DESIGN QUALITY OF ASPECT-ORIENTED AND

OBJECT-ORIENTED PROGRAMS

An Empirical Comparison

Konstantina Georgieva, Ayaz Farooq and Reiner R. Dumke

Inst. of Distributed Sys., Faculty of Computer Science, OvG University Magdeburg

P.O.Box 4120, 39106 Magdeburg, Germany

Keywords:

Aspect-oriented programming, Object-oriented programming, Software metrics, Java, AspectJ, Empirical

analysis.

Abstract:

The aspect-oriented programming introduces the next step in the software design approaches in the sense

of overcoming the imperfections in the object-oriented paradigm by separating the cross-cutting concerns

and manipulating them in a separate manner. While the use of aspect-orientation attempts to tackle some of

the issues with object-oriented approach, it raises some others. How far is the design quality improved by

the use of aspect-orientation over object-orientation. What are common design trends among AO programs

in comparison with OO programs? A comparison among design metrics for both these kinds of programs

can help ﬁnd answers to these questions. We have used Chidamber & Kemerer metrics suite to empirically

compare design properties of AO programs with OO programs. Results from Chidamber & Kemerer metrics

for eleven AspectJ benchmark programs and those for dozens of Java libraries and programs are analyzed.

The experiment reveals few interesting design trends and some limitations of the used metrics suite.

1 INTRODUCTION

As an another innovation to the hierarchy of pro-

gramming technologies, aspect-oriented program-

ming (AO) is aimed at enabling better software de-

signs. It introduces clear responsibilities for in-

dividual modules, consistent implementation, im-

proved reusability, and improved skill transfer (Lad-

dad, 2003). The new aspect-oriented paradigm is

aimed to improve designs over its well-known pre-

decessor, object-oriented programming,by separating

the cross-cutting concerns and in this way maintain-

ing better modularity. In this new type of code it is be-

lieved that better cohesion, maintainability, reusabil-

ity and quality are achieved.

Evaluation of aspect-oriented vs. object-oriented

programming designs can be studied in different

ways. One of these methods is the comparison be-

tween AO and OO programs for analyzing the ben-

eﬁts or improvements brought by either of the two

technologies. Such comparisons can test validity of

the claims that AO programming scores well on some

of the quality attributes in comparison to OO pro-

gramming. It has been observed that empirical re-

search of this kind is yet at an early stage (Guy-

omarc’h and Guhneuc, 2005).

Within this context, this paper performs an em-

pirical comparison between AO and OO program

designs by using the well-known and widely used

Chidamber & Kemerer (subsequently referred to as

C&K) metrics. This metrics suite was initially deﬁned

for object-oriented systems (Chidamber and Kemerer,

1994) and has been extended for aspect-oriented pro-

grams (Ceccato and Tonella, 2004). The interpreta-

tion and measurement of thesemetrics for AO and OO

programs is certainly slightly different. However, the

comparisons between these metrics can still be mean-

ingful since they measure the same property of design

in both cases with the same scale. We have used mea-

surement results from a very large source of object-

oriented measurements (Farooq et al., 2005) which

holds C&K metrics results for several Java standard

libraries. The results are compared with the measure-

ment of AO version of the same metrics calculated for

a set of programs written in AspectJ.

This experiment’s beneﬁt is twofold. On one

hand it will help us compare complexity, maintain-

ability, usability and other quality attributes of typ-

ical AO and OO programs. Secondly, it will re-

veal the typical design and implementation trends for

287

Georgieva K., Farooq A. and R. Dumke R. (2009).

DESIGN QUALITY OF ASPECT-ORIENTED AND OBJECT-ORIENTED PROGRAMS - An Empirical Comparison.

In Proceedings of the 4th International Conference on Software and Data Technologies, pages 287-290

DOI: 10.5220/0002237102870290

 SciTePress

aspect-oriented programs. This information can also

be used for benchmarking among AO programs. A

few other works have compared aspect-oriented and

object-oriented programs (Tsang et al., 2004), (Guy-

omarc’h and Guhneuc, 2005), (Kulesza et al., 2006).

However, our experiment is based on relatively very

large set of measurements which gives strength to

generalizability of our results.

2 ASPECT-ORIENTED AND

OBJECT-ORIENTED DESIGN

METRICS

As already mentioned above, several sets of met-

rics exist for aspect-oriented programs and even more

available for object-oriented programs as well. To be

able to make the planned comparisons, we needed a

common and resembling set of metrics for both AO

and OO programs. Chidamber & Kemerer metrics

appeared an excellent choice as two variants of this

metrics suite are available for aspect-oriented (Cec-

cato and Tonella, 2004) and object-oriented pro-

grams (Chidamber and Kemerer, 1994). C&K met-

rics for OO program used in this experiment are

Weighted Method per Class (WMC), Depth of Inher-

itance Tree (DIT), Number of Children (NOC), Cou-

pling between Objects (CBO), Response for a Class

(RFC), and Lack of Cohesion in Methods (LCOM).

C&K metrics for AO programs used in this exper-

iment are Weighted Operations in Module (WOM),

Depth of Inheritance Tree (DIT), Number Of Children

(NOC), Response for a Module (RFM), Coupling be-

tween Modules (CBM), and Lack of Cohesion in Op-

erations (LCO).

3 THE EXPERIMENT

A very large collection of C&K metrics val-

ues for object-oriented programs is already

provided by an online measurement repository

called OOMJ (available at http://donner.cs.uni-

magdeburg.de:8080/oomj). This resource is based

on measurement of thousands of classes from

several Java standard libraries and programs. To

calculate C&K metrics for aspect-oriented pro-

grams we have used an open source tool called

aopmetrics (available at http://aopmetrics.tigris.org).

In our evaluation, we have measured the sam-

ple programs located under the examples folder

of AspectJ 1.6.3 distribution, LoD (found at

http://www.ccs.neu.edu/home/lorenz/papers/aosd2003

Table 1: Size metrics for measured programs.

Number

of classes

Number

of meth-

ods

Avg. LOC

per class

AOP 114 502 35

OOP 14372 140784 642

Table 2: Descriptive statistics: C&K metrics for AO pro-

grams.

Min. Max. Mean Std. Dev.

DIT 0 4 0.63 0.9

NOC 0 3 0.2 0.63

WOM 0 23 4.4 4.68

CBM 0 10 1.62 6.366

RFM 0 235 13.067 2.12

LCO 0 219 10.39 35.99

lod/), and TETRIS (found at http://www.guzzzt.com/

coding/aspecttetris.shtml). The next section sum-

marizes some initial measurements of the measured

programs.

3.1 Size Metrics

Table 1 gives some size metrics for our experiment.

The sample size of object-oriented programs is enor-

mously large which provides stable trends of C&K

metrics values for OO programs. Comparatively

fewer AO programs could be measured. However, the

sample is large enough for any meaningful and rep-

resentative comparisons between both of the AO and

OO paradigms.

4 RESULTS

4.1 Descriptive Statistics

Table 2 shows minimum, maximum, mean, and stan-

dard deviation values of C&K metrics for AO pro-

grams. A relatively high value of standard deviation

for CBM and LCO indicates a high variation among

the values of these metrics.

Table 3 compares values of mean and standard de-

viation for the similar C&K metrics among both types

of OO and AO programs. The results show smaller

averages for number of operations per class (WOM),

response for class (RFM), and lack of cohesion (LCO)

values for aspect-oriented programs. The rest of the

metrics show almost similar trends. Furthermore, the

smaller standard deviations for almost all of the AO

ICSOFT 2009 - 4th International Conference on Software and Data Technologies

288

Table 3: Comparison between the C&K metrics for OO and

AO programs

Mean Std. Deviation

Metric OO AO OO AO

DIT 0.7 0.6 1.0 0.9

NOC 0.4 0.2 2.0 0.6

WMC/WOM 11.9 4.4 15.8 4.7

CBO/CBM 1.9 1.6 3.9 6.4

RFC 19.5 13.1 28.6 2.0

LCOM/LCO 139.4 10.4 802 36.0

Table 4: 95% Conﬁdence interval: mean of C&K metrics

for AO and OO programs.

Metric OOP

(lower/upper

limit)

AOP

(lower/upper

limit)

DIT 0.57–0.61 0.47–0.80

NOC 0.14–0.16 0.09–0.33

WMC/WOM 8.54–8.85 3.54–5.26

CBO/CBM 2.92–3.09 1.23–2.01

RFC/RFM 14.45–15.10 4.47–7.20

LCOM/LCO 35.88–39.14 3.79–17.00

metrics make these averages more meaningful and re-

liable.

Sometimes, mean value of an entity may be mis-

leading specially when there is a very large variation

among the values. For a further deeper analysis of

the mean values shown in table 3 we calculated conﬁ-

dence intervals for each of the metrics. Table 4 shows

conﬁdence intervals at 95% level for each of the met-

rics.

To give another orientation to the aforementioned

statistics, ﬁgure 1 presents graphical view of the OO

and AO metrics distributions. Metrics values are

drawn on the horizontal axis while on the vertical axis

are the frequencies of these values.

4.2 Measurement Analysis

General observations: The ﬁrst comment is about the

overall range of metric values. Table 3 indicates a

smaller standard deviation for all of AO metrics as

compared to their OO counterparts. This is due to

the fact that most metrics values fall within a nar-

row range with very few outliers. Figure 1 shows fre-

quency distributions for these metrics which corrobo-

rate this fact. This makes the results of table 4 more

representative and reliable which is aimed to tell us a

most probable value for each of these metrics.

WMC/WOM: A small WMC/WOM is considered

a good design practice which is believed to reduce

complexity. Separation of concerns in AO designs

0 1 2 3 4+

%ofClasses

DIT

0Ͳ5 6Ͳ10 11Ͳ15 16Ͳ20 21+

%ofClasses

WMC/WOM

100

101+

%ofClasses

RFC/RFM

100

0 1 2 3 4+

%ofClasses

NOC

100

0Ͳ4 5Ͳ8 9Ͳ12 13Ͳ16 17+

%ofClasses

CBO/CBM

100

101+

100

0Ͳ50 51Ͳ100 101Ͳ150 151Ͳ200 201+

%ofClasses

LCOM/LCO

OO/Java AO/AspectJ

Figure 1: C&K metrics distributions for AO and OO pro-

grams.

DESIGN QUALITY OF ASPECT-ORIENTED AND OBJECT-ORIENTED PROGRAMS - An Empirical Comparison

289

should reduce the size of operations. This is clearly

seen by a smaller value of this metric observed in the

performed measurements.

DIT/NOC: Both these metrics as indicators of in-

heritance show slightly smaller values for AO pro-

grams vs. OO programs. Based on their character-

istics, these metrics should not be much affected with

the change of paradigm from OO to AO. Perhaps that

is the reason that these metrics do not show any sig-

niﬁcant changes in program design strategies.

Coupling related measures, to be discussed

next, are more important for analysis since aspect-

orientation mainly concerns with this issue.

RFM: This metric, similar to RFC of C&K met-

rics suite, is a kind of coupling measure. Both the

ﬁgure 1 and the tables 3 and 4 show that its value is

between 0-25 in about 95% of the cases. This metric

reﬂects reduced coupling for AO programs. It shows

a considerable design improvement of the AOP over

the OOP.

CBO/CBM: CBM of AO is the closest parallel to

the CBO of OO programs. It is perhaps the most im-

portant measure characterizing the couplings which is

the main motivation for the paradigm shift from OOP

to AOP. The ﬁgure 1 and table 3 show considerably

reduced coupling between modules. This seems to

be a consequence of AO’s separation of concerns and

woven code.

LCO: In an OO or AO design, there should be

high cohesion between methods/operations. Non-

cohesive operations indicate a need for splitting

classes/aspects. Despite being in the range of 0-50,

a very high variation of this metric was observed for

object-oriented programs. However, in case of AO, it

shows a very small average with a very low value of

standard deviation, see table 3. Table 4 shows that its

value could be between 3 and 17. A higher majority

of measured AO programs have showed a very high

cohesion which reﬂects an improved design practice.

5 CONCLUSIONS

This paper presented a metrics-based comparison

among object-oriented and aspect-oriented designs

using several samples of OO/AO libraries and pro-

grams. Amplitude of the OO measurement data made

this empirical study unique among the very few ex-

periments of this kind ever performed. From the pro-

grams and libraries that we measured, it was found

that both kinds of programs exhibited similar inheri-

tance related characteristics. Small coupling and high

cohesion was observed in the measured AO programs

w.r.t OO programs. Based on these quality indica-

tors we were able to conﬁrm the thesis that aspect-

orientation improves modularity and maintainability.

Overall, AO reﬂected (usually considered) better de-

sign characteristics in contrast to OO programs.

A very large proportionof measured OO programs

consisted of standard libraries, and not normal pro-

grams/applications. The study needs to be extended

with a larger and diverse sample of both real OO and

AO programs using more coupling related and other

relevant metrics.

REFERENCES

Ceccato, M. and Tonella, P. (2004). Measuring the effects

of software aspectization. In Proceedings of 1st Work-

shop on Aspect Reverse Engineering.

Chidamber, S. R. and Kemerer, C. F. (1994). A metrics

suite for object-oriented design. IEEE Trans. Softw.

Eng., 20(6):476–493.

Farooq, A., Braungarten, R., and Dumke, R. R. (2005). An

empirical analysis of object-oriented metrics for Java

technologies. In INMIC 2005: Proceedings of the

9th IEEE International Multitopic Conference, pages

548–553. IEEE Computer Society.

Guyomarc’h, J.-Y. and Guhneuc, Y.-G. (2005). On the

impact of aspect-oriented programming on object-

oriented metrics. In Proceedings of the 9th

ECOOP workshop on Quantitative Approaches in

Object-Oriented Software Engineering, pages 42–47.

Springer-Verlag.

Kulesza, U., Sant’Anna, C., Garcia, A., Coelho, R., von

Arndt Staa, and Lucena, C. (2006). Quantifying the

effects of aspect-oriented programming: A mainte-

nance study. In ICSM ’06: Proceedings of the 22nd

IEEE International Conference on Software Mainte-

nance, pages 223–233. IEEE Computer Society.

Laddad, R. (2003). Aspect-oriented programming will im-

prove quality. Software, IEEE, 20(6):90–91.

Tsang, S. L., Clarke, S., and Baniassad, E. (2004).

Object metrics for aspect systems: Limiting em-

pirical inference based on modularity. Technical

report, Trinity College Dublin, Ireland. Avail-

able at http://www.cse.cuhk.edu.hk/˜elisa/papers/OO-

AOMetrics.pdf.

ICSOFT 2009 - 4th International Conference on Software and Data Technologies

290