MiG.4: A Curated Dataset of Library Migrations in Java and Python

MiG.4: A Curated Dataset of Library Migrations in Java and Python Description MiG.4 is a high-quality, manually curated dataset designed to support the development and rigorous evaluation of automated migration tools, particularly those driven by Large Language Models (LLMs). To address the noise often found in automated mining, this dataset provides a robust "ground truth" of real-world, developer-performed code migrations. The dataset spans both Java and Python ecosystems, containing 800 instances of isolated code migrations. Dataset Composition The dataset focuses on four highly relevant library pairs, covering testing, mocking, HTTP requests, and cloud SDKs: Java: JUnit and TestNG Java: Mockito and EasyMock Python: Urllib and Requests Python: Boto and Boto3 Structure and Classification The dataset is organized with one file for each migration case. Each record is richly structured and contains eight specific fields, including: Code snippets before the change. Code snippets after the change. Migration Complexity Classification: Each instance is labeled as either Simple or Complex. This distinction allows researchers to perform fine-grained assessments regarding the semantic and structural transformations required during the migration process. Potential Applications MiG.4 serves as a benchmark resource for: Evaluating LLM performance on code translation and refactoring tasks. Developing novel automated migration solutions. Conducting empirical studies on software evolution and third-party component migration.