ARQ - Writing Property Functions

ARQ - Writing Property Functions

See also Writing Filter Functions.

Applications can add SPARQL property functions to the query engine. This is done by first implementing the PropertyFunction interface, and then either registering that function or using the fake java: URI scheme to dynamically load the function.

Writing SPARQL Property Functions

Similar to SPARQL Filter Functions, a SPARQL Property Function is an extension point of the SPARQL query language that allows a URI to name a function in the query processor. A key difference is that Property Functions may generate new bindings.

Just like org.apache.jena.sparql.function.Function there are various utility classes provided to simplify the creation of a Property Function. The selection of one depends on the 'style' of the desired built-in. For example, PFuncSimple is expected to be the predicate of triple patterns ?such ex:as ?this, where neither argument is an rdf:list, and either may be a variable. Alternatively, PFuncAssignToObject assumes that the subject will be bound, while the object will be a variable.

PropertyFunction
    |
    |--PropertyFunctionBase
          |
          |--PropertyFunctionEval
                |
                |--PFuncSimpleAndList
                |
                |--PFuncSimple
                     |
                     |--PFuncAssignToObject
                     |
                     |--PFuncAssignToSubject
                |
                |--PFuncListAndSimple
                |
                |--PFuncListAndList

The choice of extension point determines the function signature that the developer will need to implement, and primarily determines whether some of the arguments will be org.apache.jena.graph.Nodes or org.apache.jena.sparql.pfunction.PropFuncArgs. In the latter case, the programmer can determine whether the argument is a list as well as how many arguments it consists of.

Registration

Every property function is associated with a particular org.apache.jena.sparql.util.Context. This allows you to limit the availability of the function to be global or associated with a particular dataset. For example, a custom Property Function may expose an index which only has meaning with respect to some set of data.

Assuming you have an implementation of org.apache.jena.sparql.pfunction.PropertyFunctionFactory (shown later), you can register a function as follows:

final PropertyFunctionRegistry reg = PropertyFunctionRegistry.chooseRegistry(ARQ.getContext());
reg.put("urn:ex:fn#example", new ExamplePropertyFunctionFactory);
PropertyFunctionRegistry.set(ARQ.getContext(), reg);

The only difference between global and dataset-specific registration is where the Context object comes from:

final Dataset ds = DatasetFactory.createMem();
final PropertyFunctionRegistry reg = PropertyFunctionRegistry.chooseRegistry(ds.getContext());
reg.put("urn:ex:fn#example", new ExamplePropertyFunctionFactory);
PropertyFunctionRegistry.set(ds.getContext(), reg);

Note that org.apache.jena.sparql.pfunction.PropertyFunctionRegistry has other put methods that allow registration by passing a Class object, as well.

Implementation

The implementation of a Property Function is actually quite straight forward once one is aware of the tools at their disposal to do so. For example, if we wished to create a Property Function that returns no results regardless of their arguments we could do so as follows:

public class ExamplePropertyFunctionFactory implements PropertyFunctionFactory {
    @Override
    public PropertyFunction create(final String uri)
    {   
        return new PFuncSimple()
        {
            @Override
            public QueryIterator execEvaluated(final Binding parent, final Node subject, final Node predicate, final Node object, final ExecutionContext execCxt) 
            {   
                return QueryIterNullIterator.create(execCtx);
            }
        };
    }
}

Node and PropFuncArg objects allow the developer to reflect on the state of the arguments, and choose what bindings to generate given the intended usage of the Property Function. For example, if the function expects a list of three bound arguments for the object of the property, then it can throw a ExprEvalException (or derivative) to indicate incorrect use. It is the responsability of the developer to identify what parts of the argument are bound, and to respond appropriately.

For example, if ?a ex:f ?b were a triple pattern in a query, it could be called with ?a bound, ?b bound, or neither. It may make sense to return new bindings that include ?b if passed a concrete value for ?a, or conversely to generate new bindings for ?a when passed a concrete ?b. If both ?a and ?b are bound, and the function wishes to confirm that the pairing is valid, it can return the existing binding. If there are no valid solutions to return, then an empty solution may be presented.

There are several extremely useful implementations of QueryIterator within the Jena library that make it easy to support typical use cases.

Of particular note:

The second two cases require instances of Binding objects which can be obtained through static methods of BindingFactory. Creation of Binding objects will also require references to Var and NodeFactory

Note that it can make a lot of sense to generate the Iterator<Binding> for QueryIterPlainWrapper by means of Jena's ExtendedIterator. This can allow domain-specific value to be easily mapped to Binding objects in a lazy fashion.