Large language models (LLMs) show promise in code translation due to their ability to generate idiomatic code. However, a significant limitation when using LLMs for code translation is scalability: existing works have shown a drop in translation success rates for code exceeding around 100 lines. We overcome this limitation by developing a modular approach to translation, where we partition the code into small code fragments which can be translated independently and semantically validated (that is, by checking I/O equivalence). When this approach is applied naively, we discover that LLMs are unreliable when translating features of the source language that do not have a direct mapping to the target language, and that the LLM often gets stuck in repair loops when attempting to fix errors. To address these issues, we introduce two key concepts: (1) feature mapping , which integrates predefined translation rules with LLM-based translation to guide the LLM in navigating subtle language differences and producing semantically accurate code; and (2) type-compatibility , which facilitates localized checks at the function signature level to detect errors early, thereby narrowing the scope of potential repairs. We apply our approach to translating real-world Go codebases to Rust, demonstrating that we can consistently generate reliable Rust translations for projects up to 9,700 lines of code and 780 functions, with an average of 73% of functions successfully validated for I/O equivalence, considerably higher than any existing work.
6 November 2023 in Journal of Functional Programming
Bidirectional transformations (BXs) are a mechanism for maintaining consistency between multiple representations of related data. The lens framework, which usually constructs BXs from lens combinators, has become the mainstream approach to BX programming because of its modularity and correctness by construction. However, the involved bidirectional behaviors of lenses make the equational reasoning and optimization of them much harder than unidirectional programs. We propose a novel approach to deriving efficient lenses from clear specifications via program calculation, a correct-by-construction approach to reasoning about functional programs by algebraic laws. To support bidirectional program calculation, we propose contract lenses , which extend conventional lenses with a pair of predicates to enable safe and modular composition of partial lenses. We define several contract-lens combinators capturing common computation patterns including $\textit{fold}, \textit{filter},\textit{map}$ , and $\textit{scan}$ , and develop several bidirectional calculation laws to reason about and optimize contract lenses. We demonstrate the effectiveness of our new calculation framework based on contract lenses with nontrivial examples.
Large language models (LLMs) show promise in code translation due to their ability to generate idiomatic code. However, a significant limitation when using LLMs for code translation is scalability: existing works have shown a drop in translation success rates for code exceeding around 100 lines. We overcome this limitation by developing a modular approach to translation, where we partition the code into small code fragments which can be translated independently and semantically validated (that is, checking I/O equivalence). When this approach is applied naively, we discover that LLMs are unreliable when translating features of the source language that do not have a direct mapping to the target language, and that the LLM often gets stuck in repair loops when attempting to fix errors. To address these issues, we introduce two key concepts: (1) feature mapping, which integrates predefined translation rules with LLM-based translation to guide the LLM in navigating subtle language differences and producing semantically accurate code; and (2) type-compatibility, which facilitates localized checks at the function signature level to detect errors early, thereby narrowing the scope of potential repairs. We apply our approach to translating real-world Go codebases to Rust, demonstrating that we can consistently generate reliable Rust translations for projects up to 6,600 lines of code and 369 functions, with an average of 73% of functions successfully validated for I/O equivalence, considerably higher than any existing work.