Researchers have introduced FeatLens, a novel approach for repository-level code generation that focuses on retrieving task-relevant code dependencies. Unlike previous methods that rely on similarity or graph-based retrieval, FeatLens uses natural-language feature descriptions to build an index that links features to code entities. This allows for more precise selection of dependencies needed to implement specific functional requirements. Experiments on DevEval and EvoCodeBench benchmarks demonstrated that FeatLens achieves superior dependency recall and supports higher non-oracle Pass@1 scores when used with models like DeepSeek-V3.2 and GPT-5-mini, notably without requiring LLM calls during the retrieval phase. AI
IMPACT Enhances code generation by improving the precision of dependency retrieval, potentially leading to more accurate and efficient software development.
RANK_REASON This is a research paper detailing a new method for code generation. [lever_c_demoted from research: ic=1 ai=1.0]
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
