ARQ - Writing Filter Functions

Applications can add SPARQL functions to the query engine. This is done by writing a class implementing the right interface, then either registering it or using the fake java: URI scheme to dynamically call the function.

Writing SPARQL Value Functions

A SPARQL value function is an extension point of the SPARQL query language that allows URI to name a function in the query processor.

In the ARQ engine, code to implement function must implement the interface org.apache.jena.sparql.function.Function although it is easier to work with one of the abstract classes for specific numbers of arguments like org.apache.jena.sparql.function.FunctionBase1 for one argument functions. Functions do not have to have a fixed number of arguments.

The abstract class FunctionBase, the superclass of FunctionBase1 to FunctionBase4, evaluates its arguments and calls the implementation code with argument values (if a variable was unbound, an error will have been generated) 

It is possible to get unevaluated arguments but care must be taken not to violate the rules of function evaluation. The standard functions that access unevaluated arguments are the logical 'or' and logical 'and' operations that back || and && are special forms to allow for the special exception handling rules.

Normally, function should be a pure evaluation based on it's argument. It should not access a graph nor return different values for the same arguments (to allow expression optimization). Usually, these requirements can be better met with a property function. Functions can't bind a variables; this would be done in a property function as well.

Example: (this is the max function in the standard ARQ library):

public class max extends FunctionBase2
{
    public max() { super() ; }
    public NodeValue exec(NodeValue nv1, NodeValue nv2)
    {
        return Functions.max(nv1, nv2) ;
    }
}

The function takes two arguments and returns a single value. The class NodeValue represents values and supports value-based operations. NodeValue value support includes the XSD datatypes, xsd:decimal and all it's subtypes like xsd:integer and xsd:byte, xsd';double, xsd:float, xsd:boolean, xsd:dateTime and xsd:date. Literals with language tags are also treated as values in additional "value spaces" determined by the language tag without regard to case.

The Functions class contains the core XML Functions and Operators operations. Class NodeFunctions contains the implementations of node-centric operations like isLiteral and str.

If any of the arguments are wrong, then the function should throw ExprEvalException.

Example: calculate the canonical namespace from a URI (calls the Jena operation for the actual work):

public class namespace extends FunctionBase1
{
    public namespace() { super() ; }

    public NodeValue exec(NodeValue v)
    {
        Node n = v.asNode() ;
        if ( ! n.isURI() )
            throw new ExprEvalException("Not a URI: "+FmtUtils.stringForNode(n)) ;
        String str = n.getNameSpace() ;
        return NodeValue.makeString(str) ;
    }
}

This throws an evaluation exception if it is passed a value that is not a URI.

The standard library, in package org.apache.jena.sparql.function.library, contains many examples.

Registering Functions

The query compiler finds functions based on the functions URI.  There is a global registry of known functions, but any query execution can have it's own function registry.

For each function, there is a function factory associated with the URI. A new function instance is created for each use of a function in each query execution.

// Register with the global registry.
FunctionRegistry.get().put("http://example.org/function#myFunction", new MyFunctionFactory()) ;

A common case is registering a specific class for a function implementation so there is an addition method that takes a class, wraps in a built-in function factory and registers the function implementation.

// Register with the global registry.
FunctionRegistry.get().put("http://example.org/function#myFunction", MyFunction.class) ;

Another convenience route to function calling is to use the java: URI scheme. This dynamically loads the code, which must be on the Java classpath. With this scheme, the function URI gives the class name. There is automatic registration of a wrapper into the function registry. This way, no explicit registration step is needed by the application and queries issues with the command line tools can load custom functions.

PREFIX f: <java:app.myFunctions.>
...
   FILTER f:myTest(?x, ?y)
...
   FILTER (?x + f:myIntToXSD(?y))
...

ARQ documentation index