Grant: Discovery Launch Supplement

With the rapid evolution of software systems, software maintenance tasks including testing and repair to ensure software quality are indispensable and yet time-consuming. To reduce manual efforts in software maintenance, many automated program generation approaches have been proposed. Given a specification in natural language or an input program, these approaches automatically produce new programs to fulfill the specification or modify the input programs. Automated program generation is essential in (1) automated test program generation, and (2) automated program repair. To automatically test systems that take programs as input (e.g., program analyzers), automated program generation plays the role of producing programs to identify potential bugs in these systems. In automated program repair, automated program generation produces code that serves as patches for fixing bugs. Despite recent advancement in automated program generation techniques, there are several limitations that hinder their wide usage: (1) lack of systematic study that analyzes the characteristics of input programs used and those produced by recent automated program generation techniques for test generation and automated program repair, (2) automated program generation techniques may produce low-quality code (e.g., uncompilable code or programs that misalign with the given specification), (3) the diversity of the programs produced by automated program generation techniques is limited (i.e., they can trigger only limited types of bugs or fix specific types of bugs in automated program repair). The goal of this research is to improve existing automated program generation techniques to address the aforementioned challenges by focusing on three research objectives. RO1: To systematically study the characteristics of programs in automated program generation techniques, this research program will analyze the characteristics of programs in test generation and automated program repair (e.g., grammar coverage, bug-triggering properties, and comparing the characteristics of programs in automated program repair with those in test program generation). RO2: To enhance the quality of generated programs, this research program plans to enhance the current representations of programs used for automated program repair and testing. RO3: To address the lack of diversity in the generated programs, this research program will develop diverse program generation techniques for test program generation and automated program repair. The proposed research will have a significant impact on the software engineering research since it will advance the program generation field by providing technical insights, new datasets, and proposing new automated program generation techniques for effective test generation and automated program repair. Furthermore, the proposed research will train and enable highly qualified personnel (HQP) to contribute to the state-of-the-art software engineering research.