C Code Repository Data For Program Repair Research

Encompassing raw data extracted from GitHub, GitLab, and Bitbucket repositories, used to train models for the CoCoNuT program repair paper. The data is not shuffled or tokenized, and the newest commit in the dataset is from 2005.

Use Cases

• Train context-aware neural translation models for automated program repair using raw C code data.
• Analyze commit patterns and code changes in C language repositories from 2005 for software evolution studies.
• Benchmark ensemble learning methods for source code transformation tasks on this specific corpus of C code.

Strengths

• Data is sourced from multiple major code hosting platforms: GitHub, GitLab, and Bitbucket.
• The dataset is part of a published research effort, providing a defined use case for program repair.

Limitations

• The dataset's temporal coverage is limited, with the newest commit dated 2005, making it potentially outdated for modern code analysis.
• The raw, unprocessed nature of the data (not shuffled or tokenized) requires significant preprocessing for machine learning applications.

Provenance

Source

GitHub, GitLab, and Bitbucket repositories.

Collection Method

Raw data extraction from version control commits.

Time Range

Commits up to and including the year 2005.

Freshness

Dataset was last updated on the platform in September 2023, but the underlying data is from 2005.