/*
 * Copyright (c) 1998, 2020 Oracle and/or its affiliates. All rights reserved.
 * Copyright (c) 1998, 2020 IBM Corporation. All rights reserved.
 *
 * This program and the accompanying materials are made available under the
 * terms of the Eclipse Public License v. 2.0 which is available at
 * http://www.eclipse.org/legal/epl-2.0,
 * or the Eclipse Distribution License v. 1.0 which is available at
 * http://www.eclipse.org/org/documents/edl-v10.php.
 *
 * SPDX-License-Identifier: EPL-2.0 OR BSD-3-Clause
 */

// Contributors:
//     Oracle - initial API and implementation from Oracle TopLink
//     07/19/2011-2.2.1 Guy Pelletier
//       - 338812: ManyToMany mapping in aggregate object violate integrity constraint on deletion
//     08/01/2012-2.5 Chris Delahunt
//       - 371950: Metadata caching
//     10/25/2012-2.5 Guy Pelletier
//       - 374688: JPA 2.1 Converter support
//     09 Jan 2013-2.5 Gordon Yorke
//       - 397772: JPA 2.1 Entity Graph Support
//     02/11/2013-2.5 Guy Pelletier
//       - 365931: @JoinColumn(name="FK_DEPT",insertable = false, updatable = true) causes INSERT statement to include this data value that it is associated with
//     06/03/2013-2.5.1 Guy Pelletier
//       - 402380: 3 jpa21/advanced tests failed on server with
//         "java.lang.NoClassDefFoundError: org/eclipse/persistence/testing/models/jpa21/advanced/enums/Gender"
//     10/19/2016-2.6 Will Dazey
//       - 506168: Make sure nestedTranslation map is new reference when cloned
//     03/22/2018-2.7.2 Lukas Jungmann
//       - 441498: @ElementCollection on Map<@Embeddable,String> cause NullPointerException when @Embeddable has a FK
package org.eclipse.persistence.mappings;

import java.beans.PropertyChangeListener;
import java.util.ArrayList;
import java.util.Arrays;
import java.util.Collection;
import java.util.Enumeration;
import java.util.HashMap;
import java.util.Hashtable;
import java.util.Iterator;
import java.util.List;
import java.util.Map;
import java.util.Map.Entry;
import java.util.Set;
import java.util.Vector;

import org.eclipse.persistence.descriptors.ClassDescriptor;
import org.eclipse.persistence.descriptors.FetchGroupManager;
import org.eclipse.persistence.descriptors.changetracking.AttributeChangeTrackingPolicy;
import org.eclipse.persistence.descriptors.changetracking.DeferredChangeDetectionPolicy;
import org.eclipse.persistence.descriptors.changetracking.ObjectChangeTrackingPolicy;
import org.eclipse.persistence.exceptions.DatabaseException;
import org.eclipse.persistence.exceptions.DescriptorException;
import org.eclipse.persistence.exceptions.QueryException;
import org.eclipse.persistence.expressions.Expression;
import org.eclipse.persistence.internal.descriptors.DescriptorIterator;
import org.eclipse.persistence.internal.descriptors.ObjectBuilder;
import org.eclipse.persistence.internal.expressions.SQLSelectStatement;
import org.eclipse.persistence.internal.helper.DatabaseField;
import org.eclipse.persistence.internal.helper.DatabaseTable;
import org.eclipse.persistence.internal.identitymaps.CacheKey;
import org.eclipse.persistence.internal.queries.ContainerPolicy;
import org.eclipse.persistence.internal.queries.EntityFetchGroup;
import org.eclipse.persistence.internal.queries.JoinedAttributeManager;
import org.eclipse.persistence.internal.queries.MappedKeyMapContainerPolicy;
import org.eclipse.persistence.internal.sessions.AbstractRecord;
import org.eclipse.persistence.internal.sessions.AbstractSession;
import org.eclipse.persistence.internal.sessions.AggregateChangeRecord;
import org.eclipse.persistence.internal.sessions.ChangeRecord;
import org.eclipse.persistence.internal.sessions.MergeManager;
import org.eclipse.persistence.internal.sessions.ObjectChangeSet;
import org.eclipse.persistence.internal.sessions.UnitOfWorkChangeSet;
import org.eclipse.persistence.internal.sessions.UnitOfWorkImpl;
import org.eclipse.persistence.logging.SessionLog;
import org.eclipse.persistence.mappings.converters.Converter;
import org.eclipse.persistence.mappings.foundation.AbstractTransformationMapping;
import org.eclipse.persistence.mappings.foundation.MapKeyMapping;
import org.eclipse.persistence.mappings.querykeys.DirectQueryKey;
import org.eclipse.persistence.mappings.querykeys.QueryKey;
import org.eclipse.persistence.queries.DeleteObjectQuery;
import org.eclipse.persistence.queries.FetchGroup;
import org.eclipse.persistence.queries.FetchGroupTracker;
import org.eclipse.persistence.queries.ObjectBuildingQuery;
import org.eclipse.persistence.queries.ObjectLevelReadQuery;
import org.eclipse.persistence.queries.ReadAllQuery;
import org.eclipse.persistence.queries.ReadObjectQuery;
import org.eclipse.persistence.queries.ReadQuery;
import org.eclipse.persistence.queries.WriteObjectQuery;
import org.eclipse.persistence.sessions.DatabaseRecord;
import org.eclipse.persistence.sessions.Project;

/**
 * <p><b>Purpose</b>:Two objects can be considered to be related by aggregation if there is a strict
 * 1:1 relationship between the objects. This means that the target (child or owned) object
 * cannot exist without the source (parent) object.
 *
 * In TopLink, it also means the data for the owned object is stored in the same table as
 * the parent.
 *
 * @author Sati
 * @since TOPLink/Java 1.0
 */
public class AggregateObjectMapping extends AggregateMapping implements RelationalMapping, MapKeyMapping, EmbeddableMapping {

    /**
     * If <em>all</em> the fields in the database row for the aggregate object are NULL,
     * then, by default, the mapping will place a null in the appropriate source object
     * (as opposed to an aggregate object filled with nulls).
     * To change this behavior, set the value of this variable to false. Then the mapping
     * will build a new instance of the aggregate object that is filled with nulls
     * and place it in the source object.
     */
    protected boolean isNullAllowed;

    protected DatabaseTable aggregateKeyTable  = null;

    /** Map the name of a field in the aggregate descriptor to a field in the source table. */
    /** 322233 - changed to store the source DatabaseField to hold Case and other colunm info*/
    protected Map<String, DatabaseField> aggregateToSourceFields;

    /**
     * Map of nested attributes that need to apply an override name to their
     * a nested aggregate mapping's database field. Aggregate to source fields
     * map is the existing EclipseLink functionality and works well when all
     * embeddable mappings have unique database fields. This map adds specific
     * attribute to database field override.
     * @see #addFieldTranslation
     */
    protected Map<String, Object[]> nestedFieldTranslations;

    /**
     * List of many to many mapping overrides to apply at initialize time to
     * their cloned aggregate mappings.
     */
    protected List<ManyToManyMapping> overrideManyToManyMappings;

    /**
     * List of unidirectional one to many mapping overrides to apply at
     * initialize time to their cloned aggregate mappings.
     */
    protected List<UnidirectionalOneToManyMapping> overrideUnidirectionalOneToManyMappings;

    /**
     * List of converters to apply at initialize time to their cloned aggregate mappings.
     */
    protected Map<String, Converter> converters;

    /**
     * List of maps id mappings that need to be set to read only at initialize
     * time on their cloned aggregate mappings.
     */
    protected List<DatabaseMapping> mapsIdMappings;

    /**
     * Default constructor.
     */
    public AggregateObjectMapping() {
        aggregateToSourceFields = new HashMap(5);
        nestedFieldTranslations = new HashMap<>();
        mapsIdMappings = new ArrayList<>();
        overrideManyToManyMappings = new ArrayList<>();
        overrideUnidirectionalOneToManyMappings = new ArrayList<>();
        converters = new HashMap<>();
        isNullAllowed = true;
    }

    /**
     * INTERNAL:
     */
    @Override
    public boolean isRelationalMapping() {
        return true;
    }

    /**
     * INTERNAL:
     * Used when initializing queries for mappings that use a Map
     * Called when the selection query is being initialized to add the fields for the map key to the query
     */
    @Override
    public void addAdditionalFieldsToQuery(ReadQuery selectionQuery, Expression baseExpression){
        for (DatabaseField field : getReferenceDescriptor().getAllFields()) {
             if (selectionQuery.isObjectLevelReadQuery()) {
                 ((ObjectLevelReadQuery)selectionQuery).addAdditionalField(baseExpression.getField(field));
             } else if (selectionQuery.isDataReadQuery()) {
                ((SQLSelectStatement) selectionQuery.getSQLStatement()).addField(baseExpression.getField(field));
            }
        }
    }

    /**
     * Add a converter to be applied to a mapping of the aggregate descriptor.
     */
    @Override
    public void addConverter(Converter converter, String attributeName) {
        converters.put(attributeName, converter);
    }

    /**
     * INTERNAL:
     * Used when initializing queries for mappings that use a Map
     * Called when the insert query is being initialized to ensure the fields for the map key are in the insert query
     */
    @Override
    public void addFieldsForMapKey(AbstractRecord joinRow){
        for (DatabaseMapping mapping : getReferenceDescriptor().getMappings()) {
            if (!mapping.isReadOnly()) {
                for (DatabaseField field : mapping.getFields()) {
                    if (field.isUpdatable()){
                        joinRow.put(field, null);
                    }
                }
            }
        }
    }

    /**
     * PUBLIC:
     * Add a field name translation that maps from a field name in the
     * source table to a field name in the aggregate descriptor.
     */
    public void addFieldNameTranslation(String sourceFieldName, String aggregateFieldName) {
        // 322233 - changed to store the sourceField instead of sourceFieldName
        addFieldTranslation(new DatabaseField(sourceFieldName), aggregateFieldName);
    }

    /**
     * PUBLIC:
     * Add a field translation that maps from a field in the
     * source table to a field name in the aggregate descriptor.
     */
    @Override
    public void addFieldTranslation(DatabaseField sourceField, String aggregateFieldName) {
        //AggregateObjectMapping does not seem to support Aggregates on multiple tables
        String unQualifiedAggregateFieldName = aggregateFieldName.substring(aggregateFieldName.lastIndexOf('.') + 1);// -1 is returned for no ".".
        getAggregateToSourceFields().put(unQualifiedAggregateFieldName, sourceField);
    }

    /**
     * INTERNAL:
     * In JPA users may specify a maps id mapping on a shared embeddable
     * descriptor. These mappings need to be set to read-only at initialize
     * time, after the reference descriptor is cloned.
     */
    public void addMapsIdMapping(DatabaseMapping mapping) {
        mapsIdMappings.add(mapping);
    }

    /**
     * INTERNAL:
     * Add a nested field translation that maps from a field in the source table
     * to a field name in a nested aggregate descriptor. These are handled
     * slightly different that regular field translations in that they are
     * unique based on the attribute name. It solves the case where multiple
     * nested embeddables have mappings to similarly named default columns.
     */
    @Override
    public void addNestedFieldTranslation(String attributeName, DatabaseField sourceField, String aggregateFieldName) {
        // Aggregate field name is redundant here as we will look up the field
        // through the attribute name. This method signature is to  satisfy the
        // Embeddable interface. AggregateCollectionMapping uses the aggregate
        // field name.
        nestedFieldTranslations.put(attributeName, new Object[]{sourceField, aggregateFieldName});
    }

    /**
     * INTERNAL:
     * In JPA users may specify overrides to apply to a many to many mapping
     * on a shared embeddable descriptor. These settings are applied at
     * initialize time, after the reference descriptor is cloned.
     */
    @Override
    public void addOverrideManyToManyMapping(ManyToManyMapping mapping) {
        overrideManyToManyMappings.add(mapping);
    }

    /**
     * INTERNAL:
     * In JPA users may specify overrides to apply to a unidirectional one to
     * many mapping on a shared embeddable descriptor. These settings are
     * applied at initialize time, after the reference descriptor is cloned.
     */
    @Override
    public void addOverrideUnidirectionalOneToManyMapping(UnidirectionalOneToManyMapping mapping) {
        overrideUnidirectionalOneToManyMappings.add(mapping);
    }

    /**
     * INTERNAL:
     * For mappings used as MapKeys in MappedKeyContainerPolicy.  Add the target of this mapping to the deleted
     * objects list if necessary
     *
     * This method is used for removal of private owned relationships.
     * AggregateObjectMappings are dealt with in their parent delete, so this is a no-op.
     *
     * @param object
     * @param deletedObjects
     */
    @Override
    public void addKeyToDeletedObjectsList(Object object, Map deletedObjects){
    }

    /**
     * INTERNAL:
     * Return whether all the aggregate fields in the specified
     * row are NULL.
     */
    protected boolean allAggregateFieldsAreNull(AbstractRecord databaseRow) {
        Vector fields = getReferenceFields();
        int size = fields.size();
        for (int index = 0; index < size; index++) {
            DatabaseField field = (DatabaseField)fields.get(index);
            Object value = databaseRow.get(field);
            if (value != null) {
                return false;
            }
        }
        return true;
    }

    /**
     * PUBLIC:
     * If <em>all</em> the fields in the database row for the aggregate object are NULL,
     * then, by default, the mapping will place a null in the appropriate source object
     * (as opposed to an aggregate object filled with nulls). This behavior can be
     * explicitly set by calling #allowNull().
     * To change this behavior, call #dontAllowNull(). Then the mapping
     * will build a new instance of the aggregate object that is filled with nulls
     * and place it in the source object.
     * In either situation, when writing, the mapping will place a NULL in all the
     * fields in the database row for the aggregate object.
     *
     * Note: Any aggregate that has a relationship mapping automatically does not allow
     * null.
     */
    public void allowNull() {
        setIsNullAllowed(true);
    }

    /**
     * INTERNAL:
     * Return whether the query's backup object has an attribute
     * value of null.
     */
    protected boolean backupAttributeValueIsNull(WriteObjectQuery query) {
        if (query.getSession().isUnitOfWork()) {
            Object backupAttributeValue = getAttributeValueFromObject(query.getBackupClone());
            if (backupAttributeValue == null) {
                return true;
            }
        }
        return false;
    }

    /**
     * INTERNAL:
     * Clone and prepare the selection query as a nested batch read query.
     * This is used for nested batch reading.
     */
    public ObjectBuildingQuery prepareNestedQuery(ObjectBuildingQuery sourceQuery) {
        if (sourceQuery.isObjectLevelReadQuery()) {
            ObjectLevelReadQuery objectQuery = (ObjectLevelReadQuery)sourceQuery;
            ObjectLevelReadQuery nestedObjectQuery = objectQuery.getAggregateQuery(this);
            if (nestedObjectQuery != null) {
                return nestedObjectQuery;
            }
            nestedObjectQuery = objectQuery;
            String attributeName = getAttributeName();
            if ((objectQuery.isPartialAttribute(attributeName))) {
                // A nested query must be built to pass to the descriptor that looks like the real query execution would.
                nestedObjectQuery = (ObjectLevelReadQuery)objectQuery.clone();
                // Must cascade the nested partial/join expression and filter the nested ones.
                if (objectQuery.hasPartialAttributeExpressions()) {
                    nestedObjectQuery.setPartialAttributeExpressions(extractNestedExpressions(objectQuery.getPartialAttributeExpressions(), nestedObjectQuery.getExpressionBuilder()));
                }
            }
            if (objectQuery.isAttributeBatchRead(this.descriptor, attributeName)) {
                if (nestedObjectQuery == objectQuery) {
                    // A nested query must be built to pass to the descriptor that looks like the real query execution would.
                    nestedObjectQuery = (ObjectLevelReadQuery)nestedObjectQuery.clone();
                }
                // Must carry over properties for batching to work.
                nestedObjectQuery.setProperties(objectQuery.getProperties());
                // Computed nested batch attribute expressions.
                nestedObjectQuery.getBatchFetchPolicy().setAttributeExpressions(extractNestedExpressions(objectQuery.getBatchReadAttributeExpressions(), nestedObjectQuery.getExpressionBuilder()));
                nestedObjectQuery.computeBatchReadAttributes();
            }
            FetchGroup parentQueryFetchGroup = sourceQuery.getExecutionFetchGroup(this.descriptor);
            if (parentQueryFetchGroup != null) {
                if (nestedObjectQuery == objectQuery) {
                    // A nested query must be built to pass to the descriptor that looks like the real query execution would.
                    nestedObjectQuery = (ObjectLevelReadQuery)nestedObjectQuery.clone();
                }
                FetchGroup targetFetchGroup = parentQueryFetchGroup.getGroup(getAttributeName());
                if (targetFetchGroup != null && sourceQuery.getDescriptor().hasFetchGroupManager()) {
                    //if the parent object has a fetchgroup manager then aggregates can support a fetchgroup manager
                    nestedObjectQuery.setFetchGroup(targetFetchGroup);
                } else {
                    targetFetchGroup = null;
                    nestedObjectQuery.setFetchGroup(null);
                    nestedObjectQuery.setFetchGroupName(null);
                }
            }
            if (nestedObjectQuery != sourceQuery) {
                objectQuery.setAggregateQuery(this, nestedObjectQuery);
                return nestedObjectQuery;
            }
        }
        return sourceQuery;
    }

    /**
     * INTERNAL:
     * Build and return an aggregate object from the specified row.
     * If a null value is allowed and all the appropriate fields in the row are NULL, return a null.
     * If an aggregate is referenced by the target object, return it (maintain identity)
     * Otherwise, simply create a new aggregate object and return it.
     */
    public Object buildAggregateFromRow(AbstractRecord databaseRow, Object targetObject, CacheKey cacheKey, JoinedAttributeManager joinManager, ObjectBuildingQuery sourceQuery, boolean buildShallowOriginal, AbstractSession executionSession, boolean targetIsProtected) throws DatabaseException {
        if (databaseRow.hasSopObject()) {
            Object sopAggregate = getAttributeValueFromObject(databaseRow.getSopObject());
            if ((targetObject != null) && (targetObject != databaseRow.getSopObject())) {
                setAttributeValueInObject(targetObject, sopAggregate);
            }
            return sopAggregate;
        }

        // check for all NULLs
        if (isNullAllowed() && allAggregateFieldsAreNull(databaseRow)) {
            return null;
        }

        // maintain object identity (even if not refreshing) if target object references the aggregate
        // if aggregate is not referenced by the target object, construct a new aggregate
        Object aggregate = null;
        ClassDescriptor descriptor = getReferenceDescriptor();
        boolean refreshing = true;
        if (targetObject != null){
            if (descriptor.hasInheritance()) {
                Class newAggregateClass = descriptor.getInheritancePolicy().classFromRow(databaseRow, executionSession);
                descriptor = getReferenceDescriptor(newAggregateClass, executionSession);
                aggregate = getMatchingAttributeValueFromObject(databaseRow, targetObject, executionSession, descriptor);
                if ((aggregate != null) && (aggregate.getClass() != newAggregateClass)) {
                    // if the class has changed out from underneath us, we cannot preserve object identity
                    // build a new instance of the *new* class
                    aggregate = descriptor.getObjectBuilder().buildNewInstance();
                    refreshing = false;
                }
            } else {
                aggregate = getMatchingAttributeValueFromObject(databaseRow, targetObject, executionSession, descriptor);
            }
        }

        // Build a new aggregate if the target object does not reference an existing aggregate.
        // EL Bug 474956 - build a new aggregate if the the target object references an existing aggregate, and 
        // the passed cacheKey is null from the invalidation of the target object in the IdentityMap.
        if (aggregate == null || (aggregate != null && cacheKey == null)) {
            aggregate = descriptor.getObjectBuilder().buildNewInstance();
            refreshing = false;
        }

        ObjectBuildingQuery nestedQuery = prepareNestedQuery(sourceQuery);
        FetchGroup targetFetchGroup =  null;
        if (nestedQuery.isObjectLevelReadQuery()) {
            targetFetchGroup = ((ObjectLevelReadQuery)nestedQuery).getFetchGroup();
            if (refreshing && descriptor.hasFetchGroupManager()) {
                descriptor.getFetchGroupManager().unionEntityFetchGroupIntoObject(aggregate, descriptor.getFetchGroupManager().getEntityFetchGroup(targetFetchGroup), executionSession, true);
                //merge fetchgroup into aggregate fetchgroup that may have been there from previous read.
            }
        }
        if (buildShallowOriginal) {
            descriptor.getObjectBuilder().buildAttributesIntoShallowObject(aggregate, databaseRow, nestedQuery);
        } else if (executionSession.isUnitOfWork()) {
            descriptor.getObjectBuilder().buildAttributesIntoWorkingCopyClone(aggregate, buildWrapperCacheKeyForAggregate(cacheKey, targetIsProtected), nestedQuery, joinManager, databaseRow, (UnitOfWorkImpl)executionSession, refreshing);
        } else {
            descriptor.getObjectBuilder().buildAttributesIntoObject(aggregate, buildWrapperCacheKeyForAggregate(cacheKey, targetIsProtected), databaseRow, nestedQuery, joinManager, nestedQuery.getExecutionFetchGroup(descriptor), refreshing, executionSession);
        }
        if ((targetFetchGroup != null) && descriptor.hasFetchGroupManager() && cacheKey != null
                && !refreshing && sourceQuery.shouldMaintainCache() && !sourceQuery.shouldStoreBypassCache()) {
            // Set the fetch group to the domain object, after built.
            EntityFetchGroup entityFetchGroup = descriptor.getFetchGroupManager().getEntityFetchGroup(targetFetchGroup);
            if (entityFetchGroup != null) {
                entityFetchGroup = (EntityFetchGroup)entityFetchGroup.clone();
                entityFetchGroup.setRootEntity((FetchGroupTracker) cacheKey.getObject());
                entityFetchGroup.setOnEntity(aggregate, executionSession);
            }
        }
        return aggregate;
    }

    /**
     * INTERNAL:
     * Wrap the aggregate represented by this mapping in a cachekey so it can be processed my
     * methods down the stack.
     * @param owningCacheKey - the cache key holding the object to extract the aggregate from
     * @return
     */
    protected CacheKey buildWrapperCacheKeyForAggregate(CacheKey owningCacheKey, boolean targetIsProtected) {
        if (!this.descriptor.getCachePolicy().isProtectedIsolation()) {
            return owningCacheKey;
        }
        if (!targetIsProtected || this.isMapKeyMapping || (owningCacheKey == null)) {
            return owningCacheKey;
        }
        CacheKey aggregateKey = owningCacheKey;
        Object object = owningCacheKey.getObject();
        if (owningCacheKey.getObject() != null) {
            Object aggregate = getAttributeValueFromObject(object);
            aggregateKey = new CacheKey(null, aggregate, null);
            aggregateKey.setProtectedForeignKeys(owningCacheKey.getProtectedForeignKeys());
            aggregateKey.setRecord(owningCacheKey.getRecord());
            aggregateKey.setIsolated(owningCacheKey.isIsolated());
            aggregateKey.setReadTime(owningCacheKey.getReadTime());
        }
        return aggregateKey;
    }

    /**
     * INTERNAL:
     * Write null values for all aggregate fields into the parent row.
     */
    protected void writeNullReferenceRow(AbstractRecord record) {
        List<DatabaseField> fields = getReferenceFields();
        int size = fields.size();
        boolean nullInserted = false;
        for (int index = 0; index < size; index++) {
            DatabaseField field = fields.get(index);
            // EL Bug 393520
            if (!field.isReadOnly() && (field.isUpdatable() || field.isInsertable())) {
                record.put(field, null);
                nullInserted = true;
            }
        }
        if (size > 0 && nullInserted) {
            // EL Bug 319759 - if a field is null, then the update call cache should not be used
            record.setNullValueInFields(true);
        }
    }

    /**
     * INTERNAL:
     * Used to allow object level comparisons.
     * In the case of an Aggregate which has no primary key must do an attribute
     * by attribute comparison.
     */
    @Override
    public Expression buildObjectJoinExpression(Expression expression, Object value, AbstractSession session) {
        Expression attributeByAttributeComparison = null;
        Expression join = null;
        Object attributeValue = null;

        // value need not be unwrapped as it is an aggregate, nor should it
        // influence a call to getReferenceDescriptor.
        ClassDescriptor referenceDescriptor = getReferenceDescriptor();
        if ((value != null) && !referenceDescriptor.getJavaClass().isInstance(value)) {
            throw QueryException.incorrectClassForObjectComparison(expression, value, this);
        }
        Enumeration mappings = referenceDescriptor.getMappings().elements();
        for (; mappings.hasMoreElements();) {
            DatabaseMapping mapping = (DatabaseMapping)mappings.nextElement();
            if (value == null) {
                attributeValue = null;
            } else {
                attributeValue = mapping.getAttributeValueFromObject(value);
            }
            join = expression.get(mapping.getAttributeName()).equal(attributeValue);
            if (attributeByAttributeComparison == null) {
                attributeByAttributeComparison = join;
            } else {
                attributeByAttributeComparison = attributeByAttributeComparison.and(join);
            }
        }
        return attributeByAttributeComparison;
    }

    /**
     * INTERNAL:
     * Used to allow object level comparisons.
     */
    @Override
    public Expression buildObjectJoinExpression(Expression expression, Expression argument, AbstractSession session) {
        Expression attributeByAttributeComparison = null;

        //Enumeration mappingsEnum = getSourceToTargetKeyFields().elements();
        Enumeration mappingsEnum = getReferenceDescriptor().getMappings().elements();
        for (; mappingsEnum.hasMoreElements();) {
            DatabaseMapping mapping = (DatabaseMapping)mappingsEnum.nextElement();
            String attributeName = mapping.getAttributeName();
            Expression join = expression.get(attributeName).equal(argument.get(attributeName));
            if (attributeByAttributeComparison == null) {
                attributeByAttributeComparison = join;
            } else {
                attributeByAttributeComparison = attributeByAttributeComparison.and(join);
            }
        }
        return attributeByAttributeComparison;
    }

    /**
     * INTERNAL:
     * Write the aggregate values into the parent row.
     */
    protected void writeToRowFromAggregate(AbstractRecord record, Object object, Object attributeValue, AbstractSession session, WriteType writeType) throws DescriptorException {
        if (attributeValue == null) {
            if (this.isNullAllowed) {
                writeNullReferenceRow(record);
            } else {
                throw DescriptorException.nullForNonNullAggregate(object, this);
            }
        } else {
            if (!session.isClassReadOnly(attributeValue.getClass())) {
                getObjectBuilder(attributeValue, session).buildRow(record, attributeValue, session, writeType);
            }
        }
    }

    /**
     * INTERNAL:
     * Write the aggregate values into the parent row for shallow insert.
     */
    protected void writeToRowFromAggregateForShallowInsert(AbstractRecord record, Object object, Object attributeValue, AbstractSession session) throws DescriptorException {
        if (attributeValue == null) {
            if (this.isNullAllowed) {
                writeNullReferenceRow(record);
            } else {
                throw DescriptorException.nullForNonNullAggregate(object, this);
            }
        } else {
            if (!session.isClassReadOnly(attributeValue.getClass())) {
                getObjectBuilder(attributeValue, session).buildRowForShallowInsert(record, attributeValue, session);
            }
        }
    }

    /**
     * INTERNAL:
     * Write the aggregate values into the parent row for update after shallow insert.
     */
    protected void writeToRowFromAggregateForUpdateAfterShallowInsert(AbstractRecord record, Object object, Object attributeValue, AbstractSession session, DatabaseTable table) throws DescriptorException {
        if (attributeValue == null) {
            if (!this.isNullAllowed) {
                throw DescriptorException.nullForNonNullAggregate(object, this);
            }
        } else {
            if (!session.isClassReadOnly(attributeValue.getClass()) && !isPrimaryKeyMapping()) {
                getObjectBuilder(attributeValue, session).buildRowForUpdateAfterShallowInsert(record, attributeValue, session, table);
            }
        }
    }

    /**
     * INTERNAL:
     * Write the aggregate values into the parent row for update before shallow delete.
     */
    protected void writeToRowFromAggregateForUpdateBeforeShallowDelete(AbstractRecord record, Object object, Object attributeValue, AbstractSession session, DatabaseTable table) throws DescriptorException {
        if (attributeValue == null) {
            if (!this.isNullAllowed) {
                throw DescriptorException.nullForNonNullAggregate(object, this);
            }
        } else {
            if (!session.isClassReadOnly(attributeValue.getClass()) && !isPrimaryKeyMapping()) {
                getObjectBuilder(attributeValue, session).buildRowForUpdateBeforeShallowDelete(record, attributeValue, session, table);
            }
        }
    }

    /**
     * INTERNAL:
     * Build and return a database row built with the values from
     * the specified attribute value.
     */
    protected void writeToRowFromAggregateWithChangeRecord(AbstractRecord record, ChangeRecord changeRecord, ObjectChangeSet objectChangeSet, AbstractSession session, WriteType writeType) throws DescriptorException {
        if (objectChangeSet == null) {
            if (this.isNullAllowed) {
                writeNullReferenceRow(record);
            } else {
                Object object = ((ObjectChangeSet)changeRecord.getOwner()).getUnitOfWorkClone();
                throw DescriptorException.nullForNonNullAggregate(object, this);
            }
        } else {
            if (!session.isClassReadOnly(objectChangeSet.getClassType(session))) {
                getReferenceDescriptor(objectChangeSet.getClassType(session), session).getObjectBuilder().buildRowWithChangeSet(record, objectChangeSet, session, writeType);
            }
        }
    }

    /**
     * INTERNAL:
     * Build and return a database row built with the changed values from
     * the specified attribute value.
     */
    protected void writeToRowFromAggregateForUpdate(AbstractRecord record, WriteObjectQuery query, Object attributeValue) throws DescriptorException {
        if (attributeValue == null) {
            if (this.isNullAllowed) {
                if (backupAttributeValueIsNull(query)) {
                    // both attributes are null - no update required
                } else {
                    writeNullReferenceRow(record);
                }
            } else {
                throw DescriptorException.nullForNonNullAggregate(query.getObject(), this);
            }
        } else if ((query.getBackupClone() != null) && ((getMatchingBackupAttributeValue(query, attributeValue) == null) || !(attributeValue.getClass().equals(getMatchingBackupAttributeValue(query, attributeValue).getClass())))) {
            getObjectBuilder(attributeValue, query.getSession()).buildRow(record, attributeValue, query.getSession(), WriteType.UPDATE);
        } else {
            if (!query.getSession().isClassReadOnly(attributeValue.getClass())) {
                WriteObjectQuery clonedQuery = (WriteObjectQuery)query.clone();
                clonedQuery.setObject(attributeValue);
                if (query.getSession().isUnitOfWork()) {
                    Object backupAttributeValue = getMatchingBackupAttributeValue(query, attributeValue);
                    if (backupAttributeValue == null) {
                        backupAttributeValue = getObjectBuilder(attributeValue, query.getSession()).buildNewInstance();
                    }
                    clonedQuery.setBackupClone(backupAttributeValue);
                }
                getObjectBuilder(attributeValue, query.getSession()).buildRowForUpdate(record, clonedQuery);
            }
        }
    }

    /**
     * INTERNAL:
     * Clone the attribute from the original and assign it to the clone.
     */
    @Override
    public void buildClone(Object original, CacheKey cacheKey, Object clone, Integer refreshCascade, AbstractSession cloningSession) {
        Object attributeValue = getAttributeValueFromObject(original);
        Object aggregateClone = buildClonePart(original, clone, cacheKey, attributeValue, refreshCascade, cloningSession);

        if (aggregateClone != null && cloningSession.isUnitOfWork()) {
            ClassDescriptor descriptor = getReferenceDescriptor(aggregateClone, cloningSession);
            descriptor.getObjectChangePolicy().setAggregateChangeListener(clone, aggregateClone, (UnitOfWorkImpl)cloningSession, descriptor, getAttributeName());
        }

        setAttributeValueInObject(clone, aggregateClone);
    }

    /**
     * INTERNAL:
     * Build a clone of the given element in a unitOfWork
     * @param attributeValue
     * @param parent
     * @param parentCacheKey
     * @param refreshCascade
     * @param cloningSession
     * @param isExisting
     * @param isFromSharedCache
     * @return
     */
    @Override
    public Object buildElementClone(Object attributeValue, Object parent, CacheKey parentCacheKey, Integer refreshCascade, AbstractSession cloningSession, boolean isExisting, boolean isFromSharedCache){
        Object aggregateClone = buildClonePart(attributeValue, parent, parentCacheKey, refreshCascade, cloningSession, !isExisting);
        if (aggregateClone != null && cloningSession.isUnitOfWork()) {
            ClassDescriptor descriptor = getReferenceDescriptor(aggregateClone, cloningSession);
            descriptor.getObjectChangePolicy().setAggregateChangeListener(parent, aggregateClone, (UnitOfWorkImpl)cloningSession, descriptor, getAttributeName());
        }
        return aggregateClone;
    }

    /**
     * INTERNAL:
     * Set the change listener in the aggregate.
     */
    @Override
    public void setChangeListener(Object clone, PropertyChangeListener listener, UnitOfWorkImpl uow) {
        Object attributeValue = getAttributeValueFromObject(clone);
        if (attributeValue != null) {
            ClassDescriptor descriptor = getReferenceDescriptor(attributeValue, uow);
            descriptor.getObjectChangePolicy().setAggregateChangeListener(clone, attributeValue, uow, descriptor, getAttributeName());
        }
    }

    /**
     * INTERNAL:
     * A combination of readFromRowIntoObject and buildClone.
     * <p>
     * buildClone assumes the attribute value exists on the original and can
     * simply be copied.
     * <p>
     * readFromRowIntoObject assumes that one is building an original.
     * <p>
     * Both of the above assumptions are false in this method, and actually
     * attempts to do both at the same time.
     * <p>
     * Extract value from the row and set the attribute to this value in the
     * working copy clone.
     * In order to bypass the shared cache when in transaction a UnitOfWork must
     * be able to populate working copies directly from the row.
     */
    @Override
    public void buildCloneFromRow(AbstractRecord databaseRow, JoinedAttributeManager joinManager, Object clone, CacheKey sharedCacheKey, ObjectBuildingQuery sourceQuery, UnitOfWorkImpl unitOfWork, AbstractSession executionSession) {
        // This method is a combination of buildggregateFromRow and buildClonePart on the super class.
        // None of buildClonePart used, as not an orignal new object, nor do we worry about creating heavy clones for aggregate objects.
        // Ensure that the shared CacheKey is passed, as this will be set to null for a refresh of an invalid object.
        Object clonedAttributeValue = buildAggregateFromRow(databaseRow, clone, sharedCacheKey, joinManager, sourceQuery, false, executionSession, true);
        if (clonedAttributeValue != null) {
            ClassDescriptor descriptor = getReferenceDescriptor(clonedAttributeValue, unitOfWork);
            descriptor.getObjectChangePolicy().setAggregateChangeListener(clone, clonedAttributeValue, unitOfWork, descriptor, getAttributeName());
        }
        setAttributeValueInObject(clone, clonedAttributeValue);
    }

    /**
     * INTERNAL:
     * Builds a shallow original object.  Only direct attributes and primary
     * keys are populated.  In this way the minimum original required for
     * instantiating a working copy clone can be built without placing it in
     * the shared cache (no concern over cycles).
     */
    @Override
    public void buildShallowOriginalFromRow(AbstractRecord databaseRow, Object original, JoinedAttributeManager joinManager, ObjectBuildingQuery sourceQuery, AbstractSession executionSession) {
        Object aggregate = buildAggregateFromRow(databaseRow, original, null, joinManager, sourceQuery, true, executionSession, true);// shallow only.
        setAttributeValueInObject(original, aggregate);
    }

    /**
     * INTERNAL:
     * Certain key mappings favor different types of selection query.  Return the appropriate
     * type of selectionQuery
     * @return
     */
    @Override
    public ReadQuery buildSelectionQueryForDirectCollectionKeyMapping(ContainerPolicy containerPolicy){
        ReadAllQuery query = new ReadAllQuery();
        query.setReferenceClass(referenceClass);
        query.setDescriptor(getReferenceDescriptor());
        query.setContainerPolicy(containerPolicy);
        return query;
    }

    /**
     * INTERNAL:
     * Build and return a "template" database row with all the fields
     * set to null.
     */
    protected AbstractRecord buildTemplateInsertRow(AbstractSession session) {
        AbstractRecord result = getReferenceDescriptor().getObjectBuilder().buildTemplateInsertRow(session);
        List processedMappings = (List)getReferenceDescriptor().getMappings().clone();
        if (getReferenceDescriptor().hasInheritance()) {
            for (ClassDescriptor child : getReferenceDescriptor().getInheritancePolicy().getChildDescriptors()) {
                for (DatabaseMapping mapping : child.getMappings()) {
                    // Only write mappings once.
                    if (!processedMappings.contains(mapping)) {
                        mapping.writeInsertFieldsIntoRow(result, session);
                        processedMappings.add(mapping);
                    }
                }
            }
        }
        return result;
    }

    /**
     * INTERNAL:
     * Cascade discover and persist new objects during commit to the map key
     */
    @Override
    public void cascadeDiscoverAndPersistUnregisteredNewObjects(Object object, Map newObjects, Map unregisteredExistingObjects, Map visitedObjects, UnitOfWorkImpl uow, boolean getAttributeValueFromObject, Set cascadeErrors){
        ObjectBuilder builder = getReferenceDescriptor(object.getClass(), uow).getObjectBuilder();
        builder.cascadeDiscoverAndPersistUnregisteredNewObjects(object, newObjects, unregisteredExistingObjects, visitedObjects, uow, cascadeErrors);
    }

    /**
     * INTERNAL:
     * Cascade perform delete through mappings that require the cascade
     */
    @Override
    public void cascadePerformRemoveIfRequired(Object object, UnitOfWorkImpl uow, Map visitedObjects, boolean getAttributeValueFromObject) {
        Object objectReferenced = null;
        if (getAttributeValueFromObject){
            //objects referenced by this mapping are not registered as they have
            // no identity, however mappings from the referenced object may need cascading.
            objectReferenced = getAttributeValueFromObject(object);
        } else {
            objectReferenced = object;
        }
        if ((objectReferenced == null)) {
            return;
        }
        if (!visitedObjects.containsKey(objectReferenced)) {
            visitedObjects.put(objectReferenced, objectReferenced);
            ObjectBuilder builder = getReferenceDescriptor(objectReferenced.getClass(), uow).getObjectBuilder();
            builder.cascadePerformRemove(objectReferenced, uow, visitedObjects);
        }
    }

    /**
     * INTERNAL:
     * Cascade perform delete through mappings that require the cascade
     */
    @Override
    public void cascadePerformRemoveIfRequired(Object object, UnitOfWorkImpl uow, Map visitedObjects) {
        cascadePerformRemoveIfRequired(object, uow, visitedObjects, true);
    }

    /**
     * INTERNAL:
     * Cascade perform removal of orphaned private owned objects from the UnitOfWorkChangeSet
     */
    @Override
    public void cascadePerformRemovePrivateOwnedObjectFromChangeSetIfRequired(Object object, UnitOfWorkImpl uow, Map visitedObjects) {
        Object attributeValue = getAttributeValueFromObject(object);
        if (attributeValue == null) {
            return;
        }
        if (!visitedObjects.containsKey(attributeValue)) {
            visitedObjects.put(attributeValue, attributeValue);
            ObjectBuilder builder = getReferenceDescriptor(attributeValue, uow).getObjectBuilder();
            // cascade perform remove any related objects via ObjectBuilder for an aggregate object
            builder.cascadePerformRemovePrivateOwnedObjectFromChangeSet(attributeValue, uow, visitedObjects);
        }
    }

    /**
     * INTERNAL:
     * Cascade registerNew for Create through mappings that require the cascade
     */
    @Override
    public void cascadeRegisterNewIfRequired(Object object, UnitOfWorkImpl uow, Map visitedObjects, boolean getAttributeValueFromObject) {
        Object objectReferenced = null;
        //aggregate objects are not registered but their mappings should be.
        if (getAttributeValueFromObject){
            objectReferenced = getAttributeValueFromObject(object);
        } else {
            objectReferenced = object;
        }
        if ((objectReferenced == null)) {
            return;
        }
        if (!visitedObjects.containsKey(objectReferenced)) {
            visitedObjects.put(objectReferenced, objectReferenced);
            ObjectBuilder builder = getReferenceDescriptor(objectReferenced.getClass(), uow).getObjectBuilder();
            builder.cascadeRegisterNewForCreate(objectReferenced, uow, visitedObjects);
        }
    }

    /**
     * INTERNAL:
     * Cascade registerNew for Create through mappings that require the cascade
     */
    @Override
    public void cascadeRegisterNewIfRequired(Object object, UnitOfWorkImpl uow, Map visitedObjects) {
        cascadeRegisterNewIfRequired(object, uow, visitedObjects, true);
    }

    /**
     * INTERNAL:
     * Clone the aggregate to source field names. AggregateCollectionMapping
     * needs each nested embedded mapping to have its own  list of aggregate
     * to source field names so that it can apply nested override names to
     * shared aggregate object mappings.
     */
    @Override
    public Object clone() {
        AggregateObjectMapping mappingObject = (AggregateObjectMapping) super.clone();

        Map<String, DatabaseField> aggregateToSourceFields = new HashMap<>();
        aggregateToSourceFields.putAll(getAggregateToSourceFields());
        mappingObject.setAggregateToSourceFields(aggregateToSourceFields);

        Map<String, Object[]> nestedTranslations = new HashMap<>();
        nestedTranslations.putAll(getNestedFieldTranslations());
        mappingObject.setNestedFieldTranslations(nestedTranslations);

        return mappingObject;
    }

    /**
     * INTERNAL:
     * Return the fields handled by the mapping.
     */
    @Override
    protected Vector<DatabaseField> collectFields() {
        return getReferenceFields();
    }

    /**
     * INTERNAL:
     * Aggregate order by all their fields by default.
     */
    @Override
    public List<Expression> getOrderByNormalizedExpressions(Expression base) {
        List<Expression> orderBys = new ArrayList(this.fields.size());
        for (DatabaseField field : this.fields) {
            orderBys.add(base.getField(field));
        }
        return orderBys;
    }

    /**
     * INTERNAL:
     * This method is used to store the FK fields that can be cached that correspond to noncacheable mappings
     * the FK field values will be used to re-issue the query when cloning the shared cache entity
     */
    @Override
    public void collectQueryParameters(Set<DatabaseField> record){
        for (DatabaseMapping mapping : getReferenceDescriptor().getMappings()){
            if ((mapping.isForeignReferenceMapping() && !mapping.isCacheable()) || (mapping.isAggregateObjectMapping() && mapping.getReferenceDescriptor().hasNoncacheableMappings())){
                mapping.collectQueryParameters(record);
            }
        }
    }

    /**
     * INTERNAL:
     * Convert all the class-name-based settings in this mapping to actual
     * class-based settings. This method is used when converting a project that
     * has been built with class names to a project with classes.
     * @param classLoader
     */
    @Override
    public void convertClassNamesToClasses(ClassLoader classLoader) {
        super.convertClassNamesToClasses(classLoader);

        for (Converter converter : converters.values()) {
            // Convert and any Converter class names.
            convertConverterClassNamesToClasses(converter, classLoader);
        }
    }

    /**
     * INTERNAL
     * Called when a DatabaseMapping is used to map the key in a collection.  Returns the key.
     */
    @Override
    public Object createMapComponentFromRow(AbstractRecord dbRow, ObjectBuildingQuery query, CacheKey parentCacheKey, AbstractSession session, boolean isTargetProtected){
        Object key = buildAggregateFromRow(dbRow, null, parentCacheKey, null, query, false, session, isTargetProtected);
        return key;
    }

    /**
     * INTERNAL:
     * Creates the Array of simple types used to recreate this map.
     */
    @Override
    public Object createSerializableMapKeyInfo(Object key, AbstractSession session){
        return key; // Embeddables have no identity so they are not reduced to PK.
    }

    /**
     * INTERNAL:
     * Create an instance of the Key object from the key information extracted from the map.
     * This may return the value directly in case of a simple key or will be used as the FK to load a related entity.
     */
    @Override
    public List<Object> createMapComponentsFromSerializableKeyInfo(Object[] keyInfo, AbstractSession session){
        return Arrays.asList(keyInfo); // Embeddables have no identity so they are not reduced to PK.
    }

    /**
     * INTERNAL:
     * Create an instance of the Key object from the key information extracted from the map.
     * This key object may be a shallow stub of the actual object if the key is an Entity type.
     */
    @Override
    public Object createStubbedMapComponentFromSerializableKeyInfo(Object keyInfo, AbstractSession session){
        return keyInfo;
    }

    /**
     * INTERNAL
     * Called when a DatabaseMapping is used to map the key in a collection and a join query is executed.  Returns the key.
     */
    @Override
    public Object createMapComponentFromJoinedRow(AbstractRecord dbRow, JoinedAttributeManager joinManger, ObjectBuildingQuery query, CacheKey parentCacheKey, AbstractSession session, boolean isTargetProtected){
        return createMapComponentFromRow(dbRow, query, parentCacheKey, session, isTargetProtected);
    }

    /**
     * INTERNAL:
     * Create a query key that links to the map key
     * @return
     */
    @Override
    public QueryKey createQueryKeyForMapKey(){
        return null;
    }

    /**
     * INTERNAL:
     * For mappings used as MapKeys in MappedKeyContainerPolicy, Delete the passed object if necessary.
     *
     * This method is used for removal of private owned relationships.
     * AggregateObjectMappings are dealt with in their parent delete, so this is a no-op.
     *
     * @param objectDeleted
     * @param session
     */
    @Override
    public void deleteMapKey(Object objectDeleted, AbstractSession session){
    }

    /**
     * PUBLIC:
     * If <em>all</em> the fields in the database row for the aggregate object are NULL,
     * then, by default, the mapping will place a null in the appropriate source object
     * (as opposed to an aggregate object filled with nulls). This behavior can be
     * explicitly set by calling #allowNull().
     * To change this behavior, call #dontAllowNull(). Then the mapping
     * will build a new instance of the aggregate object that is filled with nulls
     * and place it in the source object.
     * In either situation, when writing, the mapping will place a NULL in all the
     * fields in the database row for the aggregate object.
     *
     * Note: Any aggregate that has a relationship mapping automatically does not allow
     * null.
     */
    public void dontAllowNull() {
        setIsNullAllowed(false);
    }

    /**
     * INTERNAL:
     * This method is called to update collection tables prior to commit.
     */
    @Override
    public void earlyPreDelete(DeleteObjectQuery query, Object object) {
        // need to go through our reference's pre-delete mappings
        for (DatabaseMapping mapping : getReferenceDescriptor().getPreDeleteMappings()) {
            Object nestedObject = getRealAttributeValueFromObject(object, query.getSession());

            // If we have an aggregate object, go through the pre-delete.
            if (nestedObject != null) {
                mapping.earlyPreDelete(query, nestedObject);
            }
        }
    }

    /**
     * INTERNAL:
     * Extract the fields for the Map key from the object to use in a query.
     */
    @Override
    public Map extractIdentityFieldsForQuery(Object object, AbstractSession session){
        Map keyFields = new HashMap();
        ClassDescriptor descriptor =getReferenceDescriptor();
        boolean usePrimaryKeyFields = descriptor.getPrimaryKeyFields() != null && ! descriptor.getPrimaryKeyFields().isEmpty();
        Iterator <DatabaseMapping> i = descriptor.getMappings().iterator();
        while (i.hasNext()){
            DatabaseMapping mapping = i.next();
            if (!mapping.isReadOnly() && (!usePrimaryKeyFields || (usePrimaryKeyFields && mapping.isPrimaryKeyMapping()))){
                Iterator<DatabaseField> fields = mapping.getFields().iterator();
                while (fields.hasNext()){
                    DatabaseField field = fields.next();
                    if (field.isUpdatable()){
                        Object value = descriptor.getObjectBuilder().extractValueFromObjectForField(object, field, session);
                        keyFields.put(field, value);
                    }
                }
            }
        }

        return keyFields;
    }

    /**
     * INTERNAL:
     * Return any tables that will be required when this mapping is used as part of a join query
     * @return
     */
    @Override
    public List<DatabaseTable> getAdditionalTablesForJoinQuery(){
        return getReferenceDescriptor().getTables();
    }

    /**
     * INTERNAL:
     * Return the selection criteria necessary to select the target object when this mapping
     * is a map key.
     *
     * AggregateObjectMappings do not need any additional selection criteria when they are map keys
     * @return
     */
    @Override
    public Expression getAdditionalSelectionCriteriaForMapKey(){
        return null;
    }

    /**
     * INTERNAL:
     * Return a collection of the aggregate to source field  associations.
     */
    public Vector<Association> getAggregateToSourceFieldAssociations() {
        Vector<Association> associations = new Vector(getAggregateToSourceFields().size());
        Iterator aggregateEnum = getAggregateToSourceFields().keySet().iterator();
        Iterator sourceEnum = getAggregateToSourceFields().values().iterator();
        while (aggregateEnum.hasNext()) {
            associations.addElement(new Association(aggregateEnum.next(), sourceEnum.next()));
        }

        return associations;
    }

    /**
     * INTERNAL:
     * Return the hashtable that stores aggregate field name to source fields.
     */
    public Map<String, DatabaseField> getAggregateToSourceFields() {
        return aggregateToSourceFields;
    }

    /**
     * INTERNAL:
     * Return the hashtable that stores the nested field translations.
     */
    public Map<String, Object[]> getNestedFieldTranslations() {
        return nestedFieldTranslations;
    }

    /**
     * PUBLIC:
     * The classification type for the attribute this mapping represents
     */
    @Override
    public Class getAttributeClassification() {
        return getReferenceClass();
    }

    /**
     * INTERNAL:
     * Return the classification for the field contained in the mapping.
     * This is used to convert the row value to a consistent Java value.
     */
    @Override
    public Class getFieldClassification(DatabaseField fieldToClassify) {
        DatabaseMapping mapping = getReferenceDescriptor().getObjectBuilder().getMappingForField(fieldToClassify);
        if (mapping == null) {
            return null;// Means that the mapping is read-only
        }
        return mapping.getFieldClassification(fieldToClassify);
    }

    /**
     * INTERNAL:
     * Return the fields that make up the identity of the mapped object.  For mappings with
     * a primary key, it will be the set of fields in the primary key.  For mappings without
     * a primary key it will likely be all the fields
     * @return
     */
    @Override
    public List<DatabaseField> getIdentityFieldsForMapKey(){
        ClassDescriptor descriptor =getReferenceDescriptor();
        if (descriptor.getPrimaryKeyFields() != null){
            return descriptor.getPrimaryKeyFields();
        } else {
            return getAllFieldsForMapKey();
        }
    }

    /**
     * INTERNAL:
     * Get all the fields for the map key
     */
    @Override
    public List<DatabaseField> getAllFieldsForMapKey(){
        return getReferenceDescriptor().getAllFields();
    }


    /**
     * INTERNAL:
     * Return a Map of any foreign keys defined within the the MapKey
     * @return
     */
    @Override
    public Map<DatabaseField, DatabaseField> getForeignKeyFieldsForMapKey(){
        return null;
    }

    /**
     * INTERNAL:
     * This is used to preserve object identity during a refreshObject()
     * query. Return the object corresponding to the specified database row.
     * The default is to simply return the attribute value.
     */
    protected Object getMatchingAttributeValueFromObject(AbstractRecord row, Object targetObject, AbstractSession session, ClassDescriptor descriptor) {
        return getAttributeValueFromObject(targetObject);
    }

    /**
     * INTERNAL:
     * This is used to match up objects during an update in a UOW.
     * Return the object corresponding to the specified attribute value.
     * The default is to simply return the backup attribute value.
     */
    protected Object getMatchingBackupAttributeValue(WriteObjectQuery query, Object attributeValue) {
        return getAttributeValueFromObject(query.getBackupClone());
    }

    /**
     * INTERNAL:
     * Return the query that is used when this mapping is part of a joined relationship
     *
     * This method is used when this mapping is used to map the key in a Map
     * @return
     */
    @Override
    public ObjectLevelReadQuery getNestedJoinQuery(JoinedAttributeManager joinManager, ObjectLevelReadQuery query, AbstractSession session){
        return null;
    }

    /**
     * INTERNAL:
     * Since aggregate object mappings clone their descriptors, for inheritance the correct child clone must be found.
     */
    @Override
    public ClassDescriptor getReferenceDescriptor(Class theClass, AbstractSession session) {
        if (this.referenceDescriptor.getJavaClass() == theClass) {
            return this.referenceDescriptor;
        }

        ClassDescriptor subDescriptor = this.referenceDescriptor.getInheritancePolicy().getSubclassDescriptor(theClass);
        if (subDescriptor == null) {
            throw DescriptorException.noSubClassMatch(theClass, this);
        } else {
            return subDescriptor;
        }
    }


    /**
     * INTERNAL:
     * Return the fields used to build the aggregate object.
     */
    protected Vector<DatabaseField> getReferenceFields() {
        return getReferenceDescriptor().getAllFields();
    }
    /**
     * INTERNAL:
     * If required, get the targetVersion of the source object from the merge manager.
     *
     * Used with MapKeyContainerPolicy to abstract getting the target version of a source key
     * @return
     */
    @Override
    public Object getTargetVersionOfSourceObject(Object object, Object parent, MergeManager mergeManager, AbstractSession targetSession){
        if (mergeManager.getSession().isUnitOfWork()){
            UnitOfWorkImpl uow = (UnitOfWorkImpl)mergeManager.getSession();
            Object aggregateObject = buildClonePart(object, parent, null, null, targetSession, uow.isOriginalNewObject(parent));
            return aggregateObject;
        }
        return object;
    }

    /**
     * INTERNAL:
     * Return the class this key mapping maps or the descriptor for it
     * @return
     */
    @Override
    public Object getMapKeyTargetType(){
        return getReferenceDescriptor();
    }

    /**
     * INTERNAL:
     * Return if the mapping has any ownership or other dependency over its target object(s).
     */
    @Override
    public boolean hasDependency() {
        return getReferenceDescriptor().hasDependencyOnParts();
    }

    /**
     * INTERNAL:
     * For an aggregate mapping the reference descriptor is cloned. The cloned descriptor is then
     * assigned primary keys and table names before initialize. Once the cloned descriptor is initialized
     * it is assigned as reference descriptor in the aggregate mapping. This is a very specific
     * behavior for aggregate mappings. The original descriptor is used only for creating clones and
     * after that the aggregate mapping never uses it.
     * Some initialization is done in postInitialize to ensure the target descriptor's references are initialized.
     */
    @Override
    public void initialize(AbstractSession session) throws DescriptorException {
        AbstractSession referenceSession = session;
        if( session.hasBroker()) {
            if (getReferenceClass() == null) {
                throw DescriptorException.referenceClassNotSpecified(this);
            }
            referenceSession = session.getSessionForClass(getReferenceClass());
        }
        super.initialize(session);

        ClassDescriptor clonedDescriptor = (ClassDescriptor)getReferenceDescriptor().clone();

        List<AttributeAccessor> accessorTree = getDescriptor().getAccessorTree();
        if (accessorTree == null){
            accessorTree = new ArrayList();
        }else{
            accessorTree = new ArrayList<>(accessorTree);
        }
        accessorTree.add(getAttributeAccessor());
        clonedDescriptor.setAccessorTree(accessorTree);
        if (isMapKeyMapping() && clonedDescriptor.isAggregateDescriptor()){
            clonedDescriptor.descriptorIsAggregateCollection();
        }
        if (clonedDescriptor.isChildDescriptor()) {
            ClassDescriptor parentDescriptor = session.getDescriptor(clonedDescriptor.getInheritancePolicy().getParentClass());
            initializeParentInheritance(parentDescriptor, clonedDescriptor, session);
        }

        setReferenceDescriptor(clonedDescriptor);

        // Apply any override m2m mappings to their cloned mappings.
        for (ManyToManyMapping overrideMapping : overrideManyToManyMappings) {
            DatabaseMapping mapping = clonedDescriptor.getMappingForAttributeName(overrideMapping.getAttributeName());

            if (mapping.isManyToManyMapping()) {
                ManyToManyMapping mappingClone = (ManyToManyMapping) mapping;
                mappingClone.setRelationTable(overrideMapping.getRelationTable());
                mappingClone.setSourceKeyFields(overrideMapping.getSourceKeyFields());
                mappingClone.setSourceRelationKeyFields(overrideMapping.getSourceRelationKeyFields());
                mappingClone.setTargetKeyFields(overrideMapping.getTargetKeyFields());
                mappingClone.setTargetRelationKeyFields(overrideMapping.getTargetRelationKeyFields());
            }

            // Else, silently ignored for now. These override mappings are set
            // and controlled through JPA metadata processing.
        }

        // Apply any override uni-directional 12m mappings to their cloned mappings.
        for (UnidirectionalOneToManyMapping overrideMapping : overrideUnidirectionalOneToManyMappings) {
            DatabaseMapping mapping = clonedDescriptor.getMappingForAttributeName(overrideMapping.getAttributeName());

            if (mapping.isUnidirectionalOneToManyMapping()) {
                UnidirectionalOneToManyMapping mappingClone = (UnidirectionalOneToManyMapping) mapping;
                mappingClone.setSourceKeyFields(overrideMapping.getSourceKeyFields());
                mappingClone.setTargetForeignKeyFields(overrideMapping.getTargetForeignKeyFields());
            }

            // Else, silently ignored for now. These override mappings are set
            // and controlled through JPA metadata processing.
        }

        // Mark any mapsId mappings as read-only.
        for (DatabaseMapping mapsIdMapping : mapsIdMappings) {
            DatabaseMapping mapping = clonedDescriptor.getMappingForAttributeName(mapsIdMapping.getAttributeName());

            if (mapping != null) {
                mapping.setIsReadOnly(true);
            }

            // Else, silently ignored for now. Maps id mappings are set and
            // controlled through JPA metadata processing.
        }

        // disallow null for aggregates with target foreign key relationships
        if (isNullAllowed) {
            if (getReferenceDescriptor().hasTargetForeignKeyMapping(session)) {
                isNullAllowed = false;
                session.log(SessionLog.WARNING, SessionLog.METADATA, "metadata_warning_ignore_is_null_allowed", new Object[]{this});
            }
        }

        initializeReferenceDescriptor(clonedDescriptor, referenceSession);
        //must translate before initializing because this mapping may have nested translations.
        translateNestedFields(clonedDescriptor, referenceSession);
        clonedDescriptor.preInitialize(referenceSession);
        clonedDescriptor.initialize(referenceSession);

        // Apply any converters to their cloned mappings (after initialization
        // so we can successfully traverse dot notation names)
        for (String attributeName : converters.keySet()) {
            String attr = attributeName;

            ClassDescriptor desc = clonedDescriptor;

            int idx;
            while ((idx = attr.indexOf('.')) > -1) {
                desc = desc.getMappingForAttributeName(attr.substring(0, idx)).getReferenceDescriptor();
                attr = attr.substring(idx + 1);
            }

            DatabaseMapping mapping = desc.getMappingForAttributeName(attr);

            if (mapping != null) {
                // Initialize and set the converter on the mapping.
                converters.get(attributeName).initialize(mapping, session);
            }

            // Else, silently ignored for now. These converters are set and
            // controlled through JPA metadata processing.
        }
        translateFields(clonedDescriptor, referenceSession);

        if (clonedDescriptor.hasInheritance() && clonedDescriptor.getInheritancePolicy().hasChildren()) {
            //clone child descriptors
            initializeChildInheritance(clonedDescriptor, referenceSession);
        }

        setFields(collectFields());

        // Add the nested pre delete mappings to the source entity.
        if (!isMapKeyMapping() && clonedDescriptor.hasPreDeleteMappings()) {
            getDescriptor().addPreDeleteMapping(this);
        }
    }

    /**
     * INTERNAL:
     * For an aggregate mapping the reference descriptor is cloned.
     * If the reference descriptor is involved in an inheritance tree,
     * all the parent and child descriptors are cloned also.
     * The cloned descriptors are then assigned primary keys and
     * table names before initialize.
     * This is a very specific behavior for aggregate mappings.
     */
    public void initializeChildInheritance(ClassDescriptor parentDescriptor, AbstractSession session) throws DescriptorException {
        //recursive call to the further children descriptors
        if (parentDescriptor.getInheritancePolicy().hasChildren()) {
            //setFields(clonedChildDescriptor.getFields());
            List<ClassDescriptor> childDescriptors = parentDescriptor.getInheritancePolicy().getChildDescriptors();
            List<ClassDescriptor> cloneChildDescriptors = new ArrayList();
            for (ClassDescriptor child : childDescriptors) {
                ClassDescriptor clonedChildDescriptor = (ClassDescriptor)child.clone();
                clonedChildDescriptor.getInheritancePolicy().setParentDescriptor(parentDescriptor);
                initializeReferenceDescriptor(clonedChildDescriptor, session);
                clonedChildDescriptor.preInitialize(session);
                clonedChildDescriptor.initialize(session);
                translateFields(clonedChildDescriptor, session);
                cloneChildDescriptors.add(clonedChildDescriptor);
                initializeChildInheritance(clonedChildDescriptor, session);
            }
            parentDescriptor.getInheritancePolicy().setChildDescriptors(cloneChildDescriptors);
        }
    }

    /**
     * INTERNAL:
     * For an aggregate mapping the reference descriptor is cloned.
     * If the reference descriptor is involved in an inheritance tree,
     * all the parent and child descriptors are cloned also.
     * The cloned descriptors are then assigned primary keys and
     * table names before initialize.
     * This is a very specific behavior for aggregate mappings.
     */
    public void initializeParentInheritance(ClassDescriptor parentDescriptor, ClassDescriptor childDescriptor, AbstractSession session) throws DescriptorException {
        ClassDescriptor clonedParentDescriptor = (ClassDescriptor)parentDescriptor.clone();

        //recursive call to the further parent descriptors
        if (clonedParentDescriptor.getInheritancePolicy().isChildDescriptor()) {
            ClassDescriptor parentToParentDescriptor = session.getDescriptor(clonedParentDescriptor.getJavaClass());
            initializeParentInheritance(parentToParentDescriptor, parentDescriptor, session);
        }

        initializeReferenceDescriptor(clonedParentDescriptor, session);
        Vector children = org.eclipse.persistence.internal.helper.NonSynchronizedVector.newInstance(1);
        children.addElement(childDescriptor);
        clonedParentDescriptor.getInheritancePolicy().setChildDescriptors(children);
        clonedParentDescriptor.preInitialize(session);
        clonedParentDescriptor.initialize(session);
        translateFields(clonedParentDescriptor, session);
    }

    /**
     * INTERNAL:
     * Initialize the cloned reference descriptor with table names and primary keys
     */
    protected void initializeReferenceDescriptor(ClassDescriptor clonedDescriptor, AbstractSession session) {
        if (aggregateKeyTable != null){
            clonedDescriptor.setDefaultTable(aggregateKeyTable);
            Vector<DatabaseTable> tables = new Vector<>(1);
            tables.add(aggregateKeyTable);
            clonedDescriptor.setTables(tables);
        } else {
            // Must ensure default tables remains the same.
            clonedDescriptor.setDefaultTable(getDescriptor().getDefaultTable());
            clonedDescriptor.setTables(getDescriptor().getTables());
            clonedDescriptor.setPrimaryKeyFields(getDescriptor().getPrimaryKeyFields());

            // We need to translate the mappings field now BEFORE the clonedDescriptor gets initialized.
            // During clonedDescriptor initialize, the mappings will be initialized and the field needs translated by then.
            // If we wait any longer, the initialized mapping will be set as a primary key if the field name matches primaryKeyFields
            Vector<DatabaseMapping> mappingsToTranslate = clonedDescriptor.getMappings();
            for (DatabaseMapping mapping : mappingsToTranslate) {
                DatabaseField field = mapping.getField();
                if(field != null) {
                    DatabaseField sourceField = getAggregateToSourceFields().get(field.getName());
                    translateField(sourceField, field, clonedDescriptor);
                }
            }

            if (clonedDescriptor.hasTargetForeignKeyMapping(session) && !isJPAIdNested(session)) {
                for (DatabaseField pkField : getDescriptor().getPrimaryKeyFields()) {
                    if (!getAggregateToSourceFields().containsKey(pkField.getName())) {
                        // pk field from the source descriptor will have its type set by source descriptor
                        // this only could be done if there is no aggregate field with the same name as pk field.
                        clonedDescriptor.getObjectBuilder().getFieldsMap().put(pkField, pkField);
                    }
                }
            }
        }
        if (this.getDescriptor().hasFetchGroupManager() && FetchGroupTracker.class.isAssignableFrom(clonedDescriptor.getJavaClass())){
            if (clonedDescriptor.getFetchGroupManager() == null) {
                clonedDescriptor.setFetchGroupManager(new FetchGroupManager());
            }
        }
    }

    /**
     * INTERNAL:
     * Called when iterating through descriptors to handle iteration on this mapping when it is used as a MapKey
     * @param iterator
     * @param element
     */
    @Override
    public void iterateOnMapKey(DescriptorIterator iterator, Object element){
        super.iterateOnAttributeValue(iterator, element);
    }

    /**
     * INTERNAL:
     * Return whether this mapping should be traversed when we are locking
     * @return
     */
    @Override
    public boolean isLockableMapping(){
        return true;
    }

    /**
     * INTERNAL:
     * Related mapping should implement this method to return true.
     */
    @Override
    public boolean isAggregateObjectMapping() {
        return true;
    }

    /**
     * INTERNAL:
     * Return if this mapping supports change tracking.
     */
    @Override
    public boolean isChangeTrackingSupported(Project project) {
        // This can be called before and after initialization.
        // Use the mapping reference descriptor when initialized, otherwise find the uninitialized one.
        ClassDescriptor referencedDescriptor = getReferenceDescriptor();
        if (referencedDescriptor == null) {
            Iterator ordered = project.getOrderedDescriptors().iterator();
            while (ordered.hasNext() && referencedDescriptor == null){
                ClassDescriptor descriptor = (ClassDescriptor)ordered.next();
                if (descriptor.getJavaClassName().equals(getReferenceClassName())){
                    referencedDescriptor = descriptor;
                }
            }
        }
        if (referencedDescriptor != null) {
            if (!referencedDescriptor.supportsChangeTracking(project)) {
                return false;
            }
            // Also check subclasses.
            if (referencedDescriptor.hasInheritance()) {
                for (Iterator iterator = referencedDescriptor.getInheritancePolicy().getChildDescriptors().iterator(); iterator.hasNext(); ) {
                    ClassDescriptor subclassDescriptor = (ClassDescriptor)iterator.next();
                    if (!subclassDescriptor.supportsChangeTracking(project)) {
                        return false;
                    }
                }
            }
            return true;
        }
        return false;
    }

    /**
     * INTERNAL
     * Return true if this mapping supports cascaded version optimistic locking.
     */
    @Override
    public boolean isCascadedLockingSupported() {
        return true;
    }

    /**
     * INTERNAL:
     * Flags that either this mapping or nested mapping is a JPA id mapping.
     */
    public boolean isJPAIdNested(AbstractSession session) {
        if (isJPAId()) {
            return true;
        } else {
            ClassDescriptor referenceDescriptor = getReferenceDescriptor();
            if (referenceDescriptor == null) {
                // the mapping has not been initialized yet
                referenceDescriptor = session.getDescriptor(getReferenceClass());
            }
            for (DatabaseMapping mapping : referenceDescriptor.getMappings()) {
                if (mapping.isAggregateObjectMapping() && ((AggregateObjectMapping)mapping).isJPAIdNested(session)) {
                    return true;
                }
            }
            return false;
        }
    }

    /**
     * PUBLIC:
     * Return if all the fields in the database row for the aggregate object are NULL,
     * then, by default, the mapping will place a null in the appropriate source object
     * (as opposed to an aggregate object filled with nulls).
     * To change this behavior, set the value of this variable to false. Then the mapping
     * will build a new instance of the aggregate object that is filled with nulls
     * and place it in the source object.
     *
     * Note: Any aggregate that has a relationship mapping automatically does not allow
     * null.
     */
    public boolean isNullAllowed() {
        return isNullAllowed;
    }

    /**
     * INTERNAL:
     * For an aggregate mapping the reference descriptor is cloned. The cloned descriptor is then
     * assigned primary keys and table names before initialize. Once the cloned descriptor is initialized
     * it is assigned as reference descriptor in the aggregate mapping. This is a very specific
     * behavior for aggregate mappings. The original descriptor is used only for creating clones and
     * after that the aggregate mapping never uses it.
     * Some initialization is done in postInitialize to ensure the target descriptor's references are initialized.
     */
    @Override
    public void postInitialize(AbstractSession session) throws DescriptorException {
        super.postInitialize(session);

        if (getReferenceDescriptor() != null) {
            getReferenceDescriptor().getCachePolicy().setCacheIsolation(this.descriptor.getCachePolicy().getCacheIsolation());
            // Changed as part of fix for bug#4410581 aggregate mapping can not be set to use change tracking if owning descriptor does not use it.
            // Basically the policies should be the same, but we also allow deferred with attribute for CMP2 (courser grained).
            if (getDescriptor().getObjectChangePolicy().getClass().equals(DeferredChangeDetectionPolicy.class)) {
                getReferenceDescriptor().setObjectChangePolicy(new DeferredChangeDetectionPolicy());
            } else if (getDescriptor().getObjectChangePolicy().getClass().equals(ObjectChangeTrackingPolicy.class)
                    && getReferenceDescriptor().getObjectChangePolicy().getClass().equals(AttributeChangeTrackingPolicy.class)) {
                getReferenceDescriptor().setObjectChangePolicy(new ObjectChangeTrackingPolicy());
            }

            //need to set the primary key classification as the mappings for the pk fields might not be available
            if (getReferenceDescriptor().isAggregateDescriptor()){
                getReferenceDescriptor().getObjectBuilder().setPrimaryKeyClassifications(this.getDescriptor().getObjectBuilder().getPrimaryKeyClassifications());
                getReferenceDescriptor().setHasSimplePrimaryKey(this.getDescriptor().hasSimplePrimaryKey());
            }

            getReferenceDescriptor().postInitialize(session);
        }
    }

    /**
     * INTERNAL:
     * Making any mapping changes necessary to use a the mapping as a map key prior to initializing the mapping
     */
    @Override
    public void preinitializeMapKey(DatabaseTable table) {
        setTableForAggregateMappingKey(table);
    }

    /**
     * INTERNAL:
     * Making any mapping changes necessary to use a the mapping as a map key after initializing the mapping.
     */
    @Override
    public void postInitializeMapKey(MappedKeyMapContainerPolicy policy) {
        return;
    }

    /**
     * INTERNAL:
     * Build an aggregate object from the specified return row and put it
     * in the specified target object.
     * Return row is merged into object after execution of insert or update call
     * according to ReturningPolicy.
     * If not null changeSet must correspond to targetObject. changeSet is updated with all of the field values in the row.
     */
    public Object readFromReturnRowIntoObject(AbstractRecord row, Object targetObject, ReadObjectQuery query, Collection handledMappings, ObjectChangeSet changeSet) throws DatabaseException {
        Object aggregate = getAttributeValueFromObject(targetObject);
        ObjectChangeSet aggregateChangeSet = null;

        if (aggregate == null) {
            aggregate = readFromRowIntoObject(row, null, targetObject, null, query, query.getSession(), true);
        } else {
            if(changeSet != null && (!changeSet.isNew() || (query.getDescriptor() != null && query.getDescriptor().shouldUseFullChangeSetsForNewObjects()))) {
                aggregateChangeSet = getReferenceDescriptor(aggregate, query.getSession()).getObjectBuilder().createObjectChangeSet(aggregate, (UnitOfWorkChangeSet)((UnitOfWorkImpl)query.getSession()).getUnitOfWorkChangeSet(), true, query.getSession());
            }
            AbstractRecord aggregateRow = new DatabaseRecord();
            int size = row.size();
            List fields = row.getFields();
            List values = row.getValues();
            List aggregateFields = getReferenceFields();
            for(int i=0; i < size; i++) {
                DatabaseField field = (DatabaseField)fields.get(i);
                if(aggregateFields.contains(field)) {
                    aggregateRow.add(field, values.get(i));
                }
            }

            getObjectBuilder(aggregate, query.getSession()).assignReturnRow(aggregate, query.getSession(), aggregateRow, aggregateChangeSet);
        }

        if (aggregate != null && isNullAllowed()) {
            boolean allAttributesNull = true;
            int nAggregateFields = this.fields.size();
            for (int i = 0; (i < nAggregateFields) && allAttributesNull; i++) {
                DatabaseField field = this.fields.elementAt(i);
                if (row.containsKey(field)) {
                    allAttributesNull = row.get(field) == null;
                } else {
                    Object fieldValue = valueFromObject(targetObject, field, query.getSession());
                    if (fieldValue == null) {
                        Object baseValue = getDescriptor().getObjectBuilder().getBaseValueForField(field, targetObject);
                        if (baseValue != null) {
                            DatabaseMapping baseMapping = getDescriptor().getObjectBuilder().getBaseMappingForField(field);
                            if (baseMapping.isForeignReferenceMapping()) {
                                ForeignReferenceMapping refMapping = (ForeignReferenceMapping)baseMapping;
                                if (refMapping.usesIndirection()) {
                                    allAttributesNull = refMapping.getIndirectionPolicy().objectIsInstantiated(baseValue);
                                }
                            } else if (baseMapping.isTransformationMapping()) {
                                AbstractTransformationMapping transMapping = (AbstractTransformationMapping)baseMapping;
                                if (transMapping.usesIndirection()) {
                                    allAttributesNull = transMapping.getIndirectionPolicy().objectIsInstantiated(baseValue);
                                }
                            }
                        }
                    } else {
                        allAttributesNull = false;
                    }
                }
            }
            if (allAttributesNull) {
                aggregate = null;
                setAttributeValueInObject(targetObject, aggregate);
            }
        }

        if(changeSet != null && (!changeSet.isNew() || (query.getDescriptor() != null && query.getDescriptor().shouldUseFullChangeSetsForNewObjects()))) {
            AggregateChangeRecord record = (AggregateChangeRecord)changeSet.getChangesForAttributeNamed(getAttributeName());
            if(aggregate == null) {
                if(record != null) {
                    record.setChangedObject(null);
                }
            } else {
                if (record == null) {
                    record = new AggregateChangeRecord(changeSet);
                    record.setAttribute(getAttributeName());
                    record.setMapping(this);
                    changeSet.addChange(record);
                }
                if (aggregateChangeSet == null) {
                    // the old aggregate value was null
                    aggregateChangeSet = getReferenceDescriptor(aggregate, query.getSession()).getObjectBuilder().createObjectChangeSet(aggregate, (UnitOfWorkChangeSet)((UnitOfWorkImpl)query.getSession()).getUnitOfWorkChangeSet(), true, query.getSession());
                }
                record.setChangedObject(aggregateChangeSet);
            }
        }
        if (handledMappings != null) {
            handledMappings.add(this);
        }
        return aggregate;
    }

    /**
     * INTERNAL:
     * Build an aggregate object from the specified row and put it
     * in the specified target object.
     */
    @Override
    public Object readFromRowIntoObject(AbstractRecord databaseRow, JoinedAttributeManager joinManager, Object targetObject, CacheKey parentCacheKey, ObjectBuildingQuery sourceQuery, AbstractSession executionSession, boolean isTargetProtected) throws DatabaseException {
        Object aggregate = buildAggregateFromRow(databaseRow, targetObject, parentCacheKey, joinManager, sourceQuery, false, executionSession, isTargetProtected);// don't just build a shallow original
        setAttributeValueInObject(targetObject, aggregate);
        return aggregate;
    }

    /**
     * INTERNAL:
     * Rehash any hashtables based on fields.
     * This is used to clone descriptors for aggregates, which hammer field names.
     */
    @Override
    public void rehashFieldDependancies(AbstractSession session) {
        getReferenceDescriptor().rehashFieldDependancies(session);
    }

    /**
     * INTERNAL:
     * Return whether this mapping requires extra queries to update the rows if it is
     * used as a key in a map.  This will typically be true if there are any parts to this mapping
     * that are not read-only.
     */
    @Override
    public boolean requiresDataModificationEventsForMapKey(){
        if (getReferenceDescriptor() != null){
            Iterator<DatabaseMapping> i = getReferenceDescriptor().getMappings().iterator();
            while (i.hasNext()){
                DatabaseMapping mapping = i.next();
                if (!mapping.isReadOnly()){
                    Iterator<DatabaseField> fields = mapping.getFields().iterator();
                    while (fields.hasNext()){
                        DatabaseField field = fields.next();
                        if (field.isUpdatable()){
                            return true;
                        }
                    }
                }
            }
            return false;
        }
        return true;
    }

    /**
     * INTERNAL:
     * Set a collection of the aggregate to source field name associations.
     */
    public void setAggregateToSourceFieldAssociations(Vector<Association> fieldAssociations) {
        Hashtable fieldNames = new Hashtable(fieldAssociations.size() + 1);
        for (Enumeration associationsEnum = fieldAssociations.elements();
                 associationsEnum.hasMoreElements();) {
            Association association = (Association)associationsEnum.nextElement();
            fieldNames.put(association.getKey(), association.getValue());
        }

        setAggregateToSourceFields(fieldNames);
    }

    /**
     * INTERNAL:
     * Set the hashtable that stores target field name to the source field name.
     */
    public void setAggregateToSourceFields(Map<String, DatabaseField> aggregateToSource) {
        aggregateToSourceFields = aggregateToSource;
    }

    /**
     * INTERNAL:
     * Set the hashtable that stores a field in the source table 
     * to a field name in a nested aggregate descriptor.
     */
    public void setNestedFieldTranslations(Map<String, Object[]> fieldTranslations) {
        nestedFieldTranslations = fieldTranslations;
    }

    /**
     * PUBLIC:
     * Configure if all the fields in the database row for the aggregate object are NULL,
     * then, by default, the mapping will place a null in the appropriate source object
     * (as opposed to an aggregate object filled with nulls).
     * To change this behavior, set the value of this variable to false. Then the mapping
     * will build a new instance of the aggregate object that is filled with nulls
     * and place it in the source object.
     *
     * Note: Any aggregate that has a relationship mapping automatically does not allow
     * null.
     */
    public void setIsNullAllowed(boolean isNullAllowed) {
        this.isNullAllowed = isNullAllowed;
    }

    /**
     * INTERNAL:
     * If this mapping is used as the key of a CollectionTableMapMapping, the table used by this
     * mapping will be the relation table.  Set this table.
     */
    public void setTableForAggregateMappingKey(DatabaseTable table){
        aggregateKeyTable = table;
    }

    /**
     * INTERNAL:
     * Apply the field translation from the sourceField to the mappingField.
     */
    protected void translateField(DatabaseField sourceField, DatabaseField mappingField, ClassDescriptor clonedDescriptor) {
        // Do not modify non-translated fields.
        if (sourceField != null) {
            //merge fieldInSource into the field from the Aggregate descriptor
            mappingField.setName(sourceField.getName());
            mappingField.setUseDelimiters(sourceField.shouldUseDelimiters());
            mappingField.useUpperCaseForComparisons(sourceField.getUseUpperCaseForComparisons());
            mappingField.setNameForComparisons(sourceField.getNameForComparisons());
            //copy type information
            mappingField.setNullable(sourceField.isNullable());
            mappingField.setUpdatable(sourceField.isUpdatable());
            mappingField.setInsertable(sourceField.isInsertable());
            mappingField.setUnique(sourceField.isUnique());
            mappingField.setScale(sourceField.getScale());
            mappingField.setLength(sourceField.getLength());
            mappingField.setPrecision(sourceField.getPrecision());
            mappingField.setColumnDefinition(sourceField.getColumnDefinition());

            // Check if the translated field specified a table qualifier.
            if (sourceField.hasTableName()) {
                mappingField.setTable(clonedDescriptor.getTable(sourceField.getTable().getName()));
            }

            // Tag this field as translated. Some mapping care to know which
            // have been translated in the rehashFieldDependancies call.
            mappingField.setIsTranslated(true);
        }
    }

    /**
     * INTERNAL:
     * If field names are different in the source and aggregate objects then the translation
     * is done here. The aggregate field name is converted to source field name from the
     * field name mappings stored.
     */
    protected void translateNestedFields(ClassDescriptor clonedDescriptor, AbstractSession session) {
        if (this.nestedFieldTranslations == null) {
            //this only happens when using Metadata Caching
            return;
        }
        // Once the cloned descriptor is initialized, go through our nested
        // field name translations. Any errors are silently ignored as
        // validation is assumed to be done before hand (JPA metadata processing
        // does validate any nested field translation)
        for (Entry<String, Object[]> translations : this.nestedFieldTranslations.entrySet()) {
            String attributeName = translations.getKey();
            DatabaseMapping mapping = null;
            String currentAttributeName = attributeName.substring(0, attributeName.indexOf('.'));
            String remainingAttributeName = attributeName.substring(attributeName.indexOf('.')+ 1);
            mapping = clonedDescriptor.getMappingForAttributeName(currentAttributeName);
            if (mapping.isAggregateObjectMapping()) {
                if (remainingAttributeName != null && remainingAttributeName.contains(".")){
                    //This should be the case otherwise the metadata validation would have validated
                    ((AggregateObjectMapping)mapping).addNestedFieldTranslation(remainingAttributeName, (DatabaseField)translations.getValue()[0], (String)translations.getValue()[1]);
                } else {
                    ((AggregateObjectMapping)mapping).addFieldTranslation((DatabaseField) translations.getValue()[0], (String)translations.getValue()[1]);
                }
            }
        }
    }

    /**
     * INTERNAL:
     * If field names are different in the source and aggregate objects then the translation
     * is done here. The aggregate field name is converted to source field name from the
     * field name mappings stored.
     */
    protected void translateFields(ClassDescriptor clonedDescriptor, AbstractSession session) {
        // EL Bug 326977
        Vector fieldsToTranslate = (Vector) clonedDescriptor.getFields().clone();
        for (Iterator qkIterator = clonedDescriptor.getQueryKeys().values().iterator(); qkIterator.hasNext();) {
            QueryKey queryKey = (QueryKey)qkIterator.next();
            if (queryKey.isDirectQueryKey()) {
                DatabaseField field = ((DirectQueryKey)queryKey).getField();
                fieldsToTranslate.add(field);
            }
        }

        // EL Bug 332080 - translate foreign reference mapping source key fields
        if (!clonedDescriptor.getObjectBuilder().isSimple()) {
            for (Iterator dcIterator = clonedDescriptor.getMappings().iterator(); dcIterator.hasNext();) {
                DatabaseMapping mapping = (DatabaseMapping)dcIterator.next();
                if (mapping.isForeignReferenceMapping()) {
                    Collection fkFields = ((ForeignReferenceMapping)mapping).getFieldsForTranslationInAggregate();
                    if (fkFields != null && !fkFields.isEmpty()) {
                        fieldsToTranslate.addAll(fkFields);
                    }
                }
            }
        }

        for (Iterator entry = fieldsToTranslate.iterator(); entry.hasNext();) {
            DatabaseField field = (DatabaseField)entry.next();
            //322233 - get the source DatabaseField from the translation map.
            translateField(getAggregateToSourceFields().get(field.getName()), field, clonedDescriptor);
        }

        clonedDescriptor.rehashFieldDependancies(session);
    }

    /**
     * INTERNAL:
     * Allow the key mapping to unwrap the object.
     */
    @Override
    public Object unwrapKey(Object key, AbstractSession session){
        return key;
    }

    /**
     * INTERNAL:
     * Allow the key mapping to wrap the object.
     */
    @Override
    public Object wrapKey(Object key, AbstractSession session){
        return key;
    }

    /**
     * INTERNAL:
     * A subclass should implement this method if it wants different behavior.
     * Write the foreign key values from the attribute to the row.
     */
    @Override
    public void writeFromAttributeIntoRow(Object attribute, AbstractRecord row, AbstractSession session){
        writeToRowFromAggregate(row, null, attribute, session, WriteType.UNDEFINED);
    }

    /**
     * INTERNAL:
     * Extract value of the field from the object
     */
    @Override
    public Object valueFromObject(Object object, DatabaseField field, AbstractSession session) throws DescriptorException {
        Object attributeValue = getAttributeValueFromObject(object);
        if (attributeValue == null) {
            if (isNullAllowed()) {
                return null;
            } else {
                throw DescriptorException.nullForNonNullAggregate(object, this);
            }
        } else {
            return getObjectBuilder(attributeValue, session).extractValueFromObjectForField(attributeValue, field, session);
        }
    }

    /**
     * INTERNAL:
     * Get the attribute value from the object and add the appropriate
     * values to the specified database row.
     */
    @Override
    public void writeFromObjectIntoRow(Object object, AbstractRecord databaseRow, AbstractSession session, WriteType writeType) throws DescriptorException {
        if (isReadOnly()) {
            return;
        }
        writeToRowFromAggregate(databaseRow, object, getAttributeValueFromObject(object), session, writeType);
    }

    /**
     * INTERNAL:
     * This row is built for shallow insert which happens in case of bidirectional inserts.
     */
    @Override
    public void writeFromObjectIntoRowForShallowInsert(Object object, AbstractRecord row, AbstractSession session) {
        if (isReadOnly()) {
            return;
        }
        writeToRowFromAggregateForShallowInsert(row, object, getAttributeValueFromObject(object), session);
    }

    /**
     * INTERNAL:
     * This row is built for update after shallow insert which happens in case of bidirectional inserts.
     * It contains the foreign keys with non null values that were set to null for shallow insert.
     */
    @Override
    public void writeFromObjectIntoRowForUpdateAfterShallowInsert(Object object, AbstractRecord row, AbstractSession session, DatabaseTable table) {
        if (isReadOnly() || !getFields().get(0).getTable().equals(table) || isPrimaryKeyMapping()) {
            return;
        }
        writeToRowFromAggregateForUpdateAfterShallowInsert(row, object, getAttributeValueFromObject(object), session, table);
    }

    /**
     * INTERNAL:
     * This row is built for update before shallow delete which happens in case of bidirectional inserts.
     * It contains the same fields as the row built by writeFromObjectIntoRowForUpdateAfterShallowInsert, but all the values are null.
     */
    @Override
    public void writeFromObjectIntoRowForUpdateBeforeShallowDelete(Object object, AbstractRecord row, AbstractSession session, DatabaseTable table) {
        if (isReadOnly() || !getFields().get(0).getTable().equals(table) || isPrimaryKeyMapping()) {
            return;
        }
        writeToRowFromAggregateForUpdateBeforeShallowDelete(row, object, getAttributeValueFromObject(object), session, table);
    }

    /**
     * INTERNAL:
     * Get the attribute value from the object and add the appropriate
     * values to the specified database row.
     */
    @Override
    public void writeFromObjectIntoRowWithChangeRecord(ChangeRecord changeRecord, AbstractRecord databaseRow, AbstractSession session, WriteType writeType) throws DescriptorException {
        if (isReadOnly()) {
            return;
        }
        writeToRowFromAggregateWithChangeRecord(databaseRow, changeRecord, (ObjectChangeSet)((AggregateChangeRecord)changeRecord).getChangedObject(), session, writeType);
    }

    /**
     * INTERNAL:
     * Get the attribute value from the object and add the changed
     * values to the specified database row.
     */
    @Override
    public void writeFromObjectIntoRowForUpdate(WriteObjectQuery query, AbstractRecord databaseRow) throws DescriptorException {
        if (isReadOnly()) {
            return;
        }
        writeToRowFromAggregateForUpdate(databaseRow, query, getAttributeValueFromObject(query.getObject()));
    }

    /**
     * INTERNAL:
     * Write fields needed for insert into the template for with null values.
     */
    @Override
    public void writeInsertFieldsIntoRow(AbstractRecord databaseRow, AbstractSession session) {
        if (isReadOnly()) {
            return;
        }

        AbstractRecord targetRow = buildTemplateInsertRow(session);
        for (Enumeration keyEnum = targetRow.keys(); keyEnum.hasMoreElements();) {
            DatabaseField field = (DatabaseField)keyEnum.nextElement();
            if (field.isInsertable()) {
                Object value = targetRow.get(field);
                //CR-3286097 - Should use add not put, to avoid linear search.
                databaseRow.add(field, value);
            }
        }
    }

    @Override
    public void writeUpdateFieldsIntoRow(AbstractRecord databaseRow, AbstractSession session) {
        if (isReadOnly()) {
            return;
        }

        AbstractRecord targetRow = buildTemplateInsertRow(session);
        for (Enumeration keyEnum = targetRow.keys(); keyEnum.hasMoreElements();) {
            DatabaseField field = (DatabaseField)keyEnum.nextElement();
            if (field.isUpdatable()) {
                Object value = targetRow.get(field);
                //CR-3286097 - Should use add not put, to avoid linear search.
                databaseRow.add(field, value);
            }
        }
    }

    /**
     * INTERNAL:
     * Add a primary key join column (secondary field).
     * If this contain primary keys and the descriptor(or its subclass) has multiple tables
     * (secondary tables or joined inheritance strategy), this should also know the primary key
     * join columns to handle some cases properly.
     */
    public void addPrimaryKeyJoinField(DatabaseField primaryKeyField, DatabaseField secondaryField) {
        // now it doesn't need to manage this as a separate table here,
        // it's enough just to add the mapping to ObjectBuilder.mappingsByField
        ObjectBuilder builder = getReferenceDescriptor().getObjectBuilder();
        DatabaseMapping mapping = builder.getMappingForField(primaryKeyField);
        if (mapping != null) {
            builder.getMappingsByField().put(secondaryField, mapping);
        }
    }
}