Replication Package for ASE 2026 Anonymous Submission
Summary
No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.
Summarise with VIRAT Free to try with your Science Catalogs account.Abstract
REA-CoderCode generation refers to automatically producing executable programs from user requirements. Recently, researchers have explored approaches to enhance the correctness of generated code with advanced large language models. Although achieving improvements, existing approaches focus on designing reasoning strategies or post-refinement methods to enhance code generation performance. Despite their differences, all these methods share a common assumption: the LLM can correctly understand the given requirement. However, this assumption does not always hold. To fill this gap, we propose REA-Coder, a requirement alignment approach to enhance the code generation performance of LLMs. REA-Coder involves first identifying the requirement content that does not align with LLMs and aligning the requirements. Then, based on the aligned requirements, LLMs generate code and further verify whether the generated code aligns with the requirements, iterating this process of requirement alignment and code generation until generating correct code or achieving the maximum number of iterations.Repository StructureThe project is structured to separate the core REA-Coder framework from the comparative baseline methods. . ├── REA-Coder/ # Core implementation of the REA-Coder framework │ ├── apps_controller/ # Logic and orchestration │ │ ├── questions.py # Clarification and question generation │ │ ├── requirements.py # Requirement engineering and gap summarization │ │ ├── codegen.py # Code generation and answer synthesis │ │ ├── feedback.py # Evaluation and feedback generation │ │ ├── prompts_markdown/ # System prompts │ │ └── run_eval.py # Main execution script │ ├── apps_eval/ # Execution and evaluation environment │ │ ├── executor.py # Code execution sandbox │ │ └── parallel_runner.py # Parallel evaluation runner │ └── llm/ # Large Language Model client definitions │ └── client.py # API request handlers and routing └── baseline/ # Implementations of comparative baselines ├── Icot/ # Interactive Chain-of-Thought ├── Scot/ # Structured Chain-of-Thought ├── Self-collaboration/ # Multi-role collaborative framework ├── Self-repair/ # Execution-based automated repair ├── Specfix/ # Specification-driven generation ├── Specine/ # Alignment and sanitization focus └── ufix/ # Fix generation using CodeBLEU and feedback Supported ModelsThe framework integrates a flexible LLM client interface configured to support the following advanced models for generation and routing:qwen3-coder-30b-a3b-instructgpt-5-mini-2025-08-07gemini-3-flash-previewdeepseek-chatEvaluated DatasetsREA-Coder is designed to be evaluated on a comprehensive suite of programming benchmarks to test functional correctness under varying levels of complexity:appsCodecontests_rawCodecontestsXCodeEvallivecodebenchComparative BaselinesTo demonstrate the efficacy of REA-Coder, this repository includes integrated implementations of several recent baseline methods for code generation and self-repair. These are provided for brief comparison:ICoT: Interactive Chain-of-Thought.SCoT: Structured Chain-of-Thought.Self-collaboration: A multi-role (analyst, coder, tester) LLM collaboration framework.Self-repair: Automated code repair using execution feedback.Specfix & Specine: Specification-driven generation and sanitization approaches.UFix: Evaluation and fix-generation using CodeBLEU and execution feedback.Getting StartedInstallationEnsure you have Python 3.9+ installed. Clone the repository and install the dependencies. git clone https://anonymous.4open.science/r/REA-Coder-6CFE cd REA-Coder pip install -r requirements.txt Running REA-CoderYou can execute the main REA-Coder evaluation pipeline by running the controller script: python REA-Coder/apps_controller/run_eval.py To configure API keys or switch between models, please update the settings within REA-Coder/llm/client.py and the respective routing configurations.
Fields of Study & Subjects
Software Engineering Research
Software Engineering Techniques and Practices
Topic Modeling
Publication details
- Journal / source
- Figshare
- Publication date
- 27 March 2026
- Language
- EN
- Document type
- Dataset