Bug Localization in Model-Based Systems in the Wild

The companies that have adopted the Model-Driven Engineering (MDE) paradigm have the advantage of working at a high level of abstraction. Nevertheless, they have the disadvantage of the lack of tools available to perform bug localization at the model level. In addition, in an MDE context, a bug can be related to different MDE artefacts, such as design-time models, model transformations, or run-time models. Starting the bug localization in the wrong place or with the wrong tool can lead to a result that is unsatisfactory. We evaluate how to apply the existing model-based approaches in order to mitigate the effect of starting the localization in the wrong place. We also take into account that software engineers can refine the results at different stages. In our evaluation, we compare different combinations of the application of bug localization approaches and human refinement. The combination of our approaches plus manual refinement obtains the best results. We performed a statistical analysis to provide evidence of the significance of the results. The conclusions obtained from this evaluation are: humans have to be involved at the right time in the process (or results can even get worse), and artefact-independence can be achieved without worsening the results.

Bug Reports and Deep Learning Models

Deep Learning is one of the emerging and trending research area of machine learning in various domains. The paper describes the deep learning approaches applied to the domain of Bug Reports. The paper classifies the tasks being performed for mining of Bug Reports into Bug Report Classification, Bug Localization, Bug Report Summarization and Duplicate Bug Report Detection. The paper systematically discusses about the deep learning approaches being used for the mentioned tasks, and the future directions in this field of research.

Adaptive Ranking Relevant Source Files for Bug Reports Using Genetic Algorithm

Precisely locating buggy files for a given bug report is a cumbersome and time-consuming task, particularly in a large-scale project with thousands of source files and bug reports. An efficient bug localization module is desirable to improve the productivity of the software maintenance phase. Many previous approaches rank source files according to their relevance to a given bug report based on simple lexical matching scores. However, the lexical mismatches between natural language expressions used to describe bug reports and technical terms of software source code might reduce the bug localization system’s accuracy. Incorporating domain knowledge through some features such as the semantic similarity, the fixing frequency of a source file, the code change history and similar bug reports is crucial to efficiently locating buggy files. In this paper, we propose a bug localization model, BugLocGA that leverages both lexical and semantic information as well as explores the relation between a bug report and a source file through some domain features. Given a bug report, we calculate the ranking score with every source files through a weighted sum of all features, where the weights are trained through a genetic algorithm with the aim of maximizing the performance of the bug localization model using two evaluation metrics: mean reciprocal rank (MRR) and mean average precision (MAP). The empirical results conducted on some widely-used open source software projects have showed that our model outperformed some state of the art approaches by effectively recommending relevant files where the bug should be fixed.

Retrieval from software libraries for bug localization

This retrospective on our 2011 MSR publication starts with the research milieu that led to the work reported in our paper. We brie y review the competing ideas of a decade ago that could be applied to solving the problem of identifying the les in a software library related to a query. We were especially interested in nding out if the more complex text retrieval methods of that time would be e ective in the software context. A surprising conclusion of our paper was that the reality was exactly the opposite: the more traditional simpler methods outperformed the complex methods. In addition to this surprising result, our paper was also the rst to report what was considered at that time a large-scale quantitative evaluation of the IR-based approaches to automatic bug localization. Over the years, such quantitative evaluations have become the norm. We believe that these contributions were largely responsible for the popularity of this paper in the research literature.

On Empirically Examining The Effectiveness Of Deep Learning-Based Bug Localization Models

Software Bug Localization involves a significant amount of time and effort on the part of the software developer. Many state-of-the-art bug localization models have been proposed in the past, to help developers localize bugs easily. However, none of these models meet the adoption thresholds of the software practitioner. Recently some deep learning-based models have been proposed, that have been shown to perform better than the state-of-the-art models. With this motivation, we experiment on Convolution Neural Networks (CNNs) to examine their effectiveness in localizing bugs. We also train a SimpleLogistic model as a baseline model for our experiments. We train both our models on five open source Java projects and compare their performance across the projects. Our experiments show that the CNN models perform better than the SimpleLogistic models in most of the cases, but do not meet the adoption criteria set by the practitioners.

On Empirically Examining The Effectiveness Of Deep Learning-Based Bug Localization Models

Software Bug Localization involves a significant amount of time and effort on the part of the software developer. Many state-of-the-art bug localization models have been proposed in the past, to help developers localize bugs easily. However, none of these models meet the adoption thresholds of the software practitioner. Recently some deep learning-based models have been proposed, that have been shown to perform better than the state-of-the-art models. With this motivation, we experiment on Convolution Neural Networks (CNNs) to examine their effectiveness in localizing bugs. We also train a SimpleLogistic model as a baseline model for our experiments. We train both our models on five open source Java projects and compare their performance across the projects. Our experiments show that the CNN models perform better than the SimpleLogistic models in most of the cases, but do not meet the adoption criteria set by the practitioners.