Abstract Build systems are an essential part of modern software projects. As software projects change continuously, it is crucial to understand how the build system changes because neglecting its maintenance can, at best, lead to expensive build breakage, or at worst, introduce user-reported defects due to incorrectly compiled, linked, packaged, or deployed official releases. Recent studies have investigated the (co-)evolution of build configurations and reasons for build breakage; however, the prior analysis focused on a coarse-grained outcome (i.e., either build changing or not). In this paper, we present BuildDiff, an approach to extract detailed build changes from Maven build files and classify them into 143 change types. In a manual evaluation of 400 build-changing commits, we show that BuildDiff can extract and classify build changes with average precision, recall, and f1-scores of 0.97, 0.98, and 0.97, respectively. We then present two studies using the build changes extracted from 144 open source Java projects to study the frequency and time of build changes. The results show that the top-10 most frequent change types account for 51% of the build changes. Among them, changes to version numbers and changes to dependencies of the projects occur most frequently. We also observe frequently co-occurring changes, such as changes to the source code management definitions, and corresponding changes to the dependency management system and the dependency declaration. Furthermore, our results show that build changes frequently occur around release days. In particular, critical changes, such as updates to plugin configuration parts and dependency insertions, are performed before a release day. The contributions of this paper lay in the foundation for future research, such as for analyzing the (co-)evolution of build files with other artifacts, improving effort estimation approaches by incorporating necessary modifications to the build system specification, or automatic repair approaches for configuration code. Furthermore, our detailed change information enables improvements of refactoring approaches for build configurations and improvements of prediction models to identify error-prone build files.

Source code comments contain key information about the underlying software system. Many redocumentation approaches, however, cannot exploit this valuable source of information. This is mainly due to the fact that not all comments have the same goals and target audience and can therefore only be used selectively for redocumentation. Performing a required classification manually, e.g. in the form of heuristic rules, is usually time-consuming and error-prone and strongly dependent on programming languages and guidelines of concrete software systems. By leveraging machine learning, it should be possible to classify comments and thus transfer valuable information from the source code into documentation with less effort but the same quality. We applied different machine learning techniques to a COBOL legacy system and compared the results with industry-strength heuristic classification. As a result, we found that machine learning outperforms the heuristics in number of errors and less effort.

On question and answer sites, such as Stack Overflow (SO), developers use tags to label the content of a post and to support developers in question searching and browsing. However, these tags mainly refer to technological aspects instead of the purpose of the question. Tagging questions with their purpose can add a new dimension to the identification of discussed topics in posts on SO. In this paper, we aim at automating the classification of SO question posts into seven question categories. As a first step, we harmonized existing taxonomies of question categories and then, we manually classified 1,000 SO questions according to our new taxonomy. Additionally to the question category, we marked the phrases that indicate a question category for each of the posts. We then use this data set to automate the classification of posts using two approaches. For the first approach, we manually analyzed the phrases to find patterns. Based on regular expressions, we implemented a classifier, for each of the categories, that determines whether a post belongs to a category. These regular expressions are derived by analyzing patterns in the phrases. In the second approach, we use the curated data set to train classification models of supervised machine learning algorithms (Random Forest and Support Vector Machines). For the machine learning algorithms, we experimented with 1,312 different configurations regarding the preprocessing of the text and the representation of the input data. Then, we compared the performance of the regex approach with the performance of the best configuration that uses machine learning algorithms on a validation set of 110 posts. The results show that using the regular expression approach, we can classify posts into the correct question category with an average precision and recall of 0.90, and an MCC of 0.68. Additionally, we applied the regex approach on all questions of SO that deal with Android app development and investigated the co-occurrence of question categories in posts. We found that the categories API usage , Conceptual , and Discrepancy are the most frequently assigned question categories and that they also occur together frequently. Our approach can be used to support developers in browsing SO discussions or researchers in building recommender systems based on SO.

Software developers frequently solve development issues with the help of question and answer web forums, such as Stack Overflow (SO). While tags exist to support question searching and browsing, they are more related to technological aspects than to the question purposes. Tagging questions with their purpose can add a new dimension to the investigation of topics discussed in posts on SO. In this paper, we aim to automate such a classification of SO posts into seven question categories. As a first step, we have manually created a curated data set of 500 SO posts, classified into the seven categories. Using this data set, we apply machine learning algorithms (Random Forest and Support Vector Machines) to build a classification model for SO questions. We then experiment with 82 different configurations regarding the preprocessing of the text and representation of the input data. The results of the best performing models show that our models can classify posts into the correct question category with an average precision and recall of 0.88 and 0.87 when using Random Forest and the phrases indicating a question category as input data for the training. The obtained model can be used to aid developers in browsing SO discussions or researchers in building recommenders based on SO.

On Stack Overflow, more than 38,000 diverse tags are used to classify posts. The Stack Overflow community provides tag synonyms to reduce the number of tags that have the same or similar meaning. In our previous research, we used those synonym pairs to derive a number of strategies to create tag synonyms automatically.

The increasing number of posts related to mobile app development indicates unaddressed problems in the usage of mobile APIs. Arguing that these problems result from in- adequate documentation and shortcomings in the design and implementation of the APIs, the goal of this research is to develop and evaluate two developers' issues elimination tools (DIETs) for mobile API developers to diminish the problems of mobile applications (apps) development.After categorizing the problems, we investigate their causes, by exploring the relationships between the topics and trends of posts on Stack Overflow, the app developers' experience, the API and test code, and its changes. The results of these studies will be used to develop two DIETs that support API developers to improve the documentation, design, and implementation of their APIs.

The amount of diverse tags used to classify posts on Stack Overflow increased in the last years to more than 38,000 tags. Many of these tags have the same or similar meaning. Stack Overflow provides an approach to reduce the amount of tags by allowing privileged users to manually create synonyms. However, currently exist only 2,765 synonym-pairs on Stack Overflow that is quite low compared to the total number of tags. To comprehend how synonym-pairs are built, we manually analyzed the tags and how the synonyms could be created automatically. Based on our findings, we then present TSST, a tag synonym suggestion tool, that outputs a ranked list of possible synonyms for each input tag. We first evaluated TSST with the 2,765 approved synonym-pairs of Stack Overflow. For 88.4% of the tags TSST finds the correct synonyms, for 72.2% the correct synonym is within the top 10 suggestions. In addition, we applied TSST to 10 randomly selected Android related tags and evaluated the suggested synonyms with 20 Android app developers in an online survey. Overall, in 80% of their ratings, developers found an adequate synonym suggested by TSST.

While many tutorials, code examples, and documentation about Android APIs exist, developers still face various problems with the implementation of Android Apps. Many of these issues are discussed on Q&A-sites, such as Stack Overflow. In this paper we present a manual categorization of 450 Android related posts of Stack Overflow concerning their question and problem types. The idea is to find dependencies between certain problems and question types to get better insights into issues of Android App development. The categorization is developed using card sorting with three experienced Android App developers. An initial approach to automate the classification of Stack Overflow posts using Lucene is also presented. The study highlights that the most common question types are 'How to?' and 'What is the problem?'. The problems that are discussed most often are related to 'User Interface' and 'Core Elements'. In particular, the problem category 'Layout' is often related to 'What is the problem?' and 'Frameworks' issues often come with 'Is it possible?' questions.