130419 aminata sabane - a study of the impact of antipatterns on class testability

MotivationsGoal

Empirical StudyResults

Other Findings: Refactoring for Reducing Testing CostConclusion

Outline

Motivations

Goal

Empirical Study

Results

Other Findings

Conclusion

A. Sabane, M. Di Penta, G. Antoniol, Y-G. Gueheneuc March 07th, 2013 1 / 26

MotivationsGoal



Motivations

What are Antipatterns?

Identify poor solutions to recurring design problems[Brown et al.]

Some causes

Time-market pressure

Insufficient knowledge and or experience

Misapplication of some design patterns


MotivationsGoal



Motivations

Impact of Antipatterns...


MotivationsGoal



Motivations

What about Testability and Testing?


MotivationsGoal



Goal

Study the impact of antipatterns on class testability


MotivationsGoal



Class Testability MeasureMaDUMObjectsSubjectsRQ1: Assess AP class impact on testing effort?RQ2: Which APs are the most expensiveRQ3: Cost-effectiveness of APs testing

Class Testability Measure

What is Testability?

Degree to which a system or component facilitates theestablishment of test criteria and performance of tests todetermine whether those criteria have been met [IEEEGlossary]

Testing effort [Bache et al.]

How to Measure Testability?

Metric-Based approaches [Bruntink et al.]

number of test cases required to satisfy a given coveragecriterion [Bache et al.]


MotivationsGoal




MaDUM

What is it?

Unit testing technique for object oriented programs [Bashir etal.]

Based on data slices

Core idea: The correctness of a class is equivalent to thecorrectness of all its slices

Focus on the test of methods interactions


MotivationsGoal




MaDUM


MotivationsGoal




Why MaDUM?

Why MaDUM?

Specific to OO programs

Not require specific design documentation

Possibility to compute the number of required test cases

Upper-bound for testing cost


MotivationsGoal




Objects

4 systems: Ant, ArgoUML, Checkstyle & JFreechart

Systems belong to different application domains, havebug-fixing data available and have been used in previousstudies [Abbes et al.], [Bruntink et al.]


MotivationsGoal




Subjects

13 antipatterns

Studied in previous studies [Khomh et al.], [Abbes et al.],

Detection tool: Decor [Moha et al.]

Name (Abbr) Ant ArgoUML CheckStyle JFreeChartAntipattern classes 452 901 161 245No Antipattern (None) 297 376 99 233


MotivationsGoal




RQ1: Assess AP class impact on testing effort?

How large is the MaDUM test suite for classes participating in APscompared to that of other classes?


MotivationsGoal




RQ2: Which APs are the most expensive

How does the size of the MaDUM test suite vary among classesparticipating in different kinds of APs?


MotivationsGoal




RQ3: Cost-effectiveness of APs testing

What is the potential cost-benefit achieved when focusing testingon APs, as opposed to other classes?


MotivationsGoal



RQ1RQ2RQ3

Results

Results...


MotivationsGoal



RQ1RQ2RQ3

RQ1: How large is the MaDUM test suite for classesparticipating in APs compared to that of other classes?

●

●

●●

●

●●

●

●

●●

●

●

●●

●●

●

●

●

●

●

●

●

●

●

●●

●●

●●●●●●

●

●●●●●

●●

●

●

●

●

●

●●

●

●

●

●●

●

●

●

●

●

●

●

●●

●

●

●

●

●●

●

●

●

●

●

●●●

●●

●

●●

●

●

●

●

●

●

●●

●

●

●

●

●

●

●

●

●

●

●

●

●●

●

●●

●

●

●

●

●

●

●

●

●

●

●

●●

●

●

●●

●

●

●

●

●

●

●

●●

●

●

●

●●

●

●●●

●

●

●

●

●

●

●●●●●●●●●●●●

●

●

●

●

●●●

●

●

●

●

●

●

●●

●●

●

●

●

●

●

●

●●

●

●

●

●

●

●

●

●

●

●

●

●

●

●●

●

●

●

●●●●●

●

●●●●●●●

●

AP NAP AP NAP AP NAP AP NAP

020

4060

80

Ant ArgoUML Checkstyle JFreeChartSystemClasses

# of

Tes

t Cas

es (<

=100

)

Figure: Number of test cases in the MaDUM test suite for classesparticipating (AP) and not (NAP) in antipatterns.


MotivationsGoal



RQ1RQ2RQ3

RQ1: How large is the MaDUM test suite for classesparticipating in APs compared to that of other classes?

System Mean TCs AP Mean TCs NAP p-Value Cliff’s dAnt 18 9 < 0.01 0.23 (Small)

ArgoUML 10 3 < 0.01 0.35 (Medium)CheckStyle 9 6 = 0.01 NAJFreeChart 26 13 < 0.01 0.22 (Small)


MotivationsGoal



RQ1RQ2RQ3

RQ2: How does the size of the MaDUM test suite varyamong classes participating in different kinds of APs?

●●●●●

●●●

●

●

● ●●●

●●●●

●

●●

●●

●●●

●

●

●●●●●

●●

●

●

●

●

●

●●●●●●●●●●●●●●●●●●●●●

●

AS BCSBA B CDSBP CC LzC LM LPL MC RPB SC SG SAK NONE

02

04

06

08

01

00

Kind of Antipattern

# o

f Te

st

Ca

se

s (

<=

10

0)


MotivationsGoal



RQ1RQ2RQ3

RQ2: How does the size of the MaDUM test suite varyamong classes participating in different kinds of APs?

●

●

●

●

●

●

●

●

●

●●●●●●●●

●●●●●

●●●

●

AS BCSBA B CDSBP CC LzC LM LPL MC RPB SC SG SAK NONE

02

04

06

08

01

00

Kind of Antipattern

# o

f Te

st

Ca

se

s (

<=

10

0)

Classes participating in APs related to “excess of responsibility” aremore expensive


MotivationsGoal



RQ1RQ2RQ3

RQ3: What is the potential cost-benefit achieved whenfocusing testing on APs, as opposed to other classes?

0 1000 2000 3000 4000 5000 6000 7000

020

4060

80100

# of Test Case (cum.)

Defec

ts in te

sted c

lasses

(cum.)

It is cost-effective to analyze/test APs classes with a higher prioritythan other classes


MotivationsGoal



Other Findings: Refactoring for Reducing Testing Cost

Effect of traditional refactoring on testing cost?


MotivationsGoal



Other Findings: Refactoring for Reducing Testing Cost

Before Afterrefactoring refactoring

Class (system) Type TRS TCs TRS TCsTokenFilter (Ant) CDSBP 5 263 2 27PropPanel (ArgoUML) Blob 5 271 3 43BooleanExpressionComplexityCheck

(Checkstyle)LPL 6 732 5 132

AxisState (JFreeChart) NAP 5 248 1 11DynamicTimeSeriesCollection(JFreeChart)

Blob 4 208 2 122


MotivationsGoal



Conclusion

Findings

Most of APs classes are expensive for unit testing

APs classes with excess of responsibilities require more testingeffort than other

Prioritize the testing of APs classes can be more cost-effective

Specific refactoring can help to reduce testing cost


MotivationsGoal



Conclusion

Future Work

Study more systems

Investigate the symptoms that make APs classes moreexpensive

Study the impact of APs on testing effort from other aspects

Investigate specific refactoring for testability


MotivationsGoal



Thank you very much for your attention!


130419 aminata sabane - a study of the impact of antipatterns on class testability

Technology

Transcript of 130419 aminata sabane - a study of the impact of antipatterns on class testability