Transcript Object-Oriented Query Languages and Views

ADBIS - DASFAA'2000
Fourth International Symposium on Advances in Databases and Information Systems
September 5-8, 2000, Prague, Czech Republic
Tutorial:
Object-Oriented Query Languages and Views
Part 1: Basic concepts and issues
Lecturer:
Kazimierz Subieta
Polish-Japanese Institute of Information Technology,
Warsaw, Poland
Institute of Computer Science
Polish Academy of Sciences, Warsaw, Poland
[email protected]
http://www.ipipan.waw.pl/~subieta
K.Subieta. Object-Oriented Query Languages and Views, slide 1
Sept. 2000
What Is a Query Language?
A lot of views...
User-friendly: A language for a person who does not want to knows anything
about databases, but wants to operate with them.
Theoretical achievement: A syntactic variant of some famous mathematical
theory, e.g. logic. Used mainly to produce a next paper to a next conference (:-)).
Ad hoc interactive database language: A facility for quick retrieval and simple
updates through formalized commands (select...from..., update...set..., ...) or
through some simple visual interfaces: forms, graphs, menus, etc.
A very-high-level programming construct to be embedded into a popular
programming language, e.g. SQL embedded into C.
A very-high-level programming construct integrated with a
programming language, e.g. PL/SQL, many 4GLs, and (recently) SQL3.
K.Subieta. Object-Oriented Query Languages and Views, slide 2
new
Sept. 2000
Properties of Query Languages
Abstract, conceptual, data independent: no concepts related to physical level of
data (file organizations, indices, moving data between disk and main memory, etc.
Declarative (non-procedural): determines what is to be done rather than how.
Macroscopic: from the point of the user a query determines simultaneous
(parallel) actions on many data.
Natural: supporting a natural way of thinking of the user, supporting conceptual
modeling, easy to learn and use.
Efficient: acceptable response time (performance) through automatic query
optimization.
Universal: giving the possibility to express every (reasonable) request.
Independent of an application domain: supporting all applications of the given
DBMS.
Lately bound, interpreted: supporting ad hoc queries and various abstractions
stored in a database (views, stored procedures, active rules, etc.).
K.Subieta. Object-Oriented Query Languages and Views, slide 3
Sept. 2000
A Query Language in a Database Environment
A tool for an end user enabling him/her interactive (ad hoc) querying,
generating reports and updating data stored in a database.
Very-high-level conceptual language constructs for database programming.
Defining integrity constraints preventing illegal operations/states.
Defining subschemas and access restrictions.
Defining virtual views, materialized views, derived (replicated) data, and
procedures stored in a database.
Components of scripts in 4GL-s and RAD tools.
Defining active rules (triggers) and deductive rules.
Determining data to be selected and transmitted in distributed databases;
interoperability between heterogeneous/remote databases (ODBC, JDBC, ...).
...... several other applications
K.Subieta. Object-Oriented Query Languages and Views, slide 4
Sept. 2000
Query Optimization
A query language is unacceptable without automatic optimization of queries.
Typical optimization methods:
Rewriting. A query q1 is substituted with a semantically equivalent query q2
promising better performance. E.g. performing selections before joins or
removing dead (not used) parts of queries. The methods have no negative side
effects. They require, however, regularity and homogeneity of the language
definition and deep understanding of formal semantics;
Auxiliary data structures and special data organizations: indices, tables of
pointers, hash coding, etc;
Caching results of queries and then reusing them (materialized views);
Simultaneous optimization of many queries;
Selecting an optimal query evaluation plan.
Optimization of object-oriented QLs is in infancy. There is a lot of wishful thinking
and poorly justified hopes concerning the optimization potential of particular theories.
This is one of the reasons of slow adoption of object QLs in the commercial world.
K.Subieta. Object-Oriented Query Languages and Views, slide 5
Sept. 2000
Optimizable and Non-optimizable Queries
An entire query language
Efficient (optimizable) queries
Most useful queries
In any query language there are non-optimizable queries.
The non-optimizable part of a QL should be as small as possible.
In real QLs not all useful queries are optimizable. Poorly defined, informal or
irregular features of a QL decrease the optimization potential.
K.Subieta. Object-Oriented Query Languages and Views, slide 6
Sept. 2000
OQL and SQL3 on One Slide
OQL is a part of the ODMG standard. It is claimed to be a compatible extension
of SQL, but actually OQL retains some syntactic patterns of SQL only.
Semantically OQL is very different from SQL, because it follows an object model,
which is incompatible with the relational model. OQL does not deal with updating
and does not define SQL-like facilities such as views, triggers and stored
procedures. OQL statements can be embedded into Java, C++ and Smalltalk, with
a lot of impedance mismatch. The semantics of OQL is defined poorly and
inconsistently, thus probably it is not fully implementable.
SQL3 is a new SQL standard developed by ANSI and ISO. In contrast to its
predecessors SQL3 is assumed to be a programming language with full
computational power. The main data structure is a table, equipped however with a
lot of options (thus using the term “relational” makes no sense). SQL3 supports
user-defined abstract data types, including methods, object identifiers, subtypes,
inheritance and polymorphism. Further facilities include control statements and
parameterized types. Together with an extremely rich collection of various
features, SQL3 is claimed to be downward compatible with SQL-92 and follows
the sweet select...from...where... syntax. The standard is eclectic and extremely
huge (more than 1000 pages), thus probably it is not fully implementable.
K.Subieta. Object-Oriented Query Languages and Views, slide 7
Sept. 2000
Is Java an Alternative to Query Languages?
There is a lot of discussion around the role of Java in database programming.
Java is a very important language.
But...
Java offers rather classical low-level (object-oriented) programming.
The portability of Java bytecode is low-level. There are many details of database
interfaces outside the Java bytecode standard.
The Java object model is not powerful enough to be a database model.
The Java database interfaces (JDBC, SQLJ,...) are not object-oriented. They
present SQL interfaces to relational databases, wrapped into the Java syntax.
Java does not solve the problem of storing important abstractions within a
database (views, database procedures, triggers, etc.)
Java itself is not an answer to database problems. It presents much lover
level of database programming than programming via query languages.
Java + any query language do not avoid the impedance mismatch.
K.Subieta. Object-Oriented Query Languages and Views, slide 8
Sept. 2000
Requirements to Object Query Languages
Conceptual simplicity, generality and minimality: clean and precise
semantics, a small set of semantic primitives, no redundant constructs.
Pragmatic universality: the possibility to formulate any request.
Universality of an approach to semantics: no black holes in the semantics,
treating every semantic inconsistency or irregularity as a very big problem.
Compositionality and orthogonality: no big syntactic constructs, every
reasonable combination of constructs should be allowed.
Modularity: the possibility to create complex encapsulated reusable units.
Homogeneous and consistent approach to all concepts of the underlying
object model: complex objects, classes, types, interfaces, ADTs, inheritance,
methods, encapsulation, polymorphism, etc.
Homogeneous and consistent approach to integration of the query language
with programming constructs (updating, etc.) and abstractions (views,
methods, stored procedures, parameters of procedures, etc.).
High potential for query optimization.
K.Subieta. Object-Oriented Query Languages and Views, slide 9
Sept. 2000
Coupling a QL with a PL
Loose coupling (“embedding”): A QL is developed independently of a PL. An
additional interface (“glue”) is implemented, enabling the use of QL within the
underlying PL. The complexity of the interface presents a problem. The
incompatibility between QL and PL is referred to as impedance mismatch. Typical
cases: SQL + C, JDBC+Java, OQL + C++, OQL + Java
Advantages: the programmers can deal with their favorite PL, the API to a
database is independent of PLs.
Disadvantages: aesthetically ugly, a lot of limitations, programs are longer than
necessary, tricky programming, poor maintainability.
Tight coupling (“seamless integration”): Queries are building blocks for
programming constructs. No special interface between QL and programming
constructs is provided.
Typical cases: Oracle PL/SQL, many 4GLs, DBPL, LOQIS, SQL3
Advantages: a consistent homogeneous solution, no impedance mismatch.
Disadvantages: implies a new programming language, which is difficult to
promote in the current commercial world.
K.Subieta. Object-Oriented Query Languages and Views, slide 10
Sept. 2000
What is Impedance Mismatch?
Incompatibility between a PL and a QL to be embedded. It concerns:
Syntax: Two different grammars within one programming interface;
Type systems: different types, lack of bulk types in PL, no static typing of QL;
Semantics and pragmatics: declarative QL vs. procedural PL;
Abstraction levels: data independence of QL vs. deep data dependence of PL;
Binding phases and mechanisms: late binding of QL vs. early binding of PL;
Name spaces and scoping rules: two incompatible name spaces the programmer
deals with, stack based scoping rules of PL are not respected by QL;
Null values: they are ignored in PL, special mapping tricks are required;
Iteration mechanisms: implicit in QL (selection, projection,...), explicit in PL;
Persistence: QL - only persistent data, PL - only volatile data; in PL special
mechanisms are required to copy persistent data into volatile memory and v/v.
Generic programming: reflection in QL (see dynamic SQL), other techniques
(e.g. casting, templates) in PL.
Looking at the above, claims that ODMG Java + OQL avoid the impedance
mismatch are at least dubious.
K.Subieta. Object-Oriented Query Languages and Views, slide 11
Sept. 2000
Object-Orientedness in Databases
The commercial world: manifestos, standards and products - no agreement.
OODB Manifesto, 3rd Generation DB Manifesto, Third Manifesto, ODMG standard,
SQL3 standard (SQL 1999, SQL 2000), persistent Java, XML as a database model,
CORBA as a database model, Gemstone, Versant, O2, ObjectStore, Poet, Objectivity/DB,
Uni SQL, Oracle 8, Informix Dynamic Server, and others.
Useful. But... Eclectic solutions, legacy burden, design monsters, underspecified
semantics, non-universal solutions, redundant constructs, many inconsistencies.
The academic world: theories and prototypes - no agreement.
Nested relational algebras, F-logic, comprehensions, monoid calculus, object algebras,
object calculi, structural recursion, functional approaches, and others.
Theories neglect vital aspects of object-orientedness, present wishful thinking
(e.g. concerning a mapping from OQL), are limited, are conceptually or
mathematically invalid (e.g. object algebras). False stereotypes inherited from the
relational model (e.g. concerning the role of an algebra in query optimization).
The today’s state-of-the-art is PREMATURE (despite thousands of papers).
The stability - not sooner than after 5-10 years. Thus in this tutorial we will
discuss concepts without referring to a particular standard, product or theory.
K.Subieta. Object-Oriented Query Languages and Views, slide 12
Sept. 2000
Complex Objects
 Many data hierarchy levels (no limitations);
 Nested repeating attributes (collections);
 Large objects (BLOBs) as regular values;
 References (pointers) to other objects.
Conceptual
modeling
EMPLOYEE
COMPANY IBM
ENO E127
JOB designer
PREV_JOB
WHEN 1975-77
NAME Smith
JOB analyst
COMPANY ICL
PHOTO
PREV_JOB
WHEN 1977-90
WORKS_IN
K.Subieta. Object-Oriented Query Languages and Views, slide 13
Sept. 2000
Classes
Bad definitions:
A class is a collection of objects (wrong: what about methods, inheritance, ...?);
A class is a blueprint for objects (wrong: only creation of objects is regarded).
Correct definition:
A class is a conceptual (imaginary) entity storing invariant properties of objects.
(Invariant properties are factored out from objects to their classes.)
Typical invariant properties include:
• Names and types of objects’ attributes (i.e. a type of an object);
• Methods that can be applied to an object.
But invariant properties can be of various kind:
• A name of an object (ODMG);
• Rules for events or exceptions that can occur on objects (CORBA, ODMG);
• Links (pointers, references relationships) to other objects (ODMG);
• Active rules (triggers) and integrity constraints;
• Default values for attributes;
• ...
Sometimes a class is a regular run-time object (Self).
K.Subieta. Object-Oriented Query Languages and Views, slide 14
Sept. 2000
Encapsulation, Interfaces
Encapsulation and information hiding
are basic principles of any engineering, including software engineering.
A TV set encapsulates a lot details, which the user does not need to know.
De-encapsulation of these details may result in an electric shock of the user.
A similar threat concerns a programmer working on non-encapsulated software.
Interface - general definition: It consists of everything that the programmer can
use or has to know in order to correctly process the object. An interface should not
include unnecessary (physical) details of objects’ construction or operation.
Interface - particular definition (CORBA, ODMG, Java): An interface a
specification of all public properties (attributes and methods) of an object that the
programmer can use in a particular context.
Interface is a concept different from class (e.g. classes can be sold, interfaces - not);
Interface is a concept different from type (e.g. exceptions are not relevant to types).
Bad understanding of encapsulation (all attributes are private, only some methods
are public) has led to the absurd thesis on contradiction between encapsulation
and query languages.
K.Subieta. Object-Oriented Query Languages and Views, slide 15
Sept. 2000
Inheritance
If two or more classes have common invariants, then they can be factored out to
another class. Hence classes are organized in a hierarchy.
An object inherits invariants from its class and from all its superclasses.
Multi-inheritance – many superclases are allowed.
Employee
name
date of birth
salary
age
salary net
Student
name
date of birth
faculty
grades
age
average grade
K.Subieta. Object-Oriented Query Languages and Views, slide 16
Person
name
date of birth
age
Employee
salary
salary net
Student
faculty
grades
average grade
Sept. 2000
Methods and Messages
A method is a procedure stored within a class.
It acts on an environment consisting of:
internal environment (attributes) of the currently processed object;
private and public properties of the same class and public properties of all its
superclasess;
base environment, which includes database, volatile variables/objects of the user
session and global environment (environment variables, libraries);
public properties of currently active program modules.
A message is a call of a method.
Message passing does not mean parallel asynchronous communication between
autonomous agents (this was false association made by OO pioneers).
The usual syntax for messages: object . methodName [( parameters )]
e.g.
(Person where name = “Smith”). age
Query languages can introduce other syntax for messages, e.g.
e.g.
K.Subieta. Object-Oriented Query Languages and Views, slide 17
(Person where age > 30) . name
Sept. 2000
Types
A type is an expression, which constraints the content of objects or value.
Types determine input/output properties of operators, functions, procedures
and methods. Specification of types is obligatory in strongly typed languages.
Types formally restrict a context of the use of objects, operators, methods, etc.
Strong (static) type checking (strong typing): every use of objects, operators,
functions, methods, etc. is checked at compilation time.
Typing safety: more than 80% of programmer’s errors are detected at compilation.
Dynamic type checking (dynamic typing): types are checked at run-time. Less
efficient w.r.t. detecting errors.
ODMG OQL is strongly typed, but the typing system is inconsistent (S.Alagic).
SQL and SQL3 are dynamically typed.
Strong typing decreases the power of a language (generic programming) and
presents a difficulty for developers of DBMS. Thus the attitude of the commercial
world to strong typing is undetermined (officially approved, practically neglected).
K.Subieta. Object-Oriented Query Languages and Views, slide 18
Sept. 2000
Links (references, pointers)
Objects can be connected by explicit pointer links. Links support conceptual
modeling (see OMT, UML, etc.). Links can be directly used in queries (through path
expressions). They much simplify queries and are more efficient than joins.
EMPLOYEE
NAME Brown
EMPLOYEE
NAME Jones
EMPLOYEE
NAME Smith
SALARY 3500
SALARY 2500
SALARY 2000
WORKS_IN
WORKS_IN
WORKS_IN
COMPANY
BOSS
EMPLOYS
NAME Syntex
LOC London
EMPLOYS
EMPLOYS
A path expression in SBQL:
Name of the Smith’s boss:
(EMPLOEE where NAME = “Smith”).WORKS_IN.COMPANY.BOSS.EMPLOYEE.NAME
K.Subieta. Object-Oriented Query Languages and Views, slide 19
Sept. 2000
Class Extents
An extent is a named collection of objects being current members of a class.
The concept has roots in the relational model, where a declaration of a table (a
protoplast of the class concept) is sticked with creation of the table (i.e. extent).
In PLs declarations of types/classes are separated from declaration/creation of
corresponding variables/objects.
An extent is a bit doubtful concept. For example, having a class Person and its
subclass Employee the extent for Person and the extent for Employee has a nonempty intersection: some parts of objects Employee are “virtual” parts of the extent
for Person. This can be the source of inconsistencies and programmers’ errors.
It is also easy to imagine the situation, where a class must have not one but many
extents. E.g. the class EmployeePhotoAlbum has an extent for each employee.
Probably, the extent concept gives very little for the database designers and requires
additional attention of programmers. In my opinion, it should be dropped.
K.Subieta. Object-Oriented Query Languages and Views, slide 20
Sept. 2000
Collections
The relational model deals only with one collection - a table (relation).
Object models (ODMG) assume more collections, in particular:
sets (no duplicates, no order);
bags (duplicates are allowed, no order);
sequences (duplicates are allowed, the order is informative);
arrays - as sequences, no inserting/deleting elements except top, access through
order numbers.
Collections can be nested, e.g. collection-valued attributes are allowed.
Moreover, collection-valued pointer links are allowed too.
Collections have a big meaning for conceptual modeling.
Nested collections simplify queries (navigations instead of joins).
Typical object-oriented programming languages have no explicit collection types;
collections must be modelled by some tricks.
K.Subieta. Object-Oriented Query Languages and Views, slide 21
Sept. 2000
The OODBMS Ideals
Orthogonal persistence: the same types can be applied to persistent and
volatile objects. Ergo: a query language should process persistent and volatile
data uniformly. The ideal much reduces the complexity of a QL.
Object relativism: each object consists of objects; there are no other concepts
that are used for description of objects. Ergo: (atomic) attributes are objects,
links are objects, each object can be a component of a higher-level object, etc.
The ideal much reduces the complexity of a QL.
Total internal identification: each run-time program entity, which can be
separately retrieved, bound, updated, inserted, indexed, protected, locked, etc.,
must possess an unique internal identifier. Internal identifiers can be used as
references (l-values) by various language constructs (e.g. updating). The ideal
much simplifies semantics of a QL and makes it consistent.
Unfortunately, the above ideals are not respected by commercial OODBMS-s
and standards.
K.Subieta. Object-Oriented Query Languages and Views, slide 22
Sept. 2000
Syntax, Semantics and Pragmatics of Languages
Each language, including query languages, has three aspects:
Syntax: describes how to build correct expressions of the language from the
alphabet (basic symbols).
Semantics: describes what expressions of the language denote. Semantics is the
basis for implementation, in particular for query optimization. Semantics is usually
syntax-driven: semantic rules are built on top of syntax rules.
Pragmatics: describes how to use the language to accomplish practical needs. It
deals with mapping concrete problems or tasks onto expressions of the language.
Pragmatics is informal, important for teaching, frequently the most difficult for
users. (Even the developers of SQL3 have problems how to use their own creature!)
Many languages are explained through syntax and pragmatics. Few languages are
specified through precise formal semantics.
If semantics is underspecified, then each implementation of the language augments
the specification on its own way. This is the reason of low portability of languages.
K.Subieta. Object-Oriented Query Languages and Views, slide 23
Sept. 2000
Semantics of Query Languages
Semantics of a query is a function which maps a state into a result.
For any query language, what we have to define?
Query - a syntactic domain of all queries;
Result - a domain containing all possible results of queries;
State - a domain containing all possible states.
General definitions of semantics:
sem : Query  (State  Result)
For queries with no side effects;
sem : Query  (State  (Result  State)) For queries with side effects;
sem : Query  (State  State))
For imperative queries (e.g. the
update clause of SQL).
Formal semantics can be different from implementation (see SQL).
Stateless approaches to query languages (logic, algebra) have severe limitations.
K.Subieta. Object-Oriented Query Languages and Views, slide 24
Sept. 2000
Four sides of a query language
As follows from the previous slide, description of any query language must involve
four sides:
Description of data stored in a database that are to be queried, i.e. the description
of data/object model (formal definition of the set State);
Description of syntax (formal definition of the set Query);
Description of results returned by queries (formal definition of the set Result);
Description of the mapping from queries into results (formal definition of the
mapping sem).
Usually in practical languages, these definitions are incomplete, inconsistent or
even sloppy. If one has to be serious with implementation and query optimization
all these definitions must be as clean and precise as possible.
Unfortunately, it is not easy to present all sides in detail during 180 minutes
of the tutorial.
K.Subieta. Object-Oriented Query Languages and Views, slide 25
Sept. 2000
What is “state”?
For real QLs state is more than database state. A state includes:
database: all data, objects, classes, methods, etc. in the database;
volatile objects/variables/... of the run-time environment of a user session;
local objects/variables/... of all currently executed procedures, functions, methods;
global environment: environment variables, libraries, files, etc.
A state includes temporary internal structures of the query processing machine:
Get 10 best-paid employees:
select * from Employee as x
where count( select * from Employee as y where y.salary > x.salary ) < 10
If the subquery select * from Employee as y where y.salary > x.salary
has to be evaluated independently of the context, then the “free” variable x must be
stored on an internal structure of the query processing machine.
This internal structure (stack) augments the concept of “state”.
K.Subieta. Object-Oriented Query Languages and Views, slide 26
Sept. 2000
The Closure Property
It is a property of a query language (or a theoretical framework) saying that
the input and output of queries should belong to the same formal domain.
For relational QLs, the input consists of tables and the output is a table. For objectoriented QLs the input is a set of objects and the output is a set of objects too.
According to its advocates, the closure property is a condition for nesting queries.
I disagree. Essentially, the closure property is a false inconsistent stereotype
inherited from the relational model. The closure property does not hold even for
SQL. E.g. input tables are named and output tables are unnamed; hence there is a
big semantic difference between input and output tables.
For object-oriented QLs the closure property is a conceptual nonsense.
In particular, it leads to subdivision of queries onto “object preserving” and
“object generating”, which is a nonsense too.
We will formulate QLs semantics in consistent and formally correct terms,
without subdividing queries onto “object preserving” and “object generating”.
K.Subieta. Object-Oriented Query Languages and Views, slide 27
Sept. 2000
Results of Queries
In ODMG terms, queries return literals. We generalize this concept.
The recursive definition below defines the domain Result:
Each atomic value  Result.
Each reference (to an object, attribute, link, method, view, etc.)  Result.
If v  Result, n is a name, then n(v)  Result. Such results will be called
binders.
If v1, v2, v3, ...  Result, then row{ v1, v2, v3, ...}  Result. In general, the order
of elements in the row is essential. This construct generalizes a tuple known from
relational systems.
If v1, v2, v3, ...  Result, then set{ v1, v2, v3, ... }, bag{ v1, v2, v3, ... }, sequence{
v1, v2, v3, ... }, ...  Result.
There is no other results.
In our terms queries never return objects, but can return references to objects,
or more precisely, some structures built upon references, values and names.
K.Subieta. Object-Oriented Query Languages and Views, slide 28
Sept. 2000
Example results of queries
Atomic:
25, "Smith", i11, i18
Complex:
i1, i2 ,... - references
row{i1, i21}
bag{ row{i1, i21}, row{i7, i26} }
bag{row{ 2, Lecture(i26), Stud( bag{
row{ n("Russel"), y(i36) },
row{ n("Black"), y(i30) }})}
We present bags of rows as rectangular tables (similarly to relational tables),
for example:
4
i33
i40
i47
p(i1)
p(i7 )
K.Subieta. Object-Oriented Query Languages and Views, slide 29
“Russell”
“Jones”
“Black”
i15
i2
i8
52
44
lect(i21)
lect(i26)
Sept. 2000
ADBIS - DASFAA'2000
Fourth International Symposium on Advances in Databases and Information Systems
September 5-8, 2000, Prague, Czech Republic
Tutorial:
Object-Oriented Query Languages and Views
Part 2: The Stack-Based Approach
Lecturer:
Kazimierz Subieta
Polish-Japanese Institute of Information Technology,
Warsaw, Poland
Institute of Computer Science
Polish Academy of Sciences, Warsaw, Poland
[email protected]
http://www.ipipan.waw.pl/~subieta
K.Subieta. Object-Oriented Query Languages and Views, slide 30
Sept. 2000
Why the stack-based approach (SBA) to QLs?
The motto of SBA (frequently neglected by database researchers):
Each, even apparently small semantic problem is a big problem.
We would like to achieve:
Universality concerning both data structures an QL/PL functionalities;
Modularity and compositionality (the hierarchy of conceptual abstractions);
Regularity, full orthogonality of the concepts;
Minimality of semantic primitives;
Clean and precise (formal) semantics;
Uniform approach and full integration with procedural capabilities:
updating, procedures, views, methods, etc.;
Precise treatment of object-oriented concepts (classes, encapsulation, ...);
High potential for query optimization.
K.Subieta. Object-Oriented Query Languages and Views, slide 31
Sept. 2000
The Environment Stack (ES)
In PLs the environment stack is a basic mechanism to accomplish:
Abstraction: the programmer can abstract from internal details of procedures.
Semantic independence and program reuse: the meaning and the behaviour of
procedural abstractions is independent on the context of its use.
Recursion: A procedure (function, method, view) can call other procedures, in
particular, can call itself. Encapsulation of local environments is preserved.
Consistent binding: Name x is bound to the most local definition or declaration
of x. Other definitions or declarations of the name x should be allowed.
Parameter passing: The stack makes it possible to store and manage parameters
of procedures and to accomplish consistently parameter passing methods.
Proper scoping: a program entity should act only on the data environment and
name space that the programmer who has programmed the entity was aware of.
In SBA the environment stack has a new role: consistent semantic
mechanism for definition and implementation of query operators.
K.Subieta. Object-Oriented Query Languages and Views, slide 32
Sept. 2000
Assumptions of the SBA: syntax
Unification of PL expressions and queries:
2+2
(x + y) * z
(x + (EMP where (NAME = “Smith”)).SAL) * z
2, “Smith”, 1000,...
x, y, z, EMP, NAME, SAL, ...
+,- ,*, /, =, >, where, ., ...
sin, sqrt, sum, count, distinct,...
}
EMP, NAME, SAL
persistent data
x, y, z
volatile data
atomic queries
binary operators
unary operators
q1, q2 are queries,  is a binary operator 
q is a query,
 is a unary operator 
q1  q2 is a query
(q)
is a query
Abstract syntax and compositionality: no big syntactic/semantic constructs,
e.g. no famous select ... from ... where ... group by ... having ... order by ...
Big constructs decrease orthogonality, maintainability, reusability and optimization
potential, are more difficult in implementation, are the reason of irregularities.
K.Subieta. Object-Oriented Query Languages and Views, slide 33
Sept. 2000
Assumptions of the SBA: semantics
The naming-scoping-binding principle:
Each name occurring in a query is bound to run-time database/program entities
(persistent objects, volatile objects, attributes, procedures, parameters, views,
methods,...) according to the actual scope for the name.
This concerns:
• names of persistent objects;
• names of objects’ attributes, sub-attributes, ...;
• auxiliary names (“variables”) defined within a query;
• names of transient objects, programming variables and their attributes;
• names of procedures, methods, operators, ...;
• names of parameters of procedures, methods,...;
• ..... any other name.
• Scopes are organized in ES with the “search-from-the-top” rule.
• ES is separated from the object store.
• Binding of a name can be multi-valued (macroscopic binding).
In SBA the domain State consists of: Object Store + Environment Stack.
K.Subieta. Object-Oriented Query Languages and Views, slide 34
Sept. 2000
An Abstract Store Model
The component of a ‘state”.
< i, n, v > atomic object
< i1, n, i2 > link object
< i, n, T > complex object
T is a set of objects
I - a set of internal identifiers
N - a set of external names
V - a set of atomic values, blobs,
compiled procedures, ...
Store:
A set of objects +
A set of identifiers (roots)
+
some obvious constraints
(uniqueness of identifiers,
referential integrities)
No record, tuple, array, set, and bag constructors in the model: essentially all
of them are collections of objects (“environments”).
No uniqueness of external names on any level of data hierarchy: modeling
bulk data.
Uniform treatment of relational, object-relational and pure object databases.
Classes, inheritance and encapsulation require extension.
K.Subieta. Object-Oriented Query Languages and Views, slide 35
later
Sept. 2000
Tiny Database
i1 EMP
i5 EMP
i9 EMP
i2 NAME Brown
i6 NAME Smith
i10 NAME Jones
i3 SAL 2500
i7 SAL 2000
i11 SAL 1500
i4 WORKS_IN i13
i8 WORKS_IN i17
i12 WORKS_IN i17
i13 DEPT
i17 DEPT
i14 DNAME Toys
i18 DNAME Sales
i15 LOC Paris
i19 LOC Berlin
EMP[0..*]
NAME
SAL
JOB[0..1]
age
WORKS_IN
DEPT[0..*]
DNAME
LOC[1..*]
i16 LOC London
K.Subieta. Object-Oriented Query Languages and Views, slide 36
Sept. 2000
Universality of the Store Model
Complex hierarchical objects can be defined (no limits of levels);
programming variables are treated as objects;
Orthogonal persistence: we abstract from the persistence status of objects and
variables, i.e. we define in the same way persistent and transient objects;
Object relativity: no difference in treatment of objects on any hierarchy level big advantage for the universality, minimality and simplicity of semantics.
Total internal identification: each entity stored in the store model, including
attributes, links, BLOBs, methods, views, etc. has a unique internal identifier;
Binary relationships (associations) can be defined via link objects; as in the
ODMG model we do not deal with ternary and higher order relationships and
attributes of relationships (they must be reduced to binary ones);
Bulk data: we deal with sets/bags. They are modeled by the same name assigned
to many objects on the same hierarchy level;
Relational structures: each tuple is understood as an object with subobjects;
K.Subieta. Object-Oriented Query Languages and Views, slide 37
Sept. 2000
What is binding?
Binding is substituting a name occurring in a query or a program
by a run-time program entity (entities).
For example:
• procedure name occurring in a program is substituted by a call of a machine code;
• variable name is substituted by an address of a main memory
• attribute name is substituted by an offset relatively to the beginning of a structure;
• object name is substituted by an object identifier.
Binding is early or static, if the substitution is made before the program is
executed (i.e. during compilation and linking).
Binding is late or dynamic, if the substitution is made during run time.
In query languages binding is usually dynamic, because of dynamic database features
(inserting new data, removing data, creating/removing views, etc.)
Static binding is sometimes used for optimization.
K.Subieta. Object-Oriented Query Languages and Views, slide 38
Sept. 2000
What is binder?
Binder is an internal structure to determine bindings.
A binder consists of two parts:
An external name defined by the
application designer or
programmer
An internal run-time program entity,
e.g. an object identifier, a value, a
procedure code.
This is an abstract view. In implementation binders may be not so explicit.
Binding: for each external name occurring in a query a proper binder is found;
then the name is substituted by the corresponding internal entity.
Binders are written as n(x), where n is an external name, x is an internal entity.
For a binder n( i ) name n may be different from the name of the object identified by i.
In query languages binders have additional roles.
Binders can be nested, i.e. x may consist of binders.
In general, a binder will be understood as a query result equipped with a name.
General definition of binders:
K.Subieta. Object-Oriented Query Languages and Views, slide 39
n  N, r  Result  n( r ) is a binder
Sept. 2000
The Environment Stack
It consists of sections. Each section is a set of binders.
The stack is growing and shrinking according to program/query nesting.
The most local data
are at the top.
......
Binders to local entities of
currently executed method
.....
NAME(i2) SAL(i3)
WORKS_IN(i4)
Binders to global entities
of the user session
The most global data
are at the bottom.
EMP(i1) EMP(i5) EMP(i9)
DEPT(i13) DEPT(i17)
The section of the
currently processed object
The section of the
“Tiny database”
Binders to entities of
the global environment
K.Subieta. Object-Oriented Query Languages and Views, slide 40
Sept. 2000
Binding through the environment stack
Binding a name search from the top:
......
G(“Mary”) X(i221)
.....
NAME(i2) SAL(i3)
WORKS_IN(i4)
Binding ( G ) = “Mary”
Binding ( X ) = i221
Binding( SAL ) = i3
Binding( EMP ) = {i1, i5, i9 }
Binding( DEPT ) = {i13, i17}
...
EMP(i1) EMP(i5) EMP(i9)
DEPT(i13) DEPT(i17)
...
• First the top section is visited, then lower sections are visited.
• The search is finished after a binder with the proper name is found.
• All binders with the proper name form the result of the search.
K.Subieta. Object-Oriented Query Languages and Views, slide 41
Sept. 2000
Opening a new scope by a query operator
In PLs opening a new scope (an activation record) at the top of an environment
stack is associated with an activation of a block or a procedure.
In SBA a new scope at the top of the environment stack is opened to evaluate a
query component in the context determined by another component.
A context
Operator
EMP
A subquery evaluated in the context
where
SAL > 1000
The ES state (in one iteration):
The ES state:
The new scope
opened
by where
NAME(i2) SAL(i3)
WORKS_IN(i4)
EMP(i1) EMP(i5) EMP(i9)
DEPT(i13) DEPT(i17)
EMP(i1) EMP(i5) EMP(i9)
DEPT(i13) DEPT(i17)
EMP is bound to {i1, i5, i9}
SAL is bound to i3
K.Subieta. Object-Oriented Query Languages and Views, slide 42
Sept. 2000
Function “nested”
Given identifier i of a complex object, the function nested returns binders
to direct sub-objects of the object identified by i.
For “Tiny database”:
A context
A subquery evaluated in the context
EMP where SAL > 1000
yields {i1, i5, i9}
nested( i1 ) = { NAME( i2 ), SAL( i3 ), WORKS_IN( i4 )}
nested( i5 ) = { NAME( i6 ), SAL( i7 ), WORKS_IN( i8 )}
Function nested determines a new scope - the environment in which the subquery
will be evaluated. The scope is pushed at the top of the environment stack.
Function nested is naturally generalized for any r  Result.
If l is a link, then nested( l) returns the binder to the entity that the link points to.
If b is a binder, then nested ( b ) returns b (no change).
For rows, nested ( row{v1, v2, ...} ) = nested(v1)  nested(v1) ...
K.Subieta. Object-Oriented Query Languages and Views, slide 43
Sept. 2000
The Language SBQL
A formalized variant of SQL-like languages, including ODMG OQL and SQL3.
It is relevant to relational, object-relational and object-oriented models.
Abstract syntax, free of sugar.
Syntax:
Literals, names, unary or binary operators, parentheses.
Orthogonality:
1000
EMP
SAL
Examples queries:
2+2
SAL > 1000
EMP where (SAL > 1000)
((( EMP where (SAL > 1000)) . WORKS_IN ) . DEPT ) . (DNAME, LOC)
Relativity:
Each query is evaluated relatively to the state of the environment stack.
Thus queries such as SAL > 1000 have semantics independent from the
context (providing ES contains a SAL binder).
Query operators are subdivided into algebraic and non-algebraic.
Algebraic operators do not deal with the environment stack.
Non-algebraic operators operate on the environment stack.
K.Subieta. Object-Oriented Query Languages and Views, slide 44
Sept. 2000
SBQL - Algebraic Operators
• Numerical and string comparisons, operators and functions:
=, <, +, *, concatenation, sqrt, sin, log, ...
• Boolean and, or, and not
• Aggregate arithmetic functions sum, max, min, avg
• Function count, function for removing duplicates, function exists
• Equality of complex query results (shallow, deep)
• Dereferencing operator (usually implicit)
• Coercion operators (changing representation and types);
• Operators for bags (union, intersections, difference, equality, ...)
• Operators for sets (union, intersections, difference, equality, containment)
• Operators for sequences (concatenation, i-th element, sorting,...)
• Cartesian product
• ... many other operators
Examples of use: 2+2
SAL > 1000
NAME = “Smith” and SAL > 1000
implicit dereferencing
K.Subieta. Object-Oriented Query Languages and Views, slide 45
Sept. 2000
SBQL - Declaration of an Auxiliary Name
Unary algebraic operator
Syntax:
q as n
n - auxiliary name,
q - a query returning a single-column table,
e.g. bag{x1, x2, ... xn}
Semantics:
x1
x2
Let q return a bag: ...
xk
n(x1 )
n(x2 )
...
Then q as n returns the bag of binders: n(xk )
Each value returned by q is
equipped with the name n.
Applications:
SQL, OQL correlation variables (“synonyms”):
EMP as e
Variables bound by quantifiers:
 EMP as e ( ... )
Cursors in “for each” statements:
for each EMP as x do ...
DEPT as d
Virtual names in SQL-like views.
K.Subieta. Object-Oriented Query Languages and Views, slide 46
Sept. 2000
SBQL - Non-algebraic Operators
If  is a non-algebraic operator, then in q1  q2 the evaluation order of q1 and q2 is
essential. Hence non-algebraic operators do not follow basic properties of algebraic
expressions. In contrast to relational/object algebras, our non-algebraic operators are
not indexed by (informal) meta-language expressions. No informal treatment of
names: each name in a query, including names of attributes, links, etc. precisely
follows the same scoping-binding discipline.
q1 where q2
 q1 ( q2 )
q1 order by q2 (ordering)
Syntax:
q1 . q2
q1 join q2
 q1 ( q2 )
dependent join
q1 closed by q2 (transitive closure)
(plus possibly other operators)
Semantics - the uniform homogeneous idea:
For each element r of the collection returned by q1 the query q2 is evaluated
with the environment stack augmented by nested( r ).
A partial result is a combination of r and the result returned by q2.
All partial results are merged into the final result.
All these operators are implemented in SBQL.
K.Subieta. Object-Oriented Query Languages and Views, slide 47
Sept. 2000
SBQL: Selection
q1 where q2
For each row r returned by q1 query q2 is evaluated
with the stack ES augmented by nested( r ).
The row r belongs to the final result, iff q2 returns TRUE for it.
EMP
i1
i5
ES
EMP(i1) EMP(i5) EMP(i9)
DEPT(i13) DEPT(i17)
i9
K.Subieta. Object-Oriented Query Languages and Views, slide 48
where
NAME( i2 ) SAL( i3 )
WORKS_IN( i4 )
EMP(i1) EMP(i5) EMP(i9)
DEPT(i13) DEPT(i17)
NAME( i6 ) SAL( i7 )
WORKS_IN( i8 )
EMP(i1) EMP(i5) EMP(i9)
DEPT(i13) DEPT(i17)
NAME( i10 ) SAL( i11 )
WORKS_IN( i12 )
EMP(i1) EMP(i5) EMP(i9)
DEPT(i13) DEPT(i17)
The
final
result
SAL > 1800
i3
1800
TRUE
i1
1800
TRUE
i5
1800
FALSE
(2500)
i7
(2000)
i11
(1500)
Sept. 2000
SBQL: Projection, navigation
q1 . q2
For each row r returned by q1 query q2 is evaluated
with the stack ES augmented by nested( r ).
The final result is the union of tables returned by q2 .
EMP
i1
i5
ES
EMP(i1) EMP(i5) EMP(i9)
DEPT(i13) DEPT(i17)
i9
K.Subieta. Object-Oriented Query Languages and Views, slide 49
.
NAME( i2 ) SAL( i3 )
WORKS_IN( i4 )
EMP(i1) EMP(i5) EMP(i9)
DEPT(i13) DEPT(i17)
NAME( i6 ) SAL( i7 )
WORKS_IN( i8 )
EMP(i1) EMP(i5) EMP(i9)
DEPT(i13) DEPT(i17)
NAME( i10 ) SAL( i11 )
WORKS_IN( i12 )
EMP(i1) EMP(i5) EMP(i9)
DEPT(i13) DEPT(i17)
SAL
The
final
result
i3
i3
i7
i7
i11
i11
Sept. 2000
SBQL: Navigational (Dependent) Join
q1 join q2
For each row r returned by q1 query q2 is evaluated
with the stack ES augmented by nested( r ).
A partial result is a concatenation of r with each row returned
by q2 . The final result is the union of partial results.
EMP
i1
i5
ES
EMP(i1) EMP(i5) EMP(i9)
DEPT(i13) DEPT(i17)
i9
K.Subieta. Object-Oriented Query Languages and Views, slide 50
join
NAME( i2 ) SAL( i3 )
WORKS_IN( i4 )
EMP(i1) EMP(i5) EMP(i9)
DEPT(i13) DEPT(i17)
NAME( i6 ) SAL( i7 )
WORKS_IN( i8 )
EMP(i1) EMP(i5) EMP(i9)
DEPT(i13) DEPT(i17)
NAME( i10 ) SAL( i11 )
WORKS_IN( i12 )
EMP(i1) EMP(i5) EMP(i9)
DEPT(i13) DEPT(i17)
SAL
i3
The
final
result
i1 i3
i7
i5 i7
i9 i11
i11
Sept. 2000
Examples in OQL and SBQL
OQL:
select e.NAME, d.DNAME, count(select x from d.EMPLOYS as x)
from EMP as e, e.WORKS_IN as d
where e.JOB = "manager"
SBQL:
((((EMP as e) join (((e.WORKS_IN). DEPT) as d))
where ((e.JOB) = "manager"))) .
(((e.NAME), (d.DNAME)), count((((d.EMPLOYS). EMP) as x) . x))
Syntactic differences, but striking similarity of the ideas.
K.Subieta. Object-Oriented Query Languages and Views, slide 51
Sept. 2000
Classes, inheritance and encapsulation on ES
If some operator opens a section on ES for an object,
then sections of its classes and super-classes are inserted onto ES, in proper order
(shown on the picture below).
PERSON
....
The query: EMP where ... X ...
The order of search during binding name X
Binders to attributes of the tested EMP object
EMP
....
DEPT
....
Binders to properties of the EMP class
Binders to properties of the PERSON class
Some invisible sections (for lexical scoping)
i1 EMP
i6 EMP
...
i1... EMP
...
...
...
...
...
i17 DEPT
i13 DEPT
...
...
......
Binders to properties of the current session
Binders to database objects, views, ...
...
...
...
The state of the object store
Binders to global procedures, variables,...
The state of the environment stack
Encapsulation:
some sections inserted on ES contain only binders to public properties.
Polymorphism - a a side effect of the above ES rules.
K.Subieta. Object-Oriented Query Languages and Views, slide 52
Sept. 2000
Object views
Expectations:
Customization, conceptualization, encapsulation. The user works with a part of a
database that is relevant to his/her area of interest in a way that is convenient for
his/her everyday processing routines and concepts. Views have the potential of
significant improvement of programmer’s productivity.
Security, privacy, and autonomy. Views restrict possible accesses only to a
relevant part of a database. In federated databases this restriction is required to
enable the autonomy of local databases
Interoperability, heterogeneity, schema integration, legacy applications. Views
enable integration of heterogeneous databases, allowing understanding and
processing foreign, legacy or remote databases according to a common, unified
schema.
Data independence, schema evolution. Views enable the user to change database
organization and schema without affecting already written applications.
The current state-of-the-art is immature, the above expectations are actually wishes.
K.Subieta. Object-Oriented Query Languages and Views, slide 53
Sept. 2000
Views are Stored Functional Procedures
Function
declaration:
function RichEmp
begin return EMP where SAL > 1800 end;
Function
invocation:
(RichEmp where JOB = "designer"). NAME
The name RichEmp is bound and the function is invoked. A new ES section contains
only a return address. Then, the query within the body of this function is executed.
It returns bag{i1,i5}, containing identifiers of Brown and Smith objects. This bag is a
function output.
The section for
view invocation
NAME(i2) JOB(..)
SAL(i3) WORKS_IN(i4)
The section for
EMP class
age(...)
EMP(i1) EMP(i5) EMP(i9)
DEPT(i13) DEPT(i17)
EMP(i1) EMP(i5) EMP(i9)
DEPT(i13) DEPT(i17)
EMP(i1) EMP(i5) EMP(i9)
DEPT(i13) DEPT(i17)
EMP(i1) EMP(i5) EMP(i9)
DEPT(i13) DEPT(i17)
The initial ES state
After invocation of
the RichEmp function
After where within
the RichEmp function
After return from
the function
K.Subieta. Object-Oriented Query Languages and Views, slide 54
.....
Sept. 2000
Two points of view on the database content
EMP
EMP
EMP
DEPT
DEPT
Objects
RichEmp
Stored
functional
procedure
EMP
EMP
EMP
a) The database seen
by the view definer
DEPT
DEPT
Objects
RichEmp
RichEmp
Imaginary
objects
b) The database seen
by a view user
The imaginary RichEmp objects adopt the entire semantics of EMP objects:
(RichEmp where age < 30).
WORKS_IN.DEPT.DNAME
RichEmp as r (r.age > 50
and "Rome"  r.WORKS_IN.DEPT.LOC)
K.Subieta. Object-Oriented Query Languages and Views, slide 55
Sept. 2000
Generalized functions
The stack-based semantics makes no restriction in calling functions from inside of
functions, i.e. to define a view through other views. Loqis makes it possible to define
arbitrary stored functional procedures, including procedures with parameters (being
arbitrary queries), with local objects, recursive, with side effects, and higher-order
(procedures with parameters being procedures).
For example, the WellPaid function has a parameter JobPar and a local variable a:
function WellPaid( JobPar )
begin
local a: real;
a := avg(EMP.SAL);
return EMP where JOB = JobPar and SAL > 2*a;
end;
The algorithm can be
complex.
Get names of departments of well paid clerks:
LowPaid( "clerk "). WORKS_IN. DEPT. DNAME
K.Subieta. Object-Oriented Query Languages and Views, slide 56
Sept. 2000
Query modification to process views
Macro-substitution of view invocation by the view body.
Function
declaration:
function RichEmp
begin return EMP where SAL > 1800 end;
Function
invocation:
(RichEmp where JOB = "designer"). NAME
After
macrosubstitution:
((EMP where SAL > 1800) where JOB = "designer"). NAME
Because of uniformity and universality of the stack-based semantics
query modification in SBA is very easy to implement.
Query modification makes it possible to use other optimization methods:
rewriting, access through indices.
K.Subieta. Object-Oriented Query Languages and Views, slide 57
Sept. 2000
Performance of views
Various approaches to object views do not consider this issue.
If this issue is unsolved, there will be no chance for object views.
What can be done?
Caching view results (materialized views): there is a side problem of keeping
materialized views consistent with stored data (incremental updating algorithms).
Query modification: macro-substitution of invocation of views by bodies of view
definitions. Applicable only to views defined by single queries. The method can be
difficult for more complex views.
Predicate-move-around: a predicate occurring in a query invoking a view modifies
the view definition. The method is not tested practically (probably).
K.Subieta. Object-Oriented Query Languages and Views, slide 58
Sept. 2000
Updatable views
A challenging theme.
Typical theoretical approaches (algebras, calculi, logic, functional...) are stateless,
hence updating operations are non-expressible. There are naive believes that
translations of view updates can be done in some auto-magical way, e.g. through
employing some constraints.
Oracle, with „do instead” triggers, is perhaps on the right way.
Updating through methods encapsulated in a class of virtual objects (no generic
updating operations). An idealistic view on object-orientedness and bad
understanding of encapsulation.
Updating through references returned by view invocations (undesirable side
effects, no universality).
Overriding generic updating operations by codes written by the view definer
(the mentioned Oracle way).
K.Subieta. Object-Oriented Query Languages and Views, slide 59
Sept. 2000
Conclusions
Object-oriented and object- relational query languages are a very important feature
of future DBMS.
Query languages have the potential to increase significantly the efficiency of
programming, as well as the maintainability and reusability of programs.
The current state-of-the-art of object query languages is immature. For us,
researchers, this is optimistic opportunity – we have a lot of work to do.
Commercial achievements, in particular OQL and SQL3, are a big progress, but
suffer from inconsistencies, underspecified semantics and eclecticism.
Theoretical proposals have many limitations, are mathematically and/or
conceptually inconsistent (object algebras), present wishful thinking.
Object views are still not developed to a satisfactory degree, both theoretically and
practically.
Optimization of object queries and views is not developed to a satisfactory degree
The stack-based approach to object query languages and views is a right
paradigm, which is able to make significant progress in the area.
K.Subieta. Object-Oriented Query Languages and Views, slide 60
Sept. 2000