Towards python program repair with generative pre-trained transformer (GPT-3.5)

ChatGPT has a great potential using a simple prompt design, which means further studies can be done to investigate the effect of different prompt designs. Complex software often contains hidden bugs in its source code. Recent study suggests that OpenAI’s ChatGPT can perform numerous operations including code-to-code operations like code completion, translation, repair, and summarization, along with language-to-code operations such as code explanation and search. ChatGPT’s dialogue capability can assist in generating more accurate bug fixes. However, it sometimes offers solutions without seeking further information, which can mislead users. To address this issue and enhance user experience, we conducted three design iterations to develop “D-bugger” — a system enabling programmers to fix bugs more effectively. We conducted a survey to gauge the need for refinement, designed a low-fidelity prototype with key features, and then created a high-fidelity prototype evaluated by Python users. Our work aims to enhance the debugging process and user engagement.