ManySStuBs4J: Simple Java Bug Fixes from GitHub

ManySStuBs4J contains single-statement bug fixes mined from open-source Java projects on GitHub, classified into 16 syntactic templates called SStuBs. The dataset has two variants: one mined from 100 Java Maven Projects and another from the top 1000 Java Projects. Bug commits were identified using the SZZ heuristic and keyword filtering in commit messages, achieving an estimated 94% accuracy.

Use Cases

• Evaluating program repair techniques based on single-statement fixes.
• Training models to detect or generate bug fixes using annotated syntactic patterns.
• Analyzing common bug patterns (SStuBs) in Java code from popular projects.
• Benchmarking automated software engineering tools on real-world, simple bug changes.

Strengths

• Bug-fix classification method was evaluated on a sample, achieving 94% accuracy.
• Refactoring changes like variable and function renaming were explicitly detected and excluded.
• Bugs fitting 16 predefined patterns are annotated in a separate JSON file.

Limitations

• Row count, size, and last updated date are unknown across all sources.
• The dataset may include some non-bug changes, as the keyword-based classification is not perfect.
• Metadata completeness is low; specific column names and exact project counts are not provided.

Provenance

Source

Open-source Java projects hosted on GitHub.

Collection Method

Bug-fixing changes collected using the SZZ heuristic, filtered to single-statement changes, and classified via keywords in commit messages.