Regular Expressions in Software

Testing

Tailored
from Book Chapter
(Introduction to Software Testing)
BY
Paul Ammann & Jeff Offutt
 Graphs – Circles and Arrows

• It is usually easier for humans to understand graphs by viewing

nodes as circles and edges as arrows

• Sometimes other forms are used to manipulate the graphs

– Standard data structures methods are appropriate for tools
– An algebraic representation (regular expression) allows certain useful
mathematical operations to be performed

Introduction to Software Testing (Ch 2),

© Ammann & Offutt 2
www.introsoftwaretesting.com
 Representing Graphs Algebraically
• Assign a unique label to each edge
– Could represent semantic information, such as from an activity graph or
an FSM
– Could be arbitrary – we use lower case letters as an abstraction
• Operators:
– Concatenation (multiplicative) : if edge a is followed by edge b, they are
concatenated as “a * b”, or more usually, “ab”
– Selection (additive) : if either edge a or b can be taken, they are summed
as “a + b”
• Path Product : A sequence of edges “multiplied” together is
called a product of edges, or a path product
• Path Expression : Path products and possibly ‘+’ operators
– A = ab + cd
• Loops : represented as exponents

Introduction to Software Testing (Ch 2),

© Ammann & Offutt 3
www.introsoftwaretesting.com
 Examples
2 5
b d f h
a j
0 1 4 7 8
c 3 e g i
6

Double-diamond graph
Path Expression = abdfhj + abdgij + acegij + acefhj

a b e g
0 1 2 4 5
c d
f
3

Path Products : A = abeg, B = (cd)*, C = cf

Path Expression = ab (cd)* (e + cf) g

4
 Let’s Look at the Math
• Path Products
– Commutative : Path product is not ( AB != BA )
– Associative : Path product is ( A(BC) = (AB)C = ABC )
• Path Summation
– Commutative : Summation is ( A + B = B + A )
– Associative : ( (A+B)+C = A+(B+C) = A + B + C )

• These are close to the “usual” arithmetic addition and

multiplication, so most of the standard algebraic laws can be
used
– In normal math, multiplication is commutative

• (Don’t worry … we’ll get back to testing soon …)

Introduction to Software Testing (Ch 2),

© Ammann & Offutt 5
www.introsoftwaretesting.com
 Algebraic Laws on Graph Expressions
• Distributive : A ( B + C ) = AB + AC
• Distributive : ( B + C ) D = BD + CD
• Absorption rule : A + A = A
• Shortcut notation for loops
– At least one iteration : A+ = AA*
– Bounds on iteration : A3 = A0 + A1 + A2 + A3
• More generally : An = A0 + A1 + … + An
• Absorbing exponents
– An + Am = A max(n,m)
– AnAm = An+m
– An A* = A*An = A*
– A n A+ = A +A n = A +
– A * A+ = A +A * = A +

© Ammann & Offutt 6

 Identity Operators
• Multiplicative : λ ( an empty path )
• Additive : Φ ( a null path – set of paths that has no paths )
• Multiplicative identity laws
1
– λ+λ =λ a
– λA = Aλ = A 0
n n * + b c
– λ =λ =λ =λ =λ 2 3
– λ+ + λ = λ * = λ
pe = bc + aΦ = bc
• Additive identity laws
– A+Φ=Φ+A=A a
– A Φ = ΦA = Φ
* 2
b
–Φ =λ+Φ+Φ +…=λ 0 1 2
λ
pe = (a + λ)b = ab + λb = ab + b

Introduction to Software Testing (Ch 2),

© Ammann & Offutt 7
www.introsoftwaretesting.com
 Graphs to Path Expressions
• General algorithms for reducing finite state machines to regular
expressions are widely known
• This lecture presents a four step, iterative, special case process
• The four steps are illustrated on a simple graph

b e
a d f h i
0 1 2 3 4 5 6
c
g

Introduction to Software Testing (Ch 2),

© Ammann & Offutt 8
www.introsoftwaretesting.com
 Step 1 : Sequential Edges
• Combine all sequential edges
• Multiply the edge labels
• Precisely: For any node that has only one incoming and one
outgoing edge, eliminate the node, combine the two edges, and
multiply their path expressions
• Example: Combine edges h and i
b e
a d f h i
0 1 2 3 4 5 6
c
g
b e
a d f hi
0 1 2 3 4 6
c
g

9
 Step 2 : Parallel Edges
• Combine all parallel edges
• Add the edge labels
• Precisely: For any pair of edges with the same source and target
nodes, combine the edges and add their path expressions
• Example : Combine edges b and c

b e
a d f hi
0 1 2 3 4 6
c
g

e
a b+c d f hi
0 1 2 3 4 6

g
10
 Step 3 : Self-Loops
• Combine all self-loops (loops from a node to itself)
• Add a new “dummy” node
• An incoming edge with exponent
• Merge the three resulting sequential nodes with multiplication
• Precisely: For any node n1 with a self-loop X, incoming edge A
and outgoing edge B, remove X, add a new node n1’ and edge
with label X*, then combine into one edge AX*B
e
a b+c d f hi
0 1 2 3 4 6

a b+c d e* f hi
0 1 2 3 de*f 3’ 4 6

Introduction to Software Testing (Ch 2),

© Ammann & Offutt 11
www.introsoftwaretesting.com
 Step 4 : Remove Tester-Chosen Nodes
• Use for “other” special cases, when steps 1-3 do not apply
• Choose a node that is not initial or final
• Replace it by inserting edges from all predecessors to all
successors
• Multiply path expressions from all incoming with all outgoing
edges
AC
1 4 1 4
A C
AD
3 3
B D BC

2 5 2 5
BD

Introduction to Software Testing (Ch 2),

© Ammann & Offutt 12
www.introsoftwaretesting.com
 Step 4 : Eliminate Node 2
• Remove node 2
• Edges (1, 2) and (2, 4) become one edge
• Edges (4, 2) and (2, 4) become a self-loop

a b+c hi
0 1 2 de*f 4 6

a b +bde*f
c + cde*f hi
0 1 2 de*f 4 6

g
gde*f

Introduction to Software Testing (Ch 2),

© Ammann & Offutt 13
www.introsoftwaretesting.com
 Repeat Steps 1 Through 4
Continue until only one edge is left …

a bde*f + cde*f hi
0 1 4 6
gde*f

abde*f + acde*f hi
0 4 6
gde*f

abde*f + acde*f (gde*f)* hi

0 4 4’ 6

abde*f (gde*f)* hi + acde*f (gde*f)* hi

0 6

Introduction to Software Testing (Ch 2),

© Ammann & Offutt 14
www.introsoftwaretesting.com
 Applications of Path Expressions

1. Deriving test inputs

2. Counting paths in a flow graph

3. Minimum number of paths to satisfy All Edges

4. Complementary operations analysis

Introduction to Software Testing (Ch 2),

© Ammann & Offutt 15
www.introsoftwaretesting.com
 1. Deriving Test Inputs
• Very simple … find a graph and cover it
• Cover regular expressions by covering each separate path
product
• This is a form of the specified path coverage criterion
• Loops are represented by exponents – if an unbounded exponent
appears, replace with a constant value based on domain
knowledge of the program

• Test requirements for running example:

– Final path expression: abde*f (gde*f)* hi + acde*f (gde*f)* hi
– Test Requirements
abde5f (gde5f)5 hi
acde5f (gde5f)5 hi

Introduction to Software Testing (Ch 2),

© Ammann & Offutt 16
www.introsoftwaretesting.com
 2. Counting Paths in a Flow Graph
• It is sometimes useful to know the number of paths in a graph

• The path expressions allow this computation with

straightforward arithmetic

• Cycles mean we have an infinite number of paths, so we have to

make assumptions to approximate

• Put a reasonable bound on the number of iterations by replacing

the ‘*’ with an integer value
– The bound may be a true maximum number of iterations
– The bound may represent a tester’s assumption that executing the loop ‘N
times’ is enough

Introduction to Software Testing (Ch 2),

© Ammann & Offutt 17
www.introsoftwaretesting.com
 2. Counting Paths in a Flow Graph
Example
Edge weights
1 b
1 a e
0 1 2 1 3
1 c
(0 – 2)
d 1 Bounds the loop
– max 2, min 0
pe = a (b + c) (d (b + c) )2 e iterations
= 1 * (1 + 1) * (1 * (1 + 1) )2 * 1
= 1 * 2 * 22 * 1
= 2 * ( Σ2i=0 2i ) * 1
= 2 * ( 20 + 21 + 22 ) * 1
=2*(1+2+4)*1
=2*7*1
= 14
Introduction to Software Testing (Ch 2),
© Ammann & Offutt 18
www.introsoftwaretesting.com
 2. Counting Costs of Executing Paths
Edge Weights
• It sometimes helps to have the number of paths include a
measure of cost for executing each path
– Number of paths in a function call
– Expensive operations
– An operation our intuition tells the tester to avoid or encourage
• Edge weights default to 1 – otherwise marked by tester
• Cycle weight: the maximum number of iterations allowed
– Only mark one edge per cycle !
• Compute path expression and substitute weights for edges
– Path expression : A+B becomes weight (A) + weight (B)
– Path product : AB becomes weight (A) * weight (B)
– Loop is sum of the weight of all iterations
• The result of this computation is the estimated total cost of
executing all paths
Introduction to Software Testing (Ch 2),
© Ammann & Offutt 19
www.introsoftwaretesting.com
3. Minimum Number of Paths to Satisfy All Edges
• If we wish to satisfy All Edges, how many paths are needed?
• How many tests will be needed?
• Similar to computing the maximum number of paths
– Different computation

• Specifically:
– Path expression : A+B becomes weight (A) + weight (B)
– Path product : AB becomes max (weight (A), weight (B))
– Loop : An is either 1 or weight (A)
– Judgment of tester – if all paths in the loop can be taken in one test, use 1,
else use weight(A)

Introduction to Software Testing (Ch 2),

© Ammann & Offutt 20
www.introsoftwaretesting.com
 3. Min Number of Paths to Satisfy All Edges
Example
1 b
1 a e
0 1 2 1 3
1 c
(0 – 2)
d 1
pe = a (b + c) (d (b + c) )2 e

Conservatively assume the same edge

from 1 to 2 must be taken every iteration
through the loop –use the edge weight …

= 1 * 2 * (1 * (2))2 * 1

= 1 * 2 * (1 * 2) * 1

= max (1, 2, 1, 2, 1)

=2
Introduction to Software Testing (Ch 2),
© Ammann & Offutt 21
www.introsoftwaretesting.com
 4. Complementary Operations Analysis
• A method for finding potential anomalies
– Def-use anomalies (use before a def)
– FileADT example (closing before writing)
• A pair of operations are complementary if their behaviors negate
each other, or one must be done before the other
– push & pop
– enqueue & dequeue
– getting memory & disposing of memory
– open & close
• Edge weights are replaced with one of three labels
– C – Creator operation
– D – Destructor operation
– 1 – Neither

Introduction to Software Testing (Ch 2),

© Ammann & Offutt 22
www.introsoftwaretesting.com
 4. Complementary Operations Analysis
Arithmetic Operations
• The addition and multiplication operators are replaced with two
tables

* C D 1 + C D 1

C C2 1 C C C C+D C+1

D DC D2 D D D+C D D+1

1 C D 1 1 C+1 D+1 1

• Not the same as usual algebra

– C*D = 1, C+C = C, D+D = D
– Multiplication is not commutative because the operations are not
equivalent … a destructor cancels a creator, but not vice versa

Introduction to Software Testing (Ch 2),

* © Ammann & Offutt 23
www.introsoftwaretesting.com
 4. Complementary Operations Analysis Example

C C
C C 1 1
0 1 2 3 4 5 6
1 D
(n)
D

pe = C (C+1) C (C+D) 1 (D (C+D) 1)n 1

reduced pe = (CCCC + CCCD + CCC + CCD) (DC + DD)n

C*D = 1, canceling
out

final pe = (CCCC + CC + CCC + C) (DC + DD)n

Introduction to Software Testing (Ch 2),

© Ammann & Offutt 24
www.introsoftwaretesting.com
4. Using Complementary Operations Analysis
final pe = (CCCC + CC + CCC + C) (DC + DD)n

• Ask questions:
– Can we have more destructors than creators?

• CCCD (DD)n, n > 1

• CCCD (DD)n, n > 0
• CCC (DDDCDD)
– Can we have more creators than destructors?

• CCCC
• CCD (DC)n, forall n

• Each “yes” response represents a specification for a test that

might cause anomalous behavior

Introduction to Software Testing (Ch 2),

© Ammann & Offutt 25
www.introsoftwaretesting.com
 Summary of Path Expressions

• Having an algebraic representation of a graph can be very

useful

• Most techniques involve some human input and subjectivity

• The techniques have not been sufficiently quantified

• We know of no commercial tools to support this analysis

26