Implementation of Nested Relations in a

Database Programming Language

Hongbo HE

School of Cornputer Science

McGill University, Montreal

September 1997

A thesis submitted to the Faculty of Graduate Studies and Research

in partial fuifiUment of the requirements for the degree of
Master of Science in Cornputer Science.

Copyright @ Hongbo HE 1997

 Acquisiions and Acquisitions et
Bibliogcaphic Services senrices bibliographiques

The author has granted a non- L'auteur a accordé une Licence non
exclusive licence allowing the exclusive permettant a la
National Libfary of Canada to Bibliothèque nationale du Cana& de
reproduce, loan, distniute or seli reproduire, prêter, distribuer ou
copies of this thesis in microfoim, vendre des copies de cette thèse sous
paper or electronic formats. la forme de rnicrofiche/film, de
reproduction sur papier ou sur format
électronique.

The author retains ownershrp of the L'auteur conserve la propriété du

copyright in this thesis. Neither the droit d'auteur qui protège cette thèse.
thesis nor substantial extracts from it Ni la thèse ni des elctraits substantiels
may be printed or otherwise de celle-ci ne doivent être imprimés
reproduced without the author's ou autrement reproduits sans son
permission. autorisation.
 Abstract

This t hesis discusses the design and implementatioo of nested relations

in Relix, a relational database programming language. The purpose of this
thesis is to integrate nested relations into Relix.
While a flat relation is defined over a set of atomic attributes, a nested
relation is defined over attributes which can include non-atomic ones, Le.
a data item itself can be a relation. To show t h e power of relational
database systems, it is desirable to have nested relations in Relix. Our
implementation was done using existing relational functionalities of Relix,
without any modification of the physicd data representation. Instead
of Iocusing on nesting and unnesting as the major research direction of
nested relations. we built nested relations on top of Bat relations and
we built nested queries by allowing the domain algebra to subsume the

relat ional algebra.

Users are able to take advantage of nested relations in Relix with only
minimal new syntax being added to the system.
 Résumé

Cette thèse a pour objectif la spécification et l'implémentation des

relations imbriquées dans Relix, un langage de programmation de base de
données relationnelles. Le but de cette thèse est d'intégrer les relations
imbriquées dans Relix.
Une relation plate est definie sur un ensemble d'attributs atomiques,
alon qu'une relation imbriquée est définie sur des attributs qui sont non
atomiqucs,i.e., une donnée pourrait être une relation. Pour montrer la
puissance des systems de base de données relationnelles, il est desirable
d'avoir des relations imbriquées dans Relix. Notre implémentation est
basée sur les fonct ionali tés relationnelles déjà existantes dans Relix, au-
cune modification a u niveau de la représentation physique des données
n'a été apportée. Au lieu de focaliser notre axe de recherche sur les pro-
priétés d'imbrication et de non-imbrication des relations imbriquées, nous
avons construit des requêtes imbriquées permettant à l'algèbre relation-
nelle d'être une composante du domaine algébrique.
Les utilisateurs peuvent tirer profit des relations imbriquées dans Relix
à l'aide d'une nouvelle syntaxe minimale qui a été ajoutée au système.
 Acknowledgements

I would like to express my gratitude to my thesis supervisor, Professor

T. H. Memett, for his attentive guidance, invaluable advice, and endless
patience throughout the research and preparation of this thesis. 1 would
also like to thank him for his financial support.
I would like t o thank my colleagues in the ALDAT [ab, especially Xi-
aoyan Zhao and Rebecca Lui for their assistance on the usage of facilities
in the lab and their consultation on the existing Relix system. Special
thanks goes to Abdelkrim Hebbar who translated the abstract of this
thesis to French and Anne Vogt who proofread this thesis.
1 would also like to t hank al1 the secretaries of the School of Cornputer
Science for their kind help, especially Ms. Josie Vallelonga and Ms. Franca
Cianci.
I wish t o thank al1 my friends dunng my years at McGill, Pung Hay,
Xinan Tang, Shaohua Han and Marcia Cavalcmte for their endless en-
couragement.
Thanks must also go to my father, my brothers for their love and
constant support.
Finally, 1 would like to dedicate this thesis to my mother, for her bless
in my iife to date and forever.

iii
Contents

Abstract

Résumé

Acknowledgements iii

1 Introduction 1
1.1 Relaliional Mode1 ............................. 1
1.1.1 O perat ions on Relations . . . . . . . . . . . . . . . . . . . . . 2
1.12 Operations on Dornains . . . . . . . . . . . . . . . . . . . . . 3
1.2 Object Oriented Model .......................... 3
1.3 Object Relational .........................
Mode1 4
1.4 Nested Relation Mode1 . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.4.1 Nested Relations . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.4.2 Nesting and Umesting . . . . . . . . . . . . . . . . . . . . . . 7
1.4.3 Our Approach . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
1.5 Thesis Aim and Outline . . . . . . . . . . . . . . . . . . . . . . . . . 11

2 Relix 12
2.1 Overview . ................................. 12
2.1.1 Domains and Relations . . . . . . . . . . . . . . . . . . . . . . 13
2.1.2 Basic Commands in Relix . . . . . . . . . . . . . . . . . . . . 14
 2.2 Relational Algebra ............................ 16
2.2.1 Projection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
2.2.2 Selection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
2.2.3 Joins . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
2.3 Domain Algebra . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
2.3.1 HorizontalOperations . . . . . . . . . . . . . . . . . . . . . . 23
2.3.3 Reduction (Vertical Operations) . . . . . . . . . . . . . . . . . 25
2.3.3 Nested Relations . . . . . . . . . . . . . . . . . . . . . . . . . 26
2.4 ijoin, ujoin, sjoin are Associative and Commutative . . . . . . . . . . 27
2.4.1 Definition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
3.4.2 Commutative . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
2.4.3 Associative . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
2.1.4 Another Approach . . . . . . . . . . . . . . . . . . . . . . . . 30

3 User's Manual on Nested Relations 31

3.1 The Nested Relations and Relation Data Type . . . . . . . . . . . . . 31
3.2 Operations on Nested Relations . . . . . . . . . . . . . . . . . . . . . 34
3.2.1 Vertical Operations . . . . . . . . . . . . . . . . . . . . . . . . 34
3.2.2 Horizontal Operations ...................... 40

4 Implementation of Nested Relations 45

4.1 Implementation of Relix ......................... 45
4.1.1 System Relations . . . . . . . . . . . . . . . . . . . . . . . . . 46
4.1.2 Parser and Lnterpreter . . . . . . . . . . . . . . . . . . . . . . 47
4.1.3 lmplementation of Domain Operations . . . . . . . . . . . . . 50
4.3 Declaration and Initialization of Nested Relations . . . . . . . . . . . 53
4.2.1 Declaration of Relation Data Type . . . . . . . . . . . . . . . 53
4.22 Lnitiahation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
 4.3 Operations ................................. 58
4.3.1 Implementation of Reduction . . . . . . . . . . . . . . . . . . 59
4.3.2 Honzootal Operation . . . . . . . . . . . . . . . . . . . . . . . 67

5 Conclusion 74
5.1 Surnmary ................................. 74
5.2 Future Work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75

Bibliography
Chapter 1

Introduction

This thesis discusses the implementation of nested relations in Relix, a relational

database system developed at McGill.
The relational model for representing data was proposed by Codd [Cod701 in
the early seventies. Since then. it has gained an undisputable key position in the
commercial database industry. The nested relational model [Maki71 was developed
as an extension of the relational model and has gained significant importance in non-
traditional database applications (such as C AD/C AM databases, text and pictorial
databases).

1.1 Relational Mode1

In the relational model,information is represented in a table format with the foiiowing
properties:

a Al1 rows are distinct from each other.

a The ordering of the rows is unimportant.

a Each column is unique and the ordering of the columns is immaterial.

CHAPTER 1. INTRODUCTION

The value in each row under a given column is atomic, i.e., it is nondecompos-
able.

Each row is called a tuple and a column is referred to as a domain, A name is

given to the domain of a relation to release the usen from remembering the dornain
ordering of the relation. They are called attributes. From a mathematical perspective,
a relation is a subset of the Cartesian product of its domains.

1.1.1 Operations on Relations

Operations on relations form the relational algebra, and can be thought of as a
collection of methods for building new tables that constitute answers to queries.
Codd defined a set of relationai operations and proved that they are "relationally
completen' [Cod72].
Relations are considered atomic objects in the relational algebra, and access to
tuples within a relation is precluded. Thus the notation and manipulations that must
be done are greatly simplified [Mer84]. The operations are defined as following:

0 unary operations

- projection
- selection

binary operations

- p-joins: applied to relations that are union compatible

- a-joins: support set operations on relations

'An algebra or calculus is reldionally complete if, given any finite collection of relations
Ri,R2,...,Rn in simple normal form, the expressions of the algebra or calculus permit definition of
any relation fiom R i , R a , . .., R, by using a set of N range predicates in one-to-one correspondence
with RI, R2,. . ., R,.
CHAPTER 1. INTRODUCTION

1.1.2 Operations on Domains

The need for arithmetic and similar processing of the values of attributes in individual
tuples is apparent. The domain algebra was proposed [Mer761entirely to avoid tuple-
at-a-time operations for processing attributes in individual tuples. It allows the user
to create new domains from existing ones, and allows the generation of new values
from many values within a tuple or from values along ôn attribute. The domain
algebra operations are defined as:

0 horizontal operations

- Constant
- Rename

- Function
- If- t hen-else

0 vertical operations

- Reduction

- Equivdence Reduction
- Functional Mapping
- Partial Functional Mapping

1.2 Object Oriented Mode1

Object-oriented techniques are becoming popular for designing and implementing user
interfaces, applications and systems. O DBMS (Object-oriented Database Manage-
ment System) is the result of objected-onented techniques irnplemented in database
management systems.
CHAPTER 1. INTROD UCTiON

Object-oriented techniques include the following key points:

a Encapsulation: combiaing data and functions in a single unit, the object.

a Polymorphism: the ability to treat different objects the same way by sending

them the sarne message, which elicits a semantically similar function in each
object.

a Class instant iation: creating different objects of the sarne general description
from the same class.

a Inheritance: extending one or more existing objects to create new objects that
share data, behavior, and methods in terms of 00 terminology.

Generally, ODBMSs are the database systems that allow data to be stored beyond
the tabular format of the relational model. They can deal with complex data stnic-
tures as in prograrnming languages. Another possible way of thinking of ODBMSs is
as an object-oriented programming language wit h persistent data, in the sense t hat
data in the progams lives beyond the life of the programs. The ability to manipulate
data and perforrn computations within one single system is the strong point that
has been claimed to solve the problem of the misrnatch between data manipulation
laquages (e.g. SQL) in the relational model and ordinary prograrnming languages.

1.3 Object Relational Mode1

h o t her database model is the object-relat iooal database management system, which
was proposed by Stonebraker et. al. (Stone961.
It has four major features:

Support for base data type extension. These include dynamic linking of user-
defmed funct ions, client /semer activation of user-defmed funct ions, secure user-
CHAPTER 1. INTROD UCTION

defined functions, callback in user-defined functions, user-defined access meth-

ods, and =bit r q - l e n g t h data types.

0 Support for complex objects. Three basic type constructors are available: com-
posites, sets and references. Full featured user-defined functions c m be imposed
on complex objects. Cornplex data types can be of arbitrary-length and have
SQL support.

0 Support for inheritance. Both data and function inheritance are supported.
Overloading is also available, as well as multiple inheritance.

0 Support for a production rule system. Events and actions are retrieved as well
a s updates. Rules are integrated with inheritance and type extension. There
are rich execution semantics for rules and no infinite loops.

Stonebraker predicted "object-relational DBMS to be the next great wave in

dat abase technology" [Stonegô].

1.4 Nested Relation Mode1

Most work on the relational model of Codd [Cod701 involved the first normal f o m
(1NF) assumption, Le., that al1 elements of a tuple of a relation axe atomic values
(undecomposable). This has the advantage of simplifying the data model. However,
lrom the programming laquage point of view, this is an arbitrary restriction. Ways
of relaxing 1NF have been investigated which retain much of the advantages of the
relational model. The need to introduce complex objects into relations to make them
more qualified to handle non-business data processing applications such as picture
and map proceasing, computer aided design and scientific applications was realized
in the late 197O3s,thus leading to the introduction of nested relations [Mak77] and
the non-first-normal-form ( NF2)[Jae82].
CHAPTER 1. INTRODUCTION

I Project
Manager

Fr- Detail
PName Budget(K)

1 Sue
---

1 ~2 / 30

Figure 1.1: Nesting

1.4.1 Nested Relations

The relation Project in Figure 1.1 gives an example of nesting. Relation Project
consists of 2 tuples each having two attributes:

Manager: The name of the manager who is in charge. The data is of type string
(atomic).

Detail: A nested relation containing the projects of which the manager is in

charge. Each tuple in relation Detail contains a whole relation as an attribute
value. The first tuple contains a relation with 2 tuples. The second tuple
contains a relation with 3 tuples.

In [Sch82][Pis861[Lev92], the authors claim that N F 2 relations have some advan-

tages over 1NF relations, such as:

Nested relations minimize redundancy of data. Related information can be

stored in one relation only without redundancy. For example, if relation Project
in Figure 1.1 were to be represented by INF, it would be either have had to
CHAPTER 1 . INTRODLICTION 7

have redundant values for attribute Manager, or it would have had to be split
into two different relations (Project and Detail), with a foreign key, PName.

a Nested relations allow efficient query pmcessing since some of the joins are
realized within the nested relations themselves. In our example in figure 1.1,
if information about the manager's budget needs to b e retrieved in the 1NF
representation a join must be perfomed between Manager and Detail, while no
join is needed in the NF2 representation.

a Low level implementation techniques such as clustering and repeating fields c m

be represented using the formalism defined by the nested relation mode1 [Kor89].

1A.2 Nesting and Unnesting

In the literature, defining a nested relational mode1 was done by extending relational
operators t o nested relations, and adding t wo restructuring operators, NEST and
UNNEST [Jae82][Fissa]. The NEST operator creates partitions which are based
on the formation of equivalence classes [Kor89]. Tuples are equivalent if the vdues
of t h e same attributes which are not nested are the sarne in the different tuples.
Al1 equivalent tuples are replaced with a single tuple in the resulting relation; the
attributes of this tuple consists of al1 the attributes that are not nested, having the
cornmon value in the original tuples, as well as a nested relation whose tuples are the
values of the attribute to be nested. Figure 1.2 shows an example of the use of the
NEST operator. Relation Project is nested on attnbute Member.
The UNNEST operator undoes the result of the NEST operator. It creates a new
relation whose tuples are the concatenation of all the tuples in t h e relation being
umested to t h e ot her attributes in the relation [Kor89]. Thus:
UNNESThk(NESTMmb,(Project)) = Project [Jae82]
But, the reverse does not hold, i.e.:
CHAPTER 1. INTRODUCTION

Project N E S T M - ~(Project)
~ 1
Pl Joe 1
Pl Sue Sue
Pl Sam
P2 Joe
P2 Mary

Fiy r e 1.2: Nesting on Member

Figure 1.3: NESTB(CrNNESTB(R))O R

N NESTAtttibutC(Relation))
"NESTAttrsUte(U = Relationn is not always true.
The case in Figure 1.3 gives an example.
As the price of the advantages over 1NF relations, nested relations pose a non-
trivial problem of data representation [Tak89].There are generdy alternative rep-
resentations of data in a nested relation, while the data is uniquely represented by a
1NF relation. This is illustrated by the foliowing example:
In left side of Figure 1.2, we have a simple 1NF relation Project on ProjName
and Member. This relation is a unique representation of a set of 7 tuples.
CHAPTER 1. INTRODUCTION

1 ProjName 1 Member 1

Figure 1.4: Relation: N ESTp,,_N,,e( Project )

We can nest Project on att ribute Member as shown in the right side of Figure 1.2.
We can also nest Project on attrîbute Proj-Name, as illustrated in Figure 1.4.
Thus, it might be controversial whether or not these two relations are regarded
as the same relation. There are two different assumptions with respect to the inter-
pretation [Tak89]
:

1. To consider each tuple in the relation to be meaningful. Hence, the relation

in the right side of Figure@ 1.2 gives a list of projects and their members,
while the relation in Figure@ 1.4 gives the list of members and the projects
they participate. They carry different meanings, therefore, each nested relation
should be recognized as distinct. Thus, it would be difficult to identify a nested
relation with a 1N F relation. It 'poses a semantic gap between 1NF and nested
form relations although it enables us to represent complex objects in a natural
way by using nested relations" [Tak89].

'2. Conversely, to assume that each tuple is just a union of single values rather
than a specific object, which d o w s the identification of the two nested relations
 in the right side of Figures 1.2 and 1.4 and the identification of them with the
original INF relation. Many research papers implicitly use this assumption such
as t hose proposing t ransfonnation operators [Jae82][Fis85], and t hose designing
nested relations [Ozy87][0zy89].

Significant progress has been made in the field of nested relations during the past
decade. A generalizat ion of the ordinary relat ional model, allowing relations wit h
set-valued attributes and adding two restructuring operators, nest and unnest, was
int roduced [Jae82][00M87].Fisher and Van Gucht (Fis851 discussed one-level nested
relations and their characterization by a new family of dependencies, and furthermore,
t hey developed a polynomial-time algorithm to test if a structure is a one-level nested
relation. Thomas and Fischer generalized their work on the one-level model and d-
lowed nested relations of arbitrary, but fixed depth [Tho86]. In [RKS86], Roth, Korth
and Silberschatz defined a normal form called "Partitioned Normal Form(PNF)" for
nested relation, and also defined algebra and calculus query languages for thern; how-
ever, their proofs and method were later questioned by Tansel and Gamett [Tag92].
Numerous query languages have been introduced for the nested rnodel [RKS86], and
mt.rnsions have been proposed to practical query languages such as SQL to accom-
modate nest ing [Pis861[Kor89]. Implementation of databases based on the nested
relation rnodel are dso amiilable such as of in [Sps87][Des88][Sab89]. These are either
built on top of existing relational databases, or from scratch.

1.4.3 Our Approach

We view nested relations in a different light. We do not restrict our approach to
nesting and umesting. We build nested relations to facilitate nested queries. We do
this by extending domain operations to include relational operations.
In our approach, we observe that:
CHAPTER 1. INTRODUCTION 11

Using flat relations, we can model nested relations. We can use a set of sumo-
gates to keep links between parent relations and their nested child relations.

0 We can build a nested relation query facility in the context of flat relations.
Since an attribute itself can be a relation, relational operations can be included
in domain operations.

1.5 Thesis Aim and Outline

The purpose of this thesis is to extend Relix with nested relations and to integrate
t h e relational algebra into the domain algebra.

0 Chapter 1 contains a literature review of t h e relational model, the object ori-

ented model, object-relationai mode1 and nested relations.

0 Chapter '1 provides a generd overview of the Relix database programming

language-the relational database programming language developed at McGill
University. The syntax and intemal operation of Relix that are relevant to the
work done in this thesis are discussed in this chapter.

a Chapter 3 is the user's manual on nested relations. It shows the semantics and
syntax for nested relation definitions and operations.

O Chapter 4 gives a detailed description of the implementation of nested relations

in Relix.

Chapter 5 concludes the thesis with a summary and proposds for future work.
Chapter 2

Relix

Relix is briefly described in t his Chapter. The purpose of this Chapter is to provide
readers with enough background to understand the rest of the thesis. Since al1 the
design and implementation work in this thesis follows the conceptual framework of
the existing Relix system, we will present only the subset of Relix related to this
thesis. The theoretical foundation on which the development of Relix is based can
be found in [Mer84], while the basic reference of Relix can be found in [Ld86].

2.1 Overview
Relix is a REIational database programming Laquage in U N M . It is an interpreted
language written in C. It can accept and execute commands or statements from the
command line. It cm also accept Relix commands and statements batch files.
Relix deals primaxily with two kinds of data models: domains and relations. There
are two categories of operations: domain algebra and relational algebra.
CHAPTER 2. RELIX

2.1.1 Domains and Relations

A relation is defined on one or more attributes, and the data for a given attribute is
€rom a particular domain of values. The domain of a given attribute determines its
data type.
For example the Student relation in Figure 2.1 is defined on four att ributes: Stu-id,
Enter-year, Naame, Canadian. The domains of Stu-id and Enter-year attributes are
integer. The domain of Name attribute is string. And the domain of Canadian
attribute is boolean.

9546900 1995 Joe true

9602324 1996 Sue true
9701087 1937 Jin false
9702340 1997 Jin false

Figure 2.1: Student relation

There are six atomic data types in Relix as shown in Figure 2.2. Note that we

also have a special data type, relation, which wiil be introduced in Chapter 3.
In Relix, we can declare the domains of relation Student as follows:
> dornain Stu-id integer ;
> domain Enter-year integer ;
> domain Name string ;
> domain Canadian boolean ;

The relation Student can then be declared and initialized:

 Data m e Short Form Domain
integer int singed integer
long long signed long integer
short short sighed short integer
real real sighed floating point
string s trg sequence of characters
(with limitations )
boolean bool true or false

Figure 2.2: Atomic Data Type in Relix

> relation Student(Stu-id, Enter-year, Name, Canadian) <-

{(9546900, 1995, "Joe ", true),
(960284, 1996, "Sue ".true),
(9701087, 1997, "Jin ", fdse),
(9702340, 1997, "Jin ",falsej} ;

We can also declare a relation without initialkation, i.e., a relation without any

data :
> relation Student (Sttu-id, Enter-year, Name, Canadian)

2.1.2 Basic Commands in Relix

In Relix, there are basic commands to show, pnnt and delete domains and relations
declazed in the database.
The grammar for the commands is:
 <commandnarne> ( ! or !!<parameters>).
Where <commandmame> includes reserved words which will be introduced in
the following paragraphs and ! means that the programmer is prompted for the
parameters, while !! requires command line parameters.

Show C o m m a n d s

0 sd! or sd!!<domainaame>

Relix will show the name, type and other information associated with al1 do-
mains in the database or the specified domain. For example:

> sd!! Stu-id

will show the information of domain Stu-id.

a sr! or sr!!<relationname>

Relix will show the narne, degree and other information of al1 relations in the
database or the specified relation. For instance:

> sr!! Student

will show the information of relation Student.

a srd! or srd!! <relationaarne>

Relix will show all relations and their domains in the database or the specified
relation and its domains. For example:

> srd!! Student

will show relation Student and its domains.

a pr! !<relationname>

Relix will p i n t dl data in the specified relation. For instance:

> pr!! Student

wili p r h t a l l data in relation Student.
CHAPTER 2. RELIX

dd!! <domainname>

Relix will delete the specified domain. If it is still in use, Relix will give an
error message and the domain will not be deleted.

> dd!!Year
will delete domain Year, if it is not in use.

Relix will delete the specified relation.

> dd!! Student

will delete relation Student.

q!

This command can be used to quit the Relix system.

2.2 Relat ional Algebra

The relational algebra consists of a set of operations on relations. Both operands and
resiilts are relations.
In Relational Algebra operations, we have unary operations and binary operations.
As the narnes indicate, unary operators take one relation as an operand, and binaq
operators take two relations as operands. In unary operations, there are projection
and selection; in binary operations, there are joins.

2.2.1 Projection
Projection is as operation on the attributes of a given relation. The results of a
projection is a relation whose attribut- are the spetified attributes in the projection
list. Duplicate tuples in the resulting relation are removed. For example, we can
project the Name of Student relation as follows:
CHAPTER 2. RELIX

> Stu-name < - [ Nome ] in Student ;

S tu-name
- - d e -

Jin
Joe

Sue

Selection is an operation on a relation to select t uples that meet the condit ion specified
in the selection clause, which is called T-selector(tup1e selector). We can do the
following selection to extract t h e student information about who is a Canadian.
> Ca-stu < - where Canadian = tme in Student ;
or
> Castu < - where Canadian in Student ;

9546900 1995 Joe true

9602324 1996 Sue true

We can combine projection and selection in a single statement. First Relix will
do selection on the input relation based on the selection clause, then do projection
on the output of the selection. We can extract the Stuid numbers of students who
are Canadian using the following statement:
CHAPTER 2. RELIX

> Ca-stu-id < - [ Stu-id ] where Canadian in Student ;

-Ca-stu-id
-----
Stu-id
------
9546900
9602324

2.2.3 Joins
There are two classes of join operations in Relix: p-joins, the family of set-valued set
operat ions; and o-joins, the family of logical-vdued set operations [Mer84].

p-joins are derived from the set operators such as intersection, union, difference, etc.
The p-joins on two relations, R(X,Y) and S(Y,Z), are based on three parts:

A
0 crnter = {(x,y, 2) 1 (2,y ) E R and (y,=) E S}

A
0 left wing = {(x,y, DC)1 (x, y) E R and V ~ ( yz, ) S}
A
a right wing = { ( D C , y , r )1 ( y , z ) S~a n d V x ( x , y ) $ R )

We will explain these three basic p-joins in detail in this section. The two relations
in Figure 2.3 are used to illustrate the operations:

0 The most used p-join is the natural join (ijoin or natjoin), which gives us the
center part of the operand relations. It combines tuples of the two relations
that have equal values on the join attributes. Thus, it is the intersection of the
two relations on the join attributes, which gives us ijoin.
CHAPTER 2. RE:LIX

Courses
-----------
S tu-id c-name
----------- - - c - - - - - - c -

9546900 Joe 9576701 Math

9602324 Sue 9546900 Physics
9701087 Jin 9602324 Histow
9602324 Math
9702340
----------- Jin
-----------

Figure 2.3: Student and Courses relations

The natural join of relations R and S is defined as [Cod70]:

A
R natjoin S = { ( a ,b: c) 1 R(a, 6 ) and S(b,c)}
where (a,b,c) is a tuple in the new relation, of which (a,b) is a tuple of R and
(b,c) is a tuple of S.

The following Relix st atement performs a natjoin between relation Student and
relation Courses.

> SijoinC < - Student ijoin Courses ;

S tu-id Name C-Narne

- - - - - - a - - - - - - - - - -

9546900 Joe Physics

9602324 Sue History
-----------------
9602324 Sue Math

a The union join (ujoin) is an operation that is a union of the set of tuples from
the natural join, together with the tuples from the relations of both sides that
axe not equal to each other in the join attributes, and the missing attnbutes
CHAPTER 2. RELIX 20

axe filled up with DC1 null value. It gives us the union of the lefl, center, right
parts of the operand relations.

> SujoinC < - Student ujoin Courses;

-----------------
9546900 Joe Physics
9576701 DC Math
9602324 Sue History
9602324 Sue Math
9701087 Jin DC
9702340 Jin DC

* The symmetric difference join (sjoin)is the set of tuples from the relations of
both sides that are not equd to each other in the join attributes, the rnissing
attributes are filled up with DC null value. It gives us the union of the lefi,
rignt parts of the operand relations.

> S.sjoinC < - Student sjoin Courses;

- - œ - - - - - - - - œ - - - œ

9576701 DC Math
9701087 Jin DC
9702340 Jin DC

The overall p-join operations axe shown in Figure 2.4.

- -

DC,Don't Care, describes irrelevant values.

CHAPTER 2. RE:LIX

p-ioins p-ioin-o~erator Resultina Relation

Natural Join 'natjoin' or 'ijoin' centre

Union Join 'ujoin' left U centre U right

Left Join 'Ijoin' left U centre

Right Join 'rjoin' right U centre

Left Difference Join 'djoin' or 'dijoin' left

Right Difference Join 'drjoin' right

Symmetric Difference Join 'sjoin' left U right

Figure 2.4: p-join operations

CHAPTER 2. RELU(

o-joins

The family of O-joins are based on set cornparis oper ors. In opecat ions, the tuples
in each of the operand reiat ions are grouped such that for each group, al1 the non-join
attributes on both sides axe identical. The set comparison operator is then applied
to the Cartesian product of the groups. The values of the non-join attributes of the
comparing groups are accepted if the specified set comparison on t h e join attributes
is satisfied.
There are five a-joins:

a sup or div or gejoin, the superset operator, a generalization of 2. 'div' stands

for 'division', which extends Codd's definition of relational diviaion [Cod72].

a sub or lejoin, subset, a generalization of C.

a eqjoin, equai set, a generalization of =.

a sep, intersection empty? a generalization of R.

0 icomp, intersection not empty, a generalization of @.

Considering the two relations Student and Ch.- in Figure 2.5.

Student Class
-----------
Name Course
----------- - - - - - - - - -Room
Course
--
Joe Ma th Math 286
Joe Physics Physics 286
Sue Physics Chemistry 302
Physics 3 12
-----------
Jin Math - - - c - - - - - - -

Figure 2.5: Student and Class relations

To answer Eoliowing query: Find students and the classrooms such that the courses
the student has taken is a subset of the courses which are given in this classroom.
CHAPTER 2. RELIX

> StuRoom < - Student sub Class;

--------
Joe 286
Jin 286
Sue 286
--------

The overall o-join operations are shown in Figure 2.6.

2.3 Domain Algebra

Relat ional algebra considers relations to be data primitives [Mer841 and therefore
does not give the user the power to manipulate attributes. To overcome this problem,
Merrett proposed domain algebra [Mer77].
Besides creating a domain by declaring its type as i n section 2.1.1, one can build
a new domain by expressing the domain as operation o n existing domains. It allows
operations over a single tuple (horizontal operations) and operations over sets of
tuples (vertical operations). Domains defined in this way are 'virtud' in the sense
that they are expressions and no actual values are associôted with them. The values
of the virtual domains are actualized in a Relix statement, notably, projection or
selection.

2.3.1 Horizontal Operations

Horizontal operations work on a single tuple of relation. We can define constants,
perform renaming and arithmetic functions, as weîl as if-then-else expressions.
CHAPTER 2. RELIX

Set Com~arison a-ioin O~erator

Superset 'div' or 'sup' or 'gejoin'

Equal Set

Subset 'sub' or 'lejoin'

Intersection Empty 'sep'

Proper Superset 'gtjoin'

Proper Subset 'Itjoin'

Not Supenet

Not Equal Set

Not Subset

Intersection Not Empty 'icomp'

Not Proper Supenet

Not Proper Subset

Figure 2.6: o-join operations

 0 constants
Iet t w o be 2;
let myname be "marc";

renaming
let stuaame be name;

a arit hmet ic functions

let Sin b e sin(ang1e) ;
let area be sqrt(a**2 + b**2 + c**2) / 2;

if-t hen-else
let Grade b e if Mark > 60 then "Pass" else "Fail";

Al1 above domains defined are virtual domains. For example, we can actualize

Crade as following:
> CRADES < - [ Student, Crade ] in MARKS

MARKS GRADES
------- --------
Name Mark Nante Grade
------- -Joe- - - -Fail
---
Joe 50
Jin 80 Jin Pass
Sue 90 Sue Pass
-------

2.3.2 Reduction (Vertical Operat ions)

Reduction are domain algebra operations which combine values from more thkn one
t uple - the 'vertical' operation [Mer84].

Simple Reduction
 Simple reduction produces a single result from the values from al1 tuples of a
single attribute in the relation (Mer84j. The operator in simple reduction must
be both commutative and associative, such as plus (+), multiplication (*). For
exarnple:

l e t Total be red + of Grade;

Transcript
-----------
Name Dept Grade (Total)
-----------
Joe CS 85
Jin CS 90
Sue EE 80
Weny ME 75
-----------

Equivalence Reduct ion

Equivalence reduct ion is like simple reduction but produces a different result
[rom different sets of tuples in the relation. Each set is characterized by al1
tuples having the same value for some specified attributes - an 'equivalence
class" in mathematical terminology [Mer84]:

let Subtotal be equiv + of Grade by Dept;

Transcript
-----------
Name Dept Grade ( Subtotal)
i - - - - - - - - - -

Joe CS 85 175
Jin CS 90 175
Sue EE 80 80
Weny ME: 75 75

2.3.3 Nested Relations

In this thesis, we extend Relk to support nested relations. In chapter 3 and chapter 4,
we will discuss nested relations in detail, including a user manual and implementation
CHAPTER 2. RELU(

techniques.

2.4 ijoin, ujozn, sjoin are Associative and Cornmu-

t at ive
From Section 2.3.2, we know that in simple and equivdence reduction, the operator
needs to satisfy the commutative and associative criteria. In the following sections,
we prove that ujoin, zjoin, and sjoin al1 have these two characteristics .

2.4.1 Definition
For relations, R(X, Y) and S(Y,Z), these three sets of tuples are each defined on the
attributes(or attribute groups) X, Y, 2.
We first define three disjoint sets of tuples which are set operations between R
and S [Mers$]:

a
1. center = { ( x . y , z ) 1 ( x , y ) E Rand ( y , ~ E
) S}

a
'2. 1eft uiny = {(r,y , DC) 1 ( r ,y ) F R and V s(y: tj $ S}

A
3. right wzng = y , z ) 1 (y,z) E S and Vx(x, y)
{(DC, R}

The joins' definitions are based on these 3 sets:

a
1. R ijoin S = center

A
2. R ujoin S = Ieft uring U center U right ving
A
3. R sjoin S = left wing U right wing
2 A.2 Commutative
By definition, an binary operator 8 is commutative iff A 8 B = B O A.
Remark 1: R ijoin S = S ijoin R.
Proof:
R ijoin S = {(x,y, z) 1 (x, y ) E R and (y, z ) E S} (from definition)
*
R ijoin S = { ( r , y , 2) 1 (z, y ) E S and (9,s) E R} (from the commutativity of
and)
*
R ijoin S = S ijoin R
Remark 2: R sjoin S = S sjoin R.
Proof:
R sjoin S = {(2,y, DC)1 (x,y ) E R and tf z(y,:) y, t) 1 (y,z) E
$ S } u {(DC,
5' and V x ( z , y ) $Ri} (from definition)
*

-
R sjoin S = {(z, y, DC) 1 (-,y) y, x ) 1
E S and V ~ ( yx,) 6 S} U {(DC, (y, x ) E
R and V z ( z , y) S ) (from symmetry and the commutativity of U)

R sjoin S = S sjoin R
Remark 3: R ujoin S = S ujoin R.
Since R ujoin S = (R ijoin S) U (R sjoin S) (from the definition)
And from Remark 1 and Remark 2, the proof is trivial.

2.4.3 Associative
By definition , an b i n q operator 19 is associative iff ( A 8 B)0 C = A O ( B 0 C)
Suppose we have 3 relations, R(X,Y), S(Y,Z), T(Z,W)
Remark 4: (R ijoin S) ijoin T = R ijoin (S ijoin T)
 Proof:

-
( R ijoin S) ijoin T = ((2,y, z ) ( ( x ,y ) E R and ( y , z ) E S} ijoin T (from the
definition)

( R ijoin S) ijoin T = {(x, y, r , w ) 1 (x,y ) E R and (y, z) E S and (2,w ) E T }

( from the definition)
*
( R ijoin S) ijoin T = { ( x , y, z , w ) 1 (x,y ) E R and ( ( y , z ) E S and (2,w ) E T ) }
( from the associativity of and)
*
( R ijoin S) ijoin T = R ijoin {(y,z, w ) 1 (y, t) E S and ( 2 , ~ ) E T } (from
definit ion)
*
R ijoin S ) ijoin T = R ijoin (S ijoin T) (from definition)
Remark 5: ( R sjoin S) sjoin T = R sjoin (S sjoin T)
Proof:

-
( R sjoin S ) sjoin T = ( l e f t ~ i n g ( ~ U, sr)i g h t ~ i n g ( ~sjoin

( R sjoin S) sjoin T = ({(z, y, DC)1 (2,y ) E R and Vz(y,z)

, ~ ) )T (from definition)

6 S ) U {(DC,y, r ) 1
(y, z ) E S and V x(x, y ) 4 R})sjoin T (from definition)
*
( R sjoin S) sjoin T = { ( x ,y, DC,DC)1 ( 2 ,y) E R and Vz(y, z) @ S and Vw(DC,w )
T ) ü {(DC,y,z,DC) 1 (y,z) E S and Vx(x,y) 6R and V w ( r , w ) 6 T} U
D C , r , w ) / (qw)E T and
{(DC, V y(y,t) 4 S and V x(x,DC) 6 R ) (from def-
inition)
In the same way, w e can get:
R sjoin (S sjoin T) = {(x,y, DC,DC)1 (x,y ) E R and Vz(y, z ) 6 S and Vw(DC,w )6
Tl U z, DC) I (Y,z)
{(WY, E S m d 'd z(z,Y) 4 R zmd v ~ ( 2 1 6 T} U
~ )
CHAPTER 2. RELIX

DG',2,w ) 1 (2,W ) E T and V y(y, 2 ) S and V ~ ( xDC)

{(DG', , $ R)
Thus
( R sjoin S) sjoin T = R sjoin (S sjoin T)
Remark 6: ( R ujoin S) ujoin T = R ujoin ( S ujoin T)
Proof: From Rernark 4 and Remark 5, the proof of Remark 6 is trivial.

2.4.4 Another Approach

Let x be a tuple, and let X be a binary va.riable such that if x E some relation R,
then X has value 1, otherwise O.

L. for R = RI ujoin R a . . .R, and for some tuple x, if XI+ X2 + ... + Xn = 1,

x E

2. for R = Ri ijoin R2...& and for some tuple x, if XI * -Y2* . . . * .Y, = 1,

e x E R.3

3. for R = Ri sjoin R2.. .R, and for some tuple x, if Xi $ ,Y2 $ .. . $ *Yn$ =
0,-x E R."
From characteristics of 8, we can conclude that if x appears odd times in
relations Ri. .. &, t hen x f R.

2Here + means logicai operation OR, which is commutative and associative

3Here * means logicai operation AND, which is commutative and associative
4Here @ means logical operation XOR, which is commutative and associative
Chapter 3

User's Manual on Nested

Relations

This chapter describes how to define and manipulate nested relations in Relix. Sec-
tion 3.1 explains the basic concept of nested relations in Relix and presents the
ini tializat ion of nested relations. Section 3.2 illustrates the operations that can b e
imposed on nested relations.

3.1 The Nested Relations and Relation Data Type

To introduce nested relations, we add a relation data type to Relix. The opera-
tions imposed on it are those relational operations on regular relations with some
limitations.
We will show an example tir& then we will explain how to declare and initialize
nested relations, and f i n d y we explain the intemal data representations.
CHAPTER 3. USER'S MANUAL ON NESTED RELATIONS

The above Relix commands are used to initidize the sample nested relation in
Figure 3.1.

TEST

Figure 3.1: Sample nested relation: schema tree and value table

W e have three regular domains A, B and C, which are defined as integers, and a
nested domain S,which is defined upon A and B. When w e declare TEST,it includes
the nested dornain S. Relix will consider S as a domain as well as a relation.
The data in S is stored in another relation outside the parent relation TEST,
which has the same name as S. References to the data (cded RELATION .id) are
stored in attribute S of relation TEST. However, this method of implementation is
lilrgely transparent to users, who manipulate the attributes of nested domains as if
CHAPTER 3. USER'S MANUAL ON NESTED RELATIONS

the data were stored directly in the parent relations.

relation: "TEST"has "2"tuple(s)

Figure 3.2: What is shown in Relix

Any Relix operation that displays an attribute of type RELATION will display
the attribute as a number. The actual data of the attribute is printed below it as
a separate relation whose .id field links it to its parent. In above print command,
TEST and its nested domain S are printed out. In child relation S, .id is mapped to
attribute S of TEST.
CHAPTER 3. USER'S MANUAL ON NESTED RELATIONS

The formal syntax of declaration and iaitialization is as follows:

<declaration> := 'domain' <domainname> '(' <at tri b u t e l s t > ')'
<initialization> := 'relation1 <relationname> '(' ~ a t t r i b u t e l i s>t 7'

Note in the following sections, we will use the conceptual format as shown in
Figure 3.1 to show the example, while in Relix, t h e actual format will be as in pr!,
Le. as shown in Figure 3.2.
So far, we have only implemented two levels of nesting. Future work is needed to
gain multiple level nesting.

3.2 Operations on Nested Relations

In this section, we will show by example how to conduct operations on nested rela-
tions. We will show vertical operations, followed by horizontal operations.
The schema of nested relation is represented by the schema tree [Ozy87], as shown
in Figure 3.3. The nested relation schema of the Faculty of Engineering database is:
Dept, Building, Professor and Secretary, in which Dept and Building are regular
simple dornains, and Pm/essor and Secretury are neisted domains, which are lurther
defined by Name, Salary and Comrnif.
The nested relation, FactEng, over the schema tree of Figure 3.3, is shown in
Figure 3.4.

3.2.1 Vertical Operations

This section is for the purpose of extending reductions (vertical operations) £rom
scalar attributes to nested relation attributes.
CHAPTER 3. USER'S MANUAL ON NESTED RELATIONS

Faculty of Engineering (FactEng)

Narne

Figure 3.3: The schema tree of the sample

Dept Building Professor Secretary

-
Name 1 Salarv 1 Commit
Pat 65 PADS Sal 35 PODS
Paul 55 PODS Sue 38 PODS
Pully 50 SIGM
Pat 65 PADS Sandy 36 IEEE
Paul 55 PODS Sharon 35 PODS
Piree 54 IEE Sam 40 PODS

Pat 65 PADS Sandra 35 MEE

Ping 57 MEE SY~ 37 MDS

Figure 3.4: The nested relation, Engineering Department, over the schema in Fig.3.3
CHAPTER 3. MER'S MANUAL ON NESTED RELATIONS

Simple Reduction

Recall that we already proved that ijoin, ujoin and sjoin are al1 commutative and
associative (see Section U )we
, cm now extend the reduction operations to ijoin,
ujoin, and sjoin.
We start with the following example: Suppose we want to find al1 the professors
in the faculty of engineering, we can do the following query:
> let EngPmfbe red ujoin ofProfessor
> AllEngProf <-[EngPiof]inFactEng
> pr!! AllEngProf

EngProf
1 1
t

Name Salary Commit

Pat 65 PADS
Paul 55 PODS
Piree 54 IEE
Ping 57 MEE
Pully 50 SIGM

Figure 3.5: Al1 Professors of Faculty of Engineering

The formal syntax of simple reduction is as follows:

<simple~eductionstatement> := 'let' cnewnested-domainname> 'be red'

~nested-domainname>
<binary-operator> := 'ijoin' 1 'ujoin' 1 'sjoin'
Now we introduce the universal professor, who works in every unit of an education
organization.
CHAPTER 3. USER'S MANUAL ON NESTED RELATIONS

Query: Find al1 the universal engineering professors.

> let UnivEngProj be red ijoin of Professor

> UEP < - [ UnivEngPro/] in FactEng
> pr!! WEP

I UEP I

1 Pat 65 PADS 1
Figure 3.6: Al1 uni versal engineering professors

If we do sjoin on the attribute ProIessor, we obtain professors who are assigned

an odd number of positions (see Section 2.4.4 for explanation). Thus we have the
following query:
Find al1 the engineering professors who are assigned an odd number of positions.
> let OddProj be red sjoin of Professor
> OPrd <-[OddProflin ED
> pr!!OProf

Pat 65 PADS
Ping 57 MEE
Piree 54 IEE
Pully 50 SIGM

Figure 3.7: Professors with an odd number of positions

CHAPTER 3. U S E R S MANUAL ON NESTED RELATIONS

Equivalence Reduction

Like simple reduction, equivôlence reduction is extended to ujoin, ijoin and sjoin as
well.
Query: Find the professors by each building.

> let ProfigBuild be equiv ujoin of Professor by Building

> PbB < -[Building, ProfiyBuild] in FactEng
> pr!!PbB

PbB

Nume 1 Salary 1 Commit

Pat 65 PADS
Paul 55 PODS
Piree 54 IEE
hlly 50 SIGM

Pat 65 PADS
Ping 57 MEE

Figure 3.8: Professors in each building

Query: Find the universal professors by building. (we introduced the idea of a
universal professor in the last section. Here a universal professor in each building
works in each department of the building)

> let IlnivbuilProf be equiv ijoin of Prolessor by Building

> UBP < - [ Building, UnivBuilProf ] in FactEng
> pr!! UBP
CHAPTER 3. USER'S MANUAL ON NESTED RELATIONS

I UBP I
Building UnivBuiidProf
Nume 1 S a h y 1 Commit
MC Pat 65 PADS
Paul 55 PODS

I "" I Fb, 65
57 MEE
PADS 1
Figure 3.9: Universal Professors in each Building

Query: Find the professors in each building who are assigned odd department
positions in that building.

> let OddBuzlPmf be equiv sjoin of Professor by Building

> OBP < - [ Building, PureBuilProf ] in FactEng
> pr!! OBP

1 OBP 1
1 Building 1 PureBuilProf I
MC Piree 54 IEE
Pully 50 SIGM

Figure 3.10: Professors who are assigned odd positions in the building

Syntax:
CHAPTER 3. USER'S MANUAL ON NESTED RELATIONS 40

<equiv~eductionstatement> := 'let' ~newnested-domainname>'be' 'equiv'

<binary aperator> 'of' <nested-domainname>
'by' <attribute-list>
:= 'ijoin' 1 'ujoin' 1 'sjoin'

3.2.2 Horizontal Operations

Horizontal operations consists of binary operations and general operations.

Binary Operations

Binary relational operations take two relations as operands and produce a relation as
a result. We extend those operations to nested domains, and take two nested domains
as operands and produce a nested domain as a result, which itself is a relation data

t~pe.
Query: Find al1 the staff of the faculty of engineering.
> let Staff be Professor ujoin Secreiary
> FaetEdtafl < - [ D e p t , Building, Stafl] in FactEng
> pr!! IiactEngSLafl

The result is in Figure 3.1 1.

The formal syntax is as follows:
~binarystaternent> := 'let' <newnested-domainlime> 'be'
<nesteddomainname> <binary -operator>

<binaryaperator> := 'ijoin' 1 'ujoin' 1 'sjoin'

CHAPTER 3. USER'S MANUAL ON NESTED RELATIONS

Dept Building Staff

Name 1 Salan, 1 Commit
Pat PADS
Paul PODS
Pully SIGM
Sa1 PODS
Sue PODS

Pat PADS
Piree iEE
Sandy EEE
Sharon PODS
Sam PODS
Pat PADS
Ping MEE
Sandra MEE
SY~ MDS

Figure 3.11: Staff of the Faculty of Engineering

CHAPTER 3. USER'S MANUAL ON NESTED RELATIONS

General Operation

We can also embed general relational expressions into dornain algebra. This is c d e d
general operation. "Ceneral" here means more general t han the operation we intro-
duced before in this Chapter. However, it is not arbitraxily general. We will show
t h e limitations irnposed o n it at the end of this Chapter.
In the Faculty of Engineering, rich professors are professors whose yearly salary
equals or exceeds 55 K. We have the query: Find the rich engineering professors
together with their salary and department. The following expression will answer the
query :
> let RichProf be "< [ Name, Salary ] where Salary>=55 in Pro/essor Y;
> RP < - [ Dept. RichProf 1 in FactEng,
> pr!! RP;

The result is shown in Figure 3.12.

Dept RichProf
Name 1
Salary
CS Pat 65
Paul 55

EE Pat 65

Figure 3.12: Rich Professors of Engineering Depart ment s

We can make more complicated generd operations. For example, we can do sjoin
on different domain names in t wo nested domainrelations.
CHAPTER 3. USER'S MANUAL ON NESTED RELATIONS 43

Query: Find professors and secretaries such that the secretary works for al1 the
cornmittees to which the professor belongs.
> let Pnarne be Name
> let Srrame be Narne
> let ProfSew be
( [ Pname, Cornmil] in P~ofessor)sub ( [ Sname, Commit ] in Secretary) >"
> PSC < - [ Dept, ProJSecr ] in ED
> pr!! PSC

PSC
Rept
1

CS Paul Sa1
Paul Sue

EE Pirre Sandy

ME Ping Sandra

Figure 3.13: Professors and Secretary in Cornmittes

The formal syntax:

<domainrelationals tatement > := 'let' <aest,domainaame> 'be'
' <' <relational-expression> ' >" '
II

<relat ional-expression > is an expression of relational algebra operat ions wit h

some limits. The T-selector in the following paragraph illustrates this. Note that we
quote <relat ional~xpression>using < >"" , and during declaation, it is treated
"II

as string, yet during the actualization, the Relix statement included in the string will
be evaluated.
CHAPTER 3. USER'S MANUAL ON NESTED RELATIONS

<T-selector> := '[' <attributelist> '1' 'where7

~selection
xlause> 'in' <nested,domain>

<selection~c1ause> is a comma-separated list of simple logic domain expression

that can be evaluated horizontally to true or faIse on each tuple of the operand
~nested-domain>(which is a relation as well).
We have not been able to implement verticai domain operations within the syntax
of general operations (in <relational~xpression>).
Chapter 4

Implementation of Nest ed
Relations

This chapter deals wit h the implementation of nested relations. Section 4.1 gives an
overview of the implementation of Relix. Section 4.2 describes how nested relations
are represented and declared. Section 4.3 illust rates the implementation of nested
relation operat ions.

4.1 Implementation of Relix

Relix is an interactive rnulti-user system written in C, and is portable across different
platforms running the UNIX operating system. Extensions in Relix require that the
modules to be added are compatible with the existing code. Therefore, in this section
we overview the implementation of Relix that is related to the work of this thesis. A
complete documentation for its first implementation can be found in [Lal86].
CHAPTER 4. IMPLEMENTATION OF NESTED RELATIONS

4.1.1 System Relations

A relation is stored in a UNIX file whose name corresponds to the name of the
relation. A database, which is a collections of relations, is equivalent to a UNIX
directory. Every Relix database maintains a set of system relations which represents
the data dictionary of the database and are stored permanently as UNIX hidden
files.' Three basic system relations are used to store information about domains and
relations in the database.

1. .rel (.rebname. .sort-status, .rank, . n t ~ p l e s ) ~

The .rel system relation stores information about al1 the relations in the database.

0 .relaarne is the name of the relation

.sortstatus specifies the type of sorting for t h e relation, such as sorted,

non-sorted and partly sorted

a .rank is the number of sorted attributes in the relation

.ntuples is the number of tuples in the relation

2. .dom (.dom-name, .type)

The .dom system relation stores information about dl the domains in the
database.

.dommame is the name of the domain

a .type is the data type of the domain. There are 6 atomic data types (see
Figure 2.2)
'File names beginning with a period (.) are UNIX hidden files which are not normaiiy h t e d
under the UNIX List directory command.
?In Relix convention, the names which begin with a period (.) are system names.
CHAPTER 3. IMPLEMENTATlON OF NESTED RELATIONS

3. .rd (.rel-name, .dominame, .dom-pos, .dom-count )

The .rd system relation stores information that links the relations with the
domains on which they are defined.

r .relnarne is the name of the relation

a .domname is the narne of the domain

r .dornpos is the byte position of the domain in the relation

r .dorn,count is the nurnber of domains in the relation

In our implementation of nested relations, we use two system relations to store

the interface information for the nested relations declared in the database.

1. .nst (sup-name,
.sub-name)
The .nst system relation contains information about parent relations and their
child relations.

r .supaame is the name of the parent relation

r .subname is the name of the child relation

2. .ne&-dom (.domain-name? .domain-rej)

The .nest-dom system relation contains information about the nested domains.

a .domainname is the name of the nested domain (child relation)

r .domain~efis the number of reference times of this domain

4.1.2 Parser and Interpreter

Relix consists of two main modules: a parser and an interpreter. The parser, which
is generated by Lex [Les751 and Yacc [Joh75],performs syntax analysis and gener-
ates intermediate codes. The interpreter is written in C, it reads instructions from
CHAPTER 4. IMPLEMENTATION OF NESTED RELATIONS 48

the intemediate code and calls particular C functions to perform the operations.
Figure 4.1 summarizes the main fiow of Relix.

Load sy&m 1
.- - - -
-

I
Wait for input from the user
1
1

I Scan input into tokens 1

I I

'-.Le--LILL--I-code-----------------l
I Interoreter Module I
Y
I
lnterpret I a d e I I
I

'1 Write system relations back to disk

Figure 4.1: Relix Execution Flowchart

W e wiii show an example from an implementation point of view to exemplify how

Relix operates.
Suppose we have:
CHAPTE3R 4. IMPLEMENTATlON OF NESTED RELATIONS

The parser performs syntax analysis and fmds that the above statement fits the
following grammar rules.

domain-declaration:
DOMAIN-DEC ident ifer
( translater( DOMAIN-DEC);)
TIPE
( translater( IDENTIFIER) ; translater( TYPE) ; 3

Actions in Yacc are C codes enclosed in a pair of curly brackets. The t ransla-
tor function is a C function which performs various tasks according to the actual
parameters. The tasks of the translater function include:

a rnaintaining a scaiar stack for storing and retrieving identifiers

O maintainhg a set of flags and counters

0 generating 1-code

For instance, the cal1 'translator(1DENTIFIER)' pushes the value of the identifier
onto the scalar stack.
Some of the parameters produce 1-code. For example:
paramet er Lcode
DOMAINDEC global-dom
TYPE push-name a domain

'a' is a string obtained by popping an item fiom the scalar stack. The 1-code for
the exarnple statement is shown below:
 global-dom /*set t h e flag n o t i f y i n g that the following
declared domain i s a global domain. */
a
push-name /* Push the next s t r i n g onto the stack.*/
long
push-name
a
domain /* Pop a from t h e stack, and actually declare
a as an integer domain. */
halt /* Update system r e l a t i o n s and return. */
The comments on the right hand side describe the interpreter actions for the
corresponding 1-codes. The interpreter maintains a stack for storing and retrieving
operaads. The 'push-name' pushes an operand onto the stack. The 'domain' is a
collection of C functions that the interpreter needs to cal1 with predefined arguments,
which are obtained by popping the operands from the stack. Note that ' h a l t ' is
required at the end of the 1-code for the interpreter to stop execution.

4.1.3 Implementation of Domain Operations

Suppose we define a virtual domain D as a function of other domains (see Section
2.3). In the implementation, we have routines which will locate these domains in
relation R, calculate the corresponding values of D from these operands and append
these values of D to the appropraite tuples of the original relation.
The following example will show how domain operations work in Relix:
We declare a constant atrribute as follows:
> letabe3;
After the declaration, domain 'a' is recored in the system as:
CHAPTER 4. IMPLEMENTATION OF NESTED RELATIONS

Name Actual Visited Label Type

a FALSE TRm 1 short
ûperator: constant
Value : a+OOOOSm

Note that the 'Actz le of dornain a is /aise, which means that a is a virtual
domain, and the following Relix statement requires it to be actualized.
> ACT < - [ a ] in TEST;
The 1-code for the example statement is shown below.

push-name /* Push the next string onto the stack. */

ACT

constant - r e l a t ion /* C a l 1 funct ion const ant-relation t O

create a new relation using t h e name
on the stack */
push-name
TEST
push-name

/* Push a counter onto the stack. */

project /* Cal1 funct ion pro je c t t o

create a nev r e l a t i o n according to
the attributes required */
ass ign-scalar /* Pop item A and B from the stack, and
cal1 f a c t i o n assign,scalar to
assign item A t o item B . */
/* Update system relations and return. */
CHAPTER 4. IMPLEMENTATION OF NESTED RELATIONS 52

In above 1-code, when the iotepreter reads project, it will call a C function
*projeet()'t o perform the actual projection. In turn, porjecto will call yet another
lunction 'actionizeifany-virtual()' to actualize the virtual domains ('a' in this case).
The algorithm for routine pmject() is as follows:
project ( list-R, r-name)
where listR is a lznked list which contains the domains to be projected and
r-narne is the name O/ the relation on whzch the domains are t o be projected.

1. Check lisLR, make sure no duplicates are included.

2. Actualzre lîst-R jrom r-name to R ( a ternporaryjile). Sort R o n 1ist-R. Cal1 the

routine actualize-ij-any().

3. Do actual projection according to list-R.

4. Return the Jile name O/ the results O/ projection.

The algorithm for routine actualire-if-any-Grtual() is :

actualize-$-an y-uirlual (R-name, E-list)

where Raame is the name of the relation being processed and Elist is a list of
attributes of the relation i n R s a m e , including both the original attributes a n d virtual
attributes which are d e f i e d as a function of the original attributes.

1. Traverse the attribute list and find i f there are any virtual domains.

2. If there are no virtual domains, return the original relation.

3. 11 there exist mrtual dornains.

(a) Traverse each tupie of the orginal relation.

(6) Actualize the virtual dornain value according to the definition of the virtual
dornain.
CHAPTER 4. IMPLEMENTATION OF NESTED RELATIONS

(cl Put al1 the tuples in a tempomry relation.

(d) Return the tempomry relation.

In our example, the prograrn flow is as follows:

F%en project() is called, the valves in the two pummeters are:

(a) IistR, which points to a list which includes only one item, 'a'.

( b ) rname, zvhich is 'TEST'.

Then actualize-if-any() is called with the parameters' values as:

( a ) E l i s t , which points to a list which is the same as listR in project(), i.e.,

'a'.

(b) rname, which is the same as rname in project(), i.e., *TEST9.

In actualire-if-any(), the sytern Jnds that 'a' is a virtual attribute, and there-
afler, domain a is actualized by asszgning the value of 5 t o the attribute 'a' of
every tuple in TEST.

Artunlire-ii_ong(l returns the name of the temporary relation to project (1. which
in lurn projects the a ' domain and retvrns the resuit to systern.

Update system tables.

Declarat ion and Initialization of Nest ed Re-

lat ions
4.2.1 Declaration of Relation Data Type
We can declare a regular integer domain S and a regular relation S with domains a
and b as follows:
CHAPTER 4. IMPLEMENTATION OF NESTED RELATIONS

> relation S ( a , 6 );
We have already explained the 1-codes of domain declaration (see Section 4.1.2). The
1-codes of the relation declaration is as follows:

push-name
no-cp-ln /* S e t the flag that only declare,
no data input*/
push-name /* Push the next string onto the stack.*/

push-name

push-name
b
push-
/* number of domains */
push-name
S
relation /* Pop domain list (a and b) from t h e stack,
pop S from the stack, and declare S a s a
relation */
halt /* Update system relations and return. */
To declare a relation data type, we combine the above two cases and add the
following grammar to yacc:
<nested,domain-dechration> := 'domain' <identifier> <domainiist>
For instaace:
CHAPTER 4. IMPLEMENTATION OF NESTED RELATIONS

> domain S ( a , b );
The 1-codeare dso combined from above:

push-name
no-cp-ln
push-name

push-name
. id /* Add a system domain .i d t o refer to
the parent r e l a t i o n */
push-name
a
push-name
b
push-count
3
relation
global-dom
S
push-name
relation
push-name
S
domain
end-don-code
halt

The compazison of the above three cases is s h o w in Figure 4.2.

CHAPTER 4. IMPLEMENTATION OF NESTED RELATIONS

domain S intg; relation S (a,b); domain S (a,6);

global-dom push-name push-narne

S no-cpln no-cpln
push-narne push-name push-name
long
push-name push-name push-narne
S a .id
domain push-name push-name
end-dom-code b a
push-coun t push-name
2 b
push-name push-count
S 3
relation relation
global-dom
S
push-narne
relation
push-name
S
domain
end-dom-code

Figure 4.2: Cornparison of the nested domain declaration with the regular domain
declaration and the regukr relation declaration
CHAPTER 4. IMPLEMENTATION OF NESTED RELATIONS 57

Each nested domain has its declaration entry in both .dom system table and .ml
system table. The .type in table .dom of aay nesteddomain, i.e., relation data type, is
set to a constant 'RELATION', which equals 11 in the current version. The following
entry in .dom table is for the nested domain S
.dom (.domname, .type)
S 11
The following entry in .rel table is also for the nested domain S

S O O O
Because nested domain S is a relation itself, its information and that of its domains
are stored in another system table .rd. The following entry is for S:
.rd ( .relname, .domname, .dom-pos, .dom-count )
S .id O -3
S a I -3
S b 2 -3
Note that S has three domains, among which .id is added by t h e system in order to
refer it to the parent relation.
S also has an entry in the system table .nesLdom.
.nest ,dom ( .domainname, .domainref )
S O

Init ialization of relations can be achieved by supplying the initialization data directly
on the command line:

> relation Simple (a,b) c- ( (I,2),(3$)};

For Bat relations, the algorithm of initialkat ion is:

 CHAPTER 4. IMPLEMENTATION OF NESTED RELATIONS 58

1. Parse the relation identifier and parse the domain identijiers. In the above case,
:
'Sample ', 'a ', and 'b then create a file named Simple '.

2 Parse the constants, and Save the constants lo jile Simple '.

Recall that we declare the nested domain:

> dornainS(a, 6);

For nested relations, we can initialize as follows:

since we include a nested domain S here, we need to revise the algorithm to achieve
the desired effects.

1. Parse the relation identifier and the domain identijiers, and record the nested
:
subrelations (nested domains). Then create a jile named 'Test also create files
according to subrelations, in this case we have 5 '.

2. Purse the constants. When we rneet a curly brace '{ ', we create a surrogate
to the parent uttribute, and put the conesponding real constants into the cor-
responding subrelations. For example, for { (1 $), (8, y)}, the surrogate is O and
!OP the surrogate is 1. Thus,
{(6,5),(4,9)},

(a) In file TEST, we have (3,O), (7,l);

(b) In file S , we have (0,1,2),(0,8,7),(1,6,5),(1,4,9);

4.3 Operations
In this section, we present the implementation for operations on nested child relations
. (nested domains).
CHAPTER 4. IMPLEbIENTATION OF NESTED RELATIONS

4.3.1 Implementation of Reduct ion

We will show by example how reduction operates on nested relations in Relix. Since
we based our implementation on the existing implementation of reduction on scalar
attributes, we will first present the implementation of reduction on scalar attributes.

Reduction on scalar attributes

Scalar attributes' data types are atomic as surnmarized in Figure 2.2. Recall that
in Chapter 2, we already listed that what scalar operations can be conducted on
both simple reductions and equivlant reduction. Now we will show how they are
implemented by using an example of '+', the add operator.
Suppose we have a database order as in Figure 4.3.

---------------------
Customer Product Amount
- - - - - - - - - - - - - - - - - - c _ c

Ann
Ann
Ping
Sam

Figure 4.3: Order table

Ln order to gain the total order Arnount of al1 the customers, we can use our 'red
+' operator, and impose it on the domain Amount.
> let Total be red + of Amount ;
Domain Total is kept in the systern as:
CHAPTER 3. IMPLEMENTATTON OF NESTED RELATIONS

Name Actual Visited Label Type

Total FALSE TRUE 51 long
ûperator: red-plus
ûperand-1: Amount

Whenever a Relix statement wants to include Total, the system will cal1 Actual-
ise-$-nny() to actualize it.
As we can see, Total is defined on Amount.
The algorit hm is as follows:

1. Initialire an accurnulator accordhg to Amount (In this case, its data type is
long).

2. Scan through each tuple O/ the relation Order. Extract the value of Amount,
add it to the accumulator (Recall that operator of Total is '+y.
9. rlsszgn the vulue in Le accumulator to the Total attribvte of each tuple.

Thus we can actualize Total and the result is shown in Figure 4.4.

- - - - - - - - - - c - - - - - - - - - -

- t-orner
-Cus - - - - - -Product
- - - - - - - - Amount
---- (Total)
Ann W 10 100
Ann X 40 100
Ping M 20 100
Sam Y 30 100

Fiy r e 4.4: Values of Total &ter act ualization

Furthemore, we would like to know the total amount of the products each cus-
torner ordered. The follwing ReIix statement can help us to perform this task:
CHAPTER 4. IMPLEMENT.4TION OF NESTED RELATIONS

> let CusTotal be equiv + of Amount b y Customer ;

I t is stored in the system as:

Name Actual Visited Label Type

CusTotal FALSE TRUE 52 long

Operator: equiv-plus
Operand-1: Amount
By-list: Customer

We can see in the system data structure that CusTotal actually has an item called

by-list, which includes Customer, and that the resulting CusTotal will be based on
this list.
With following steps we can actualize Cus Total:

1. Sort original relation Order on by-List (i.e., 'Custorner 'J.

2. Initialire an accumulator storage according to CusTotal

9. Scan through tuples of Order, i j the tuple's value is kept the same in uttribute
Customer, add it to the accu.rnulator, othenuise append the value O/ the accu-
rnulator to the previous tuples, and reset the accumulator.

This way we can actualize GusTotal as shown in Figure 4.5.

Reduction on Nested Attributes

In this section, we will present the general aigorithms of reduction on nested attributes
first and then show some examples.
The operator of reductions on nested attributes f d s in one of the following groups:
CHAPTER 4. LMPLEMENTATION OF NESTED RXLATIONS

(CusTotal)

Figure 4.5: Value of CusTotal after actualization

(simplereduction equivalence-reduction)
red-ijoin equivijoin
red-ujain equiv-ujoin
redsjoin equivsjoin

General Algorithm

0 Simple Reduction

In this case, the operator belongs to the Jrst group.

2. In the parent relation level, tue assign each tuple in &heposition of the
operand domain a constant O. For simple nduction, the valve of this at-
tribute should have the same value for al1 tvples in the relation.

2. In the nested relation leoel, acco~dingto the operator, do ujoin, ijoin and
sjoin with the subrelations (which are actually stored in the same physical
table).

(a) ujoin: Project al1 the attributes except .id. The obtained result is the
required ujoin operations on those sub-relations. Then, append a new
.id to it, in oîder to keep links lmth the parental relation. The value
is a constant 0.
CHAPTER 4. IMPLEMENTATION OF NESTED RELATIONS 63

(b) ijoin: Sort the table according to the number of tuples in each sub-
relation, select the sub-relations one by one according to the value
O/ .id and do ijoin on them. in this way, we can improve the join
eficiency, since during the join procedure, the result might be empty
befon we reach the last subrelation.
(c) sjoin: The algorithm is the snme as ijoin, except we do not need to
sort the table.

Equivalence Reduction

In this case, the operator belongs to the second group.

1. Sort the original relation on by-list.

2. Determine equivalence classes, for each class, do inside reduction, zuhich

will be presented nezt.

Inside Reduction

1. Initialize an accurnulator, which is an empty temporary relation.

2. For each tuple:

Extract the value of the nested domain, i.e., the pointer to the underiying
subrelation;
Eztract tuples of subrelation according to the mapping between the parent
nested domain and .id, store thenz in a temporary file.
Perform the appropriate join (ijoin, ujoin, sjozn) with the accumulator.

Examples
In Figure 4.6, we have a relation Order-book with domains Customer and Order,
which is a subreIation with domain Product.
CHAPTER 4. IMPLEMENTATION OF NESTED RELATIONS

Order-Book Order
----------- ---------
C u s torner
- - - - - - 3 - - -
Order
- -.id- - - -Produc
- - -t-

Figure 4.6: Relation OrderBook and its subrelation Order

We have three Relix statements:

1. > let AllProduct be red ujoin of Order ;

2. > let [Product be red ijoin of Order ;
3. > let CustProduct be equiv ijoin of Order by Customer ;

The first Relix statement above Ends al1 the products ordered by the customers.
The second one finds products which are ordered in each individual order. The third
one finds al1 the products ordered in every order by each customer.
Tu actualize AllProduct , we can run the Relix statement:
> Order-Bookf < - [Customer, Alfproduct ] in OrderBook ;

System m i n g flow:

1. Operator red ujoin belongs to the Jrst group

8. In OrderBook, tue assign AUProdoct a constant O

3. In the nested relation leuel, i.e., ALlProduct, the operator is red ujoin and the
operand is Order. We project [ f ~ o d u c t from
] Order, and append a new .id to
CHAPTER 4. IMPLEMENTATION OF hrESTED RELATIONS 65

each tuple of the new obtained relation, in order to keep links with AllProduct
in OrderBook. Thus we have a new subrelation AllProduct.

4. Update system tables.

The actualized AllProduct is shown in Figure 4.7.

AllProduct

Anri
Ping
Sam
O
O
0
,---- - - - - ( :
]
;
r

1
O M
wX
- - - - - - - - - O -
O Y

AIIProduct: red ujoin of Oder

Figure 4.7: AllProduct in relation Order-book1

To actualize IProduct , we can run the Relix statement:

> OrderBook2 < - [Custorner, Producl ] in O n l e d o o k ;

System running flow:

1. Operator red ijoin belongs to the jrst group

2. In OrderBook, v e assign to IProduct a constant O

Y. In the nested relation level (Le., IProduct) the operator is red ijoin and the
operand is Order. We do ijozn between the diffeient set of Prodaet values ac-
cording to .id. They a n {(W), (X)},{ ( W ) ) ,{(M),(W))
and {(Y),(W)} respec-
tiuely. The result is { (W)}. In order to keep links with IProduct in OrderBook,
CHAPTER 4. IMPLEMENTATION OF NESTED RELATIONS 66

we append a new .id to each tuple of the new obtained relation. T h w we have
a new subrelation IProduct.

4. Update system tables.

The actualize [Product is shown in Figure 4.8.

Ann O ,' ,----'O W

Ping 0 j---- _____----
Sam O ;
- - - - - - - - - O -

IProduct: red ijoin of Order

Figure 4.8: IProduct in relation Order-book2

To actualize CustProduct, the following Relix statement can satisfy the require-
ment:
> Order-Book3 < - [Customer, CuslProduct] in OrderBook ;

System m i n g flow:

1. Operator equiv ijoin belongs to the second group

2. Sort Order-book on Customer

3. For each Custorner: detemin e equivu~enceclasses, and conduct ijoin within

each class. For ezample, for customer Ann, we fist extract {(W),(X)}, then
{ ( W ) ) . After doing ijoin between them, we get { ( W ) } ;
CHAPTER 4. IMPLEMENTATION OF NESTED RELATIONS

4. Update systern tables.

The actualized CustProduct is shown in Figure 4.9.

Order-Book2 CustProduct

-----------
Custorner CustProduct

CustProduct: equiv ujoin of Order

Figure 4.9: Cust Product in relation Order-book3

4.3.2 Horizontal Operation

Binary Operation

The operators of binary operation are: ujoin, ijoin, and sjoin.

General Algorithm

1. In the parent relation level, copy the value /rom one of the operands' to the new
domain.

2. In the subrelation level, cal1 ReZix again to obtain the new subrelation.

3. Join back the obtained subnlation to the parent relation on subrelation's .id
uttribute un'th parental relation3 attribute.

4. Update system table.

CHAPTER 4. IMPLEMENTATION OF NESTED RELATIONS

Example
In Figure 4.10, we have relation OrderBook with domains OldOrd, Customer and
NezuOrd. OldOrd and NewOld are aested domains.

W *1- ------= a-lO- Ann 0

- - - -- - -0

_
œ = = - - W
Y
z & * -
- ,==2 Ann
Ping
1
2 - - - - - -' 21
*
-0 X
Z
W
X
y/
H M
0 @
,3 Sam 3
. -3
=-
- - - - - - - - - - - - - -:--
-
M
W
-
W
- m e - - -
3' -------
-3 Y

Figure 4.10: Relation OrderBook wit h subrelations OldOrd and NewOld

Suppose we have:

> let Order be OldOrd ujoin NewOld ;

and we can actualize Order using the following statement:
> Order-Book4 < - [~~~.stlistomer,,
Order] in Order-Book ;

The procedure of actualizing Order:

Copy OldOrd to Order. This way, we can keep a set of surrogates of Order in
parent relation OrderBook.

Gall Relzx again to get Order, i e . , run "Order < - OldOrd ujoin NewOrdn in
Relix. Since both OldOrd and NewOrd have same attributes, .id and Product,
we do ujoin on them to get Order.

Join back the obtained subrelation to the pannt relation on subrelationk .id at-
tribute with the parent relation's attribute Order. GOrderBook < - OrderEook
C'HAPTER 4. IMPLEMENTATION OF NESTED RELATIONS

[Order ijoin .id] Order"

The final result is shown in Figure 4.11.

Order-Book5 Order
- - - - - - - - O - -
--------
- - - - - - - -Order
Custorner --- .id
- - - - - O
Product
- -

Order: OldOrd ujoin NewOrd

Figure 4.11: Actualized result of Order in relation OrderBook

General Operation

General Operations are stored as strings when they are declared. Suppose we have
the relation as shown in Figure 4.12 and the following query:
> let BigOrd be "< [ Product ] where Amount > 8 in Order >" ;

Domain BzgOrd is stored as:

Name Actual Visited Label Type

BigOrd FALSE TFLUE 52 relation
ûperator: t-dom
ûperand: [Product] where Amount > 8 in Order
CHAPTER 4. 1MPLEMENTATION OF NESTED RELATIONS

Order-Book Order

Arin
Anil 1 - - - -- -- - 0
O *=----O
- W
X
9
6
Ping 2 - - - - - -1 Z 10
Sam
- - - - - - O - - . - -
3 =- \
-
-
'2
-

=3
-3
M
Y
W
12
10
7

Figure 4.12: Relation OrderBook

And the following staternent will actualize BigOrd:

> Order-Book5 < - [Cuslomer, Bigorder] in OrderBook ;

The procedure of actualizing BigOrd is as follows:

1. In the parent leoel, copy Order to BigOrd.

2. Extract the relational statement /rom the string, parse it (the parser will be de-
scribed in next section); the string wdl be altered /rom "[Product] where A mount
> 8 in Order" to "[id, Product] where Arnovnt > 8 in Order".

3. Cal1 Relzz to get the resulting subrelation, YBigOrd < - [.id, Product,] where
Amount > 8 in Order".

4. Join back the resulting subnlation &th the parent relation on .id. Y&de~Book
< - OrderBook [BigOrd ijoin .id] BigOrd".
5. Update system tables.

The result is shown in Figure 4.13

CHAPTER 4. IMPLEMENT,-1TION OF NESTED RELATlONS

BigOrd
-----------
-----
Customer - -BigOrd
--- -,id
- - Product
- - - - -Amount
---
Ann 0 - - - - - -O W 9
Ann 1 ------1 2 10
Ping 2 -----*2 M 12
- - - - - - - - - - - - - - - - 3 _ - - -Y- - - - - - -
Sam 3 1O

Figure -1.13: hctualized BigOrd

Parser

In general domain algebra operations. we can write regular relational expressions with
some limitations. i.e.. we can not include vertical operations in the quoted relational
expression.
Since w e cal1 Reiix again to get the resulting relation, we need to preprocess the
statement. W e bbuilcl a small parser to preprocess the expression.
For example. *[Produet] where . - h o u n t > 8 i n Order' will beconie *[.id. Product]
where Amo~tnl> 8 i n Order'. T h e automaton of the parser is shown in Figure 4.14.
Suppose we have -A [a ijoin b] BE. The flow of its automaton is:

1. The automaton reads 'A'. It stays at the start. The output is "A'.

2. The automaton reads '['. It goes to state 1. The outputs is "A [".

3. The automaton reads 'a'. It stays a t state 1. The output is *.4 [ an

4. The automaton reads 'ijoin'. It stays at state 1. The output is -A [ a, .id ijoin"

5. The automaton reads 'b'. It stays at state 1. The output is "A [ a, .id ijon b"

6. The automaton reads '1'. It goes back to the start. The output is &A [ a, .id
ijoin b. .idln
CHAPTER 4. IMPLEMENTATION OF NESTED RELATIONS 72

7. The automaton reads 'B'. It stays at the start. The output is 'A [ a, .id ijoin
b, .id] B"

S. The automaton reads EOF. It stops and returns the obtained output.
CHAPTER 4. IMPLEMENTATZON OF NESTED RELATIONS

w
her than 'ï' other than

Algonthm:

For state start:

if next token is go to state 1, else stoy at state start
For state 1:
if next token is " l n
add .id before
go to state start
else if next token is any join token
add .id before the join token, for example, '' id i j o i n "
stay at state 1

join token: ijoin, djoin, ujoin, sjoin, ljoin, rjoin,

drjoin, natjoin dljoin, natjoin, dljoin,
gtjoin, sup, eqjoin, sub, ltjoin, sep,
qejoin , lejoin, iejoin, div, -gejoin, -sup,
-eqjoin, -sub, -Itjoin, icomp, natcomp

Figure 4.14: The parser to parse the ernbedded general relational expression
Chapter 5

Conclusion

Nested relations have been explored thoroughly in past decades, with the major re-
search direct ion focused on nesting and unnesting [Jae82][Fis851[K0r89][TakSS]. In
Our approach, we build nested relations upon Bat relations. We show that Bat rela-
t ions are powerful enough to model nested relations and to facilitate nested relation
queries. The purpose of this thesis is to begin to integrate nested relations into a ce-
Iational database programming language (Re1ix)by integating the relationai algebra
into the domain algebra.

5.1 Summary
We built our nested relation model upon the original Relix database model. Relix is
powerful enough to support nested relations. No modifications have been made to
the original database engine itself. However some extensions were made to facilitate
the process of integration and to provide new features.

a A new system attribute .id has been added to Relix , which provides a way of
linking the parent relation to its included nested relations.

a One level of nesting has been integrated into Relix.

 a A part of the relational operator can be added to the domain algebra. This
partially eliminates the difference between domains and relations.

Our irnplementation showed that Relix is powerful enough to include nested rela-
tions, and that it is convenient to add nested relations to the system. The relational
operations, such as ujoin, sjoin, ijoin, which are added to dornain operations, function
well.
However, the surrogate mechanism we used is a bit simple, and we bave not been
able to include more information in the surrogates except to use it to keep links
between nested child relations and the parent relation. No large-scale tests have been
done, since it is beyond the scope of this M.Sc. thesis.

5.2 Future Work

So Ear, we have only implemented one level of nesting in Relix, which is the first
step towards fully implementing the features of nested relations. There axe still more
features that can be added such as:

a Implementing multiple nesting and recursive nesting. 'To date, we have oniy im-
plemented one level of nesting, which provides a prototype for multiple nesting.
Theoretically, it is possible to build infinite levels of nested relations.

a F d y integrating the relational algebra into the domain algebra. Only a part
of relational algebra has been integrated into domain algebra to date. Further
work can be done on functional mapping and partial function mapping on nested
relations.

Cornbining nested relations with procedure abstraction and to implement com-

plex objects. A p r o c e d w facility has been recently added to the Relix s y t
tem [Lui96]. We could extend certain procedures to nested relations. Those
procedures can be viewed as methods to manipulate a certain nested relation,
which cm then be treated as a complex object.
Bibliography

[CodiO] E. F. Codd. A Relational Model of Data for Large Shared Data Banks.
Communications o j the ACM 13(6). Oct. 1970. pp.337-387

[Codi'2] E. F Codd. A Data Base Sublanguage Founded on the Relational Calcii-

lus. Proceedings of 1971 ACM SIGFIDET Workshop on D a t a Descn'p-
tion, Access and G n t r o l .

[ DesSS] A. Deshpande. D. Van Gucht. An implementation for Nested Relational

Database. Proceedings of the L4t h Internat ional Conference on Véry
L a y e Rn!« Bases. April 1988. pp. 266-274

[Fis851 P. C. Fischer, D. Van Gucht. Determining when a Structure is a Nested

Relation. Proceedings of the 11th Internat ional Con ference o n Venj Large
Data Baes. August 1985, pp. 171-180

[.J aeS?] G. Jaeschke. H-J. Schek. Remarks o n the hlgebra of Xon-First-Normal-

Form Relations. Proceedings of the First A CiVI SICA CT-SICiCIOD Sym-
posium on Pnncipfes of Databnse Systems. Mar& L 9-2, pp. 124- 13s

[JohX] S. C. Johnson. kacc: Yet another compiler-compiler. Technical Report 32.

A T k T Bell Laboratories, Murray Hill. N.J., 1975.
[KorY91 H. F. Korth, M. A. Roth. Query Languages for Nested Relational
Databases. lrlested Relations and Cornplex Objects in Database. Lecture
Notes in Cornomputer Science, Springer-Verlag, New York 1989.

[La1861 N. Laliberté. Design and implementation of a Primary Memory Version

of Aldat. Master's thesis, McGili University, Montreal, Canada, 1986.

[Les7 4 M. E. Lesk. Lex: a lexical analyzer generator. Technical Report 39.

ATScT Bell Laboratories, Murray Hill. N.J., 1955.
[LevS?] M. Levene. T h e Nested Universal Relational Database Model. Lecture
Notes in Cornputer Science. Spcinger-Verlag? New York. 1992
R. Lui. Implementation of Procedure in a Database Programming L m -
guage. Master's t hesis, McGill University, Montreal, Canada, 1996.

A. Makinouchi. A consideration on normal forrn of not-necessarily-

normalized relation in the relational data model. Proceedings of 3rd In-
ternational Conference on VLDB, Tokyo, pp. 447-453, 1977.
T. H. Merrett. MRDS: An Algebraic Relational Database System. In
Canadian Cornputer ConJennce, Montreal, pp. 102-124, May 1976
T. H . Merret t . Relations as programming language elements. lnformation
Processing Letters, 6(1):29-33, Feb. 1977.
T. H. Merrett. Relational Information Systems. Reston Publishing Com-
pany, Reston, Virginia, 1984.

G. Ozsoyoglu, 2. hl. Ozsoyoglu, V. Matos. Extending relational dgebra

and relat ional calculus with set-valued attributes and aggregate functions.
ACM Transaction on Database Systems, 12(4) Dec. 1987, pp. 566-593
2. M. Ozsoyoglu k L. Y Yuan. A design method for nested relational
databases. Proceedings of 3rd IEEE conference on Data Engineering,
Los Angeles, pp. 599-608, 1987

2. M. Ozsoyoglu & L. Y Yuan. On Nonnalization in Nested Relatonal

Databases. Nested Relations and Complex Objects in Database. Lecture
Notes in Cornputer Science, Springer-Verlag, New York, 1989.

P. Pistor, F. Anderson. Designing a Generalized N F 2 Model With An

SQL-Type language Interface. Proceedings of the 12th International Con-
ference on Very Large Data Bases, August 1986, pp. 278-285.

J. Paredaens, D. Van Gucht. Converting Nested Algebra Expressions

into Flat Algebra Expressions. ACM Transactions on Database Systems
17(1), March 1992, pp. 6.193.

M. A. Roth, H. F. Korth, A. Silberschatz. Extended algebra and calculus

for nested relational databases. ACM Transactions on Database Systems
13(4), Dec. 1988, pp. 390-417.

H. J. Schek, P. Pistor. Data Structure for an Integrated Data Base

Management and Information Retrieve System. Proceedings of the 8t h
International Conference on Very Large Data Bases, Sep. 1982, pp. 197-
207.
M. H. Scholl, H. B. Paul, H. J Scholl. Supporting Flot Relations by a
Nested Relational Kernel. Proceedings of the 13th International Confer-
ence on Very Large Data Bases, Sep. 1987, pp. 137-147.

M. Scholl, S. Abiteboul, F. Bancilhon, N. Bidoit, S. Garnerman,

D.Plateau, P. Richard, A. Verroust. VERSO: A Database Machine Based
on Nested Relations. Nested Relations and Complex Objects in Database,
Lecture Notes in Computer Science, Springer-Verlag, NY, 1989.
M. Stonebraker. Object-Relationai DBMSs. Morgan Kaufmann Publish-
ers Inc., San Francisco, California, 1996.

K. Takeda. On the Uniqueness of Nested Relations. Nested Relations

and Complex Objects in Databases, Lecture notes in Computer Science,
Springer-Verlag, New York, 1989.
A. U. Tansel, L. Garnett. On Roth, Korth, and Silberschatz's Extended
Aigebra and Calculus for Nested Relational Databases. ACM Transac-
tions on Database Systems, 17(2), June 1992, pp. 374-383.

S. Thomas, P. Fischer. Nested relational structures. In Advances in

Computing Research III, The Theory of Databases, P.C.Kanellakis, Ed.
J AI Press, Greenwich, Conn., 1986.
 MASE EVAiuNiÛN
TEST TARGET (QA-3)

APPLIED 4 IMAGE, lnc

- 1653 East Main Street

=
-
.---- 0
Rochester. NY 14809 USA
Phone: 71ô/48~-O300
Fax: 716228&5989