Bun's experimental Rust rewrite hits 99.8% test compatibility on Linux x64 glibc
Bun's experimental Rust rewrite, reportedly powered by AI (Claude), has achieved 99.8% test compatibility in just six days. This rapid progress highlights the astonishing capabilities of LLMs in code translation, sparking intense debate on the future of programming, language ecosystems, and developer roles. The community is buzzing with questions about the implications of this AI-driven development for reliability, maintainability, and the economics of software engineering.
The Lowdown
Jarred Sumner, the creator of Bun, recently announced a significant milestone in the project's experimental Rust rewrite: 99.8% of Bun's pre-existing test suite now passes on Linux x64 glibc. This rapid transformation, reportedly completed in just six days with the help of Anthropic's Claude Mythos LLM, has ignited widespread discussion across the Hacker News community.
Key aspects of the story and its implications include:
- The rewrite targets Rust, moving away from Bun's original implementation in Zig, citing a desire to address memory leaks, crashes, and stability issues.
- The astonishing speed of the rewrite (six days) is attributed directly to the leverage of advanced LLMs, demonstrating a new paradigm for large-scale code migration.
- The effort follows Bun's recent acquisition by Anthropic, leading many to view this as both a technical experiment and a strategic marketing showcase for Anthropic's AI capabilities.
- The high test compatibility (99.8%) highlights the critical role of comprehensive, pre-existing test suites in enabling successful AI-assisted code translation.
The successful, albeit experimental, migration of such a large codebase in a remarkably short period forces a re-evaluation of traditional software development timelines and methodologies, while also raising complex questions about code ownership, maintainability, and the evolving relationship between human developers and AI.
The Gossip
AI's Astonishing Acceleration
Many commenters expressed awe at the speed and scale of the rewrite, completed in mere days with LLM assistance. This ignited discussions about a new era of software development, where access to powerful AI (like Anthropic's Mythos) could create a 'haves and have-nots' divide based on compute and token availability. While some are excited by the potential for rapid innovation and achieving '10x' productivity, others fear the implications for the developer job market and the industry's changing dynamics.
Zig vs. Rust: A Language Lament
The rewrite sparked a fervent debate over language choice. Many argue that Bun's original implementation in Zig was plagued by memory bugs and crashes, making Rust's memory safety a clear technical upgrade. Conversely, Zig proponents countered that Bun's issues were due to poor Zig usage, not inherent language flaws, pointing to other stable Zig projects. There's a strong sentiment among some that Bun's move, especially given its new Anthropic ties and Zig's anti-AI stance, is a 'political' betrayal and damages Zig's reputation as a viable language for large projects.
The 'Vibe-Coded' Quandary
A significant portion of the discussion centered on the maintainability and trustworthiness of 'vibe-coded' or AI-generated software. Concerns were raised about whether any human developer, including the original author, could truly understand such a massive codebase created in such a short time. Questions about code ownership, legal implications of AI-generated code (e.g., public domain), and the long-term viability of a runtime where core components might lack deep human comprehension were prominent. Many commenters expressed skepticism, viewing the project as a marketing stunt for Anthropic rather than a purely technical endeavor.
Test Suite's Crucial Contribution
Commenters widely acknowledged that the rewrite's success was largely enabled by Bun's 'beyond standard extensive and comprehensive test suite.' There was a consensus that LLMs excel at code translation when provided with a robust, verifiable specification and an existing reference implementation. This insight led to speculation that future software development might involve a workflow where developers focus on crafting thorough tests and detailed specifications, then rely on AI to generate or translate the actual code, effectively turning tests into the primary 'spec' for AI agents.