The SearchRepair automated C program repair tool (see
Java C ANTLR Python
Switch branches/tags
Nothing to show
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Failed to load latest commit information.
reminder of common bug

This project demonstrates the feasibility of repairing program with semantic code search. The project contains code and data described in "Repairing Programs with Semantic Code Search" by Yalin Ke, Kathryn T. Stolee, Claire Le Goues, and Yuriy Brun, published in Proceedings of the 30th IEEE/ACM International Conference on Automated Software Engineering (ASE), 2015, pp. 295–306 (doi).


In executors/, there exist several several compiled utilities that are called by the Java program. Update z3, gcc and gdb to point to your local versions/binaries (symlinks should work fine). pathgen and transform are produced by the OCaml code in the supportive associated repository SearchRepair-SymEx (


Let PROJECTROOT denotes the root directory of this project

Several things to setup and rerun:

  1. Firstly, switch to yalin branch for this project.

  2. Secondly, import this project in Eclipse

  3. Install python, mysql, z3, z3 ocaml api, jdk, gcc in local computer.

  4. If folder "./bughunt" exists, delete that folder.

  5. Create a database in mysql, and go to PROJECTROOT/configuration/configuration file, edit three fields user, password, database:




    NOTE: please set the actual values of these three fields in your local computer. Above are the mysql setting in one author yalin's computer.

  6. Go to PROJECTROOT/src/test/, run the main function of that java file, then the experiments will be re run. IMPORTANT: if there were a lot of running time IO exceptions being printing out at console in the beginning, this is probably because C program can not be compiled by GCC in java. This problem might happen a minute after starting rerun experiment. If it did happen, delete "./bughunt" and rerun this experiments. Usually a second rerun will be successful. A successful sign of this is that, the console is printing a list of name of directories, which are the directories of every version in IntroClass folder. After the experiments are completed, a csv file PROJECTROOT/csvlog/log will be generated, which contains all of data presented in the paper.

  7. notice that, there are four fields in the main function of PROJECTROOT/src/test/, which are:

    Field 1: repositoryType. repositoryType=0 means search repository built by linux, repositoryType=1 means search repository built by introclass, repositoryType=2 means search repository by old versions and other students' submissions.

    Field 2: operation. operation=1 means rerun the experiments, operation=0 means generating csvData using existing results. I have backed up my results in case you wants the csv data without rerun the whole project

    Field 3: wb. wb=false, means rerun bb testcases, evaluating on wb. wb=true, means rerun wb testcases, evaluating on bb.

     Field 4: introclasspath. which is the absolute path of introclass benchmark set. When new versions are added
             in introclass set, please empty the folder PROJECTROOT/bughunt, deleting all of subfolders and files 
             in PROJECTROOT/bughunt but keep bughunt folder. Just make it empty. Then set introclasspath to the
             absolute path of the new introclass.

NOTICE: rerun experiments on syllables versions takes a lot of time.