Undergraduate Samantha McLoughlin and LIVE doctoral student Zach Karas, in partnership with LIVE core faculty member Dr. Yu Huang and collaborators from the University of Notre Dame and Louisiana State University, have had their paper “Programmers’ Visual Attention on Function Call Graphs During Code Summarization“ accepted by the Automated Software Engineering (ASE) conference, one of the top conferences for software engineering. They will present their work later this month in Seoul, South Korea. Their paper was also selected for an ACM SIGSOFT Distinguished Paper Award, one of the highest awards for all papers, given to less than 10% of papers accepted.
Functional Call Graphs as a Meaningful Framework for Studying Project-level Code Comprehension
McLoughlin, Karas and Huang’s work takes an important step forward in understanding how developers extract meaning from real-world software systems, where code is interconnected across many different interconnected functions and files. While current code comprehension research tends to focus on isolated pieces of code with self-contained functionality, their work offers a way to study code comprehension at a project level in order to provide targeted assistance to developers in this context.
Karas & McLoughlin addressed the longstanding shortcoming with an eye-tracking study where programmers summarized pieces of code from large projects. Their team developed novel methods to analyze programmers’ cognitive patterns by relating the specific code they read with an underlying structure called the function call graph, which shows how parts of a codebase are connected, to reveal how programmers integrate information across large codebases. They showed how the function call graph is a meaningful framework for studying project-level code comprehension, and how different strategies for navigating the function call graph related to the quality of participants’ summaries. In particular, they found that traversing more of the function call graph was associated with significant decreases in summary quality, suggesting that “less is more” for finding contextual information from related code. This work paves the way for future research on real-world code comprehension, assisting tool design, as well as Computer Science education.