Canonical backs AI-driven C-to-Rust translation PhD

The PhD project responds to a known problem: translating C to Rust often preserves the original structure too literally, resulting in code that compiles but still relies on unsafe operations and awkward idioms. The researchers plan to use LLMs to decompose large codebases into smaller components before rewriting them, a step that The Register's desk suspects may be the hardest part.

The project's focus on mature codebases acknowledges that they contain knowledge that programmers never consciously documented, encoded through years of fixes. A machine translation might preserve this behavior, whereas a clean human rewrite could miss it. This is a key argument from Joel Spolsky's 2000 essay, which the article cites.

AppArmor's relevance is notable because openSUSE 16 replaced it with SELinux last year; a hardened Rust implementation could benefit its remaining users. The Register's desk remains skeptical about viability, but welcomes the investment as a way to produce evidence, rather than hype.