Program Comprehension Challenges in Software Code Review

Software engineers spend more time understanding code than writing it (with up to 70% of their time being devoted to this). A key activity where developers spend a lot of time reading and understanding code is software code review. Yet, little is known on the challenges concerning program comprehension during software code review. This study provides insight into the types of comprehension challenges that occur during code reviews and their causes. We find that missing design rationale is the most common reason for comprehension challenges in software code review. Comprehension challenges occur most commonly around five topics: “program logic,” “code design,” “defensive coding,” “condition checking,” and “concurrency”. We also show that machine learning (ML) can be used to automatically detect comprehension challenges with 74.3% precision and a 66.7% recall. We discuss potential improvements to code review support tools based on our findings.

These files contain the trained ML algorithms and all the data collected and used for this study.

Data was collected in 2018 with analysis performed in 2018/2019, but completion was delayed due to impacts of COVID19.