Ladybird adopts Rust, with help from AI

The Problem

Ladybird faced a critical challenge in transitioning from C++ to Rust for their LibJS JavaScript engine. The need for a memory-safe language was paramount, especially given the complexities of their existing codebase. The existing reliance on Swift had not matured sufficiently for cross-platform support, prompting the need for a robust solution that could ensure identical functionality while leveraging modern programming paradigms.

The Approach

To tackle this, Ladybird employed AI-assisted programming tools, specifically Claude Code and Codex. The process was not fully autonomous; rather, it was a human-directed initiative where the developer defined the scope of the porting process. This involved determining the order of components to port and the expected structure of the Rust code. The primary goal was to achieve byte-for-byte identical output between the original C++ implementation and the new Rust version, ensuring no regressions in functionality.

Implementation

The porting process began with the lexer, parser, Abstract Syntax Tree (AST), and bytecode generator of LibJS. These components were chosen due to their self-contained nature and the extensive test coverage provided by test262. The developer issued hundreds of targeted prompts to the AI tools, guiding them through the nuances of the translation. Each prompt was crafted to ensure that the AI-generated code adhered to the desired output specifications. The use of a comprehensive conformance testing suite like test262 was crucial, as it allowed for rigorous validation of the output against the original implementation.

Results

The outcome was impressive: approximately 25,000 lines of Rust code were generated in just two weeks, a task that would have traditionally taken several months. The verification process confirmed that every AST produced by the Rust parser matched the C++ version, and all bytecode generated was identical to that of the C++ compiler. This zero-regression outcome not only validated the effectiveness of the AI tools but also demonstrated the potential for significant productivity gains in critical coding projects.

Key Takeaways

1. Utilize AI tools like Claude Code and Codex for code translation tasks to enhance productivity and accuracy.
2. Establish a robust testing framework such as test262 to ensure conformance and reliability during the porting process.
3. Embrace human-directed AI programming to maintain control over the output while leveraging the efficiency of AI-generated code.
4. Focus on self-contained components for initial porting efforts to simplify the transition and validation process.
5. Document the porting process meticulously to facilitate future transitions and knowledge sharing within the team.