56 International Journal of Applied Metaheuristic Computing, 3(2), 56-74, April-June 2012

Copyright 2012, IGI Global. Copying or distributing in print or electronic forms without written permission of IGI Global is prohibited. Copyright 2012, IGI Global. Copying or distributing in print or electronic forms without written permission of IGI Global is prohibited.
Keywords: Control Flow Graph (CFG), Intelligent Water Drop (IWD), Software Testing, Test Data
Generation, Test Path Discovery
INTRODUCTION
In software engineering area, software testing is
a crucial phase (Kewen, Zilu, & Wenying, 2009)
to assess the quality of the software. Software
testing contains almost 50% of total cost of the
software development (Edvardsson, 1999). Aim
of the software testing is to uncover errors and
faults present in the program, so that customer
requirement can be properly fulfilled. Testing
phase includes in the review of specification,
analysis, design, and implementation part of
the Software Development Life Cycle (SDLC).
Manual generation of test data for testing the
program, results in low reliability and high
cost (Li & Lam, 2005; Pressman, 2010). Due
to the lack of cost and reliability, automation
of testing process is necessary, so that the cost
of testing can be reduced.
Software engineering is a knowledge-
intensive activity, which requires access and
manipulation of large quantity of information
about the project domain (Naaz Raza, 2009). By
managing these facts manually, cost becomes
very high. Artificial Intelligence (AI) based
techniques can help in removing this situation.
Test Data Generation Based
on Test Path Discovery Using
Intelligent Water Drop
Praveen Ranjan Srivastava, Birla Institute of Technology and Science, Pilani, India
Amitkumar Patel, Birla Institute of Technology and Science, Pilani, India
Kunal Patel, Birla Institute of Technology and Science, Pilani, India
Prateek Vijaywargiya, Birla Institute of Technology and Science, Pilani, India
ABSTRACT
Automatic test data generation is required to generate test cases dynamically for a specifc software program.
Manual generation of test data is too tedious and a time consuming task. This paper proposes a technique
using Intelligent Water Drop (IWD) for automatic generation of test data. Correctly generated test data helps
in reducing the effort while testing the software. This paper discusses different algorithms based on IWD to
generate test data and path coverage over Control Flow Graph. Test data is generated keeping in mind all of
the programming constraints like if, while, do while, etc., available in the program.
DOI: 10.4018/jamc.2012040105
International Journal of Applied Metaheuristic Computing, 3(2), 56-74, April-June 2012 57
Copyright 2012, IGI Global. Copying or distributing in print or electronic forms without written permission of IGI Global is prohibited. Copyright 2012, IGI Global. Copying or distributing in print or electronic forms without written permission of IGI Global is prohibited.
AI based technique helps in solving the prob-
lem by using fast and proper judgments rather
than using step by step deduction (Naaz Raza,
2009). From the last decade various techniques
like genetic algorithm, ant colony optimization
and other metaheuristic techniques have been
introduced for generating test path and test data
(Korel, 1990; Michael et al., 1997; Michael &
McGraw, 1998; McMinn, 2004; Kewen et al.,
2009; Srivastava, Baby, & Raghurama, 2009).
The main issues of various metaheuristic tech-
niques (McMinn, 2004) are optimal test data
generation and complete software coverage.
Another new optimization technique, Intelligent
Water Drop (IWD) (Shah-Hosseini, 2009), an
algorithm which is based on swarm optimiza-
tion techniques, can be used for generation of
test data. IWD algorithm works similar to the
natural water drop of a river bed.
This paper applies IWD algorithm to gen-
erate test data and also discover existing paths
in the program. This paper is structured as fol-
lows; first, we describe the background work
of software testing. Then, we present the basic
IWD optimization approach. The next section
applies IWD approach to generate test path and
sequence generation. We then describe the case
study of the suggested approach for test data
generation, while the next section discusses the
analysis part. Finally we conclude the paper
along with future scope of the applied approach.
BACKGROUND
Various techniques (Korel, 1990; Pedrycz &
Peters, 1998; Briand, 2002; McMinn, 2004;
Harman, 2007; Kewen et al., 2009; Srivastava
et al., 2009; Srivastava & Baby, 2010) have
been proposed for automated testing to reduce
efforts to a remarkable extent. Swarm optimi-
zation techniques are widely used for test data
generation. Ant Colony Optimization (ACO)
is one of them (Li & Lam, 2005; Kewen et al.,
2009; Srivastava et al., 2009). Automatic test
data generation based on ACO (Kewen et al.,
2009) has introduced a model to generate test
data using the branch function technique. It has
solved the problem of local optimization, but
this approach is applicable only for numeric
data types and also the model is not suitable
for object oriented programs and other types
of input behavior. Software test data generation
using ACO (Li & Lam, 2005) has also proposed
a solution for state-based software testing but
complete coverage is not possible by the pro-
posed approach.
Genetic algorithm has been applied for
dynamic test data generation (Michael et al.,
1997) and the results are pretty impressive over
random or exhaustive test data generation, but
this model was not applicable for boolean and
string type of variables.
Another approach is suggested for genera-
tion of test data using genetic algorithm and
hamming distance concept (Srivastava et al.,
2010) but optimal test data are uncertain, and
stuck in the local optima and explore more
repetitive paths. An approach for software
testing has been recommended (Srivastava et
al., 2009) in which ACO procedure is used to
discover effective paths which are promising
for full path coverage.
Genetic algorithm and tabu search have
been applied on control-dependence graph
(Rathore et al., 2011). This approach uses only
the number of common predicates with the
target node for finding objective value, but few
important factors such as distance from initial
and final node, criticality, etc., that it fails to
take into account.
Many papers (Korel, 1990; McMinn,
2004; Srivastava et al., 2009) have suggested
techniques for full path coverage and other
have suggested approaches for optimal test
data generation (Li & Lam, 2005; Srivastava et
al., 2008; Kewen et al., 2009; Srivastava et al.,
2010), but none of them have suggested com-
bined approaches for tackle both the problems
together though they are very closely related.
This paper discusses the newly emerging ap-
proach for combined solutions to both of these
problems together.
IWD algorithm (Shah-Hosseini, 2007;
Shah-Hosseini, 2009) is a swarm-based opti-
mization algorithm, simulated from observing
58 International Journal of Applied Metaheuristic Computing, 3(2), 56-74, April-June 2012
Copyright 2012, IGI Global. Copying or distributing in print or electronic forms without written permission of IGI Global is prohibited. Copyright 2012, IGI Global. Copying or distributing in print or electronic forms without written permission of IGI Global is prohibited.
natural water drops in river. IWD has been
applied to various problems like Travelling
Salesman Problem (TSP), N-queen puzzle,
Multidimensional Knapsack Problem (MKP),
etc. These results have proved the significance of
IWD algorithm over other swarm optimization
algorithms. Another solution for TSP using IWD
algorithm (Shah-Hosseini, 2007) is introduced
where proposed algorithm converges very fast
to the optimum solution.
The improved IWD algorithm (Duan, Liu,
& Lei, 2008) has been applied to solve the air
robot path planning in dynamic environments
and results are quite impressive over genetic
algorithm and ACO algorithm.
Since IWD has not yet been applied to the
area of software testing and the effective results
have been produced for various problems, this
paper tries to derive a solution model for soft-
ware testing using IWD in the hope that expected
results will be more significant than the current
solutions available for test data generation.
Before moving to the proposed solution of
IWD, general introduction is provided which
describes its strategy along with available
metrics in it.
INTELLIGENT WATER DROP
IWD algorithm is a new swarm-based optimi-
zation algorithm inspired from natural rivers.
In a natural river, water drops move towards
centre of the earth, due to some gravitational
force acting on it. Due to this the water drop
follows the straight and the shortest path to
its destination (Shah-Hosseini, 2009). Picto-
rial representation of basic IWD is shown in
Figure 1. In ideal conditions it is observed that
the optimal path will be obtained. Water drop
flowing in the river has some velocity which
is affected by another actor, i.e., soil.
Some changes that occurred while transi-
tion of water drop from one point to another
point are:
1. Velocity of water drop is increased.
2. Soil content in the water drop is also
increased.
3. Amount of soil in the riverbed from source
to destination get decreased.
Water drop in the river picks up some soil
in it when its velocity gets high and it releases
the soil content when its velocity is less (Shah-
Hosseini, 2007)
Some of the prominent properties of the
natural water drop are taken, based on which
IWD is suggested. IWD has the two following
important properties,
Figure 1. Pictorial representation of IWD
International Journal of Applied Metaheuristic Computing, 3(2), 56-74, April-June 2012 59
Copyright 2012, IGI Global. Copying or distributing in print or electronic forms without written permission of IGI Global is prohibited. Copyright 2012, IGI Global. Copying or distributing in print or electronic forms without written permission of IGI Global is prohibited.
1. The amount of soil the water drop carries,
which is represented by Soil (IWD) (or
soil
IWD
).
2. The velocity of water drop with which it is
moving now, denoted by Velocity (IWD)
(or vel
IWD
).
Value of both the properties may change
during the transition. Environment contains
lots of paths from source to destination (Duan
et al., 2008) which may be known or unknown.
When the destination is known, IWD follows
the best path to reach the destination (best is in
terms of cost and any other desired measure).
When destination is unknown it finds the op-
timal destination.
From the current location to the next loca-
tion Velocity (IWD) is increased by an amount,
which is nonlinearly proportional to the inverse
of the amount of soil between the two locations,
referred to as the change in velocity. The Soil
(IWD), is also increased by extracting some soil
of the path between two locations. The amount
of soil added to the IWD is inversely (and
nonlinearly) proportional to the time needed
for the IWD to pass from its current location
to next location. IWD chooses the path with
less soil content.
In the proposed approach, IWD is applied
over the Control Flow Graph (CFG) to obtain
the number of paths available in the program.
The CFG depicts the logical control flow of
the program (Pressman, 2010). All linearly
independent paths could be obtained by CFG.
Independent path is the path in the program that
determines at least one new set of processing
statement. In other words it introduces at least
one new edge in the graph. Number of available
paths can be obtained by finding the cyclomatic
complexity of the graph (Pressman, 2010).
PROPOSED SOLUTION
Software testing is a mandatory phase of soft-
ware life cycle model but it increases the time
duration as well as cost of the development. It
should avoid the exhaustive search for generat-
ing the test cases. This paper tries to implement
the solution model which is useful for generating
test data for program dynamically. The proposed
algorithm using IWD is shown in Figure 2.
The proposed model for test data generation
consists of two major phases. In the first phase
Lex and Yacc (Brown, Levine, & Mason, 1995)
parsing tool is used to generate output file which
contains the details about control flow graph.
It also generates the predicate node along with
its existing conditions. Second phase takes
output file of the first phase as an input which
contains the control flow graph.
Second phase works in three steps.
STEP 1: Find out cyclomatic complexity (CC)
to each node of CFG.
STEP 2: Extract all possible paths for traversing
the CFG from start node to end node using
proposed IWD algorithm.
STEP 3: Generate test data for all independent
paths.
For processing the noted steps, three al-
gorithms, namely, Algorithm 1, Algorithm 2
and Algorithm 3, are suggested. Algorithm 1
describes the working of Step-1 which assigns
cyclomatic complexity to each node of the CFG.
Cyclomatic complexity of a node is ob-
tained from Algorithm 1 which shows the
number of paths available from the current node
to reach the end node of CFG. This information
is fed into Algorithm 2, which extracts all the
independent paths of the CFG using IWD al-
gorithm.
From the algorithm, all available indepen-
dent paths are obtained. Now, in Algorithm 3,
test data is generated for each of the independent
path.
In Algorithm 3, all individual paths are
collected and for each particular path, test data
is generated. Evaluation of the proposed algo-
rithm is discussed below. Depth analysis of
proposed approach, few case studies have been
wisely chosen such that comparison studies can
be done properly. For experimental study a tool
has been developed for which snapshots are
available in the Appendix (Figures 9 through
11).
60 International Journal of Applied Metaheuristic Computing, 3(2), 56-74, April-June 2012
Copyright 2012, IGI Global. Copying or distributing in print or electronic forms without written permission of IGI Global is prohibited. Copyright 2012, IGI Global. Copying or distributing in print or electronic forms without written permission of IGI Global is prohibited.
EXPERIMENTAL STUDY
This section shows the result of the execution of
algorithm, in order to determine the effective-
ness and feasibility of the algorithm. Algorithms
demonstrate that it is a reasonably accurate
technique in terms of the paths covered using
IWD and test data generation, as compared to
the exhaustive methods. The proposed IWD
approach is implemented using Java program-
ming environment on an Intel Pentium 4 system
running Windows XP Service Pack-2. In experi-
mental study, IWD is applied on two programs,
one is triangle classifier problem (Classical
Figure 2. Architecture of IWD
Algorithm 1. Assign CC to each node of CFG
Initialization:
Global array visited [N]
/* empty array of size N, where N = number of nodes in the graph */
Input: root node of Graph(N,E)
/* Graph(N,E) = Graph containing N nodes and E edges */
Output: cyclomatic_complexity of each node
Step 1. If node is already visited OR node is an end node then return 1;
Step 2. Add node to visited list;
Step 3. Loop for all branches
Step 3.1 Increment result by cyclomatic complexity of noderesult += cyclomatic_
complexity(node);
/* cyclomatic complexity is calculated for the given node */
Goto Step-1
Step 3.2 End loop
Step 4. return result;
International Journal of Applied Metaheuristic Computing, 3(2), 56-74, April-June 2012 61
Copyright 2012, IGI Global. Copying or distributing in print or electronic forms without written permission of IGI Global is prohibited. Copyright 2012, IGI Global. Copying or distributing in print or electronic forms without written permission of IGI Global is prohibited.
continued on the following page
Algorithm 2. Extract all possible paths using IWD
Initialization:
Max_Iteration - Maximum time Iteration to be performed
Init_Soil - Initial soil on the path
soil(i,j) - Available amount of soil between node (i) and node (j)
Visited_Path - Contains information about node that has been visited
paths(i) - Number of path from node (i) yet to be explored
p(i,j) - Probability of choosing path between node (i) and node (j)
Input: Graph(N,E), cyclomatic_complexity of each node
Output: Path_List = Contains all extracted paths (initially empty list)
Step 1. Initialize all static parameters (from Step-1.1 to Step-1.8)
Step 1.1 Graph(N,E) having total N nodes
Step 1.2 Max_Iteration = CC which we have found using Algorithm 1
Step 1.3 Velocity updating parameters
a
v
= 1, b
v
= 0.01, c
v
= 1 and = 1
Here, the a
v
, b
v
, c
v
and are user-selected positive parameters.
Step 1.4 Soil updating parameters
a
s
= 1, b
s
= 0.01 and c
s
= 1
Here, the a
s
, b
s
and c
s
are user-selected positive parameters.
Step 1.5 Initial soil on each edge of graph
Init_Soil = 10000, (It is assigned to each edge, i.e., soil(i,j) =
Init_Soil)
Step 1.6 For all IWD, Soil(IWD) = 0
Step 1.7 Initial velocity of each IWD
Init_Vel = 200(It is assigned to each drop)
Step 1.8 Visited_Path = empty list
Step 2. Put IWD on root node of the graph(CFG)
Step 3. Calculate probability for choosing next node (j) from available path
of node (i).
Step 3.1 Probability can be find using formulae

p i j
paths i
paths j
soil i j
Init Soil
soil i k
In
,
,
_

,
( )
=
( )
( )
+
( )
( )

iit Soil
No of k
_
.
where, paths (i) = number of path from node (i) yet to be explored (CC(i)),
soil(i,k) = Sum of the soil of every path i to k, ik. Probability formulae
is used for finding probability of a path when there are two nodes are avail-
able for moving forward from the current node (i). This function can also
tackle the blocked path situation. Along with that for handling the situation
of paths having same CC, the concept of soil has also been introduced in the
fitness function which is likely to be varied for the different paths
Step 3.2 Using the formulae as mentioned in Step-3.1, find probability for all
outgoing paths from the current node (i)
Step 4. Choose next path which is having greatest probability because it is
the optimal path where many other paths are yet to be explored, e.g.,
p(1,2) > p(1,3), then choose path(1,2) and add it to the visited path
list.
Step 5. Update the Velocity(IWD) (denoted by vel
IWD
) moving from node (i) to
node (j).

vel t vel t
a
b c soil i j
IWD IWD v
v v
+
( )
=
( )
+
+
( )

1
2

( , )

where, vel
IWD
(t+1) is the updated velocity of IWD
Step 6. Update time parameter for IWD.

time i j vel t
HUD j
vel
IWD
IWD
, ; ( )
( )
+
( )
= 1
62 International Journal of Applied Metaheuristic Computing, 3(2), 56-74, April-June 2012
Copyright 2012, IGI Global. Copying or distributing in print or electronic forms without written permission of IGI Global is prohibited. Copyright 2012, IGI Global. Copying or distributing in print or electronic forms without written permission of IGI Global is prohibited.
problem used in many research work) (Kewen
et al., 2009; Edvardsson, 1999) and another is
maximum of three inputs.
In the study-1, IWD is applied on the
triangle program (Program 1). This program
takes 3 sides of triangle as input, and classifies
the triangle as equilateral, isosceles, scalene
and not a triangle. Program-1 shows triangle
problem and the corresponding CFG which
contains the nodes and flows between them is
shown in Figure 3.
In the CFG, individual cyclomatic com-
plexity for each node is assigned by the Algo-
rithm 1 as given in Table 1.
As explained in Algorithm 1, number of
branches is calculated recursively. It means that
for finding the number of branches for Node-1,
HUD (j) = CC of CFG CC at node (j)
where, a local heuristic function HUD(j) has been defined for a given
problem to measure the undesirability of an IWD to move from one
location to the next.
Step 7. For the IWD, moving on path node (i) to node (j), it computes the
change of soil.

soil i j
a
b c time i j vel t
s
s s
IWD
,
( , ; ( ) )
( )
=
+ +
( )
2
1
where, soil(i,j) which IWD loads from path while traversing through
that path
Step 8. Update the Soil(IWD) (denoted by soil
IWD
) by some amount which it has
loaded from path soil
IWD
= soil
IWD
+ soil(i,j)
Step 9. Update the soil of path between node (i) and node (j). soil(i,j) =
soil(i,j) - soil(i,j)
Step 10. Repeat Step-3 to Step-9 until it encounters the end node or already
visited node.
Step 11. Store the whole path as one of the final independent path in Path_
List and decrease CC by 1
Step 12. Repeat Step-2 to Step-11 until, CC (root node) != 0
Algorithm 2. Continued
Algorithm 3. Generate test data for all independent paths
Step 1. Collect all individual paths from Path_List (output of Algorithm 2).
Step 2. Select the appropriate range of variables which reduce the total time
by increasing the probability of satisfying the condition used in each path
with less number of iterations which is explained pictorially in more detail,
in the Analysis section.
Step 3. Initialize Conditions used in path to empty, i.e., Conditions = empty
Step 4. Repeat the following steps for each node in the path
Step 4.1 If node is predicate node than concatenate it to conditions.
Step 4.2 Else Continue
Step 5. Generate random data for only the variables present in the conditions,
which optimize the total time and iteration by avoiding the generation of
unused variable for the given path.
Step 6. Repeat Step-4 and Step-5 for all paths
Step 7. Fire generated data for the given condition.
Step 7.1 If condition evaluated to be true then write the generated values as
the test data for the given path.
Step 7.2 Else repeat Step-5 to generate new value for given condition.
International Journal of Applied Metaheuristic Computing, 3(2), 56-74, April-June 2012 63
Copyright 2012, IGI Global. Copying or distributing in print or electronic forms without written permission of IGI Global is prohibited. Copyright 2012, IGI Global. Copying or distributing in print or electronic forms without written permission of IGI Global is prohibited.
the branch value is calculated first for each
child node.
For example, the branch value for Node-8 is
4. Now for calculating this value, branch values
for the nodes 9, 10, 11, 12, 13, 14 and 15 needs
to be calculated. Then sum up only the branch
values of immediate children of Node-8, which
gives the branch value for Node-8. Immediate
children for Node-8 are Node-9 and Node-12.
Branch values for Node-9 and Node-12 are
2 and 2, respectively. The branch value for
Node-8 is 4 (2+2).
Algorithm 2 finds out the entire individual
paths available in the program. Experimental
result for the exploration of Path-1 is given in
Table 1. Cyclomatic complexity of each node for the triangle problem
Node Branches
1 10
2 9
3 8
4 7
5 6
6 5
7 1
8 4
9 2
10
11 1
12 2
13 1
14 1
15 0
Program 1. Triangle Problem
int TriType(int i, int j, int k)
{
1. if (i <= 0 ||
2. j <= 0 ||
3. k <= 0 ||
4. i+j-k <= 0 ||
5. i+k-j <= 0 ||
6. j+k-i <= 0)
7. type = Not a Triangle
8. else if (i == j){
9. if (j == k)
10. type = Equilateral
11. else type = Isosceles
12. else{ if (j == k)
13. type = Isosceles
14. else type = Scalene}
15. return type;
}
64 International Journal of Applied Metaheuristic Computing, 3(2), 56-74, April-June 2012
Copyright 2012, IGI Global. Copying or distributing in print or electronic forms without written permission of IGI Global is prohibited. Copyright 2012, IGI Global. Copying or distributing in print or electronic forms without written permission of IGI Global is prohibited.
Figure 3. CFG for triangle problem
Table 2. Experimental result of triangle problem for path exploration
Path Probability
Chosen
Path
Velocity
of
IWD
Time
of
IWD
Change
In
Soil
Soil
of
IWD
(1,2) 0.9
(1,2) 200.01 0.449 1.468 1.468 (1,7) 0.1
(2,3) 0.888
(2,3) 200.01 0.039 1.556 3.0254 (2,7) 0.111
(3,4) 0.875
(3,4) 200.03 0.035 1.662 4.687 (3,7) 0.125
(4,5) 0.857
(4,5) 200.03 0.029 1.793 6.480 (4,7) 0.143
(5,6) 0.833
(5,6) 200.04 0.024 1.961 8.441 (5,7) 0.166
(6,7) 0.2
(6,8) 200.05 0.019 2.187 10.629 (6,8) 0.7999
(8,9) 0.5
(8,9) 200.06 0.009 3.065 41.443 (8,12) 0.5
(9,10) 0.5
(9,10) 200.07 0.004 12.39 13.695 (9,10) 0.5
(10,15) 1 (10,15) 200.08 0.004 4.281 17.976
Path explored: 1 2 3 4 5 6 8 9 10 15
International Journal of Applied Metaheuristic Computing, 3(2), 56-74, April-June 2012 65
Copyright 2012, IGI Global. Copying or distributing in print or electronic forms without written permission of IGI Global is prohibited. Copyright 2012, IGI Global. Copying or distributing in print or electronic forms without written permission of IGI Global is prohibited.
Table 2. Brief description for Path-1 is given
below.
Put IWD on the top most node of CFG,
i.e., Node-1, it can have two possible out going
paths, either Node-2 or Node-7. As per formu-
lae given in Step-3 of Algorithm 2, it calculates
the probability for both nodes by using prob-
ability fitness function.
Probability (1, 2)
= 0.9 & Probability (1, 7) = 0.1
Probability value shows the probability of
selecting a particular path among all available
paths from the current node. It chooses the path
with the highest probability. Probability (1, 2) =
0.9, it chooses path (1, 2) and ignores path (1,
7). Probability of the path is affected by all the
metrics of IWD. Velocity (IWD) moving from
Node-1 to Node-2 is updated as per the equation
shown in Step-5 of Algorithm 2, i.e., Velocity
(IWD) = 200.01. In Step-6 of Algorithm 2,
update time parameter for IWD. It is calculated
as time(1,2;IWD) = HUD(2) / Velocity(IWD) =
90/200.01 = 0.449. Where HUD (2) = 10 (CC
of CFG) 9 (CC at Node-2) = 90. In Step-7 of
Algorithm 2, change of soil is calculated, i.e.,
soil(1,2) = 1.468 as per Step-7 of Algorithm
2. Update the Soil (IWD) = 1.468 loaded from
path between Node-1 and Node-2 as per Step-8
of Algorithm 2. In Step-9 of Algorithm 2, soil
of path between Node-1 and Node-2 is updated,
i.e., soil(1,2) = soil(1,2) - soil(1,2) = 10,000
- 1.468 = 9,998.532.
By using the same approach, Algorithm
2 iterates for the exploration of the remaining
paths. It finds all possible paths which count
to cyclomatic complexity (i.e., 10 for the given
triangle problem) of CFG. The available paths
in the given triangle problem are as follows.
Path 1: 1 2 3 4 5 6 8 9 10 15
Path 2: 1 2 3 4 5 6 8 12 13 15
Path 3: 1 2 3 4 5 6 8 9 11 15
Path 4: 1 2 3 4 5 6 7 15
Path 5: 1 2 3 4 5 7 15
Path 6: 1 2 3 4 7 15
Path 7: 1 2 3 7 15
Path 8: 1 2 7 15
Path 9: 1 7 15
Path 10: 1 2 3 4 5 6 8 12 14 15
From each of the paths obtained, test data
is generated based on the condition available
Table 3. Experimental result of triangle problem
Path No. Condition
Variable
i j k
1
!(i<=0) && !(j<=0) && !(k<=0) && !(i+j-k<=0) && !(i+k-j<=0) && !(j+k-i<=0) && (i==j) &&
(j==k)
4 4 4
2
!(i<=0) && !(j<=0) && !(k<=0) && !(i+j-k<=0) && !(i+k-j<=0) && !(j+k-i<=0) && !(i==j)
&& (j==k)
6 7 7
3
!(i<=0) && !(j<=0) && !(k<=0) && !(i+j-k<=0) && !(i+k-j<=0) && !(j+k-i<=0) && (i==j) &&
!(j==k)
6 6 1
4 !(i<=0) && !(j<=0) && !(k<=0) && !(i+j-k<=0) && !(i+k-j<=0) && (j+k-i<=0) 5 2 2
5 !(i<=0) && !(j<=0) && !(k<=0) && !(i+j-k<=0) && (i+k-j<=0) 3 8 5
6 !(i<=0) && !(j<=0) && !(k<=0) && (i+j-k<=0) 3 6 9
7 !(i<=0) && !(j<=0) && (k<=0) 9 7 0
8 !(i<=0) && (j<=0) 9 0 5
9 (i<=0) 0 7 3
10
!(i<=0) && !(j<=0) && !(k<=0) && !(i+j-k<=0) && !(i+k-j<=0) && !(j+k-i<=0) && !(i==j)
&& !(j==k)
6 4 5
66 International Journal of Applied Metaheuristic Computing, 3(2), 56-74, April-June 2012
Copyright 2012, IGI Global. Copying or distributing in print or electronic forms without written permission of IGI Global is prohibited. Copyright 2012, IGI Global. Copying or distributing in print or electronic forms without written permission of IGI Global is prohibited.
on that path. Table 3 shows the result of pro-
posed algorithm for generating test data from
the obtained path which contains the attributes,
namely, path number, condition, and values
of the variable. For particular path, test data
is generated based on the conditions available
on that path.
Detailed Calculation on Path-1: 1 2 3 4 5
6 8 9 10 15, is given.
Condition on any path can be extracted by
parsing the input program using YACC parsing
tool (Brown et al., 1995), only for each node
present in the condition; append all extracted
condition using AND operation (shown as &&).
Conditions present in Path-1 are as shown.
!(i<=0) && !(j<=0) && !(k<=0) &&
!(i+j-k<=0) && !(i+k-j<=0) && !
(j+k-i<=0) && (i==j) && (j==k)
Algorithm 3 tries to fire all different
combinations of variables i, j and k. Execute
generated combination on the given condition
until condition evaluates to true.
For example, randomly generated values
are
Variable: i, value: 1
Variable: j, value: 8
Variable: k, value: 5
Given combinations are not satisfying the
above condition, so it tries to generate different
combinations.
At some point it generates values which
satisfies the above condition for the variable
i, j and k like,
Variable: i, value: 4
Variable: j, value: 4
Variable: k, value: 4
These values are given as output as the test
data for Path-1: 1 2 3 4 5 6 8 9 10 15.
In the same way it generates test data for
the rest of the paths explored during Algorithm
2 and assigns unique combination to each one.
Table 3 shows the results for each path as per
condition that has been encountered.
Similarly, the proposed strategy is applied
to the min-max problem and results observed
are also very impressive.
In the study-2, proposed IWD approach is
being applied on another problem, determining
maximum of three inputs, as listed in Program
2. This program takes three values of i, j, k as
an input, and finds the maximum amongst the
three values.
The cyclomatic complexity for Program-2
is 4 which can be calculated after constructing
the CFG for the program. Table 4 shows the
result of the proposed algorithm by listing
Figure 4. Graph showing number of iterations per path
International Journal of Applied Metaheuristic Computing, 3(2), 56-74, April-June 2012 67
Copyright 2012, IGI Global. Copying or distributing in print or electronic forms without written permission of IGI Global is prohibited. Copyright 2012, IGI Global. Copying or distributing in print or electronic forms without written permission of IGI Global is prohibited.
conditions extracted for respective path and
generated test.
Table 4 contains attributes, namely, Path
Number, Traversed Path, Condition, and
values of the Variable. For particular path, test
data is generated based on the condition avail-
able on that path.
The above two studies show that, it dis-
covers all paths available in the CFG with less
number of iterations and also generates the test
data for the obtained paths. Comparison of these
approaches with other approaches is detailed
in the next section.
ANALYSIS
IWD approach is used for exploring the number
of paths available in the program. As already
shown, the algorithm is effectively generating
test data based on the obtained paths. This
section gives the comparison of the proposed
approach with other approaches.
Figure 4 shows the comparison of the IWD
with artificial bee colony algorithm (Dahiya,
Chhabra, & Kumar, 2010) for triangle problem
(which finds type of triangle from three sides
of the triangle). Comparison criteria have been
taken as the number of iterations needed to
generate test data for each path. In artificial bee
colony algorithm, the number of iterations for
triangle problem implemented by code having
CC=7 is 6197, whereas in the proposed IWD
implementation it needs only 518 iterations
for the same problem implemented by code
having CC = 10.
Figure 5 shows the comparison graph for
the generation of character type test data in
addition to numeric test data where the proposed
approach overcomes the conceptual limitations
for character test data generation. The proposed
approach can generate numeric as well as
character data but the other two approaches,
i.e., automatic test data generation based on
ACO (Kewen et al., 2009) and application of
genetic algorithm and Tabu search in software
Table 4. Experimental result for maximum problem
Path
No.
Path
Traversed
Condition
Variable
i j k
1 1 2 3 8 (i>j) && (j>k) 4 2 1
2 1 5 6 8 !(i>j) && (j>k) 1 2 0
3 1 2 4 8 (i>j) && !(j>k) 3 0 0
4 1 5 7 8 !(i>j) && !(j>k) 0 1 4
Figure 5. Graph showing character data support for the proposed approach
68 International Journal of Applied Metaheuristic Computing, 3(2), 56-74, April-June 2012
Copyright 2012, IGI Global. Copying or distributing in print or electronic forms without written permission of IGI Global is prohibited. Copyright 2012, IGI Global. Copying or distributing in print or electronic forms without written permission of IGI Global is prohibited.
Figure 7. Graph showing iterations vs. range
Program 2. Maximum of three input
int findMax(int i, int j, int k)
{
int max;
1. if(i>j) then
2. if(i>k) then
3. max = i;
4. else max=k;
5. elseif(j>k)
6. max = j;
7. else max = k;
8. return(max);
}
Figure 6. Graph showing comparison of number of iterations
International Journal of Applied Metaheuristic Computing, 3(2), 56-74, April-June 2012 69
Copyright 2012, IGI Global. Copying or distributing in print or electronic forms without written permission of IGI Global is prohibited. Copyright 2012, IGI Global. Copying or distributing in print or electronic forms without written permission of IGI Global is prohibited.
testing (Rathore et al., 2011) are applicable only
for numeric test data.
Figure 6 shows the comparison results of
the number of iterations during path generation
using IWD and random process. The number
of iterations required during path exploration
using random process is too high as compared to
the iterations required using IWD. The random
process is independent of the number of paths
but as graph of IWD is linearly monotonically
increasing it means it is directly dependent on
the number of the paths. The advantage of this
algorithm is that, it requires exactly the same
number of iterations as that of available paths.
Figure 7 contains the graph that shows the
number of iterations is closely related to the
range of variable per execution, e.g., In triangle
problem if range is set to be 0..5 then number of
iterations required for the test data generation for
each path is less than that with range 0..255. In
addition to this, as depicted in the graph, number
of iterations for generation of data increased
drastically for worst case conditions like (a==b),
because once value for variable a is gener-
ated, probability of generating same value for
variable b is 1/max(Range). In the same way
it can become worse if conditions like (a==b)
&& (b==c) has been given. So, probability of
generating the same value for variable b as
well as c again is 1/(max(Range))^2. In short,
number of iterations can be reduced by choos-
ing the range wisely for test data generation.
The developed tool provides option to select
the range of variables from user (developer/
tester) because he/she is the only person who
can take decision between trade-off of the se-
lecting the wide range v/s number of iterations
for generating test data.
Figure 8 shows the optimization techniques
for generating values of the variables present
in the path of predicate node. The technique is
applied during test data generation. This saves
unnecessary effort which would have been ap-
plied and wasted in randomly generating data
for unnecessary variable in each path.
As shown in Figure 8, consider any pro-
gram containing 200 variables and N number
of paths. Randomly chosen strategy generates
the values for each of the 200 variables during
test data generation of every path, whereas the
proposed approach generates test data for each
path according to the variables that are present
in the predicate node of the given path.
Figure 8. Graph showing number of variable need to generate per path
70 International Journal of Applied Metaheuristic Computing, 3(2), 56-74, April-June 2012
Copyright 2012, IGI Global. Copying or distributing in print or electronic forms without written permission of IGI Global is prohibited. Copyright 2012, IGI Global. Copying or distributing in print or electronic forms without written permission of IGI Global is prohibited.
CONCLUSION
This paper presents strategies for automatic test-
path exploration using IWD algorithm which is
able to find optimal solutions. The efficiency and
practicability of IWD is proved by the testing
of several real problems, and the results were
compared to that of random processes. From
the results, it is proved that IWD is more ef-
ficient than random process, and it also avoids
the problem of consuming a large number of
iterations. The results of the experimental study
support the claims that the proposed algorithm
is superior to other related strategies.
This paper has provided solutions for the
test data generation taking into consideration
only the numeric and character types. In future,
more works are needed to be carried out for
extending the proposed solutions to the string
data type.
REFERENCES
Briand, L. C. (2002). On the many ways Software
Engineering can benefit from Knowledge Engineer-
ing. Proceeding in 14th SEKE, (pp. 3-6).
Brown, D., John, L., & Tony, M. (1995). lex & yacc
(Second ed.). OReilly Media.
Dahiya, S. S., Chhabra, J. K., & Kumar, S. (2010).
Application of Artificial Bee Colony Algorithm to
Software Testing. 21st Australian Software Engi-
neering Conference 2010(ASWEC), (pp. 149-154).
Dahiya, S. S., Chhabra, J. K., & Kumar, S. (2010).
Application of Artificial Bee Colony Algorithm to
Software Testing. 21st Australian Software Engi-
neering Conference 2010(ASWEC), (pp. 149-154).
Doug, B., John, L., & Tony, M. (1995). lex & yacc
(Second ed.). OReilly Media.
Duan, H., Liu, S., & Lei, X. (2008). Air Robot Path
Planning Based on Intelligent Water Drops Optimiza-
tion. IEEE International Joint Conference on Neural
Networks, (pp. 1397-1401).
Edvardsson, J. (1999). A Survey on Automatic Test
Data Generation. In Proceedings of the Second
Conference on Computer Science and Engineering,
(pp. 21-28).
Hamed, S. (2007). Problem solving by intelligent
water drops. IEEE Congress on Evolutionary Com-
putation, (pp. 3226-3231).
Harman, M. (2007). The Current State and Future
of Search Based Software Engineering. 29th Int.
Conference on Software Engineering (ICSE 2007),
Future of Software Engineering (FoSE).
Huaizhong, L., & C. Peng, L. (2005). Software Test
Data Generation using Ant Colony Optimization.
Journal of World Academy of Science Engineering
and Technology, 557-560.
Kewen, L., Zilu, Z., & Wenying, L. (2009). Auto-
matic Test Data Generation Based On Ant Colony
Optimization. Fifth International Conference on
Natural Computation Institute of Electrical and
Electronics Engineers, (pp. 216-220).
Korel, B. (1990). Automated Software Test Data
Generation. IEEE Transactions on Software En-
gineering, 16(8), 870879. doi:10.1109/32.57624
Li, H., & Lam, C. P. (2005). Software Test Data
Generation using Ant Colony Optimization. Jour-
nal of World Academy of Science Engineering and
Technology, 557-560.
McMinn, P. (2004). Search-Based Software Test Data
Generation: A Survey. Software Testing . Verification
and Reliability, 14(3), 212223.
Michael, C., & Gary, M. (1998). Automated Soft-
ware Test Data Generation for Complex Programs.
Proceeding in 13th IEEE International Conference
on Automated Software Engineering, (pp. 136-146).
Michael, C., & McGraw, G. (1998). Automated Soft-
ware Test Data Generation for Complex Programs.
Proceeding in 13th IEEE International Conference
on Automated Software Engineering, (pp. 136-146).
Michael, C., McGraw, G. E., & Michael, A. S., &
Curtis C., W. (1997). Genetic Algorithms for Dynamic
Test Data Generation. Proceeding in 12th IEEE
International Conference on Automated Software
Engineering, (pp. 307-308).
Naaz Raza, E. (2009). Artificial Intelligence Tech-
niques in Software Engineering. In Proceedings of
the International MultiConference of Engineers and
Computer Scientists, (pp. 1-3).
Pedrycz, W., & Peters, J. F. (1998). Computational
Intelligence in Software Engineering. World Scien-
tific Publishing Co. Pte. Ltd.
Pressman, R. S. (2010). Software Engineering:
A Practitioners Approach (7 ed.). McGraw-Hill
Higher Education.
International Journal of Applied Metaheuristic Computing, 3(2), 56-74, April-June 2012 71
Copyright 2012, IGI Global. Copying or distributing in print or electronic forms without written permission of IGI Global is prohibited. Copyright 2012, IGI Global. Copying or distributing in print or electronic forms without written permission of IGI Global is prohibited.
Praveen Ranjan Srivastava is working under the Software Engineering and Testing Research
Group in the Computer Science and Information Systems Department at the Birla Institute of
Technology and Science (BITS) Pilan at the Pilani Campus in Rajasthan, India. He is currently
doing research in the area of software testing. His research areas are software testing, quality
assurance, quality attributes ranking, testing effort, software release, test data generation, test
effort estimation, agent oriented software testing, and metaheuristic approaches. He has published
more than 80 research papers in various leading international journals and conferences in the
area of software engineering and testing. He has been actively involved in reviewing various
research papers that he has submitted in his field for different leading journals and various in-
ternational and national level conferences. He has received various funds from different agencies
like Microsoft, IBM, Google, DST, CSIR, etc.
Amitkumar Patel is pursuing Master of Engineering specializing in Software Systems from
Computer Science and Information Systems Department at the Birla Institute of Technology
and Science (BITS) Pilani, Pilani Campus, Rajasthan, India. He was working with one of the IT
giant Wipro Technologies before joining BITS. His areas of interests include research in field
of software engineering & algorithms designs.
Kunal Patel is pursuing Master of Engineering specializing in Software Systems from Computer
Science and Information Systems Department at the Birla Institute of Technology and Science
(BITS) Pilani, Pilani Campus, Rajasthan, India. His areas of interests are software development,
software testing and parallel computing. He is an open source enthusiast and has been actively
contributing to the community.
Rathore, A., Bohara, A., Gupta, P. R., Lakshmi
Prashanth, T. S., & Srivastava, P. R. (2011). Ap-
plication of Genetic Algorithm and Tabu Search
in Software Testing. In Proceedings of the Fourth
Annual ACM Bangalore Conference (COMPUTE
11), (pp. 1-4).
Shah-Hosseini, H. (2007). Problem solving by intel-
ligent water drops. IEEE Congress on Evolutionary
Computation, (pp. 3226-3231).
Shah-Hosseini, H. (2009). The intelligent water
drops algorithm: a nature-inspired swarm-based
optimization algorithm. International Journal of Bio-
Inspired Computation, 1(1/2), 7179. doi:10.1504/
IJBIC.2009.022775
Srivastava, P. R., & Baby, K. (2010). Automated
Software Testing Using Metahurestic Technique
Based on an Ant Colony Optimization. International
Symposium on Electronic System Design (ISED),
(pp. 235 - 240).
Srivastava, P. R., Baby, K., & Raghurama, G. (2009).
An Approach of Optimal Path Generation using Ant
Colony Optimization. TENCON 2009 - IEEE Region
10 Conference 2009, (pp. 1-6).
Srivastava, P. R., Jain, P., Baheti, S., & Samdani, P.
(2010). Test Case generation using Genetic Algo-
rithm. International journal Information Analysis
and Processing (IJIAP), 3 (1), 7-13.
Srivastava, P. R., Km, B., & Raghurama, G. (2009).
An Approach of Optimal Path Generation using Ant
Colony Optimization. TENCON 2009 - IEEE Region
10 Conference 2009, (pp. 1-6).
Srivastava, P. R., Ramachandran, V., Kumar, M.,
Talukder, G., Tiwari, V., & Sharma, P. (2008).
Generation of Test Data Using Meta Heuristic Ap-
proach. TENCON 2008 - IEEE Region 10 Confer-
ence, (pp. 1-6).
72 International Journal of Applied Metaheuristic Computing, 3(2), 56-74, April-June 2012
Copyright 2012, IGI Global. Copying or distributing in print or electronic forms without written permission of IGI Global is prohibited.
Prateek Vijaywargiya is currently a first year Masters (Masters of Engineering-Software System)
student at Birla Institute of Technology & Science, Pilani and has just completed a Bachelors
Degree in Computer Science and Engineering in the university of Rajiv Gandhi Technical Uni-
versity, Bhopal. He has most of his interest in the Development Area within the area of his
specialization that is C/C++/Java. Apart from his regular curriculum, he was an active member
of the institute for two years and achieved FIRST position in the Software Debugging. He has
also achieved FIRST position in the TAEK-WON-DO at State and District Level.
International Journal of Applied Metaheuristic Computing, 3(2), 56-74, April-June 2012 73
Copyright 2012, IGI Global. Copying or distributing in print or electronic forms without written permission of IGI Global is prohibited.
APPENDIX
The application of the tool that has been developed for simulation of proposed solution has been
described through following screen shots. Simulated results displayed in the output text area.
Status messages shown in the respective section of the UI.
Figure 9. The simple GUI for simulating the results of the proposed intelligent water drop algorithm
Figure 10. As the input CFG, all the independent paths have been extracted as explained in
Algorithm 2 in above sections. Execution time is presented in the status.
74 International Journal of Applied Metaheuristic Computing, 3(2), 56-74, April-June 2012
Copyright 2012, IGI Global. Copying or distributing in print or electronic forms without written permission of IGI Global is prohibited.
The output in the text area shows the generated test data according to each path and the condi-
tions extracted in each path. In the status part, elapsed time is displayed along with the total
number of iterations.
Figure 11. Simulation of the Algorithm 3 is elaborated in this figure