Queries and Query Languages

 A query is a question which is posed to

a database.

 Formally, queries are considered to be

passive. That is, they do not modify the
state of the database.

 Often, active queries, or updates, are

also included in the general context of
database queries.

 In any case, most practical query

languages also provide facilities for
specifying updates.

20061031:slides5: 1 of 27
Query Languages to be Covered in the
Course

 Ideal Query Languages

 The Relational Algebra

 The relational algebra is based upon a set of

operations on relations. Each operation
takes one or more relations as its
arguments, and produces a new relation as
the result. Since the operations may be
composed with one another, they form an
algebra.

 A query is then a functional composition of

operations which yields the (derived)
relations which comprise the answer to the
query.

 From a logic point of view, the relational

algebra is a set of operations on the models
of the theory.

20061031:slides5: 2 of 27
 The Relational Calculus

 In the relational calculus, a query is specified

as a logical formula in a first-order logic
based upon the database schema. The set
of tuples for which the formula is true
constitutes the answer to the query.

 Equivalence result: Under very reasonable

assumptions of which operators and formulas are
allowed, the relational algebra and relational
calculus may be shown to be equivalent. Thus, it
is a matter of ease of use, rather than power,
which suggests that one should be chosen over
the other.

 A Practical Query Language

 SQL (Structured Query Language)

 SQL has become an industry standard.

 It incorporates aspects of both the relational

algebra and the relational calculus.

20061031:slides5: 3 of 27
 The Relational Algebra

We start with the most basic class: SPJ-

queries

 SPJ queries are constructed from three

basic operations,
 Select: horizontal restriction
 Project: vertical restriction
 Join: vertical reconstruction

20061031:slides5: 4 of 27
The Select Operation:

 The basic syntax is

<cond>(<rel_arg>)
Here:
  is the symbol used to denote a select
operation.
 <cond> is the condition of the selection.
 <rel_arg> is the relation to which the selection
is applied.

Example: Call the following instance r.

R
A B C
a1 b1 c1
a1 b2 c2
a1 b3 c3
a2 b4 c4

Then the answer to the query

σ A=a   r 
1
is
A B C
a1 b1 c1
a1 b2 c2
a1 b3 c3

20061031:slides5: 5 of 27
 The same notation may be used to represent the
query on a schema. For example,
σ  A=a   R
1

may be used to represent the query to be applied

to schema R.

 The most basic operation just allows equality

selection on domain values, but more general
operations may include arithmetic comparison.

Example: Call the following instance s.

S
A B C
1 b1 c1
3 b2 c2
3 b3 c3
4 b4 c4

The solution to the query (A3)(s) is

A B C
3 b2 c2
3 b3 c3
4 b4 c4

20061031:slides5: 6 of 27
 Disjunctive formulas are also possible. Construct
the answer to the following query.

σ  A= 1∨ A= 4  s 

 When working with selections, it is critical to

specify exactly the allowable selection
conditions.

20061031:slides5: 7 of 27
The Project Operation:

 Just as the select operation is a “horizontal”

restriction which retains selected rows of a
relation, so too is the projection operation a
“vertical” restriction, which retains selected
columns.

The general syntax is <cols>(<rel_arg>).

Example: Call the following instance r.

R
A B C
a1 b1 c1
a1 b2 c1
a2 b3 c3
a2 b4 c3

The solution to the query {A,C}(r), or just AC(r), is

A C
a1 c1
a2 c3

Notice that duplicate cells are omitted.

The associated schema query is AC(R).

20061031:slides5: 8 of 27
Join:
The join operation is much more complex than the
previous two. We begin with the simplest, yet most
widely used version.
Natural Join:
 It is easiest to explain with an example. Let r and
s be the instances associated with the following
database.

R S
A B B C
a1 b1 b1 c1
a2 b1 b1 c2
a2 b3 b3 c3
a3 b3 b4 c3

Then the natural join of r and s, denoted rs, is

T
A B C
a1 b1 c1
a1 b1 c2
a2 b1 c1
a2 b1 c2
a2 b3 c3
a3 b3 c3

20061031:slides5: 9 of 27
The idea is to match the tables on their common
attributes.
 A join is said to be lossless if the original
relations may be recovered from the appropriate
projections on the join. In this example, this
means that
AB(rs) = r and BC(rs) = s.

 The join of this example is not lossless.

 The natural join of the following two instances is
lossless.

R S
A B B C
a1 b1 b1 c1
a2 b1 b1 c2
a2 b3 b3 c3
a3 b3

● In general, for a join to be lossless, the

projections on the matching columns must
match.

20061031:slides5: 10 of 27
Other Forms of Join:
 General equality join: Even if attribute names do
not match, a join may be constructed by
specifying which columns to match. In this case,
the duplicate columns are often retained.

 Example: R {B,D}S

R S
A B D C
a1 b1 b1 c1
a2 b1 b1 c2
a2 b3 b3 c3
a3 b3

Result:

T
A B C D
a1 b1 c1 b1
a1 b1 c2 b1
a2 b1 c1 b1
a2 b1 c2 b1
a2 b3 c3 b3
a3 b3 c3 b3

20061031:slides5: 11 of 27
 Cartesian product: In this context, the join on an
empty set of columns (i.e., R S) yields the
Cartesian product of the two relations.

 Theta joins: More general matching operators

may be applied in situations in which they make
sense. These are called theta joins.

 Example: R {B D}S (Here the values for

attributes B and D must be amenable to an
ordering relation. The operator  is called the
theta operator.

20061031:slides5: 12 of 27
Join Conventions and Notation

The textbook introduces some relatively

nonstandard notation for joins. In these slides, the
more common notation, which is used in this
course, is presented. This notation should be used
on obligatory exercises, and on the examination.

Conventions and notation for the natural join:

 The “bowtie” symbol  without any subscript

always denotes the natural join. This means
that:

 the two relations are joined on their common

attributes, and

 the common attributes are not repeated.

Thus, the R[ABCD]  S[BDE] is a relation on
attributes ABCDE.

 Note that the natural join is a Cartesian product

iff the attribute sets of the two relations are
disjoint. For example R[ABCD]  S[EFG] is a
Cartesian product, with attributes ABCDEFG.

20061031:slides5: 13 of 27
Conventions and notation for other types of join:

 With subscripts, the bowtie symbol may only be

applied to relations which do not have any
attribute names in common. (Rename if
necessary.) In this case, all attributes of both
relations are present in the resulting relation.

 Equality joins are represented as in the following

example:
R[ABC] {B=D} S[DEF]

 A Cartesian product is indicated by a “” as

subscript:
R[ABC]  S[DEF]

 Other joins are indicated by the appropriate

operations in the subscript:
R[ABC] {BD} S[DEF]

20061031:slides5: 14 of 27
Other Relational Algebra Operators:
 In addition to the “SPJ” trio, it is often the case
that set-based operations are allowed in the
relational algebra. The most common operations
are the following:
 Union ()
 Set difference ()
 Intersection (), which can in fact be
constructed from union and difference.
Note that, for queries to make sense, these set
operations may only be performed on relations over
identical attribute sets, or at least over relations
with a bijective correspondence between the
domains of their attribute sets.

Format of queries:
 Officially, the correct format is to write queries in
functional composition format.
 In a less elegant representation intermediate
variables are introduced and queries are written
in an imperative program style.
 Examples will be given later.

20061031:slides5: 15 of 27
 The Relational Calculus

 The relational calculus uses logical formulas to

specify queries.
 This is natural, since the relational model is
closely based upon first-order predicate logic.
 There are two versions of the relational calculus,
domain calculus and tuple calculus.
 In each case, the formulas must be controlled
(limited to safe formulas) to avoid meaningless
queries.

20061031:slides5: 16 of 27
The Domain Calculus

 The domain calculus shows clearly the

connection between the relational model and
first-order predicate logic.
 Each relation symbol becomes a relation name in
a first-order logic.
 For each attribute A, there is a set of variables
whose values may range over the domain
assigned to that attribute. Such variables will be
represented with a superscript matching the
attribute name. (Example: xA, yA, for domain A.)

 Specific domain values may be represented with

constant symbols.
 There are no non-nullary function symbols.

20061031:slides5: 17 of 27
Example: Consider the problem of computing the
AC-projection of the join of the following two
relational instances (i.e., AC(RS)):

R S
A B B C
a1 b1 b1 c1
a2 b1 b1 c2
a2 b3 b3 c3
a3 b3 b4 c3

To make things work, we need to assign an order to

the attributes. We use the ordering indicated in the
picture. The domain calculus query to compute the
projection of the natural join is as follows.

{(xA,xC) | (xB)(R(xA,xB)S(xB,xC))}

20061031:slides5: 18 of 27
Equality and even arithmetic comparisons may also
be used in a limited fashion within the logic. Here
is an example of a selection query revisited:

S
A B C
1 b1 c1
3 b2 c2
3 b3 c3
4 b4 c4

The solution to the query (A3)(s) is

A B C
3 b2 c2
3 b3 c3
4 b4 c4

And this query is represented in the domain

calculus by

{(xA,xB,xC) | S(xA,xB,xC)  (xA 3)}

20061031:slides5: 19 of 27
 These queries are safe, in the sense that the
answer only involves values extracted from the
relations in the database. In formulating a
precise notion of safe queries, we must be
careful to ensure this condition. An example of
an unsafe query is:

{ xA | (xB)(R(xA,xB))}

 The formalization of safety is rather technical,

and will not be presented here. For a query
designed by hand, it is usually obvious whether
or not it is safe. Of course, a query processor
must be able to detect safety, or lack thereof.
Another example: Here is an informal example of
an unsafe query. Consider the database of all
people living in Sweden, and suppose that the
Last_Name field may contain any string of up to 64
characters. Then the following is an unsafe query
which is easily represented in the domain calculus.
“Give me the set of strings of length at most 64
which do not represent the last name of someone
living in Sweden.”

20061031:slides5: 20 of 27
The Tuple Calculus
 The tuple calculus differs from the domain
calculus in that the variables range over entire
tuples from relations, rather than over single
domain values.
 The idea is best illustrated by example: Suppose
that we have the following simple relational
database schema.

R S
A B B C

The composition AC(RS) is reconstructed via the

following formula in the tuple calculus:

{(x.A,y.C) | R(x)  S(y)  (x.B = y.B) }

 Here x and y are tuple variables, and their values
range over (binary in this case) tuples for the
associated relations.

 x.A, for example, denotes the A-indexed

component of tuple x.

 Note that x.A and y.C are free variables; an

existential quantifier is not used.
 As in the case of the domain calculus, a notion of
safe query may be formulated.

20061031:slides5: 21 of 27
Division – a Special Operator in the
Relational Algebra

Suppose that we have the following simple

relational database schema.

Employee Project
Emp_Num Name Emp_Num Proj_Num

Consider the query “Provide the names of those

employees who work on every project.”

This is easy in the tuple calculus.

{[Link] | Employee(x) 
(y)(z)(Project(y)  (Project(z) 
(y.Proj_Num = z.Proj_Num) 
(x.Emp_Num = z.Emp_Num)))}

20061031:slides5: 22 of 27
At first glance it seems impossible within the
relational algebra. However, there is an operation,
called division, which makes it possible.

Definition: Let R[A] and S[B] be relation schemata,

and assume that B  A. Then R  S, called R
divided by S, is the relation schema on attribute set
A \ B defined by the following formula.

R  S = A\B(R) \ A\B((A\B(R )  S) \ R)

This quantity is also called the quotient of R by S.

Despite its apparently indecipherable definition, it
has a relatively straightforward characterization.
Let r be an instance of R[A], and let s be an
instance of S[B]. Then

r  s = { t[A\B] |
( tS  s)( tR  r)(( tR[B] = tS)  (tR[A\B] = t))}

Here is the solution to the query:

{Name}((Employee  Project)  Proj_Num(Project))

● Such “double negation” constructs occur

frequently when writing queries with universal
quantification in SQL, as will be seen later.
C'est simple comme bonjour, n’est-ce pas?

20061031:slides5: 23 of 27
 Renaming – Another Useful Operator in
the Relational Algebra

Suppose that we again have the following simple

relational database schema.

Employee Project
Emp_Num Name Emp_Num Proj_Num

Consider the query “Provide the names of those

employees who work on exactly one project.”

This is easy in the tuple calculus.

{[Link] | Employee(x) 
(y)(Project(y)  (x.Emp_Num = y.Emp_Num)) 
(z)(w)((Project(z)  Project(w) 
(x.Emp_Num = z.Emp_Num) 
(x.Emp_Num = w.Emp_Num)) 
(z.Proj_Num = w.Proj_Num))}

20061031:slides5: 24 of 27
Again, at first glance it seems impossible with the
relational algebra. However, it is easily realized
with the aid of renaming. In renaming, a copy of a
relation schema is made, with the name of one or
more attributes changed.
Copy-with-Rename(
Project, Project',{(Proj_Num, Proj_Num')})
creates the following relation schema:

Project'
Emp_Num Proj_Num'

To solve the query, let us proceed step-by-step.

First, join Project with Project':

R1  Project  Project'.

R1
Emp_Num Proj_Num Proj_Num'

20061031:slides5: 25 of 27
Now, restrict this relation to those tuples with
different values for Proj_Num and Proj_Num':

R2   (Proj_Num  Proj_Num')(R1)

The following formula identifies the employees who

work on more than one project.

R3  Emp_Num(R2)

To identify the employees who work on no projects,

use

R4  Emp_Num(Employees)  Emp_Num(Project).

Then the difference below identifies the employees

who work on exactly one project.

R5  Emp_Num(Employees)  (R3  R4).

The query asks for names, however. Join the

above result with the Employee relation:

R6  R5  Employee.
Finally, project the names:

R7  Name(R6).

20061031:slides5: 26 of 27
As a single expression in the relational algebra, this
becomes:

Name(
(Emp_Num(Employee) 
Emp_Num(
 (Proj_Num  Proj_Num')(Project  Project'))

(Emp_Num(Employee)  Emp_Num(Project))
 Employee)

Exercise: Build a query in the relational algebra

which gives the names of those employees who
work on exactly two projects.

20061031:slides5: 27 of 27