Towards the Assisted Decomposition of Large-Active Files

Tightly coupled and interdependent systems inhibit productivity by requiring developers to carefully coordinate their changes, even when modifying subsystems that should be independent of one another. Poor architectural decisions frequently lead to the presence of large, change-prone source files that are at the center of complexes of technical debt. This kind of technical debt quickly incurs interest both through coordination costs and error-proneness. In this paper, we present a new approach for decomposing these ``large-active'' files to pay down critical technical debt. We package our approach as a refactoring recommendation system. Each recommendation is determined by analyzing patterns of co-change and mutual dependency among files. Each recommendation corresponds to a responsibility the large-active file has in relation to the rest of the system. By moving recommended functionality from the large-active file into smaller files, developers can reduce the impact of a debt-laden file and clarify its essential responsibilities. A key advantage of this approach over prior work is that we better focus effort; we avoid spending developer effort refactoring code that is only superficially problematic. We achieve this by incorporating revision history into both determining and ranking recommendations. Each recommendation corresponds to some change-prone responsibility. We present some examples of this approach in action and outline our future plans.