CallGraphite

CallGraphite 是一个 Neovim 插件，旨在通过深度优先搜索（DFS）方式探索大型代码库，收集函数体并将其发送到语言模型（LLM）进行分析。该插件利用语言服务器协议（LSP）和树解析器（Tree-sitter）技术，自动识别和分析函数调用关系，生成调用图和流程图，帮助开发者快速理解复杂代码结构。

功能特点

• 函数文本捕获：使用 Tree-sitter 精确识别和提取当前光标下的函数体
• 深度优先遍历：通过 LSP 引用查询，自动遍历函数调用关系
• LLM 智能分析：将函数代码发送到 LLM 进行分析，识别关键调用和变量
• 两阶段分析：初步分析单个函数，然后基于子调用结果进行综合分析
• 可视化展示：生成 ASCII 和 Mermaid 格式的调用图和流程图
• 导航历史：维护跳转栈，支持在分析过的函数间前进和后退导航
• 结果缓存：缓存分析结果，避免重复请求，提高效率

项目架构

graph TD
    A[插件入口 __init__.py] --> B[函数捕获 capture.py]
    A --> C[遍历管理 traversal.py]
    A --> D[LLM客户端 llm.py]
    B --> E[Lua工具 lua_utils]
    C --> B
    C --> D
    C --> E
    D --> F[配置管理 config.py]
    E --> G[函数体提取 get_current_function_body.lua]
    E --> H[引用查询 get_buf_references.lua]

Loading

flowchart TD
    Start[用户启动分析] --> Capture[捕获当前函数文本]
    Capture --> Analyze[LLM初步分析函数]
    Analyze --> FindCalls[查找函数调用位置]
    FindCalls --> Prioritize[优先处理重要调用]
    Prioritize --> Traverse[递归遍历子函数]
    Traverse --> Comprehensive[综合分析结果]
    Comprehensive --> Visualize[生成可视化图表]
    Visualize --> Display[显示分析结果]

Loading

Installation

Use your favourite Neovim plugin manager. Examples are shown below.

Using Plug

Plug 'mikewong23571/CallGraphite'

After running :PlugInstall, restart Neovim and run :UpdateRemotePlugins so the Python plugin is registered.

Using packer.nvim

use {
  'mikewong23571/CallGraphite',
  run = ':UpdateRemotePlugins'
}

Congiguration

插件默认使用 ~/.config/callgraphite/config.json 作为配置文件。您可以创建此文件并自定义以下选项：

{
  "llm": {
    "endpoint": "https://api.openai.com/v1/chat/completions",
    "api_key": "your-api-key-here",
    "model": "gpt-3.5-turbo",
    "temperature": 0.3,
    "max_tokens": 500
  },
  "visualization": {
    "enabled": true,
    "ascii_graph": true,
    "mermaid_graph": true,
    "mermaid_flow": true,
    "max_depth": 5
  },
  "analysis": {
    "comprehensive": true,
    "cache_results": true
  }
}

Usage

Open a project supported by your LSP. Then run the command:

:CallGraphite

The plugin traverses the files via LSP, extracts each function body and prints the text that would be sent to the LLM. You should see output similar to:

Collected 42 functions
Uploading to LLM...

The exact output depends on the LLM integration, but you can expect a summary once all functions are processed.

You can also capture the text of the function under the cursor using:

:CaptureFunction

This command uses tree-sitter to locate the nearest parent function and prints its source.

Navigation through analysed locations is possible using:

:CallGraphiteBack
:CallGraphiteForward

These commands move through the internal jump stack maintained by the plugin, mirroring Neovim's own jump list.

Running Tests

Install the Python dependencies and run pytest:

pip install -r requirements.txt
pytest

License

This project is licensed under the Apache License 2.0. See LICENSE for details.