MIGRATION-BENCH: Repository-Level Code Migration Benchmark from Java 8
Linbo Liu, Xinle Liu, Qiang Zhou, Lin Chen, Yihan Liu, Hoan Nguyen, Behrooz Omidvar-Tehrani, Xi Shen, Jun Huan, Omer Tripp, Anoop Deoras
2025-05-21
Summary
This paper talks about MIGRATION-BENCH, a new tool and dataset designed to test how well AI models can help update old Java code to newer versions across entire software projects.
What's the problem?
Updating large amounts of code from older versions of Java to newer ones is really hard and time-consuming, and there hasn't been a good way to measure how well AI models can handle this complicated task.
What's the solution?
The researchers built a big collection of real Java projects that need to be updated and created a system to test and compare how well different AI models perform these updates, including a new method that gives feedback to help the AI improve.
Why it matters?
This work matters because it helps developers and researchers understand how good AI is at making big code upgrades, which can save a lot of time and reduce mistakes when keeping software up to date.
Abstract
A new benchmark MIGRATION-BENCH evaluates large language models on Java code migration tasks, providing a dataset and framework for repository-level migration assessment.