Consistency-Preserving Operators for Efficient Product Line Configuration

Replication package

This repository contains all the resources and artifacts of the paper entitled We're Not Gonna Break It! Consistency-Preserving Operators for Efficient Product Line Configuration.

Consistency-preserving configuration operators (CPCOs) are mutation operators that bundle coherent sets of changes in a feature model configuration, specifically, the activation or deactivation of a particular feature together with other (de)activations that are needed to preserve validity.

Project structure

This repository is organized into three parts:

• aCaPulCO: A new search-based optimization tool based on the CPCOs. aCaPulCO implements the IBEA algorithm with custom mutation and crossover operators that apply the CPCOs to a binary vector-based encoding o fconfigurations. CPCOs are represented as transformation rules in the model transformation language Henshin.
• Comparator tools: The two state-of-the-art tools for multi-objective optimization in software product lines: MODAGAME and SATIBEA. This contains the adaptations performed in terms of quality attributes, algorithms and parameters settings, in order to enable a fair comparison between the tools.
• Results: Complete experimentation results.

The image below explains the project structure (editable source).

Requirements

• Java JDK 13+
• Eclipse 2020 (4.18.0)+
• Python 3+ (only for statistical analysis tests)

Execution and experiment replication

1. Clone/Download this repository.
2. Import all projects into Eclipse.
3. Dependencies.
   • Eclipse Henshin is needed. Version 1.7.0 was used. The required plugin is org.eclipse.emf.henshin.model.
   • Eclipse Xtend is needed to build the acapulco.cpcogen project. Version 2.26.0 was used.
4. The entrypoint is located at acapulco.pipeline.PipelineRunnerExternal. Execute it with the following program arguments:

   • To generate all required artifacts for the tools (models, quality attributes,...) including our CPCOs for aCaPulCO:

     -p -fm <feature_model_name>

     Example: -p -fm WeaFQAs (without the extension file).

     This takes the feature model located in the input folder and generates all artifacts in the generated folder, both at the acapulco.pipeline project.

   • To execute the experiments for a specific tool (aCaPulCO, MODAGAME, or SATIBEA):

     -fm <feature_model_name> -t <tool_name> -sc <stopping_condition> -sv <stopping_value> -r <runs>

     where:

     <feature_model_name> is the name of the feature model without the extension file (as in the previous use case).

     <tool_name> can be acapulco, modagame, or satibea.

     <stopping_condition>: use evols for a number of evolutions (default), or time for a specific timeout.

     <stopping_value> is an integer number: in case of evols as stopping condition it represents the number of evolutions (default 50), in case of time it represents the timeout in milliseconds.

     <runs>: an integer specifing the number of runs to be performed.

     Example: -fm WeaFQAs -t acapulco -sc evols -sv 50 -r 30

     This will execute the optimization search and generate raw results in the output folder.

   • Generate the metrics (hypervolume, generational distance, and execution times) and calculate statistics (medians, means, and standard deviations) from the raw results. This last step should be done after executing the search with all tools:

     -fm <feature_model_name> -ptf -r <runs>

     Example: -fm WeaFQAs -ptf -r 30.

     This will generate the final results in the output folder. Two result files are generated for each tool:

     acapulco_weafqas_30runs_results.dat contains the results for each run performed.

     acapulco_weafqas_statsResults.dat contains the summary of the results with statistics.

   • To execute the statistical analysis tests in order to compare the HV-values and the runtime of the three algorithms, use the Python scripts stests.py located at acapulco.statisticstests:

     python stests.py

     This applies the Mann-Whitney U test using the thescipy.stats.mannwhitneyu package of SciPy.org. The script reports p-values, where a value below 0.05 means that the comparison is statistically significant at the 95% confidence level.