@furongjun1999/dsh-memory 0.4.3 → 0.4.5

package/md_cg/whitebox_kb/wisdom/code_compose.py

@@ -0,0 +1,730 @@
+# -*- coding: utf-8 -*-
+"""code_compose.py · 白箱自举第二阶段·代码编写主线原型 v1
+理论：《白箱自举·角色扮演与代码编写》（§3 代码=编程规律条件单元）
+核心：代码编写=带条件的知识问答（条件=编程任务需求）——
+  ① 代码条件单元 CODE_UNITS（{任务 → 代码模式模板}，带 {参数} 占位）
+  ② 任务方向识别（问题 → 任务：排序/去重/计数/最大/反转/求和）
+  ③ 组合生成（任务模板 × 参数 → 代码，未预写完整代码）
+  ④ 自校验三层：L1 语法（ast.parse）→ L2 样例（输入→期望断言运行）→ L3 边界
+  ⑤ 固化：验证通过代码 → 固化 JSON → 同任务直出
+零 LLM：全部白箱确定性生成 + 自校验（语法/测试是物理基底裁决）。
+"""
+import sys, json, os, ast
+sys.stdout.reconfigure(encoding='utf-8')
+
+
+# ============ 一、代码条件单元库（任务 → 代码模式模板 + 验证样例） ============
+# pattern: 代码模板（{fn}=函数名占位）；cases: (输入, 期望输出) 自校验样例
+CODE_UNITS = {
+    "排序-冒泡": {
+        "task": "排序",
+        "pattern": (
+            "def {fn}(arr):\n"
+            "    # 排序：冒泡法（相邻比较交换，较小元素浮到头部）\n"
+            "    n = len(arr)\n"
+            "    for i in range(n):\n"
+            "        for j in range(n - 1 - i):\n"
+            "            if arr[j] > arr[j + 1]:\n"
+            "                arr[j], arr[j + 1] = arr[j + 1], arr[j]\n"
+            "    return arr\n"),
+        "cases": [([3, 1, 2], [1, 2, 3]), ([5, 4, 3, 2, 1], [1, 2, 3, 4, 5]),
+                  ([7], [7]), ([], [])],
+        "params": ["fn"],
+    },
+    "去重-保序": {
+        "task": "去重",
+        "pattern": (
+            "def {fn}(arr):\n"
+            "    # 去重：保留首次出现顺序（seen 集合判重）\n"
+            "    seen = set()\n"
+            "    out = []\n"
+            "    for x in arr:\n"
+            "        if x not in seen:\n"
+            "            seen.add(x)\n"
+            "            out.append(x)\n"
+            "    return out\n"),
+        "cases": [([1, 2, 2, 3, 1], [1, 2, 3]), ([], []), ([1, 1, 1], [1])],
+        "params": ["fn"],
+    },
+    "计数-频率": {
+        "task": "计数",
+        "pattern": (
+            "def {fn}(arr):\n"
+            "    # 计数：元素出现频率（Counter 统计）\n"
+            "    from collections import Counter\n"
+            "    return dict(Counter(arr))\n"),
+        "cases": [([1, 2, 2, 3], {1: 1, 2: 2, 3: 1}), ([], {}), (["a", "a"], {"a": 2})],
+        "params": ["fn"],
+    },
+    "最大值": {
+        "task": "最大",
+        "pattern": (
+            "def {fn}(arr):\n"
+            "    # 最大值：线性扫描取最大（空列表返回 None）\n"
+            "    if not arr:\n"
+            "        return None\n"
+            "    m = arr[0]\n"
+            "    for x in arr[1:]:\n"
+            "        if x > m:\n"
+            "            m = x\n"
+            "    return m\n"),
+        "cases": [([3, 1, 2], 3), ([7], 7), ([], None), ([-1, -5, -2], -1)],
+        "params": ["fn"],
+    },
+    "反转列表": {
+        "task": "反转",
+        "pattern": (
+            "def {fn}(arr):\n"
+            "    # 反转：切片逆序（倒序排列）\n"
+            "    return arr[::-1]\n"),
+        "cases": [([1, 2, 3], [3, 2, 1]), ([], []), (["a", "b"], ["b", "a"])],
+        "params": ["fn"],
+    },
+    "求和": {
+        "task": "求和",
+        "pattern": (
+            "def {fn}(arr):\n"
+            "    # 求和：累加所有元素（空列表返回 0）\n"
+            "    total = 0\n"
+            "    for x in arr:\n"
+            "        total += x\n"
+            "    return total\n"),
+        "cases": [([1, 2, 3], 6), ([], 0), ([5], 5)],
+        "params": ["fn"],
+    },
+    "最大公约数": {
+        "task": "最大公约数",
+        "pattern": (
+            "def {fn}(a, b):\n"
+            "    while b:\n"
+            "        a, b = b, a % b\n"
+            "    return a\n"),
+        "cases": [((48, 36), 12), ((7, 5), 1), ((0, 5), 5), ((54, 24), 6)],
+        "params": ["fn"],
+    },
+    "斐波那契": {
+        "task": "斐波那契",
+        "pattern": (
+            "def {fn}(n):\n"
+            "    if n <= 0:\n"
+            "        return []\n"
+            "    fib = [0, 1]\n"
+            "    while len(fib) < n:\n"
+            "        fib.append(fib[-1] + fib[-2])\n"
+            "    return fib[:n]\n"),
+        "cases": [(1, [0]), (5, [0, 1, 1, 2, 3]), (10, [0, 1, 1, 2, 3, 5, 8, 13, 21, 34])],
+        "params": ["fn"],
+    },
+    "素数": {
+        "task": "素数",
+        "pattern": (
+            "def {fn}(n):\n"
+            "    if n < 2:\n"
+            "        return False\n"
+            "    i = 2\n"
+            "    while i * i <= n:\n"
+            "        if n % i == 0:\n"
+            "            return False\n"
+            "        i += 1\n"
+            "    return True\n"),
+        "cases": [(2, True), (3, True), (4, False), (17, True), (1, False), (0, False)],
+        "params": ["fn"],
+    },
+}
+
+# 任务方向识别（问题 → 任务，最长关键词优先）
+TASK_KEYWORDS = {
+    "排序": ["排序", "排好", "从小到大", "升序", "从大到小", "降序", "整理顺序"],
+    "去重": ["去重", "重复", "不重复", "唯一", "去掉重复"],
+    "计数": ["数一数", "计数", "出现次数", "几次", "统计", "频率"],
+    "最大": ["最大", "最大值", "最高的", "最大的数", "最多"],
+    "反转": ["反转", "倒过来", "倒序", "反过来"],
+    "求和": ["求和", "加起来", "总和", "相加", "累加", "的和", "之和"],
+    "最大公约数": ["最大公约数", "最大公因数", "公约数", "公因数", "gcd"],
+    "斐波那契": ["斐波那契", "fibonacci", "fib"],
+    "素数": ["素数", "质数", "是不是质数", "判断质数", "isprime", "prime"],
+}
+
+# ============ 四、多语言代码单元（第四阶段·代码深学：Rust / JavaScript） ============
+# 每单元带 py_ref（Python 等价实现）——Rust/JS 逻辑样例用 py_ref 验证（模板逻辑正确性）
+
+RUST_UNITS = {
+    "排序-冒泡": {
+        "task": "排序", "lang": "rust",
+        "pattern": ("fn {fn}(arr: &mut Vec<i32>) {\n"
+                    "    let n = arr.len();\n"
+                    "    for i in 0..n {\n"
+                    "        for j in 0..n - 1 - i {\n"
+                    "            if arr[j] > arr[j + 1] {\n"
+                    "                arr.swap(j, j + 1);\n"
+                    "            }\n"
+                    "        }\n"
+                    "    }\n"
+                    "}\n"),
+        "cases": [([3, 1, 2], [1, 2, 3]), ([5, 4, 3, 2, 1], [1, 2, 3, 4, 5]),
+                  ([], [])],
+        "py_ref": "def solve(arr):\n    arr = arr[:]\n    n = len(arr)\n"
+                  "    for i in range(n):\n        for j in range(n - 1 - i):\n"
+                  "            if arr[j] > arr[j + 1]:\n"
+                  "                arr[j], arr[j + 1] = arr[j + 1], arr[j]\n"
+                  "    return arr\n",
+    },
+    "求和": {
+        "task": "求和", "lang": "rust",
+        "pattern": "fn {fn}(arr: &[i32]) -> i32 {\n    arr.iter().sum()\n}\n",
+        "cases": [([1, 2, 3], 6), ([], 0), ([5], 5)],
+        "py_ref": "def solve(arr): return sum(arr)\n",
+    },
+    "最大值": {
+        "task": "最大", "lang": "rust",
+        "pattern": ("fn {fn}(arr: &[i32]) -> Option<i32> {\n"
+                    "    arr.iter().copied().max()\n}\n"),
+        "cases": [([3, 1, 2], 3), ([7], 7), ([], None)],
+        "py_ref": "def solve(arr): return max(arr) if arr else None\n",
+    },
+    "去重-保序": {
+        "task": "去重", "lang": "rust",
+        "pattern": ("fn {fn}(arr: &[i32]) -> Vec<i32> {\n"
+                    "    let mut seen = std::collections::HashSet::new();\n"
+                    "    arr.iter().copied().filter(|x| seen.insert(*x)).collect()\n}\n"),
+        "cases": [([1, 2, 2, 3, 1], [1, 2, 3]), ([], []), ([1, 1, 1], [1])],
+        "py_ref": "def solve(arr):\n    seen = set(); out = []\n"
+                  "    for x in arr:\n        if x not in seen: seen.add(x); out.append(x)\n"
+                  "    return out\n",
+    },
+    "反转": {
+        "task": "反转", "lang": "rust",
+        "pattern": ("fn {fn}(arr: &[i32]) -> Vec<i32> {\n"
+                    "    arr.iter().rev().copied().collect()\n}\n"),
+        "cases": [([1, 2, 3], [3, 2, 1]), ([], []), ([7], [7])],
+        "py_ref": "def solve(arr): return arr[::-1]\n",
+    },
+}
+
+JS_UNITS = {
+    "排序": {
+        "task": "排序", "lang": "javascript",
+        "pattern": "function {fn}(arr) {\n    return arr.slice().sort((a, b) => a - b);\n}\n",
+        "cases": [([3, 1, 2], [1, 2, 3]), ([5, 4, 3, 2, 1], [1, 2, 3, 4, 5]),
+                  ([], [])],
+        "py_ref": "def solve(arr): return sorted(arr)\n",
+    },
+    "求和": {
+        "task": "求和", "lang": "javascript",
+        "pattern": "function {fn}(arr) {\n    return arr.reduce((s, x) => s + x, 0);\n}\n",
+        "cases": [([1, 2, 3], 6), ([], 0), ([5], 5)],
+        "py_ref": "def solve(arr): return sum(arr)\n",
+    },
+    "反转": {
+        "task": "反转", "lang": "javascript",
+        "pattern": "function {fn}(arr) {\n    return arr.slice().reverse();\n}\n",
+        "cases": [([1, 2, 3], [3, 2, 1]), ([], []), (["a", "b"], ["b", "a"])],
+        "py_ref": "def solve(arr): return arr[::-1]\n",
+    },
+    "去重": {
+        "task": "去重", "lang": "javascript",
+        "pattern": "function {fn}(arr) {\n    return [...new Set(arr)];\n}\n",
+        "cases": [([1, 2, 2, 3, 1], [1, 2, 3]), ([], []), ([1, 1, 1], [1])],
+        "py_ref": "def solve(arr):\n    seen = set(); out = []\n"
+                  "    for x in arr:\n        if x not in seen: seen.add(x); out.append(x)\n"
+                  "    return out\n",
+    },
+    "最大": {
+        "task": "最大", "lang": "javascript",
+        "pattern": "function {fn}(arr) {\n    return arr.length ? Math.max(...arr) : null;\n}\n",
+        "cases": [([3, 1, 2], 3), ([7], 7), ([], None)],
+        "py_ref": "def solve(arr): return max(arr) if arr else None\n",
+    },
+}
+
+LANG_UNITS = {"python": CODE_UNITS, "rust": RUST_UNITS,
+              "javascript": JS_UNITS}
+
+
+# ============ 五、域单元注册（第六阶段·白箱自举正式管线接管） ============
+# 四套域单元库（编译器/语言机制/图数据库/操作系统）接入正式自举管线：
+# 域识别 → 单元匹配 → 模板填充 → verify_code 三层自校验 → 固化 JSON → 固化直出
+try:
+    from compiler_code_units import COMPILER_UNITS as _CU
+    from python_code_units import PYTHON_UNITS as _PU
+    from graph_db_units import GRAPH_UNITS as _GU
+    from os_units import OS_UNITS as _OU
+    from browser_units import BROWSER_UNITS as _BU
+    from net_units import NET_UNITS as _NU
+except Exception:
+    _CU = _PU = _GU = _OU = _BU = _NU = {}
+
+DOMAIN_UNITS = {"compiler": _CU, "pylang": _PU,
+                "graph": _GU, "os": _OU, "browser": _BU, "net": _NU}
+
+# 域识别词表（问题 → 域）
+DOMAIN_KEYWORDS = {
+    "compiler": ["编译", "词法", "语法", "VM", "字节码", "序列化", "调试",
+                 "类型推断", "类型检查", "名实", "道德经", "中文编译器",
+                 "函数", "递归", "定义", "注释", "逻辑表达式", "链式",
+                 "常量折叠", "死代码", "寄存器", "优化", "闭包", "词法作用域",
+                 "捕获环境", "断点", "回溯", "监视", "内联展开", "循环展开",
+                 "尾调用", "文件头校验", "完整性校验", "紧凑编码", "圈复杂度",
+                 "活跃变量", "调用图", "字符串字面量", "数字字面量",
+                 "数组字面量", "作用域分析", "常量传播", "指令重排", "引用计数",
+                 "指令剖析", "栈保护", "名实一致", "类型转换", "数据流分析",
+                 "三元表达式", "复合赋值", "位运算", "标识符解析", "操作符解析", "函数签名", "栈操作", "算术执行", "比较执行", "信任检查", "信息差追踪", "条件空间类型", "条件断点", "调用计数", "覆盖率", "窥孔优化", "指令融合", "循环不变式", "字典字面量", "元组解析", "转义序列", "条件空间符号类型", "布尔字面量", "空值字面量", "行号跟踪", "关键字识别", "常量池", "语句分隔", "异常处理", "表达式树", "栈深度分析", "基本块", "支配树", "中间表示", "循环检测", "指令调度", "寄存器溢出", "数组操作", "污点分析", "边界检查消除", "字符类别", "括号匹配", "指令选择", "循环开销", "字符串拼接", "指令大小", "控制流图", "内联缓存", "寄存器着色", "数字后缀", "逃逸分析", "去虚拟化", "名实绑定", "信任流分析", "短路求值", "devirtualize"],
+    "pylang": ["Python", "表达式", "闭包", "闭包机制", "控制流", "作用域", "栈机",
+               "优先级", "逻辑短路", "异常", "抛出", "捕获", "try", "raise",
+               "工厂", "延迟绑定", "nonlocal", "生成器", "yield", "迭代器",
+               "推导", "继承", "多态", "类定义", "装饰器", "上下文",
+               "with", "属性", "类型注解", "运行时检查", "协议", "类型",
+               "异步", "协程", "事件循环", "并发", "元编程", "动态建类",
+               "元类", "描述符", "运算符重载", "枚举", "数据类", "正则匹配",
+               "日期时间", "JSON序列化", "队列栈", "格式化", "排序键控",
+               "集合运算", "计数器", "分组", "自定义异常", "对象组合",
+               "深拷贝", "超时控制", "任务取消", "异步信号量", "映射", "过滤",
+               "归约", "字符串拆分", "字符串替换", "字符串判断", "数学函数", "数值舍入", "数值统计", "默认参数", "关键字参数", "多行字符串", "文件读取", "性能计时", "环境查询", "链表", "进制转换", "异常链", "二叉树", "迭代工具", "字典合并", "最小堆", "函数缓存", "异步生成器", "切片操作", "矩阵运算", "类装饰器", "默认字典", "排列组合", "二分查找", "字典推导", "异步队列", "模板渲染", "有序字典", "随机采样", "字节编解码", "字符串哈希", "跳表", "时间格式化", "字符串对齐", "顺序去重", "滑动均值", "行程压缩", "位掩码", "众数统计", "分位数", "文本分词", "笛卡尔积", "列表分块", "嵌套扁平化", "循环轮转", "字典反转", "直方图", "峰值检测", "累加器", "成对迭代", "打乱", "并查集", "最长公共前缀", "编辑距离", "解包赋值", "集合推导", "切片赋值"],
+    "graph": ["图数据库", "图遍历", "图查询", "条件路由图", "遍历", "路径",
+              "路由", "持久化", "条件链", "建图", "图存储", "模式匹配",
+              "聚合", "匹配", "事务", "索引", "子图", "PageRank", "连通",
+              "拓扑", "执行计划", "批量", "布隆", "相似度", "同构", "社区",
+              "属性", "快照", "版本", "时序", "增量", "分片", "复制",
+              "分布式", "分区", "力导向", "布局", "邻接矩阵", "可视化",
+              "备份", "恢复", "一致性", "压缩", "指标", "健康", "度分布",
+              "监控", "嵌入", "特征", "推荐", "图学习", "权限", "租户",
+              "加密", "安全", "ACL", "读写分离", "慢查询", "在线扩容",
+              "最小生成树", "二分图", "中心性", "正则路径", "查询缓存",
+              "物化视图", "环形布局", "视口变换", "社区着色", "邻接表压缩",
+              "图合并", "属性边", "最大流", "欧拉路径", "图直径", "条件合并",
+              "信任传播", "信息差收敛", "布隆过滤", "租户隔离", "增量备份", "一致性快照", "一致性哈希", "邻居查询", "路径过滤", "三角形计数", "导出子图", "时间窗口", "边活跃度", "模式路径", "节点标签", "模糊匹配", "条件回溯", "信任聚合", "审计日志", "审计记录", "强连通分量", "二分匹配", "可达性判定", "传递闭包", "图着色", "最小割", "聚类系数", "介数中心性", "边索引", "割点", "独立集", "桥检测", "游标遍历", "路径规划", "节点大小", "最大团", "旅行商", "标签约束", "度序列", "生成树计数", "路径计数", "支配集", "弦图判定", "标签计数", "汉密尔顿路径", "图差分", "双层邻居", "树重心", "最大割", "图规范化", "顶点覆盖", "最近公共祖先", "条件分解", "出度直方图"],
+    "os": ["进程", "调度", "内存", "文件系统", "路径解析", "管道", "IPC",
+           "inode", "页置换", "缺页", "状态机", "最短作业", "SJF",
+           "块管理", "位图", "优先级调度", "互斥", "首次适配", "文件块",
+           "页表", "虚拟", "页面", "物理帧", "MMU", "目录树", "文件描述符",
+           "字符设备", "设备", "中断", "向量", "上下文", "嵌套", "抢占",
+           "挂载", "权限", "进程树", "系统调用", "信号", "参数", "校验",
+           "日志", "恢复", "监控", "守护", "信号量", "读写锁", "生产者",
+           "命名空间", "cgroup", "容器", "虚拟化", "RAID", "条带", "奇偶",
+           "快照", "引导", "启动", "固件", "init", "初始化", "访问控制",
+           "审计", "能力", "ACL", "性能", "分析", "瓶颈", "调优", "系统",
+           "配额", "锁", "限额", "ulimit", "安全启动", "TPM", "度量",
+           "哈希", "可信", "校验", "热插拔", "即插即用", "设备树", "PnP",
+           "文件权限", "消息队列", "共享内存", "邮箱", "消息槽", "多级反馈",
+           "EDF", "文件系统调用", "伙伴系统", "写时复制", "内存压缩", "硬链接",
+           "软链接", "文件元数据", "内存映射", "屏障同步", "工作池",
+           "进程生命周期", "磁盘调度", "写时复制快照", "磨损均衡", "强制访问控制", "系统调用过滤", "加密文件系统", "服务管理", "日志轮转", "定时任务", "配置管理", "权限提升", "环境变量", "网络接口", "设备驱动", "电源管理", "文件压缩", "存储池", "文件版本", "条件变量", "自旋锁", "时间片轮转", "优先级继承", "内存池", "稀疏文件", "碎片整理", "段式管理", "时钟节拍", "孤儿进程", "内存碎片", "多核均衡", "僵尸进程", "内存热插拔", "文件系统日志", "进程组", "页缓存", "亲和性", "软中断", "工作窃取", "模块加载"],
+    "browser": ["浏览器", "HTTP", "响应解析", "DOM", "HTML", "CSS", "选择器",
+                "渲染", "布局", "网页", "标签解析", "URL", "请求", "主机",
+                "端口", "协议", "盒模型", "级联", "样式", "padding", "border",
+                "绘制", "布局树", "画布", "事件", "冒泡", "监听", "动画帧",
+                "存储", "localStorage", "会话存储", "Worker", "并行",
+                "Fetch", "缓存", "Cookie", "同源", "CSP", "安全", "XSS",
+                "转义", "Service Worker", "推送", "IndexedDB", "渲染优化",
+                "懒加载", "防抖", "节流", "性能", "应用清单", "缓存策略",
+                "安装事件", "合成分层", "重排", "关键渲染路径", "历史记录",
+                "书签", "标签页", "下载管理", "扩展管理", "网络记录", "表单验证",
+                "拖放", "资源完整性", "媒体播放", "地理位置", "全屏模式", "预加载", "请求合并", "资源优先级", "权限API", "CSP报告", "支付请求", "响应式断点", "离线队列", "会话恢复", "CORS", "跨域资源共享", "文本排版", "剪贴板", "沙箱隔离", "滚动容器", "在线状态", "命中测试", "标签页通信", "页面可见性", "像素光栅化", "触控手势", "设备方向", "渐变填充", "网络信息", "字体加载", "颜色转换", "振动反馈", "混合模式", "边框圆角", "屏幕方向", "空闲调度", "摄像头", "蓝牙扫描", "传感器", "语音合成", "语音识别", "扫码", "画中画", "屏幕捕获", "文件系统访问", "File System", "网页共享", "唤醒锁", "窗口管理", "翻译", "语言检测", "文本摘要", "联系人", "形状检测", "存储配额", "内置聊天", "图像生成", "提示词", "游戏手柄", "虚拟键盘", "音频上下文", "事件委托", "弹窗拦截", "混合内容"],
+    "net": ["网络", "TCP", "UDP", "IP分片", "握手", "校验和", "广播",
+            "局域网", "蜂群", "socket", "中继", "分片", "路由表", "重传",
+            "停等", "去重", "确认", "分帧", "会话", "滑动窗口", "拥塞",
+            "累积确认", "慢启动", "CIDR", "子网", "距离矢量", "NAT",
+            "DNS", "状态码", "负载均衡", "HTTP", "WebSocket", "帧封装",
+            "流式", "多路复用", "连接池", "QUIC", "BGP", "Anycast", "CRC",
+            "IPv6", "隧道", "VLAN", "MQTT", "遥测", "物联网", "消息队列",
+            "FIFO", "CDN", "边缘计算", "内容路由", "前缀", "流量", "延迟", "异常",
+            "RTT", "监控", "检测", "报文", "解析", "抓包", "解码", "协议",
+            "令牌桶", "服务发现", "加密握手", "快速重传", "选择性确认", "SACK",
+            "端口转发", "QoS", "QoS队列", "链路聚合", "滑动窗口限流",
+            "反向代理", "组播", "Reno", "RTO", "吞吐量", "链路状态",
+            "策略路由", "多径传输", "访问令牌", "压缩传输", "会话亲和", "链路加密", "流量镜像", "网络切片", "分块传输", "重定向", "内容协商", "路径MTU", "端口扫描", "会话超时", "消息路由", "API限流", "数据序列化", "DHCP", "租约", "ARP", "地址解析", "ICMP", "Ping", "广播风暴", "心跳保活", "报文重排序", "CSMA", "路由收敛", "链路预算", "RSSI", "NTP同步", "报文分片", "前向纠错", "尽力交付", "带宽时延积", "多宿主", "RTT平滑", "路由汇聚", "序列号回绕", "漏桶限流", "报文调度", "链路利用率", "带宽分配", "连接迁移", "重传统计", "路由衰减", "流分类", "数据包采样", "时隙", "跳频", "误码率", "拓扑发现", "路径备份", "FRR", "数据去重", "网关", "端口镜像", "包过滤", "帧解析", "MAC学习", "证书校验"],
+}
+
+
+def detect_domain(question):
+    """域识别：问题 → 域（compiler/pylang/graph/os/browser/net/None）
+    命中计数优先（「CSS级联+样式优先级」= browser 3 词 > pylang 1 词），
+    平局取最长关键词（「路由表」len3 > 「路由」len2）；
+    空格归一化（「JSON 序列化」→「JSON序列化」——与关键词连写对齐）"""
+    best = None
+    best_cnt_h = best_cnt_d = best_len = best_uid_bonus = 0
+    q_compact = question.replace(" ", "")
+    # 双层计分制（T1 缺口根治 v2）：跨域共享词（校验/序列化/审计日志）靠
+    # 手工表无法消歧——引入 uid 完整串强信号（单元名含于问题 = +100），
+    # 手工表词命中 = 常规分；两类分数独立累加后统一排序。
+    q_hits = {}
+    for domain, kws in DOMAIN_KEYWORDS.items():
+        hit = {k for k in kws if k in question or k in q_compact}
+        if hit:
+            q_hits[domain] = hit
+    unit_pool = {}
+    for domain, units in DOMAIN_UNITS.items():
+        pool = set()
+        uid_full = set()
+        for uid, u in units.items():
+            uid_n = uid.replace("-", "")
+            pool.add(uid)
+            pool.update(uid.replace("-", "").split())
+            pool.add(u.get("task", ""))
+            pool.update(t for t in (u.get("triggers") or []) if t)
+            # uid 完整串（含连字符原样与去连字符形态）强信号
+            if uid in question or uid in q_compact:
+                uid_full.add(uid)
+            elif uid_n and (uid_n in q_compact):
+                uid_full.add(uid)
+        pool.discard("")
+        unit_pool[domain] = (pool, uid_full)
+    all_domains = sorted(set(DOMAIN_KEYWORDS) | set(DOMAIN_UNITS))
+    for domain in all_domains:
+        cnt_h = cnt_d = max_len = uid_bonus = 0
+        # 手工表（人工维护强信号）：命中数与最长词都参与竞争
+        hit = q_hits.get(domain, set())
+        if hit:
+            cnt_h = len(hit)
+            max_len = max(len(k) for k in hit)
+        # 动态池（白箱自举自动扩展）：仅新增词计 cnt_d，**不抬 max_len**
+        # ——自动词是弱信号，只加数量分，防止子串词翻盘人工词
+        pool, uid_full = unit_pool.get(domain, (set(), set()))
+        if pool:
+            hit2 = {k for k in pool if k in question or k in q_compact}
+            cnt_d = len(hit2 - hit)
+            if uid_full:
+                uid_bonus = 100
+                max_len = max(max_len, max(len(u2) for u2 in uid_full))
+        if cnt_h == 0 and cnt_d == 0 and uid_bonus == 0:
+            continue
+        score_tuple = (uid_bonus, cnt_h, cnt_d, max_len)
+        best_tuple = (best_uid_bonus, best_cnt_h, best_cnt_d, best_len)
+        # 确定性：sorted 遍历序固定；平局保持先到者（≥ 已含于 > 语义，
+        # 此处仅 > 判定——平局不翻转，消除 PYTHONHASHSEED 非确定性波动）
+        if score_tuple > best_tuple:
+            best, best_cnt_h, best_cnt_d, best_len, best_uid_bonus =                 domain, cnt_h, cnt_d, max_len, uid_bonus
+    return best
+
+
+def compose_domain_code(question, domain=None, unit_id=None):
+    """域组合生成：域识别 → 单元匹配（task/uid 关键词）→ 模板填充"""
+    domain = domain or detect_domain(question)
+    if domain is None:
+        return None, "域未识别（诚实边界：不属编译器/语言机制/图数据库/操作系统域）", None, None, None
+    units = DOMAIN_UNITS.get(domain, {})
+    # 单元匹配：uid/task 拆词（"编译-指令"→[编译,指令]）+ 原文（最长优先）
+    def _unit_kws(u, uid):
+        """单元级触发词聚合：triggers 表优先，拼接 uid/task 及其连字符词元拆解（T1 自举数据层）。"""
+        # v0.2 T1（自举数据层）：单元级 triggers 触发词——
+        # 缺口检测器（route_gap_audit）白箱自举产出的增强数据
+        triggers = [t for t in (u.get("triggers") or []) if t]
+        return triggers + [uid, u["task"]] + uid.replace("-", "").split() \
+            + u["task"].replace("-", "").split() \
+            + [p for p in uid.split("-") if p]
+
+    best_uid, best_cnt, best_len = None, 0, 0
+    q_compact = question.replace(" ", "")  # 空格归一化（"IP 分片"→"IP分片"）
+    for uid, u in units.items():
+        kws = _unit_kws(u, uid)
+        hit = list({kw for kw in kws
+                    if kw and (kw in question or kw in q_compact)})
+        if not hit:
+            continue
+        cnt, max_len = len(hit), max(len(kw) for kw in hit)
+        if cnt > best_cnt or (cnt == best_cnt and max_len > best_len):
+            best_uid, best_cnt, best_len = uid, cnt, max_len
+    if unit_id is not None and unit_id in units:
+        best_uid = unit_id
+    if best_uid is None:
+        return None, f"域[{domain}]无匹配单元（条件链不完整）", None, None, domain
+    unit = units[best_uid]
+    code = unit["pattern"]
+    return unit["task"], best_uid, code, unit, domain
+
+
+def _verify_with_verifier(code, unit, unit_id=""):
+    """本地校验器六层校验（替代 verify_code 三层）——返回 (ok, checks) 兼容格式。
+
+    接入（阶段 4）：code_compose 生成 → verifier 校验 → 固化，全程零 LLM。
+      - L1语法/L2样例/L3边界/规范符合性/集成 六层 + 指纹本地缓存
+      - 相同请求第二次 → 指纹命中 → 零计算直接返回（替代 LLM 缓存命中）
+      - unit_id 传入（与 --audit 共享同一指纹空间——同一单元同一指纹同一缓存条目）
+      - 注入型单元（needs_inject）L2/L3 由集成测试覆盖；'call' 标记样例跳过
+    """
+    try:
+        from verifier import Verifier, VerifyRequest        # 裸模块（同目录 sys.path，开发态）
+    except ImportError:
+        from wisdom.verifier import Verifier, VerifyRequest  # 包内相对（wheel 发布态）
+    v = Verifier()
+    req = VerifyRequest(
+        task=unit.get("task", ""), code=code, unit_id=unit_id,
+        cases=list(unit.get("cases", [])),
+        expected_structure={"inject": True} if unit.get("needs_inject") else {},
+    )
+    r = v.verify(req)
+    checks = []
+    for c in r.checks:
+        lv = c.get("level", "?")
+        ok = c.get("ok")
+        mark = "✔" if ok else "✗"
+        # evidence 完整保留（[:60] 截断曾吞掉 L2 崩溃 traceback 关键诊断信息）
+        checks.append(f"{mark} {lv} {str(c.get('evidence', ''))}")
+    if r.cached:
+        checks.append(f"✔ 缓存命中（指纹 {r.fingerprint[:12]}）")
+    return r.ok, checks
+
+
+def domain_solidify(question, domain=None, uid=None):
+    """域固化：组合生成 + 本地校验器自校验通过 → 固化（自举纪律：未验证不固化）"""
+    t, u, code, unit, domain = compose_domain_code(question, domain, uid)
+    if code is None:
+        return None
+    ok, checks = _verify_with_verifier(code, unit, u)
+    if not ok:
+        return None  # 自校验未过 → 拒绝固化
+    key = f"domain:{domain}|{u}"
+    entry = {"task": t, "unit": u, "code": code, "checks": checks,
+             "source": "domain_solidified", "domain": domain}
+    CODE_SOLIDIFIED[key] = entry
+    try:
+        json.dump(CODE_SOLIDIFIED, open(_SOL_FILE, "w", encoding="utf-8"),
+                  ensure_ascii=False, indent=1)
+    except Exception:
+        pass
+    return entry
+
+
+def domain_route(question, uid=None):
+    """域路由统一入口：单元匹配 → 已固化直出 → 域组合生成 + 本地校验器校验"""
+    domain = detect_domain(question)
+    if domain is None:
+        return {"question": question, "ok": False,
+                "reason": "域未识别（诚实边界）", "domain": None, "code": None}
+    # 单元匹配先行（与组合生成同一套逻辑），匹配到的单元已固化 → 直出
+    t, u, code, unit, domain = compose_domain_code(question, domain, uid)
+    if code is not None:
+        key = f"domain:{domain}|{u}"
+        if key in CODE_SOLIDIFIED:
+            entry = CODE_SOLIDIFIED[key]
+            return {"question": question, "ok": True, "solidified": True,
+                    "code": entry["code"], "task": entry.get("task"),
+                    "unit": entry.get("unit"), "domain": domain, "checks": []}
+    if code is None:
+        return {"question": question, "ok": False, "reason": u,
+                "task": t, "code": None, "domain": domain}
+    ok, checks = _verify_with_verifier(code, unit, u)
+    return {"question": question, "ok": ok, "task": t, "unit": u,
+            "code": code, "checks": checks, "solidified": False, "domain": domain}
+
+
+def detect_language(question):
+    """语言识别：问题含 rust/rs → Rust；js/javascript → JavaScript；否则 Python"""
+    q = question.lower()
+    if "rust" in q or "rs写" in q or "用rust" in q:
+        return "rust"
+    if "javascript" in q or "js写" in q or "用js" in q or "用javascript" in q:
+        return "javascript"
+    return "python"
+
+
+def identify_task(question):
+    """任务方向识别：问题 → 编程任务（最长关键词优先）"""
+    best, best_len = None, 0
+    for task, kws in TASK_KEYWORDS.items():
+        for k in kws:
+            if k in question and len(k) > best_len:
+                best, best_len = task, len(k)
+    return best
+
+
+def extract_fn_name(question):
+    """参数解析：函数名（问题含英文标识符则用，否则默认 solve）"""
+    import re
+    m = re.search(r"\b([a-zA-Z_][a-zA-Z0-9_]*)\b", question)
+    if m:
+        name = m.group(1)
+        if name.lower() not in ("python", "def", "list", "array", "arr"):
+            return name
+    return "solve"
+
+
+# ============ 二、组合生成（任务模板 × 参数 → 代码） ============
+def compose_code(question, unit_id=None, fn_name=None, lang=None):
+    """组合生成：语言识别 → 任务识别 → 代码单元 → 模板填充 → 代码（未预写完整代码）"""
+    lang = lang or detect_language(question)
+    units = LANG_UNITS.get(lang, CODE_UNITS)
+    task = identify_task(question)
+    if task is None:
+        return None, "任务未识别（诚实边界：不知道要写什么程序）", None, None, lang
+    # 匹配代码单元（任务 → 单元；多单元时取第一个）
+    uids = [uid for uid, u in units.items() if u["task"] == task]
+    if unit_id is not None and unit_id in units:
+        uids = [unit_id]
+    if not uids:
+        return None, f"任务[{task}]在[{lang}]无代码单元覆盖（条件链不完整）", None, None, lang
+    uid = uids[0]
+    unit = units[uid]
+    fn = fn_name or extract_fn_name(question)
+    # replace 而非 format：Rust/JS pattern 含函数体 {}（format 会当占位符报错）
+    code = unit["pattern"].replace("{fn}", fn)
+    return task, uid, code, unit, lang
+
+
+# ============ 三、自校验三层（语法 → 样例 → 边界） ============
+def _run_case(fn, case):
+    """执行单个样例：兼容 (inp, exp) 单参数 / (args_tuple, exp) 多参数 / ('call', exp)"""
+    if len(case) != 2:
+        return None, f"样例格式错误: {case}"
+    inp, exp = case
+    if inp == "call":
+        return "call", exp  # 特殊标记：域注入型单元（L2 由集成测试覆盖）
+    if isinstance(inp, tuple):
+        return fn(*inp), exp
+    return fn(inp), exp
+
+
+def verify_code(code, unit, lang="python"):
+    """自校验：L1 语法 + L2 样例 + L3 边界（白箱自己发现代码错误）
+    Python：ast 语法 + exec 样例运行；Rust/JS：结构校验 + py_ref 逻辑样例
+    v6：支持多参数样例 (args_tuple, exp)；needs_inject 单元 L2 由集成测试覆盖"""
+    import re as _re
+    if lang != "python":
+        checks = []
+        # L1 结构校验：函数关键字 + 括号平衡 + 无占位残留
+        kw = "fn " if lang == "rust" else "function "
+        if kw not in code:
+            return False, [f"✗ L1 结构错误：缺 {kw.strip()} 函数声明"]
+        if "{" not in code or code.count("{") != code.count("}"):
+            return False, ["✗ L1 结构错误：函数体括号不平衡"]
+        if "{" in code and _re.search(r"\{\s*[a-z_]+\s*\}", code):
+            return False, ["✗ L1 结构错误：存在未替换的模板占位符"]
+        # L2 逻辑样例：py_ref（Python 等价实现）验证模板逻辑正确
+        ns = {}
+        try:
+            exec(unit.get("py_ref", ""), ns)
+        except Exception as e:
+            return False, [f"✗ L2 py_ref 错误: {e}"]
+        fns = [v for k, v in ns.items() if callable(v) and not k.startswith("__")]
+        if not fns:
+            return False, ["✗ L2 无 py_ref 函数"]
+        fn = fns[0]
+        for inp, exp in unit.get("cases", []):
+            try:
+                got = fn(inp)
+            except Exception:
+                return False, [f"✗ L2 样例崩溃: {inp}"]
+            if got != exp:
+                return False, [f"✗ L2 样例失败: {inp} → {got}（期望 {exp}）"]
+        checks.append(f"✔ 结构通过 | 样例 {len(unit.get('cases', []))} 组全过（py_ref 逻辑）")
+        return True, checks
+    checks = []
+    # L1 语法
+    try:
+        ast.parse(code)
+    except SyntaxError as e:
+        return False, [f"✗ L1 语法错误: {e}"]
+    # L2 样例运行（函数名必须是 solve——模板默认；用户指定名时用字典注入）
+    ns = {}
+    try:
+        exec(code, ns)
+    except Exception as e:
+        return False, [f"✗ L1 定义错误: {e}"]
+    fns = [v for k, v in ns.items() if callable(v) and not k.startswith("__")]
+    if not fns:
+        return False, ["✗ L1 无函数定义"]
+    fn = fns[0]
+    if unit.get("needs_inject"):
+        # 注入型单元（需要 Graph/run_stmts 等白箱单元组装）：L2 由集成测试覆盖
+        checks.append(f"✔ 语法通过 | 注入型单元：L2 样例由域集成测试覆盖（{len(unit.get('cases', []))} 组）")
+        return True, checks
+    cases = unit.get("cases", [])
+    for case in cases:
+        try:
+            got, exp = _run_case(fn, case)
+        except Exception as e:
+            return False, [f"✗ L2 样例崩溃: {case} → {e}"]
+        if got == "call":
+            continue  # call 特殊标记（域注入型，由集成测试覆盖）
+        if got != exp:
+            return False, [f"✗ L2 样例失败: {case} → {got}（期望 {exp}）"]
+    # L3 边界（额外边界用例：None/空/重复已含在 cases）
+    checks.append(f"✔ 语法通过 | 样例 {len(cases)} 组全过")
+    return True, checks
+
+
+# ============ 四、固化（验证通过 → 固化 → 同任务直出） ============
+_SOL_FILE = os.path.join(os.path.dirname(os.path.abspath(__file__)),
+                         "code_solidified.json")
+CODE_SOLIDIFIED = {}
+if os.path.exists(_SOL_FILE):
+    try:
+        _ld = json.load(open(_SOL_FILE, encoding="utf-8"))
+        if isinstance(_ld, dict):
+            CODE_SOLIDIFIED = _ld
+    except Exception:
+        CODE_SOLIDIFIED = {}
+
+
+def code_solidify(question, task=None, uid=None, fn=None):
+    """固化：组合生成 + 自校验通过 → 固化（自举纪律：未验证代码不固化）"""
+    lang = detect_language(question)
+    t, u, code, unit, lang = compose_code(question, uid, fn, lang)
+    if code is None:
+        return None
+    ok, checks = verify_code(code, unit, lang)
+    if not ok:
+        return None  # 自校验未过 → 拒绝固化（自举纪律）
+    key = f"{task or t}|{u}"
+    entry = {"task": t, "unit": u, "code": code, "checks": checks,
+             "source": "code_solidified", "lang": lang}
+    CODE_SOLIDIFIED[key] = entry
+    try:
+        json.dump(CODE_SOLIDIFIED, open(_SOL_FILE, "w", encoding="utf-8"),
+                  ensure_ascii=False, indent=1)
+    except Exception:
+        pass
+    return entry
+
+
+def code_route(question, uid=None, fn=None):
+    """代码条件路由统一入口：固化层 → 组合生成+自校验（多语言）"""
+    lang = detect_language(question)
+    t = identify_task(question)
+    if t is not None and lang == "python":
+        # 固化层：同任务已固化的代码直出（仅 Python——多语言不混）
+        for k, entry in CODE_SOLIDIFIED.items():
+            if entry.get("task") == t:
+                return {"question": question, "ok": True, "solidified": True,
+                        "code": entry["code"], "task": t,
+                        "unit": entry.get("unit"), "checks": [],
+                        "lang": "python"}
+    task, uid, code, unit, lang = compose_code(question, uid, fn, lang)
+    if code is None:
+        return {"question": question, "ok": False, "reason": uid,
+                "task": task, "code": None, "lang": lang}
+    ok, checks = verify_code(code, unit, lang)
+    return {"question": question, "ok": ok, "task": task, "unit": uid,
+            "code": code, "checks": checks, "solidified": False, "lang": lang}
+
+
+if __name__ == "__main__":
+    print("=== 白箱自举·代码编写主线（组合生成 + 自校验 · 零 LLM） ===\n")
+    QS = [
+        "写一个函数把数组从小到大排序",
+        "写一个函数去掉数组里重复的元素",
+        "写一个函数数一数数组里每个元素出现几次",
+        "写一个函数找出一组数里的最大值",
+        "写一个函数把列表反转",
+        "写一个函数把数组加起来求和",
+        "写一个函数计算圆的面积",   # 任务未识别 → 诚实边界
+    ]
+    results = []
+    for q in QS:
+        r = code_route(q)
+        results.append(r)
+        if r.get("ok") and r.get("code"):
+            mark = "✔"
+            print(f"[{mark}] {q}  [任务={r['task']} 单元={r['unit']}]")
+            print(f"   {r['code'].replace(chr(10), ' | ')}")
+            print(f"   自校验: {r['checks'][0] if r['checks'] else '✔'}")
+        else:
+            print(f"[✘] {q} -> {r.get('reason')}")
+    # 固化演示
+    print("\n=== 固化演示 ===")
+    e = code_solidify("写一个函数把数组从小到大排序")
+    r2 = code_route("写一个函数把数组从小到大排序")
+    print(f"固化: {'✔' if e else '✘'} | 再问: {'固化直出' if r2.get('solidified') else '重新组合'}")
+    # 判定
+    hit = sum(1 for r in results if r.get("ok"))
+    print(f"\n=== 判定 ===\n组合生成测试通过率: {hit}/{len(QS)} = {hit/len(QS)*100:.0f}%（目标≥80%）")