Object SQL, Pointers and Encapsulation

Oracle Database Tips by Donald Burleson

One of the basic constructs of object-oriented programming is encapsulation. Encapsulation is defined as the ability to access objects only via their behaviors. This is contradictory to a basic principle the relational database model, data independence, which says that any data may be accessed in an ad-hoc, independent fashion.

At first glance, it seems that these two concepts cannot be reconciled, because it would be impossible to have data tables which are independent of the application, while at the same time supporting encapsulation, which tightly-couples the objects and their behaviors. However, these concepts are not contradictory. Because the behaviors are stored in the database, they are not external, and will not jeopardize the independence of applications from data.

For example, one could only access the CREDIT_RATING field in the CUSTOMER table by invoking the PLACE_ORDER behavior. The SQL language, of course, would allow access and update to any data items which are allowed within the system security tables. Any authorized user could view the credit rating of a customer without ever invoking an object-oriented method.

Another conceptual limitation of SQL which has legitimate ramifications for object-oriented databases is the inability of SQL to associate a behavior with a data item. The properties of an object and its operational semantics must be coded within an external entity (the application program), and SQL has no built-in method for incorporating behaviors into tables. However, there is a solution to this problem, and many of the object/relational database vendors have created "methods" for database objects that have a one-for-one correspondence with the SQL operators. For example, a database might automatically create a method called insert_customer, which would invoke the appropriate SQL statement to insert a row into a customer table. A database might automatically create methods to insert, update and delete rows from the target table, much as C++ allows for constructors and destructors for objects. While this works fine for a simple operation, there is still a problem for more sophisticated methods that access and alter more than one data column.

Many of the built-in functions of SQL also violate the encapsulation rule. For example, instead of writing a methods to compute the gross pay for an employee, we could directly use SQL to perform this operation, thereby bypassing the method:

SELECT
     hours_worked*payrate
FROM
     timesheet, payrates
WHERE
     emp_id = 123
AND
     week = '03/98';

The same is true when using the SUM, AVG and any one of the dozens of other SQL functions that are offered by the major relational database vendors.

SQL and Pointers

One of the greatest mismatches with SQL and objects lies in the arena of pointers. The introduction of pointers into the relational model has led to a situation where the declarative nature of SQL is being radically changed.

For example, the use of the DEREF operator in the new object/relational SQL, allows an SQL statement to de-reference a row pointer, essentially navigating from one table to the next table. Rather than relying on the SQL optimizer to take care of the database access, the developer now has the option of embedding SQL statements into their programs that will allow them to navigate through the database, visiting tables that have been linked together with pointers. This is a very foreign idea for most SQL developers but it is now a reality.

For example, the following SQL could be used to navigate from a customer to the order rows for the customer:

SELECT
   DEREF(order_list)
FROM
   CUSTOMER
WHERE
   customer_id = "JONES";

This would be the equivalent to the traditional SQL:

SELECT
   order_stuff
FROM
   CUSTOMER, ORDER
WHERE
   customer_id = "JONES"
AND
customer_ID = order_ID;

The SQL becomes even more confounding when we start dealing with the more abstract uses of pointers in a relational/object model. As we recall from earlier chapters, these pointer constructs use object-Ids (OID) and include some very abstract data structures:

Pointers to individual rows in other tables.
Repeating groups of pointers to rows.
Pointers to arrays of pointers to rows.
Pointers to whole tables.
Multidimensional arrays of pointers.

Many of the relational vendors have extended their SQL syntax to provide the following constructs to deal with pointers:

DEREF - This SQL operator accepts an OID and returns the contents of the row that the OID points to:

    SELECT
        DEREF(order_oid)
    FROM
        CUSTOMER;

CAST & MULTISET - These SQL operators casts a multiple input data stream into the appropriate data types for the SQL operation:

INSERT INTO
   COURSE (STUDENT_LIST)
   (CAST
      (MULTISET
          (SELECT
                student_name,
                student_address,
                grade
            FROM
                 GRADE, STUDENT
            WHERE
                 GRADE.course_name = 'CS101'
                 AND
                 GRADE.student_name = STUDENT.student_name
           )
    )
);

With all of these new pointer constructs and extensions to SQL, it will be several years before the mainstream programming community arrives at a general agreement about the use of pointer-based navigation within SQL programming.

SQL and inheritance

As we recall from Chapter 7, the object/relational database allow the developer to create a class hierarchy of related data items, where each item in the class hierarchy is a sub-type of another data type. These types of IS-A relationships, while valid from a data modeling viewpoint, do not have a simple implementation in object/relational databases. Since most object/relational databases do not support hierarchical relationships, it is impossible to directly represent the fact that a database entity has sub-entities.

Once an object has been instantiated, the data items within the object have been defined and the object may be used within SQL to retrieve the data items. The problem is that a separate table must exist for all sub-classes within the class hierarchy, since each my have separate data items. Therefore, if we have a class hierarchy with ten separate classes, there we may have ten separate tables to hold the instantiated objects. This can make the SQL more complicated, especially since we must know the type of the object when we are making the query. In the following example, we are querying the data items from a luxury_sedan vehicle, and we must know that it is of the type "luxury_sedan" in order to formulate the query:

SELECT
        vehicle_number,
        registration_number,
        number_of_doors,
        type_of_leather_upholstery
FROM
        LUXURY_SEDAN
WHERE
   sedan.key = 8383635;

Perhaps future implementation of object/relational databases may remove the SQL restriction of having to explicitly specify the target table by name, but there is no way the any SQL optimizer will be able to tell the sub-type of the target object without some kind of a hint.

Summary

Now that we have addressed the basic issues relating to SQL and objects, let's take a look at how many of the database developers are interfacing objects with SQL by developing object-oriented applications that use relational database to store their object information.

��