Cerberus started as a command line interface tool but with the creation of the Program Repair Competition, we have developed a more advanced configuration module, which would allow for easier orchestration of running various subjects on different tools in different profiles. An example config file config.json is provided in the root directory of the repository to allow for a template cerberus -c config/config.json.
The config file is based on 4 big chunks - general, tasks, profiles.
- The
generalobject contains the configuration options which are specific for the current invocation of Cerberus. - The
taskscontains two objects -defaultwhich is a template for a task (this will be expanded upon later) and thechunksarray, which contains a list of chunks. - The
profilesobject contains two lists - one for the task profiles and another for the container profiles.
A chunk is considered a task type, a list of benchmarks with the specified bug indices or bug subjects and a list of tools with associated extra parameters. For example, in the file config/config_example.json, there are 3 chunks, all of which are for the repair task - the first chunk takes the 28th and 29th bugs of the vulnloc benchmark and runs the extractfix and vulnfix tools. The second chunk only executes extractfix on the 31st, 32nd and 33rd bugs. The last chunk runs the refactory tool on the first three bugs of the refactory benchmark.
As most runs will be the same, the default object is used as the set of configuration parameters, but chunks can override these fields if required. For example, one can add more repair profiles to run the extractfix tool.