Towards the exploration strategies by mining Mylyns' interaction histories

When developers perform a maintenance task, they always explore the program, i.e., move from one program entity to another. However, even though maintenance is a crucial task, the exploration strategies (ES) used by developers to navigate through the program entities remain unstudied. This lack of study prevents us from understanding how developers explore a program and perform a change task, from recommending strategies to developers, and (ultimately) from critically evaluating a developer's exploration performance. As a first step towards understanding ES, we mined interaction histories (IH) gathered using the Eclipse Mylyn plugin from developers performing a change task on four open-source projects (ECF, Mylyn, PDE, and Eclipse Platform). An ES is defined and characterized by the way (how) the developers navigate through the program entities. Using the Gini inequality index on the number of revisits of program entities, we observe that ES can be either centralized (CES) or extended (EES). We automatically classified interaction histories as CES or EES and performed an empirical study to ascertain the effect of the ES on the task duration and effort. We found that, although an EES requires more exploration effort than a CES, an EES is less time consuming than a CES. Extensive work (number of days spent performing a task) typically imply a CES. Our results show that developers who follow an EES have a methodical investigation of source code while developers who follow a CES have an opportunistic exploration of source code.

Mots clés

Software Maintenance, Program Exploration, Interaction Histories, Exploration Strategies, Mylyn