Dealing with Inheritance in OO Evolutionary...

GECCO 2009 Dealing with Inheritance in OO Evolutionary Testing 1 of 16

Dealing with Inheritance in OO Evolutionary Testing

Javier Ferrer

, Francisco Chicano and Enrique [email protected]

[email protected]

[email protected]

Dealing with Inheritance in OO Evolutionary Testing

Javier Ferrer

, Francisco Chicano and Enrique [email protected]

[email protected]

[email protected]

IntroductionThe Test Case Generator

Distance for instanceofExperimentsConclusions


3

Table of ContentsTable of Contents

Introduction

The Test Case Generator

Distance for instanceof

Experiments

Conclusions

1

2

4

5




After codification, software products require a test phase

The objective is to find errors

Object Oriented paradigm is followed by most software developers

Inheritance is an important issue of this paradigm

We propose a distance measure for the instanceof

operator that use the information of the class hierarchy

We define two mutation operators based on the distance

IntroductionIntroduction




Select a Partial Objective

Optimization Algorithm

End

Continue?yes

no

Test Case Generator

Test case

Objective function

Program

The

Test

Case GeneratorThe

Test

Case Generator

The

test

case generator

breaks

the

global objective

into

several

partial

objectives

Our generator creates a coverage table with the values that traverses each branch



Genetic Algorithm


Distance

for

instanceofDistance

for

instanceof

We

defined

an

objective

function

(fitness) to

be minimized



truefalse

o1

instanceof

c1

o2 instanceof

c2

Cond

C

Cond

B


Distance

for

instanceofDistance

for

instanceof



d=3h+2a

Hierarchical

walk= 3Approximation

walk= 2

ArrayList

instanceof

Integer


Distance

for

instanceofDistance

for

instanceof



d=2h+2a

ArrayList

instanceof

SetArrayList

instanceof

Integer


Distance-based

and

Uniform

MutationDistance-based

and

Uniform

Mutation

O1 O2 O3 O4 O5C1 C2 C3 C4 C5

O1 O2 O3 O4 O5C1 C2 C3 C4 C5O1 O2 O3 O4 O5C1 C2 C3 C4 C5




Experiments: ProgramsExperiments: Programs


Distance for instanceofExperiments

Conclusions

Setting

ParametersMutation

Operators

intanceof

expression

GECCO 2009 Dealing with Inheritance in OO Evolutionary Testing10 of 16

Experiments

: Approximation

and

Hierarchical

ConstantsExperiments

: Approximation

and

Hierarchical

Constants

h = 1 h = 25 h = 50 h = 100

a = 200 75.45 % 75.33 % 74.93 % 75.68 %

a = 100 75.53 % 74.74 % 75.10 % 74.79 %

a = 50 74.85 % 75.81 % 74.44 % 73.56 %



Setting

ParametersMutation

Operators

Does it hold that a>h?

Yes, because a

weights how close the test case is to satisfy the condition


Experiments: Distance-based

and

Uniform

mutationExperiments: Distance-based

and

Uniform

mutation



Setting

ParametersMutation

Operators

• Average of 200 executions


Experiments: Distance-based

and

Uniform

MutationExperiments: Distance-based

and

Uniform

Mutation



Setting

ParametersMutation

Operators

• Average of 200 executions

• MDn

is better than MU

• MU is faster at the beginning

New proposal:

Adaptive Mutation


Adaptive

Mutation

: New

ProposalAdaptive

Mutation

: New

Proposal



Setting

ParametersMutation

Operators

Distance

Prob

abilit

y

Distance Distance

Prob

abilit

y

Prob

abilit

y

Mutation

probability

VS Distance

between

classes

Adaptive

Speed


Adaptive

Mutation: ExperimentsAdaptive

Mutation: Experiments



Setting

ParametersMutation

Operators

100% coverage

obtained

in all

programs

The

GA with

adaptive

mutation

is

much

better

than

the

Random

Search

Difficult

to

test

is

correlated

to

the

expression

i + j

λ

Evals

i + j = 3

i + j = 4

i + j = 5

i + j = 6

i + j = 7


Combine our proposal with other OO features

Analysis of the impact of our proposal in real-world software

Conclusions

& Future

WorkConclusions

& Future

Work



We created a test case generator able of dealing with inheritance

A new branch distance has been defined for inheritance

We have proposed and compared two mutation operators based on the

distance

The MDn

operator is better when using a greedy seeding of the GA

The number of atomic conditions and the nesting degree have a great

influence on the automatic testing complexity


THANKS FOR YOUR ATTENTION

Javier [email protected]

FINALFINAL



Dealing with Inheritance in OO Evolutionary...

Documents

Transcript of Dealing with Inheritance in OO Evolutionary...