.ClassName?from=file:///c:/lib/Samples.dll". The URI can be given as an absolute URI, or as a relative URI which is interpreted relative to the base URI of the expression containing the function call.

For production running it is probably more appropriate to place the assembly holding extension functions in the global assembly cache. It can then generally be referenced in one of two ways:

• By partial name, for example xmlns:my="clitype:Namespace.ClassName?partialname=Samples"

• By fully-qualified name, for example xmlns:my="clitype:Namespace.ClassName?asm=Samples;ver=3.0.0.1;loc=neutral;sn=e1f2a3b4e1f2a3b4"

The following example shows how to call system methods in the .NET framework:

Identifying and Calling Specific Methods

The rest of this section considers how a .NET method, property, or constructor is identified. This decision (called binding) is always made at the time the XPath expression is compiled.

There are three cases to consider: static methods, constructors, and instance-level methods. In addition, a public property in a class is treated as if it were a zero-argument method, so static properties can be accessed in the same way as static methods, and instance-level properties in the same way as instance-level methods. (Note that the property name is used directly: it is not prefixed by "get".)

In XSLT, the system function function-available(String name) returns true if there appears to be a method available with the right name. The function also has an optional second argument to test whether there is a method with the appropriate number of arguments. However, it is not possible to test whether the arguments are of appropriate types. If the function name is "new" it returns true so long as the class is not an abstract class or interface, and so long as it has at least one constructor.