Discovering undocumented knowledge through visualization of agile software development activities

摘要

In agile software development projects, software engineers prioritize implementation over documentation. Is the cost of missing documentation greater than the cost of producing unnecessary or unused documentation? Agile software engineers must still maintain other software artifacts, such as tickets in an issue tracking system or source code committed to a version control system (VCS). Do these artifacts contain useful knowledge? In this paper, we examine undocumented knowledge in a multi-case exploratory case study of industrial agile software development projects. The first is an internal project with 159 source code commits and roughly 8000 lines of code. The second is an external project with 760 source code commits and roughly 50,000 lines of code. We introduce a ticket-commit network chart (TCC) that visually represents time-series commit activities along with filed issue tickets. We also implement a tool to generate the TCC using both commit log and ticket data. Our case study revealed that software engineers committed source code to the VCS without a corresponding issue ticket in a non-trivial minority of instances. If these commits were based on and linked to individual issue tickets, then these “unissued” tickets would have accounted for a non-trivial minority (5–21%) of the knowledge needed for future software modification and operations. End users and requirements engineers also evaluated the contents of these commits. They found that the omission of links to individual tickets had an important impact on future software modification or operation with between 22 and 49% of these instances resulting in undocumented knowledge.