Abordagem para seleção de exemplos de código de sistemas distribuídos para a criação de exemplos trabalhados de engenharia de software

Abstract

This study investigates the evolution of open-source Distributed Systems (DS) projects, including Apache Kafka and ZooKeeper, which aims at selecting code snippets as candidates for creating worked examples that demonstrate important concepts of Software Engineering (SE) and DS, applicable to the teaching of these subjects. To achieve this, the tool called Worked-Example-Miner (WEM) was developed, implementing a heuristic that applies static code quality metrics, such as coupling and complexity, as well as detecting refactoring patterns that indicate improvements in software design. The application of the heuristic to various open-source repositories resulted in the identification of promising candidates for the creation of worked examples. The integration with the Artificial Intelligence (AI) model Google Gemini allowed the identification of candidates addressing specific DS topics, such as distributed consensus and fault tolerance. The results demonstrate that the developed heuristic successfully selects several dozen candidates within the analyzed projects. Considering that the viability of such a heuristic depends on selecting a reduced and relevant number of candidates for expert analysis, this figure represents a reduction in the number of candidates by up to 99.760%, with an average of 95.740% and a small standard deviation of 3.877% relative to the total number of classes in a repository, with thousands of files and tens of thousands of changes, demonstrating that the developed heuristic yields promising results. The evaluation of the candidates by experts in DS and SE confirmed the relevance of the selected code segments for educational purposes.