Of Change And Software

Staff - Faculty of Informatics

Start date: 1 December 2008

End date: 2 December 2008

Mr. Romain Robbes, December 1st 2008 at 16h00, Auditorium

 

The dissertation committee is composed of:

  • Prof. Michele Lanza, USI Università della Svizzera italiana (advisor)
  • Prof. Mehdi Jazayeri, USI Università della Svizzera italiana (internal member)
  • Prof. Mauro Pezzè, USI Università della Svizzera italiana (internal member)
  • Prof. Stéphane Ducasse, INRIA Lille, France (external member)
  • Prof. Jacky Estublier, CNRS Grenoble, France (external member)
  • Prof. Jean-Marc Jézéquel, INRIA Rennes, France (external member)
  • Prof. Andreas Zeller, University of Saarbrücken, Germany (external member)

 

ABSTRACT:

Software changes. Any long-lived software system has maintenance costs dominating its initial development costs as it is adapted to new or changing requirements. Systems on which such continuous changes are performed inevitably decay, making each maintenance task harder. This problem is not new:  The software evolution research community has been tackling it for more than two decades. However, most approaches have been targeting individual tasks using an ad-hoc model of software evolution.

Instead of only addressing individual maintenance tasks, we propose to take a step back and address the software evolution problem at its root by treating change as a first-class entity. We apply the strategy of reification, used with success in other branches of software engineering, to the changes software systems experience.

Our thesis is that a reified change-based representation of software enables better evolution support for both reverse and forward engineering activities.

To this aim, we present our approach, Change-based Software Evolution, in which first-class changes to programs are recorded as they happen.

We implemented our approach to record the evolution of several systems. We validated our thesis by providing support for several maintenance task. We found that:

  • Change-based Software Evolution eases the reverse engineering and program comprehension of systems by providing access to historical information that is lost by other approaches. The fine-grained change information we record, when summarized in evolutionary measurements, also gives more accurate insights about a system’s evolution.
  • Change-based Software Evolution facilitates the evolution of systems by integrating program transformations, their definition, comprehension and possible evolution in the overall evolution of the system. 

Further, our approach is a source of fine-grained data useful to both evaluate and improve the performance of recommender systems that guide developers as they change a software system.

These results support our view that software evolution is a continuous process, alternating forward and reverse engineering activities that requires the support of a model of software evolution integrating these activities in a harmonious whole.