展示 HN:Eulix - 大型代码库的代码导航
2 分•作者: Nurysso•5 天前
我一直在开发 Eulix,一个用于浏览大型代码库的工具。
它会将一个仓库解析成符号、调用图和其他结构化信息,然后将其与关键字和语义检索相结合,以查找相关代码。
我在 OpenStack(约 690 万行代码 / 29,000 个文件)上对其进行了测试。一次关于 Nova 的 PCI 直通调度的查询,在索引完成后,不到一秒钟就检索出了相关的过滤器、辅助函数和相关的调用路径。
有些查询根本不需要 LLM,因为 Eulix 可以直接从结构化的代码库数据中给出答案。
它是开源的,并且可以在本地运行:
[https://github.com/Nurysso/eulix](https://github.com/Nurysso/eulix)
我特别希望那些在代码搜索、静态分析或大型单体仓库方面有工作经验的人能提供反馈。
另外,它可能也能处理 3000 万行以上的代码库,我之所以没能测试这么大的仓库,是因为我没有足够好的 GPU 来嵌入解析器的输出! :)
查看原文
Hey I've been working on Eulix, a tool for navigating large codebases.<p>It parses a repository into symbols, call graphs and other structural information, then combines that with keyword and semantic retrieval to find relevant code.<p>I tested it on OpenStack (~6.9M LOC / 29k files). One query about Nova's PCI passthrough scheduling pulled back the relevant filters, helpers and related call paths in well under a second once indexed.<p>Some queries don't need an LLM at all, since Eulix can answer directly from the structured codebase data.<p>It's open source and runs locally:<p><a href="https://github.com/Nurysso/eulix" rel="nofollow">https://github.com/Nurysso/eulix</a><p>I'd especially like feedback from people who've worked on code search, static analysis, or large monorepos.<p>on a side note it may be able to handle 30M+ loc codebase too, I haven't been able to test such huge repos cause I don't have a good enough gpu to embed parsers output! :)