Enter the email address you signed up with and we'll email you a reset link. Need an account? Click here to sign up. Download Free PDF. Contributor turnover in libre software projects Open Source Systems, Jesus Gonzalez-Barahona. Gregorio Robles.
A short summary of this paper. Download Download PDF. Translate PDF. A common problem that management faces in software companies is the high instability of their staff. In libre free, open source software projects, the permanence of developers is also an open issue, with the potential of causing problems amplified by the self-organizing nature that most of them exhibit.
Hence, human resources in libre soft- ware projects are even more difficult to manage: developers are in most cases not bound by a contract and, in addition, there is not a real management structure concerned about this problem. This raises some interesting questions with respect to the composition of development teams in libre software projects, and how they evolve over time. There are projects lead by their original founders some sort of "code gods" , while others are driven by several different developer groups over time i.
In this paper, we propose a quan- titative methodology, based on the analysis of the activity in the source code management repositories, to study how these processes developers leaving, developers joining affect libre software projects. The basis of it is the analysis of the composition of the core group, the group of de- velopers most active in a project, for several time lapses.
We will apply this methodology to several large, well-known libre software projects, and show how it can be used to characterize them. In addition, we will discuss the lessons that can be learned, and the validity of our proposal. Keywords: open source, human resources, turnover, mining software repositories 1 Introduction Employee turnover the ratio of the number of workers replaced in a given period to the average number of workers , is known to be high in the proprietary software industry [1].
Most of the attention has been focused on the organizational structure of the In this paper we will use the term "libre software" to refer to any software licensed under terms compliant with the FSF definition of "free software", and the OSI definition of "open source software", thus avoiding the controversy between those two terms. Please use the following format when citing this chapter: Robles, G.
Damiani, E. Gonzalez-Barahona projects, and how developers move to a central position in that structure, but not on how they are replaced when leave it. In this line, probably the best known model about the organizational struc- ture of Hbre software projects is the onion model [2, 3, 4], a visual analogy that represents how developers and users are positioned in communities.
In this model, it is possible to differentiate among core developers those who have a high involvement in the project , co-developers with punctual, but frequent contributions , active users that contribute only occasionally and passive users [8, 7]. But the onion model provides only a static picture of a project, lacking the time dimension that is required for studying joining and leaving processes.
Advancing to fill in this shortcoming, it has been complemented by Ye et al. According to this refinement, a core developer is supposed to go through all the outlying roles, starting as a user, until she eventually reaches the core group. An alternative approach is proposed by Jensen and Scacchi [6], who have studied and modelled the processes of role migration for some libre software communities, focusing on end-users who become developers.
They have found different paths for this process, concluding that the organizational structure of the studied projects is highly dynamic in comparison to traditional software development organizations. With respect to abandonment, it is worth mentioning a study [9] which analyzes how many Debian developers leave the project, and how this affects it i. The authors propose a half-Hfe parameter, defined as the time required for a certain group of contributors to fall to half of its initial population, which is of 7.
Given these precedents, the research goal for the study presented in this pa- per is to gain further understanding of the evolution of libre software developers, and especially of the "core group", those most active.
Therefore, we will study the evolution over time of this core in some libre software projects. We consider at least two possible scenarios: one in which the first core group is highly stable and does not change over time, and one in which the core group "regenerates".
This first case which will be codenamed the code gods scenario assumes that projects rely heavily on their initiators and that their absence would suppose a great loss for, if not the death of, the project. The second scenario assumes that as time passes, the core group changes its composition with some of the initial members leaving the project, but others joining and filling the gap. One of the main goals of this study is to find which of these two approaches is the most common in fibre software projects.
This paper is organized as follows. First, the methodology that has been used to extract information from source code management systems is described.
The next section contains the results of applying the methodology to 21 large libre software projects. Finally, conclusions are drawn and some hints about further research are given. Contributor Turnover in Libre Software Projects 2 Methodology Our methodology is based on mining source code management system reposito- ries, in our case the well-known and widely-used CVS systems.
We analyze the log history of the versioning systems using CVSAnalY [10], a tool that retrieves the information related to every commit to the repository, and inserts it to a database where it can be conveniently analyzed. To identify the "generations" of the core developers, we split the project life-time into ten equally large intervals, which means that intervals will be of different lengths depending on the project, but will have the same relative importance in the context of its history.
Selected projects for this study are all at least three years old, and therefore the minimum length of the time interval is more than three months which is considered to be significative enough, although further research should clarify if this is a correct assumption or not.
For each interval we consider the activity measured in terms of commits to the repository. Therefore, for each project we identify ten different core groups, one per interval. Of course, the composition of the core group in each interval depends on the total number of commiters in that interval.
If only 10 commiters participated in the first interval, the core group would be composed of two persons. If in the last interval the number of participants is 19 commiters, the core group would have 4 members. This means that the core group has always the same relative importance, despite the growth in number of developers in the project.
Some other possibilities for selecting the fraction of commiters that form a core groups, or the duration of the intervals could be considered. Using these other approaches would lead to different definitions of "core group" and "gener- ation". The technique we use is based on visualizing the contribution of the core groups over time. We identify the core group in the first interval, and then plot its contribution not only for the first time interval, but also for all the others.
In the end, we obtain ten curves one per core;group which show the evolution of the contributions for all of them since the beginning to the end of the project. To better understand the visual information that the plots provide, it is important to notice that core groups may have members in common.
In addition, we will discuss the lessons that can be learned, and the validity of our proposal. Documents: Advanced Search Include Citations. Authors: Advanced Search Include Citations. Abstract Abstract. First Monday , 10 2 , February Rishab A. Ghosh and Vipul Ved Prakash.
The orbiten free software survey. First Monday , 5 7 , May Chris Jensen and Walter Scacchi. Modeling recruitment and role migration pro cesses in OSSD projects. Louis, May Stefan Koch and Georg Schneider. Information Systems J , 12 1 —42, CrossRef Google Scholar. Audris Mockus, Roy T. Fielding, and James D. Two case studies of Open Source software development: Apache and Mozilla. Gregorio Robles, Jesus M. Gonzlez-Barahona, and Martin Michlmayr.
0コメント