@hupan56/wlkj 2.7.12 → 3.1.2

package/templates/qoder/scripts/domain/kg/kg_capabilities.py

@@ -0,0 +1,182 @@
+# -*- coding: utf-8 -*-
+"""
+kg_capabilities.py - 知识图谱共享查询能力 (Strategy 模式, 消除 CLI↔MCP 重复)
+
+v3.0 f3: kg.py (CLI) 和 kg_mcp_server.py (MCP) 有近字重复的查询逻辑
+(workflow/hop/wiki/design_system 等)。本模块提取共享查询, 两边都调它,
+消除重复, 且保证 CLI 和 MCP 行为完全一致。
+
+核心 API (都接收已打开的 duckdb con, 由调用方管理连接):
+    workflow_query(con, q) -> str    业务流程链查询
+    # hop_query / wiki_query / design_system_query 后续添加
+
+设计: 查询逻辑 (数据清洗/SQL/格式化) 在此单一真源;
+      kg.py 和 kg_mcp_server 只负责"开连接 + 调本模块 + 返回"。
+"""
+
+__all__ = ["workflow_query", "hop_query", "wiki_query"]
+
+
+def workflow_query(con, q):
+    """workflows 表查询 + 脏数据清洗 + 中文兜底。
+
+    Args:
+        con: 已打开的 duckdb 连接 (read_only)
+        q: 模块名 (英文 assets/cost/hr 或中文触发兜底)
+
+    Returns:
+        格式化的业务流程链文本
+    """
+    q = (q or '').strip()
+    # 1) 精确 + 模糊匹配 workflows.module（英文）
+    hits = con.execute(
+        "SELECT DISTINCT module FROM workflows WHERE module = ? OR module LIKE '%' || ? || '%' ORDER BY module",
+        [q.lower(), q.lower()]
+    ).fetchall()
+    if not hits:
+        # 2) 中文兜底: 查 entities.cn 返回英文候选
+        cn_hits = con.execute(
+            "SELECT canonical, cn FROM entities WHERE type IN ('FEATURE','MODULE') AND cn <> '' AND (cn LIKE '%' || ? OR cn LIKE '%' || ? || '%')",
+            [q, q]
+        ).fetchall()
+        if cn_hits:
+            cands = ', '.join('%s(%s)' % (c[0], c[1]) for c in cn_hits[:8])
+            return "未在 workflows 表找到模块 '%s'。但 entities 表有相近中文名，可换这些英文模块名重试:\n  %s" % (q, cands)
+        all_mods = [r[0] for r in con.execute("SELECT DISTINCT module FROM workflows ORDER BY module").fetchall()]
+        return "未找到模块 '%s'。现有模块: %s" % (q, ', '.join(all_mods))
+
+    out = []
+    for (mod,) in hits:
+        # 清洗: 过滤空端点和模板泄漏按钮
+        rows = con.execute("""
+            SELECT state, seq, button, endpoint FROM workflows
+            WHERE module = ?
+              AND endpoint <> 'None' AND endpoint <> ''
+              AND button NOT LIKE '{{%'
+            ORDER BY seq, state
+        """, [mod]).fetchall()
+        if not rows:
+            continue
+        out.append('=== %s (%d 步, 已清洗脏数据) ===' % (mod, len(rows)))
+        cur_seq = -1
+        for st, sq, btn, ep in rows:
+            if sq != cur_seq:
+                out.append('  [%d] %s' % (sq, st))
+                cur_seq = sq
+            out.append('        %-16s -> %s' % (btn, ep))
+        out.append('')
+    return '\n'.join(out).strip() or "模块 '%s' 清洗后无有效操作链" % q
+
+
+def hop_query(con, symbol, depth, multi_hop_fn=None):
+    """多跳遍历: 别名解析(三级兜底) + 递归CTE + 按depth分组展示。
+
+    Args:
+        con: duckdb 连接
+        symbol: 起始符号 (中文/端点/函数名, 自动别名解析)
+        depth: 最大跳数 1-5
+        multi_hop_fn: kg_duckdb.multi_hop 函数 (避免硬依赖)
+    Returns:
+        格式化的多跳遍历文本
+    """
+    # 1) 别名解析（三级兜底）
+    start_id = symbol
+    resolved = con.execute(
+        "SELECT entity_id FROM aliases WHERE alias = ?", [symbol.lower()]
+    ).fetchone()
+    if resolved:
+        start_id = resolved[0]
+    else:
+        fuzzy = con.execute(
+            "SELECT id FROM entities WHERE id = ? OR id LIKE '%' || ? || '%' LIMIT 1",
+            [symbol, symbol]
+        ).fetchone()
+        if fuzzy:
+            start_id = fuzzy[0]
+        else:
+            edge_hit = con.execute(
+                "SELECT from_id FROM edges WHERE from_id LIKE '%' || ? || '%' LIMIT 1",
+                [symbol]
+            ).fetchone()
+            if edge_hit:
+                start_id = edge_hit[0]
+
+    # 2) 递归 CTE 遍历
+    if multi_hop_fn is None:
+        from domain.kg.storage import kg_duckdb as K
+        multi_hop_fn = K.multi_hop
+    try:
+        hops = multi_hop_fn(con, start_id, max_depth=depth)
+    except Exception as e:
+        return "遍历出错 (起点=%s): %s" % (start_id, str(e))
+
+    if not hops or (len(hops) == 1 and hops[0][1] == 0):
+        return "从 '%s' (解析为 %s) 出发无可达节点。可能该符号不在图谱中，或无出边。" % (symbol, start_id)
+
+    # 3) 按 depth 分组展示
+    from collections import defaultdict
+    by_depth = defaultdict(list)
+    for eid, d in hops:
+        by_depth[d].append(eid)
+
+    out = ['=== 多跳遍历: %s (解析为 %s) ===' % (symbol, start_id)]
+    out.append('可达 %d 个实体, 最大 %d 跳:' % (len(hops), max(by_depth.keys())))
+    out.append('')
+    for d in sorted(by_depth.keys()):
+        eids = by_depth[d]
+        out.append('[第 %d 跳] %d 个实体:' % (d, len(eids)))
+        shown = 0
+        for eid in eids:
+            if shown >= 12:
+                out.append('  ... 还有 %d 个' % (len(eids) - 12))
+                break
+            info = con.execute(
+                "SELECT type, canonical, cn FROM entities WHERE id = ?", [eid]
+            ).fetchone()
+            if info:
+                t, c, cn = info
+                label = '%s / %s' % (c, cn) if cn and cn != c else c
+                out.append('  %-14s %-12s %s' % ('[%s]' % t, eid[:40], label))
+            else:
+                out.append('  %-14s %s' % ('[edge]', eid[:60]))
+            shown += 1
+        out.append('')
+    return '\n'.join(out).strip()
+
+def wiki_query(kw):
+    """Repo Wiki 搜索 + 格式化 (CLI/MCP 共用)。
+
+    Args:
+        kw: 搜索关键词
+    Returns:
+        格式化的 wiki 匹配结果文本
+    """
+    import os
+    from foundation.core.paths import DATA_INDEX_DIR as _idx, PROJECT_ROOT as _base
+    wiki_path = os.path.join(str(_idx), 'wiki-index.json')
+    if not os.path.isfile(wiki_path):
+        return ("wiki-index.json 不存在。请手动构建(一次性):\n"
+                "  python .qoder/scripts/domain/kg/search/search_index.py wiki")
+    try:
+        from repowiki import search_wiki
+        from foundation.integrations.terms import get_cn_map_with_auto
+    except ImportError as e:
+        return "Wiki 搜索模块缺失: %s" % str(e)
+    code_dir = os.path.join(_base, 'data', 'code')
+    try:
+        cn_map = get_cn_map_with_auto()
+        results = search_wiki(kw, str(_idx), code_dir, cn_map)
+    except Exception as e:
+        return "Wiki 搜索出错: %s" % str(e)
+    if not results:
+        return "No wiki pages match: %s" % kw
+    out = ['Wiki pages matching: %s (%d found)' % (kw, len(results)), '']
+    for i, r in enumerate(results, 1):
+        out.append('  %d. [%s] %s' % (i, r.get('project', '?'), r.get('title', '')))
+        deps = r.get('dependent_files') or []
+        if deps:
+            out.append('     依赖: %s%s' % (', '.join(deps[:2]),
+                      (' (共 %d 个文件)' % len(deps)) if len(deps) > 2 else ''))
+        if r.get('md_path'):
+            out.append('     文档: %s' % r['md_path'])
+    return '\n'.join(out)

package/templates/qoder/scripts/{context_pack.py → domain/kg/search/context_pack.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
+
 """
 QODER Context Pack - 一次调用返回写 PRD/原型所需的全部上下文
 
@@ -19,7 +31,7 @@ Usage:
 """
 
 import os
-from common.paths import BUILD_CACHE_DIR
+from foundation.core.paths import BUILD_CACHE_DIR
 import sys
 import json
 import time
@@ -34,9 +46,9 @@ except (AttributeError, TypeError, OSError, IOError):
         pass
 
 sys.path.insert(0, os.path.dirname(os.path.abspath(__file__)))
-from common.terms import CN_MAP, BUSINESS_PATH_MAP, CN_TO_EN, get_platform_map
+from foundation.integrations.terms import CN_MAP, BUSINESS_PATH_MAP, CN_TO_EN, get_platform_map, expand_chinese_query
 
-BASE = os.path.dirname(os.path.dirname(os.path.dirname(os.path.abspath(__file__))))
+BASE = get_repo_root()
 INDEX_DIR = os.path.join(BASE, 'data', 'index')
 PLATFORM_MAP = get_platform_map()
 MIN_FUZZY_LEN = 4
@@ -45,7 +57,7 @@ MIN_FUZZY_LEN = 4
 def load(filename):
     """Load index from DuckDB (preferred) or JSON fallback."""
     try:
-        from search_index import _duckdb_load
+        from domain.kg.search.search_index import _duckdb_load
         result = _duckdb_load(filename)
         if result is not None:
             return result
@@ -62,11 +74,14 @@ def load(filename):
 
 
 def expand_query(query):
-    words = []
-    if query in CN_MAP:
-        words.extend(CN_MAP[query].split())
-    if query in BUSINESS_PATH_MAP:
-        words.extend(p.strip('/-').lower() for p in BUSINESS_PATH_MAP[query])
+    """词扩展: 委托给 expand_chinese_query (三级匹配: 精确/分词/兜底)。
+
+    旧版只查精确匹配, 复合词如 "保险单OCR" / "异常记录" 查不到就 0 命中。
+    现在用 expand_chinese_query, 能拆 "保险单OCR" → "保险" + "ocr",
+    "异常记录" → "异常" + "记录", 大幅减少单词查询的 0 命中。
+    """
+    words = expand_chinese_query(query)
+    # 兜底: 拆不出词就用原词小写 (让英文索引做子串匹配)
     return sorted(set(w.lower() for w in words)) or [query.lower()]
 
 
@@ -132,7 +147,7 @@ def _save_cache(key, content):
         os.replace(tmp, cache_path)
         # LRU 清理: 保留最新 30 个, 删更老的
         try:
-            from search_index import _cleanup_cache_dir
+            from domain.kg.search.search_index import _cleanup_cache_dir
             _cleanup_cache_dir(cache_dir, 'ctx-cache-', max_files=30)
         except Exception:
             pass
@@ -212,6 +227,21 @@ def _compute_and_print(query, platform, page_type):
         print('- {} (命中: {})  -> {}'.format(f, ','.join(sorted(kws)[:3]), full))
     if not hits:
         print('- (无命中 — 换业务词重试, 或该功能为全新)')
+    # 语义补充: 关键词命中稀疏时, 用 bge-small-zh 向量检索补一轮 (中→英跨语言)。
+    # 模型未装/无 embedding → 静默跳过, 不报错。
+    if len(hits) <= 2:
+        try:
+            from domain.kg.graph.kg_semantic import semantic_search_files, filter_semantic_by_platform
+            sem = semantic_search_files(query, top_k=8)
+            if sem:
+                filtered = filter_semantic_by_platform(sem, platform, BASE)
+                if filtered:
+                    print('- [语义补充] (关键词命中少, 用向量检索跨语言补充):')
+                    for fp, cn, score in filtered[:6]:
+                        full = 'data/code/' + fp
+                        print('  - %.2f %s -> %s' % (score, cn[:20], full))
+        except Exception:
+            pass
 
     # ---- 2. 同类型页面真实示例 ----
     si = load('ui-style.json') or {}
@@ -279,8 +309,8 @@ def _compute_and_print(query, platform, page_type):
     code_dir = os.path.join(BASE, 'data', 'code')
     wiki_results = []
     try:
-        from common.repowiki import search_wiki, build_wiki_index
-        from common.terms import get_cn_map_with_auto
+        from repowiki import search_wiki, build_wiki_index
+        from foundation.integrations.terms import get_cn_map_with_auto
         wiki_idx_path = os.path.join(BUILD_CACHE_DIR, 'wiki-index.json')
         if not os.path.isfile(wiki_idx_path):
             build_wiki_index(code_dir, wiki_idx_path)
@@ -315,10 +345,9 @@ def _compute_and_print(query, platform, page_type):
 
     # ── 8. 数据库表结构 (从 kg_db.duckdb, 合并了原来要单独调 MCP 的查询) ──
     try:
-        import duckdb as _ddb2
-        db_path = os.path.join(INDEX_DIR, 'kg_db.duckdb')
-        if os.path.isfile(db_path):
-            con = _ddb2.connect(db_path, read_only=True)
+        from domain.kg.search.search_index import _get_kg_db
+        con = _get_kg_db()  # 惰性缓存的只读连接, 不在此 close (供后续调用复用)
+        if con is not None:
             # 关键词匹配表名 (英文词或snake_case, camelCase自动转)
             kw_lower = query.lower()
             # camelCase → snake_case (qualityCase → quality_case)
@@ -349,7 +378,6 @@ def _compute_and_print(query, platform, page_type):
                     if ents:
                         print('  {} → Entity: {}'.format(
                             tn, ', '.join(e[0] for e in ents)))
-            con.close()
     except Exception:
         pass  # kg_db不存在/被占用 → 跳过, 不影响其它段