Missing link smell occurs when developers collaborate in source code without communication. This can afect software maintenance by the means of lacking mutual awareness, mistrust and knowledge gap. Existing studies have investigated the relationship of missing link smell with code smell and diferent socio-technical factors like turnover. This study aims to understand how many developers are involved with missing link smell, by calculating the percentage of smelly developers for a project. The study also investigates the relationship between the number of contributions and the number of missing link involvements of a developer. The result shows that the percentage of smelly developers involved with missing link smell is 8.7% on average. The result also suggests a moderate positive correlation between the contribution of a developer to the project and the involvement in smell.
software artifacts have been analyzed in previous
studCommunity smells are the organizational and social anti-ies. S. Magnoni proposed the identification pattern of
patterns in a development community [
Missing link community smell decreases communi- source community [
RQ2: How does missing link smell relate with a developer contribution?
This RQ focuses on the involvement of individual developer in missing link smell. This RQ relates an important characteristic of a developer, i.e., contribution, to missing link smell. This finding will help project managers understanding which type of developers involve more in missing link smell. This information can be used to decide which developers can be monitored to control missing link smell in the community from the beginning of a project. Figure 1: Developer Social Network
In this study, missing link smells are analyzed on seven open-source projects ofApache ecosystem. These projects are selected for being large enough to analyse and thein the defined communication channel, i.e., mailing list. availability of communication data, i.e., mailing list. First,
Two developers are connected through an edge if they the instances of missing link smell are detected in each replied in the same e-mail within a given time frame3[]. project. The missing link smell is identified by finding A communication network is illustrated in Figure3. cases where a collaboration does not have its communi- Missing Link Community Smell: A missing link cation counterpart. Then the developers associated with community smell occurs when a couple of developers each smell are identified by extracting the instance of collaborate with each other but show uncooperative besmell. The fraction of developers involved with missing haviors by not communicating. This smell can be identilink smell is calculated to check whether a subset of de
ifed by detecting collaboration between two developers
velopers are involved with this type of smell. Then thethat do not have the communication counterpart in
decorrelation is investigated between the contribution of ifned communication channel, e.g., development mailing
developers and their involvement in missing link smells.list [
The results of the study show that a small part of the An example of DSN is illustrated in Figur1e. The uptotal developers involved with missing link community per part of the graph represents communication and the smell. On average, 8.7% of the total developers of a projectlower part represents the collaboration among developare involved with missing link smell. This study also finds ers. The developers are connected with a solid line if a significant moderate positive correlation between the they communicate with each other. The developers are developer contribution and their involvement in missing connected to the file icon through a dashed line if they link smell. contribute to that source code file.
The collaboration and communication network can be 2. Background generated separately from this DSN. Figure2 and Figure 3 represent the collaboration and the communication This section provides some important terminologies to network respectively. The missing link smell can be idenbetter understand the missing link community smell. tified comparing the collaboration network with the com
Developer Social Network (DSN): A network of a munication network. There is a link between developer software development community where a node repre- E and F in the collaboration network (Figur2e) but there sents developer and relationships between developers,is no corresponding link between these two developers e.g., communication, coordination, are represented by an in the communication network (Figure 3). DeveloperE edge. and F are collaborating on the same part of source code
Collaboration Network: A specific type of DSN which but they are not connected through any communication
indicates the collaboration in a development community. link. Thus, this is considered as an instance of a missing
Here, a node represents a developer who contributes to link between developerE and F.
the project in the version control system. Two
developers are connected through an edge if they contribute to 3. Related Work
the same part of source code within a given time frame
[
Communication Network: A specific type of DSN
which indicates the communication within the defined
porate the organizational and social aspects of developer
community in software engineering research. Some
studies focused on defining diferent community smells that
communication channel of a development community. can lead to unforeseen project costs1[], [
They found that gender diverse team had a lower number of community smells than non-gender diverse team.
They also showed that gender diversity and women participation were important factors for Black Cloud and Radio Silence whereas organizational Silo and Lone wolf were found partially related.
The existing studies have focused on community smells and the impact of these smells on software artifacts. The phenomenon of community smells is surrounded with developers in a development community. However, developer involvement in missing link smell and the relation between missing link smell and developer contributions have not been investigated yet. So, the developers involved with community smells and how their contribution relate to missing link smell need to be explored.
project are involved in missing link smell. This study also
wants to assess the relationship between a developer’s
smells on diferent software artifacts [
Tamburri et al. first introduced the concept of social
debt in software engineering [
Magnoni proposed the identification pattern of four In this work, seven large open-source projects belonging
out of nine community smells [
Tamburri et al. analysed the distribution of community the development mailing list archives are available on smells in open-source projects 4[]. They also assessed the Gmane2. relation between community smells and existing socio- The selected projects are large enough in terms of technical quality factors, e.g., socio-technical congruence,community members and the number of commits. The communicability, turnover etc. projects have 668 community members on average. All
Palomba et. al examined the relationship between so-the projects have a substantial number of commits, with cial and technical debt2[], [8]. They assessed the impact an average of 10359. Thus the study has enough collaboof community smells on code smells. They found commu- ration and communication data for analysis. nity smells significantly influencing code smell intensity.
They also proposed a community-aware code smell intensity model in which both technical and community 4.2. Missing Link Smell Detection related factors were considered. The selected projects are analysed using a six-month anal
Catolino et al. analysed the role of gender diversityysis window. The analysis period of a project starts from and women participation in community smell 5[]. They when both communication in mailing list and change considered four types of community smell i.e., organi- history in repository are available. A few more months
are excluded to make the analysis period divisible by six To calculate the percentage of smelly developers in a months. The analysis period for each project is given in project, the total number of developers of that project is to detect missing link community smell in this study. 4.4. Correlation Analysis the aforementioned way from project repository and
development mailing list. The tool requires the link ofveloper’s contribution and involvement in missing link source code repository and mailing list archive as input.smell. To address this RQ, the correlation between folThen the tool returns a list of missing link instances for lowing two measures is analysed: each window of the project. A missing link instance is represented by a pair of developers. For example(, ) represents a missing link instance between develop er and .
repository ing link smell
tive form of coding contribution [9]. So, the contribution A developer involved with a missing link smell is consid-of a developer in a project is measured by the number ered as a smelly developer. An instance of missing linkof commits of that developer in the project repository. smell consists of two collaborating developers who doThe number of commits of every individual developer is not communicate with each other. Thus for every missing link smell, there are two smelly developerCs.ODEFACE4SMELLS outputs a missing link instance as a pair be obtained from the list of missing link instances of a of developers. So, the smelly developers can be obtainedproject. First, the developers are extracted from all the by extracting all missing link instances of a project. Themissing link instances of the project. Then the number be the number of elements in
. smelly developers of a project can be denoted by a set . The number of smelly developers of the project will developer occurs in the list.
of involvement is calculated counting how many times a
volvement in smells of a developer are converted into retrieved from the source code repository.
∑=1
× 100%
volvement of a developer in percentage. oper and is the total number of smelly developers.
is the number of commits of devel
each project.
Equation 3 is used to calculate missing link smell in-opers for each project. For example, Apache Cassandra in missing link smells of developer and is the total among 7 communities. Tamburri et. al. found that the is the number of involvement developers (21.1%). This is also the smallest community number of smelly developers.
number of community smell grows quadratically with Finally, the correlation analysis is performed betweenthe number of community members until the threshold involved in missing link community smell?
(2) considered. For every project, the number of total developers and the number of smelly developers are calculated.
×100% (3) community are involved in missing link smellsA.pache project has 1380 total developers and 205 smelly developers which is 14.9% of total developers. It is observed that on average 10.5% of total developers of a software
Cayenne community has the highest percentage of smelly
and
and
individually. Kendall’s tau-b10[] is used to assess the
degree of association between these two variables. Both number of total developers inApache Cayenne
commufor each project
of 200 community members [
These results suggest that only a small portion of dedicates the strength of the correlation. has a range velopers in an open-source software community are inof value from -1.0 to 1.0. As closes to 0, it indicates volved with missing link smells. They do not commuless correlation between two variables. As approaches to -1.0 or +1.0, the strength of correlation between two orative developers. Thus, they contribute to the total variables is increased. The positive value o findicates a number of community smells in a software community.
nicate appropriately with their co-committing or collaba negative correlation between two variables. positive correlation and the negative value o f indicates
This section presents the result analysis and discussion of this study. All the missing link smells found in selected projects are analysed to answer the two research questions. Analysis and discussion for both research questions are provided as follows.
This section discusses the potential threats that may af# Project Name Tau-b p-value fect the validity of this study. 1 Apache Cassandra 0.508 < 0.01 Threats to external validity: Threats to external 2 Apache Cayenne 0.543 < 0.01 validity concern the generalization of the obtained results. 3 Apache CXF 0.528 < 0.01 In this study, seven projects from Apache are analysed. 4 Apache Jackrabbit 0.589 < 0.01 Thus the generalisation requires more projects belonging 5 Apache Jena 0.452 < 0.01 to diferent systems. However, to mitigate this threat 6 Apache Mahout 0.409 < 0.01 large and diverse projects are selected that have a long 7 Apache Pig 0.513 < 0.01 change history - 11 years on average.
Overall 0.612 < 0.01 Threats to internal validity: Threats to internal
validity concern the factors that can influence the result but
are not accounted for. In this study,CODEFACE4SMELLS
example, the correlation coeficient for Apache Cassan- tool is used for the detection of missing link smell. The
dra project is 0.508 and it represents a moderate positive outputs of CODEFACE4SMELLS are directly incorporated
correlation. The value of correlation coeficient is sig- in this study without checking whether there is any
denificant with a p-value less than 0.01. All seven projects fect in the tool. However, the capability of this tool of
of this study show a moderate positive correlation be- identifying missing link smell was evaluated in3][. This
tween number of commits and number of smells which tool is also used in other studies in detecting community
is statistically significant with p<0.01. smells [
Another correlation analysis is performed after com- Moreover, this tool relies on mailing list to detect bining the data from all the projects. The value of thecommunication among developers. But there may excorrelation coeficient is slightly increased to 0.612 but ist other communication channels, e.g., Skype, Facebook still falls under the range of moderate positive correlatione.tc., where developers communicate with each other. The This result is also statistically significant with a p-valueresult can be changed if these communication source are less than 0.01. considered. However, mailing list represents the main
These results suggest that a developer who contributes communication channel for the projects analysed in this
more in a project tends to have more missing link smells. study according to the contribution guidelines of these
This can happen because a developer, who contributes projects. Besides, mailing list is used as the
communicamore, have to communicate more with other develop- tion source in other related studies [
From another point of view, a developer having more 7. Conclusion contribution to a project is likely to be more familiar and This study explores the percentage of developers in a experienced with that project. As he knows most of the software development community involved in missing aspects of that project, he may take the communication link smells. Furthermore, the relationship between develwith co-committers lightly while contributing. However oper contribution and involvement in missing link smell further analysis is required to find out the causes of in- is examined. At first, missing link smells are detected for volving in more smells.
all the projects. Next, the smelly developers are identified by extracting missing link instances. The percentage of and Human Aspects of Software Engineering, IEEE, smelly developers are calculated for every project. The 2013, pp. 93–96. number of appearances of a developer in missing link [7] M. Joblin, W. Mauerer, S. Apel, J. Siegmund, smell is counted. The contribution of a developer to a D. Riehle, From developer networks to verified project is measured by the number of commits. Finally, communities: a fine-grained approach, in: 2015 correlation analysis is done between contribution and IEEE/ACM 37th IEEE International Conference on involvement in smell. Software Engineering, volume 1, IEEE, 2015, pp.
This study analyses seven open-source projects of 563–573.
Apache. The result shows that the number of developers [8] F. Palomba, D. A. Tamburri, A. Serebrenik, A. Zaidinvolved in missing link smells is 8.7% on average. This man, F. A. Fontana, R. Oliveto, Poster: How do study also founds that there is a moderate positive cor- community smells influence code smells?, in: 2018 relation between the number of commits of a developer IEEE/ACM 40th International Conference on Softand the number of involvement in missing link smells. ware Engineering: Companion, IEEE, 2018, pp. The developers who contribute more tend to involve in 240–241. more missing link smell. [9] S. Daniel, R. Agarwal, K. J. Stewart, The efects of di
In future, projects from other systems can be analysed versity in global, distributed collectives: A study of to assess the generalization of the result. Besides, other open source project success, Information Systems types of community smell, e.g., organizational silo, radio Research 24 (2013) 312–333. silence, can be examined to find their association with [10] M. G. Kendall, Rank correlation methods, 1948. developers contribution. [11] F. GIAROLA, Detecting code and community smells in open-source: an automated approach (2018).