Multi-level code comprehension model for large scale software, A

Abstract

For the past 20 years researchers have studied how programmers understand code they did not write. Most of this research has concentrated on small-scale code understanding. We consider it necessary to design studies that observe programmers working on large-scale code in production environments. We describe the design and implementation of such a study which included 11 maintenance engineers working on various maintenance tasks. The objective is to build a theory based on observations of programmers working on real tasks. Results show that programmers understand code at different levels of abstraction. Expertise in the application domain, amount of prior experience with the code, and task can determine the types of actions taken during maintenance, the level of abstraction at which the programmer works, and the information needed to complete a maintenance task. A better grasp of how programmers understand large scale code and what is most efficient and effective can lead to better tools, better maintenance guidelines, and documentation.