Beyond the Code: Leveraging Socio-Technical Knowledge to Improve the Performance of Automated Approaches to support Logging Activities

Abstract

Logs are generated through logging statements inserted within software code and serve as a valuable resource for identifying execution anomalies, as well as preventing and debugging issues. Due to the complex nature of logging, several machine learning approaches have been proposed to guide developers on where, what, and at which level to log. However, modern software systems, with long maintenance histories, multi-component structures, and multiple maintaining teams lead to diverse and evolving logging strategies across components, teams, and over time, which complicates ML-based automation efforts. This thesis investigates whether automated approaches for supporting logging activities can be improved by incorporating socio-technical knowledge that exists beyond the code. Using data from large, modern software systems, we study three socio-technical signals: components, ownership, and evolution over time and examine their effect on automated tools that assist developers in selecting the appropriate logging levels. Overall, our work shows that leveraging software engineering knowledge beyond the code can significantly improve the predictive performance of automated tools to support logging activities.