Quoting Rick Brewster
Paint.NET now includes a from-scratch, clean-room reverse-engineered rewrite of Direct2D (in `PaintDotNet.Windows.Direct2D1.Managed.dll`) to enable WINE compatibility, triggered via the `/wine` flag The entire 180,000-line implementation was generated by Claude, which the author describes as "vibe coded" — largely unreviewed and trust-based due to the sheer volume Claude exhibited highly variable quality: at times producing genius-level work (especially in reverse-engineering Direct2D effect lib
Analysis
TL;DR
- Paint.NET now includes a from-scratch, clean-room reverse-engineered rewrite of Direct2D (in
PaintDotNet.Windows.Direct2D1.Managed.dll) to enable WINE compatibility, triggered via the/wineflag - The entire 180,000-line implementation was generated by Claude, which the author describes as "vibe coded" — largely unreviewed and trust-based due to the sheer volume
- Claude exhibited highly variable quality: at times producing genius-level work (especially in reverse-engineering Direct2D effect library formulas), but also making significant errors like omitting COM
AddRef()calls for reference-counted objects - Direct2D has long been the biggest obstacle for running Paint.NET on WINE, and since the feature cannot simply be disabled, this rewrite was the only viable path forward
- The author had to actively supervise Claude's output, correcting bad design decisions and resource management issues throughout the process
Why It Matters
This is a notable real-world case study of an AI model generating a massive, complex systems-level codebase (180K lines) from scratch, demonstrating both the extraordinary capability and the current limitations of AI-assisted software engineering. It highlights the emerging pattern of AI as a force multiplier for reverse engineering and compatibility layer development, while also underscoring that human oversight remains essential for correctness-critical code.
Technical Details
- Architecture: A managed .NET assembly (
PaintDotNet.Windows.Direct2D1.Managed.dll) that reimplements the Direct2D 1 API from scratch, bypassing the need for WINE to support the native Windows graphics API - Scale: 180,000 lines of generated code — roughly 25% the size of Paint.NET's entire existing codebase (~700,000 lines, 20+ years of development)
- Key technical achievement: Claude reverse-engineered the mathematical formulas required for Direct2D's built-in effects library, a notoriously complex component
- Resource management issues: Claude initially failed to implement COM reference counting correctly (missing
AddRef()equivalents), requiring human intervention - Quality characterization: Described as "vibe coded" — not thoroughly reviewed, relying on trust rather than systematic verification, due to the impracticality of manually reviewing 180K lines
Industry Insight
- AI-generated code at this scale (180K lines) is now feasible for complex system-level libraries, but the "vibe coded" reality means organizations must invest in rigorous review processes or automated verification before deploying such code in production
- The hybrid human-AI workflow demonstrated here — where the human acts as architect and reviewer while AI handles massive code generation — is likely to become a standard pattern for compatibility layer development and reverse engineering projects
- Projects relying on incomplete compatibility layers (like WINE for Direct2D) should consider clean-room AI-assisted reimplementation as a viable strategy, but must budget significant human oversight time for correctness-critical components like memory management and resource handling
Disclaimer: The above content is generated by AI and is for reference only.