Release Notes ============= ==== Jena 2.6.5 ** Java 6 is now required for running Jena. ** The deprecate RDB subsystem has been removed. Please use TDB or SDB. ==== Jena 2.6.4 + Maintenace release ==== Jena 2.6.3 GraphBase: o Added limits to the number of triples thatr GraphBase.toString() will serialise. Model/Resource API: o Took advantage of covariant return types to allow Resource.inModel() and Property.inModel() to have Resource resp. Property return types, allowing casts of calls to inModel() to (mostly) be eliminated. o As per request on mailing list, ModelGraphInterface now has a method `Resource wrapAsResource(Node)` [name changed to fit better] which wraps a node as a Resource and throws if the node is a Literal. o As per request on mailing list, added Resource method x.getPropertyResourceValue( Property p ) which answers some R such that x p R if any exists, otherwise null o As per request on mailing list, there's now an OWL2 vocabulary class with constants for the terms of OWL2. NOTE: there is no OWL2 ontology support present or implied by this vocabulary class. o getModel() has been pulled up from Resource to RDFNode. This means that Literal now supports getModel() and inModel(). o RDFNode has .asLiteral() and .asResource() methods, which deliver the Literal or Resource respectively if the RDF Node has that type, and otherwise throw a *RequiredException. Misc: o Patched AnonId using UID when running in environments like GAE o Fixed problem with comparison on XSDDurations with fractional seconds Reasoners: o added makeSkolem builtin o fixed deadlock problem with hybrid rule systems o changed interpretation of Functors so they are neither literals nor objects, this allows validation of nested reasoners to cope with the introduction of functors representing inferred individuals o improved RETE initialization so that fewer triples are injected into the rule network in cases such as the RDFS no-resource rules Ontology API: o Fixed NPE bug when an ObjectProperty was not the inverseOf another property Schemagen: o Internal refactoring, which should not affect the external behaviour but should make it easier to extend the schemagen class o Individuals in an included namespace now appear in the output, even if their rdf:type is not included. ==== Jena 2.6.2 + Resource.addLiteral() now has an overload that takes a Literal object and uses it as the statement object unaltered. (Previously such a Literal would be treated as an Object and converted to a String literal, oops). Similarly model.addLiteral() has a Literal overloading. + N3 pretty writer now write Turtle (no use of "=" for owl:sameAs) + Maven-style codebase layout. + Update to iri.jar (version 0.7). + Change the way URIs are split into namespace and local part so as to not split %-encoded sequences. Reasoners: o Fixed problems with pure forward rules generating functor values (fixed looping behaviour, fixed selective filtering of functors). o Generalized rule reasoner to permit triples in subject position at Graph level. Returned InfGraph and deductions graphs are safe but getRawGraph on deductions graph will return generalized graph. o Fixed handling of unbound groups in regex. Core API: o Fixed bug in comparison of dateTimes with fractional seconds and with round-tripping from calendars ARQ: removed from Jena development The following jars have been removed from the lib/ directory - they are only used by ARQ: arq.jar, arq-extra.jar, lucene-core-2.3.1.jar, json.jar, stax-api-1.0.jar, wstx-asl-3.0.0.jar (Jena 2.6.1 is an internal tag, not a release) ==== Jena 2.6.0 ** Codebase converted to Java 5 o Convert to use SLF4J for logging facade o Exceptions AlreadyExistsException, DoesNotExistException and RulesetNotFoundException no longer inherit via RDFRDBException Jar changes: o Removed antlr-2.7.5.jar (no longer used) o Removed concurrent.jar (no longer used) o Removed xml-apis.jar (no longer used) o Removed commons-logging-1.1.1.jar o Add slf4j-api-1.5.6.jar, slf4j-log4j12.jar o junit.jar upgraded to junit-4.5.jar Deprecation clearance. Methods, classes and statics marked deprecated in previous releases have been largely removed. ObjectF deprecations The interface ObjectF and the methods that use it have been deprecated. They are hangovers from before the time RDF acquired typed literals; nobody should be using them any more. They will be removed in the next Jena release. Reasoners: o Extended validity checking for OWL configurations to include datatype range checking that can cope with user datatypes as done for RDFS. Places the culprit Triple as the ValidityReport.Report extension value. o Removed deprecated reasoners (RDFS1, RDFSExpt, OWLExpt) and associated support. o Fixed RDFS-SIMPLE to include properties being reflexive o Removed DIG 1.0 reasoner interface JenaModelSpec: o This long-deprecated vocabulary (which supported the old ModelSpec system) has been removed. ModelSpec was replaced by Assembler long ago. o some Maker methods that implicitly relied on this vocabulary (but which had not been deprecated) have also been removed. This affects GraphMaker, SimpleGraphMaker, BaseGraphMaker, and ModelMakerImpl: the methods used were getMakerClass, augmentDescription, addDescription, and getDescription. These methods were only relevant to uses of the old ModelSpec system, which was removed in previous releases. N3: o N3 parser is now the Turtle parser. The dialect of N3 that Jena supported (which is quite old) has some idioms that are not legal Turtle although they are nowadays not correct N3 either. o Turtle writer now speeded up when many namespaces in use. Ontology API: o Removed support for no-longer-supported OWL syntax checker. o Removed previously-deprecated namespace management methods from OntDocumentManager. Namespaces should be managed via the PrefixMapping API. o Unified code for checking individuals in OntResource.isIndividual() and OntModel.listIndividuals(), and updated the check to account for some user-reported edge cases. Datatypes: o Added a getTypeByClass call to TypeMapper and registered additional primitive types based on suggestion and patch supplied by Thorsten Moeller. ==== Jena 2.5.7 Database layer: o Compatibility fixes for PostgreSQL 8.3 (no schema changes) o Compatibility fixes for MS SQL Server 2008 (no schema changes) ReasonerFactoryAssembler o Now takes note of ja:schema properties & binds such a schema to the reasoners that the factory produces. Reasoners: o Modified @include processing and Rule.rulesFromURL to support FileManager redirects. o Fixed problem withLPTopGoalIterator which caused one extra level of lookahead in some circumstances. ==== Jena 2.5.6 Typed literals: o Fixed bug in retrieval of decimal values with trailing fractional zeros by canonicalizing values after parsing and improving canonicalization. ==== Jena 2.5.5 Jar changes: o lucene-core has been upgraded from lucene-core-2.0.0 to lucene-2.2.0.jar. o logging-commons-1.1.jar upgraded to logging-commons-1.1.1.jar Deprecated Method Removal Frenzy o A bunch of methods in Resource, Model, ModelCon, and Statement have for a while now been marked as deprecated. These methods took literal values for the object of a statement and used the old treat-it-as-a-string approach rather than the newer typed-literal approach. o These deprecated methods have been removed. However, to allow migrators an easier life and to avoid requiring many many calls of createTypedLiteral (which needs a model, not always conveniently to hand), typed-literal versions of these methods have been added to the interfaces. o Model has grown: * addLiteral( Resource, Property, Various ) * listResourcesWithProperty( Property, Various ) * createLiteralStatement( Resource, Property, Various ) * listLiteralStatements( Resource, Property, Various ) * listResourcesWithProperty( Property, Various ) o Resource has grown: * addLiteral( Property, Various ) * hasLiteral( Property, Various ) o Statement has grown: * changeObject( Various ) where `Various` is one of float, double, long,boolean, char, short, int, or object [short & int may be missing in some cases]. o As signalled in the Jena 2.5 release, the dedicated DAML API has now been removed. Users who still process DAML+OIL ontologies should switch to using the generic OntModel with the DAML profile. PrefixMapping loses deprecated usePrefix o The deprecated PrefixMapping method `usePrefix` has been removed. (Use `shortForm` instead.) ModelSpec o ModelSpec, which has been deprecated for a while, has been removed, along with ModelSpecImpl. OntModelSpec, which used to implement ModelSpec (via OntModelSpecObsolete, also removed), doesn't any more. RDF/XML-ABBREV o Aesthetic improvement changes most RDF/XML-ABBREV output. The treatment of URI nodes and blank nodes has been made more uniform. Please give feedback if for your data this is inappropriate. Schemagen o Member variables (i.e. for classes, properties and individuals) are now lexically sorted to create more stability in change-sets when generated Java classes are placed under source code control. o Now tries to guess the namespace of an ontology file in the absence of specific hints (hints such as: an option, prefix binding for the empty prefix, or an owl:Ontology or daml:Ontology resource). The algorithm is to look for the most prevalent URI that is not owl:, rdf: etc. o Added a new option, --includeSource, which causes schemagen to include the source code of the input ontology into the generated Java object. Ontology API updates o Added methods to Individual to allow testing and manipulation of the classes to which an individual belongs. o Added a method to OntProperty to list the Restrictions that mention the property o Clarified the contract of OntModel.listOntProperties, and added a new method OntModel.listAllOntProperties that is more robust in the absence of a reasoner, but potentially slower. o Added a utility class of basic ontology manipulation tools, com.hp.hpl.jena.ontology.OntTools. So far, this includes: * compute the lowest common ancestor of two classes * compute minimum path between two nodes * compute the named roots of a class hierarchy Reasoners: o Changed the equality contract of Rule objects to include the rule name. o Extended OWLMicro to include support for hasValue to match documentation. o Fixed problem with OntModel.getDeductionsModel not always triggering prepare. o Fixed problem with reasoner configuration when using typed literals Change of Behaviour in getByte/Short/Int on typed literals o The methods getByte(), getShort(), and getInt(), when applied to typed literals whose value is out of range, delivered a truncated value [as opposed to those methods on plain literals, which throw an exception]. This was deemed to be an error, and those methods will now throw IllegalArgumentExceptions. o These methods may be deprecated & removed in later releases. Assemblers o When the Assembler system (ie AssemblerGroup et al) load a class, that class is loaded before any of the implementsWith of that group are consulted. Thus the loaded class can update existing groups inplementsWith tables. o Because such an update need not be done to the global group Assembler.general, the group loading the class must be available, so using a static initialiser isn't sufficient. o Instead, if the class has a (static) method called `whenRequiredByAssembler` with an AssemblerGroup argument, that method is called passing in the loading AssemblerGroup. o Users are discouraged from using static initialisers which update Assembler.general, since this will not work as they might have expected when other assembler groups are being used. o OntModelSpecs have acquired an optional likeBuiltinSpec property for defining the "base" on which an OntModelSpec can be built by specifying other properties. In particular, this allows OntModelSpecs /with no reasoning/ to be constructed, which was a mysterious absence in earlier assemblers. Typed literals: o Fixed bug in equality checking of instances of xsd:decimal, and creation of xsd:decimal or xsd:integer instances from BigDecimal or BigInteger instances Reasoners: o Changed the equality contract of Rule objects to include the rule name. ==== Jena 2.5.4 Graph o Graph has grown a new method, getStatisticsHandler o null is an acceptable result o Graphs that extend GraphBase have this defined for them o It is intended for the ARQ optimiser optimising in-memory graphs Assemblers o ContentAssembler now respects the fileManager property for external content. o withFullModel now does subclass closure (as it should have done) o the code for findSpecificType is now better N3 & Turtle Writer o Correct writing nested bnodes with no properties to make strict Turtle. o No longer automatically add default namespace prefix. Reasoners o Added rebindAll() to FBRuleInfGraph o Extended OWLMini/full to include validation of maxCardinality(1) over individuals o The (forward) deductions model is now preserved across rebind() operations so that listeners on deductions models are safe o added now(?x) builtin to return the current xsd:dateTime Misc o Extended ResourceFactory.createTypedLiteral to handle Calendar objects ==== Jena 2.5.3 Statement.Util o The interface Statement has grown a Util class containing the three constants getSubject, getPredicate, and getObject, which are Map1 objects with the obvious meanings. Assembler modification-and-fix o There was a bug with the behaviour of OntModelSpec assemblers. The "feature" that if a ja:OntModelSpec had no properties it was treated as an OntModelSpec constant name did not work, since the generalAssembler obligingly inferred properties for it. The behavior has changed (I hope for the better ...). Instead, an OntModelSpec is built by modifying the properties of [a copy of] a default spec, which is normally OWL_MEM_RDFS_INF. However, if the root resource is ja:, then that constant provides the default. This gives the intended behaviour of "I specified no properties" and additionally allows specs to be specified as tweaks-to-a-constant. ContentAssemblers can now be constructed with default FileManager arguments. External content is loaded using that FileManager. The object of an externalContent statement can be a string literal or a URI. AssemblerGroups (which includes the instance Assembler.general) now support a `copy` method which makes a new group initialised with the same mapping. ReasonerFactoryAssembler now respects JA:reasonerClass properties, which specify the factory by giving its class name. ModelSpec and family REMOVED o the long-obsolete and recently-deprecated ModelSpec has been removed, along with its related tests, descendants, registry, and references. OntModelSpec remains (but is no longer a ModelSpec), with the old ModelSpec-oriented constructors deprecated and new Assembler statics introduced. OntModelSpec now extends OntModelSpecObsolete, which holds (almost) all of the OntModelSpec's deprecated ModelSpec-oriented machinery. OMSO will vanish after the next Jena release, when the remaining ModelSpec code will evaporate. Deprecated XSD.NS in favour of XSD.getURI() which reports namespace with a trailing # character. Reasoners: o Extended rule syntax to accept URIs wrapped in <..> and file: uris Typed literals o Fixed problem which prevented Calendars being used to instantiate XSDDates rather than full XSDDateTimes IRI o Fixed conformance problem with mailto: o Rewrote relativize code, fixing a #frag bug. ==== Jena 2.5.2 RDQL support now in ARQ o Removed old RDQL engine Migrate to SPARQL, use ARQ instead and package com.hp.hpl.jena.query http://jena.sf.net/ARQ/ -- ARQ also supports RDQL as a legacy language RDFException removed o Removed RDFException. It was obsolete (and recently deprecated). It's subclasses are now subclasses of JenaException. rdfcat tool o The handling of command line arguments to rdfcat has changed slightly. The input format (N3, RDF/XML, n-triple) argument, -n, -x and -t respectively, now *only* affect the following argument. In previous versions, the format was 'sticky' in that it persisted until another input format argument was supplied. To set the default input format, a new parameter -in has been added. See the rdfcat Javadoc for full details. ==== Jena 2.5.1 Fix for test.bat not setting the classpath correctly. Fix for PropertyNotFoundException from OntResourceImpl.getPropertyValue, not a null return. ==== Jena 2.5 ModelJEB is no longer supported o The ModelJEB BDB storage for graphs is no longer supported. (It has been unofficially unsupported for quite a while without complaint.) It's job -- cheap-and-cheerful persistent models -- has been taken over by HSQLDB. RDFException has been deprecated o It has been obsolete [because of the JenaException revisions] for yonks. No-one should be using it (other than some internal uses to be excised.) It will be removed post-2.5. Single-Element Unions [hence, OntModels] o If a union graph contains a single element, both find and queryHandler delegate directly to that single element. This means that find does not need to remember all the found elements to discard duplicates, and query gets to use any specialised code for that single graph (rather than the general nested-finds that a full union needs). In particular, if that single graph is a database graph, it will be able to use fastpath. A specific use of this is an OntModel with no imports in a database. (If it has imports, then this optimisation cannot apply.) In case something breaks as a result, the optimisation can be disabled by setting the system property jena.union.optimise to anything other than "yes". Reification with Inference Models o A partial fix to a reported bug with reification on inference models. Previously the built-in inference models had the reifier of their base model, and hence deduced triples did not contribute to the ReifiedStatement API calls, so deduced reified statements were invisible. Now those inference models have their own reifier and the reified statements of an inference model are the reified statements of the base model and of the deductions model. Reified statements inferred by backwards inference are /not/ reported. Reified statements formed from fragments spread between the base model and the deductions model are /not/ reported. Implementing those reifications would be expensive both for the implementors and for any users. PropertyImpl.getOrdinal() [internal] o now computed on demand rather than on construction. Also log-on-too-big replaced by throw-an-exception. I doubt very very much that this will ever matter. There was no TestPropertyImpl code, but there is now. It should be extended to cover the "illegal URI detection", or perhaps we should remove that since it's RDF/XML-specific and checked by the XML output code anyway? Turtle support o New Turtle parser (uses javacc, not antlr). Accessed for files with extension ".ttl" and language names "TTL", "TURTLE", "Turtle" Database backends: o Changed MS SQL Server driver to use NVARCHAR instead of VARCHAR, to fix i18n support. o Fixed i18n bug in Oracle backend and also switched to NVARCHAR. o No requirement to compile Jena for Oracle use. The Oracle driver now works in the standard Jena distribution. ModelFactory: o added methods to construct models using Assembler system. Updated ModelFactory howto to include them. Deprecated the ModelSpec methods. Removed them from ModelFactory howto. ModelSpec: o deprecated the ModelSpec class. Modified the ModelSpec docs to say it's deprecated. Assembler: o missing subclass declarations added to vocabulary. Bogus `domain` replaced by correct `range`. FileManagerAssembler now by default builds a FileManager with the standard locators. To test this, made the existing standard locator classes have appropriate .equals() and .hashCode() methods. o ConnectionAssembler now records subject resource in ConnectionDescription. ConnectionDescription falls over usefully if the URL or type is null when it's opened. RDF/XML output: o bug whereby a literal containing the sequence "]]>" generated illegal XML (because the "]]>" went through unchanged, and is illegal except as CDATA end) fixed by converting any ">" in element content into ">". o unencodable characters (any control character except \n, \t, \r, also \uFFFF and \uFFFE) now throw exceptions. We may allow a new parameter setting to switch this off later. o the entity-conversion code has been rewritten using Java regex's to avoid multiple conversions. o the new Writer property "showDoctypeDeclaration" is used by the RDF/XML writers to force inclusion of a !DOCTYPE header declaring as entities each of the prefix-mappings of the model. Attributes that start with the URIs of the mapping are written to use the entity invocation instead. RDF/XML Input: o When reading from a URL, content negotiation is now used prefering: RDF/XML, then XML, over other content types. o An over-eager and confusing warning was suppressed. o Behaviour on very large files has been modified. For such files, checks for illegal reuse of rdf:ID are not made after the first 10000 have been seen. A warning is issued about this change when reading a large file. IRI: Release synchronized with an IRI release (iri 0.3). Minor bug fixes. Untyped literal operations deprecated (char, boolean, long, float, double, Object) These operations do not work on typed literals. They converted their arguments to plain strings and parsed strings to retrive values. They are a ghost from pre-datatyping RDF. ** The forms on RDFNode, String, String+lang, and XML literals remain. ** Forms taking a typed literal or a lexical form and RDF datatype added (or already exist). Model: .add, .listStatements, .listSubjectsWithProperty Resource: .hasProperty, .addProperty JAR file changes: Added: arq-extra.jar Changed: wstx-asl-2.8.jar ==> wstx-asl-3.0.0.jar EnhNodes/UnsupportedPolymorhismException/QualifiedRestriction o Any EnhNode now automatically canAs any type already on its view ring if that view is still isValid. Previously it required finding a relevant Personality and Implementation and trying canWrap, which failed if the EnhNode had no EnhGraph, eg a Resource allocated by ResourceFactory. o This exposed problems with the QualifiedRestriction family of classes, which didn't override isValid. They do now. o Attempting to .as() an EnhNode with no EnhGraph no longer generates a NullPointerException, instead giving an explicit UnsupportedPolymorphism exception. o Attempting to canAs an EnhNode with no existing implementation of the required type and no EnhGraph returns false. OntResourceImpl o some internal modifications to simplify/correct the code: - corrected uses of .remove duing an iteration - replaced some loops by existing single Graph methods - some uses of getRequiredProperty + exception handling (probably hangovers from the old days when getProperty threw an exception) replaced by uses of getProperty and tests for null ResourceImpl o implementation of removeAll simplified in two ways: uses the Model removeAll method, and no longer tests for an InfModel, since the InfGraph implementation of removeAll works on the base model anyway so no special-case code required. Also avoids iffy (but currently safe) use of remove(StmtIterator) where the StmtIterator is active over the same Model. Statement o the method getWellFormed has been deprecated in favour of a new method hasWellFormedXML(), paralleling Literal's newish isWellFormedXML(). Model o Model has acquired a new method isClosed(), true iff the model has been .close()d o Graph ditto o The built-in Graph and Model implementations have been modified to implement this method. Typed literals o Fixed bug in serialization of xsd:duration types. o Added getBigSeconds to XSDDuration to avoid rounding errors in use of getSeconds. OntModel o the prefix mapping of an OntModel will not offer as a default URI the default URI of any non-base component. This change is implemented via Polyadic and hence any MultiUnion will have this behaviour. o added some extra checks so that importing a previously closed model URI no longer causes ClosedException (SF bug 1474220) o aliased method 'listImportedModels' to 'listSubModels' (the preferred name), and introduced a Boolean flag to allow the listed sub-models to be presented without their own import sub-models if required OntDocumentManager o added a 'load hook' so that custom behaviours can be executed just before and/or just after a URI is read in to the contents of an OntModel or one of its sub-models BooleanClassDescription (UnionClass, IntersectionClass, etc) o canAs() checking in the language profile is now more liberal for OntClass resources, to prevent partially-formed class descriptions triggering a ConversionException (SF bug 1608765) Legacy DAML API o This entire API has been deprecated, and will be removed in Jena 2.6. Ontology users are encouraged to work with OWL rather than DAML+OIL, however the DAML+OIL profile for OntModel will continue to work and be supported. The deprecated API is in the package: com.hp.hpl.jena.ontology.daml and was originally provided to ease transition of existing code from Jena 1 to Jena 2. This transitional capability is no longer supported. Reasoners o Fixed a ConcurrentModificationException bug in TransitiveReasoner o added simple string hanlding builtins regexp, strConcat, uriConcat o fixed bug with arithmetic builtins which caused matching against an already bound but incorrect solution to pass. o fixed bug in backward reasoner which caused to try redundant solutions to purely grounded calls. This may improve efficiency in some cases. o fixed bug in duplicate removal in GenericRuleReasoner in pure backward mode o fixed bug in parsing of typed literals to handle user defined prefixes (thanks to Steve Cranefield for the fix) o removed MAX_VARS limit from the old forward rule engine ==== Jena 2.4 JAR file changes: Added: wstx-asl-2.8.jar json.jar iri.jar Removed: stax-1.1.1-dev.jar jakarta-oro-2.0.8.jar N3/Turtle: o The syntax form 2.3 is now treated as an XSD decimal, not an XSD double This is inline with chnages to N3 and Turtle specifications and compatible with SPARQL. General o Fixed various security violations which prevent use within Applets Model o Model[Con] has acquired a new method, remove(S, P, O), which removes the triple (S, P, O) from the model. None of S, P, O can be null. O is of type RDFNode (no overloading for primitive types etc). ExtendedIterator [hence NiceIterator and ResultSetIterator] o added toSet() and toList(), which bundle the [remaining] elements of the iterator up as a Set or List. Testing these is a bit tricky; for the moment I've cheated by testing specific instacnes of WrappedIterator, which is built on NiceIterator, and NiceIterator is the base class for /almost/ all the ExtendedIterators in the system, except ResultSetIterator for which I added special code. Node o added getBlankNodeLabel() as shortcut to the label string of a blank node. AnonId o added getLabelString(), which delivers the label of this AnonId and (unlike toString()) is guaranteed to do /only/ that. GraphMatcher o following up on a performance issue, replaced the idiom find().hasNext() for containment in GraphMatch with proper use of contains(). After fixing a hashing problem in LiteralLabel [see below], this worked and sped up the checker /lots/ in some circumstances. NEW: Assembler specifications o Assembler specifications replace the old ModelSpecs. They are documented in doc/assembler/index.html and linked pages. JenaTestBase o Added new test method assertInstanceOf to do instanceof testing with a nice failure message; updated (some) test code to use it. FileGraph[Maker] o The recent change to FileGraph which meant that closed FileGraph's were removed from their owning Maker also meant that they were not deleted when the Maker was closed. Amongst other things, this left crud in the tmp directory when running the tests. Fixed: the Maker remembers /all/ the FileGraphs, even the closed ones. o Added a constructor that allows the file language to be explicitly specified. LiteralLabel o reported bug whereby a LiteralLabel can be constructed with the value part null has been fixed [by putting in the missing 'this.'] o discovered a problem with literal labels whose values are arrays: hashing doesn't work. Added delegation so that an RDFDataType can provide a specialised hash function and arranged that the relevant XSD datatypes used it. Database: o Added a based MS SQL Server 2000/MSDE 2000 driver. o Suppressed the "reifier must fix over-specification" messages from tests. Reasoners: o Fixed bug in the forward rule engine which prevented firing of axiomatic rules with non-empty bodies (i.e. no pattern match but calls to builtins). o Changed contract for TransitiveReasoner so the same reasoner can be bound to multiple datasets without interference between them o Added error check for backward rules with multiple head clauses OntModelSpec: o Changed contract for reasoners so that if a ReasonerFactory is supplied then the generated reasoner instances won't be cached. This allows the same spec to be reused multiple times with reasoners which can only be bound once. o OntModelSpecs now export access to their to-be-deprecated ModelMakers via the importModelGetters. This is part of the move to using Assemblers. OntDocumentManager o When the OntDocumentManager asks for a model for an import, it goes via the OntModelSpec's ModelGetter. Since the default ModelGetter just wraps the spec's ModelMaker, this doesn't change the usual behaviour, but it allows the OntModelSpec to use weak Getters. Jar changes: o stax-1.1.1-dev.jar ==> stax-1.1.2-dev.jar o new iri.jar required ==== Jena 2.3 ARQ - SPARQL for Jena o ARQ added to Jena See doc/ARQ/index.html for details JAR changes jena.jar has been split into jena.jar and jenatest.jar (the test packages). Replaced and upgraded with name changes: antlr.jar => antlr-2.7.5.jar jakarta-oro-2.0.5.jar => jakarta-oro-2.0.8.jar log4j-1.2.7.jar => log4j-1.2.12.jar icu4j.jar => icu4j_3_4.jar Xerces jars updated to Xerces 2.7.1 New Jarfiles: arq.jar jenatest.jar stax-1.1.1-dev.jar stax-api-1.0.jar Constraint rewriter o the .graph.query.Rewrite class recognises certain RDQL regex idioms and rewrites them. The rewritten expressions contained errors: (a) the case-insensitive classes stored the lowercased version fo the test string, which broken the RDB generated code; (b) against all his principles, kers had used toString() on the results of expression evaluation, which broke the comparison on typed or languaged literals. These Have Been Fixed. PrefixMappings o new boolean method samePrefixMappingAs(PrefixMapping) compares mappings for equality but has opportunity to avoid creating intermediate maps. o NOTE: cannot overload .equals() because Model::equals() is already defined. Making Model implement PrefixMapping may have been a mistake ... ARP o Total rewrite of internals. Now approx four times faster. User code will experience significantly less speed up, depending on the percentage of runtime taken up with parsing (as opposed to reading the data from the network or disk, and adding the triples to a Model). o The contract concerning behaviour after syntax errors has changed. (See Javadoc for ARPOptions#setErrorMode, for details of the new contract) o Some changes in the error codes produced for ill-formed input. o The treatment of interrupts has changed. Instead of throwing an InterruptedIOException, an error is produced: ERR_INTERRUPTED, and reported through the error handler as a fatal error. This normally throws a ParseException. o A few public classes that were in the old ARP package, but labelled as not part of the API have been removed. o DOM2Model public constructors have been deprecated, and replaced with factories. o SAX2Model and SAX2RDF factories methods have been deprecated and replaced. o SAX2RDF, SAX2Model protected constructors have changed in a not backwardly compatible fashion. o NTriple command-line option documentation changed, in manner that is theoretically not backwardly compatible. -r is now default (in documentation, follows previous implementation). -R option added to cancel -r. Node/Node_Literal/Literal o Node has gained the method getIndexingValue(), which is the value to use when indexing this Node for GraphMem lookup (and other such things). Non-literal nodes return themselves. Literal nodes return an appropriate value; the current implementation defers to the getIndexingValue() method of the associated LiteralLabel. o Node has gained the method getLiteralValue(), which fails if the node is not a Node_Literal and otherwise returns the value of the associated literal. This method allows uses of getLiteral().getValue() to be replaced, so that external code need not know about getLiteral() as much. o Literal::getWellFormed() has been deprecated; it is replaced by Literal::isWellFormedXML(). There is a missing API method eg isWellFormed() which would apply to any typed literal; this will arrive in due course. o Support for indexing of typed literals added. NodeToTripleMaps now use an indexing object to represent the Node rather than the Node itself. That object should implement the appropriate semantic equality: index(x) == index(y) <=> sameValueAs(x, y) If future datatyping extensions can't meet this contract it could be weakened to: index(x) != index(y) => ! sameValueAs(x, y) at the expense of post-processing find results with a sameValueAs test. Currently the index objects used are: plain literals, no lang and xsd:string literals -> the lexical form plain literals, lang tag -> the Node XMLLiterals -> the Node known typed literals -> the java getValue() object unknown typed literals -> a cons of the lexical value and datatype URI RDFNode/Resource/Literal o The method Resource::getNode() has been deprecated in favour of RDFNode::asNode(). o The method Resource::isAnon() has been moved up into RDFNode. o The (pointless) method Literal::isLiteral() has been moved up into RDFNode() (where it is pointful). o RDFNode has acquired `boolean isURIResource()` and `boolean isResource()`. o This allows all three what-kind-of-node tests to be applied to an RDFNode. Note that heavy use of these methods is a likely design smell - visitWith() may be a better solution to the classification problem. GraphMem/GraphTripleStore/NodeToTriplesMap o Performance analysis suggests that a chunk of the time in find() was taken up with redundant comparisions when filtering the intermediate iterator with the triple pattern. [EG: the hashmap index field is always right; the ANY nodes are never relevant.] So, after a go or two around the houses, this was optimised to only test the non-wild remaining fields, using the new Triple.Field operations plus the new Filter operations. o The default memory-based graph is now GraphMemFaster, which does optimised query handling. Triple o S/P/O fields made final (dunno why they weren't already). o Added Field class, which gives three constants (getSubject, getObject, Predicate) which have a getField(Triple) method to extract that field. Fields also have filterOn() methods to create filters over nodes and fields of triples; filtering over ANY nodes delivers any(), which composes cheaply. Filters & Iterators o Removed a performance infelicity in the default andThen implementation (it kept calling the left-hand hasNext even when it had been exhausted). Later replaced the implementation with one that keeps a list of pending iterators and itself implements .andThen() by extending the pending list. o Added FilterDropiterator so that filterDrop doesn't need to create a new negated Filter and suffer an extra indirection layer; added filterKeepIterator for symmetry. o Filter is now a class, not an interface, to make it easier to add new operations to it (otherwise the API changes would be grossly visible to all users). o Filter has new methods: .and(Filter), producing a Filter that passes only elements that pass both filters, and .filterKeep(Iterator), which filters the iterator in the same way as ExtendedIterator's filterKeep operation does. o any() has fast implementations for these operations, allowing it to be used as a fairly cheap identity element. ModelSpecs o Fixed a bug: the loadFile property did not work on inference models. The fix ensures that any descendant of ModelSpecImpl implements createModel() using a method doCreateModel() and then loading the specified files. Schemagen o Schemagen now by default includes individuals whose class is in one of the target namespaces for the document, even if the individual itself is not. This behaviour can be turned off with option strictIndividuals. Typed literals: o Fixed bug in unparsing of xsd:time values. o Added normalization step so that creating a typed literal from an XSDDateTime will use narrow types (e.g. xsd:date) when appropriate. o Fixed bug in sameValueAs when comparing an integer to a float/double. Reasoners: o Extended rule parser to support typed literals using N3 type syntax such as 'true'^^xsd:byte. o Fixed bug with rule sets which include a proprocessing hook to ensure, the hooks are rerun after new triple adds which should invoke the hook. o Fixed two bugs with derivation logging of backward rules. o Modified processing of non-monotonic rulesets (involving drop/remove) so that each entry in the conflict set is fired separately and all the consquences propagated before attempting to fire the next rule. To avoid performance hits, rulesets not involving such operators execute as before. User defined Builtins which remove data should be marked as such using the isMonotonic method. o Fixed bug in TransitiveGraphCache which had resulted in some transitive reductions being incompletely reduced (i.e. some indirect property instances were being incorrectly reported as being direct). o Added "drop" operator as an alternative to "remove" when performing non-monotonic rewrites. o Fixed bug in rebind/reset of infgraphs which use TGC and failed to reset the transitive engine. o Optimized Resource.remove operations for the case where the parent model is an InfModel. o The default DIG reasoner used in the documentation examples has been changed from Racer to Pellet. Pellet is free and open-source, while Racer has switched to a commercial license model. Ontology API o OntDocumentManager now delegates file resolution and model caching to FileManager, which means that FileManager's resolution strategies can be used to locate ontology files (e.g. from the classloader). o Prefix mapping and ontology language selection in the OntDocumentManager has been deprecated, and will be removed in a future version of Jena. Command line utilities o New utility jena.rdfcat, which can merge any number of individual rdf documents together into one model, and perform syntax translation (e.g. RDF/XML to N3). o New query utilities for SPARQL ModelLock o Deprected in favour of Lock (in shared). o Two implementations: LockMRSW and LockMutex ==== End Jena 2.3 MySQL ----- There is a problem when using MySQL j-connection 3.1.* and MySQL 4.1.*. It manifests itself as truncation of long literals. Systems not using long literals should not see any problem. Using j-connector 3.0.* or the development versions of j-connector 3.2 do not exhibit the problem. Post 2.2beta change list ARP o Fixed XMLLiteral bug relevant to OWL-S o Added workaround for ICU bug. The workaround may slow processing of Tamil and other langauges which use Unicode composing characters. If you are processing large volumes of Tamil using a patched version of icu4j may be faster. Ask on jena-dev for more information. o Improved character encoding support: all character encodings supported by Java are now supported by Jena. FileGraph and ModelMakerImpl o ModelMakerImpl now implements [the obsolescent] createModelOver using maker.openModel( ... ) rather than .createModel( ... ). This fixes a problem with existing files in the directory not being consulted for their contents. o FileGraphs now (weakly) support (non-nested) transactions, using checkpoint files to record their state at a begin() and restoring that state at an abort(). A commit() writes the current state to the backing file and deletes the checkpoint. InfModelBase (hence, any inference model) o Inference models now (weakly) support transactions; they delegate them to their base model. Additionally, an abort() will do a rebind(). JMS & ModelSpec o Added new modelName property to the vocabulary and schema, ready to properly support named models as well as model makers. o The specification for RDB models has changed: it is not the maker, but it's /hasConnection/ value, that has the connection properties. (This allows the connection to be shared, or to be prespecified.) o The vocabulary class JMS has been renamed to JenaModelSpec. There is a (deprecated) JMS subclass to allow legacy use of the vocabulary. o OntModelSpec understands the `modelName` property; it gives the name of the model (in the baseModelMaker) which is to be used in a ontModelSpec.create() call. OWL Syntax Checker o No longer in the Jena download. o A separate contribution. o Can be separately download from the Jena project files page. PrefixMappings o the requirement that adding `(prefix, uri)` to a prefix mapping remove any existing prefix for `uri` has been removed. Calls that run the mapping backward (eg qNameFor()) will get a correct answer, not necessarily the same one each time; if possible, it will be the "most recently bound" prefix. (The "not possible" cases are those where a prefix has been removed and the inverse mapping has been regenerated.) Ontology API o As part of a move to provide more consistent behaviour, listDeclaredProperties has been completely re-written. The new behaviour, which in some important respects differs from the old behaviour, is now documented in doc/how-to/rdf-frames.html. There has also been a non backwards-compatible change in the meaning of the Boolean flag passed to listDeclaredProperties. o An OntModel can now list the root classes in the local class hierarchy, see OntModel.listHierarchyRoots() DIG reasoner o Fixed a bug that meant that and were not being translated to owl:Thing and owl:Nothing, and hence not appearing in output RDF/XML Output o Improved character encoding support: all character encodings supported by Java are now supported by Jena. Post Jena 2.1 change list [up to 2.2beta] RDF API o Fixed bug in typed literals support which caused isValidLiteral tests to fail on user defined types derived indirectly from simple types. o Fixed bugs in conversion of Calendar to XSDDateTime o Fixed XSDDouble isValidValue test to check for Double o Fixed XSDbase64Binary, XSDhexBinary returning strings instead of byte[] o GraphExtract operations made available through ModelExtract plus StatementBoundary[Base] & StatementTripleBoundary. o Model has gained read(String url, String baseURI, String lang ). Database o Fixed user-reported problem where removeNsPrefix() wasn't persistent o handles some constraints internally o A minor change to ModelCom has improved the speed with which DB models with several prefixes are opened. GraphBase and Reification SPI o Reifier::getHiddenTriples() and getReificationTriples() REMOVED and replaced by iterator-returning find(TripleMatch) [for all quadlets], findExposed(TripleMatch) [for exposed quadlets, ie Standard], and findEither(TripleMatch, boolean showHidden) [for exposed or hidden quadlets]. o Reworking of GraphBase to clean up reification and fix bug with duplicated fully-reified triples. GraphBase now implements find() by appending triples from reifier with triples from local triple store. find() is not over-ridable; over-ride graphBaseFind() instead. Similarly size -> graphBaseSize, contains -> graphBaseContains. o Reworking of SimpleReifier to express it in terms of implementations of a store for fragments and a store for complete triples, with new interfaces. This should allow implementors of persistent Graphs an easier time of it. [Driven by GraphBDB work, so there's an example to hand.] o Reifiers must also implement size[OfExposedTriples]() and close() methods. Model & Graph removeAll(), remove(S, P, O) o added new API operation Model::removeAll() which removes "all" statements from a model [currently there are issues about inference models] o added removeAll() to Graph BulkUpdateHandler o added Model::remove(S, P, O) which removes all statements matching (S, P, O) with nulls as wildcards o added BulkUpdateHandler.remove(S,P,O) removing triples matching (S, P, O), Node.ANY as wildcard o BulkUpdateHandler generates events for these o ModelFactory has gained createUnion(Model, Model) which creates a dynamic union of the two argument models. o The class GraphExtract and its related interface TripleBoundary have been created to allow the extraction of rooted subgraphs from a graph, terminating at triples satisfying some boundary condition. GraphMem, SmallGraphMem o GraphMem has had the redundant Set(Triple) excised, and changes made to NodeToTriplesMap to push the triple-matching inwards and simplify GraphMem's find code. It will use a little less memory and should be a tad faster. o a new memory-based Graph, SmallGraphMem, has been introduced. This *only* holds a Set(Triple); no indexing is done. Hence it is unsuitable for graphs with more than "a few" statements, unless memory footprint is (much) more important than search speed. It is primarily an extreme to compare other Graphs against. Graph Capabilities gains findContractSafe() o used to tell prettywriter that its use of find() works, otherwise it falls back to the ordinary writer. Graph Query handling o Query now rewrites (some) RDQL pattern matches which are equivalent to startsWith, endsWith, or contains to use new Expression nodes with labels J_startsWith, J_endsWith, J_contains, to allow back ends to optimise those. RDQL o Added langeq operator o Remove ResultBinding.getValue() (which was an internal use operation) as part of move to more directly using the graph level queryhandling. o ResultBinding becomes an interface. See also ResultBindingImpl. Event handling o added new graph.GraphEvents as holder of event constants with one such, .removeAll, issued by BulkUpdateHandler for removeAll(), and a static method for creating removed-by-pattern triple values. o the standard Jena readers generate startRead and finishRead events o all the Graph events now come with their source Graph attached as an argument to the event method. o added test case and fixed resulting issues to ensure that .remove() on the iterator from find() generated a notifyDeleteTriple event. Reasoners o Changed processing of PROPruleSet on GenericRuleReasoners to accept multiple rulesets and merge them o Added support for @prefix and @include to the Rule parser utility o Suppressed internal properties (rb:xsdRange etc) from leaking out of infmodel o Fix space leak in backward chainer o Added subProperty inheritance of domain/range values o Fixed validation of owl Functional properties to handle literals o Changed validation report of un-instantiatable classes to be warnings rather than errors, report name is "Inconsistent class" o During validation the culprit resource is now made available via the error report's getExtension method o Fixed bug in backward reasoner which caused it to occasionally miss some results in non-deterministic ways o Fixed performance problem with listing all triples in a transitive reasoner o Fixed bug 927644, mishandling of cycles in transitive reasoner o Fixed ommission in handling of someValuesFrom applied to an rdfs:Datatype. o Fixed bug in hide() table not being carried across from prebuilt reasoner caches, which resulted in the prototypical instance of owl:Thing being visible in listIndividuals. o Fixed bug in TransitiveReasoner reported by wangxiao o Changed delete for RETE reasoner to be non-incremental to work around bug without demanding full reference counting o Added experimental OWLMini and OWLMicro reasoners o [kers] Added WrappedReasonerFactory class and RuleReasoner interface. Some tweaking to "implements" clauses. Refactored out some setRules code into FBRuleReasoner and BaseRuleReasonerFactory, to make it easy to share code for ModelSpec's reasoner specs. ModelSpecs o The modelspec language has been extended. Reasoner specs can now specify multiple rulesets by URL or by literal strings. Schemas may also be specified by URL. o Internal refactoring has cleaned up the API somewhat, and there is, optionally, config information read from etc/modelspec-config.n3 At the moment, this only allows the Java class that implements a given jms:ModelSpec type to be specified. o The JMS (Java Model Spec) vocabulary class has had its schema extracted - it is now loaded from vocabularies/jena-model-spec.n3. That vocabulary element is now added to jena.jar (as are some other vocabulary elements used by the system). PrefixMappings o Standard no longer contains vcard, jms, or rss. o Extended introduced = Standard + vcard, jms, rss. o usePrefix deprecated - use shortForm instead. o qnameFor added; result is legal qname or null. o w.withDefaultMapping(x) added, which adds mappings from x which don't clash with those already in w. o the restriction that namespaces must not end with name characters has been removed. Exceptions o added WrappedException extends JenaException for wrapped exceptions o added WrappedIOException extends WrappedException Node o Node cache hash table replaced by specialised implementation o new Node methods getNameSpace, getLocalName, hasURI(String) o minor adjustments to Node.create(String) to allow for specifying language or type for literals, use of default prefix, and elimination of the nn-NN hack for numeric literals. o default Node.toString() uses quoting and @/^^ for literals Triple o Triple.create now uses a cache Resource o new method hasURI(String) Ontology API o Added a new method getOWLLanguageLevel() to OntModel, which returns the OWL sublanguage - lite, DL or full - and error messages o Fixed ont language profiles to allow .as() on e.g. owl:SymmetricProperty as OntProperty.class. Previously this relied on the reasoner; the change was needed to support the DIG interface. o OntModels created over existing base models don't include elements from their document manager's PrefixMapping if they would clash with those already existing in the base. o Fixed bug 985258 - schemagen can now create RDFS vocabularies o Fixed bug 940570 - solved a problem with listIndividuals when using the default OWL reasoner o Fixed bug 948995 - .as() on owl:InverseFunctionalProperty for datatype properties failing o Various fixes to prevent cycles in the graph confusing listSubClasses, listSuperClasses, etc o fixed profiles to allow owl:Thing .as(OntClass.class) even if no reasoner present o In response to bug 1065103, DAML models now by default use rdfs:subClassOf, rdfs:subPropertyOf, rdfs:range and rdfs:domain in preference to their daml: equivalents. The old (Jena 2.1) behaviour is available by switching to the DAML_OILLegacyProfile in the OntModelSpec. The new version more closely matches what typical DAML ontologies do. o Added createIndividual() to OntClass o Added listDataRanges() to OntModel DIG reasoner interface o Various bug fixes, plus a significant performance fix for listIndividuals() OWL Syntax Checker o Added support for OntDocumentManager o Improved command line, supports N3 etc, OntDocumentManager File Utilities o ModelLoader retired by deprecation. Use FileManager instead, specifically, FileManager.get() for the global FileManager (or create your own). Create new model with FileManager.get().loadModel Read into an existing model with FileManager.get().readModel o In schemagen, inference is now *not* used by default on input models; a new option --inference has been added to allow inference to be turned on. Creating sets/maps o util.CollectionFactory has static methods for creating hashed Maps and hashed Sets. These are used in the internals to allow the implementing classes to be changed (eg to use the trove library). Non-hashed-collection create methods may follow. o This was initially called HashUtils; that class remains, with the initial method names, but it is deprecated and will disappear post J2.2. ARP o New support for non-Xerces SAX sources. o Support for DOM sources (Java 1.4.2 and later). o ARP setup rationalized, a few methods deprecated as a result. o Improved documentation, covering new features, (see doc/ARP) o *Removed* StanfordImpl, assuming noone uses it. Please complain (jena-dev@yahoogroups.com) if assumption was false. Utilities o new class IteratorCollection with methods iteratorToSet and iteratorToList (heavility used in tests and useful in general) N3 o Resolve relative URIs if there is a base. Jena 2.1 - Main changes from Jena 2.0 Installation The names of some jars have changed be sure to update your classpath. The name of the xerces jars are now: xercesImpl.jar and xml-apis.jar We also require Jakarta commons logging: commons-logging.jar See readme.html for the full list of jars. OWL Syntax Checker Major update from alpha version (Jena 2.0), to production version (Jena 2.1). API created. Many fold performance improvement (orders of magnitude) Now conformant with OWL Recommendation. Streaming mode added, suitable for lower memory environments or large input. Command-Line jena.owlsyntax program added. Still to do: better error messages. RDF/XML-ABBREV output Changes default configuration to not use property attributes, which seem unpopular. ARP Extended API to show bnode scope and XML Namespaces. Discovered memory leak which has been present since the beginning. This is not fixed. Users of ARP and Jena in memory limited, or long-running applications, or reading lots of varied RDF/XML, should read the updated Javadoc for the package com.hp.hpl.jena.rdf.arp. Reasoner Small bug fixes (see below). Xerces Now requires Xerces 2.6.0 or better. The included jars are Xerces 2.6.1. Ontology API General bug fixes and improvements based on jena-dev feedback. The default document manager policy (etc/ont-policy.rdf) no longer re-directs imports of owl.owl and daml+oil.daml to a cached copy in the 'vocabulary' directory. This is becuase the vocabulary directory is not included in jena.jar, and this default re-direction was causing problems in some applet or web service environments. The Jena 2.0 behaviour can be restored by replacing ont-policy.rdf with ont-policy-test.rdf. Instance detection has been improved, with the side-effect that DAML ontologies using the DAML micro rule reasoner may now report that instances have rdf:type daml:Thing in addition to other types. Jena 2.1 changes from Jena 2.1-dev-3 Minor bug fixes in OWL Syntax Checker. Streaming mode in OWL Syntax Checker. Documented ARP memory leak. Jena 2.1-dev-3 Implements W3C RDF and OWL Proposed Recommendations OWL Syntax Checker - much faster, new API (error msgs still being worked on) RDF/XML-ABBREV output various bug fixes RDF/XML-ABBREV msg added requesting bug reports on rdf:parseType="Collection" ARP new extended handler for scope of blank nodes and namespace handler ARP improved syntax error messages OWL Syntax checker prolog source included in download (see tools dir) Jena 2.1-dev-2: Developers' release to include recent bug fixes (notably handling of typed literals in validation rules). Do not use this version unless you need one or more bug fixes not in Jena 2.0. Jena 2.1-dev-1: This is a developers' release, particularly intended for users of the OWL Syntax Checker. Most users should continue to use Jena 2.0. For changes, see directly below. Documentation may not be up to date. Do not use this version unless: - you need a conformant OWL Syntax Checker - you need one or more of the bug fixes not in Jena 2.0 RDF API: o Bug fixes: - fixed issue with typed literals sometimes treating lexically different, sameValueAs resources as equal. - fixed bug in Model::remove(StmtIterator) at the expense of manifesting the iterator into a List - fixed bug in .remove on StmtIterators returned from listStatements() on a memory model o ModelFactory improvements: - ModelSpecs can now be identified by URI as well as Model values - ModelRDBMaker.createModel will now return a ModelRDB rather than plain Model Reasoner subsystem: o Fixed delete bug in FORWARD_RETE configurations, remove of statements should now remove the consequents of that statement. o Added a check for constructing one OWL reasoner instance layered on top of another because this can have a large performance impact to no benefit. o Added a "hide" primitive which marks nodes as hidden. When querying an inference model no triples containing hidden subject or object nodes will be included in the result iterator. Used this to hide class-prototype instances o Extended the comparision builtins (equal, le etc) to support comparison of XSDDataTime instances. Many thanks to Bradley Schatz (Bradley@greystate.com) for supply the patches for this. o Extended OWL rules to include more explicit representation of XSD knowledge. o Various bug fixes in OWL rules (maxCardinality bug, hasValue in intersection lists fixed, bug in someValuesFrom fixed, misssing property subclass axioms). o Fixed bug in RETE engine which could loop when deleting non-deletable triples. o Fixed bug in LP engine which could lead to loss of variable bindings (manifested as "Internal error in LP reasoner: variable in triple result") o Extended is/notDTtype to check for ill-formed typed literals. OWL Syntax Checker o Now conforms with OWL Proposed Rec of December 2003 - Performance much improved. There is about a one second delay on start-up. - Error messages still somewhat cryptic (Should be better in next release) RDF/XML-ABBREV output o Failed to fix bug concerning rdf:parseType="Collection" - added new message trying to generate sufficient user feedback on jena-dev to track down bug. o Fixed other bugs on bug list RDQL o Improved handling of character sets in qnames N3 o Improved handling of character sets in qnames Graph query SPI [NB NOT visible at the model level] o replaced use of Graph for constraints by new Expression interface as part of ongoing query improvement.