Automated Program Repair (APR) is an emerging technology which can automatically fix program errors. Decades of research have led to the evolution of APR as search-based, template-based, constraint-based, and learning- based techniques. Learning-based techniques in particular has demonstrated substantial advancements in generating correct patches. However, it continues to face challenges in producing multi-line patches. We identified two fundamental limitations of the existing learning-based APR tools that hinder them from generating multi-line patches.

First, the length of the input sequence of existing APR tools is limited, restricting them from gathering information from compacted code contexts. Second, they fail to capture semantic dependencies among generated patches. We introduce FusionRepair, a transformer-based approach designed to capture more context information from broader contexts and fix bugs by understanding the semantic dependencies between generated patches. For this purpose, we have adapted the Fusion-in-Decoder(FiD) architecture to provide an expanded context. We utilize an iterative program repair paradigm to capture the semantic dependencies among generated patches effectively. Our experiment results with Defects4J v2.0, shows FusionRepair can produce 55 single-line fixes and 28 multi-line fixes, identical to developer patch. Comparison with state of the art tools such as ITER and DEAR shows 35% and 18% improvements respectively. Our results show that FusionRepair has significantly outperformed current state of the art tools in addressing bugs that require multi-line patches.