SaraCoder: Orchestrating Semantic and Structural Cues for Profit-Oriented Repository-Level Code Completion

Abstract: Retrieval-augmented generation (RAG) for repository-level code completion commonly relies on superficial text similarity, leading to results plagued by semantic misguidance, redundancy, and homogeneity, while also failing to resolve external symbol ambiguity. To address these challenges, we introduce Saracoder, a Hierarchical Feature-Optimized retrieval framework. Its core Hierarchical Feature Optimization module systematically refines candidates by distilling deep semantic relationships, pruning exact duplicates, assessing structural similarity with a novel graph-based metric that weighs edits by their topological importance, and reranking results to maximize both relevance and diversity. Furthermore, an External-Aware Identifier Disambiguator module accurately resolves cross-file symbol ambiguity via dependency analysis. Extensive experiments on the challenging CrossCodeEval and RepoEval-Updated benchmarks demonstrate that Saracoder significantly outperforms existing baselines across multiple programming languages and models. Our work proves that systematically refining retrieval results across multiple dimensions provides a new paradigm for building more accurate and robust repository-level code completion systems.