Table of Contents

  1. About this Guide
  2. Inheritance and the Object Relational Map
  3. A Class Hierarchy
  4. Overriding the Default Behavior

About this Guide

The inheritance guide is intended to help those who want to use class hierarchies with Torque. For example, you would use Torque's inheritance support if you have a class that extends another class, and both classes should be saved into and read from the database.

Inheritance and the Object Relational Map

Torque can handle object-oriented inheritance. There are generally considered to be 3 methods of object-relational mapping designs. Torque uses one of the fastest, mapping all objects in a class hierarchy to a single table. All attributes for every class in the hierarchy are stored in the table. Consider an abstract ComputerComponent class that has Monitor and Keyboard subclasses. There would only be one table - both Monitor and Keyboard objects would be persisted to the same place. The table would consist of all ComputerComponent attributes, any unique Monitor attributes, and any unique Keyboard attributes. Keyboard table rows would have NULL for any unique Monitor data columns, and vice versa.

The other fast method is to map each concrete class to a distinct table. Every object stores all attributes in a single row in the class table. An example would be that if we had a Kitchen class that inherited from Room, two tables would be needed for storage. The Kitchen table would contain all of the columns of the Room table, plus any additional data columns needed to describe the additional Kitchen attributes.

The slowest, but most object-oriented method is to store each class in its own table. Only attributes that are added to a derived class are stored in its table. The persistence layer would need to join tables to read an object out of storage. Saving objects would be more complex, because objects will need to be distributed across multiple tables. For our Kitchen and Room example, there would also be two tables, Kitchen and Room, but the Kitchen table would only contain those attributes which weren't part of the Room class.

One of the advantages of the first method (the one Torque uses) is that it does not require joins like the third method described above. Another advantage is that the data model is easier to maintain than the second method. It falls short in modelling a class hierarchy where the related classes have a non intersecting collection of attributes, as in this case a row in the table will have several null columns.

For more information, visit Scott Ambler's excellent web site, AmbySoft.com, where he discusses object mapping to relational databases.

A Class Hierarchy

       A
       |
     -----
    |     |
    B     C
    |
    D

There are two ways that are built into the torque generated Peers in order to specify what class a particular row in the table A represents. A row will need to have some information that can distinguish the class. You should specify the column in the table that serves this purpose with the attribute "inheritance"

<table name="A"...>
  ...
  <column name="FOO" inheritance="single" type="VARCHAR".../>
</table>

In this case you would need to specify the full className in column FOO, so the valid values of FOO would be:

com.mycompany.project.om.A
com.mycompany.project.om.B
com.mycompany.project.om.C
com.mycompany.project.om.D

This is slightly inefficient in storage and also generates some inefficient code in the Peers because the Peer cannot know what classes are available until it gets each record and so will call Class.forName(FOO's value) for each row.

The efficiency can be improved in the case where the class hierarchy is known, which would be in most circumstances. So you can specify the classes in the xml specification:

<table name="A"...>
  ...
  <column name="FOO" inheritance="single" type="CHAR" size="1"...>
    <inheritance key="B" class="B" extends="com.mycompany.project.om.A"/>
    <inheritance key="C" class="C" extends="com.mycompany.project.om.A"/>
    <inheritance key="D" class="D" extends="com.mycompany.project.om.B"/>
  </column>
</table>

where in the above we are using NULL (or any other value) to stand for class "A". An numeric column could also be used for the key. Using the above method, torque will cache a copy of each class, so the Class.forName is only done during APeer's initial load into memory.

Overriding the Default Behavior

The following example came from Scarab, an issue tracking system. In Scarab a class hierarchy definition was described in a few tables, this provided for an evolving hierarchy. This arrangement can be provided for using the following extensions. In the xml specification, the column responsible for determining the class is marked using the inheritance="single" attribute.

<table name="SCARAB_ISSUE_ATTRIBUTE_VALUE" idMethod="none"
        javaName="AttributeValue">
    <column name="ISSUE_ID" primaryKey="true" required="true"
            type="INTEGER"/>
    <column name="ATTRIBUTE_ID" primaryKey="true" required="true"
            type="INTEGER" inheritance="single"/>
    <column name="OPTION_ID" required="false" type="INTEGER"/>
...
    <foreign-key foreignTable="SCARAB_ISSUE">
        <reference local="ISSUE_ID" foreign="ISSUE_ID"/>
    </foreign-key>
    <foreign-key foreignTable="SCARAB_ATTRIBUTE">
        <reference local="ATTRIBUTE_ID" foreign="ATTRIBUTE_ID"/>
    </foreign-key>
    <foreign-key foreignTable="SCARAB_ATTRIBUTE_OPTION">
        <reference local="OPTION_ID" foreign="OPTION_ID"/>
    </foreign-key>
...
</table>

It might be interesting to note that the column responsible for the determining the class is also a primary and foreign key. Marking the column this way will cause torque to generate an BaseAttributeValuePeer.getOMClass method.The code in this method will be attempting to create a class from the information provided in column which is an integer. This is obviously wrong, but it gives us a method to override to provide the correct information.

So in AttributeValuePeer, we override the method:

/**
 * Get the className appropriate for a row in the
 * SCARAB_ISSUE_ATTRIBUTE_VALUE table
 */
public static Class getOMClass(Record record, int offset)
    throws Exception
{
    NumberKey attId = new NumberKey(record.getValue(offset-1 + 2).asString());
    Attribute attribute = Attribute.getInstance(attId);
    String className = attribute.getAttributeType().getJavaClassName();

    TurbineGlobalCacheService tgcs =
        (TurbineGlobalCacheService) TurbineServices
        .getInstance().getService(GlobalCacheService.SERVICE_NAME);

    String key = getClassCacheKey(className);
    Class c = null;
    try
    {
        c = (Class) tgcs.getObject(key).getContents();
    }
    catch (ObjectExpiredException oee)
    {
        c = Class.forName(className);
        tgcs.addObject(key, new CachedObject(c));
    }
    return c;
}

where in the above method, we use the foreign key(s) to traverse the tables to get the class information. Then we cache the Class to avoid the inefficiency of Class.forName on each row. (We also cache the contents of the class hierarchy tables, since the dataset is quite small and static.)