Codeflaws

The Codeflaws benchmark is a collection of C programs with 3902 defects. Each defect are where the dataset is crawled from Codeforces

Download:

Codeflaws is available for download at tar-link

How to use codeflaws?

1. Install the time command:

sudo apt-get install time

2. Download codeflaws:

wget http://www.comp.nus.edu.sg/~release/codeflaws/codeflaws.tar.gz
tar xf codeflaws.tar.gz

3. Download individual repair tool
4. Select defects (there are in general 3902 defects) to run by creating a file filename that is a copy of the codeflaws/all-script/codeflaws-defect-detail-info.txt file.
5. Modify ./run-version-.sh by setting the appropriate variables. For example, in the run-version-genprog.sh file, you need to modify the following variables:

rundir="$rootdir/genprog-run" # directory in which genprog is called from, a temporary output directory where everything will be copied to during the repair
versiondir="$rootdir/codeflaws" #directory where the codeflaws.tar.gz is extracted
filename="$rootdir/run1" #should be a copy of the codeflaws-defect-detail-info.txt
genprog="/home/ubuntu/genprog-source-v3.0/src/repair"

6. run the script: ./run-version-genprog.sh

Structure of each subject folder

All the subject programs are in the benchmark directory. Each subject folder is named using the following convention: <contestid>-<problem>-bug-<buggy-submisionid>-<accepted-submissionid> Each folder contains:

• Buggy submission with name <contestid>-<problem>-<buggy-submisionid>.c
• Accepted submission with name <contestid>-<problem>-<accepted-submisionid>.c
• Two sets of test scripts:
  1. Repair Test script (test suite given to repair tools for generating repair): test-genprog.sh is for search-based repair tools (GenProg, SPR, Prophet), test-angelix.sh is for Angelix as it requires inserting special instrumentation.
  2. Test script for patch validation (held-out test suite): test-valid.sh is for validating the correctness of patches
• Test input files: input[0-9]+ file used by Test suite (i), and heldout-input[0-9]+ file used by Test suite (ii)
• Test output files: output[0-9]+ file used by Test suite (i), and heldout-output[0-9]+ file used by Test suite (ii)
• Makefile for compiling the buggy submission. This contains the CFLAGS options recommended by Codeforces. To compile the accepted submission, use the command make FILENAME=10-A-13543524
• Makefile.genprog for compiling the buggy submission using cilly. This is for GenProg experiments as GenProg works on CIL representation.
• Test configuration for SPR that specify the name for pass/fail test: <contestid>-<problem>-<buggy-submisionid>.c.revlog

Script for running each repair tool

All the files mentioned below are stored in the all-script directory

Angelix

Use the following files for running Angelix:

• File for running Angelix: run-version-angelix.sh

GenProg

Use the following files for running GenProg:

• File for GenProg general configuration: configuration-default
• File for compilation configuration: compile.pl
• File for running GenProg: run-version-genprog.sh
• File for validating patches generated by GenProg (This script is called from run-version-genprog.sh): validate-fix-genprog.sh

SPR

Use the following files for running SPR:

• File for compilation configuration: code-build.py
• File for test configuration: run-test.py
• File for running SPR: run-version-spr.sh
• File for validating patches generated by SPR (This script is called from run-version-spr.sh): validate-fix-spr.sh

Prophet

Use the following files for running Prophet:

• File for compilation configuration: code-build.py
• File for test configuration: run-test.py
• File for running Prophet: run-version-prophet.sh
• File for validating patches generated by Prophet (This script is called from run-version-prophet.sh): validate-fix-prophet.sh
• Parameter file with learned model (From Prophet original experiment): para-rext-all.out

Citing Codeflaws

If you use Codeflaws in an academic work, we would be really glad if you cite our paper using the following bibtex:

@inproceedings{Tancodeflaws,
  author    = {Tan, Shin Hwei and Yi, Jooyong and Yulis and Mechtaev, Sergey and Roychoudhury,Abhik},
  title     = {Codeflaws: A Programming Competition Benchmark for Evaluating Automated Program Repair Tools},
  booktitle = {2017 IEEE/ACM 39th International Conference on Software Engineering Companion (ICSE-C)}, 
  year      = {2017},
  pages = {180-182}
}

Other information

For more information/questions about the benchmark, refer to the following website: https://codeflaws.github.io/