A Naive Approach for Automatic Line-level Code Completion (2402.19120v1)

Abstract: Coding is an integral aspect of programming. A programmer can automatically complete a code fragment after writing a few tokens, and the process of automatic completion is known as code completion. Several research studies on code completion have previously been conducted for method body completion and method parameter completion. However, this fundamental study explores the automatic completion of any program statement that might not even be part of a method. The goal is to provide suggestions to the programmer for completing code throughout the codebase by identifying and analyzing code similarities. The proposed methodology can be regarded as a fundamental framework for automated code completion. From the investigation of hundreds of revisions of four subject systems written in C and Java, it is observed that the proposed method can automatically complete around 22% of code statements with an average accuracy of 87% that a programmer writes during development, accelerating software development time. The empirical analysis further demonstrates that the approach can be used with programming language neutrality. The study concludes by illustrating that taking 10 characters as prefixes before invoking completion provides maximum precision.