TOWARDS A KNOWLEDGE BASE TO IMPROVE

REUSABILITY OF DESIGN PATTERN

Cédric Bouhours, Hervé Leblanc and Christian Percebois

IRIT – MACAO team (Models, Aspects, and Components for Object-oriented Architectures)

Université Paul Sabatier

118 Route de Narbonne

F-31062 Toulouse Cedex 9

Keywords: Software architectures, Design patterns, Design review.

Abstract: In this paper, we propose to take directly into account the knowledge of experts during a design review

activity. Such activity requires an ability to analyze and to transform models, in particular to inject design

patterns. Our approach consists in identifying model fragments which can be replaced by design patterns.

We name these fragments “alternative models” because they solve the same problem as the pattern, but with

a more complex or different structure than the pattern. In order to classify and to explain the design defects

of this alternative models base, we propose the concept of strong point. A strong point is a key design

feature which permits the pattern to resolve a problem most efficiently.

1 CONTEXT

The emergent MDE community, aiming to give a

productive character to models, has proposed model-

driven process development. However, these

processes should be able to reuse the knowledge of

experts generally expressed in terms of analysis

(Fowler, 1997), design (Gamma et al., 1995) or

architectural (Buschmann, 1996) patterns approved

by the community. Given the existence of “code

review” activities (Dunsmore, 1998) in some

development processes, we would like to introduce a

“design review” activity, directed by design patterns,

to improve object model quality. We limit our

approach to design patterns, because we consider

that analysis patterns are business domain specific,

and the use of architectural patterns must be planned

before the design stage.

This activity may be decomposed into four sub-

activ

ities. First, model preparation puts the model to

review in a minimal quality, for example, to impose

that one class implements at least one interface or to

impose that all attributes are private or protected.

Then, a research based on structural and behavioral

similarities determines the model fragments which

may be substitutable with a pattern. Next, a design

expert validates the patterns proposed to substitute

the fragments found in the previous sub-activity,

considering the designers’ intentions. Lastly, the

designers integrate the validated patterns into their

models. This integration is dealt with by automatic

parameterized transformations.

Up to now, in spite of the efforts to improve

reu

sability of design patterns, thanks to assistance

tools to guide pattern integration in models by

precise modeling (Guennec, 2000) (France, 2004),

and thanks to pattern wizards dedicated to integrate

patterns by code refactorings (Eden, 1997)

(O'Cinnéide, 1999), we do not find a model

inspecting tool that urges the use of patterns in the

most automatic way possible. To do the detection of

the substitutable fragments, we use a match method.

Rather than using an approximate design-pattern

match detection based on a similarity research

(Arcelli Fontana, 2004), we do exact pattern

matching of models substitutable with a design

pattern. Then, we seek a set of substitutable models

for each structural pattern proposed by Gamma et al.

We name these models “alternative models”.

According to the taxonomy proposed by Chikofsky

and Cross (Chikofsky, 1990), the implemented

technique can be connected to a redocumentation

technique so as to permit model restructuring.

In a first part, we present the concept of

altern

ative model, and how to collect and to use

421

Bouhours C., Leblanc H. and Percebois C. (2007).

TOWARDS A KNOWLEDGE BASE TO IMPROVE REUSABILITY OF DESIGN PATTERN.

In Proceedings of the Second International Conference on Software and Data Technologies - SE, pages 421-424

DOI: 10.5220/0001325104210424

 SciTePress

them. Next, we present a problem solvable with the

Composite design pattern, and its corresponding

alternative models. Moreover, to characterize the

design defects of these alternative models, we

deduce them with some modification to the pattern

structure.

2 ALTERNATIVE MODELS

An alternative model is a model which solves the

same problem as the pattern, but with a more

complex or different structure than the pattern.

Therefore, in agreement with the hypotheses on

design patterns and class design defects (Guéhéneuc,

2001), it is a candidate model for substitution with a

pattern.

Each alternative model is characterized like a

pattern. A set of structural features is associated

with each alternative model role. For the moment,

these features concern inter-class relations only: i.e.,

associations, generalizations and aggregation-

composition links, but neither interfaces nor class

semantics. We deduce these features both

associating corresponding pattern roles with each

class of alternative models, and studying their

structure (Bouhours, 2006).

2.1 Discovery

For this study, we choose to collect a set of models

which do not use any patterns. If these models solve

a problem solvable with a particular pattern, they

may be considered as an alternative model.

We have organized an experiment which consists

of the design of the seven standard problems

solvable with the seven GoF structural patterns, in

UML notation. We use the examples presented in

the “motivations” section of the GoF catalogue,

when they are relevant. Each problem admits a

solution using a pattern, but participants have solved

these problems without any pattern knowledge.

From three hundred models obtained, we have

selected fifteen of them which present significant

structural variants with patterns under consideration,

the others were either incorrect or duplicated design.

We consider these models valid because they permit

a solution to the problems under study, namely, they

implement functionalities required by this problem.

Each model obtained (between two and six for each

pattern) constitutes a plausible alternative to just one

pattern.

2.2 Use

This collection method allows us to constitute a

catalogue for each pattern and its associated

alternative models that we consider as potentially

bad design practices. One entry of the catalogue

corresponds to one pattern with its alternative

models classified by the strong points of the pattern.

A strong point is a key design feature which permits

the pattern to resolve a problem most efficiently. For

example, the Composite pattern resolves the

problem: “How to compose and use object

hierarchies as simply as possible for a client in

keeping the extensibility possibilities on

components?”. So the two strong points for this

pattern are “uniform processing” and “decoupling

and extensibility”.

The strong points are the “essence” of the

patterns. They are characterized by criteria of

object-oriented architecture and software

engineering quality, partially deduced from the

“consequences” section of the GoF catalogue and

from the study of the design defects of alternative

models. As pattern injection may alter some object

oriented metrics (Huston, 2001), they allow us to

compute dedicated pattern metrics to classify the

alternative models and to help the estimation of the

pertinence of pattern injection in a design model.

In order to characterize the design defects, we

deduce the alternative models in perturbing the

strong points of each pattern. A perturbation may

either delete a strong point or simply damage it. So

to specify the degree of damage, we add sub-features

for some strong points. Moreover, thanks to these

perturbations, we should build new alternative

models not taken from experiments. And, if we

reverse these perturbations and if we apply them on

alternative models, we would deduce a sequence of

structural refactoring operations (Sunyé, 2001) that

automatically perform the pattern integration in the

models to review.

3 COMPOSITE ALTERNATIVE

MODELS

The problem “Design a system enabling to draw a

graphic image: a graphic image is composed of

lines, rectangles, texts and images. An image may

be composed of other images, lines, rectangles and

texts.” may be solvable with the Composite design

pattern.

ICSOFT 2007 - International Conference on Software and Data Technologies

422

The below figures represent this problem

instantiation

(Model 0) and the five alternative models

(Models 1 to 5) taken from our experiment.

Thanks to the analysis of these alternative

models, we find two strong points with their sub-

features for the Composite pattern:

1 Decoupling and extensibility.

Image

<<Composite>>

Graphic

<<Component>>

Line

<<Leaf>>

Rectangle

<<Leaf>>

Text

<<Leaf>>

0: Composite problem instantiation

1.1 Maximal factorization of the composition.

1.2 Addition or removal of a leaf does not

need code modification.

1.3 Addition or removal of a composite does

not need code modification.

2 Uniform processing.

2.1 Uniform processing on operations of

composed object.

Graphic

<<Component>>

Image

<<Com posite>>

Line

<<Leaf>>

Rectangle

<<Leaf>>

Text

<<Leaf>>

1: Development of the composition on «Component»

2.2 Uniform processing on composition

managing.

2.3 Unique access point for the client.

In order to validate these strong points, we

deduce now each alternative model in perturbing the

strong points:

First, if we use the instantiated pattern

(Model 0)

and if we replace the inheritance links by inverted

composition links and composition links by

inheritance links, we obtain a first alternative model

(Model 1) where the first strong point is deleted and

the second is damaged on the first sub-feature.

Indeed, without the inheritance link, there is no

guarantee that the processing produces conformity

between «Composite» and «Leaf» classes.

In the first alternative model

(Model 1), if we

replace the inheritance link by its composition

equivalence, we keep on damaging the second

strong point, in deleting the second sub-feature.

Indeed, this alternative model

(Model 2) has only

composition relationships that impose to manage the

composition in «Component» and «Composite»

classes.

From the second alternative model

(Model 2), if we

factor «Composite» composition links on

«Component», we obtain an alternative model

(Model

without a single strong point. Indeed, this

factorization adds a cycle between «Composite» and

«Component» that produces two access points for

the client.

From the instantiated pattern

(Model 0), by

developing composition from «Composite» to

«Component» over every sub-class of

«Component», we obtain a new alternative model

(Model 4). The first strong point is deleted, but the

second is only damaged on the first sub-feature.

Although the access point is not in a good place,

namely on «Composite», we consider that the third

sub-feature is present.

Lastly, if we add an intermediary class between

«Composite» and «Component» in the problem

Graphic

<<Component>>

Image

<<Composite>>

Line

<<Leaf>>

Rectangle

<<Leaf>>

Text

<<Leaf>>

* *

* * *

2: Development of the composition on

«Component» and «Composite»

Image

<<Composite>>

Graphic

<<Com ponent>>

Line

<<Leaf>>

Rectangle

<<Leaf>>

Text

<<Leaf>>

* *

3: Recursive composition

Graphic

<<Com ponent>>

Image

<<Composite>>

Line

<<Leaf>>

Rectangle

<<Leaf>>

Text

<<Leaf>>

** *

4: Development of the composition on «Composite»

Graphic

Image

<<Composite>>

GraphicComponent

<<Component>>

Line

<<Leaf>>

Text

<<Leaf>>

Rectangle

<<Leaf>>

5: Indirect composition on «Composite»

TOWARDS A KNOWLEDGE BASE TO IMPROVE REUSABILITY OF DESIGN PATTERN

423

instantiation (Model 0), we damage the two strong

points. In the first, the damage is due to the first

sub-feature: the factorization is not maximal. For

the second strong point, the composition

management is done in «Composite» and “Graphic”

classes, and there are two redundant access points

(Model 5).

Table 1 resumes the state of every strong point

for each Composite alternative model. For each

strong point, we represent sub-features in this table

with a “

+” if it is present and with a “-” if it is

deleted. In a first approach, we define a “quality

score” simply based on strong points. We consider

the strong points qualitatively equivalent, and

compute the metrics with their degree of

perturbation.

4 CONCLUSION

In order to validate our approach, we have applied

OCL rules on UML models. So, it has been

necessary to implement detection of each alternative

model by several rules. When a rule is not validated

in the model, the NEPTUNE (Neptune, 2003)

platform returns the context of the error, which is the

model fragment substitutable by a pattern. In a first

attempt, we have applied these rules on industrial

models and then on OMG meta-models.

To increase the range of our catalog, which is,

for now, only constituted of alternative models taken

by our experiments, we are currently developing a

collaborative web site allowing to share knowledge

about object misconception.

REFERENCES

Arcelli Fontana F., Raibulet C., Tisato F., “Design Pattern

Recognition”, in Proceedings of the ISCA 13th IASSE,

Nice, France, July 1-3, 2004, pages 290-295.

Bouhours C., Leblanc H., Percebois C., “Structural

variants detection for design pattern instantiation”, in

1st International Workshop on DPD4RE, Benevento,

Italy, October 2006.

Buschmann F., Meunier R., Rohnert H., Sommerlad P.,

Stal M., “Pattern-Oriented Software Architecture”,

John Wiley \& Sons, August 1996.

Chikofsky E. J., Cross J. H., “Reverse engineering and

design recovery: A taxonomy”, in IEEE Software,

7(1), page 13-17, January 1990.

O'Cinnéide M., Nixon P., “A Methodology for the

Automated Introduction of Design Patterns”, in ICSM

'99: Proceedings of the IEEE International

Conference on Software Maintenance, IEEE

Computer Society, 1999, pages 463.

Dunsmore A.P., Comprehension and Visualisation of

Object-Oriented code for Inspections, Technical

Report, EFoCS-33-98, Computer Science Department,

University of Strathclyde, 1998.

Eden A. H., Yehudai A., Gil J., “Precise specification and

automatic application of design patterns”, in ASE '97:

Proceedings of the 12th international conference on

Automated software engineering (formerly: KBSE),

IEEE Computer Society, 1997, pages 143.

Fowler M., “Analysis patterns: reusable objects models”,

Addison Wesley Longman Publishing Co, Inc., 1997

France R. B., Kim D., Ghosh S., Song E., “A UML-Based

Pattern Specification Technique”, in IEEE Trans.

Softw. Eng., IEEE Press, 2004, 30, pages 193-206.

Gamma E., Helm R., Johnson R., Vlissides J., “Design

Patterns: Elements of Reusable Object-Oriented

Software”, Addison Wesley Professional, 1995.

Guennec A. L., Sunyé G., Jézéquel, J., “Precise Modeling

of Design Patterns”, in UML, 2000, pages 482-496.

Guéhéneuc Y. G., Albin-Amiot. H., “Using Design

Patterns and Constraints to Automate the Detection

and Correction of Inter-Class Design Defects”, in

Proceedings conference TOOLS, July 2001, pages

296-305.

Huston B., “The effects of design pattern application on

metric scores”, in Journal of Systems and Software,

58(3), Elsevier Science, September 15, 2001, pages

261-269.

Neptune, [w] http://neptune.irit.fr, 2003.

Sunyé G., Pollet D., Le Traon Y., Jézéquel J.M.,

“Refactoring UML Models”, in Proceedings of UML

2001, pages 134-148.

Table 1: State of the Strong Points of the Composite Alternative Models.

Alternative model number

Strong point Sub-features 5 1 4 2 3

- - --

- - - -

+ - - --

- + + --

Quality score:

50% 33.3% 33.3% 16.7% 0%

ICSOFT 2007 - International Conference on Software and Data Technologies

424