Experiences of debugger-based architecture comprehension

Architecture comprehension is crucial for appropriately maintaining, evolving and reusing large scale software. However, in an actual software development, architecture descriptions are often insufficient, obsoleted or at worst missing, and most of the maintenance tasks are performed without sufficient understanding of the architecture. While many techniques to extract architectural information from existing source code have been proposed to support architecture comprehension, debuggers are still powerful for more exact comprehension of architectures. However, a debugger-based comprehension task usually becomes complicated and cumbersome. In this paper, we study debugger-based manual comprehension processes of real-world architectures. Through the case studies, we found that manual comprehension processes are basically driven by backward tracking of runtime flow of multiple non-primitive objects and such tracking task is cumbersome and time-consuming when only a debugger is available. We expect that the experiences of the case studies will be beneficial to exploring a dynamic analysis approach which can support exact comprehension of architectures.