Skip to content

Latest commit

 

History

History
132 lines (92 loc) · 2.83 KB

ARTIFACT.md

File metadata and controls

132 lines (92 loc) · 2.83 KB

LIBERATOR Artifact – Replication Guide

This guide summarizes how to reproduce the experiments from the paper:
"Liberating libraries through automated fuzz driver generation"

Requirements

Hardware

  • At least 64 GB of RAM
  • Minimum 4 TB of disk space
  • Multi-core server recommended (~2–3 days runtime)

Software

  • Ubuntu 20.04
  • Ability to run Docker
  • Internet connection

Dependencies

  • curl, git, docker, pip
  • Python dependencies: 
    ./preinstall.sh

Installation

# Clone the repository
git clone https://github.com/HexHive/liberator.git
cd liberator

# Switch to the artifact tag
git checkout fse-25-artifact

# Install Python dependencies (on the host)
pip install -r requirements_host.txt

All experiments run in Docker containers. Local runs are not supported.

Experiment 1 (E1): Trade-off Analysis

  • Goal: Measure the impact of driver generation time (tgen) vs test time (ttest)
  • Claim supported: (C1)
  • Time: 2 min human, ~60 compute-hours

Steps

# Optional: Install static analysis results (faster)
./install_analysis_result.sh

# Run all 4 campaigns
./run_campaign_artifact.sh

# Generate results
./fig3.sh
./tab4.py

Output

  • fig3.png (matches Figure 3 in the paper)
  • Table similar to first 5 columns of Table 4

Experiment 2 (E2): Library Exploration

  • Goal: Evaluate how LIBERATOR explores library internals
  • Claim supported: (C2)
  • Time: 2 min human, ~10 compute-hours
  • Depends on: E1 results

Steps

./fig4.sh
./fig5.sh

Output

  • edge_coverage.png (per library)
  • api_function_coverage.png (per library)

Experiment 3 (E3): Ablation Study

  • Goal: Measure contribution of each module in LIBERATOR
  • Claim supported: (C3)
  • Time: 2 min human, ~24 compute-hours
  • Depends on: E1 results

Steps

./tab9.sh
./tab10.sh

Output

  • Tables similar to Table 9 and Table 10 from the paper

Major Claims Supported

  • (C1) LIBERATOR captures the trade-off between creating and testing drivers
  • (C2) LIBERATOR diversifies coverage across libraries
  • (C3) LIBERATOR’s design is modular and effective (confirmed by ablation)

Resources