Challenges in C++ to Rust Translation with Large Language Models: A Preliminary(…)
summary
Empirical evaluation uses GPT-4-turbo and Deepseek-v3 on the roset/Codex-like and Konet datasets, translating small complete units (classes, functions, templates) with dependency extraction, syntax checks and execution-based semantic checks. GPT-4-turbo produces 45 syntax errors and excels at class translation, while Deepseek-v3 shows strengths on functions and templates; simple features hit 100% on enums/classes but complex templates, macros and global state cause most failures, and IO/output mismatches drive semantic errors.
AI-written summary · about
chapters (7)
- 0:00C++ to Rust translation challenges
- 3:03Current translation approaches
- 4:26LLM study framework
- 11:14Experimental setup & metrics
- 12:19Model performance on correctness
- 14:03Translation error root causes
- 16:55Summary & future research
related talks
15mOptimizing Type Migration for LLM-Based C-to-Rust Translation: A Data Flow Graph Approach
20mScalable, Validated Code Translation of Entire Projects using Large Language Models
37mGPT-3-Powered Type Error Debugging: Investigating the Use of Large Language Models...
01h 01mC++hatGPT: How AI Tools Impact Your Life as a Systems Programmer
01h 09mCodex presentation & demo
34mGood or Bad Vibes? Coding with AI in Rust
36mHow NOT to Train Your Hack Bot: Dos and Don'ts of Building Offensive GPTs
41mTeaching Exotic Programming Languages to Large Language Models