oracle.toplink.essentials.queryframework.ReportQuery Maven / Gradle / Ivy
The newest version!
/*
* DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS HEADER.
*
* // Copyright (c) 1998, 2007, Oracle. All rights reserved.
*
*
* The contents of this file are subject to the terms of either the GNU
* General Public License Version 2 only ("GPL") or the Common Development
* and Distribution License("CDDL") (collectively, the "License"). You
* may not use this file except in compliance with the License. You can obtain
* a copy of the License at https://glassfish.dev.java.net/public/CDDL+GPL.html
* or glassfish/bootstrap/legal/LICENSE.txt. See the License for the specific
* language governing permissions and limitations under the License.
*
* When distributing the software, include this License Header Notice in each
* file and include the License file at glassfish/bootstrap/legal/LICENSE.txt.
* Sun designates this particular file as subject to the "Classpath" exception
* as provided by Sun in the GPL Version 2 section of the License file that
* accompanied this code. If applicable, add the following below the License
* Header, with the fields enclosed by brackets [] replaced by your own
* identifying information: "Portions Copyrighted [year]
* [name of copyright owner]"
*
* Contributor(s):
*
* If you wish your version of this file to be governed by only the CDDL or
* only the GPL Version 2, indicate your decision by adding "[Contributor]
* elects to include this software in this distribution under the [CDDL or GPL
* Version 2] license." If you don't indicate a single choice of license, a
* recipient has the option to distribute your version of this file under
* either the CDDL, the GPL Version 2 or to extend the choice of license to
* its licensees as provided above. However, if you add GPL Version 2 code
* and therefore, elected the GPL Version 2 license, then the option applies
* only if the new code is made subject to such option by the copyright
* holder.
*/
package oracle.toplink.essentials.queryframework;
import java.util.*;
import oracle.toplink.essentials.expressions.*;
import oracle.toplink.essentials.internal.expressions.*;
import oracle.toplink.essentials.internal.queryframework.*;
import oracle.toplink.essentials.exceptions.*;
import oracle.toplink.essentials.internal.helper.*;
import oracle.toplink.essentials.mappings.*;
import oracle.toplink.essentials.internal.security.PrivilegedAccessHelper;
import oracle.toplink.essentials.internal.sessions.AbstractRecord;
import oracle.toplink.essentials.internal.sessions.UnitOfWorkImpl;
import oracle.toplink.essentials.internal.sessions.AbstractSession;
import oracle.toplink.essentials.descriptors.ClassDescriptor;
/**
* Purpose: Query for information about a set of objects instead of the objects themselves.
* This supports select single attributes, nested attributes, aggregation functions and group bys.
*
* Attribute Types:
* - addAttribute("directQueryKey") is a short cut method to add an attribute with the same name as its corresponding direct query key.
*
- addAttribute("attributeName", expBuilder.get("oneToOneMapping").get("directQueryKey")) is the full approach for get values through joined 1:1 relationships.
*
- addAttribute("attributeName", expBuilder.getField("TABLE.FIELD")) allows the addition of raw values or values which were not mapped in the object model directly (i.e. FK attributes).
*
- addAttribute("attributeName", null) Leave a place holder (NULL) value in the result (used for included values from other systems or calculated values).
*
* Retrieving Primary Keys: It is possble to retrieve the primary key raw values within each result, but stored in a separate (internal) vector. This
* primary key vector can later be used to retrieve the real object.
* @see #retrievePrimaryKeys()
* @see ReportQueryResult#readObject(Class, Session)
* If the values are wanted in the result array then they must be added as attributes. For primary keys which are not mapped directly
* you can add them as DatabaseFields (see above).
*
* @author Doug Clarke
* @since TOPLink/Java 2.0
*/
public class ReportQuery extends ReadAllQuery {
/** Simplifies the result by only returning the first result. */
public static final int ShouldReturnSingleResult = 1;
/** Simplifies the result by only returning one value. */
public static final int ShouldReturnSingleValue = 2;
/** Simplifies the result by only returning the single attribute(as opposed to wrapping in a
ReportQueryResult). */
public static final int ShouldReturnSingleAttribute = 3;
/** For EJB 3 support returns results without using the ReportQueryResult */
public static final int ShouldReturnWithoutReportQueryResult = 4;
/** Specifies whether to retreive primary keys, first primary key, or no primary key.*/
public static final int FULL_PRIMARY_KEY = 2;
public static final int FIRST_PRIMARY_KEY = 1;
public static final int NO_PRIMARY_KEY = 0;
//GF_ISSUE_395
protected static final Boolean RESULT_IGNORED = Boolean.valueOf(true);
//end GF_ISSUE
/** Flag indicating wether the primary key values should also be retrieved for the reference class. */
protected int shouldRetrievePrimaryKeys;
/** Collection of names for use by results. */
protected Vector names;
/** Items to be selected, these could be attributes or aggregate functions. */
protected Vector items;
/** Expressions representing fields to be used in the GROUP BY clause. */
protected Vector groupByExpressions;
/** Expression representing the HAVING clause. */
protected Expression havingExpression;
/** Can be one of (ShouldReturnSingleResult, ShouldReturnSingleValue, ShouldReturnSingleAttribute)
** Simplifies the result by only returning the first result, first value, or all attribute values
*/
protected int returnChoice;
/** flag to allow items to be added to the last ConstructorReportItem **/
protected boolean addToConstructorItem;
protected Class resultConstructorClass;
protected Class[] constructorArgTypes;
protected List constructorMappings;
/* GF_ISSUE_395 this attribute stores a set of unique keys that identity results.
* Used when distinct has been set on the query. For use in TCK
*/
protected HashSet returnedKeys;
/**
* INTERNAL:
* The builder should be provided.
*/
public ReportQuery() {
this.queryMechanism = new ExpressionQueryMechanism(this);
this.items = new Vector();
this.shouldRetrievePrimaryKeys = NO_PRIMARY_KEY;
this.groupByExpressions = new Vector(3);
this.havingExpression=null;
this.addToConstructorItem = false;
// overwrite the lock mode to NO_LOCK, this prevents the report query to lock
// when DEFAULT_LOCK_MODE and a pessimistic locking policy are used.
this.setLockMode(ObjectBuildingQuery.NO_LOCK);
}
public ReportQuery(Class javaClass, Expression expression) {
this();
this.defaultBuilder = expression.getBuilder();
setReferenceClass(javaClass);
setSelectionCriteria(expression);
}
/**
* PUBLIC:
* The report query is require to be constructor with an expression builder.
* This build must be used for the selection critiera, any item expressions, group bys and order bys.
*/
public ReportQuery(Class javaClass, ExpressionBuilder builder) {
this();
this.defaultBuilder = builder;
setReferenceClass(javaClass);
}
/**
* PUBLIC:
* The report query is require to be constructor with an expression builder.
* This build must be used for the selection critiera, any item expressions, group bys and order bys.
*/
public ReportQuery(ExpressionBuilder builder) {
this();
this.defaultBuilder = builder;
}
/**
* PUBLIC:
* Add the attribute from the reference class to be included in the result.
* EXAMPLE: reportQuery.addAttribute("firstName");
*/
public void addAttribute(String itemName) {
addItem(itemName, getExpressionBuilder().get(itemName));
}
/**
* PUBLIC:
* Add the attribute to be included in the result.
* EXAMPLE: reportQuery.addAttribute("city", expBuilder.get("address").get("city"));
*/
public void addAttribute(String itemName, Expression attributeExpression) {
addItem(itemName, attributeExpression);
}
/**
* PUBLIC:
* Add the attribute to be included in the result. Return the result as the provided class
* EXAMPLE: reportQuery.addAttribute("city", expBuilder.get("period").get("startTime"), Time.class);
*/
public void addAttribute(String itemName, Expression attributeExpression, Class type) {
addItem(itemName, attributeExpression, type);
}
/**
* PUBLIC:
* Add the average value of the attribute to be included in the result.
* Aggregation functions can be used with a group by, or on the entire result set.
* EXAMPLE: reportQuery.addAverage("salary");
*/
public void addAverage(String itemName) {
addAverage(itemName, getExpressionBuilder().get(itemName));
}
/**
* PUBLIC:
* Add the average value of the attribute to be included in the result and
* return it as the specified resultType.
* Aggregation functions can be used with a group by, or on the entire result set.
* EXAMPLE: reportQuery.addAverage("salary", Float.class);
*/
public void addAverage(String itemName, Class resultType) {
addAverage(itemName, getExpressionBuilder().get(itemName), resultType);
}
/**
* PUBLIC:
* Add the average value of the attribute to be included in the result.
* Aggregation functions can be used with a group by, or on the entire result set.
* EXAMPLE: reportQuery.addAverage("managerSalary", expBuilder.get("manager").get("salary"));
*/
public void addAverage(String itemName, Expression attributeExpression) {
addItem(itemName, attributeExpression.average());
}
/**
* PUBLIC:
* Add the average value of the attribute to be included in the result and
* return it as the specified resultType.
* Aggregation functions can be used with a group by, or on the entire result set.
* EXAMPLE: reportQuery.addAverage("managerSalary", expBuilder.get("manager").get("salary"), Double.class);
*/
public void addAverage(String itemName, Expression attributeExpression, Class resultType) {
addItem(itemName, attributeExpression.average(), resultType);
}
/**
* PUBLIC:
* Add a ConstructorReportItem to this query's set of return values.
* @param ConstructorReportItem - used to specify a class constructor and values to pass in from this query
* @see ConstructorReportItem
*/
public void addConstructorReportItem(ConstructorReportItem item){
addItem(item);
}
/**
* PUBLIC:
* Include the number of rows returned by the query in the result.
* Aggregation functions can be used with a group by, or on the entire result set.
* EXAMPLE:
* Java:
* reportQuery.addCount();
* SQL:
* SELECT COUNT (*) FROM ...
* @see #addCount(java.lang.String)
*/
public void addCount() {
addCount("COUNT", getExpressionBuilder());
}
/**
* PUBLIC:
* Include the number of rows returned by the query in the result, where attributeExpression is not null.
* Aggregation functions can be used with a group by, or on the entire result set.
* Example:
*
* TopLink: reportQuery.addCount("id");
* SQL: SELECT COUNT (t0.EMP_ID) FROM EMPLOYEE t0, ...
*
* @param attributeName the number of rows where attributeName is not null will be returned.
* @see #addCount(java.lang.String, oracle.toplink.essentials.expressions.Expression)
*/
public void addCount(String attributeName) {
addCount(attributeName, getExpressionBuilder().get(attributeName));
}
/**
* PUBLIC:
* Include the number of rows returned by the query in the result, where attributeExpression is not null.
* Aggregation functions can be used with a group by, or on the entire result set.
* Set the count to be returned as the specified resultType.
* Example:
*
* TopLink: reportQuery.addCount("id", Long.class);
* SQL: SELECT COUNT (t0.EMP_ID) FROM EMPLOYEE t0, ...
*
* @param attributeName the number of rows where attributeName is not null will be returned.
* @see #addCount(java.lang.String, oracle.toplink.essentials.expressions.Expression)
*/
public void addCount(String attributeName, Class resultType) {
addCount(attributeName, getExpressionBuilder().get(attributeName), resultType);
}
/**
* PUBLIC:
* Include the number of rows returned by the query in the result, where attributeExpression
* is not null.
* Aggregation functions can be used with a group by, or on the entire result set.
* Example:
*
* TopLink: reportQuery.addCount("Count", getExpressionBuilder().get("id"));
* SQL: SELECT COUNT (t0.EMP_ID) FROM EMPLOYEE t0, ...
*
* Example: counting only distinct values of an attribute.
*
* TopLink: reportQuery.addCount("Count", getExpressionBuilder().get("address").distinct());
* SQL: SELECT COUNT (DISTINCT t0.ADDR_ID) FROM EMPLOYEE t0, ...
*
* objectAttributes can be specified also, even accross many to many
* mappings.
* @see #addCount()
*/
public void addCount(String itemName, Expression attributeExpression) {
addItem(itemName, attributeExpression.count());
}
/**
* PUBLIC:
* Include the number of rows returned by the query in the result, where attributeExpression
* is not null.
* Aggregation functions can be used with a group by, or on the entire result set.
* Set the count to be returned as the specified resultType.
* Example:
*
* TopLink: reportQuery.addCount("Count", getExpressionBuilder().get("id"), Integer.class);
* SQL: SELECT COUNT (t0.EMP_ID) FROM EMPLOYEE t0, ...
*
* Example: counting only distinct values of an attribute.
*
* TopLink: reportQuery.addCount("Count", getExpressionBuilder().get("address").distinct());
* SQL: SELECT COUNT (DISTINCT t0.ADDR_ID) FROM EMPLOYEE t0, ...
*
* objectAttributes can be specified also, even accross many to many
* mappings.
* @see #addCount()
*/
public void addCount(String itemName, Expression attributeExpression, Class resultType) {
addItem(itemName, attributeExpression.count(), resultType);
}
/**
* ADVANCED:
* Add the function against the attribute expression to be included in the result.
* Aggregation functions can be used with a group by, or on the entire result set.
* Example: reportQuery.addFunctionItem("average", expBuilder.get("salary"), "average");
*/
public void addFunctionItem(String itemName, Expression attributeExpression, String functionName) {
Expression functionExpression = attributeExpression;
functionExpression = attributeExpression.getFunction(functionName);
ReportItem item = new ReportItem(itemName, functionExpression);
addItem(item);
}
/**
* PUBLIC:
* Add the attribute to the group by expressions.
* This will group the result set on that attribute and is normally used in conjunction with aggregation functions.
* Example: reportQuery.addGrouping("lastName")
*/
public void addGrouping(String attributeName) {
addGrouping(getExpressionBuilder().get(attributeName));
}
/**
* PUBLIC:
* Add the attribute expression to the group by expressions.
* This will group the result set on that attribute and is normally used in conjunction with aggregation functions.
* Example: reportQuery.addGrouping(expBuilder.get("address").get("country"))
*/
public void addGrouping(Expression expression) {
getGroupByExpressions().addElement(expression);
//Bug2804042 Must un-prepare if prepared as the SQL may change.
setIsPrepared(false);
}
/**
* PUBLIC:
* Add the expression to the query to be used in the HAVING clause.
* This epression will be used to filter the result sets after they are grouped. It must be used in conjunction with the GROUP BY clause.
* Example: reportQuery.setHavingExpression(expBuilder.get("address").get("country").equal("Canada"))
*/
public void setHavingExpression(Expression expression) {
havingExpression = expression;
setIsPrepared(false);
}
/**
* INTERNAL:
* Method used to abstract addToConstructorItem behavour from the public addItem methods
*/
private void addItem(ReportItem item){
if (addToConstructorItem && (getItems().size()>0) &&(((ReportItem)getItems().lastElement()).isContructorItem() )){
((ConstructorReportItem)getItems().lastElement()).addItem(item);
}else{
getItems().addElement(item);
}
//Bug2804042 Must un-prepare if prepared as the SQL may change.
setIsPrepared(false);
}
/**
* ADVANCED:
* Add the expression value to be included in the result.
* EXAMPLE: reportQuery.addItem("name", expBuilder.get("firstName").toUpperCase());
*/
public void addItem(String itemName, Expression attributeExpression) {
ReportItem item = new ReportItem(itemName, attributeExpression);
addItem(item);
}
/**
* ADVANCED:
* Add the expression value to be included in the result.
* EXAMPLE: reportQuery.addItem("name", expBuilder.get("firstName").toUpperCase());
*/
public void addItem(String itemName, Expression attributeExpression, List joinedExpressions) {
ReportItem item = new ReportItem(itemName, attributeExpression);
item.getJoinedAttributeManager().setJoinedAttributeExpressions_(joinedExpressions);
addItem(item);
}
/**
* INTERNAL:
* Add the expression value to be included in the result.
* EXAMPLE: reportQuery.addItem("name", expBuilder.get("firstName").toUpperCase());
* The resultType can be specified to support EJBQL that adheres to the
* EJB 3.0 spec.
*/
protected void addItem(String itemName, Expression attributeExpression, Class resultType) {
ReportItem item = new ReportItem(itemName, attributeExpression);
item.setResultType(resultType);
addItem(item);
}
/**
* PUBLIC:
* Add the maximum value of the attribute to be included in the result.
* Aggregation functions can be used with a group by, or on the entire result set.
* EXAMPLE: reportQuery.addMaximum("salary");
*/
public void addMaximum(String itemName) {
addMaximum(itemName, getExpressionBuilder().get(itemName));
}
/**
* PUBLIC:
* Add the maximum value of the attribute to be included in the result.
* Aggregation functions can be used with a group by, or on the entire result set.
* EXAMPLE: reportQuery.addMaximum("managerSalary", expBuilder.get("manager").get("salary"));
*/
public void addMaximum(String itemName, Expression attributeExpression) {
addItem(itemName, attributeExpression.maximum());
}
/**
* PUBLIC:
* Add the minimum value of the attribute to be included in the result.
* Aggregation functions can be used with a group by, or on the entire result set.
* EXAMPLE: reportQuery.addMinimum("salary");
*/
public void addMinimum(String itemName) {
addMinimum(itemName, getExpressionBuilder().get(itemName));
}
/**
* PUBLIC:
* Add the minimum value of the attribute to be included in the result.
* Aggregation functions can be used with a group by, or on the entire result set.
* EXAMPLE: reportQuery.addMinimum("managerSalary", expBuilder.get("manager").get("salary"));
*/
public void addMinimum(String itemName, Expression attributeExpression) {
addItem(itemName, attributeExpression.minimum());
}
/**
* PUBLIC:
* Add the standard deviation value of the attribute to be included in the result.
* Aggregation functions can be used with a group by, or on the entire result set.
* EXAMPLE: reportQuery.addStandardDeviation("salary");
*/
public void addStandardDeviation(String itemName) {
addStandardDeviation(itemName, getExpressionBuilder().get(itemName));
}
/**
* PUBLIC:
* Add the standard deviation value of the attribute to be included in the result.
* Aggregation functions can be used with a group by, or on the entire result set.
* EXAMPLE: reportQuery.addStandardDeviation("managerSalary", expBuilder.get("manager").get("salary"));
*/
public void addStandardDeviation(String itemName, Expression attributeExpression) {
addItem(itemName, attributeExpression.standardDeviation());
}
/**
* PUBLIC:
* Add the sum value of the attribute to be included in the result.
* Aggregation functions can be used with a group by, or on the entire result set.
* EXAMPLE: reportQuery.addSum("salary");
*/
public void addSum(String itemName) {
addSum(itemName, getExpressionBuilder().get(itemName));
}
/**
* PUBLIC:
* Add the sum value of the attribute to be included in the result and
* return it as the specified resultType.
* Aggregation functions can be used with a group by, or on the entire result set.
* EXAMPLE: reportQuery.addSum("salary", Float.class);
*/
public void addSum(String itemName, Class resultType) {
addSum(itemName, getExpressionBuilder().get(itemName), resultType);
}
/**
* PUBLIC:
* Add the sum value of the attribute to be included in the result.
* Aggregation functions can be used with a group by, or on the entire result set.
* EXAMPLE: reportQuery.addSum("managerSalary", expBuilder.get("manager").get("salary"));
*/
public void addSum(String itemName, Expression attributeExpression) {
addItem(itemName, attributeExpression.sum());
}
/**
* PUBLIC:
* Add the sum value of the attribute to be included in the result and
* return it as the specified resultType.
* Aggregation functions can be used with a group by, or on the entire result set.
* EXAMPLE: reportQuery.addSum("managerSalary", expBuilder.get("manager").get("salary"), Float.class);
*/
public void addSum(String itemName, Expression attributeExpression, Class resultType) {
addItem(itemName, attributeExpression.sum(), resultType);
}
/**
* PUBLIC:
* Add the variance value of the attribute to be included in the result.
* Aggregation functions can be used with a group by, or on the entire result set.
* EXAMPLE: reportQuery.addVariance("salary");
*/
public void addVariance(String itemName) {
addVariance(itemName, getExpressionBuilder().get(itemName));
}
/**
* PUBLIC:
* Add the variance value of the attribute to be included in the result.
* Aggregation functions can be used with a group by, or on the entire result set.
* EXAMPLE: reportQuery.addVariance("managerSalary", expBuilder.get("manager").get("salary"));
*/
public void addVariance(String itemName, Expression attributeExpression) {
addItem(itemName, attributeExpression.variance());
}
/**
* PUBLIC: Call a constructor for the given class with the results of this query.
* @param constructorClass
*/
public ConstructorReportItem beginAddingConstructorArguments(Class constructorClass){
ConstructorReportItem citem = new ConstructorReportItem(constructorClass.getName());
citem.setResultType(constructorClass);
//add directly to avoid addToConstructorItem behaviour
getItems().add(citem);
//Bug2804042 Must un-prepare if prepared as the SQL may change.
setIsPrepared(false);
this.addToConstructorItem=true;
return citem;
}
/**
* PUBLIC: Call a constructor for the given class with the results of this query.
* @param constructorClass
* @param constructorArgTypes - sets the argument types to be passed to the constructor.
*/
public ConstructorReportItem beginAddingConstructorArguments(Class constructorClass, Class[] constructorArgTypes){
ConstructorReportItem citem =beginAddingConstructorArguments(constructorClass);
citem.setConstructorArgTypes(constructorArgTypes);
return citem;
}
/**
* INTERNAL:
* Construct a result from a row. Either return a ReportQueryResult or just the attribute.
*/
public Object buildObject(AbstractRecord row, Vector toManyJoinData) {
ReportQueryResult reportQueryResult = new ReportQueryResult(this, row, toManyJoinData);
//GF_ISSUE_395
if (this.returnedKeys != null){
if (this.returnedKeys.contains(reportQueryResult.getResultKey())){
return RESULT_IGNORED; //distinguish between null values and thrown away duplicates
} else {
this.returnedKeys.add(reportQueryResult.getResultKey());
}
}
//end GF_ISSUE_395
if (this.shouldReturnSingleAttribute()) {
return reportQueryResult.getResults().firstElement();
}
if (this.shouldReturnWithoutReportQueryResult()){
if (reportQueryResult.getResults().size() == 1){
return reportQueryResult.getResults().firstElement();
}
return reportQueryResult.toArray();
}
return reportQueryResult;
}
/**
* INTERNAL:
* Construct a container of ReportQueryResult from the rows.
* If only one result or value was asked for only return that.
*/
public Object buildObjects(Vector rows) {
if (shouldReturnSingleResult() || shouldReturnSingleValue()) {
if (rows.isEmpty()) {
return null;
}
ReportQueryResult result = (ReportQueryResult)buildObject((AbstractRecord)rows.firstElement(), rows);
if (shouldReturnSingleValue()) {
return result.elements().nextElement();
}
return result;
}
ContainerPolicy containerPolicy = getContainerPolicy();
Object reportResults = containerPolicy.containerInstance(rows.size());
// GF_ISSUE_395
if (shouldDistinctBeUsed()){
this.returnedKeys = new HashSet();
}
//end GF_ISSUE
//If only the attribute is desired, then buildObject will only get the first attribute each time
for (Enumeration rowsEnum = rows.elements(); rowsEnum.hasMoreElements();) {
// GF_ISSUE_395
Object result = buildObject((AbstractRecord)rowsEnum.nextElement(), rows);
if (result != RESULT_IGNORED){
containerPolicy.addInto(result, reportResults, getSession());
}
//end GF_ISSUE
}
return reportResults;
}
/**
* INTERNAL:
* The cache check is done before the prepare as a hit will not require the work to be done.
*/
protected Object checkEarlyReturnImpl(AbstractSession session, AbstractRecord translationRow) {
// Check for in-memory only query.
if (shouldCheckCacheOnly()) {
throw QueryException.cannotSetShouldCheckCacheOnlyOnReportQuery();
} else {
return null;
}
}
/**
* INTERNAL: Required for a very special case of bug 2612185:
* ReportItems from parallelExpressions, on a ReportQuery which is a subQuery,
* which is being batch read.
* In a batch query the selection criteria is effectively cloned twice, meaning
* the ReportItems need to be cloned an extra time also to stay in sync.
* Each call to copiedVersionFrom() will take O(1) time as the expression was
* already cloned.
*/
public void copyReportItems(Dictionary alreadyDone) {
items = (Vector)items.clone();
for (int i = items.size() - 1; i >= 0; i--) {
ReportItem item = (ReportItem)items.elementAt(i);
Expression expression = item.getAttributeExpression();
if ((expression != null) && (alreadyDone.get(expression.getBuilder()) != null)) {
expression = expression.copiedVersionFrom(alreadyDone);
}
items.set(i, new ReportItem(item.getName(), expression));
}
if (groupByExpressions != null) {
groupByExpressions = (Vector)groupByExpressions.clone();
for (int i = groupByExpressions.size() - 1; i >= 0; i--) {
Expression item = (Expression)groupByExpressions.elementAt(i);
if (alreadyDone.get(item.getBuilder()) != null) {
groupByExpressions.set(i, item.copiedVersionFrom(alreadyDone));
}
}
}
if (orderByExpressions != null) {
for (int i = orderByExpressions.size() - 1; i >= 0; i--) {
Expression item = (Expression)orderByExpressions.elementAt(i);
if (alreadyDone.get(item.getBuilder()) != null) {
orderByExpressions.set(i, item.copiedVersionFrom(alreadyDone));
}
}
}
}
/**
* PUBLIC:
* Set if the query results should contain the primary keys or each associated object.
* This make retrieving the real object easier.
* By default they are not retrieved.
*/
public void dontRetrievePrimaryKeys() {
setShouldRetrievePrimaryKeys(false);
//Bug2804042 Must un-prepare if prepared as the SQL may change.
setIsPrepared(false);
}
/**
* PUBLIC:
* Don't simplify the result by returning the single attribute. Wrap in a ReportQueryResult.
*/
public void dontReturnSingleAttribute() {
if (shouldReturnSingleAttribute()) {
returnChoice = 0;
}
}
/**
* PUBLIC:
* Simplifies the result by only returning the first result.
* This can be used if it known that only one row is returned by the report query.
*/
public void dontReturnSingleResult() {
if (shouldReturnSingleResult()) {
returnChoice = 0;
}
}
/**
* PUBLIC:
* Simplifies the result by only returning a single value.
* This can be used if it known that only one row is returned by the report query and only a single item is added
* to the report.
*/
public void dontReturnSingleValue() {
if (shouldReturnSingleValue()) {
returnChoice = 0;
}
}
/**
* PUBLIC:
* Simplifies the result by only returning a single value.
* This can be used if it known that only one row is returned by the report query and only a single item is added
* to the report.
*/
public void dontReturnWithoutReportQueryResult() {
if (shouldReturnWithoutReportQueryResult()) {
returnChoice = 0;
}
}
/**
* PUBLIC:
* Used in conjunction with beginAddingConstructorArguments to signal that expressions should no longer be
* be added to the collection used in the constructor
*
* Get the rows and build the object from the rows.
* @exception DatabaseException - an error has occurred on the database
* @return Vector - collection of objects resulting from execution of query.
*/
public void endAddingToConstructorItem(){
this.addToConstructorItem=false;
}
/**
* INTERNAL:
* Execute the query.
* Get the rows and build the object from the rows.
* @exception DatabaseException - an error has occurred on the database
* @return Vector - collection of objects resulting from execution of query.
*/
public Object executeDatabaseQuery() throws DatabaseException {
// ensure a pessimistic locking query will go down the write connection
if (isLockQuery() && getSession().isUnitOfWork()) {
UnitOfWorkImpl unitOfWork = (UnitOfWorkImpl)getSession();
// Note if a nested unit of work this will recursively start a
// transaction early on the parent also.
if (isLockQuery()) {
if ((!unitOfWork.getCommitManager().isActive()) && (!unitOfWork.wasTransactionBegunPrematurely())) {
unitOfWork.beginTransaction();
unitOfWork.setWasTransactionBegunPrematurely(true);
}
}
}
if (getContainerPolicy().overridesRead()) {
return getContainerPolicy().execute();
}
if (getQueryId() == 0) {
setQueryId(getSession().getNextQueryId());
}
Vector rows = getQueryMechanism().selectAllReportQueryRows();
// If using -m joins, must set all rows.
return buildObjects(rows);
}
/**
* INTERNAL:
* Return the group bys.
*/
public Vector getGroupByExpressions() {
return groupByExpressions;
}
/**
* INTERNAL:
* Return the Having expression.
*/
public Expression getHavingExpression() {
return havingExpression;
}
/**
* INTERNAL:
* return a collection of expressions if PK's are used.
*/
public Vector getQueryExpressions() {
Vector fieldExpressions = new Vector(getItems().size());
// For bug 3115576 and an EXISTS subquery only need to return a single field.
if (shouldRetrieveFirstPrimaryKey()) {
if (!getDescriptor().getPrimaryKeyFields().isEmpty()) {
fieldExpressions.addElement(getDescriptor().getPrimaryKeyFields().get(0));
}
}
if (shouldRetrievePrimaryKeys()) {
fieldExpressions.addAll(getDescriptor().getPrimaryKeyFields());
}
return fieldExpressions;
}
/**
* INTERNAL:
* return a collection of expressions from the items. Ignore the null (place holders).
*/
public Vector getItemExpressions() {
Vector fieldExpressions = new Vector(getItems().size());
// For bug 3115576 and an EXISTS subquery only need to return a single field.
if (shouldRetrieveFirstPrimaryKey()) {
if (!getDescriptor().getPrimaryKeyFields().isEmpty()) {
fieldExpressions.addElement(getDescriptor().getPrimaryKeyFields().get(0));
}
}
if (shouldRetrievePrimaryKeys()) {
fieldExpressions.addAll(getDescriptor().getPrimaryKeyFields());
}
for (Enumeration itemsEnum = getItems().elements(); itemsEnum.hasMoreElements();) {
ReportItem item = (ReportItem)itemsEnum.nextElement();
Expression fieldExpression = item.getAttributeExpression();
if (fieldExpression != null) {
fieldExpressions.addElement(fieldExpression);
}
}
return fieldExpressions;
}
/**
* INTERNAL:
* @return ReportQueryItems defining the attributes to be read
*/
public Vector getItems() {
return items;
}
/**
* INTERNAL:
* Clear the ReportQueryItems
*/
public void clearItems() {
items = new Vector();
setIsPrepared(false);
}
/**
* INTERNAL:
* Lazily initialize and return the names of the items requested for use in each result object
*/
public Vector getNames() {
if (names == null) {
names = new Vector();
for (Enumeration e = getItems().elements(); e.hasMoreElements();) {
names.addElement(((ReportItem)e.nextElement()).getName());
}
}
return names;
}
/**
* PUBLIC:
* Return if this is a report query.
*/
public boolean isReportQuery() {
return true;
}
/**
* INTERNAL:
* Prepare the receiver for execution in a session.
* Initialize each item with its DTF mapping
*/
protected void prepare() throws QueryException {
// Oct 19, 2000 JED
// Added exception to be thrown if no attributes have been added to the query
if (getItems().size() > 0) {
try {
for (Enumeration itemsEnum = getItems().elements(); itemsEnum.hasMoreElements();) {
((ReportItem)itemsEnum.nextElement()).initialize(this);
}
} catch (QueryException exception) {
exception.setQuery(this);
throw exception;
}
} else {
if ((!shouldRetrievePrimaryKeys()) && (!shouldRetrieveFirstPrimaryKey())) {
throw QueryException.noAttributesForReportQuery(this);
}
}
super.prepare();
}
/**
* INTERNAL:
* Prepare a report query with a count defined on an object attribute.
* Added to fix bug 3268040, addCount(objectAttribute) not supported.
*/
protected void prepareObjectAttributeCount(Dictionary clonedExpressions) {
prepareObjectAttributeCount(getItems(), clonedExpressions);
}
private void prepareObjectAttributeCount(List items, Dictionary clonedExpressions) {
int numOfReportItems = items.size();
//gf675: need to loop through all items to fix all count(..) instances
for (int i =0;i
// treat COUNT(entity) as COUNT(entity.pk)
DatabaseMapping pk = getMappingOfFirstPrimaryKey(newDescriptor);
Expression countArg = baseExp.get(pk.getAttributeName());
if (distinctUsed) {
countArg = countArg.distinct();
}
count.setBaseExpression(countArg);
count.getChildren().setElementAt(countArg, 0);
} else if (!distinctUsed) {
// case 2: composite PK, but no DISTINCT =>
// pick a PK column for the COUNT aggregate
DatabaseMapping pk = getMappingOfFirstPrimaryKey(newDescriptor);
Expression countArg = baseExp.get(pk.getAttributeName());
while (pk.isAggregateObjectMapping()) {
newDescriptor = ((AggregateObjectMapping)pk).getReferenceDescriptor();
pk = getMappingOfFirstPrimaryKey(newDescriptor);
countArg = countArg.get(pk.getAttributeName());
}
count.setBaseExpression(countArg);
count.getChildren().setElementAt(countArg, 0);
} else if (!outerJoin) {
// case 3: composite PK and DISTINCT, but no
// outer join => previous solution using
// COUNT(*) and EXISTS subquery
// If this is a subselect baseExp is yet uncloned,
// and will miss out if moved now from items into a selection criteria.
if (clonedExpressions != null) {
if (clonedExpressions.get(baseExp.getBuilder()) != null) {
baseExp = (QueryKeyExpression)baseExp.copiedVersionFrom(clonedExpressions);
} else {
baseExp = (QueryKeyExpression)baseExp.rebuildOn(getExpressionBuilder());
}
}
// Now the reference class of the query needs to be reversed.
// See the bug description for an explanation.
ExpressionBuilder countBuilder = baseExp.getBuilder();
ExpressionBuilder outerBuilder = new ExpressionBuilder();
ReportQuery subSelect = new ReportQuery(getReferenceClass(), countBuilder);
subSelect.setShouldRetrieveFirstPrimaryKey(true);
// Make sure the outerBuilder does not appear on the left of the subselect.
// Putting a builder on the left is desirable to trigger an optimization.
if (getSelectionCriteria() != null) {
outerBuilder.setQueryClass(newDescriptor.getJavaClass());
subSelect.setSelectionCriteria(baseExp.equal(outerBuilder).and(getSelectionCriteria()));
} else {
subSelect.setSelectionCriteria(baseExp.equal(outerBuilder));
}
setSelectionCriteria(outerBuilder.exists(subSelect));
count.setBaseExpression(outerBuilder);
count.getChildren().setElementAt( outerBuilder, 0);
setReferenceClass(newDescriptor.getJavaClass());
changeDescriptor(getSession());
} else {
// case 4: composite PK, DISTINCT, outer join =>
// not supported, throw exception
throw QueryException.distinctCountOnOuterJoinedCompositePK(
newDescriptor, this);
}
}
}
}
}
}
/** */
private DatabaseMapping getMappingOfFirstPrimaryKey(ClassDescriptor descriptor) {
if (descriptor != null) {
for (Iterator i = descriptor.getMappings().iterator(); i.hasNext(); ) {
DatabaseMapping m = (DatabaseMapping)i.next();
if (m.isPrimaryKeyMapping()) {
return m;
}
}
}
return null;
}
/**
* INTERNAL:
* Prepare the mechanism.
*/
protected void prepareSelectAllRows() {
prepareObjectAttributeCount(null);
getQueryMechanism().prepareReportQuerySelectAllRows();
}
/**
* INTERNAL:
* Prepare the receiver for being printed inside a subselect.
* This prepares the statement but not the call.
*/
public synchronized void prepareSubSelect(AbstractSession session, AbstractRecord translationRow, Dictionary clonedExpressions) throws QueryException {
if (isPrepared()) {
return;
}
setIsPrepared(true);
setSession(session);
setTranslationRow(translationRow);
checkDescriptor(getSession());
if (descriptor.isAggregateDescriptor()) {
// Not allowed
throw QueryException.aggregateObjectCannotBeDeletedOrWritten(descriptor, this);
}
try {
for (Enumeration itemsEnum = getItems().elements(); itemsEnum.hasMoreElements();) {
((ReportItem)itemsEnum.nextElement()).initialize(this);
}
} catch (QueryException exception) {
exception.setQuery(this);
throw exception;
}
prepareObjectAttributeCount(clonedExpressions);
getQueryMechanism().prepareReportQuerySubSelect();
setSession(null);
setTranslationRow(null);
}
/**
* PUBLIC:
* Set if the query results should contain the primary keys or each associated object.
* This make retrieving the real object easier.
* By default they are not retrieved.
*/
public void retrievePrimaryKeys() {
setShouldRetrievePrimaryKeys(true);
//Bug2804042 Must un-prepare if prepared as the SQL may change.
setIsPrepared(false);
}
/**
* PUBLIC:
* Simplify the result by returning a single attribute. Don't wrap in a ReportQueryResult.
*/
public void returnSingleAttribute() {
returnChoice = ShouldReturnSingleAttribute;
}
/**
* PUBLIC:
* Simplifies the result by only returning the first result.
* This can be used if it known that only one row is returned by the report query.
*/
public void returnSingleResult() {
returnChoice = ShouldReturnSingleResult;
}
/**
* PUBLIC:
* Simplifies the result by only returning a single value.
* This can be used if it known that only one row is returned by the report query and only a single item is added
* to the report.
*/
public void returnSingleValue() {
returnChoice = ShouldReturnSingleValue;
}
/**
* PUBLIC:
* Simplifies the result by only returning a single value.
* This can be used if it known that only one row is returned by the report query and only a single item is added
* to the report.
*/
public void returnWithoutReportQueryResult() {
this.returnChoice = ShouldReturnWithoutReportQueryResult;
}
/**
* PUBLIC:
* Set if the query results should contain the primary keys or each associated object.
* This make retrieving the real object easier.
* By default they are not retrieved.
*/
public void setShouldRetrievePrimaryKeys(boolean shouldRetrievePrimaryKeys) {
this.shouldRetrievePrimaryKeys = (shouldRetrievePrimaryKeys ? FULL_PRIMARY_KEY : NO_PRIMARY_KEY);
}
/**
* ADVANCED:
* Sets if the query results should contain the first primary key of each associated object.
* Usefull if this is an EXISTS subquery and you don't care what fields are returned
* so long as it is a single field.
* The default value is false.
* This should only be used with a subquery.
*/
public void setShouldRetrieveFirstPrimaryKey(boolean shouldRetrieveFirstPrimaryKey) {
this.shouldRetrievePrimaryKeys = (shouldRetrieveFirstPrimaryKey ? FIRST_PRIMARY_KEY : NO_PRIMARY_KEY);
}
/**
* PUBLIC:
* Simplifies the result by only returning the attribute (as opposed to wrapping in a ReportQueryResult).
* This can be used if it is known that only one attribute is returned by the report query.
*/
public void setShouldReturnSingleAttribute(boolean newChoice) {
if (newChoice) {
returnSingleAttribute();
} else {
dontReturnSingleAttribute();
}
}
/**
* PUBLIC:
* Simplifies the result by only returning the first result.
* This can be used if it known that only one row is returned by the report query.
*/
public void setShouldReturnSingleResult(boolean newChoice) {
if (newChoice) {
returnSingleResult();
} else {
dontReturnSingleResult();
}
}
/**
* PUBLIC:
* Simplifies the result by only returning a single value.
* This can be used if it known that only one row is returned by the report query and only a single item is added
* to the report.
*/
public void setShouldReturnSingleValue(boolean newChoice) {
if (newChoice) {
returnSingleValue();
} else {
dontReturnSingleValue();
}
}
/**
* PUBLIC:
* Simplifies the result by returning a nested list instead of the ReportQueryResult.
* This is used by EJB 3.
*/
public void setShouldReturnWithoutReportQueryResult(boolean newChoice) {
if (newChoice) {
returnWithoutReportQueryResult();
} else {
dontReturnWithoutReportQueryResult();
}
}
/**
* PUBLIC:
* Return if the query results should contain the primary keys or each associated object.
* This make retrieving the real object easier.
*/
public boolean shouldRetrievePrimaryKeys() {
return (shouldRetrievePrimaryKeys == FULL_PRIMARY_KEY);
}
/**
* PUBLIC:
* Return if the query results should contain the first primary key of each associated object.
* Usefull if this is an EXISTS subquery and you don't care what fields are returned
* so long as it is a single field.
*/
public boolean shouldRetrieveFirstPrimaryKey() {
return (shouldRetrievePrimaryKeys == FIRST_PRIMARY_KEY);
}
/**
* PUBLIC:
* Answer if we are only returning the attribute (as opposed to wrapping in a ReportQueryResult).
* This can be used if it is known that only one attribute is returned by the report query.
*/
public boolean shouldReturnSingleAttribute() {
return returnChoice == ShouldReturnSingleAttribute;
}
/**
* PUBLIC:
* Simplifies the result by only returning the first result.
* This can be used if it known that only one row is returned by the report query.
*/
public boolean shouldReturnSingleResult() {
return returnChoice == ShouldReturnSingleResult;
}
/**
* PUBLIC:
* Simplifies the result by only returning a single value.
* This can be used if it known that only one row is returned by the report query and only a single item is added
* to the report.
*/
public boolean shouldReturnSingleValue() {
return returnChoice == ShouldReturnSingleValue;
}
/**
* PUBLIC:
* Simplifies the result by returning a nested list instead of the ReportQueryResult.
* This is used by EJB 3.
*/
public boolean shouldReturnWithoutReportQueryResult() {
return returnChoice == ShouldReturnWithoutReportQueryResult;
}
}