The use of large language models (LLMs) to automate the porting of programs from C to Rust is effective and cost-efficient.
Property testing, specifically fuzz testing, was used to ensure the correctness of ported code, comparing the output between C and Rust versions.
The process involves translating C code into Rust using LLMs, generating fuzz tests, and iterating until both versions produce identical results.
While the ported Rust code retains a 'C-like' structure, this approach allows for easy cross-validation and future improvements.
Automation of the porting process is promising but still requires human intervention to handle complex cases and refine LLM outputs.
The study shows that LLM-driven porting could significantly reduce costs and time compared to traditional methods, offering a scalable approach in maintaining large codebases.
Get notified when new stories are published for "🇺🇸 Hacker News English"