Exploring the Role of Commercial Stakeholders in Open
Source Software Evolution
Andrea Capiluppi1, Klaas-Jan Stol2, and Cornelia Boldyreff3
1Brunel University, United Kingdom
2Lero—The Irish Software Engineering Research Centre
University of Limerick, Ireland
3University of East London, United Kingdom
Abstract. It has been lately established that a major success or failure factor of
an OSS project is whether or not it involves a commercial company, or more ex-
tremely, when a project is managed by a commercial software corporation. As
documented recently, the success of the Eclipse project can be largely attributed
to IBM’s project management, since the upper part of the developer hierarchy is
dominated by its staff. This paper reports on the study of the evolution of three dif-
ferent Open Source (OSS) projects — the Eclipse and jEdit IDEs and the Moodle
e-learning system — looking at whether they have beneﬁted from the contribution
of commercial companies. With the involvement of commercial companies, it is
found that OSS projects achieve sustained productivity, increasing amounts of
output produced and intake of new developers. It is also found that individual and
commercial contributions show similar stages: developer intake, learning effect,
sustained contributions and, ﬁnally, abandonment of the project. This preliminary
evidence suggests that a major success factor for OSS is the involvement of a com-
mercial company, or more radically, when project management is in hands of a
Governance and control in Open Source Software (OSS) has been dramatically chang-
ing . The traditional volunteer-based OSS project model is now being accompanied
by sponsored OSS, where commercial stakeholders provide effort beyond voluntary
programmers. It has been argued that OSS projects have become increasingly hybrid
with respect to this type of contributing stakeholders .
Since their inception in the early 1980s, OSS projects were mostly volunteer-based
(or Traditional OSS, right end of Figure 1), heavily relying on personal efforts and non-
monetary recognition, and reportedly suffering from communication and coordination
Nowadays, so-called Sponsored OSS projects have also been documented as more
similar to Closed Source systems (as in far left of Figure 1). They could be industry-led
OSS projects, where a commercial stakeholder plays a major role in the development
and decision making, as in the case of the Eclipse project by IBM [25,24,19,33]. They
I. Hammouda et al. (Eds.): OSS 2012, IFIP AICT 378, pp. 178–200, 2012.
IFIP International Federation for Information Processing 2012
Exploring the Role of Commercial Stakeholders in Open Source Software Evolution 179
Community Open Source
Sponsored Open Source
Restricted governance Open governance
Fig. 1. Continuum of governance in software projects: proprietary software (left) to Open Source
can also be industry-involved OSS projects when they are driven by an OSS community,
but often have one or several companies or institutions (e.g., universities) among their
stakeholders, as in the case of the Moodle Content Management System (CMS) .
Both the industry-led and industry-involved scenarios introduce new challenges to
OSS projects: the ﬁrst is based on one (or a small subset of) critical stakeholder(s),
which could eventually halt the project if they decide to abandon it1. In the case of
Eclipse, for example, IBM staff have been identiﬁed as the top contributors, with only
a few external developers working on the core system . For Community OSS, espe-
cially in the case of large and complexOSS systems, there is a need of proper incentives
for different types of stakeholders, with complementary expertise and requirements, in
particular when their contributions are relevant to a system’s core functionality.
This paper aims to explore these three scenarios and to study whether the involve-
ment of commercial companies can help sustaining the evolution of OSS projects. To
that end, the paper presents different analyses of the evolution of a commercial and a
traditional OSS systems (Eclipse and jEdit), sharing the same application domain, and
one community OSS project (Moodle).
By exploring the type of activities performed by commercial stakeholders, and by
comparing the results achieved by similar OSS projects (sharing the same application
domain, but with different involvement of stakeholders), this paper explores a research
area that only recently started to be covered in the literature [28,27,31].
Existing literature typically distinguishes between community-managed (also called au-
tonomous ) and sponsored communities. However, current terminology for this has
some issues. For instance, OSS projects that are led by ﬁrms are referred to as commer-
cial OSS, whereas an OSS project that involves commercial stakeholders (companies),
1This happened with Netscape Navigator (then Mozilla) when Netscape Communications Cor-
poration (NCC) released it as open source, but without further evolving it.
180 A. Capiluppi, K. Stol, and C. Boldyreff
but is led by an OSS community (consisting of “traditional” community members) is
referred to as Community OSS. We argue that both terms are not precisely deﬁned and
need further reﬁnement. Commercial OSS suggests that proﬁt is made from the OSS
project. The term “Community OSS” does not clearly distinguish projects that involve
companies from “traditional” OSS projects (that do not involve companies). Therefore,
in this paper we propose the following new terminology for the various models of
Traditional OSS projects are those projects in which no companies are involved.
Industry-involved OSS projects are projects in which commercial ﬁrms are
involved as contributors, but the project is still managed by the “community”.
Industry-led OSS projects are projects that are led by a commercial ﬁrms. The
wider community can contribute (as with any OSS project), but since a company has
control over the project, it deﬁnes the evolution strategy.
Together, industry-involved and industry-led projects are Sponsored OSS projects,
whereas industry-involved and traditional projects are both forms of Community
projects (see Figure 1) as they are led by a community (as opposed to a company).
1.2 Structure of This Paper
The remainder of this paper is structured as follows: Section 2 presents the goals, ques-
tions and metrics of the study. Section 3 presents the research design. Sections 4 and 5
presents the Eclipse and jEdit case studies, respectively. Section 6 focuses on the Moo-
dle system as an example of an industry-involved OSS system, and explores the rele-
vance of the commercial stakeholders, and how they differ from individual developers.
Section 7 discusses the results followed by conclusions in Section 8.
2 Background and Related Work
This section provides a brief overview of relevant backgroundand related studies. Most
reports on participation of ﬁrms in OSS projects present results from large-scale
Bonaccorsi and Rossi studied contributions to OSS projects by commercial ﬁrms.
They conducted a large-scale survey among 146 Italian companies that provide soft-
ware solutions and services based on Open Source Software . One of the ﬁndings
was that approximately 20 per cent of companies were coordinating an OSS project.
Furthermore, almost half of the companies (46.2%) had never joined an OSS project. It
is important to note that these results were published in 2004, and that these numbers
may have changed signiﬁcantly over the last eight years; we suggest that a replication
of this study would be a valuable contribution.
Bonaccorsi and Rossi have further studied (using data from the same survey) moti-
vations of ﬁrms to contribute to OSS projects [7,29].
Bonaccorsi et al.  have investigated whether and how ﬁrms contribute to OSS
projects. Their study investigated which activities ﬁrms undertake in OSS projects, as
well as whether the presence of ﬁrms affect the evolution of OSS projects. To address
these questions, Bonaccorsi et al. conducted a survey of 300 OSS projects hosted on
Exploring the Role of Commercial Stakeholders in Open Source Software Evolution 181
SourceForge.net. They found that almost one in three of the studied projects had one
or more ﬁrms involved. In a survey of 1,302 OSS projects by Capra et al. , similar
results were found, namely that ﬁrms were involved in 31% of the projects. Different
types of involvement were identiﬁed: (1) project coordination,(2) collaboration in code
development, and (3) provision of code. Capra et al.  made a slightly different clas-
siﬁcation of participation models: the Management model (for project coordination),
the Support model (sponsoring through ﬁnancial or logistic support) and the Coding
model (contributing code, bug ﬁxes, customization, etc.). In most cases, it was found
that the ﬁrms founded the OSS project, but in some cases ﬁrms took over by replacing
a project’s coordinator.
Aaltonen and Jokinen  studied the inﬂuence in the Linux kernel community and
found that ﬁrms have a large impact in the project’s development.
Martinez-Romo et al.  have studied collaboration between an OSS community
and a company.They conducted case studies of two OSS projects: Evolution and Mono.
Companies can sponsor OSS projects in different ways. Berdou  investigated the
dynamics of cooperation in community-led projects that involve paid contributors, and
proposed a framework to understand this relationship.
Dahlander and Magnusson  proposed a typology consisting of symbiotic (win-
win), commensalistic (ﬁrm gains, community indifferent) and parasitic (ﬁrm gains,
community loses) approaches to characterize ﬁrm–community relationships. These re-
lationships only apply in community-led projects.
The last decade of research in OSS has well established the relationship between
ﬁrms and OSS projects. This relationship has been shown to have a direct effect on a
project’s sustainability. However, what kind of effects this relationship has on a project’s
evolution has not been studied. Therefore, we set out to explore this by means of a
comparative case study. The next section outlines the research design.
3 Research Design
This section presents the research design of the empirical study following the Goal-
Question-Metric (GQM) approach .
The long term objective of this research is to understand whether there are (and there
likely will be) differences in the maintenance and evolution activities of OSS projects
as long as commercial stakeholders join or drive the development.
This paper addresses the following research questions:
1. Are there differences in the evolution of similar-scoped OSS applications, as long
as one (or more) commercial stakeholders play a major role in the development?
2. When considering projects in the same application domain, are different “cate-
gories” achieving different results or patterns of maintenance?
182 A. Capiluppi, K. Stol, and C. Boldyreff
3. From an effort perspective, do commercial stakeholders behave similarly to indi-
3.3 Method and Metrics
Given the exploratory nature of this topic, we decided to perform an exploratory multi-
ple case study. Since this topic has not been studied in depth, this multiple case study
can be considered as a revelatory case study . Rather than seeking to make gener-
alizations with respect to the inﬂuence of commercial stakeholders in OSS evolution,
we have aimed at exploring this phenomenon with the purpose to identify more precise
hypotheses that can be studied in more depth and with different research methods (e.g.,
The choice of the studied projects was grounded in the fact that they are appropriate
examples of the three types of involvement models mentionedearlier, and was also sup-
ported by the fact that the ﬁrst author was familiar with these projects through previous
studies [8,9,35] as well as ongoing (as of yet unpublished)studies.
Our study is a quantitative analysis of the studied projects, for which two types of
metrics are used: code metrics and effort metrics. These are discussed below.
Code Metrics Given the available (public) releases, a set of data was extracted from
the studied projects: two systems (Eclipse and jEdit) are implemented mostly in Java,
while Moodle is implemented in PHP, and partially relying on OO features, evidenced
by a visible number of PHP classes. The terminology and associated deﬁnitions for
these metrics are extracted from related and well-known past studies, for example, the
deﬁnition of common and control coupling ([2,21,16]).
– Methods (or functions in PHP): the lowest level of granularity of the present analy-
sis. Within this attribute, the union of the sets of OO methods, interfaces, construc-
tors and abstract methods was extracted.
– Classes: as containers of methods, the number of classes composing the systems
has been extracted. Differently from past studies , anonymous and inner
classes  were also considered as part of the analysed systems.
–Size: the growth in size was evaluated in number of SLOCs (physical lines of code),
number of methods, classes and packages.
– Coupling: this is the union of all the dependencies and method calls (i.e.,thecom-
mon and control coupling) of all source ﬁles as extracted through Doxygen2.The
three aggregations introduced above (methods, classes and packages) were consid-
ered for the same level of granularity (the method-to-method,class-to-class and the
package-to-package couplings). A strong coupling link between package A and B
is found when many elements within A call elements of package B.
– Complexity: the complexity was evaluated at the method level. Each method’s
complexity was evaluated via its McCabe index .
2http://www.doxygen.nl, supporting both the Java and PHP languages.
Exploring the Role of Commercial Stakeholders in Open Source Software Evolution 183
Effort Metrics A second set of data was extracted based on the availability of CMS
servers: this data source represents a regular, highly parsable set of atomic transactions
(i.e., ‘commits’) which details the actions that developers (i.e., ‘committers’) perform
on the code composing the system. Two metrics were extracted:
–Effort: the effort of developers was evaluated by counting the number of unique
(or distinct, in a SQL-like terminology) developers in a month.
– Output metrics: the work produced was evaluated by counting the monthly cre-
ations of, or modiﬁcations to, classes or packages. Several modiﬁcations to the
same ﬁle were also ﬁltered with the SQL distinct clause, in order to observe how
many different entities were modiﬁed in a month3.
4 Industry-Led Open Source Project: Eclipse IDE
The Eclipse project has attracted a vast amount of attention by researchers and prac-
titioners, in part due to the availability of its source code, and the openness of its de-
velopment process. Among the recent publications, several have been focused on the
“architectural layer” of this system [34,19], extracting the relevant information from
special-purposed XML ﬁles used to describe Eclipse’s features and extensions (i.e.,plu-
gins) implementing them, in this way representing some sort of “module architecture
As recently reported, the growth of the major releases in Eclipse follows a linearly
growing trend , when studying the evolution of its lines of code, number of ﬁles
and classes. The study on Eclipse’s meta-data indicated that, over all releases, the size
of the architecture has increased more than sevenfold (from 35 to 271 plugins).
The present study is instead performed at the method level, and on two release
streams (trunk and milestones). Regarding Eclipse, 26 releases composing the stream
of “major” and “minor” releases of Eclipse (from 1.0 to 3.5.1) and some 30 additional
releases tagged as “milestones” (M) or “release candidates” (RC), were considered in
this study, spanning some 8 years of evolution. For each release, we performed an anal-
ysis of the source code with the Doxygentool. This latter analysis lasted a few hours for
the early releases, but it required more than one day of parsing for the latest available
releases, mostly due to the explosion in size of the project (490,000 SLOCs found in
the 1.0 release of Eclipse, up to more than 3 million SLOCs found in the 3.6 releases4).
Overall, it required more than one month to perform the analysis on the whole batch of
The remainder of this section presents the results of the analysis of Eclipse. Sub-
section 4.1 presents the results of the evolution of the size of Eclipse. Subsection 4.2
presents the evolution of Eclipse’s complexity.
3In speciﬁc cases, speciﬁc committer IDs were excluded, when it was clear that they are respon-
sible for automatic, uninteresting, commits; it was also excluded from this metric any activity
concerning the ’Attic’ CMS location (which denotes deleted source material).
4Statistics were collected with SLOCCount, http://www.dwheeler.com/sloccount/
184 A. Capiluppi, K. Stol, and C. Boldyreff
4.1 Results – Eclipse Size
This study considered the “main” releases (3.0, 3.1, etc.), and the “milestone” releases
(e.g., 3.2M1, 3.2M2, etc.) and “release candidates” (e.g., 3.3RC1, 3.3RC2, etc.) release
streams of the Eclipse project. The overall growth is almost ﬁvefold, while it is also
evident from Figure 2 that the main stream of releases has a stepwise growth, the steps
being the major releases5.
Fig. 2. Growth in the “main” branch of Eclipse
Major releases of Eclipse are regularly devoted to new features, while milestone and
release candidates releases are devoted to maintaining existing ones (Figure 3). The
milestones stream has a more linear path: plotting the number of methods against the
“build date” of the relative release, a linear ﬁt is found with an appropriate goodness of
ﬁt (R2=0.98). The step-wise growth for the main release stream, and the linear trend
for the milestones release also reﬂect what was found when studying the evolution of
Eclipse at a larger granularity level, i.e. its plugins .
4.2 Results – Eclipse Complexity
The study at the method level shows a distribution of the McCabe cyclomatic indexes
which is constant along the two streams of releases (main and milestones) of Eclipse.
This is visible when assigning the cyclomatic complexity of each method (cci)inthe
four following clusters:
3. 10 ≤cci<15
5The overall size growth has been normalized to 1 for easing the reading of the graph.
Exploring the Role of Commercial Stakeholders in Open Source Software Evolution 185
Fig. 3. Growth and maintenance patterns in the “milestones” branch of Eclipse
Figure 4 shows the relative evolution of the fourth cluster, and reveals a quasi-constant
evolutionary trend (for reason of clarity, the other trends are not displayed, although
they follow a similar evolutionary pattern). The amount of highly complex methods
(cc > 15, ) present in the system never reaches the 2% of the overall system. As
reported in other works, this shows a profound difference from other traditional Open
Source projects, where this ratio (for C and C++ projects) has been observed at around
10% of the system .
Fig. 4. Patterns of highly complex methods (McCabe index >15) in the main branch of Eclipse
186 A. Capiluppi, K. Stol, and C. Boldyreff
4.3 Results – Eclipse Coupling
The number of couplings (i.e., unique method calls) has been counted for each of the
two streams of releases. The set of added,deleted and kept couplings has been evaluated
between two subsequent releases in each stream, and plotted in Figure 5. As shown,
these ﬁndings conﬁrm previous ones  regarding Eclipse’s maintenance patterns: in
the main stream, a large amount of modiﬁcations to its existing connections is made
between minor and major releases, reaching more than 60% of new couplings added
during the transition between the subsequent versions 2.1.3 and 3.0.
Fig. 5. Distribution of coupling in the main branch of Eclipse
On the other hand, the Milestones stream (Figure 6) conﬁrms a recurring pattern,
where the milestones show a great deal of added and removed couplings, whereas the
Release Candidates (RC’s) show a much lower activity in the same activity of coupling
restructurings (the amount of shared couplings between two subsequent releases is not
shown for clarity purposes).
4.4 Results – Eclipse Cohesion
The cohesion of classes or packages was measured by counting the number of elements
connected with other internal elements, and then cumulated for all the classes or pack-
ages. Figure 7 shows the evolution of cohesion at the package level, and it conﬁrms
the observations achieved when evaluating the highly complex methods (Figure 4). Al-
though there is a vast increasein the number of methods and classes, most of the connec-
tions are conﬁned within the same package, keeping the cohesion constant throughout
the life-cycle until the latest observed release. This measurement is also found higher
in the earliest releases (some 73%), and declining sharply until release 3.0, where it
stabilizes to some 69 −70% for the last 6 years.
Exploring the Role of Commercial Stakeholders in Open Source Software Evolution 187
Fig. 6. Distribution of coupling in the milestones branch of Eclipse
Fig. 7. Patterns of cohesion of the two branches of Eclipse
5 Traditional Open Source Project: jEdit
Given the results from the above study, a community-driven OSS project (i.e., where no
commercial company is “sponsoring” the development ) was studied in a similar
way to evaluate and compare in some way the quantitative results of Eclipse. Although
not exactly implementing all the features within Eclipse, the jEdit project also aims to
be a fully-ﬂedged IDE, beneﬁting from a large number of add-ons and plugins, indepen-
dently developed and pluggable in the core system. Though any two software systems
are always different to some degree, this study was not performed for the purpose of
comparing features, but for the sake of observing whether the patterns observed in a
188 A. Capiluppi, K. Stol, and C. Boldyreff
very large and articulated project are similarly found in a much smaller project, and
whether good practices should be inferred in any direction.
Similarly to the Eclipse project, the 14 releases available of jEdit were therefore col-
lected on the largest OSS portal (i.e., SourceForge), from 3.0 to 4.3.1 (earlier releases
do not provide the source code). Being a much smaller project, collecting the informa-
tion via Doxygen was much quicker,both at the beginning of the sequence (57 kSLOCs,
jEdit-3.0) and at the end (190 kSLOCs, jEdit-4.3.1). The 14 considered releases are the
ones made available to the community, and span some 10 years of development.
5.1 Results – jEdit Size
The second system also shows a linear growth, with an adequate goodness of ﬁt (R2=
0.97), albeit with a lower slope than what found in Eclipse, as to summarise a slower
linear growth in Figure 8. A similar linear trend is found in the evolution of methods,
classes and packages. The most evident difference with the evolution of Eclipse is the
pace of the public releases in jEdit: between releases 4.2 and 4.3 some 5 years passed,
although the jEdit conﬁguration management system contains information on the ongo-
ing activity by developers.
Fig. 8. Evolution of size in jEdit
5.2 Results – jEdit Complexity
Regarding jEdit, the evolution of the complexity at the methods’ level brings an inter-
esting insight: for this project, it was found that more than 25% of the methods are
constantly over a threshold of high complexity, at any time of jEdit’s evolution. This
complexity pattern has been observed also in other OSS systems . Large and com-
plex methods are typically a deterrent to the understandability and maintainability of a
software system, and a vast refactoring of these methods has been achieved in the last
two public releases, as visible in the graph, where a signiﬁcant drop of highly complex
methods is achieved even in the presence of a net increase in the number of methods.
Exploring the Role of Commercial Stakeholders in Open Source Software Evolution 189
5.3 Results – jEdit Coupling
The maintenance patterns of jEdit present a more discontinuous proﬁle, with changes
between major releases typically presenting large additions of new couplings (see Fig-
ure 9, bottom), and minor releases where less of such modiﬁcations were made. More
importantly, the maintenance of couplings appears not to be planned, where the largest
modiﬁcations (between 4.2 and 4.3) appear after a long hiatus of ﬁve years, and rep-
resent a full restructuring of the underlying code architecture, with added and deleted
couplings representing three-times and twice as many couplings as the maintained ones,
Fig. 9. Coupling in jEdit
6 Industry-involved OSS Project: Moodle
As per the deﬁnition of an industry-involved OSS project, Moodle’s development is
primarily centered around the OSS community, but various other actors have interest in
its development. A number of organizations across the world are directly contributing
to the development of Moodle by way of funding or contributing their expertise, and
have been deﬁned as “Moodle partners”.
Similarly to the other two case studies, we extracted the size, complexity and cohe-
sion of the PHP code contained in the publicly available releases6: overall we studied
some 90 releases of this project. By checking on the ofﬁcial website, it can be observed
that Moodle was evolved in one single stream of release until version 1.7: from 1.8
onwards, several branches have been evolved at the same time (e.g., 1.7.x, 1.8.x, 1.9.x
etc). For each of these branches we kept the results on size, coupling and complexity
separated from the other branches.
6A list of the releases (with the relative releasing date) since 2002 is available at http://
190 A. Capiluppi, K. Stol, and C. Boldyreff
6.1 Results – Moodle Size
As observed in Eclipse, the evolution of Moodle resembles a step-like pattern (see Fig-
ure 10), where the major releases consist of the addition of a large number of ﬁles,
classes and functions, and theminor releases show smaller additionsin all the measured
metrics. From release 1.8 onwards, all the various branches maintain the same pattern
as well, albeit the growth is intertwined in time with all the other branches (Figures 11
middle and bottom): during the interim releases between minor (e.g., 1.8) and develop-
ment (e.g., 1.8.1) releases, the growth in number of functions, classes and source ﬁles is
minimal, while the step-wise growth pattern is observed between minor releases (e.g.,
between 1.8 and 1.9). Therefore, for this system the increase in size has changed the
approach to development, requiring the project to deﬁne and maintain various branches
at the same time.
Fig. 10. Growth of size in the main branch of Moodle (up to release 1.7)
6.2 Results – Moodle Complexity
Since Moodle is written in the PHP programming language, which is based on pro-
cedural and object-oriented constructs, we evaluated the complexity of the functions
contained in the source code. This was plotted per release, as above, and the percentage
of highly complex functionstracked throughout. The summary in Figure 12 shows how
the excessivecomplexity (i.e., the sum of functions whose McCabe cyclomatic index is
>15, and depicted in the continuous line) has been kept under control even though the
system constantly increases the number of its functions (depicted as a continuous line in
the same ﬁgure). What is quite evident is also the major refactoring that was undertaken
between releases 1.x and 2.x. In the latter, a larger number of functionswere introduced,
in a step-wise growth,while parallel work was done to reduce the amountof complexity
in existing and new functions, with a step-wise descent of highly complex functions.
Exploring the Role of Commercial Stakeholders in Open Source Software Evolution 191
Fig. 11. Growth of size in the parallel branches of Moodle (after release 1.7)
192 A. Capiluppi, K. Stol, and C. Boldyreff
Fig. 12. Evolution of complexity in Moodle (continuous line) and overall increase in number of
6.3 Results – Moodle Coupling
The functions composing the releases of Moodle were also analysed in terms of their
connections, and which of the connections were added or removed between major and
minor releases, and between branches. As done for the previous cases, the releases were
analysed by the Doxygen engine, extracting all the links between low level entities, that
were later lifted to ﬁle-to-ﬁle dependencies.
As reported for the size growth, it becomes clear that the minor and development
releases have become central in Moodle to perform several adjustments, that trail off
in proximity of the next release, similarly to what is found in the Eclipse environment
Fig. 13. Added and removed couplings in Moodle (branches 1.7.x and 1.8.x)
Exploring the Role of Commercial Stakeholders in Open Source Software Evolution 193
(see Figure 13, displaying the 1.7.x and 1.8.x branches of releases). This has evolved in
Moodle: the earlier branches (e.g., Moodle-1.1.x or Moodle-1.2.x) did not display long
sequences of development(e.g., only Moodle-1.1 and Moodle-1.1.1 have been released
within the Moodle-1.1.x branch). With more recent releases, the pattern observed in
Eclipse is also visible in Moodle, with longer sequences of development releases (14
development releases in Moodle-1.8.x, 17 in Moodle-1.9.x), in which fewer and fewer
couplings are added and removed, until the release is being discontinued and not sup-
The two cases of Moodle and jEdit show that similar issues are faced by the develop-
ers: even if companies are involved in development of the Moodle project, they do not
drive the development, as for Eclipse. Given it is taken for granted that industry-led
OSS projects do not have an issue of long-term sustainability, industry-involved and
traditional OSS projects need to address the issue of how to attract and maintain the
existing contributors in the development loop. In the following subsections, we analyse
how effectively developers and contributors are attracted and maintained within the two
projects, and whether lessons learned can be drawn in both cases.
7.1 Contributions on the Periphery
In both the Moodle and jEdit projects, the “core” of the system is separated from
the “plugins” or “contributors” section. We assume that contributing to the “core” of
a project is more time-consuming, and requires more skills, than contributing to the
“modules” or the “plugins” sections7. Therefore we investigated whether a sustained
intake of contributors is achieved in Moodle and jEdit, or whether these projects face
an issue in this respect.
Moodle – Two main directories are found in the CMS server: the core ‘Moodle’ direc-
tory (which makes for the public releases, that we consider as “core”), and the ‘contrib’
folder, organized in ‘plugins’, ‘patches’ and ‘tools’ (but not wrapped in the ofﬁcial re-
leases). As visible in Figure 14 (left), the evolution of the core Moodle system follows
the typical pattern of an early (or ‘cathedral’ ) OSS project: few contributors are vis-
ible in the ﬁrst months (mostly the main Moodle developer), with few other contributors
being active in a discontinuous way. A further, sustained period is also visible, where
the number of active developers follows a growing trend with peaks of over 30 devel-
opers a month contributing, and revealing a ‘bazaar’ phase . The main issue that is
visible in the Moodle “core” system is revealed at around 3/4of its life-cycle, where
the number of active developers start to decline. From the point of view of the sustain-
ability, we posit that this could represent a serious issue in the long-term evolution of
7This is because writing plugins or additional modules, where the system is modular, should be
possible without modifying other ﬁles, but just using the system’s APIs.
194 A. Capiluppi, K. Stol, and C. Boldyreff
On the other hand, the activity of Moodle has been devoted more and more to the
‘contrib’ folder, rather than in the ‘core’: this reﬂects a more and more distributed par-
ticipation to the Moodle development, and a low barrier to entry, albeit not all the con-
tributed modules are selected for inclusion in the publicly available releases. The over-
all distribution of changes throughout the Moodle evolution proceeds on a linear trend
(R2=0.78): in recent months, the inﬂection of productivity in the “core” Moodle has
been balanced by the late growth of contributions to the other parts. That reﬂects a more
and more distributed participation to the Moodle development, and a low barrier to en-
try, but several of the proposedmodules have not been selected for inclusion in the main
Fig. 14. Active monthly contributors in the “core” (left) and in the overall Moodle project (right)
jEdit – The main difference between jEdit and Moodle in the intake of developers is
visible in Figure 15 (below):albeit the ‘core’ (or ‘trunk’) is separated from the ‘plugins’,
few contributors were added in the latter, following a cyclic developmentpattern overall.
Differently from Moodle, the intake of contributors does not follow a linear pattern: the
presence of developers in the “core” declines at around 3/4of the life-cycle, and so
does the number of contributors working on the periphery of the system. This makes
jEdit even more brittle to sustainability issues, speciﬁcally around the intake of new
7.2 Three-Layered Contributions
The study of Moodle as an industry-involved OSS project resulted in an in-depth analy-
sis of the types of contributors who actively produce code for the system. Interesting in-
sights were discovered when studying each developer’s actual contribution to the code:
in a ﬁrst attempt to categorize the intake, the contributions, and the developers leav-
ing the project, three categories are clearly distinguishable, not based on the amount of
effort inputed in the system, but purely on the length of the activity of each developer:
1. Sporadic developers: this refers to the extremely low presence of certain contrib-
utors in the development. Within Moodle, 60 developers have been active for just
one month; other 70 developers have been active between 2 and 6 (not necessarily
Exploring the Role of Commercial Stakeholders in Open Source Software Evolution 195
Fig. 15. Active monthly contributors in the “core” and in the “plugins” parts of jEdit
2. Seasonal developers: as reported recently , most OSS projects beneﬁt seasonal
developers, i.e., those developers who are active for a short period of time (we are
not referring to ‘recurring’ or ‘returning’ developers).
3. Stable developers: those developers showing a sustained involvement (say, more
than 24 months for the Moodle system). Both seasonal and stable developers can
be part of the top 20% developing most of the system, as in the deﬁnition of ’gen-
eration of OSS developers’ given in the past .
Some of the Moodle partners have been found acting as seasonal developers; the Cat-
alyst partner8has so far provided a large number of modiﬁcations to the core Moodle,
by deploying several developers who became active contributors within the commu-
nity. The proﬁle of the contributed outputs is visible in Figure 16 (bottom), and can be
deﬁned as a ‘seasonal’ effort pattern, meaning a large contribution on a very speciﬁc
time interval, and lower levels of effort before and after it. Comparing this curve to a
selection of seasonal Moodleindividual developers(Figure 16, top), a similar pattern is
visible: an initial period of low commit rates, followed by a peak were a high level of
contributions is observed, ﬁnally a leveling-off.
7.3 Limitations of This Study
We are aware of a few limitations of this study, which we discuss below. Yin  lists
four types of threats to validity, namely, construct,internal and external validity, and
Construct Validity. Construct validity is concerned with establishing correct opera-
tional measures for the concepts that are studied . In this study, construct validity
196 A. Capiluppi, K. Stol, and C. Boldyreff
Fig. 16. Output produced by one of the partners (Catalyst, top), as compared to seasonal develop-
ers in Moodle
relates to the measures we have used to collect and analyze the data, namely, code
metrics such as size, coupling and complexity, and effort metrics such as number of
developers and number of modiﬁcations made. We argue that these are well established
metrics that appropriately represent the concepts being studied.
Internal Validity. Internal validity is concerned with establishing a causal relationship.
In our study, the relationship that we have explored is between the nature of the stake-
holders (i.e., commercial versus non-commercial) and the evolution of OSS. The results
of this exploratory study suggest that there is, in fact, an inﬂuence from the presence of
commercial stakeholders. However, changes in evolutionary patterns may not be due to
the involvement of commercial stakeholders. Further research is needed to establish the
nature of this relationship in more detail.
Exploring the Role of Commercial Stakeholders in Open Source Software Evolution 197
External Validity. External validity is concerned with the extent to which ﬁndings
of a study can be generalized to other settings. A common critique of the case study
methodology is that ﬁndings cannotbe generalized. However, the purpose of conducting
case study research is not to look for statistical generalizability, such as aimed for in
large-scale quantitative surveys, but rather to seek theoretical generalization . In
other words, in thispaper we have started to explorea theory relating to the inﬂuence of
commercial stakeholders on the evolution of OSS. We like to emphasize that our study
is of exploratory nature, and as such serves the purpose of exploring our initial ideas
and deﬁning more focused hypotheses for further research.
Reliability. Reliability of a study refers to the degree to which a study can be repeated
and attaining the same results. One strategy to increase a study’s reliability is to estab-
lish an audit trail . Our audit trail consists of the extracted data as well as spread-
sheets that contain the analysis.
8 Conclusions and Future Work
The terminology around the OSS phenomenon has been radically changing in the past
few years. This research has studied how commercial stakeholders can have an inﬂu-
ence on the evolution and maintenance of OSS systems. Eclipse has been studied as
an industry-led OSS system, since it is backed by the IBM corporation; the Java IDE
jEdit was selected as an exemplar of a traditional OSS system; while Moodle was cho-
senasanexemplarofindustry-involved system, built mostly by the OSS community,
although several commercial stakeholders have write-access to it. The public releases
of each system, and their conﬁguration management systems (CMS), were jointly ana-
lyzed, to determine the best type of information to draw results from.
The study of the releases allowed us to focus on the main points along the evolu-
tion of the studied systems. The industry-led OSS system presents several “best prac-
tices” of software engineering: low complexity of units, continuous evolution and reg-
ular maintenance cycles. The traditional OSS system, in the same application domain,
achieves very different results: 1 in 4 units are too complex, discontinuous evolution,
and the maintenance is not regularly achieved. Finally, the industry-involved system
shows more and more regular patterns of evolution, increasing control of complexity
and alignment of its maintenance cycles to multi-branch, large software systems with
parallel maintained releases.
On the other hand, the study of the CMSs allowed the effort of the contributors to
be tracked along the life-cycle of these systems, with the speciﬁc objective of deter-
mining issues in the sustainability of OSS systems. Analysing the industry-led project,
we posit that it does not present (yet) issues of sustainability, as it is backed by a large
corporation. The industry-involved project shows that the amount of active developers
and the output produced follow an increasing, linear trend. Factors for these trends were
found in the increasing number of contributions and plug-ins, and the presence of com-
mercial partners driving the evolution, that act exactly as typical developers, joining in
the projects, producing contributions, and then leaving. As observed, and different from
198 A. Capiluppi, K. Stol, and C. Boldyreff
Eclipse, the studied commercial stakeholder in Moodle is a seasonal contributor, after
some time trailing off and leaving the project.
The study of the effort in the traditional system shows instead that, even with a
sustained number of releases, jEdit has fewer and fewer developers in both the “core”
system as well as in the periphery, showing more issues of sustainability than the other
What these ﬁndings demonstratecould have a profound impact on what is considered
as “Open Source” development and raises the following questions:
–Is the presence of commercial stakeholders a necessary condition to achieve sus-
–Are “traditional” OSS projects eventually destined to trail off and be abandoned?
–Is the lack of adherence to basic software engineering principlesan obstacle to OSS
These are fundamental questions to be answered by further research studies in order to
understand how the OSS phenomenon will change in the future.
Acknowledgments. The authors wish to thank Dr Fernández-Ramil for his extensive
comments on an earlier draft of this paper. We thank the two anonymous reviewers who
provided constructive feedback on this paper. This work was supported, in part, by Sci-
ence Foundation Ireland grant 10/CE/I1855 to Lero—The Irish Software Engineering
Research Centre (www.lero.ie).
1. Aaltonen, T., Jokinen, J.: Inﬂuence in the linux kernel community. In: Feller, J., Fitzgerald,
B., Scacchi, W., Sillitti, A. (eds.) Open Source Development, Adoption and Innovation. IFIP,
vol. 234, pp. 203–208. Springer, Heidelberg (2007)
2. Arisholm, E., Briand, L.C., Foyen, A.: Dynamic coupling measurement for object-oriented
software. IEEE Transactions on Software Engineering 30(8), 491–506 (2004)
3. Basili, V.R., Caldiera, G., Rombach, D.H.: The goal question metric approach. In: Encyclo-
pedia of Software Engineering, pp. 528–532. John Wiley & Sons (1994)
4. Berdou, E.: Insiders and outsiders: paid contributors and the dynamics of cooperation in com-
munity led f/os projects. In: Damiani, E., Fitzgerald, B., Scacchi, W., Scotto, M., Succi, G.
(eds.) Open Source Systems 2006. IFIP, vol. 203, pp. 201–208. Springer, Heidelberg (2006)
5. Bonaccorsi, A., Lorenzi, D., Merito, M., Rossi, C.: Business ﬁrms’ engagement in commu-
nity projects. empirical evidence and further developments of the research. In: Proc. First
International Workshop on Emerging Trends in FLOSS Research and Development. IEEE
Computer Society, Washington, DC (2007)
6. Bonaccorsi, A., Rossi, C.: Contributing to os projects. a comparison between individual and
ﬁrms. In: Proc. 4th Workshop on Open Source Software Engineering (WOSSE), pp. 18–22
7. Bonaccorsi, A., Rossi, C.: Intrinsic motivations and proﬁt-oriented ﬁrms. do ﬁrms practise
what they preach? In: Proc. First International Conference on Open Source Systems, pp. 241–
Exploring the Role of Commercial Stakeholders in Open Source Software Evolution 199
8. Capiluppi, A., Baravalle, A., Heap, N.W.: Engaging without Over-Powering: A Case Study
of a FLOSS Project. In: Ågerfalk, P., Boldyreff, C., González-Barahona, J.M., Madey, G.R.,
Noll, J. (eds.) OSS 2010. IFIP AICT, vol. 319, pp. 29–41. Springer, Heidelberg (2010)
9. Capiluppi, A., Baravalle, A., Heap, N.W.: From “community” to “commercial” ﬂoss – the
case of moodle. In: Proc. Third Workshop on Emerging Trends in Free/Libre/Open Source
Software Research and Development, pp. 11–16. ACM (2010)
10. Capiluppi, A., Fernández-Ramil, J.: Studying the evolution of open source systems at dif-
ferent levels of granularity: Two case studies. In: Proc. 7th International Workshop on Prin-
ciples of Software Evolution (IWPSE), pp. 113–118. IEEE Computer Society, Washington,
11. Capiluppi, A., Michlmayr, M.: From the cathedral to the bazaar: An empirical study of the
lifecycle of volunteer community projects. In: Feller, J., Fitzgerald, B., Scacchi, W., Silitti,
A. (eds.) Open Source Development, Adoption and Innovation. IFIP, vol. 234, pp. 31–44.
Springer, Heidelberg (2007)
12. Capra, E., Francalanci, C., Merlo, F.: An empirical study on the relationship between soft-
ware design quality, development effort and governance in open source projects. IEEE Trans-
actions on Software Engineering 34(6), 765–782 (2008)
13. Capra, E., Francalanci, C., Merlo, F., Rossi Lamastra, C.: A Survey on Firms’ Participation in
Open Source Community Projects. In: Boldyreff, C., Crowston, K., Lundell, B., Wasserman,
A.I. (eds.) OSS 2009. IFIP AICT, vol. 299, pp. 225–236. Springer, Heidelberg (2009)
14. Crowston, K., Wei, K., Howison, J., Wiggins, A.: Free/libre open-source software develop-
ment: What we know and what we do not know. ACM Computing Surveys 44(2) (2012)
15. Dahlander, L., Magnusson, M.G.: Relationships between open source software companies
and communities: Observations from nordic ﬁrms. Research Policy 34(4), 481–493 (2005)
16. Fenton, N.E., Pﬂeeger, S.L.: Software metrics: a practical and rigorous approach. Thomson
17. Guba, E.: Criteria for assessing the trustworthiness of naturalistic inquiries. Educational
Communication and Technology 29(2), 75–91 (1981)
18. Hofmeister, C., Nord, R., Soni, D.: Applied Software Architecture: A Practical Guide for
Software Designers. Addison-Wesley Professional (2000)
19. Hou, D.: Studying the evolution of the Eclipse Java editor. In: ECLIPSE 2007: Proc. OOP-
SLA Workshop on Eclipse Technology Exchange, pp. 65–69. ACM, New York (2007)
20. Igarashi, A., Pierce, B.C.: On inner classes. Information and Computation 177(1), 56–89
21. Li, W., Henry, S.: Object-oriented metrics that predict maintainability. The Journal of Sys-
tems and Software 23(2), 111–122 (1993)
22. Martinez-Romo, J., Robles, G., González-Barahona, J.M., Ortuño-Perez, M.: Using social
network analysis techniques to study collaboration between a ﬂoss community and a com-
pany. In: Russo, B., Damiani, E., Scott Hissam, B.L., Succi, G. (eds.) Open Source Develop-
ment, Communities and Quality, pp. 171–186. Springer, Heidelberg (2008)
23. McCabe, T.J., Butler, C.W.: Design complexity measurement and testing. Communications
of the ACM, 1415–1425 (December 1989)
24. Mens, T., Fernández-Ramil, J., Degrandsart, S.: The evolution of Eclipse. In: Proc. 24th
International Conference on Software Maintenance (ICSM), pp. 386–395 (October 2008)
25. Merlo, E., Antoniol, G., Di Penta, M., Rollo, V.F.: Linear complexity object-oriented sim-
ilarity for clone detection and software evolution analyses. In: Proc. 20th IEEE Interna-
tional Conference on Software Maintenance (ICSM), pp. 412–416. IEEE Computer Society,
Washington, DC (2004)
26. Raymond, E.S.: The Cathedral and the Bazaar. O’Reilly & Associates, Inc., Sebastopol
200 A. Capiluppi, K. Stol, and C. Boldyreff
27. Robles, G., Dueñas, S., González-Barahona, J.M.: Corporate involvement of libre software:
Study of presence in debian code over time. In: Feller, J., Fitzgerald, B., Scacchi, W.,
Sillitti, A. (eds.) Open Source Development, Adoption and Innovation, pp. 121–132.
Springer, Heidelberg (2007)
28. Robles, G., Gonzalez-Barahona, J.M., Herraiz, I.: Evolution of the core team of developers
in libre software projects. In: Proc. 6th IEEE International Working Conference on Mining
Software Repositories (MSR), pp. 167–170 (2009)
29. Rossi, C., Bonaccorsi, A.: Why proﬁt-oriented companies enter the os ﬁeld?: intrinsic vs. ex-
trinsic incentives. In: Proc. 5th Workshop on Open Source Software Engineering (WOSSE).
ACM, New York (2005)
30. Santos Jr, C.D., Kuk, G., Kon, F., Suguiura, R.: The inextricable role of organizational spon-
sorship for open source sustainability. In: Proc. 2nd Workshop Towards Sustainable Open
31. Schaarschmidt, M., von Kortzﬂieisch, H.F.: Divide et impera! the role of ﬁrms in large open
source software consortia. In: Proc. 15th Americas Conference on Information Systems,
32. Shibuya, B., Tamai, T.: Understanding the process of participating in open source commu-
nities. In: Proc. 2nd Workshop on Emerging Trends in Free/Libre/Open Source Software
Research and Development (2009)
33. Wermelinger, M., Yu, Y.: Analyzing the evolution of eclipse plugins. In: Proc. International
Working Conference on Mining Software Repositories (MSR), pp. 133–136. ACM, New
34. Wermelinger, M., Yu, Y., Lozano, A.: Design principles in architectural evolution: a case
study. In: Proc. 24th International Conference on Software Maintenance (ICSM), pp. 396–
35. Wermelinger, M., Yu, Y., Lozano, A., Capiluppi, A.: Assessing architectural evolution: a case
study. International Journal of Empirical Software Engineering, 623–666 (2011)
36. Wermelinger, M., Yu, Y., Strohmaier, M.: Using formal concept analysis to construct and
visualise hierarchies of socio-technical relations. In: Proc. 31st International Conference on
Software Engineering (ICSE), Companion Volume, pp. 327–330. IEEE (2009)
37. Yin, R.K.: Case Study Research: Design and Methods, 3rd edn. SAGE Publications (2003)