Why Object-Oriented DBMS? Abstract data types (relational data model is too simple). Interface with host programming language Managing large number of objects: –Encapsulation, reusability Semantic networks: the worls is actually a complex hierarchy.

SQL Doesn’t Work in Scotland Select name From Employee Where name > “Mc” and name < “Md” OrderBy name Doesn’t work for the McAndish clan: (they’re all the same!) McAndish MacAndish M’Andish Comparison based on ASCII Other countries are even more problematic.

Two Approaches to OO 1. Object Query Language: part of ODMG (that also gave us OQL) 2. Adding object features to SQL (row types and ADT’s). Also known as Object-relational systems. The difference: How important is a relation to you? How important is SQL to you?

Object Definition Language Resembles C++ (and Smalltalk). Basic design paradigm in ODL: –Model objects and their properties. For abstraction purposes: –Group objects into classes. What qualifies as a good class? –Objects should have common properties.

ODL Class Declarations Interface { attributes: ; relationships ; methods (in, out, inout, exceptions) }

Example Method Interface Movie { (extent Movies) attribute string title; relationship Set stars inverse Star::starredIn; float lengthInHours raises(noLengthFound); starNames (out Set ); otherMovies (in Star, out Set ) raises (noSuchStar); } Note: defining signature, not implementation. Overloading allowed.

OQL: Object Query Language SELECT can construct new objects, arbitrary structures FROM tuple variables can range over any collection; may have subqueries. WHERE pretty much the same as in SQL

Types in OQL Basic types: Atomic types (e.g., string, integer, …) Enumeration types (Monday, Tuesday, Wednesday….) Constructors: can be applied without limitations. Set: (1, 5, 6) Bag: (1, 1, 5, 6, 6 ) List: (1, 5, 6, 1, 6 ) Array: Integer[17] Struct: (name: string, address: string) struct( name: “John”, childrenAges: bag (1,1,2,2)) Declaring types for objects

Path Expressions Path expressions are needed in order to access components of objects. Attributes: a.p is the value of the attribute a of p. Relationships: a.p is the object or collection of objects related to a by p. Methods: a.p is the result of applying p to a (perhaps with a parameter). Also possible to have longer expressions: a.father.wife.child.father.wife….

Select-From-Where in OQL (simple) Example: SELECT s.name FROM Movies m, m.stars s WHERE m.title = “Sleepless in Seattle” Note: this looks a lot more procedural than SQL.

Complications in the FROM Clause SELECT a.phoneNumber FROM Movies m, (SELECT m.address FROM m.stars WHERE m.city=“Los Angeles”) AS a WHERE m.title = “Sleepless in Seattle” The FROM clause can contain arbitrary subqueries that return collections.

Complex Output Types The SELECT clause can create complex structures: SELECT Struct: (address: a, phoneNumber: a.phoneNumber) FROM Movies m, (SELECT m.address FROM m.stars WHERE m.city=“Los Angeles”) AS a WHERE m.title = “Sleepless in Seattle”

Other Features of OQL Ordering of the results ORDER BY m.title, m.year. Subqueries wherever a collection is appropriate. Quantifier expressions: FOR ALL x IN S : C(x) EXISTS x IN S:C(x) Aggregation, grouping and HAVING clauses. Set operators: UNION, INTERSECT, EXCEPT (different if operating on bags or sets). Remove duplicates: SELECT DISTINCT.

Interface with Host Language OQL is much more tightly integrated with the host language. OQL produces objects (of various types). One can simply assign the objects as values to variables with the appropriate types. No need for special interface. ELEMENT: turns a bag of one element into the single element: var1 = ELEMENT (SELECT m FROM Movies m WHERE title=“sleepless in Seattle”);

Handling Sets or Bags First: turn them into lists (using ORDER BY). Then: use host language operations to go through them. Example: movieList = select m FROM Movies m WHERE m.year > 1990 ORDER BY m.title, m.year; numberOfMovies = COUNT (movieList); for (I=0; I < numberOfMovies; I++) { movie = movieList[I]; do something with it. }

Another Example Schema: one relation Tour (S, E, C ) meaning: company C offers a tour from city S to city E. Query: find for every city, the companies giving tours of itself. SELECT [S: x.S C : (SELECT y.C from y in Tour WHERE x.E=y.S and x.S=y.E) ] FROM x in Tour

Final Example SELECT [S:x.S, C: (SELECT y.C FROM y in Tour, y’ in Tour WHERE x.S=y.S and x.E=y.E and x.E=y’.S and x.S=y’.E and y.C = y’.C)] from x in Tour. Find companies offering tours from city S to some destination E and back to S.

Row Types in SQL-3 Row types define types for tuples, and they can be nested. CREATE ROW TYPE AddressType{ street CHAR(50), city CHAR(25), zipcode CHAR(10) } CREATE ROW TYPE PersonType{ name CHAR(30), address AddressType, phone phoneNumberType }

Relations as Row Types CREATE TABLE Person OF TYPE PersonType; Recall: row types can be nested! Accessing components of a row type: (double dots) SELECT Person.name, Person.address..city FROM Person WHERE Person.address..street LIKE ‘%Mountain%’

References We can define attributes of a row type to reference objects of other row types: CREATE ROW TYPE Company( name char(30), address addressType, president REF(PersonType) ); Following references: SELECT president->name FROM Company WHERE president->address..city=“Seattle”

Abstract Data Types in SQL3 Row types provide a lot of the functionality of objects: allow us to modify objects (unlike OQL), but do not provide encapsulation. We can modify objects arbitrarily using SQL3 commands. In OQL: we can query, but not modify only via methods. Abstract data types: are used as components of tuples. CREATE TYPE ( list of attributes and their types optional declaration of the comparison functions: =, < declaration of methods for the type );

Address ADT CREATE TYPE AddressADT ( street CHAR(50), city CHAR(20), EQUALS addrEq, LESS THAN addrLT FUNCTION fullAddr (a: AddressADT) RETURNS CHAR(100); :z CHAR(10); BEGIN :z = findZip(:a.street, :a.city); RETURN (….) END; DECLARE EXTERNAL findZip CHAR(50) CHAR(20) RETURNS CHAR(10) LANGUAGE C; ); Encapsulation is obtained by making methods public/private

Differences Between OODB Approaches Programming environment: much more closely coupled in OQL/ODL than in SQL3. Changes to objects are done via the programming language in OQL, and via SQL statements in SQL3. Role of relations: still prominent in SQL 3 Row types are really tuples, ADT’s describe attributes. In OQL: sets, bags and structures are fundamental. Encapsulation: exists in OQL; not really supported by row types in SQL3, but are supported by ADT’s.