Advancements in automated program repair: a comprehensive review

Abstract

This review paper presents a comprehensive examination of automated program repair (APR) and its significant contribution to the field of modern software engineering. It elucidates how APR methodologies markedly mitigate manual debugging needs by automating the detection and resolution of software glitches. The study encompasses an in-depth exploration of three primary categories of APR techniques: template-based, machine learning, and deep learning approaches, drawing from an exhaustive evaluation of 41 APR tools. Each category showcases distinct strategies for managing diverse software errors, underscoring the breadth and effectiveness of current APR methodologies. Template-based APR solutions utilize pre-established patterns to efficiently tackle common coding issues, while machine learning-driven approaches dynamically devise repair strategies from historical bug-fix datasets. Deep learning methods extend error rectification boundaries by delving into the semantic context of code, yielding more precise adjustments. The ongoing advancement of APR technologies necessitates researchers to address critical challenges, including the integration of semantic-syntactic analyses, mitigation of data scarcity, optimization of cross-platform tools, development of context-aware approaches, enhancement of fault localization and patch validation processes, and establishment of standardized performance evaluation metrics. This comprehensive analysis underscores the pivotal role of APR in enhancing software efficiency and reliability, representing significant progress in software development and maintenance practices.