Deco Artifact

Read full text Summarise with VIRAT
Authors: Anonymous
Cited by: 0

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

# DeCo Artifact This repository contains the implementation and experiment artifacts for **DeCo**, a project for detecting Rust memory-safety bugs.It combines:- static extraction of dangerous paths from safe public entries to unsafe operations;- LLM analysis over standard-library unsafe API requirements; and- iterative PoC generation plus Miri confirmation. The repository also includes the scripts and result directories used for the paper's closed-world and open-world evaluation. `rustseek/README.md` contains details the analyzer. ## What Is Open-Sourced In short, the repository includes the source code, runnable experiment scripts, datasets, configurations, and raw results. ## Repository Layout - `rustseek/` - core analyzer built on `rustc_driver`- `experiments/` - datasets, experiment scripts, final entry scripts, and results- `vulns/` - vulnerability sources and advisory metadata used during benchmark construction- `keys.md` - local API key file used by the experiment runners ## Main Data and Results - `experiments/dataset` - closed-world and open-world benchmark- `rustseek/data` - atomic safety obligations database- `experiments/final_scripts/` - final scripts used to run the main experiments and baselines- `experiments/results/` - experiment outputs ## Environment DeCo uses:- `nightly-2025-10-09`- components: `rustc-dev`, `rust-src`, `llvm-tools-preview`, `miri` Suggested setup: ```bashrustup toolchain install nightly-2025-10-09rustup component add --toolchain nightly-2025-10-09 rustc-dev rust-src llvm-tools-preview miri``` Build the core `rustseek` analyzer: ```bashcd rustseekcargo +nightly-2025-10-09 build --release``` The main binary is `rustseek/target/release/cargo-rustseek`. ## API Keys Experiment runners load keys from the local `keys.md` file.Current model profiles are configured in `experiments/configs/model_profiles.json`.The expected key labels are:- `deepseek`- `gpt-5.2`- `claude-sonnet-4-5` `keys.md` is local-only and not versioned. ## Reproducing the Main Experiments Run from the repository root. ### 1. Closed-world main run ```bashbash experiments/final_scripts/run_gpt_rs_full.sh``` Other retained runs are in the same directory, including:- `run_gpt_rs_no_contract.sh`- `run_gpt_rs_no_rules_gates.sh`- `run_gpt_rs_no_context.sh`- `run_gpt_rs_no_ctxreq.sh`- `run_deepseek_rs_full.sh`- `run_claude_rs_full.sh`- `run_gpt_only.sh`- `run_deepseek_only.sh`- `run_claude_only.sh`- `run_safe4u_gpt.sh`- `run_rudra.sh`- `run_lockbud.sh` ### 2. Recompute closed-world tables ```bashpython3 experiments/scripts/evaluate_dataset_labeled_standard.pypython3 experiments/scripts/evaluate_dataset_labeled_safe4u.py``` ## Notes on Reproducibility - The final scripts write into fixed directories under `experiments/results/`.- The experiment entry points are documented in `experiments/final_scripts/README.md`.

Fields of Study & Subjects

Software Testing and Debugging Techniques Software Engineering Research Security and Verification in Computing

Publication details

Journal / source
Zenodo (CERN European Organization for Nuclear Research)
Publication date
27 March 2026
Language
EN