Impact Analysis by means of Unstructured Knowledge in the Context of Bug Repositories

Fixing bugs and implementing enhancements are very relevant activities in a typical software life cycle. They require, as a pre-requisite, the location of a portion of impacted code within a possibly large codebase. This operation can be extremely difficult and time-consuming particularly for developers not much familiar with the software. With that perspective we focus on a simple research question: is it possible to support impact analysis using the information available in software repositories, in particular code comments and version control log? We devised a simple and novel approach, based on Natural Language Processing techniques, that provides support in impact analysis. On the average the proposed approach is very selective with a 99% specificity and achieves a recall of 96% and a precision of 13.6% with respect to a manually built gold standard.