Fine-Grained Analysis of Similar Code Snippets

Code recommendation aims to help programmers in their coding endeavors by suggesting appropriate code snippets to complete their program. Code recommendation approaches such as code search or code repair may rely on code snippets or code templates extracted from existing projects to provide these suggestions. In this context, extracting and characterizing reusable and recurring code structures beforehand is thus essential. In this paper, we characterize recurring code structures through parametrizable code templates. Code templates can outline the common structure in code snippets along with their variation points, hence providing a convenient way to define their structural similarity. Pattern Structure is a mathematical data analysis framework for organizing objects depending on their similarity: it produces a structure supporting clustering, analysis, and knowledge discovery tasks. We propose an approach leveraging this framework and similarity defined through code templates to highlight and organize groups of similar snippets. The produced structure contains all relevant code templates as well as refinement relationships between them, and can be used to support both manual and automated analysis. We present a case study where we apply this approach to analyze snippets for the task of code sophistication, which consists of identifying and suggesting missing conditional paths in programs.