CRUST-Bench is a dataset created for evaluating the transpilation of C code into safe Rust by providing 100 C repositories with Rust interfaces and test cases.
The dataset includes interfaces in safe Rust to ensure adherence to idiomatic, memory-safe Rust patterns, and test cases to enforce functional correctness.
Evaluation of state-of-the-art large language models on CRUST-Bench reveals that generating safe and idiomatic Rust remains a challenging task for current methods.
Improvements on CRUST-Bench are crucial for enhancing transpilation systems that can handle complex scenarios and facilitate the migration of legacy C codebases into memory-safe languages like Rust.