Rars: a Rust RAR implementation, mostly written by LLMs
4 hours ago
- #AI-Assisted Development
- #Reverse Engineering
- #Rust
- The author reverse-engineered and implemented a RAR compressor in Rust using AI models (OpenAI Codex and Claude Opus) over about 5 weeks at a low cost, producing 55k lines of functional but slow code.
- The project involved creating a specification from various decompressor sources, using AI to document features, and testing with RAR binaries, eventually covering multiple RAR versions including legacy ones.
- AI workflows included using Codex for direct coding from spec, Claude for strategy and reviews, and extensive testing to correct hallucinations and ensure compatibility, though performance optimization was limited.
- Challenges included OpenAI bans due to reverse-engineering content, managing AI blind spots in UX and details, and balancing autonomous work with human oversight to avoid technical debt.
- The result is a free, open-source Rust RAR implementation (rars-cli) that compresses within 5-10% of WinRAR in some cases, available for installation via Cargo.