Canonical and University of Bristol research AI-driven C to Rust code translation

Automated C to Rust Translation Research

Canonical, in partnership with the University of Bristol, is investigating the use of automated tools to rewrite legacy C code into Rust. The primary goal is to achieve safe and maintainable Rust code without altering the original software's behavior. This research addresses the challenge of modernizing existing codebases while preserving functionality.

Testing with Security-Critical Systems

The research is testing the AI translation tool with AppArmor and snap-confine, both of which perform security-critical functions. AppArmor confines applications, and snap-confine creates sandboxed environments for snaps. These components were chosen to evaluate the translation process under high-stakes conditions, where errors could have significant security implications.

Ensuring Behavioral Equivalence

Canonical's proposed system uses language models to generate Rust code, followed by a verification process to detect and correct behavioral differences. The challenge lies not in generating code, but in proving that the translated code behaves identically to the original C code. Researchers plan to combine fuzzing with formal program analysis to identify discrepancies that traditional testing might miss.

Addressing 'Unsafe' Rust Blocks

When mismatches are detected, the system employs symbolic repair to diagnose failures and fix the code. A key focus is to minimize the overuse of 'unsafe' blocks in Rust, which allow operations not permitted in safe Rust and can reintroduce memory-safety risks present in C. The objective is to produce Rust code that is both memory-safe and functionally equivalent to the original C.