The US military agency responsible for developing new technologies is set to revolutionize the way old C code is rewritten by funding a new research challenge that aims to create an automated translator capable of converting C code into the security-focused Rust language. This initiative, known as the Translating All C to Rust (TRACTOR) project, is spearheaded by the Defense Advanced Research Projects Agency (DARPA).
DARPA will host a Proposers Day workshop on Aug. 26 to introduce the TRACTOR project and outline its objectives. The main goal of the project is to develop a system that can seamlessly translate C code into idiomatic Rust code, ultimately eliminating memory-safety errors that are a common source of software vulnerabilities in legacy codebases.
Dan Wallach, program manager in DARPA’s Information Innovation Office (I2O), emphasized the importance of automation in rewriting code, citing the high costs and labor-intensive nature of the process. By creating a highly automated system, developers can efficiently convert large volumes of C code to Rust, significantly improving the code quality and reducing vulnerabilities.
Memory-safety flaws, such as buffer overflows and double-free errors, are prevalent in C and C++ code, prompting the transition to Rust as a memory-safe alternative. Companies like Google and Microsoft have already experienced success with Rust, leading to significant improvements in performance and security.
Despite the advantages of Rust, companies must ensure that Rust code is interoperable with existing C or C++ components. Large language models (LLMs) may be necessary to bridge the gap between the two languages, although challenges persist in accurately translating C-to-Rust code due to the limited training data available for Rust.
While AI is not a strict requirement for the project, LLMs are expected to play a crucial role in the solution. Wallach highlighted the rapid pace of AI innovation, stressing the need for flexible solutions that can adapt to advancements in the field.
The development of TRACTOR will present significant challenges, particularly in the realm of intellectual property rights and legal considerations related to AI models. DARPA acknowledges that the project will require groundbreaking innovations in LLM technology and source-code translation, raising complex questions about code ownership and intent interpretation.
In conclusion, DARPA remains committed to solving complex problems through the TRACTOR project, aiming to deliver high-quality Rust code that encourages developers to transition from C. The success of the initiative will rely on advancements in LLM technology and collaborative efforts to address the multifaceted challenges of code translation and security enhancement.