public void SparqlBgpEvaluation()
{
//Prepare the Store
TripleStore store = new TripleStore();
Graph g = new Graph();
FileLoader.Load(g, "Turtle.ttl");
store.Add(g);
SparqlQueryParser parser = new SparqlQueryParser();
SparqlQuery q = parser.ParseFromString(@"PREFIX rdfs: <http://www.w3.org/2000/01/rdf-schema#>
SELECT * WHERE {?s ?p ?o . ?s rdfs:label ?label}");
Object testResult = store.ExecuteQuery(q);
ISparqlAlgebra testAlgebra = q.ToAlgebra();
if (testResult is SparqlResultSet)
{
SparqlResultSet rset = (SparqlResultSet)testResult;
Console.WriteLine(rset.Count + " Results");
foreach (SparqlResult r in rset)
{
Console.WriteLine(r.ToString());
}
Console.WriteLine();
}
//Create some Triple Patterns
TriplePattern t1 = new TriplePattern(new VariablePattern("?s"), new VariablePattern("?p"), new VariablePattern("?o"));
TriplePattern t2 = new TriplePattern(new VariablePattern("?s"), new NodeMatchPattern(g.CreateUriNode("rdfs:label")), new VariablePattern("?label"));
TriplePattern t3 = new TriplePattern(new VariablePattern("?x"), new VariablePattern("?y"), new VariablePattern("?z"));
TriplePattern t4 = new TriplePattern(new VariablePattern("?s"), new NodeMatchPattern(g.CreateUriNode(":name")), new VariablePattern("?name"));
//Build some BGPs
Bgp selectNothing = new Bgp();
Bgp selectAll = new Bgp(t1);
Bgp selectLabelled = new Bgp(new List<ITriplePattern>() { t1, t2 });
Bgp selectAllDisjoint = new Bgp(new List<ITriplePattern>() { t1, t3 });
Bgp selectLabels = new Bgp(t2);
Bgp selectNames = new Bgp(t4);
//LeftJoin selectOptionalNamed = new LeftJoin(selectAll, new Optional(selectNames));
LeftJoin selectOptionalNamed = new LeftJoin(selectAll, selectNames);
Union selectAllUnion = new Union(selectAll, selectAll);
Union selectAllUnion2 = new Union(selectAllUnion, selectAll);
Filter selectAllUriObjects = new Filter(selectAll, new UnaryExpressionFilter(new IsUriFunction(new VariableExpressionTerm("o"))));
//Test out the BGPs
//Console.WriteLine("{}");
//this.ShowMultiset(selectNothing.Evaluate(new SparqlEvaluationContext(null, store)));
//Console.WriteLine("{?s ?p ?o}");
//this.ShowMultiset(selectAll.Evaluate(new SparqlEvaluationContext(null, store)));
//Console.WriteLine("{?s ?p ?o . ?s rdfs:label ?label}");
//SparqlEvaluationContext context = new SparqlEvaluationContext(null, store);
//this.ShowMultiset(selectLabelled.Evaluate(context));
//SparqlResultSet lvnResult = new SparqlResultSet(context);
//Console.WriteLine("{?s ?p ?o . ?x ?y ?z}");
//this.ShowMultiset(selectAllDisjoint.Evaluate(new SparqlEvaluationContext(null, store)));
//Console.WriteLine("{?s ?p ?o . OPTIONAL {?s :name ?name}}");
//this.ShowMultiset(selectOptionalNamed.Evaluate(new SparqlEvaluationContext(null, store)));
Console.WriteLine("{{?s ?p ?o} UNION {?s ?p ?o}}");
this.ShowMultiset(selectAllUnion.Evaluate(new SparqlEvaluationContext(null, new InMemoryDataset(store))));
Console.WriteLine("{{?s ?p ?o} UNION {?s ?p ?o} UNION {?s ?p ?o}}");
this.ShowMultiset(selectAllUnion2.Evaluate(new SparqlEvaluationContext(null, new InMemoryDataset(store))));
Console.WriteLine("{?s ?p ?o FILTER (ISURI(?o))}");
this.ShowMultiset(selectAllUriObjects.Evaluate(new SparqlEvaluationContext(null, new InMemoryDataset(store))));
}
/// <summary>
/// Optimises BGPs in the Algebra to use Filter() and Extend() rather than the embedded FILTER and BIND
/// </summary>
/// <param name="algebra">Algebra to optimise</param>
/// <returns></returns>
public ISparqlAlgebra Optimise(ISparqlAlgebra algebra)
{
if (algebra is IAbstractJoin)
{
return ((IAbstractJoin)algebra).Transform(this);
}
else if (algebra is IUnaryOperator)
{
return ((IUnaryOperator)algebra).Transform(this);
}
else if (algebra is IBgp)
{
//Don't integerfer with other optimisers which have added custom BGP implementations
if (!(algebra is Bgp)) return algebra;
IBgp current = (IBgp)algebra;
if (current.PatternCount == 0)
{
return current;
}
else
{
ISparqlAlgebra result = new Bgp();
List<ITriplePattern> patterns = new List<ITriplePattern>();
List<ITriplePattern> ps = new List<ITriplePattern>(current.TriplePatterns.ToList());
for (int i = 0; i < current.PatternCount; i++)
{
//Can't split the BGP if there are Blank Nodes present
if (!ps[i].HasNoBlankVariables) return current;
if (!(ps[i] is TriplePattern))
{
//First ensure that if we've found any other Triple Patterns up to this point
//we dump this into a BGP and join with the result so far
if (patterns.Count > 0)
{
result = Join.CreateJoin(result, new Bgp(patterns));
patterns.Clear();
}
//Then generate the appropriate strict algebra operator
if (ps[i] is FilterPattern)
{
result = new Filter(result, ((FilterPattern)ps[i]).Filter);
}
else if (ps[i] is BindPattern)
{
BindPattern bind = (BindPattern)ps[i];
result = new Extend(result, bind.AssignExpression, bind.VariableName);
}
else if (ps[i] is LetPattern)
{
LetPattern let = (LetPattern)ps[i];
result = new Extend(result, let.AssignExpression, let.VariableName);
}
else if (ps[i] is SubQueryPattern)
{
SubQueryPattern sq = (SubQueryPattern)ps[i];
result = Join.CreateJoin(result, new SubQuery(sq.SubQuery));
}
else if (ps[i] is PropertyPathPattern)
{
PropertyPathPattern pp = (PropertyPathPattern)ps[i];
result = Join.CreateJoin(result, new PropertyPath(pp.Subject, pp.Path, pp.Object));
}
}
else
{
patterns.Add(ps[i]);
}
}
if (patterns.Count == current.PatternCount)
{
//If count of remaining patterns same as original pattern count there was no optimisation
//to do so return as is
return current;
}
else if (patterns.Count > 0)
{
//If any patterns left at end join as a BGP with result so far
result = Join.CreateJoin(result, new Bgp(patterns));
return result;
}
else
{
return result;
}
}
}
else if (algebra is ITerminalOperator)
{
return algebra;
}
else
{
return algebra;
}
}
/// <summary>
/// Optimises BGPs in the Algebra to use Filter() and Extend() rather than the embedded FILTER and BIND
/// </summary>
/// <param name="algebra">Algebra to optimise</param>
/// <returns></returns>
public ISparqlAlgebra Optimise(ISparqlAlgebra algebra)
{
if (algebra is IAbstractJoin)
{
return ((IAbstractJoin)algebra).Transform(this);
}
else if (algebra is IUnaryOperator)
{
return ((IUnaryOperator)algebra).Transform(this);
}
else if (algebra is IBgp)
{
IBgp current = (IBgp)algebra;
if (current.PatternCount == 0)
{
return current;
}
else
{
ISparqlAlgebra result = new Bgp();
List<ITriplePattern> patterns = new List<ITriplePattern>();
List<ITriplePattern> ps = new List<ITriplePattern>(current.TriplePatterns.ToList());
for (int i = 0; i < current.PatternCount; i++)
{
if (ps[i] is FilterPattern || ps[i] is BindPattern)
{
//First ensure that if we've found any other Triple Patterns up to this point
//we dump this into a BGP and join with the result so far
if (patterns.Count > 0)
{
result = Join.CreateJoin(result, new Bgp(patterns));
patterns.Clear();
}
if (ps[i] is FilterPattern)
{
result = new Filter(result, ((FilterPattern)ps[i]).Filter);
}
else
{
BindPattern bind = (BindPattern)ps[i];
result = new Extend(result, bind.AssignExpression, bind.VariableName);
}
}
else
{
patterns.Add(ps[i]);
}
}
if (patterns.Count == current.PatternCount)
{
//If count of remaining patterns same as original pattern count there was no optimisation
//to do so return as is
return current;
}
else if (patterns.Count > 0)
{
//If any patterns left at end join as a BGP with result so far
result = Join.CreateJoin(result, new Bgp(patterns));
return result;
}
else
{
return result;
}
}
}
else if (algebra is ITerminalOperator)
{
return algebra;
}
else
{
return algebra;
}
}
/// <summary>
/// Runs the optimisation against a Filter algebra
/// </summary>
/// <param name="filter">The Filter algebra to optimise</param>
/// <param name="optimisedAlgebra">Receives the optimised algebra if optimisation was performed or the input algebra otherwise</param>
/// <returns>True if an optimisation was performed, false otherwise</returns>
/// <remarks>
/// <para>This implementation currently handles the simple case of a Filter applied to a BGP where the filter
/// expression is either a single EqualsExpression or SameTermExpression or an AndExpression containing one or more
/// EqualsExpression or SameTermExpression arguments. The implementation ensures that the replaced variable is still
/// available to the outer algebra by inserting a BindPattern into the BGP. If the filter expression is a single
/// EqualsExpression or SameTermExpression, the optimiser will also strip this out of the algebra, but with an
/// AndExpression it will leave the full filter expression untouched.</para>
/// <para>The implementation will replace only URI and PlainLiteral types</para>
/// TODO: It should be possible to remove EqualsExpression and SameTermExpression instances from the AndExpression arguments and then either strip it out (if it has no remaining arguments), or optimise it to a single expression (if it has one remaining argument)
/// </remarks>
private bool OptimiseFilter(IFilter filter, out ISparqlAlgebra optimisedAlgebra)
{
if (!(filter.InnerAlgebra is Bgp))
{
// Need a BGP to be able to insert BindPatterns for replaced variables
optimisedAlgebra = filter;
return false;
}
var filterExpression = filter.SparqlFilter.Expression;
var replacementTerms = new Dictionary<string, INode>();
string var;
INode term;
bool equals;
// Currently only handle the simple filter cases of a single identity expression
// or an AND of expressions
if (IsIdentityExpression(filterExpression, out var, out term, out equals))
{
if (CanOptimize(term))
{
replacementTerms.Add(var, term);
}
}
else if (filterExpression is AndExpression)
{
foreach (var arg in filterExpression.Arguments)
{
if (IsIdentityExpression(arg, out var, out term, out equals) && CanOptimize(term))
{
replacementTerms.Add(var, term);
}
else
{
foreach (var variable in arg.Variables) {
// Cannot guarantee that the argument doesn't imply some other possible binding for the variables
replacementTerms.Remove(variable);
}
}
}
}
if (replacementTerms.Any())
{
var optimisedInner = filter.InnerAlgebra as Bgp;
foreach (var replacementEntry in replacementTerms)
{
try
{
// Replace the variable with a constant term wherever it appears and then add a Bind pattern
// to ensure that the variable is bound for use in the outer algebra
var t = new VariableSubstitutionTransformer(replacementEntry.Key, replacementEntry.Value);
optimisedInner = t.Optimise(optimisedInner) as Bgp;
optimisedInner =
new Bgp(
optimisedInner.TriplePatterns.Concat(new[]
{new BindPattern(replacementEntry.Key, new ConstantTerm(replacementEntry.Value))}));
}
catch (RdfQueryException)
{
// Could not perform this replacement.
}
}
if (filterExpression is AndExpression)
{
// Keep the filter as it may contain other necessary expressions
// TODO: Could try to remove the identity expressions here ?
optimisedAlgebra = new Filter(optimisedInner, filter.SparqlFilter);
}
else
{
// Can optimise away the filter entirely
optimisedAlgebra = optimisedInner;
}
return true;
}
optimisedAlgebra = filter;
return false;
}
/// <summary>
/// Converts the algebra back into a Graph Pattern.
/// </summary>
/// <returns></returns>
public Patterns.GraphPattern ToGraphPattern()
{
ISparqlAlgebra algebra = new Filter(new Join(_lhs, _rhs), new UnaryExpressionFilter(_expr));
return(algebra.ToGraphPattern());
}
/// <summary>
/// Converts the algebra back into a query.
/// </summary>
/// <returns></returns>
public SparqlQuery ToQuery()
{
ISparqlAlgebra algebra = new Filter(new Join(_lhs, _rhs), new UnaryExpressionFilter(_expr));
return(algebra.ToQuery());
}