前面我們已經能做到精準搜尋與 Rerank，但當工程師問「如果我改了這個函式，會影響誰」時，檢索系統就不能只看整份檔案的 import，必須精確到「函式呼叫函式」的粒度。

今天動手做變更影響分析打底：定義呼叫圖（Call Graph）的核心資料結構，並在 SQLite 中開闢保存呼叫關係的專屬資料表。

呼叫關係的最小單元：CallEdge

一次函式呼叫（Call）至少需要記錄四個維度：

1. 呼叫端（Caller）：誰發起了呼叫（包含檔案路徑與函式名稱）。
2. 被呼叫端（Callee）：呼叫了什麼名稱（AST 看到的目標函式名）。
3. 精確位置（Location）：這次呼叫發生在哪個檔案的第幾行（維持 Evidence First 原則）。
4. 解析狀態（Resolution Status）：靜態分析時是否已經能對應到真實目標檔案。

from dataclasses import dataclass
from typing import Optional

@dataclass
class CallEdge:
    caller_path: str        # 呼叫端檔案，如 "src/rag_chat.py"
    caller_symbol: str      # 呼叫端函式，如 "chat_loop"
    callee_name: str        # 被呼叫名稱，如 "build_index" 或 "client.search"
    line: int               # 呼叫發生的行號
    callee_path: Optional[str] = None   # 解析後的被呼叫端檔案 (若能確定)
    callee_symbol: Optional[str] = None # 解析後的完整符號 (若能確定)

擴充 SQLite Schema：calls 表

我們在 app/codebase.py 中擴充 SQLite 資料庫，建立專門存放跨函式呼叫的 calls 表：

CREATE TABLE IF NOT EXISTS calls (
    id INTEGER PRIMARY KEY AUTOINCREMENT,
    caller_path TEXT NOT NULL,
    caller_symbol TEXT NOT NULL,
    callee_name TEXT NOT NULL,
    callee_path TEXT,
    callee_symbol TEXT,
    line INTEGER NOT NULL
);

CREATE INDEX IF NOT EXISTS idx_calls_caller ON calls(caller_path, caller_symbol);
CREATE INDEX IF NOT EXISTS idx_calls_callee ON calls(callee_name);

核心實作：app/call_graph.py 基礎骨架

建立 app/call_graph.py，負責呼叫邊的寫入與基礎查詢：

# app/call_graph.py
import sqlite3
from typing import List, Dict, Any, Optional
from dataclasses import dataclass

@dataclass
class CallEdge:
    caller_path: str
    caller_symbol: str
    callee_name: str
    line: int
    callee_path: Optional[str] = None
    callee_symbol: Optional[str] = None

class CallGraphStore:
    def __init__(self, conn: sqlite3.Connection):
        self.conn = conn
        self._init_table()

    def _init_table(self):
        with self.conn:
            self.conn.executescript("""
                CREATE TABLE IF NOT EXISTS calls (
                    id INTEGER PRIMARY KEY AUTOINCREMENT,
                    caller_path TEXT NOT NULL,
                    caller_symbol TEXT NOT NULL,
                    callee_name TEXT NOT NULL,
                    callee_path TEXT,
                    callee_symbol TEXT,
                    line INTEGER NOT NULL
                );
                CREATE INDEX IF NOT EXISTS idx_calls_caller ON calls(caller_path, caller_symbol);
                CREATE INDEX IF NOT EXISTS idx_calls_callee ON calls(callee_name);
            """)

    def insert_edges(self, edges: List[CallEdge]):
        """批次寫入呼叫邊"""
        with self.conn:
            rows = [
                (e.caller_path, e.caller_symbol, e.callee_name, e.callee_path, e.callee_symbol, e.line)
                for e in edges
            ]
            self.conn.executemany("""
                INSERT INTO calls (caller_path, caller_symbol, callee_name, callee_path, callee_symbol, line)
                VALUES (?, ?, ?, ?, ?, ?)
            """, rows)

    def find_direct_callers_by_name(self, callee_name: str) -> List[Dict[str, Any]]:
        """給定被呼叫函式名稱，逆向找出所有直接呼叫它的 caller 與行號"""
        cur = self.conn.cursor()
        cur.execute("""
            SELECT caller_path, caller_symbol, callee_name, line
            FROM calls
            WHERE callee_name = ?
            ORDER BY caller_path, line
        """, (callee_name,))
        return [dict(row) for row in cur.fetchall()]

實作單元測試：tests/unit/test_call_graph_store.py

在 tests/unit/test_call_graph_store.py 驗證呼叫關係的寫入與逆向反查：

# tests/unit/test_call_graph_store.py
import sqlite3
from app.call_graph import CallGraphStore, CallEdge

def test_insert_and_find_callers():
    conn = sqlite3.connect(":memory:")
    conn.row_factory = sqlite3.Row
    store = CallGraphStore(conn)

    # 模擬兩處呼叫了 build_index()
    edges = [
        CallEdge(caller_path="src/cli.py", caller_symbol="main", callee_name="build_index", line=45),
        CallEdge(caller_path="src/test_pipeline.py", caller_symbol="test_run", callee_name="build_index", line=12),
        CallEdge(caller_path="src/cli.py", caller_symbol="main", callee_name="setup_logging", line=10)
    ]
    store.insert_edges(edges)

    # 逆向查詢誰呼叫了 build_index
    callers = store.find_direct_callers_by_name("build_index")

    assert len(callers) == 2
    assert callers[0]["caller_path"] == "src/cli.py"
    assert callers[0]["caller_symbol"] == "main"
    assert callers[0]["line"] == 45
    assert callers[1]["caller_path"] == "src/test_pipeline.py"

執行測試確認綠燈：

uv run pytest tests/unit/test_call_graph_store.py -v

```text
tests/unit/test_call_graph_store.py::test_insert_and_find_callers PASSED  [100%]
============================== 1 passed in 0.04s ==============================

下一步

今天完成了呼叫圖的資料底層與反查結構。

明天，我們將直接進入 AST 解析器，走訪 Python 語法樹的 ast.Call 節點，把專案中所有真實發生的函式呼叫事件一次全部抓出來！