EvoRepair: Repurposing EvoSuite for Program Repair

Instroduction

This tool uses EvoSuite so evolve both a patch pool and a test suite to find high-quality patches for program bugs.

Requirements and Setup

Docker (Recommended)

Dockerfile is recommended. Simply run

docker build -t evorepair .
docker run -it -d --name evo evorepair /bin/bash

and EvoRepair will have been set up.

Local Machine

The following dependencies need to be installed:

• python 3.8+, and pip module unidiff
• JDK 8 (only tested on azul JDK 8.0.352)
• Maven 3.6.3
• Ant 1.0.13
• dos2unix
• Defects4J. defects4j needs to be on PATH. Also, defects4j.diff needs to be applied to the Defects4j directory (patch -d /path/to/d4j -p1 -i /path/to/defects4j.diff)

With all the dependencies, run

./setup.sh

Usage

To repair a subject with the default config, run

python3 Repair.py -d --config d4j-subjects/chart_1_instr/config.json

An output directory would be created in output/.

For more parameters, see

python3 Repair.py --help

Experiment Reproduction

Each subject in d4j-subjects was run 10 times, with the random seeds 100, 200, ..., 1000, respectively. The convenience script run.py was used to run the experiments.

To run a subject with a particular seed, e.g., lang_19_instr with seed 500, run

mkdir reproduction
./run.py reproduction lang_19_instr,500

Note that the output directory, in this case reproduction, needs to be created beforehand.

The script can also take multiple tasks, e.g.,

./run.py reproduction chart_1_instr,100 math_2_instr,200

Output Structure

For an example output, see example-output/math_95_instr. The major directories are:

• perfect-patches: hall-of-fame patches that pass both the developer test suite and additionally generated test cases
• plausible-patches: plausible patches that pass the developer test suite
• repair/genX/test-adequate: patches that pass all the tests provided in the X-th generation
• repair/genX/valid: valid patch seeds found in genX, i.e., patches that pass the failing developer test cases but fail any other test case
• statistics/statistics.csv: per-generation statistics of plausible patches, killing tests, etc.

Publication

Evolutionary Testing for Program Repair
Haifeng Ruan, Hoang Lam Nguyen, Ridwan Shariffdeen, Yannic Noller and Abhik Roychoudhury
17th IEEE International Conference on Software Testing, Verification and Validation (ICST) 2024

Acknowledgements

This work was partially supported by a Singapore Ministry of Education (MoE) Tier3 grant “Automated Program Repair”, MOE-MOET32021-0001.

License

This project is licensed under the MIT License - see the LICENSE file for details

Citation

If you use Cerberus in your research work, we would highly appreciate it if you cited the following paper:

@inproceedings{evorepair,
author = {Haifeng Ruan, Hoang Lam Nguyen, Ridwan Shariffdeen, Yannic Noller and Abhik Roychoudhury },
title = {Evolutionary Testing for Program Repair},
year = {2024},
booktitle={Proceedings of the 17th IEEE International Conference on Software Testing, Verification and Validation},
location = {Toronto, Canada},
series = {ICST '24}
}