Mahal Systems BugsInPy
A Python counterpart to Defects4J for repair agents; useful as a small, reproducible pilot set before building your own.
What it asks the agent to do
- fix a real bug in one of 17 Python projects so its test passes
- reproduce a bug in a pinned environment
What to borrow from it
case template
Caveats
Old and memorized, environments are known to be hard to rebuild, and later papers quote 501 bugs instead of 493, so counts are inconsistent.
Details
| Workloads | coding |
|---|---|
| Interaction | single turn |
| Environment | containers |
| End state checked | Yes |
| Tools | shell, pytest |
| Grading | unit tests |
| Size | 493 |
| Data | real public |
| Personal data | none |
| License | unknown |
| Harness | own repo |
| Status | frozen |