Co-evolution Analysis for Software Product Lines

Anissa Benlarabi, Amal Khtira and Bouchra El Asri

IMS Team, Laboratory SIME, ENSIAS, University Mohamed V, Rabat, Morocco

Keywords:

Software Product Lines, Co-evolution, Cladistics Classiﬁcation.

Abstract:

The purpose of our approach is to study the co-evolution of the platform and the products of software product

lines. Because the platform must be able to derive all the family products, products are not allowed to evolve

independently from the plateform, thus the propagation of the products changes must be managed efﬁciently.

Instead of focusing on the change impact analysis we propose an approach to compare the evolution histories

of the products and the platform illustrated through evolutionary trees built using the biological technique

cladistics. This comparison yields important results concerning the change propagation. In this paper, we

introduce the use of cladistics for software product lines to build evolutionary trees for platform and products,

then we elaborate a mathematical analysis to compare these trees, afterwards we validate this work through a

case study (mobile media software product lines). We also provide the design of an automated tool.

1 INTRODUCTION

Today the software development tends to reuse the

knowledge more than building software from scratch.

The software product line (SPL) engineering is the

most widespread software reuse approach, it consists

in building a common platform for a set of products

dedicated for a business domain (Pohl et al., 2005).

The main advantage of this approach is the improve-

ment of the productivity by bringing to customers

products adapted to their needs at a reasonable price

and a short time (Clements et al., 2002). However, the

customers requirements evolve continuously, thus the

(SPL) must be adapted to cope with customers needs.

Among the most frequent SPL evolution prob-

lems, we mention the change propagation problem.

this issue occurs when the products evolve indepen-

dently from their common platform. In fact, if many

changes are applied to the products without propaga-

tion to the common platform, the SPL will not be able

to derive the family product. Hence, the SPL will fall

into the aging phenomenon. This issue has been tack-

led by many approaches, speciﬁcally the traceabil-

ity approaches (Anquetil et al., 2010) (Goknil et al.,

2011) that use the links between the SPL artifacts

to identify the impact of the change, the evolution

modeling approaches (Romero et al., 2013)(Ajila and

Kaba, 2008) that separates the evolution phase to el-

ementary operations for better understanding of the

impact. These approaches focus mainly on the change

understanding and the impact analysis and rely essen-

tially on human knowledge (Benlarabi et al., 2014b).

Instead of focusing on the change impact analysis

activity we proposed a new approach for co-evolution

anaysis in SPL, which consists in modelling the evo-

lution history of the platform and the derived products

in evolutionary trees and compare them to understand

how they impacted each other during their evolution

and also to discover the changes of products that were

not propagated to the platform with the aim of propa-

gating them to the platform. The comparison of evo-

lutionary trees is a prominent technique used in biol-

ogy to study the co-evolution of organisms (Ehrlich

and Raven, 1964), it allows for understanding the his-

tory of life. To model the evolution of the SPl we

use the cladistics classiﬁcation (CC) (Brinkman et al.,

2004). We illustrated our approach through a case

study on the mobile media SPL (Tizzei et al., 2011).

In this paper the focus is on the summary of our

co-evolution modelling approach and the elaboration

of an automated tool. The remainder of this paper is

structured as follows. Section 2 introduces the soft-

ware evolution problem. Section 3 reminds the bio-

logical co-evolution principal, the cladistics method

and the analogy between SPL co-evolution and bio-

logical co-evolution. Section 4 presents our approach

to study the co-evolution in SPL. Section 5 illustrates

the approach through the mobile media case study.

Section 6 investigates the design of an automated tool

and the section 7 concludes the paper.

263

Benlarabi A., Khtira A. and El Asri B..

Co-evolution Analysis for Software Product Lines.

DOI: 10.5220/0005464702630269

In Proceedings of the 10th International Conference on Evaluation of Novel Approaches to Software Engineering (ENASE-2015), pages 263-269

ISBN: 978-989-758-100-7

 2015 SCITEPRESS (Science and Technology Publications, Lda.)

2 SPLs EVOLUTIONS

The SPL engineering consists in building a platform

composed of common assets for a set of products.

By reusing these assets (Clements et al., 2002), cus-

tomers get products adapted to their needs at reason-

able price and short time. However, the users re-

quirements evolve over time, hence the SPL must be

adapted continuously to cope with the user require-

ments changes. In this section we present the SPL

evolution propagation problem. Thereby, we discuss

the previous research and we introduce our approach.

2.1 SPLs Evolution

Software evolution is an important activity because

it must accompany the evolution of business require-

ments. Compared to single software evolution, the

SPL evolution is more difﬁcult since the change must

be propagated in the two levels: the level of the prod-

ucts and the level of the platform. Hence, if a change

concerns an application product, It could not be all the

time propagated to the platform. Thereby, if products

are changed frequently and independently, the domain

artifacts of the SPL risk to be out of date and instead

of having a set of product derived from the same do-

main artifacts, we will have a set of independent soft-

ware.

2.2 Change Propagation Problem

The great majority of works done to deal with the

change propagation problem such as the traceabil-

ity (Anquetil et al., 2010)(Goknil et al., 2011) or the

evolution modelling (Romero et al., 2013)(Ajila and

Kaba, 2008) approaches focus on the change impact

analysis activity because they consider that this prob-

lem is mainly caused by a shortage in this activity.

Regarding the traceability approaches, they rely on

links deﬁned by the developers to determine the im-

pact of the change, while the evolution modelling

approaches separates the change to elementary op-

erations and identify the impact of each operation

based on a database of relationships between differ-

ent components. Both approaches presents shortcom-

ings. Firstly they rely on human knowledge and use

data that have to be manually provided which is error

prone and too expensive in practice. In addition, they

perform only pre-analysis of the change.

We aim to overcome the presented shortcomings

by studying the co-evolution of the products and the

platform. Through our approach (Benlarabi et al.,

2014b) we improve the change understanding by

summarizing the evolution history in an evolutionary

tree, we also deﬁne the impact of the change by com-

paring the evolutionary tree of the platform with the

evolutionary tree of each products, and we help cor-

recting the divergences between the platform and the

products.

3 CO-EVOLUTION FOR SPLs

In this section we introduce the co-evolution of organ-

isms with concrete examples from biology, then we

present the technique used to study the co-evolution

in biology and we discuss the analogy with SPL co-

evolution.

3.1 Co-evolution in Biology

According to Erlich (Ehrlich and Raven, 1964) the

co-evolution is the examination of patterns of inter-

action between two major groups of organisms with

a close and evident ecological relationship. The co-

evolution of host-parasite is a famous example of bi-

ological co-evolution. Because parasites cause dam-

ages to their hosts, hosts develop new capacities to re-

sist to their parasites however parasites also develop

capacities to overcome this resistance (Anderson and

M.May, 1982). Hence, a clearer understanding of

hostparasite co-evolution will point to new possibil-

ities for organic farming and reduce the application of

ecologically harmful chemicals.

3.2 Co-evolution Analysis Through

Cladistics

Co-evolution analysis consists practically on compar-

ing the evolution histories of populations depicted

by their phylogenetic trees (Fitch and Margoliash,

1967). the phylogenetic tree of a population is a

branching diagram showing the inferred evolution-

ary relationships among their members on the ba-

sis of the similarities and differences in their phys-

ical or genetic characteristics. There are two main

methods to create a phylogenetic tree (Fitch and Mar-

goliash, 1967): phenetic which is DNA sequence

based method, and cladistics which is character-based

method. We choose cladistics method (Brinkman

et al., 2004) because we rely on features. It consists

in the following steps:

Organisms selection: the organisms of the classi-

ﬁcation should be necessarily related by ancestor-

descendant relationships.

Characters extraction: the characters include physical

and behavioral characters that allow to make distinc-

tion between organisms.

ENASE2015-10thInternationalConferenceonEvaluationofNovelSoftwareApproachestoSoftwareEngineering

264

Figure 1: SPL and Product Cladograms.

Characters matrix construction: this matrix is ﬁlled by

numbers that represents the codiﬁcation of the states

of the characters in organisms.

Cladograms drawing: the cladogram is a tree. All the

organisms are in the endpoints of the tree, each node

regroups shared characters between all the next organ-

isms.

3.3 Cladistics Classiﬁcation for SPLs

The proposed approach aims at examining the co-

evolution of domain and application feature models

in SPLs, similarly to biology we investigate the co-

evolution of two major groups of organisms that are

closely related, the ﬁrst group is composed by the

common platform and the second group is composed

by the derived products. All the populations members

are derived from the same ancestor which is the ﬁrst

release of the common SPL platform. Thus, we study

the interaction between the two populations on the ba-

sis of the history of their evolution. The input of our

study is the feature models of the different releases of

the platform and the products, otherwise, for a prod-

uct we gather all the feature models of its releases then

we build the evolutionary tree of the product on the

basis of the models by applying the CC method. We

repeat this operation for each product and also for the

common platform in order to produce their evolution-

ary trees. Thereby we proceed to the comparison of

the resulted trees.

4 CO-EVOLUTION OF DOMAIN

AND APPLICATION

ENGINEERING

Similarly to biology the characters of the products are

the features and the variants of the SPL. In our pre-

vious work (Benlarabi et al., 2014b) we introduced a

new approach for establishing the evolutionary trees

of the platform and the derived products in based on

cladistics classiﬁcation, we also proposed a mathe-

matical analysis to compare these trees and to ﬁnd out

the products changes that were not propagated to the

platform. In this section, we give an overview of our

proposal.

4.1 Cladistics Application for SPLs

The ﬁrst step of the CC method is the populations

identiﬁcation. In our study we identiﬁed many pop-

ulations, each one is a set of software systems derived

from the same ancestor software. The ﬁrst population

is composed by the releases of the common platform,

while the other populations are composed by the dif-

ferent releases of the products. The second step is

characters extraction, the characters are the features.

We extract the independent features of each popula-

tion. For example, Considering the following pop-

ulation of the product P1 composed by n versions

{P1.1,P1.2,P1.3,...,P1.n}, the set of its characters

is {F1.1,F1.2,F1.3,...,F1.m} where m is the num-

ber of the independent features of the population.

After the extraction of characters we construct the

characters states matrix of each population, each fea-

ture has two states, the primitive state 0 which denotes

the non-existence of the feature in the release, and the

derived state 1 which denotes its existence. in Fig.

1 we represent an example of the features states ma-

trix of the population of the product P1. The last step

is the evolutionary tree construction, in this step we

draw a branching tree of the population based on the

characters states matrix, in the Fig. 1 ﬁgures the re-

sulted cladogram of the matrix of our population ex-

ample

4.2 Perfect Co-evolution Hypothesis

By investigating the similarities and the divergences

between the cladogram of the platform and the clado-

gram of each product, we can deduce the changes

happened to the products that were not propagated

to the platform. The branches of a cladogram repre-

sent the set of features of the population. The imper-

fect evolution is caused by branches that exist in one

cladogram and are absent from the other. However,

we eliminate the imperfection caused by branches

that exist in the SPL cladogram and are absent from

the products cladogams because features of the SPL

could not be all used in the products. We express

mathematically the relationship between the clado-

gram of a product and the cladogram of the platform

on the basis of their features states matrices. In this

subsection we formalize our hypothesis, then in the

following subsection we proceed to the comparison

and we discuss the restoration of missed features to

the platform.

Co-evolutionAnalysisforSoftwareProductLines

265

We consider two populations P1 and P2. P1 con-

sists of k releases of the SPL and P2 consists of k re-

leases of a derived product. By extracting features of

the two populations we will obtain the sets A1 com-

posed of n features and the set A2 composed of m

features respectively for P1 and P2:

A1 = {F1i while i ∈ N, i ≤ n}

A2 = {F2i while i ∈ N, i ≤ m}

We stated the following hypothesis in our previous

work as follow:

(H0) Features Independence. In the set of features

A1 and A2, the features are independent.

∀i, j ∈ N,i ≤ n, j ≤ n, F1i 6= F1 j

∀i, j ∈ N,i ≤ m, j ≤ m, F2i 6= F2 j

(H1) Domain Features Sufﬁciency. Each feature

of the set A2 has a corresponding feature in the set A1.

∀ j ∈ N, j ≤ m,∃i ∈ N,i ≤ n/F2 j = F1i

(H2) Features Exclusion. This hypothesis is deduced

from the combination of H0 and H1. Each feature of

A2 has only one corresponding feature in A1.



{F2i,F2 j/i, j ≤ m} ⊂ {F1x,F1x/x,y ≤ n}

F2i = F1x

⇒ F2 j = F1y

4.3 Perfect Co-evolution Restoration

On the basis of the hypotheses H0 and H2, we deduce

that the relationship between the two cladograms of

P1 and P2 must respect the following inequality:

B × A1 ≥ C × A2

Where B is the matrix (k x n) of features states of P1,

k is the number of taxa of P1 and P2, C is the matrix (k

x m) of features states of P2, A1 is the features vector

(n x 1) of P1 and A2 is the features vector (m x 1) of

P2.

After developing the inequality, it is reduced to a set

of k inequalities:











∑

i=1

1,i

× F1

≥

∑

j=1

1, j

× F2

∑

i=1

2,i

× F1

≥

∑

j=1

2, j

× F2

∑

i=1

k,i

× F1

≥

∑

j=1

k, j

× F2

By considering the hypothesis H1, the inequalities

can be transformed to a set of k equalities because

the difference between the two sides of each inequal-

ity i where 0 < i ≤ k will be equal to the set of fea-

tures of the SPL version P1i that did not exist in

the product version P2i. The number s represents

the difference between n the number of features of

P1 and m the number of features of P1 (s = n-m):

B × A1 −C × A2 =







1,1

·· · d

1,s

2,1

·· · d

2,s

k,1

·· · d

k,s



















The vector of features (F3

,F3

,··· , F3

) represents

the set A3 which is given by subtracting A2 from A1

(A3 = A1 − A2), and the entries d

i, j

/i, j ∈ N,0 < i ≤

k,0 < j ≤ s of the matrix D must be equal to 0 if the

feature F3

exist in the SPL version P1i or 1 if the

opposite is true.

In order to repropagate the identiﬁed changes to

the SPL we propose the algorithm of restoration (Ben-

larabi et al., 2014b) in which we add to A1 all the fea-

ture of A3 that are absent from A1, these features rep-

resent the product features that are not derived from

the domain features of the SPL. Then we extend the

features states matrix of the SPL P1 with these fea-

tures.

5 CASE STUDY

We performed a case study on the mobile media SPL

(Benlarabi et al., 2014a). It manipulates photo, mu-

sic, and video on mobile devices such as mobile

phones and has many derived products (Tizzei et al.,

2011). Firstly, we constructed a population of the SPL

formed by its different releases, it is described in the

table 1. We construct also a population for the derived

product. We started by applying our approach on the

Mobile Media and one of its derived product.

5.1 Cladistics Classiﬁcation for Mobile

Media SPL

We constructed two populations, the ﬁrst one P1 is

formed by seven versions of the SPL mobile media,

while the second one P2 is formed by seven versions

of the derived product. In Tables 1 and 2, respectively,

we give a detailed description of the populations P1

and P2, which we constructed on the basis of the fea-

ture models of the different releases of P1 and P2:

Each population is composed of six organisms, P1 has

19 features and P2 has 16 features:

A1 = {F

1,i

while i ∈ N,i ≤ 19}

A2 = {F

2,i

while i ∈ N,i ≤ 16}

The features states matrices of the two population are

constructed on the basis of Table 1 and Table 2. We

present in Table 3 an example of the matrix B of the

mobile media SPL. The matrix C of the products is

constructed similarly. We constructed cladograms on

ENASE2015-10thInternationalConferenceonEvaluationofNovelSoftwareApproachestoSoftwareEngineering

266

Table 1: Mobile Media Platform Population.

V1.0 The ﬁrst release of the mobile media SPL, this release encom-

passes the following features: Manage photos (F

1,1

), Create al-

bum (F

1,2

), Delete album(F

1,3

), Create media (F

1,4

), Delete me-

dia (F

1,5

), View media (F

1,6

), Sort media (F

1,7

), Edit media label

1,8

)

V1.1 Release 2: the features Set favorites (F

1,9

) and See favorites

1,10

) were added

V1.2 Release 3: the feature Copy media (F

1,11

) was added

V1.3 Release 4: the features Send media (F

1,12

) and Receive media

1,13

) were added

V1.4 Release 5: the feature Add music media management (F

1,14

)

was added

V1.5 Release 6: the features Add video media management (F

1,15

Capture videos (F

1,16

) and Capture photos (F

1,17

) were added

V1.6 Release 7: the features Play videos (F

1,18

) and Play music

1,19

) were added

Table 2: Derived Product Population.

V2.0 The ﬁrst release of the product, this release encompasses the

following features: Manage photos (F

2,1

), Create album (F

2,2

Delete album(F

2,3

), Create Photo (F

2,4

), Delete Photo (F

2,5

View Photo (F

2,6

), Sort media (F

2,7

), Edit media label (F

2,8

)

V2.1 Release 2: the features Set favorites (F

2,9

) and See favorites

2,10

) were added

V2.2 Release 3: the feature Copy media (F

2,11

) was added

V2.3 Release 4: the features Send media (F

2,12

) and Receive media

2,13

) were added

V2.4 Release 5: the feature Print photo (F

2,14

) was added

V2.5 Release 6: the feature Capture photos (F

2,15

) was added

V2.6 Release 7: the feature Share photo in social websites (F

2,16

) was

added

Table 3: Features States Matrix B of the Mobile Media SPL.

B F

1,1..

1,8

1,9

1,10

1,11

1,12

1,13

1,14

1,15..

1,17

1,18

1,19

V0 1 0 0 0 0 0 0

V1 1 1 0 0 0 0 0

V2 1 1 1 0 0 0 0

V3 1 1 1 1 0 0 0

V4 1 1 1 1 1 0 0

V5 1 1 1 1 1 1 0

V6 1 1 1 1 1 1 1

the basis of these matrices by grouping releases to-

gether based on their shared characters. In this step

we used the biological tool PHYLIP to generate the

coordinates of the evolutionary trees of P1 and P2

from their features states matrices. We used the out-

pout ﬁles generated by PHYLIP in the online tool

Phyﬁ to generate cladograms that correspond to the

coordinates of these ﬁle. After generating the two

cladograms we proceed to the co-evolution analysis

of the two populations

5.2 Co-evolution Analysis of Mobile

Media SPL

After applying the mathematical analysis we de-

scribed before, we noticed that two imperfections

were detected after the calculation. They are caused

by the two features F

2,14

”Print photo” and F

2,16

”Share photo in social websites”. The two features

exist in the product and are absent from the SPL. Fi-

nally we obtained the following formulate.

B × A1 −C × A2 =







0 0 0 0 0 0 0

1 0 0 0 0 −1 0

1 1 1 0 0 −1 0

1 1 1 1 1 −1 −1













1,14

1,15

1,16

1,18

1,19

2,14

2,16







The vector A3 is composed by the following features:

1,14

”Add music media management”, F

1,15

”Add

video media management”, F

1,16

”Capture videos”,

1,18

”Play videos”, F

1,19

”Play music”, F

2,14

”Print

photo”, F

2,16

”share photo in social websites”.

6 AUTOMATED TOOL

To implement our approach we have chosen to de-

sign a tool composed of three main functional blocks,

the ﬁrst block generates the features states matrix of

a population, the second block compare two clado-

grams using their features states matrices and vectors

of features and generates a log ﬁle containing the re-

sult of the comparison. and the third block proposes

on the basis of the generated log ﬁle the features that

may be added to the platform to correct the imperfect

co-evolution.

6.1 Features States Matrix Generator

In this step we use the FeatureIDE tool (Kastner et al.,

2009) to model our case study feature models. It is an

open source framework for software product line en-

gineering, the main advantage of this tool is the pos-

sibility to generate the XML scheme of the feature

model. in Fig. 2 we illustrate a domain feature model

of the mobile media software product line.

Once the xml ﬁles are generated, this block inte-

grate the features reference ﬁle and also the generated

ﬁles, then it looks for the features in each ﬁle and gen-

erate the matrix in a ﬂat ﬁle in the accepted input for-

mat for the PHYLIP tool.

Co-evolutionAnalysisforSoftwareProductLines

267

Figure 2: Mobile Media SPL feature model.

6.2 Co-evolution Comparator

The second block consists in comparing two input

matrices, by automating the mathematical analysis of

our approach. It integrates the two ﬁles generated by

the ﬁrst block and also the features reference ﬁles.

Thereby, it makes a comparison of the evolution his-

tories of the two population and generate a log ﬁle

which contains the traces of the comparison and the

resulted differences.

6.3 Perfect Co-evolution Analyzer

The third block deals with the perfect co-evolution

restoration, it integrates the log ﬁle generated by the

second blocks and displays the features that may be

restored to the platform. this block gives also the pos-

sibility to integrate the two matrices with the features

reference ﬁles and applies the perfect co-evolution

restoration algorithm which we previously elaborated

and generate a new matrix for the SPL platform.

Firstly we started by developing the ﬁrst block,

currently we are performing a list of test cases to val-

idate it. Subsequently, we intend after that to develop

the second and the third functional blocks. We envis-

age after ﬁnalizing the three blocks, integrating them

into an application for complete co-evolution analy-

sis.

7 CONCLUSIONS

Our general research work addresses the problem

of change propagation in evolving software product

lines. Thereby, we deal with the analysis of co-

evolution of the platform and the products of the soft-

ware product line to understand how they impact each

other during their evolution. In this paper, we have

ﬁrst proposed the use of biological technique cladis-

tics to illustrate the evolution history of the products

and the platform through evolutionary trees, each tree

represents the history of a product, then we estab-

lished a mathematical analysis to compare the trees

and to ﬁnd out the changes of product that were not

propagated to the platform and ﬁnally we elaborated

an algorithm to help repropagating the missing fea-

tures to the platform. We illustrated the approach

through a case study on the mobile media software

product line. There are several perspectives to our

current work. Firstly, we will develop a tool to auto-

mate the co-evolution analysis, we separated the tool

to three main functional blocks representing the three

steps of the approach. The ﬁrst block is currently

developed and under validation. Secondly, we aim

to improve the third step of our approach to act di-

rectly on the feature models of the platform and cor-

rect them.

REFERENCES

Ajila, S. A. and Kaba, A. B. (2008). Evolution support

mechanisms for software product line process. Jour-

nal of Systems and Software, 81(10):1784–1801.

Anderson, R. M. and M.May, R. (1982). Coevolution of

hosts and parasites. Parasitology, 85(02):411–426.

Anquetil, N., Kulesza, U., Mitschke, R., Moreira, A.,

Royer, J., Rummler, A., and Sousa, A. (2010).

A model-driven traceability framework for software

product lines. Software and Systems Modeling, 9:427–

451.

Benlarabi, A., A. Khtira, A., and Asri, B. E. (2014a). An

analysis of domain and application engineering co-

evolution for software product lines based on cladis-

tics: A case study. In ICSEA’14 Forthcoming (2014),

pages 495–501.

Benlarabi, A., Khtira, A., and Asri, B. E. (2014b). A co-

evolution model for software product lines: an ap-

proach based on evolutionary trees. In WCCS14. IEEE

Forthcoming (2014).

Brinkman, Fiona, S., and Leipe, D. D. (2004). Bioinformat-

ics: a Practical Guide to the Analysis of Genes and

Proteins, volume 43. John Wiley & Sons.

Clements, P., Northrop, L., and Boehm, B. W. (2002). Soft-

ware Product Lines : Practices and Patterns. Fondo

Xavier Clavigero, S.J. ITESO.

Ehrlich, P. R. and Raven, P. H. (1964). Butterﬂies and

plants: a study in coevolution evolution. JSTOR,

pages 586–608.

Fitch, W. and Margoliash, M. (1967). Construction of phy-

logenetic trees. Science, 155(760):279–284.

Goknil, A., Kurtev, I., van den Berg, K., and Veldhuis,

J. (2011). Semantics of trace relations in require-

ments models for consistency checking and inferenc-

ing. Software & Systems Modeling, 10:31–54.

Kastner, C., Thum, T., Saake, G., FeigenspanLeich, J.,

Wielgorz, T. F., and Apel, S. (2009). Featureide: A

tool framework for feature-oriented software develop-

ment. In IEEE 31st International Conference on Soft-

ware Engineering, pages 611–614. IEEE.

Pohl, K., Bckle, G., and van der Linden, F. J. (2005). Soft-

ware Product Line Engineering: Foundations, Princi-

ples and Techniques. Springer.

Romero, D., Urli, S., Quinton, C., Blay-Fornarino, M., Col-

let, P., Duchien, L., and Mosser, S. (2013). Splemma:

ENASE2015-10thInternationalConferenceonEvaluationofNovelSoftwareApproachestoSoftwareEngineering

268

A generic framework for controlled-evolution of soft-

ware product lines. In The 17th International Software

Product Line Conference, pages 59–66.

Tizzei, L. P., Dias, M., Rubira, C. M., Garcia, A., and Lee, J.

(2011). Components meet aspects: Assessing design

stability of a software product line. Information and

Software Technology, 53(2):121–136.

Co-evolutionAnalysisforSoftwareProductLines

269