The structure of the data models has grown as VIVO has developed. New models, new structures, and new means of accessing the data have been added as required by the growing code. The resulting data layer has grown more complex and more error-prone.

In release 1.5, VIVO added the RDFService interface, which has increased the flexibility of data sources, and promises to allow a more unified view of the knowledge base. However, the transition to RDFService is not complete, and so this adds another layer of complexity to the data issues. New structures have been added, but none removed.

Beyond the models

There is an incredible variety of ways to access all of these models. Some of this variety is because the models are accessed in different ways for different purposes. Additional variety stems from the evolution of VIVO in which new mechanisms were introduced without taking the time and effort to phase out older mechanisms.

Here are some of the ways for accessing data models:

Attributes on Context, Session, or Request

Previously, it was common to assign a model to the ServletContext, to the HTTP Session, or to the HttpSessionRequest like this:

OntModel ontModel = (OntModel) getServletContext().getAttribute("jenaOntModel");

Object sessionOntModel = request.getSession().getAttribute("jenaOntModel");

ctx.setAttribute("jenaOntModel", masterUnion);

Occasionally, conditional code was inserted, to retrieve a model from the Request if available, and to fall back to the Session or the Context as necessary. Such code was sporadic, and inconsistent. This sort of model juggling also involved inversions of logic, with some code acting so a model in the Request would override one in the Session, while other code would prioritize the Session model over the one in the Request. For example:

Accessing the models

There is an incredible variety of ways to access all of these models. Some of this variety is because the models are accessed in different ways for different purposes. Additional variety stems from the evolution of VIVO in which new mechanisms were introduced without taking the time and effort to phase out older mechanisms.

Here are some of the ways for accessing data models:

Attributes on Context, Session, or Request

Previously, it was common to assign a model to the ServletContext, to the HTTP Session, or to the HttpSessionRequest like this:

OntModel ontModel = (OntModel) getServletContext().getAttribute("jenaOntModel");

Object sessionOntModel = request.getSession().getAttribute("jenaOntModel");

ctx.setAttribute("jenaOntModel", masterUnion);

Occasionally, conditional code was inserted, to retrieve a model from the Request if available, and to fall back to the Session or the Context as necessary. Such code was sporadic, and inconsistent. This sort of model juggling also involved inversions of logic, with some code acting so a model in the Request would override one in the Session, while other code would prioritize the Session model over the one in the Request. For example:

public OntModel getDisplayModel(){     
    if( _req.getAttribute("displayOntModel") != null ){
        return (OntModel) _req.getAttribute(DISPLAY_ONT_MODEL);
    } else {
        HttpSession session = _req.getSession(false);
        if( session

public OntModel getDisplayModel(){     
    if( _req.getAttribute("displayOntModel") != null ){
        return (OntModel) _req.getAttribute(DISPLAY_ONT_MODEL);
    } else {
        HttpSession session = _req.getSession(false);
        if( session != null ){
            if( session.getAttribute(DISPLAY_ONT_MODEL) != null ){            
                return (OntModel) session.getAttribute(DISPLAY_ONT_MODEL);
            }else{
                if( session.getServletContext().getAttribute(DISPLAY_ONT_MODEL) != null){
                    return (OntModel)session.getServletContext().getAttribute(DISPLAY_ONT_MODEL); 
                }
            }
        }            
    }
    log.error("No display model could be found.");
    return null;                
}

...

It remains to be seen whether this uniformity can satisfy the various needs of the application. If not, at least the changes can all be made within a single point of access.

The DAO layer

This mechanism is pervasive through the code, and remains quite useful. In it, a WebappDaoFactory is created, with access to particular data models. This factory then can be used to create DAO objects which satisfy interfaces like IndividualDao, OntologyDAO, or UserAccountsDAO. Each of these object implements a collection of convenience methods which are used to manipulate the backing data models.

...

Initially, the WebappDaoFactory may have been used only with the full Union model. But what if you want to use these DAOs only against asserted triples? Or only against the ABox? This led to the OntModelSelector.

OntModelSelectors

An OntModelSelector provides a way to collect a group of Models and construct a WebappDaoFactory. With slots for ABox, TBox, and Full model, an OntModelSelector could provide a consistent view on assertions, or on inferences, or on the union. The OntModelSelector also holds references to a display model, an application metadata model, and a user accounts model, but these are more for convenience than flexibility.

...

Again, this is a change in the semantics of OntModelSelectors. It insures a consistent representation of OntModels across OntModelSelectors, but it is certainly possible that existing code relies on an inconsistent model instead.

The RDF Service

...

Model makers and Model sources

The ModelAccess class

...

• the first time that VIVO starts up,
• if a model is found to be empty,
• every time that VIVO starts up.

depending on the particular model.

• up.

depending on the particular model. 

Where are the RDF files?

In the distribution, the RDF files appear in [vivo]/rdf and in [vitro]/webapp/rdf. These directories are merged during the build process in the usual way, with files in VIVO preferred over files in Vitro.

During the build process, the RDF files are copied to the VIVO home directory, and at runtime VIVO will read them from there.

The "first time"

For purposes of initialization, "first time" RDF files are loaded if the relevant data model contains no statements. Content models may also load "first time" files if the RDFService detects that its SDB-based datastore has not been initialized.

...

Name: http://vitro.mannlib.cornell.edu/default/vitro-kb-2

Source: named graph from the RDFService

/default/vitro-kb-2

Source: named graph from the RDFService

If first setup, read the files in rdf/abox/firsttime

• In Vitro, there are none
• In VIVO, geopolitical.ver1.1-11-18-11.individual-labels.rdf

Every restartIf first setup, read the files in rdf/abox/firsttimeabox/filegraph, and create named models in the RDFService. Add them as sub-models to the base ABox. If these files are changed or deleted, update the RDFService accordingly.

• In Vitro, there are none
• In VIVOVivo, geopolitical.abox.ver1.1-11-18-11.individual-labels.rdf, vocabularySource-labels.n3-11.owl, academicDegree.rdf, continents.n3
  us-states.rdf, dateTimeValuePrecision.owl,  validation.n3, documentStatus.owl,  vocabularySource.n3

base TBox

Name: http://vitro.mannlib.cornell.edu/default/asserted-tbox

Source: named graph from the RDFService (memory-mapped)

If first setup, read the files in rdf/tbox/firsttime (without subdirectories)Every time, read the files in rdf/abox/filegraph, and create named models in the RDFService. Add them as sub-models to the base ABox. If these files are changed or deleted, update the RDFService accordingly.

• In Vitro, there are none
• In Vivo, academicDegree.rdf, continents.n3, dateTimeValuePrecision.owl, documentStatus.owl, geopolitical.abox.ver1.1-11-18-11.owl, us-states.rdf, vocabularySource.n3

base TBox

Name: http://vitro.mannlib.cornell.edu/default/asserted-tbox

Source: named graph from the RDFService (memory-mapped)

If first setup, read the files in rdf/tbox/firsttime (without subdirectories)

• In Vitro, there are none
• In VIVO, geopolitical-ver1.1-11-18-11-annotations.rdf, isDefinedBy-1.5-annotations.rdf, scires-1.5-annotations.rdf, vitro-0.7-annotations.rdf, vivo-core-1.5-annotations.rdf, vivo-core-1.5-labels_es_ES.n3

Every time, read the files in rdf/tbox/filegraph, and create named models in the RDFService. Add them as sub-models to the base TBox. If these files are changed or deleted, update the RDFService accordingly.

...

• VIVO, additionalHiding.n3  initialTBoxAnnotations.n3

Every restart, read the files in rdf/tbox/filegraph, and create named models in the RDFService. Add them as sub-models to the base TBox. If these files are changed or deleted, update the RDFService accordingly.

• In Vitro, vitro-0.7.owl, vitroPublic.owl
• In VIVO, 44 files:

README.md			    	education.owl			personTypes.n3
agent.owl			    	event.owl				process.owl
appControls-temp.n3			geo-political.owl		publication.owl
bfo-bridge.owl				grant.owl				relationship.owl
bfo.owl				    	linkSuppression.n3		relationshipAxioms.n3
classes-additional.owl  	location.owl			research-resource-iao.owl
clinical.owl				object-properties.owl	research-resource.owl
contact-vcard.owl			object-properties2.owl	research.owl
contact.owl			    	object-properties3.owl	role.owl
data-properties.owl			objectDomains.rdf		sameAs.n3
dataDomains.rdf				objectRanges.rdf		service.owl
dataset.owl			    	ontologies.owl			skos-vivo.owl
date-time.owl				orcid-interface.n3		teaching.owl
dateTimeValuePrecision.owl	other.owl				vitro-0.7.owl
documentStatus.owl		    outreach.owl			vitroPublic.owl

base Full

Source: a combination of base ABox and base TBox

...

...

Notes:

• The default WebappDaoFactory is the one backed by the unionOntModelSelector. On the request level, this is also known as the "fullWebappDaoFactory". The convenience methodsgetWebappDaoFactory() and setWebappDaoFactory(wdf) support this use.

• "baseWebappDaoFactory" and "assertionsWebappDaoFactory"  are both previous terms for the WebappDaoFactory backed by the baseOntModelSelector. The convenience methods getBaseWebappDaoFactory() and setBaseWebappDaoFactory(wdf) support this use.

• Nobody was using the "deductionsWebappDaoFactory", so we got rid of it.

Future directions?

...