Proactive detection of higher-order software design conflicts

2015-04-08A team of software architects who collaboratively evolve a software model often rely on a copy-edit-merge style version control system (VCS) via which they exchange and merge the individual changes they perform to the model. However, because the current generation of software model VCSs detect conflicts only when architects synchronize their models, the architects remain unaware of newly arising conflicts until the next synchronization, raising the risk that delayed conflict resolution will be much harder. ❧ Collaborative software implementation faces an analogous risk, and there are existing techniques and tools that proactively detect conflicts at the level of source code in order to minimize the conflict unawareness. However, it is challenging to directly apply them as they are to collaborative software design because those are constructed to manage code-level rather than model-level changes. Furthermore, no empirical data is currently available regarding the impact of proactive conflict detection on collaborative design. ❧ In order to address the risk of design conflicts, this dissertation applies proactive conflict detection to collaborative software design. Specifically, this dissertation focuses on higher-order conflicts that do not prevent merging but do violate a system's consistency rules, because higher-order conflicts are generally harder to detect and resolve than synchronization conflicts that are caused by incompatible changes and prevent merging. ❧ This dissertation presents FLAME, an extensible collaborative software design framework that detects the higher-order design conflicts in a proactive way, i.e., before an architect synchronizes her model and finally becomes aware of them. FLAME has an extensible architecture that provides facilities via which the modeling tools and consistency checkers appropriate for the target system's domain can be integrated. FLAME captures modeling changes as they are made, performs a trial merging and conflict detection in the background in order to immediately detect newly arising conflicts, and presents the results to the architects. Also, FLAME explicitly deals with the potentially resource-intensive computations necessary for higher-order conflict detection by parallelizing and offloading the burden to remote nodes. Moreover, by implementing its novel algorithm that prioritizes instances of conflict detection, FLAME guarantees that the outstanding conflicts at a given moment can be detected in a reasonable amount of time even when the available computation resources for conflict detection are scarce. ❧ This dissertation presents the results from two user studies and three systematic experiments on FLAME. The two user studies were conducted involving 90 participants, and the results indicated that the participants who used FLAME were able to create higher quality models in the same amount of time, and to detect and resolve higher-order conflicts earlier and more quickly. The results from the three systematic experiments provided evidence that FLAME minimizes delay in conflict detection, adds only negligible amount of overhead as the number of architects increases, and delivers conflict information early even when the computation resources for conflict detection are limited