AllRepair

AllRepair is a mutation-based repair tool for C programs equipped with assertions in the code.

More details on the repair method behind AllRepair can be found in our paper "Sound and Complete Mutation-Based Program Repair":

https://link.springer.com/chapter/10.1007%2F978-3-319-48989-6_36

##########################################################################################

Building instructions (currently available only for Linux Debian-like distrubutions):

1. Install all pre-requists using:

      sudo apt-get install g++ gcc flex bison make git libwww-perl patch libz-dev python-z3
   

2. Download AllRepair source code using:

      git clone https://github.com/batchenRothenberg/AllRepair.git
   

3. Obtain InGeneer submodule:

      cd AllRepair
      git submodule init python/InGeneer
      git submodule update --remote python/InGeneer
   

4. Build the translation and mutation units (altered CBMC) by running "make" in the AllRepair/src directory.

5. Build the repair unit (altered MARCO), by running "make" in the AllRepair/python/pyminisolvers directory.

##########################################################################################

Usage:

To run AllRepair, go to directory AllRepair/scripts and run the AllRepair.sh script using the following command:

                          ./AllRepair.sh [FileNames] [Options]

For example, running

                          ./AllRepair.sh Examples/ex1.c -m 1 -u 5 -s 1

Will return all ways to reapir program ex1.c (from folder Examples) using mutation level 1, with the unwinding bound set to 5 and the maximum repair size is 1 (only one mutation at a time).

To see the complete list of options and their meaning, run:

                          ./AllRepair.sh -h

Running AllRepair with a direcory name as FileName will result in the individual repair of all .c files in that directory.

AllRepair will respect the following annotation in programs: __CPROVER_assume(..) - adds an assume statement; assert(...) - adds an assertion; Functions whose name begins with "AllRepair_correct_" will not be mutated (alternatively, you may use the --no-mut flag).

##########################################################################################

Interpretting the output:

Output lines beginning with "AllRepair: " and in capital letters are output lines displayed by the AllRepair script itself. These lines should give you an idea of what is going on (translation/repair) and if it is going on succesfully. Other lines are redirected output from the translation and the repair units, including more details.

Once AllRepair finds a possible repair it displays it like this:

      \------------------
      Possible Repair: ...
      In file... 
      Replace ..
      with
      ...
      In file... 
      Replace ..
      with
      ...
      \------------------

Which represents the patch you need to apply to the original program in order to repair it. Note that since mutations are applied for a simplified version of the program, patch instructions sometimes require some translation. For example, if your program contains the instruction i++, a repair might tell you to replace i=i+1 with i=i-1, which corresponds to replacing i++ with i--.

The final line of the output should always contain the result of the repair process, from within the following options: SEARCH SPACE COVERED SUCCESSFULLY - all mutated programs in the search space were examined; ERROR DURING REPAIR - an exception or an out-of-memory error has occured; TIMEOUT - timeout was reached (-t flag); REQUESTED NUMBER OF REPAIRS FOUND - tool has found the requiested number of repairs (-r flag); MAX NUMBER OF MUTATED PROGRAMS INSPECTED - tool has inspected the requested number of programs (-p flag); MAX MUTATION SIZE REACHED - tool has succesfully covered the search space of mutated programs of at most the requested size (-s flag); INTERUPTED - tool was interrupted (e.g., by pressing ctrl+C); EXTERNALLY TERMINATED - tool was terminated externally; ORIGINAL PROGRAM IS CORRECT - the program supplied did not contain a bug, or the bug was undetectable using the given specification and unwinding bound.

Note that AllRepair will not stop once a repair is found unless you'll tell it to using the -r flag. So, the expected result in most cases is reaching some limit (timeout/repair limit/size limit/program limit).

##########################################################################################

Parsing results:

Instead of manually reviewing the output of AllRepair, you can use our script for automatically parsing its results. The script's name is ParseResults.sh and it is located in the scripts directory. To run the script simply use:

      cd scripts
      ./AllRepair.sh [arguments as you please] 2>&1 | ./ParseResults.sh [results_directory]

Output of the script are two files: a csv file with results summarized in a table, and a text file with all found repairs and the elapsed time until each of them was found. The files can be found in the supplied results_directory, or, if ommited, in a directory named AllRepairResults created under the scripts directory. If results_directory was supplied but does not exist, the script will create it.

The filename of the results csv file will be AllRepair_results_<settings>_<current_date_and_time>.csv, where <settings> is a srting created automatically from certain settings of the tool (e.g., if running with mutation level 2, unwinding bound 5 and timeout of 50 sec, <settings> wil equal "m2_u5_t50"). Similarily, the filename of the repair file will be AllRepair_repairs_<settings>_<current_date_and_time>.

Note: do not forget to redirect stderr to stdin using 2>&1, or the script will not be able to parse some of the results.

##########################################################################################

Reproducing experiments:

The instrumented TCAS and QLOSE benchmarks can be found under scripts/Benchmarks. See README in that folder for further instructions on how to reproduce experiments.

##########################################################################################

Interesting parts of code:

File scripts/AllRepair.sh is the script that puts it all together: it runs the translation and mutation unit, passes the results to the repair unit, and parses the output. Cd src contains the altered CBMC code, implementing the translation and mutation unit. The mutation process begins in file goto-symex/symex_target_equation.cpp, in function symex_target_equationt::convert_assignments. Cd python contains the altered MARCO code, implementing the repair unit. Class batMultiProgram in batMultiProgram.py implements the control part: it reads the input (a gsmt2 file), creates formulas for the SAT and SMT solvers, and handles all calls to them. Class MinicardMapSolver in batMapSolvers.py handles the SAT solving. Class Z3SubsetSolver in batSMTsolvers.py handles the SMT solving. Function enumarate in batmarco.py handles the display of results.