Git commit 9476df60687ecb133ebc9dd2f9c893edc1ab097a by Sebastian Trueg.
Committed on 31/03/2012 at 20:46.
Pushed by trueg into branch 'feature/virtuosoInference'.

New resource visibility handling.

Until now the query api would use the nao:userVisible values stored for
each resource to determine if a resource should be returned.
This requires a lot of nao:userVisible value maintenance and makes for
slower queries.
The new solution is much simpler: only types have nao:userVisible values
and the query api simply checks if the result candidate has a type
which is visible.
In contrast to the old one the new solution does not allow to hide a
sub-class of a visible class. But since we never did that anyway and
the resulting queries are faster and it allows us to remove a lot
maintenance code (which eats CPU cycles and disk space) the change is
worth it.

M  +8    -13   libnepomukcore/query/query.cpp
M  +52   -0    services/storage/ontologyloader.cpp
M  +2    -1    services/storage/ontologyloader.h

http://commits.kde.org/nepomuk-core/9476df60687ecb133ebc9dd2f9c893edc1ab097a

diff --git a/libnepomukcore/query/query.cpp b/libnepomukcore/query/query.cpp
index 943705d..6e242f4 100644
--- a/libnepomukcore/query/query.cpp
+++ b/libnepomukcore/query/query.cpp
@@ -1,6 +1,6 @@
 /*
   This file is part of the Nepomuk KDE project.
-   Copyright (C) 2008-2010 Sebastian Trueg <trueg@kde.org>
+   Copyright (C) 2008-2012 Sebastian Trueg <trueg@kde.org>

   This library is free software; you can redistribute it and/or
   modify it under the terms of the GNU Library General Public
@@ -450,22 +450,17 @@ QString Nepomuk::Query::Query::toSparqlQuery( SparqlFlags sparqlFlags ) const
    QueryBuilderData qbd( d.constData(), sparqlFlags );


-    // We restrict results to user visible types only. There is no need for this with file queries as normally all files are visible.
    //
-    // Here we use an optimizations:
-    // The storage service creates "nao:userVisible 1" entries for all visible resources. This contains two optimizations:
-    // - We use an integer instead of a boolean because Virtuoso does not support booleans and, thus, Soprano converts
-    //   booleans into a fake type which is stored as a string. This makes comparision much slower.
-    // - Instead of using crappy inference via "?r a ?t . ?t nao:userVisible true" we can directly check the resources.
-    // - Instead of restricting the visibility to 1 we trick the Virtuoso query optimizer so it won't try to use the visibility
-    //   as priority. We do this with a filter that does not test for equality. Using "?r nao:userVisible 1" has a drastic
-    //   performance impact which the filter has not.
+    // We restrict results to user visible types only. There is no need for this with file queries as normally all files are visible.
+    // We need the additional type pattern to ensure that the variable ?r is actually available to the filter (in case there is only
+    // UNIONs before.
    //
    QString userVisibilityRestriction;
    if( !(queryFlags()&NoResultRestrictions) && !d->m_isFileQuery ) {
-        userVisibilityRestriction = QString::fromLatin1("?r %1 %2 . FILTER(%2>0) . ")
-                .arg(Soprano::Node::resourceToN3(Soprano::Vocabulary::NAO::userVisible()),
-                     qbd.uniqueVarName());
+        userVisibilityRestriction = QString::fromLatin1("?r a %1 . FILTER EXISTS { ?r a [ %2 %3 ] . } . ")
+                                    .arg(qbd.uniqueVarName(),
+                                         Soprano::Node::resourceToN3(Soprano::Vocabulary::NAO::userVisible()),
+                                         Soprano::Node::literalToN3(Soprano::LiteralValue(true)));
    }


diff --git a/services/storage/ontologyloader.cpp b/services/storage/ontologyloader.cpp
index c06a297..a09665b 100644
--- a/services/storage/ontologyloader.cpp
+++ b/services/storage/ontologyloader.cpp
@@ -27,6 +27,7 @@
 #include <Soprano/PluginManager>
 #include <Soprano/StatementIterator>
 #include <Soprano/Parser>
+#include <Soprano/QueryResultIterator>

 #include <KConfig>
 #include <KConfigGroup>
@@ -42,14 +43,23 @@
 #include <kpluginfactory.h>
 #include <kpluginloader.h>

+#include <Soprano/Vocabulary/RDFS>
+#include <Soprano/Vocabulary/NAO>
+#include <Soprano/Vocabulary/XMLSchema>

 using namespace Soprano;
+using namespace Soprano::Vocabulary;
+
+namespace {
+    const char* s_typeVisibilityGraph = "nepomuk:/ctx/typevisibility";
+}

 class Nepomuk::OntologyLoader::Private
 {
 public:
    Private( OntologyLoader* p )
        : forceOntologyUpdate( false ),
+          someOntologyUpdated( false ),
          q( p ) {
    }

@@ -59,6 +69,9 @@ public:
    bool forceOntologyUpdate;
    QStringList desktopFilesToUpdate;

+    // true if at least one ontology has been updated
+    bool someOntologyUpdated;
+
    void updateOntology( const QString& filename );

 private:
@@ -110,6 +123,8 @@ void Nepomuk::OntologyLoader::Private::updateOntology( const QString& filename )
    }

    if( update ) {
+        someOntologyUpdated = true;
+
        QString mimeType = df.readEntry( "MimeType", QString() );

        const Soprano::Parser* parser
@@ -179,6 +194,7 @@ Nepomuk::OntologyLoader::~OntologyLoader()

 void Nepomuk::OntologyLoader::updateLocalOntologies()
 {
+    d->someOntologyUpdated = false;
    d->desktopFilesToUpdate = KGlobal::dirs()->findAllResources( "xdgdata-ontology", "*.ontology", KStandardDirs::Recursive|KStandardDirs::NoDuplicates );
    if(d->desktopFilesToUpdate.isEmpty())
        kError() << "No ontology files found! Make sure the shared-desktop-ontologies project is installed and XDG_DATA_DIRS is set properly.";
@@ -201,6 +217,16 @@ void Nepomuk::OntologyLoader::updateNextOntology()
    else {
        d->forceOntologyUpdate = false;
        d->updateTimer.stop();
+
+        // update graph visibility if something has changed or if we never did it
+        const QUrl visibilityGraph = QUrl::fromEncoded(s_typeVisibilityGraph);
+        if(d->someOntologyUpdated ||
+           !d->model->executeQuery(QString::fromLatin1("ask where { graph %1 { ?s ?p ?o . } }")
+                                   .arg(Soprano::Node::resourceToN3(visibilityGraph)),
+                                   Soprano::Query::QueryLanguageSparql).boolValue()) {
+            updateTypeVisibility();
+        }
+
        emit ontologyLoadingFinished(this);
    }
 }
@@ -229,6 +255,7 @@ void Nepomuk::OntologyLoader::slotGraphRetrieverResult( KJob* job )
        // TODO: find a way to check if the imported version of the ontology
        // is newer than the already installed one
        if ( d->model->updateOntology( graphRetriever->statements(), QUrl()/*graphRetriever->url()*/ ) ) {
+            updateTypeVisibility();
            emit ontologyUpdated( QString::fromAscii( graphRetriever->url().toEncoded() ) );
        }
        else {
@@ -237,4 +264,29 @@ void Nepomuk::OntologyLoader::slotGraphRetrieverResult( KJob* job )
    }
 }

+void Nepomuk::OntologyLoader::updateTypeVisibility()
+{
+    const QUrl visibilityGraph = QUrl::fromEncoded(s_typeVisibilityGraph);
+
+    // 1. remove all visibility values we added ourselves
+    d->model->removeContext(visibilityGraph);
+
+    // 2. make rdfs:Resource non-visible (this is required since with KDE 4.9 we introduced a new
+    //    way of visibility handling which relies on types alone rather than visibility values on
+    //    resources. Any visible type will make all sub-types visible, too. If rdfs:Resource were
+    //    visible everything would be.
+    d->model->removeAllStatements(RDFS::Resource(), NAO::userVisible(), Soprano::Node());
+
+    // 3. Set each type visible which is not rdfs:Resource and does not have a non-visible parent
+    d->model->executeQuery(QString::fromLatin1("insert into %1 { "
+                                               "?t %2 'true'^^%3 . "
+                                               "} where { "
+                                               "?t a rdfs:Class . "
+                                               "filter not exists { ?tt %2 'false'^^%3 .  ?t rdfs:subClassOf ?tt . } }")
+                           .arg(Soprano::Node::resourceToN3(visibilityGraph),
+                                Soprano::Node::resourceToN3(NAO::userVisible()),
+                                Soprano::Node::resourceToN3(XMLSchema::boolean())),
+                           Soprano::Query::QueryLanguageSparql);
+}
+
 #include "ontologyloader.moc"
diff --git a/services/storage/ontologyloader.h b/services/storage/ontologyloader.h
index 4e0ce26..d714249 100644
--- a/services/storage/ontologyloader.h
+++ b/services/storage/ontologyloader.h
@@ -1,5 +1,5 @@
 /* This file is part of the KDE Project
-   Copyright (c) 2007-2010 Sebastian Trueg <trueg@kde.org>
+   Copyright (c) 2007-2012 Sebastian Trueg <trueg@kde.org>

   This library is free software; you can redistribute it and/or
   modify it under the terms of the GNU Library General Public
@@ -91,6 +91,7 @@ namespace Nepomuk {
        // a little async updating
        void updateNextOntology();
        void slotGraphRetrieverResult( KJob* job );
+        void updateTypeVisibility();

    private:
        class Private;