The problem
Defects4J is the standard benchmark for Java bug studies, but researchers evaluate on it without a clear picture of what kinds of bugs and human patches it actually contains.
The approach
This companion dataset dissects the 395 bugs and their developer patches — patch size, repair actions, and recurring patterns — so the benchmark can be understood, not just run against.
In practice
It gives the community a shared, concrete picture of the most-used Java bug benchmark. Published at SANER 2018.
Highlights
- Characterises 395 bugs and their human patches
- Patch size, repair actions, and patterns
- Published at SANER 2018