Mining Common Syntactic Patterns used by Java Programmers

Abstract

Open source code repositories provide massive data as programs that have been used to develop different tools. These kinds of works have been included in the active Big Code and Mining Software Repositories research fields. Although different machine learning works already classify the syntactic constructs used by programmers, there are no reports about the most common syntactic patterns used by Java programmers. In this article, we describe a system we build to provide such a report. Our system retrieves the syntactic patterns used by Java programmers, distinguishing those utilized by experts and beginners. We also present the anomalies found in the usage of different syntactic constructs. We modify the OpenJDK compiler to double the syntactic information included in its Abstract Syntax Tree (AST), define a mechanism to translate ASTs into n-dimensional vectors, combine the information of different syntax constructs to build heterogeneous patterns, and apply the Frequent Pattern Growth algorithm to mine the syntactic patterns as association rules. The mined patterns allow expressing hierarchical subpatterns connected to one another, providing a high level of expressiveness.