Towards learning to detect meaningful changes in software

Abstract

Abstract-Software developers are often concerned with particular changes that are relevant to their current tasks: not all changes to evolving software are equally important. Specified at the language-level, we have developed an automated technique to detect only those changes that are deemed meaningful, or relevant, to a particular development task [1]. In practice, however, it is realised that programmers are not always familiar with the production rules of a programming language. Rather, they may prefer to specify the meaningful changes using concrete program examples. In this position paper, we are proposing an inductive learning procedure that involves the programmers in constructing such language-level specifications through examples. Using the efficiently generated meaningful changes detector, programmers are presented with quicker feedback for adjusting the learnt specifications. An illustrative example is used to show how such an inductive learning procedure might be applied.

From Model-Driven Software Development Processes to Problem Diagnoses at Runtime

“…Earlier we have developed the meaningful change detection tools from the specifications what changes are meaningful to different stakeholders [21], as well as the bidirectional transformation framework to propagate the meaningful changes between the template code and the user modifications [17]. One future direction to consider is how to guide the users to specify such meaningful changes or to learn from examples [18]. -Feedback loops Since boundary between development-time and runtime is disappearing, the distinction between adaptation and evolution in such systems is also getting blurred.…”