Improving Developer Efficiency through Code Reuse

Abstract

Code reuse is an integral part of modern software development, where most software is built using existing software artefacts. Ranging from the copy-pasting of code fragments to the use of third-party libraries, developers frequently turn to the internet to find already-made solutions to difficult programming tasks and save development time. However, the large amount of libraries and code online can make finding the best solution difficult, and reuse is not necessarily straightforward. Most online code snippets do not run, meaning developers need to spend time correcting errors, and when example code snippets are meant to demonstrate API usage, this can present a barrier to using new libraries. This work studies ways to aid developers in the code reuse process, in order to improve their efficiency. We look at ways to more easily connect developers to the wealth of libraries and usage examples online from within their programming environment with our tool for Node.js, Node Code Query (NCQ). We then evaluate how well developers perform compared to the conventional code reuse process and found that developers using our tool solve tasks faster and have to try fewer libraries. Additionally, we study what problems online Node.js code snippets have and how to best correct them automatically, to save developers time in this step of the reuse process. We find that through the combination of the TypeScript compiler’s error detection and codefixes, and our line deletion and custom fixes, we can increase the percentage error-free snippets in our dataset from 26.3% to 74.94%. Finally, we compare the emerging AI code snippet generation and pair programmer technologies to current online code snippet reuse practices, particularly looking at how snippets generated by GitHub’s Copilot extension and those retrieved from Stack Overflow using Google might differ. We find that for the same set of queries, Copilot returned more snippets, with fewer errors and that were more relevant. Ultimately, this work provides further evidence of how automating the code reuse process can improve developer efficiency, and proposes a series of solutions to that end. Additionally, we provide a comparison between existing and emerging reuse processes. As the state of code reuse changes, helping developers understand the strengths of weaknesses of these approaches will become increasingly important.