@hupan56/wlkj 2.7.12 → 3.1.2

package/templates/qoder/scripts/validation/test/autotest_webaccess.py

@@ -0,0 +1,571 @@
+# -*- coding: utf-8 -*-
+# v3.0 路径自举: 引导到 common/bootstrap, 统一 sys.path 逻辑
+import os as _o, sys as _s
+_f = _o.path.abspath(__file__)
+for _ in range(10):
+    _f = _o.path.dirname(_f)
+    _cp = _o.path.join(_f, 'foundation')
+    if _o.path.isfile(_o.path.join(_cp, 'bootstrap.py')):
+        break
+if _cp not in _s.path: _s.path.insert(0, _cp)
+from bootstrap import setup; setup()
+
+# 强制 stdout 无缓冲 — 被 timeout 杀掉/管道消费时也要看到输出 (实测踩坑)
+try:
+    sys.stdout.reconfigure(line_buffering=True)
+except Exception:
+    pass
+
+"""autotest_webaccess.py — 真功能测试执行器 (web-access CDP 直连已登录 Chrome).
+
+与 autotest_run.py 平级, 镜像同一个 task.json 契约, 引擎换成 web-access.
+吃同样的 --task 文件, 吐同样结构的 result JSON.
+
+为什么用 web-access 而非 opencli:
+  - opencli 走 chrome.debugger.attach (扩展代理), 在 Windows 后台 session 静默失败.
+  - web-access 走 CDP 端口直连 (chrome://inspect#remote-debugging), 不装扩展,
+    和 QoderWork chrome 连通器同源机制, 稳.
+  - 登录态天然复用 (连的是日常 Chrome, cookie/session 全在), 验证码零接触.
+
+用法: python autotest_webaccess.py --task <任务文件> [--browser chrome]
+  任务文件 JSON: {
+    "page_path": "/veh/vehicle/vehAffair/insurance",
+    "page_name": "保险管理",
+    "cases": [ {case_id, title, steps:[{intent,desc,value,anchor,check,expect}], expected} ]
+  }
+
+前置: 用户 Chrome 开 chrome://inspect/#remote-debugging 勾选允许调试.
+没开 → 友好提示 + 自动回退 playwright (50 人零报错).
+"""
+import json, sys, time, os, re, argparse, subprocess
+import urllib.request, urllib.error
+
+# ===== 配置 (路径自检测, 与 autotest_run.py 一致) =====
+SCRIPTS = os.path.dirname(os.path.abspath(__file__))
+# test 目录在 .qoder/scripts/test, 回项目根要上 3 级. 向上找含 .qoder 的目录兜底.
+_d = SCRIPTS
+while _d != os.path.dirname(_d):
+    if os.path.isdir(os.path.join(_d, '.qoder')):
+        PROJECT = _d
+        break
+    _d = os.path.dirname(_d)
+else:
+    PROJECT = os.path.normpath(os.path.join(SCRIPTS, '..', '..', '..'))
+
+WEBACCESS_DIR = os.path.join(os.path.dirname(SCRIPTS), 'web-access')  # .qoder/scripts/web-access (test 同级)
+CDP_PROXY_PORT = 3456
+PROXY_URL = 'http://localhost:%d' % CDP_PROXY_PORT
+AUTOLEARN = os.path.join(SCRIPTS, 'autotest.py')
+BASE = 'https://icsqas.inforecloud.com'
+
+
+# ===== web-access HTTP API 封装 (零依赖, 用 urllib) =====
+def _http(method, path, body=None, timeout=15):
+    """调 cdp-proxy 的 HTTP API. 返回解析后的 JSON dict."""
+    url = PROXY_URL + path
+    data = body.encode('utf-8') if body is not None else None
+    req = urllib.request.Request(url, data=data, method=method)
+    if body is not None:
+        req.add_header('Content-Type', 'text/plain')
+    try:
+        with urllib.request.urlopen(req, timeout=timeout) as r:
+            return json.loads(r.read().decode('utf-8', 'ignore'))
+    except urllib.error.URLError as e:
+        return {'error': '连接失败: %s (proxy 没起? 跑 node web-access/scripts/check-deps.mjs --browser chrome)' % str(e)[:50]}
+    except Exception as e:
+        return {'error': str(e)[:60]}
+
+
+def proxy_ready():
+    """检查 cdp-proxy 是否在跑 + 连着浏览器."""
+    h = _http('GET', '/health', timeout=5)
+    return h.get('connected') is True, h
+
+
+def start_proxy(browser='chrome'):
+    """启动 cdp-proxy (自动发现 Chrome 调试端口). 返回 (ok, msg).
+    proxy 是长驻进程 — 已在跑就直接复用, 不重复 spawn (避免冲突/卡死)."""
+    # 快路径: proxy 已在跑且连着浏览器 → 直接用
+    ok, h = proxy_ready()
+    if ok:
+        return True, h.get('browser', {}).get('label', '')
+    check = os.path.join(WEBACCESS_DIR, 'scripts', 'check-deps.mjs')
+    if not os.path.isfile(check):
+        return False, 'web-access 引擎未安装 (%s 不存在)' % check
+    # check-deps 会自动起 proxy 并连浏览器. stdout 含 proxy: ready 即成功.
+    try:
+        r = subprocess.run(['node', check, '--browser', browser],
+                           capture_output=True, timeout=40, cwd=WEBACCESS_DIR)
+        out = (r.stdout + r.stderr).decode('utf-8', 'ignore')
+    except subprocess.TimeoutExpired:
+        return False, 'check-deps 超时 (Chrome 调试开关没开?)'
+    except FileNotFoundError:
+        return False, 'node 未安装或不在 PATH (web-access 要 Node 22+)'
+    if 'proxy: ready' in out or 'proxy: connecting' in out:
+        time.sleep(2)  # 等 proxy 完全就绪
+        ok, h = proxy_ready()
+        return ok, h.get('browser', {}).get('label', '') if ok else 'proxy 起了但没连上'
+    # 没成功 → 解析失败原因给用户
+    if 'needs decision' in out or '请询问用户' in out:
+        return False, '需指定浏览器: 加 --browser chrome'
+    if 'remote debugging' in out.lower() or '调试' in out:
+        return False, 'Chrome 远程调试未开启. 地址栏开 chrome://inspect/#remote-debugging 勾选 "Allow remote debugging"'
+    return False, out.strip().split('\n')[-1][:80]
+
+
+# ===== tab 管理 =====
+def new_tab(url, wait=4):
+    """开新后台 tab, 返回 targetId 或 None."""
+    r = _http('POST', '/new', body=url, timeout=30)
+    tid = r.get('targetId')
+    if tid and wait:
+        time.sleep(wait)
+    return tid
+
+
+def navigate(tid, url, wait=3):
+    """在已有 tab 导航."""
+    r = _http('POST', '/navigate?target=%s' % tid, body=url, timeout=30)
+    if wait:
+        time.sleep(wait)
+    return 'error' not in r
+
+
+def close_tab(tid):
+    """关闭 tab."""
+    try:
+        _http('GET', '/close?target=%s' % tid, timeout=10)
+    except Exception:
+        pass
+
+
+def page_info(tid):
+    """读页面 title/url/readyState."""
+    return _http('GET', '/info?target=%s' % tid, timeout=10)
+
+
+def eval_js(tid, code, timeout=15):
+    """执行 JS, 返回 (value, error). value 是任意类型."""
+    r = _http('POST', '/eval?target=%s' % tid, body=code, timeout=timeout)
+    if 'error' in r:
+        return None, r['error']
+    return r.get('value'), None
+
+
+def click_css(tid, selector):
+    """JS 点击 (CSS 选择器). 返回 (ok, msg)."""
+    r = _http('POST', '/click?target=%s' % tid, body=selector, timeout=10)
+    return 'error' not in r, r.get('error', 'clicked')
+
+
+# ===== 登录态: 无需伪造, CDP 连的是已登录 Chrome. 但需验证登录态有效 =====
+def check_logged_in(tid):
+    """检查当前页是否在登录页 (在登录页 = 未登录/失效)."""
+    info = page_info(tid)
+    url = info.get('url', '')
+    # SPA 登录页通常 url 含 /auth/login 或 title 是登录相关
+    return '/auth/login' not in url, url
+
+
+# ===== anchor → CSS 选择器 (web-access click 用 CSS) =====
+def anchor_to_css(anchor, desc=''):
+    """把 anchor {role,name,placeholder,field,text} 转 CSS 选择器.
+    web-access 的 click 接受 CSS, 比 opencli 的语义 flag 弱一点, 但 eval 能兜底."""
+    if not anchor:
+        return ''
+    field = anchor.get('field', '')
+    if field:
+        return "input[name='%s']" % field
+    name = anchor.get('name', '')
+    ph = anchor.get('placeholder', '')
+    role = anchor.get('role', '')
+    text = anchor.get('text', '')
+    # 优先按 placeholder/name 文本匹配 input/button
+    if role == 'textbox':
+        if ph:
+            return "input[placeholder*='%s']" % ph
+        if name:
+            return "input[placeholder*='%s']" % name
+    if role == 'button':
+        # button 文本匹配用 eval 兜底更准, 这里先给个基础
+        if name:
+            return "button"
+    if text:
+        return "*"
+    return ''
+
+
+def click_by_anchor(tid, anchor, desc=''):
+    """按 anchor 点击: 先试 CSS, 失败用 eval 文本匹配兜底."""
+    css = anchor_to_css(anchor, desc)
+    name = anchor.get('name', '')
+    role = anchor.get('role', '')
+    text = anchor.get('text', '')
+    # button/link/tab 用 eval 按 innerText 精确匹配 (比 CSS :has-text 稳)
+    if role in ('button', 'link', 'tab') and (name or text):
+        kw = name or text
+        code = ("(function(){const els=[...document.querySelectorAll('%s')];"
+                "const el=els.find(e=>(e.innerText||'').trim()==='%s'||(e.innerText||'').includes('%s'));"
+                "if(el){el.scrollIntoView();el.click();return 'ok'}return 'notfound'})()" % (
+                    'button,a,[role=tab],.ant-tabs-tab' if role == 'tab' else 'button,a,[role=button]',
+                    kw, kw))
+        v, e = eval_js(tid, code)
+        return (v == 'ok'), ('clicked:%s' % kw if v == 'ok' else '未找到%s' % kw)
+    # textbox 用 CSS
+    if css:
+        ok, msg = click_css(tid, css)
+        if ok:
+            return True, 'clicked'
+    return False, '元素未找到(%s)' % desc
+
+
+def fill_by_anchor(tid, anchor, val, desc=''):
+    """填值: 用 eval 直接设 value + dispatch input event (兼容 Vue)."""
+    css = anchor_to_css(anchor, desc)
+    if not css:
+        return False, '元素未找到(anchor空:%s)' % desc
+    # Vue 需要触发 input 事件框架才认. 用原生 setter + dispatchEvent.
+    safe_val = str(val).replace("\\", "\\\\").replace("'", "\\'")
+    code = ("(function(){const el=document.querySelector(\"%s\");"
+            "if(!el)return 'notfound';"
+            "const setter=Object.getOwnPropertyDescriptor(window.HTMLInputElement.prototype,'value').set;"
+            "setter.call(el,'%s');"
+            "el.dispatchEvent(new Event('input',{bubbles:true}));"
+            "el.dispatchEvent(new Event('change',{bubbles:true}));"
+            "return el.value})()" % (css.replace('"', '\\"'), safe_val))
+    v, e = eval_js(tid, code)
+    if v == 'notfound':
+        return False, '元素未找到(%s)' % desc
+    return (v == str(val)), 'filled(actual=%s)' % _short(v)
+
+
+# ===== 表格等待 (与 autotest_run.py 等价语义) =====
+def wait_table_stable(tid, max_wait=8):
+    """轮询 .vxe-body--row 行数, 等表格加载稳定."""
+    deadline = time.time() + max_wait
+    zero_streak = 0
+    last = -1
+    while time.time() < deadline:
+        v, _ = eval_js(tid, "document.querySelectorAll('.vxe-body--row').length", timeout=5)
+        try:
+            n = int(v)
+        except (TypeError, ValueError):
+            n = 0
+        if n > 0:
+            time.sleep(0.6)
+            v2, _ = eval_js(tid, "document.querySelectorAll('.vxe-body--row').length", timeout=5)
+            try:
+                n2 = int(v2)
+            except (TypeError, ValueError):
+                n2 = 0
+            if n2 >= n:
+                return n2
+        else:
+            zero_streak += 1
+            if zero_streak >= 3:
+                return 0
+        time.sleep(0.5)
+    return last
+
+
+# ===== 单步执行 (镜像 autotest_run.py exec_step 结构) =====
+def exec_step(tid, step, timeout=8):
+    """执行一步, 返回 (ok, detail)."""
+    intent = step.get('intent', '')
+    desc = step.get('desc', '')
+    anchor = step.get('anchor', {}) or {}
+    val = step.get('value', '')
+
+    try:
+        if intent == 'goto':
+            url = BASE + val if val.startswith('/') else val
+            navigate(tid, url)
+            wait_table_stable(tid, max_wait=8)
+            return True, 'navigated'
+
+        if intent == 'fill':
+            return fill_by_anchor(tid, anchor, val, desc)
+
+        if intent == 'click':
+            ok, msg = click_by_anchor(tid, anchor, desc)
+            time.sleep(1.5)
+            return ok, msg
+
+        if intent == 'observe':
+            return True, 'observed'
+
+        if intent == 'extract':
+            # 优先 expect (数据断言, 可能是 DOM 值 或 fetch API)
+            if step.get('expect'):
+                return _check_expect(tid, step['expect'])
+            # 否则走旧语义: body 文本含 check 关键词
+            check = step.get('check', '')
+            v, _ = eval_js(tid, "document.body.innerText", timeout=10)
+            body = str(v) if v else ''
+            if check:
+                hit = check in body
+                return (hit, '命中:%s' % check if hit else '未命中:%s' % check)
+            return True, 'extracted(no check)'
+
+        if intent == 'assert':
+            return _check_assert(tid, step.get('check', ''))
+
+    except Exception as e:
+        return False, str(e)[:60]
+    return False, '未知intent:%s' % intent
+
+
+# ===== 数据断言 (web-access 核心增量: eval DOM 值 或 eval fetch API) =====
+def _check_expect(tid, expect):
+    """数据断言. expect 支持两种模式:
+    1. {dom: "css", jsonpath: "text", equals: "湖南石门"}  — 读 DOM 元素值
+    2. {fetch: "/api/...", jsonpath: "code", equals: 200}   — 调后端接口验字段
+    jsonpath 用点路径 (data.total / rows[0].name), 轻量实现不引依赖.
+    """
+    # 模式 1: DOM 值断言
+    if 'dom' in expect:
+        css = expect['dom']
+        jp = expect.get('jsonpath', 'text').lstrip('$.').lstrip('.')
+        want = expect.get('equals')
+        code = "(function(){const el=document.querySelector(\"%s\");if(!el)return null;" \
+               "const v=el.innerText||el.value||el.textContent;return v.trim()})()" % css.replace('"', '\\"')
+        v, e = eval_js(tid, code)
+        actual = str(v).strip() if v else ''
+        ok = (actual == str(want)) if want is not None else bool(actual)
+        return ok, '%s=%s(期望%s)' % (jp, _short(actual), _short(want))
+
+    # 模式 2: API 断言 (从页面上下文 fetch, 带登录态)
+    if 'fetch' in expect:
+        endpoint = expect['fetch']
+        jp = expect.get('jsonpath', '').lstrip('$.').lstrip('.')
+        want = expect.get('equals')
+        method = expect.get('method', 'POST')
+        body = expect.get('body', {})
+        body_js = json.dumps(body, ensure_ascii=False)
+        # 注意: 裸 fetch 可能 401 (没带 token). 优先建议用 dom 断言.
+        code = ("(async()=>{try{const r=await fetch('%s',{method:'%s',headers:{'Content-Type':'application/json'}," \
+                "body:%s});const d=await r.json();return JSON.stringify(d)}catch(e){return 'ERR:'+e.message}})()" % (
+                    endpoint, method, body_js))
+        v, e = eval_js(tid, code, timeout=20)
+        if e:
+            return False, 'fetch失败:%s' % e
+        try:
+            data = json.loads(v) if isinstance(v, str) else v
+        except Exception:
+            return False, '响应非JSON:%s' % _short(v)
+        actual = _dig(data, jp)
+        ok = (actual == want)
+        return ok, '%s=%s(期望%s)' % (jp, _short(actual), _short(want))
+
+    return True, 'expect(无dom/fetch字段)'
+
+
+def _check_assert(tid, check):
+    """镜像 autotest_run.py 的 assert: rows>/rows==/url_contains/input_empty."""
+    if not check:
+        return True, 'assert(no check)'
+    if check.startswith('rows>'):
+        n = int(check[5:])
+        wait_table_stable(tid, max_wait=6)
+        v, _ = eval_js(tid, "document.querySelectorAll('.vxe-body--row').length", timeout=5)
+        try:
+            rows = int(v)
+        except (TypeError, ValueError):
+            rows = 0
+        return (rows > n, '行数%d>%d' % (rows, n) if rows > n else '行数%d<=%d' % (rows, n))
+    if check.startswith('rows=='):
+        n = int(check[6:])
+        wait_table_stable(tid, max_wait=6)
+        v, _ = eval_js(tid, "document.querySelectorAll('.vxe-body--row').length", timeout=5)
+        try:
+            rows = int(v)
+        except (TypeError, ValueError):
+            rows = 0
+        return (rows == n, '行数=%d' % rows)
+    if check.startswith('url_contains'):
+        kw = check.split(':', 1)[1] if ':' in check else ''
+        info = page_info(tid)
+        cur = info.get('url', '')
+        return (kw in cur, 'url含%s' % kw if kw in cur else 'url不含%s' % kw)
+    if check.startswith('input_empty:'):
+        field = check.split(':', 1)[1]
+        v, _ = eval_js(tid, "(document.querySelector(\"input[name='%s']\")||{}).value||''" % field, timeout=5)
+        val = str(v) if v else ''
+        return (val == '', '框值=%r' % val)
+    return True, 'assert(未知check)'
+
+
+def _dig(obj, path):
+    """简单点路径取值: data.total / rows[0].name."""
+    if not path:
+        return obj
+    cur = obj
+    for part in re.split(r'\.', path):
+        m = re.match(r'([^\[]+)(\[(\d+)\])?', part)
+        if not m:
+            return None
+        key = m.group(1)
+        idx = m.group(3)
+        if isinstance(cur, dict) and key in cur:
+            cur = cur[key]
+        elif isinstance(cur, list) and key.isdigit() and int(key) < len(cur):
+            cur = cur[int(key)]
+        else:
+            return None
+        if idx is not None and isinstance(cur, list) and int(idx) < len(cur):
+            cur = cur[int(idx)]
+    return cur
+
+
+def _short(v, n=30):
+    s = str(v)
+    return s if len(s) <= n else s[:n] + '...'
+
+
+# ===== 跑一个用例 (镜像 autotest_run.py run_case) =====
+def run_case(tid, case):
+    cid = case.get('case_id', '?')
+    steps = case.get('steps', [])
+    details = []
+    all_ok = True
+    for step in steps:
+        ok, detail = exec_step(tid, step)
+        details.append('%s:%s(%s)' % (step.get('intent', '?'), '✓' if ok else '✗', detail))
+        if not ok:
+            all_ok = False
+            break
+    return {'case_id': cid, 'title': case.get('title', ''), 'pass': all_ok,
+            'expected': case.get('expected', ''), 'details': details}
+
+
+# ===== 沉淀锚点 (与 autotest_run.py 一致) =====
+def collect_anchors(tid):
+    js = '''()=>{const out=[];document.querySelectorAll('button,.ant-tabs-tab,input[placeholder],input[name],[role=tab]').forEach(el=>{const name=(el.innerText||el.getAttribute('placeholder')||el.getAttribute('name')||'').trim();const role=el.getAttribute('role')||(el.tagName==='INPUT'?'textbox':el.tagName.toLowerCase());if(name&&name.length<25&&!/^[0-9\s]+$/.test(name))out.push({role,name});});return out.slice(0,30);}'''
+    v, _ = eval_js(tid, js, timeout=10)
+    els = v if isinstance(v, list) else []
+    seen = set(); uniq = []
+    for el in els:
+        k = (el.get('role'), el.get('name'))
+        if k not in seen:
+            seen.add(k); uniq.append(el)
+    return uniq
+
+
+def sediment(page_path, anchors):
+    if not anchors:
+        return 0
+    try:
+        pj = json.dumps({'url': page_path, 'elements': [
+            {'role': a.get('role'), 'name': a.get('name'), 'desc': a.get('name')} for a in anchors
+        ]}, ensure_ascii=False)
+        subprocess.run([sys.executable, AUTOLEARN, 'learn', '--url', page_path, '--page-json', pj],
+                       capture_output=True, timeout=20)
+        return len(anchors)
+    except Exception:
+        return 0
+
+
+# ===== 友好提示 + 自动回退 =====
+def _print_not_ready(reason, explicit=False):
+    print('')
+    print('╔══ web-access 引擎未就绪 ═════════════════════════════════════')
+    print('║ 原因: %s' % str(reason)[:56])
+    print('║ ────────────────────────────────────────────────────────────')
+    print('║ 怎么修:')
+    print('║   1. 打开 Chrome, 地址栏输入 chrome://inspect/#remote-debugging')
+    print('║   2. 勾选 "Allow remote debugging for this browser instance"')
+    print('║   3. 确保 Chrome 开着 (不用切前台, 后台即可)')
+    if not explicit:
+        print('║')
+        print('║ ✅ 已自动回退: 改用 playwright 引擎 (autotest_run.py)')
+        print('║    playwright 开箱即用, 不需要开调试开关.')
+        print('║    web-access 是可选增强 —— 复用已登录 Chrome, 验证码零接触.')
+    print('╚══════════════════════════════════════════════════════════════')
+    print('')
+
+
+# ===== 主 =====
+def main():
+    ap = argparse.ArgumentParser()
+    ap.add_argument('--task', required=True, help='任务文件 JSON')
+    ap.add_argument('--browser', default='chrome', help='浏览器 (chrome/edge, 默认 chrome)')
+    ap.add_argument('--url', default='', help='临时覆盖 base url')
+    ap.add_argument('--force', action='store_true', help='强制 web-access, 没装不回退')
+    args = ap.parse_args()
+
+    global BASE
+    if args.url:
+        BASE = args.url.rstrip('/')
+
+    task = json.load(open(args.task, encoding='utf-8'))
+    page_path = task['page_path']
+    cases = task.get('cases', [])
+
+    # 1. 启动 proxy + 连浏览器 (没开调试开关 → 友好提示 + 回退)
+    ok, msg = start_proxy(args.browser)
+    if not ok:
+        _print_not_ready(msg, explicit=args.force)
+        if args.force:
+            r = {'engine': 'web-access', 'error': 'WEBAccess_NOT_READY', 'case_results': [], 'page': page_path}
+            print(json.dumps(r, ensure_ascii=False)); sys.exit(1)
+        print('>>> 自动转用 playwright 引擎 ...')
+        runner = os.path.join(SCRIPTS, 'autotest_run.py')
+        rc = subprocess.run([sys.executable, runner, '--task', args.task], cwd=PROJECT).returncode
+        sys.exit(rc)
+
+    result = {'page': page_path, 'page_name': task.get('page_name', ''),
+              'engine': 'web-access', 'case_results': [], 'anchors_count': 0, 'error': '', 'video': ''}
+    t0 = time.time()
+    tid = None
+    try:
+        # 2. 开目标页 (新 tab, 复用已登录 Chrome 的登录态)
+        tid = new_tab(BASE + page_path, wait=4)
+        if not tid:
+            result['error'] = 'OPEN_FAIL'; print(json.dumps(result, ensure_ascii=False)); sys.exit(1)
+
+        logged_in, cur_url = check_logged_in(tid)
+        if not logged_in:
+            result['error'] = 'NOT_LOGGED_IN: 请先在 Chrome 手动登录 %s' % BASE
+            print(json.dumps(result, ensure_ascii=False)); sys.exit(1)
+        print('  已连接: %s' % cur_url)
+
+        # 3. 跑每个用例
+        # 第一个用例已在目标页 (new_tab 已导航), 无需重导; 后续用例才重导保证干净状态.
+        # wait_table_stable 首次等久点 (表格可能没加载), 后续缩到 4s (页面已热).
+        first = True
+        for case in cases:
+            if not first:
+                navigate(tid, BASE + page_path, wait=2)
+            first = False
+            wait_table_stable(tid, max_wait=8 if result['case_results'] == [] else 4)
+            cr = run_case(tid, case)
+            result['case_results'].append(cr)
+            tag = '✓' if cr['pass'] else '✗'
+            print('  %s %s %s' % (tag, cr['case_id'], cr['title'][:30]))
+            for d in cr['details']:
+                print('      %s' % d)
+
+        # 4. 沉淀
+        anc = collect_anchors(tid)
+        result['anchors_count'] = len(anc)
+        sediment(page_path, anc)
+
+    except Exception as e:
+        result['error'] = str(e)[:100]
+    finally:
+        if tid:
+            close_tab(tid)
+
+    result['duration'] = round(time.time() - t0, 1)
+    passed = sum(1 for c in result['case_results'] if c['pass'])
+    total = len(result['case_results'])
+    result['summary'] = '%d/%d 通过' % (passed, total)
+    print()
+    print('汇总: %s 锚点%d 耗时%ss %s' % (result['summary'], result['anchors_count'], result['duration'], result['error']))
+    out = args.task.replace('.json', '_webaccess.json')
+    json.dump(result, open(out, 'w', encoding='utf-8'), ensure_ascii=False, indent=2)
+    print('结果: %s' % out)
+
+
+if __name__ == '__main__':
+    main()

package/templates/qoder/scripts/{benchmark.py → validation/test/benchmark.py}

@@ -1,5 +1,17 @@
 #!/usr/bin/env python3
 # -*- coding: utf-8 -*-
+# v3.0 路径自举: 引导到 common/bootstrap, 统一 sys.path 逻辑
+import os as _o, sys as _s
+_f = _o.path.abspath(__file__)
+for _ in range(10):
+    _f = _o.path.dirname(_f)
+    _cp = _o.path.join(_f, 'foundation')
+    if _o.path.isfile(_o.path.join(_cp, 'bootstrap.py')):
+        break
+if _cp not in _s.path: _s.path.insert(0, _cp)
+from bootstrap import setup; setup()
+from foundation.core.paths import get_repo_root
+
 """
 benchmark.py - QODER 流水线性能基准 (性能优化 A7)
 
@@ -21,7 +33,7 @@ Usage:
 """
 
 import argparse
-from common.paths import BUILD_CACHE_DIR
+from foundation.core.paths import BUILD_CACHE_DIR
 import json
 import os
 import sys
@@ -36,7 +48,7 @@ except (AttributeError, TypeError, OSError):
 THIS_DIR = os.path.dirname(os.path.abspath(__file__))  # .qoder/scripts
 sys.path.insert(0, THIS_DIR)
 # BASE = repo root = scripts 的上两级 (.qoder/scripts -> .qoder -> repo)
-BASE = os.path.dirname(os.path.dirname(THIS_DIR))
+BASE = get_repo_root()
 
 INDEX_DIR = os.path.join(BASE, 'data', 'index')
 
@@ -65,7 +77,7 @@ def bench_search():
     results['keyword_count'] = len(ki)
 
     # 2. 全扫描 (旧逻辑)
-    from search_index import _match_keyword
+    from domain.kg.search.search_index import _match_keyword
     queries = ['insurance', 'attendance', 'salary', 'vehicle', 'user']
     def full_scan():
         for q in queries:
@@ -75,7 +87,7 @@ def bench_search():
     results['full_scan_5q_ms'] = measure(full_scan, repeat=5)[0] / 5
 
     # 3. 倒排索引构建 (首次)
-    from common.search_engine import get_inverted_index, search_keywords_fast, build_inverted_index
+    from search_engine import get_inverted_index, search_keywords_fast, build_inverted_index
     # 清缓存测构建
     cache_path = os.path.join(INDEX_DIR, '.inverted-cache.json')
     if os.path.exists(cache_path):
@@ -103,7 +115,7 @@ def bench_search():
 
 def bench_cnmap():
     """CN_MAP 查询基准。"""
-    from common.terms import CN_MAP, BUSINESS_PATH_MAP, get_cn_map_with_auto
+    from foundation.integrations.terms import CN_MAP, BUSINESS_PATH_MAP, get_cn_map_with_auto
     results = {}
     results['cn_map_static_count'] = len(CN_MAP)
     results['business_path_map_count'] = len(BUSINESS_PATH_MAP)