EditLord: Learning Code Transformation Rules for Code Editing

2025-03-10

AI Summary

Existing code editing approaches implicitly model the task as an end-to-end mapping, overlooking its intrinsic discrete-step nature—leading to limited performance, poor robustness, and weak generalization. This paper proposes the first rule-driven framework that explicitly decomposes code editing into interpretable, verifiable steps. Leveraging large language models, it automatically induces a concise, reusable set of meta-rules; editing is then decoupled into semantically grounded intermediate operations governed by these rules. We further design a meta-rule-guided fine-tuning strategy, prompting scheme, and iterative editing mechanism to jointly optimize functional preservation and precise attribute modification. Evaluated across diverse software engineering and security tasks, our method improves editing accuracy by 22.7%, robustness by 58.1%, and functional correctness by 20.2%. It additionally enables sample self-augmentation and cross-task rule transfer.

Technology Category

• Natural Language Processing: Code Generation / Program Synthesis from Natural Language
• Search and Optimization: Metareasoning and Metaheuristics
• Computer Vision: Large Vision Models

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Economics, Online Markets and Human Computation: LLM based quality controls for crowd work
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs

Abstract

Code editing is a foundational task in software development, where its effectiveness depends on whether it introduces desired code property changes without changing the original code's intended functionality. Existing approaches often formulate code editing as an implicit end-to-end task, omitting the fact that code-editing procedures inherently consist of discrete and explicit steps. Thus, they suffer from suboptimal performance and lack of robustness and generalization. We introduce EditLord, a code editing framework that makes the code transformation steps explicit. Our key insight is to employ a language model (LM) as an inductive learner to extract code editing rules from the training code pairs as concise meta-rule sets. Such rule sets will be manifested for each training sample to augment them for finetuning or assist in prompting- and iterative-based code editing. EditLordoutperforms the state-of-the-art by an average of 22.7% in editing performance and 58.1% in robustness while achieving 20.2% higher functional correctness across critical software engineering and security applications, LM models, and editing modes.

Problem

• Explicitly modeling discrete code transformation steps
• Improving robustness and generalization in code editing
• Learning concise meta-rule sets from code pairs

Innovation

• Explicit code transformation steps framework
• LM extracts concise meta-rule sets
• Augments training samples for better performance