Experiences in scaling field studies of software developer behavior

Bug reports are widely employed to facilitate software tasks in software maintenance. Since bug reports are contributed by people, the authorship characteristics of contributors may heavily impact the performance of resolving software tasks. Poorly written bug reports may delay developers when fixing bugs. However, no in-depth investigation has been conducted over the authorship characteristics. In this study, we first leverage byte-level $N$-grams to model the authorship characteristics and employ Normalized Simplified Profile Intersection (NSPI) to identify the similarity of the authorship characteristics. Then, we investigate a series of properties related to contributors authorship characteristics, including the evolvement over time and the variation among distinct products in open source projects. Moreover, we show how to leverage the authorship characteristics to facilitate a well-known task in software maintenance, namely Bug Report Summarization (BRS). Experiments on open source projects validate that incorporating the authorship characteristics can effectively improve a state-of-the-art method in BRS. Our findings suggest that contributors should retain stable authorship characteristics and the authorship characteristics can assist in resolving software tasks.

An empirical study of software maintenance tasks was carried out in a large Norwegian organization in the period 1992–93. More than one hundred randomly selected maintenance tasks were studied through interviews with the maintainers performing the tasks immediately before they started the tasks and immediately after they had completed the tasks. The collected data is used to develop distributions and to test 33 hypotheses about software maintenance. The findings from the study indicate, for example, that Application system documentation plays only a minor role as maintenance information source. The maintenance productivity, measured in LOC/effort, is predicted by the size of the task and type of change in the source code, but seems rather independent of language level, maintainer experience, application age and application size. There is an economy of scale for software maintenance tasks. The type of programming language does not correlate with proportion of unexpected maintenance problems, task complexity or task size. The proportion of corrective maintenance decreases with increasing age of the application.

Software maintainers have long been acutely aware of the challenges involved in managing software change processes. Activities such as software migration, restructuring and reengineering all involve source code modification. They rely heavily on analysis and comprehension of the complex system structures and interactions that characterize both legacy and modern software systems. It is widely accepted that tools that support software analysis and maintenance tasks would go a long way towards addressing the constraints that software developers and maintainers work with on a day-to-day basis. Nevertheless, despite an abundance of tools, not many are used by practitioners and very few are considered essential for a particular development or maintenance task. This workshop examined design issues related to the development and application of tools to support software analysis and maintenance

How to identify class comment types? A multi-language approach for class comment classification

A software is developed to satisfy some requirements that could change with respect to technology progress or new clients needs. That is why a software evolves continually in order to fit with occurring requirements. Software evolution or maintenance represents a non-trivial issue that should be treated seriously. In fact, according to some studies software maintenance could consume about 80% of software development total cost. In most of works, such issue was treated in a source code level by proposing new techniques and tools that could optimize maintenance cost. In this paper, we treat the same issue but in an earlier phase of software development and exactly in software design. We propose here an UML profile that allows to software developers and designers to carry about maintenance tasks at the design phase in order to predict handling maintenance tasks before happening, to get an historical trace of these tasks and to express explicitly on models some design patterns helping to resolve software maintenance problems in order to make maintenance easier and cheaper.

Bug or Fault localization is a process of identifying the specific location(s) or region(s) of source code (at various granularity levels such as the directory path, file, method or statement) that is faulty and needs to be modified to repair the defect. Fault localization is a routine task in software maintenance (corrective maintenance). Due to the increasing size and complexity of current software applications, automated solutions for fault localization can significantly reduce human effort and software maintenance cost.We present a technique (which falls into the class of static techniques for bug localization) for fault localization based on a character n-gram based Information Retrieval (IR) model. We frame the problem of bug localization as a relevant document(s) search task for a given query and investigate the application of character-level n-gram based textual features derived from bug reports and source-code file attributes. We implement the proposed IR model and evaluate its performance on dataset downloaded from two popular open-source projects (JBoss and Apache).We conduct a series of experiments to validate our hypothesis and present evidences to demonstrate that the proposed approach is effective. The accuracy of the proposed approach is measured in terms of the standard and commonly used SCORE and MAP (Mean Average Precision) metrics for the task of fault localization. Experimental results reveal that the median value for the SCORE metric for JBoss and Apache dataset is 99.03% and 93.70% respectively. We observe that for 16.16% of the bug reports in the JBoss dataset and for 10.67% of the bug reports in the Apache dataset, the average precision value (computed at all recall levels) is between 0.9 and 1.0.

Most of today's industries, from engineering to agriculture to health, are run on software. In such a context, ensuring software quality play an important role in most of current working environment and have a direct impact in any scientific and technical discipline. Software maintenance and testing have the crucial goal to find or discover possible software bugs (or defects) as early as possible, enabling software quality assurance. However, software maintenance and testing are very expensive and time-consuming activities for developers. For this reason, in the last years, several researchers in the field of Software Engineering (SE) devoted their effort in conceiving tools for boosting developers productivity during such development, maintenance and testing tasks. In this talk, I will first discuss some empirical work we performed to understand the main socio-technical challenges developers face when joining a new software project. I will discuss how to address them with the use of appropriate recommender systems aimed at supporting developers during program comprehension and maintenance tasks. Then, I'll show how Summarization Techniques are an ideal technology for supporting developers when performing testing and debugging activities. Finally, I will summarize the main research advances, the current open challenges/problems and possible future directions to exploit for boosting developers productivity.

Automatic and semi-automatic approaches supporting program comprehension are sought by researchers and practitioners to facilitate software engineering tasks, such as development, maintenance, extension and so on. Using topic modeling is a promising way to automatically discover feature and structure from textual software assets. However, there are gaps between knowing and doing when applying topic modeling to software engineering practice. In this paper, we explored how to infuse topic modeling into understanding Java programs in a generic level, and summarized the whole procedure as a methodology called MAT. MAT utilizes essential information automatically generated from Java projects to establish a project overview and to bring search capability for software engineers. Experiments on two open source Java projects suggest that MAT can support program comprehension for Java software engineers during carrying on software maintenance and extension tasks.

Code clones have been used in many different software maintenance and evaluation tasks in practice (e.g., change proportion and evolution, refactoring, and vulnerability management). There are many clone detection techniques (e.g., text-based, token-based, and AST-based) which can detect code clones not only at the source code-level but also at the compiled artifacts (e.g., IR or binary) level. Unfortunately, there are few studies which thoroughly compare the results of various clone detection setups (a.k.a., different clone detection techniques applied at different artifacts), especially for C-based projects. Therefore, in this paper, we conduct a systematic study to compare the effectiveness of six different code clone detection setups. Each setup, which uses the representative one of the three clone detection techniques: token-based technique (SourcererCC), AST-based (NiCad) technique, and text-based (MSFinder) technique, is applied either at the source code-level or at the LLVM-based IR-level. We conduct our experiments on five C-based open-source systems, Apache, Python, PostgreSQL, FFmpeg, and Linux kernel. Experimental results show that the AST-based technique is better than the token-based and text-based techniques, and clone detection setups performed at the IR-level generally yield higher performance than those performed at the source code-level. The setup of AST-based technique applied at the IR-level, yields the highest performance overall, with an F-score of 84%. However, there is no one setup which can detect all the clones. Through manual qualitative analysis, we have identified ten reasons why certain clones cannot be detected at the IR level or at the source code-level, and two reasons why one of the techniques fails to detect clones. Our findings highlight the usefulness of conducting clone detection under different setups. Furthermore, this study also motivates the need for more application-oriented clone comparison studies.

Chapter 2 - Measuring and Monitoring Technical Debt

Semantic Web was proposed to make the content machine-understandable by developing ontologies to capture domain knowledge and annotating content with this domain knowledge. Although, the original idea of semantic web was to make content on the World Wide Web machine-understandable, with recent advancements and awareness about these technologies, researchers have applied ontologies in many interesting domains. Many phases in software engineering are dependent on availability of knowledge, and the use of ontologies to capture and process this knowledge is a natural choice. This chapter discusses how ontologies can be used in various stages of the system development life cycle. Ontologies can be used to support requirements engineering phase in identifying and fixing inconsistent, incomplete, and ambiguous requirement. They can also be used to model the requirements and assist in requirements management and validation. During software design and development stages, ontologies can help software engineers in finding suitable components, managing documentation of APIs, and coding support. Ontologies can help in system integration and evolution process by aligning various databases with the help of ontologies capturing knowledge about database schema and aligning them with concepts in ontology. Ontologies can also be used in software maintenance by developing a bug tracking system based upon ontological knowledge of software artifacts and roles of developers involved in software maintenance task.

An improving approach to analyzing change impact of C programs

Program comprehension is a required activity for many software development and maintenance tasks. A common means of understanding software is though UML diagrams. UML diagrams model the design and implementation of an application, describing both its structure and behaviour. However with rapid software development life cycles, UML diagrams often become inconsistent with the implementation of an application. This limits their effectiveness in program comprehension activities. This paper presents a program analysis framework to reverse engineer sequence diagrams from application source code. Part of this framework is the ability to filter out irrelevant operations from a sequence diagram in order to simplify the representation of an application. This is achieved by identifying important data, and following its use in the application. The purpose of this framework is to aid in program comprehension activities by providing up to date representations of an application; while also enabling developers to identify the logical operation of a program without interference from irrelevant or supporting operations.

Bug reproduction is a fundamental task in software maintenance and debugging, essential for fault localization, patch generation, and regression testing. However, it remains challenging due to vague descriptions, missing steps, and inconsistent environments in bug reports. These difficulties are amplified in deep learning (DL) frameworks such as TensorFlow, PyTorch, and MXNet, which involve complex computation graphs, hardware dependencies, and rapidly evolving APIs. To address this challenge, we propose ARBRGPT, a prompt-based framework that leverages large language models (LLMs) to automatically reproduce DL bugs from natural language reports. ARBRGPT extracts key entities and builds structured prompts incorporating keyword (KW) abstraction, sequence to reproduce (S2R), chain-of-thought (CoT) reasoning, and few-shot examples. These prompts guide the LLM to generate minimal executable code that faithfully reproduces the reported issue. Evaluated on 155 real-world bug reports from PyTorch, TensorFlow, and MXNet, ARBRGPT achieves an average reproduction success rate exceeding 80 %, significantly surpassing template-based baselines. The results demonstrate that carefully designed prompt components play complementary roles in boosting reproduction performance, offering actionable insights into building more effective LLMbased debugging systems for DL frameworks.

Bug localisation is a core program comprehension task in software maintenance: given the observation of a bug, where is it located in the source code files? Information retrieval (IR) approaches see a bug report as the query, and the source code files as the documents to be retrieved, ranked by relevance. Such approaches have the advantage of not requiring expensive static or dynamic analysis of the code. However, most of state-of-the-art IR approaches rely on project history, in particular previously fixed bugs and previous versions of the source code. We present a novel approach that directly scores each current file against the given report, thus not requiring past code and reports. The scoring is based on heuristics identified through manual inspection of a small set of bug reports. We compare our approach to five others, using their own five metrics on their own six open source projects. Out of 30 performance indicators, we improve 28. For example, on average we find one or more affected files in the top 10 ranked files for 77% of the bug reports. These results show the applicability of our approach to software projects without history.