C2SaferRust: Transforming C Projects into Safer Rust with NeuroSymbolic Techniques

This is the code for our paper.

Setup

You need Python 3.12 with Pip, and Rustup. If you want to re-generate our translations, you also need an OpenAI API key. Otherwise, if you'd simply like to compute metrics on our provided translations, you can skip this.

First, install the Rust toolchain and dependencies

rustup install nightly-2022-08-08
rustup component add rust-src rustc-dev llvm-tools-preview

Set the necessary environment variables.

export RUSTUP_HOME="<your_path_here>"
export RUSTFLAGS="-L $RUSTUP_HOME/toolchains/nightly-2022-08-08-x86_64-unknown-linux-gnu/lib"
export LD_LIBRARY_PATH="$LD_LIBRARY_PATH:$RUSTUP_HOME/toolchains/nightly-2022-08-08-x86_64-unknown-linux-gnu/lib"
echo 'OPENAI_API_KEY="<your_key_here>"' > models/.env

Install Python requirements.

pip install -r requirements.txt

Then, install the slicer.

cd slicer
cargo install --debug --locked --path . --force
cd ..

Then, install the metrics tool.

cd ../metrics
cargo install --debug --locked --path . --force
cd ..

Now, we are ready to run translation. You can skip this step if you only want to evaluate our provided translations.

python translate.py --code_dir coreutils/src/cat/rust --test_dir coreutils/tests/cat

In this manner, you can generate translation for all 7 utilities in coreutils, and all 10 utilities in laertes_benchmarks. Note that the Laertes Benchmarks don't come with end-to-end tests, so you can omit the --test_dir flag while generating those translations.

Finally, we evaluate the generated translations.

cd coreutils
python collect_results.py

This writes the results to a json file called coreutils_results.json. To generate Table 2 in our paper, from within the coreutils folder, run

python format_as_table.py

We can also do the same for the Laertes benchmarks to get Table 3.

cd ../laertes_benchmarks
python collect_results.py
python format_as_table.py