/**
Copyright (C) SYSTAP, LLC DBA Blazegraph 2006-2016. All rights reserved.
Contact:
SYSTAP, LLC DBA Blazegraph
2501 Calvert ST NW #106
Washington, DC 20008
licenses@blazegraph.com
This program is free software; you can redistribute it and/or modify
it under the terms of the GNU General Public License as published by
the Free Software Foundation; version 2 of the License.
This program is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
GNU General Public License for more details.
You should have received a copy of the GNU General Public License
along with this program; if not, write to the Free Software
Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
*/
/*
* Created on Nov 5, 2007
*/
package com.bigdata.rdf.rules;
import java.io.File;
import java.util.ArrayList;
import java.util.HashSet;
import java.util.List;
import java.util.Properties;
import java.util.Random;
import java.util.Set;
import org.apache.log4j.MDC;
import org.openrdf.model.URI;
import org.openrdf.model.impl.URIImpl;
import org.openrdf.model.vocabulary.OWL;
import org.openrdf.model.vocabulary.RDF;
import org.openrdf.model.vocabulary.RDFS;
import org.openrdf.rio.RDFFormat;
import com.bigdata.rdf.inf.TruthMaintenance;
import com.bigdata.rdf.internal.IV;
import com.bigdata.rdf.model.BigdataStatement;
import com.bigdata.rdf.model.BigdataURI;
import com.bigdata.rdf.model.BigdataValue;
import com.bigdata.rdf.model.BigdataValueFactory;
import com.bigdata.rdf.model.StatementEnum;
import com.bigdata.rdf.rio.StatementBuffer;
import com.bigdata.rdf.spo.ExplicitSPOFilter;
import com.bigdata.rdf.spo.ISPO;
import com.bigdata.rdf.spo.SPO;
import com.bigdata.rdf.spo.SPOComparator;
import com.bigdata.rdf.spo.SPOKeyOrder;
import com.bigdata.rdf.store.AbstractTripleStore;
import com.bigdata.rdf.store.DataLoader;
import com.bigdata.rdf.store.TempTripleStore;
import com.bigdata.rdf.store.AbstractTripleStore.Options;
import com.bigdata.rdf.store.DataLoader.ClosureEnum;
import com.bigdata.relation.accesspath.IAccessPath;
import com.bigdata.striterator.ChunkedArrayIterator;
import com.bigdata.striterator.IChunkedIterator;
import com.bigdata.striterator.IChunkedOrderedIterator;
/**
* Test suite for {@link TruthMaintenance}.
*
* @todo add a stress test where we assert random statements and then back out
* those assertions verifying that we recover the original closure?
*
* @todo run for both full and fast closure programs.
*
* @author <a href="mailto:thompsonbry@users.sourceforge.net">Bryan Thompson</a>
* @version $Id$
*/
public class TestTruthMaintenance extends AbstractInferenceEngineTestCase {
/**
*
*/
public TestTruthMaintenance() {
super();
}
/**
* @param name
*/
public TestTruthMaintenance(String name) {
super(name);
}
final private Random r = new Random();
/**
* Test for
* {@link TruthMaintenance#applyExistingStatements(AbstractTripleStore, AbstractTripleStore, IElementFilter filter)}.
*/
public void test_filter_01() {
TempTripleStore focusStore = null;
final AbstractTripleStore store = getStore();
try {
/*
* Setup some terms.
*/
final BigdataValueFactory f = store.getValueFactory();
final BigdataURI x = f.createURI("http://www.foo.org/x1");
final BigdataURI y = f.createURI("http://www.foo.org/y2");
final BigdataURI z = f.createURI("http://www.foo.org/z3");
store.addTerms(new BigdataValue[] { x, y, z });
final IV x1 = x.getIV();
final IV y2 = y.getIV();
final IV z3 = z.getIV();
/*
* Setup the database.
*/
{
final SPO[] a = new SPO[] {
new SPO(x1, y2, z3, StatementEnum.Inferred),
new SPO(y2, y2, z3, StatementEnum.Explicit)
};
store.addStatements(a, a.length);
assertTrue(store.hasStatement(x1, y2, z3));
assertTrue(store.getStatement(x1, y2, z3).isInferred());
assertTrue(store.hasStatement(y2, y2, z3));
assertTrue(store.getStatement(y2, y2, z3).isExplicit());
// nbefore = store.getStatementCount();
}
/*
* Setup a temporary store.
*/
{
final Properties properties = store.getProperties();
properties.setProperty(Options.LEXICON, "false");
focusStore = new TempTripleStore( properties );
// SPOAssertionBuffer buf = new SPOAssertionBuffer(focusStore, store,
// null/* filter */, 100/* capacity */, false/* justified */);
final SPO[] a = new SPO[] {
// should be applied to the database since already there as inferred.
new SPO(x1, y2, z3, StatementEnum.Explicit),
// should be applied to the database since already there as explicit.
new SPO(y2, y2, z3, StatementEnum.Explicit),
// should not be applied to the database since not there at all.
new SPO(z3, y2, z3, StatementEnum.Explicit),
};
/*
* add statement to the focusStore and do NOT use the focusStore
* lexicon (it does not exist) to assign sids to the statements.
*/
store.addStatements(focusStore, true/* copyOnly */,
new ChunkedArrayIterator<ISPO>(a.length, a, null/* keyOrder */),
null/* filter */);
assertTrue(focusStore.hasStatement(x1, y2, z3));
assertTrue(focusStore.getStatement(x1, y2, z3).isExplicit());
assertTrue(focusStore.hasStatement(y2, y2, z3));
assertTrue(focusStore.getStatement(y2, y2, z3).isExplicit());
assertTrue(focusStore.hasStatement(z3, y2, z3));
assertTrue(focusStore.getStatement(z3, y2, z3).isExplicit());
// assertEquals(nbefore + 1, focusStore.getStatementCount());
}
/*
* For each (explicit) statement in the focusStore that also exists
* in the database: (a) if the statement is not explicit in the
* database then mark it as explicit; and (b) remove the statement
* from the focusStore.
*/
final int nremoved = TruthMaintenance.applyExistingStatements(
focusStore, store, null/*filter*/);
// statement was pre-existing and was converted from inferred to explicit.
assertTrue(store.hasStatement(x1, y2, z3));
assertTrue(store.getStatement(x1, y2, z3).isExplicit());
// statement was pre-existing as "explicit" so no change.
assertTrue(store.hasStatement(y2, y2, z3));
assertTrue(store.getStatement(y2, y2, z3).isExplicit());
// assertEquals(nbefore, focusStore.getExactStatementCount());
assertEquals("#removed", 1, nremoved);
} finally {
if (focusStore != null)
focusStore.__tearDownUnitTest();
store.__tearDownUnitTest();
}
}
/**
* A simple test of {@link TruthMaintenance} in which some statements are
* asserted and their closure is computed and aspects of that closure are
* verified (this is based on rdfs11).
*/
public void test_assertAll_01() {
TempTripleStore tempStore = null;
final AbstractTripleStore store = getStore();
try {
final TruthMaintenance tm = new TruthMaintenance(store.getInferenceEngine());
final BigdataValueFactory f = store.getValueFactory();
final BigdataURI U = f.createURI("http://www.bigdata.com/U");
final BigdataURI V = f.createURI("http://www.bigdata.com/V");
final BigdataURI X = f.createURI("http://www.bigdata.com/X");
final BigdataURI rdfsSubClassOf = f.asValue(RDFS.SUBCLASSOF);
tempStore = tm.newTempTripleStore();
// buffer writes on the tempStore.
{
final StatementBuffer assertionBuffer = new StatementBuffer(
tempStore, store, 10/* capacity */, 10/*queueCapacity*/);
assertTrue(tempStore == assertionBuffer.getStatementStore());
assertionBuffer.add(U, rdfsSubClassOf, V);
assertionBuffer.add(V, rdfsSubClassOf, X);
// flush to the temp store.
assertionBuffer.flush();
}
if (log.isInfoEnabled())
log.info("\n\ntempStore:\n"
+ tempStore.dumpStore(store,
true, true, false, true));
// perform closure and write on the database.
tm.assertAll(tempStore);
if (log.isInfoEnabled())
log.info("\n\ndatabase:\n"
+ store.dumpStore(store, true, true, false, true));
// explicit.
assertTrue(store.hasStatement(U, rdfsSubClassOf, V));
assertTrue(store.hasStatement(V, rdfsSubClassOf, X));
// inferred.
assertTrue(store.hasStatement(U, rdfsSubClassOf, X));
} finally {
// Note: This will be torn down with the [store].
// if (tempStore != null)
// tempStore.__tearDownUnitTest();
store.__tearDownUnitTest();
}
}
/**
* A simple test of {@link TruthMaintenance} in which some statements are
* asserted, their closure is computed and aspects of that closure are
* verified, and then an explicit statement is removed and the closure is
* updated and we verify that an entailment known to depend on the remove
* statement has also been removed (this is based on rdfs11).
*
* @todo do a variant test where we remove more than one support at once in
* a case where the at least one of the statements entails the other
* and verify that both statements are removed (ie, verify that
* isGrounded is NOT accepting as grounds any support that is in the
* focusStore). This test could actually be done in
* {@link TestJustifications}.
*/
public void test_retractAll_01() {
final AbstractTripleStore store = getStore();
try {
final BigdataValueFactory f = store.getValueFactory();
final BigdataURI U = f.createURI("http://www.bigdata.com/U");
final BigdataURI V = f.createURI("http://www.bigdata.com/V");
final BigdataURI X = f.createURI("http://www.bigdata.com/X");
final BigdataURI rdfsSubClassOf = f.asValue(RDFS.SUBCLASSOF);
final InferenceEngine inf = store.getInferenceEngine();
final TruthMaintenance tm = new TruthMaintenance(inf);
// add some assertions and verify aspects of their closure.
{
final StatementBuffer assertionBuffer = new StatementBuffer(tm
.newTempTripleStore(), store, 100/* capacity */, 10/*queueCapacity*/);
assertionBuffer.add(U, rdfsSubClassOf, V);
assertionBuffer.add(V, rdfsSubClassOf, X);
// flush statements to the temp store.
assertionBuffer.flush();
// perform closure and write on the database.
tm.assertAll((TempTripleStore) assertionBuffer
.getStatementStore());
// explicit.
assertTrue(store.hasStatement(U, rdfsSubClassOf, V));
assertTrue(store.hasStatement(V, rdfsSubClassOf, X));
// inferred.
assertTrue(store.hasStatement(U, rdfsSubClassOf, X));
}
/*
* retract one of the explicit statements and update the closure.
*
* then verify that it is retracted statement is gone, that the
* entailed statement is gone, and that the other explicit statement
* was not touched.
*/
{
final StatementBuffer retractionBuffer = new StatementBuffer(tm
.newTempTripleStore(), store, 100/* capacity */, 10/*queueCapacity*/);
retractionBuffer.add(V, rdfsSubClassOf, X);
// flush buffer to temp store.
retractionBuffer.flush();
// update the closure.
tm.retractAll((TempTripleStore) retractionBuffer
.getStatementStore());
// explicit.
assertTrue(store.hasStatement(U, rdfsSubClassOf, V));
assertFalse(store.hasStatement(V, rdfsSubClassOf, X));
// inferred.
assertFalse(store.hasStatement(U, rdfsSubClassOf, X));
}
/*
* Add the retracted statement back in and verify that we get the
* entailment back.
*/
{
final StatementBuffer assertionBuffer = new StatementBuffer(tm
.newTempTripleStore(), store, 100/* capacity */, 10/*queueCapacity*/);
assertionBuffer.add(V, rdfsSubClassOf, X);
// flush to the temp store.
assertionBuffer.flush();
// update the closure.
tm.assertAll((TempTripleStore)assertionBuffer.getStatementStore());
// explicit.
assertTrue(store.hasStatement(U, rdfsSubClassOf, V));
assertTrue(store.hasStatement(V, rdfsSubClassOf, X));
// inferred.
assertTrue(store.hasStatement(U, rdfsSubClassOf, X));
}
/*
* Retract the entailment and verify that it is NOT removed from the
* database (removing an inference has no effect).
*/
{
final StatementBuffer retractionBuffer = new StatementBuffer(tm
.newTempTripleStore(), store, 100/* capacity */, 10/*queueCapacity*/);
retractionBuffer.add(U, rdfsSubClassOf, X);
// flush to the temp store.
retractionBuffer.flush();
// update the closure.
tm.retractAll((TempTripleStore)retractionBuffer.getStatementStore());
// explicit.
assertTrue(store.hasStatement(U, rdfsSubClassOf, V));
assertTrue(store.hasStatement(V, rdfsSubClassOf, X));
// inferred.
assertTrue(store.hasStatement(U, rdfsSubClassOf, X));
}
} finally {
store.__tearDownUnitTest();
}
}
/**
* A simple test of {@link TruthMaintenance} in which some statements are
* asserted, including one statement which is also produced as an inference,
* and their closure is computed and aspects of that closure are verified
* (this is based on rdfs11). After we verify the closure, we retract the
* explicit statement and then verify that the closure was updated such that
* the statement was downgraded to an inference.
*/
public void test_downgradeExplicitToInference() {
final AbstractTripleStore store = getStore();
try {
final TruthMaintenance tm = new TruthMaintenance(store.getInferenceEngine());
final BigdataValueFactory f = store.getValueFactory();
final BigdataURI U = f.createURI("http://www.bigdata.com/U");
final BigdataURI V = f.createURI("http://www.bigdata.com/V");
final BigdataURI X = f.createURI("http://www.bigdata.com/X");
final BigdataURI rdfsSubClassOf = f.asValue(RDFS.SUBCLASSOF);
{
// Note: new triple store on shared temporary store!
final TempTripleStore tempStore = tm.newTempTripleStore();
// buffer writes on the tempStore.
{
final StatementBuffer assertionBuffer = new StatementBuffer(
tempStore, store, 10/* capacity */, 10/*queueCapacity*/);
assertTrue(tempStore == assertionBuffer.getStatementStore());
assertionBuffer.add(U, rdfsSubClassOf, V);
assertionBuffer.add(V, rdfsSubClassOf, X);
/*
* Note: this statement is entailed by the other two, but we
* represent it explicitly as well in order to test the
* downgrade mechanism.
*/
assertionBuffer.add(U, rdfsSubClassOf, X);
// flush to the temp store.
assertionBuffer.flush();
}
if (log.isInfoEnabled())
log.info("\n\ntempStore:\n"
+ tempStore.dumpStore(store, true, true, false,
true));
if (log.isInfoEnabled())
log.info("Doing asserts.");
// perform closure and write on the database.
tm.assertAll(tempStore);
}
if (log.isInfoEnabled())
log.info("\n\ndatabase:\n"
+ store.dumpStore(store, true, true, false, true));
// verify statements exist.
assertTrue(store.hasStatement(U, rdfsSubClassOf, V));
assertTrue(store.hasStatement(V, rdfsSubClassOf, X));
assertTrue(store.hasStatement(U, rdfsSubClassOf, X));
// and verify their statement type.
assertEquals(StatementEnum.Explicit, store.getStatement(U,
rdfsSubClassOf, V).getStatementType());
assertEquals(StatementEnum.Explicit, store.getStatement(V,
rdfsSubClassOf, X).getStatementType());
assertEquals(StatementEnum.Explicit, store.getStatement(U,
rdfsSubClassOf, X).getStatementType());
// now retract
{
// Note: new triple store on shared temporary store!
final TempTripleStore tempStore = tm.newTempTripleStore();
// buffer writes on the tempStore.
{
final StatementBuffer retractionBuffer = new StatementBuffer(
tempStore, store, 10/* capacity */, 10/*queueCapacity*/);
assertTrue(tempStore == retractionBuffer
.getStatementStore());
/*
* Retract this statement. It is explicitly present in the
* DB. However, note that it is also inferred. Therefore, it
* MUST be downgraded to an inference.
*/
retractionBuffer.add(U, rdfsSubClassOf, X);
// flush to the temp store.
retractionBuffer.flush();
}
if (log.isInfoEnabled())
log.info("\n\ntempStore:\n"
+ tempStore.dumpStore(store, true, true, false,
true));
if (log.isInfoEnabled())
log.info("Doing retraction.");
// perform closure and write on the database.
tm.retractAll(tempStore);
}
if (log.isInfoEnabled())
log.info("\n\ndatabase:\n"
+ store.dumpStore(store, true, true, false, true));
// verify statements exist.
assertTrue(store.hasStatement(U, rdfsSubClassOf, V));
assertTrue(store.hasStatement(V, rdfsSubClassOf, X));
assertTrue(store.hasStatement(U, rdfsSubClassOf, X));
// and verify their statement type.
assertEquals(StatementEnum.Explicit, store.getStatement(U,
rdfsSubClassOf, V).getStatementType());
assertEquals(StatementEnum.Explicit, store.getStatement(V,
rdfsSubClassOf, X).getStatementType());
assertEquals(StatementEnum.Inferred, store.getStatement(U,
rdfsSubClassOf, X).getStatementType());
} finally {
store.__tearDownUnitTest();
}
}
/**
* Given three explicit statements:
*
* <pre>
* stmt a: #user #currentGraph #foo
*
* stmt b: #currentGraph rdfs:range #Graph
*
* stmt c: #foo rdf:type #Graph
* </pre>
*
* a+b implies c
* <p>
* Delete a and verify that c is NOT gone since it is an explicit statement.
*/
public void test_retractWhenStatementSupportsExplicitStatement() {
URI user = new URIImpl("http://www.bigdata.com/user");
URI currentGraph = new URIImpl("http://www.bigdata.com/currentGraph");
URI foo = new URIImpl("http://www.bigdata.com/foo");
URI graph = new URIImpl("http://www.bigdata.com/Graph");
URI rdftype = RDF.TYPE;
URI rdfsRange = RDFS.RANGE;
AbstractTripleStore store = getStore();
try {
final InferenceEngine inf = store.getInferenceEngine();
final TruthMaintenance tm = new TruthMaintenance(inf);
// add some assertions and verify aspects of their closure.
{
final StatementBuffer assertionBuffer = new StatementBuffer(tm
.newTempTripleStore(), store, 100/* capacity */, 10/*queueCapacity*/);
// stmt a
assertionBuffer.add(user, currentGraph, foo );
// stmt b
assertionBuffer.add(currentGraph, rdfsRange, graph );
// stmt c
assertionBuffer.add(foo, rdftype, graph );
// flush to the temp store.
assertionBuffer.flush();
// perform closure and write on the database.
tm.assertAll((TempTripleStore)assertionBuffer.getStatementStore());
// dump after closure.
if (log.isInfoEnabled())
log.info("\n" + store.dumpStore(true, true, false));
// explicit.
assertTrue(store.hasStatement(user, currentGraph, foo ));
assertTrue(store.hasStatement(currentGraph, rdfsRange, graph ));
assertTrue(store.hasStatement(foo, rdftype, graph));
// verify that stmt c is marked as explicit in the kb.
final BigdataStatement stmtC = (BigdataStatement) store
.getStatement(foo, rdftype, graph);
assertNotNull(stmtC);
assertEquals(StatementEnum.Explicit, stmtC.getStatementType());
}
/*
* retract stmt A and update the closure.
*
* then verify that it is retracted statement is gone and that the
* other explicit statements were not touched.
*/
{
final StatementBuffer retractionBuffer = new StatementBuffer(tm
.newTempTripleStore(), store, 100/* capacity */, 10/*queueCapacity*/);
retractionBuffer.add(user, currentGraph, foo);
// flush to the temp store.
retractionBuffer.flush();
// update the closure.
tm.retractAll( (TempTripleStore)retractionBuffer.getStatementStore());
// dump after re-closure.
if (log.isInfoEnabled())
log.info("\n" + store.dumpStore(true, true, false));
// test the kb.
assertFalse(store.hasStatement(user, currentGraph, foo));
assertTrue(store.hasStatement(currentGraph, rdfsRange, graph));
assertTrue(store.hasStatement(foo, rdftype, graph));
// verify that stmt c is marked as explicit in the kb.
final BigdataStatement stmtC = (BigdataStatement) store
.getStatement(foo, rdftype, graph);
assertNotNull(stmtC);
assertEquals(StatementEnum.Explicit, stmtC.getStatementType());
}
} finally {
store.__tearDownUnitTest();
}
}
/**
* This test demonstrates TM incorrectness (since fixed of course). I add
* two statements into store A, then remove one of them. Then I add the the
* statement that remain in store A into store B and compare the closure of
* the stores. They should be the same, right? Well, unfortunately they are
* not the same. Too many inferences were deleted from the first store
* during TM.
*/
public void test_closurecorrectness() {
final URI a = new URIImpl("http://www.bigdata.com/a");
final URI b = new URIImpl("http://www.bigdata.com/b");
final URI c = new URIImpl("http://www.bigdata.com/c");
// final URI d = new URIImpl("http://www.bigdata.com/d");
final URI sco = RDFS.SUBCLASSOF;
final AbstractTripleStore store = getStore();
try {
final TruthMaintenance tm = new TruthMaintenance(store
.getInferenceEngine());
// add two
{
final StatementBuffer assertionBuffer = new StatementBuffer(tm
.newTempTripleStore(), store, 100/* capacity */, 10/*queueCapacity*/);
assertionBuffer.add(a, sco, b );
assertionBuffer.add(b, sco, c );
// assertionBuffer.add(c, sco, d );
// write statements on the temp store.
assertionBuffer.flush();
// perform closure and write on the database.
tm.assertAll((TempTripleStore) assertionBuffer
.getStatementStore());
if (log.isInfoEnabled())
log.info("\ndump after closure:\n"
+ store.dumpStore(store, true, true, false, true));
}
// retract one
{
final StatementBuffer retractionBuffer = new StatementBuffer(tm
.newTempTripleStore(), store, 100/* capacity */, 10/*queueCapacity*/);
retractionBuffer.add(b, sco, c);
// write statements on the temp store.
retractionBuffer.flush();
// update the closure.
tm.retractAll((TempTripleStore)retractionBuffer.getStatementStore());
if (log.isInfoEnabled())
log.info("\ndump after retraction and re-closure:\n"
+ store.dumpStore(true, true, false));
}
/*
* Add statement(s) to the "control store" and compute its closure.
* This provides the basis for checking the result that we obtained
* above via retraction.
*/
{
final TempTripleStore controlStore = new TempTripleStore(store
.getProperties());
try {
// Note: maintains closure on the controlStore.
final TruthMaintenance tmControlStore = new TruthMaintenance(
controlStore.getInferenceEngine());
final StatementBuffer<?> assertionBuffer = new StatementBuffer(
tmControlStore.newTempTripleStore(), controlStore,
100/* capacity */, 10/*queueCapacity*/);
assertionBuffer.add(a, sco, b);
// assertionBuffer.add(c, sco, d );
// write statements on the controlStore.
assertionBuffer.flush();
// perform closure and write on the database.
tmControlStore.assertAll((TempTripleStore) assertionBuffer
.getStatementStore());
if (log.isInfoEnabled())
log.info("\ndump controlStore after closure:\n"
+ controlStore.dumpStore(true, true, false));
assertSameGraphs(controlStore, store);
} finally {
controlStore.__tearDownUnitTest();
}
}
} finally {
store.__tearDownUnitTest();
}
}
/**
* This test demonstrates an infinite loop in TM arising from owl:sameAs.
*/
public void test_infiniteloop() {
// if(true) fail("re-enable this test");
final URI a = new URIImpl("http://www.bigdata.com/a");
final URI b = new URIImpl("http://www.bigdata.com/b");
final URI entity = new URIImpl("http://www.bigdata.com/Entity");
final URI sameAs = OWL.SAMEAS;
// /*
// * Note: not using rdf:type to avoid entailments about (x rdf:type
// * Class) and (x rdfs:subClassOf y) that are not required by this test.
// */
// URI rdfType = new URIImpl("http://www.bigdata.com/type");
final URI rdfType = RDF.TYPE;
final AbstractTripleStore store = getStore();
try {
final InferenceEngine inf = store.getInferenceEngine();
final TruthMaintenance tm = new TruthMaintenance(inf);
// add some assertions and verify aspects of their closure.
{
final StatementBuffer assertionBuffer = new StatementBuffer(tm
.newTempTripleStore(), store, 100/* capacity */, 10/*queueCapacity*/);
// stmt a
assertionBuffer.add(a, rdfType, entity );
// assertionBuffer.add(a, x, y );
// stmt b
assertionBuffer.add(b, rdfType, entity );
// assert the sameas
assertionBuffer.add(a, sameAs, b );
// flush statements to the tempStore.
assertionBuffer.flush();
// perform closure and write on the database.
tm.assertAll((TempTripleStore) assertionBuffer
.getStatementStore());
// dump after closure.
if (log.isInfoEnabled())
log.info("\ndump after closure:\n"
+ store.dumpStore(store, true, true, false, true));
}
/*
* retract stmt A and update the closure.
*
* then verify that the retracted statement is gone and that the
* other explicit statements were not touched.
*/
{
final StatementBuffer retractionBuffer = new StatementBuffer(tm
.newTempTripleStore(), store, 100/* capacity */, 10/*queueCapacity*/);
// retract the sameas
retractionBuffer.add(a, sameAs, b);
// flush statements to the tempStore.
retractionBuffer.flush();
// update the closure.
tm.retractAll((TempTripleStore) retractionBuffer
.getStatementStore());
// dump after re-closure.
if (log.isInfoEnabled())
log.info("\ndump after re-closure:\n"
+ store.dumpStore(store, true, true, false, true));
}
} finally {
store.__tearDownUnitTest();
}
}
/**
* This is a stress test for truth maintenance. It verifies that retraction
* and assertion are symmetric by randomly retracting and then asserting
* statements while using truth maintenance and verifying that the initial
* conditions are always recovered. It does NOT prove the correctness of the
* entailments, merely that retraction and assertion are symmetric.
*
* @todo use data files that we can bundle with the distribution.
*/
public void test_stress() {
// fail("enable test");
final String ontology = "src/test/resources/data/lehigh/univ-bench.owl";
final String resource =
"src/test/resources/data/lehigh/U1";
// "../rdf-data/alibaba_data.rdf",
// "../rdf-data/alibaba_schema.rdf"
// };
// final String[] baseURL = new String[] { "", ""
// };
//
// final RDFFormat[] format = new RDFFormat[] {
// RDFFormat.RDFXML,
// RDFFormat.RDFXML
// };
//
// for(String r : resource) {
//
// if(!new File(r).exists()) {
//
// System.err.println("Resource not found: "+r+", test="+getName()+" skipped.");
//
// return;
//
// }
//
// }
final Properties properties = getProperties();
/*
* Note: overrides properties to make sure that entailments are
* not computed on load.
*/
properties.setProperty(DataLoader.Options.CLOSURE,
ClosureEnum.None.toString());
TempTripleStore tmp = null;
final AbstractTripleStore store = getStore(properties);
try {
final DataLoader dataLoader = store.getDataLoader();
final String baseURI = new File(resource).toURI().toString();
dataLoader.loadData(ontology, baseURI, RDFFormat.RDFXML);
dataLoader.loadData(resource, baseURI, RDFFormat.RDFXML);
/*
* Compute the closure of the database.
*/
final InferenceEngine inf = store.getInferenceEngine();
inf.computeClosure(null/* focusStore */);
/*
* Make a copy of the graph (statements only) that will serve as
* ground truth.
*/
{
final Properties p = new Properties(properties);
// no lexicon.
p.setProperty(Options.LEXICON, "false");
tmp = new TempTripleStore(p);
store.copyStatements(tmp, null/* filter */, false/* copyJustifications */);
}
/*
* Start the stress tests.
*/
doStressTest(tmp, inf, 10/*ntrials*/, 1/*depth*/, 1/*nstmts*/);
doStressTest(tmp, inf, 7/*ntrials*/, 1/*depth*/, 5/*nstmts*/);
doStressTest(tmp, inf, 5/*ntrials*/, 5/*depth*/, 1/*nstmts*/);
doStressTest(tmp, inf, 3/*ntrials*/, 5/*depth*/, 5/*nstmts*/);
// // very stressful.
// doStressTest(tmp, inf, 100/*ntrials*/, 10/*depth*/, 20/*nstmts*/);
} catch(Exception ex) {
fail("Not expecting: "+ex, ex);
} finally {
if(tmp != null)
tmp.__tearDownUnitTest();
store.__tearDownUnitTest();
}
}
/**
* A stress test for truth maintenance using an arbitrary data set. The test
* scans the statement indices in some order, selecting N explicit statement
* to retract. It then retracts them, updates the closure, and then
* re-asserts them and verifies that original closure was restored.
* <p>
* Note: this test by itself does not guarentee that any entailments of
* those explicit statements were removed - we need to write other tests for
* that. repeat several times on the dataset, potentially doing multiple
* retractions before we back out of them.
*
* @param tmp
* Ground truth (the state that must be recovered in order for
* truth maintenance to be symmetric).
* @param inf
* The {@link InferenceEngine} to use in the test. This reads and
* writes on the database whose closure is being maintained.
* @param ntrials
* The #of times that we will run the test.
* @param D
* The recursive depth of the retractions. A depth of ONE (1)
* means that one set of N statements will be retracted, closure
* updated, and the re-asserted and closure updated and compared
* against ground truth (the initial conditions). When the depth
* is greater than ONE (1) we will recursively retract a set of N
* statements D times. The statements will be reasserted as we
* back out of the recursion and the graph compared with ground
* truth when we return from the top level of the recursion.
* @param N
* The #of explicit statements to be randomly selected and
* retracted on each recursive pass.
*/
public void doStressTest(TempTripleStore tmp, InferenceEngine inf,
int ntrials, int D, int N) {
AbstractTripleStore store = inf.database;
/*
* Verify our initial conditions agree.
*/
assertSameGraphs( tmp, store );
for (int trial = 0; trial < ntrials; trial++) {
/*
* Do recursion.
*/
MDC.put("trial", "trial="+trial);
retractAndAssert(inf,store,0/*depth*/,D,N);
/*
* Verify that the closure is correct after all that recursive
* mutation and restoration.
*/
assertSameGraphs(tmp, store);
MDC.remove("trial");
}
}
/**
* At each level of recursion up to N explicit statements are selected
* randomly from the database, retracted, and closure is updated. The method
* then calls itself recursively, thereby building up a series of updates to
* the graph. When the recursion bottoms out, the retracted statements are
* asserted and closure is updated. This continues as we back out of the
* recursion until the graph SHOULD contain the identical RDF model.
*
* @param inf
* Used to update the closure.
* @param db
* The database.
* @param depth
* The current depth of recursion (ZERO on the first call).
* @param D
* The maximum depth of recursion (depth will always be strictly
* less than D).
* @param N
* The #of explicit statements to randomly select at each level
* of recursion for retraction from the database.
*/
private void retractAndAssert(InferenceEngine inf, AbstractTripleStore db,
int depth, final int D, final int N) {
assert depth >= 0;
assert depth < D;
/*
* Select N explicit statements at random.
*/
SPO[] stmts = selectRandomExplicitStatements(db, N);
log.info("Selected "+stmts.length+" statements at random: depth="+depth);
final TruthMaintenance tm = new TruthMaintenance(inf);
/*
* Retract those statements and update the closure of the database.
*/
{
for(SPO tmp : stmts) {
log.info("Retracting: "+tmp.toString(db));
}
final TempTripleStore tempStore = tm.newTempTripleStore();
db.addStatements(tempStore, true/* copyOnly */,
new ChunkedArrayIterator<ISPO>(stmts.length, stmts,
null/* keyOrder */), null/* filter */);
log.info("Retracting: n="+stmts.length+", depth="+depth);
// note: an upper bound when using isolated indices.
final long before = db.getStatementCount();
tm.retractAll(tempStore);
final long after = db.getStatementCount();
final long delta = after - before;
log.info("Retraction: before="+before+", after="+after+", delta="+delta);
}
if (depth + 1 < D) {
retractAndAssert(inf, db, depth+1, D, N);
}
/*
* Assert those statements and update the closure of the database.
*/
{
for(SPO tmp : stmts) {
log.info("Asserting: "+tmp.toString(db));
}
final TempTripleStore tempStore = tm.newTempTripleStore();
db.addStatements(tempStore, true/* copyOnly */,
new ChunkedArrayIterator<ISPO>(stmts.length, stmts,
null/*keyOrder*/), null/*filter*/);
log.info("Asserting: n=" + stmts.length + ", depth=" + depth);
// note: an upper bound when using isolated indices.
final long before = db.getStatementCount();
tm.assertAll(tempStore);
final long after = db.getStatementCount();
final long delta = after - before;
log.info("Assertion: before="+before+", after="+after+", delta="+delta);
}
}
/**
* Select N explicit statements from the graph at random.
*
* @param db
* The graph.
* @param N
* The #of statements to select.
*
* @return Up to N distinct explicit statements selected from the graph.
*/
public SPO[] selectRandomExplicitStatements(AbstractTripleStore db, int N) {
/*
* Count the #of distinct subjects in the graph.
*/
final int nsubjects;
{
final IChunkedIterator<IV> termIds = db.getSPORelation()
.distinctTermScan(SPOKeyOrder.SPO);
try {
int n = 0;
while (termIds.hasNext()) {
termIds.next();
n++;
}
nsubjects = n;
} finally {
termIds.close();
}
}
if (log.isInfoEnabled())
log.info("There are " + nsubjects + " distinct subjects");
/*
* Choose N distinct subjects from the graph at random.
*/
final Set<IV> subjects = new HashSet<IV>(N);
for (int i = 0; i < nsubjects && subjects.size() < N; i++) {
final IChunkedIterator<IV> termIds = db.getSPORelation()
.distinctTermScan(SPOKeyOrder.SPO);
try {
// choose subject at random.
int index = r.nextInt(nsubjects);
IV s = NULL;
for (int j = 0; termIds.hasNext() && j < index; j++) {
s = termIds.next();
}
subjects.add(s);
} finally {
termIds.close();
}
}
if (log.isInfoEnabled())
log.info("Selected " + subjects.size() + " distinct subjects: "
+ subjects);
/*
* Choose one explicit statement at random for each distinct subject.
*
* Note: It is possible that some subjects will not have any explicit
* statements, in which case we will select fewer than N statements.
*/
List<ISPO> stmts = new ArrayList<ISPO>(N);
for( IV s : subjects ) {
final IAccessPath<ISPO> accessPath = db.getAccessPath(s, NULL, NULL,
ExplicitSPOFilter.INSTANCE);
final IChunkedOrderedIterator<ISPO> itr = accessPath.iterator();
try {
if (!itr.hasNext())
continue;
// a chunk of explicit statements for that subject.
final ISPO[] chunk = itr.nextChunk();
// statement randomly choosen from that chunk.
final ISPO tmp = chunk[r.nextInt(chunk.length)];
if (log.isInfoEnabled())
log.info("Selected at random: " + tmp.toString(db));
stmts.add(tmp);
} finally {
itr.close();
}
}
if (log.isInfoEnabled())
log.info("Selected " + stmts.size() + " distinct statements: "
+ stmts);
return stmts.toArray(new SPO[stmts.size()]);
}
/**
* This is a specialized test for equality in the graphs that simply compare
* scans on the SPO index.
* <p>
* Pre-condition: The term identifiers for the graphs MUST be consistently
* assigned since the statements are not being materialized as RDF
* {@link org.openrdf.model.Value} objects.
*
* @param expected
* A copy of the statements made after the data set was loaded
* and its closure computed and before we began to retract and
* assert stuff.
*
* @param actual
* Note that this is used by both graphs to resolve the term
* identifiers.
*/
protected void assertSameGraphs(TempTripleStore expected,
AbstractTripleStore actual) {
// For the truly paranoid.
// assertStatementIndicesConsistent(expected);
//
// assertStatementIndicesConsistent(actual);
/*
* Note: You can not directly compare statement counts when using
* isolatable indices since they are upper bounds - not exact counts.
*/
// if (expected.getStatementCount() != actual.getStatementCount()) {
//
// log.warn("statementCount: expected=" + expected.getStatementCount()
// + ", but actual=" + actual.getStatementCount());
//
// }
final IChunkedOrderedIterator<ISPO> itre = expected.getAccessPath(
SPOKeyOrder.SPO).iterator();
final IChunkedOrderedIterator<ISPO> itra = actual.getAccessPath(
SPOKeyOrder.SPO).iterator();
// int i = 0;
int nerrs = 0;
int maxerrs = 10;
int nexpected = 0;
int nactual = 0;
try {
while (itre.hasNext()) {
if(!itra.hasNext()) {
fail("Actual iterator exhausted before expected: nexpected="
+ nexpected
+ ", nactual="
+ nactual
+ ", remaining=" + toString(itre, 10, expected));
}
SPO expectedSPO = (SPO)itre.next(); nexpected++;
SPO actualSPO = (SPO)itra.next(); nactual++;
if (!expectedSPO.equals(actualSPO)) {
while (SPOComparator.INSTANCE.compare(actualSPO, expectedSPO) < 0) {
log.warn("Not expecting: " + actualSPO.toString(actual));
if(!itra.hasNext()) break;
actualSPO = (SPO)itra.next(); nactual++;
if(nerrs++==maxerrs) fail("Too many errors");
}
while (SPOComparator.INSTANCE.compare(expectedSPO, actualSPO) < 0) {
log.warn("Expecting: " + expectedSPO.toString(actual));
if(!itre.hasNext()) break;
expectedSPO = (SPO)itre.next(); nexpected++;
if(nerrs++==maxerrs) fail("Too many errors");
}
}
// i++;
}
assertFalse("Actual iterator will visit more than expected", itra.hasNext());
} finally {
itre.close();
itra.close();
}
/*
* Note: This compares the #of statements actually visited by the two
* iterators rather than comparing getStatementCount(), which is an
* upper bound based on rangeCount().
*/
assertEquals("statementCount", nexpected, nactual );
}
/**
* Consumes up to max elements from the iterator and returns a
* {@link String} representation of those elements. This is used to show the
* additional elements that would be visited by an iterator when the other
* iterator is exhausted.
*
* @param itr
* The iterator.
* @param max
* The maximum #of elements to visit.
* @param db
* Used to resolve term identifiers to RDF values.
*
* @return The string representation of the visited elements.
*/
private String toString(IChunkedOrderedIterator<ISPO> itr, int max, AbstractTripleStore db) {
StringBuilder sb = new StringBuilder();
int n = 0;
while (itr.hasNext() && n < max) {
if (n > 0)
sb.append(", ");
sb.append(itr.next().toString(db));
}
return "{" + sb.toString() + "}";
}
}