An IDE-agnostic system to capture reading behaviour of C++ programs using eye-gaze tracker

Over 70% of software development effort is spent in software maintenance comprising bug fixes and version updates. These activities involve fast comprehension of large codebases authored by multiple developers. Developers are mostly trained to write code while in practice they often need to read code. Hence effective and efficient systems are needed to assess the code reading ability of developers. While there have been some studies on program comprehension, they do not deliver a working system. They use an IDE (Integrated Development Environment) that more suits the needs of program comprehension system and caters less to the needs of program development or developers. Importantly, to measure the productivity of a developer, we need to use the same IDE which the developer regularly uses and is comfortable with. In short, such systems need to be IDE-agnostic. In this paper, we build a front-end for a program comprehension assessment system using Tobii X2-30 eye-gaze tracker. The system comprises a flat screen monitor fitted with X2-30 and the developer seated within a specific distance from it. As the developer reads the C++ program, her eye movements are captured by X2-30 at 30 samples / sec. Once the reading is over, the X2-30 data stream is analysed off-line along with the program text to generate the reading sequence of the tokens as read by the developer. We use image analysis to extract regions from the displayed program text corresponding to token lexicons in C++. The lexical tokens are extracted separately from the program text. Finally we stitch the image regions, gaze points in and around the regions, and the tokens to generate the reading streams. We use several heuristics to compress the stream for semantic clarity. Several experiments have been conducted to estimate the accuracy of the generated sequences. In future, we plan to mine these reading sequences for extracting the comprehension features and develop metrics for reading skills.