Detecting and preventing the architectural roots of bugs

Abstract

Numerous techniques have been proposed to locate buggy files in a code base, but the problem of fixing one bug unexpectedly affecting other files is persistent and prevailing. Our recent study revealed that buggy files are usually architecturally connected by architecture issues such as unstable interfaces and modularity violations. We aim to detect and prevent these architecture issues that are the root causes of defects. Our contributions include (1) a new architecture model, Design Rule Space (DRSpace), that can express structural relations, quality, and evolutionary information simultaneously; (2) a method of automatically extracting defect-prone architecture roots by combining static architecture analysis with software revision history data mining. The preliminary application of our approach to dozens of open source and industry projects has demonstrated its significant potential to inform developers about how software defects should be discovered, examined, and handled.