@hupan56/wlkj 3.3.14 → 3.3.15

package/templates/qoder/scripts/domain/kg/extract/build_goal5.py

@@ -1,95 +0,0 @@
-# -*- coding: utf-8 -*-
-"""GOAL 5 统一入口: L3 弱关联推理 + C 方案重算。
-
-CLI:
-    # 限量推理弱关联 (前500个无强关联实体)
-    python -m domain.kg.extract.build_goal5 --infer --limit 500
-    # C方案重算 (过期弱边续期/删除)
-    python -m domain.kg.extract.build_goal5 --recompute
-    # 升格 (删冗余弱边)
-    python -m domain.kg.extract.build_goal5 --promote
-"""
-from __future__ import annotations
-import os, sys
-
-_THIS = os.path.dirname(os.path.abspath(__file__))
-for _i in range(8):
-    _p = os.path.dirname(_THIS)
-    if os.path.isfile(os.path.join(_p, 'foundation', 'bootstrap.py')):
-        sys.path.insert(0, _p); break
-    _THIS = _p
-
-
-def run_infer(limit: int = 500, logger=print):
-    """推理弱关联 (限量)。"""
-    from domain.kg.extract.inference import weak_link
-    from domain.kg.extract.java import pg_upsert
-    from sqlalchemy import text
-
-    engine, _ = pg_upsert._get_pg_engine()
-    if engine is None:
-        return {'error': 'PG 不可用'}
-
-    with engine.connect() as conn:
-        # 取无强关联的代码实体
-        entities = conn.execute(text(f"""
-            SELECT id, repo_id, type, canonical, cn, props FROM entities
-            WHERE type IN ('FUNCTION','BUTTON','ENDPOINT','CONTROLLER','SERVICE',
-                           'MAPPER','ENTITY','TABLE','COLUMN','FEATURE')
-              AND id NOT IN (
-                SELECT from_id FROM edges
-                WHERE source IN ('explicit','commit')
-                  AND edge_type IN ('implements_req','anchors')
-              )
-            LIMIT {int(limit)}
-        """)).all()
-        # 取全部 anchors
-        anchors = conn.execute(text("SELECT id, title, kind FROM anchors")).all()
-
-    ent_list = [{'id': r[0], 'repo_id': r[1], 'type': r[2],
-                 'canonical': r[3], 'cn': r[4], 'props': r[5] or {}} for r in entities]
-    anchor_list = [{'id': r[0], 'title': r[1], 'kind': r[2]} for r in anchors]
-    logger('  [goal5] %d 实体待推理, %d 锚点' % (len(ent_list), len(anchor_list)))
-
-    edges, st = weak_link.infer_batch(ent_list, anchor_list, logger=logger)
-
-    # 写 PG (用 pg_upsert, 它会处理 expires_at/fingerprint 字段)
-    upsert_r = pg_upsert.upsert_batch([], edges) if edges else {'ok': False, 'edges_ok': 0}
-    st['upsert'] = upsert_r
-    return st
-
-
-def run_recompute(logger=print):
-    """C 方案重算。"""
-    from domain.kg.extract.inference import recompute
-    return recompute.recompute_expired(logger=logger)
-
-
-def run_promote(logger=print):
-    """弱边升格。"""
-    from domain.kg.extract.inference import promote
-    return promote.promote_strong_over_weak(logger=logger)
-
-
-if __name__ == '__main__':
-    import argparse, json
-    ap = argparse.ArgumentParser()
-    ap.add_argument('--infer', action='store_true')
-    ap.add_argument('--recompute', action='store_true')
-    ap.add_argument('--promote', action='store_true')
-    ap.add_argument('--limit', type=int, default=500)
-    args = ap.parse_args()
-    try:
-        from dotenv import load_dotenv
-        for p in ['wlinkj-workspace/backend/.env', '.env']:
-            if os.path.isfile(p): load_dotenv(p); break
-    except: pass
-
-    if args.infer:
-        print(json.dumps(run_infer(args.limit), ensure_ascii=False, indent=2, default=str))
-    elif args.recompute:
-        print(json.dumps(run_recompute(), ensure_ascii=False, indent=2, default=str))
-    elif args.promote:
-        print(json.dumps(run_promote(), ensure_ascii=False, indent=2, default=str))
-    else:
-        ap.print_help()

package/templates/qoder/scripts/domain/kg/extract/db/__init__.py

@@ -1,8 +0,0 @@
-# -*- coding: utf-8 -*-
-"""DB schema 提取器 (GOAL 3)。
-
-子模块:
-  schema_extractor  连 information_schema → 灌 db_tables/db_columns
-  fk_extractor      外键关系 → references 边
-  data_profile      数据特征 (null率/枚举分布, 选做)
-"""

package/templates/qoder/scripts/domain/kg/extract/db/data_profile.py

@@ -1,22 +0,0 @@
-# -*- coding: utf-8 -*-
-"""数据特征 profile (选做): 表行数、列 null 率、枚举分布。
-
-价值: 让 KG 知道"这张表有 50 万行""这个字段 null 率 80%"——数据质量知识。
-本轮 best-effort, 连不上 DB 跳过。
-"""
-from __future__ import annotations
-from . import schema_extractor as _se
-
-
-def profile_tables(repo_id: str = 'fywl-ics', sample_size: int = 100, logger=print):
-    """采样表的数据特征。Returns: list[dict] 连不上返回 []。"""
-    if not _se._can_connect():
-        return []
-    try:
-        import pymysql
-    except ImportError:
-        return []
-    # 本轮简化: 只返回行数 (table_rows 已在 schema_extractor 取过)
-    # 完整版可加: SELECT COUNT(*), 列 null 率, DISTINCT 值分布
-    logger('  [db-profile] 本轮跳过详细 profile (行数已在 schema_extractor 取)')
-    return []

package/templates/qoder/scripts/domain/kg/extract/db/fk_extractor.py

@@ -1,55 +0,0 @@
-# -*- coding: utf-8 -*-
-"""外键关系提取器: KEY_COLUMN_USAGE → references 边。
-
-让 table_impact 真正能做"改表影响谁"的跨表追溯。
-"""
-from __future__ import annotations
-from . import schema_extractor as _se
-
-
-def extract_foreign_keys(repo_id: str = 'fywl-ics', logger=print):
-    """连 MySQL → 提取外键关系 → references 边。
-
-    Returns: [{from_id, to_id, edge_type:'references', ...}] 连不上返回 []
-    """
-    if not _se._can_connect():
-        return []
-    try:
-        import pymysql
-    except ImportError:
-        return []
-
-    conn = pymysql.connect(host=_se.MYSQL_HOST, port=_se.MYSQL_PORT, user=_se.MYSQL_USER,
-                           password=_se.MYSQL_PASSWORD, charset='utf8mb4',
-                           cursorclass=pymysql.cursors.DictCursor,
-                           connect_timeout=5, read_timeout=30)
-    edges = []
-    try:
-        cur = conn.cursor()
-        cur.execute("SELECT schema_name FROM information_schema.schemata "
-                    "WHERE schema_name LIKE %s", (_se.DB_NAME_PREFIX + '%',))
-        dbs = [r['schema_name'] for r in cur.fetchall()]
-        for db in dbs:
-            # 外键关系
-            cur.execute("""
-                SELECT TABLE_NAME, COLUMN_NAME, REFERENCED_TABLE_NAME, REFERENCED_COLUMN_NAME
-                FROM information_schema.KEY_COLUMN_USAGE
-                WHERE TABLE_SCHEMA=%s AND REFERENCED_TABLE_NAME IS NOT NULL
-            """, (db,))
-            for r in cur.fetchall():
-                from_tbl = r['TABLE_NAME'].lower()
-                to_tbl = r['REFERENCED_TABLE_NAME'].lower()
-                edges.append({
-                    'from_id': 'TBL:%s:%s' % (repo_id, from_tbl),
-                    'to_id': 'TBL:%s:%s' % (repo_id, to_tbl),
-                    'edge_type': 'references',
-                    'from_repo': repo_id, 'to_repo': repo_id,
-                    'confidence': 1.0, 'source': 'db_fk',
-                    'props': {'column': r['COLUMN_NAME'],
-                              'ref_column': r['REFERENCED_COLUMN_NAME'],
-                              'schema': db},
-                })
-        logger('  [db-fk] %d 条外键关系' % len(edges))
-    finally:
-        conn.close()
-    return edges

package/templates/qoder/scripts/domain/kg/extract/db/schema_extractor.py

@@ -1,90 +0,0 @@
-# -*- coding: utf-8 -*-
-"""连 MySQL information_schema → 灌 db_tables / db_columns。
-
-复用 kg_build_db.py 的 MySQL 连接配置 (内网 10.54.6.9, fywl_ics_* 库)。
-连不上内网时 best-effort 返回空, 不阻塞 (GOAL 3 主链路是 mapper XML)。
-"""
-from __future__ import annotations
-import os
-import sys
-import socket
-
-_THIS = os.path.dirname(os.path.abspath(__file__))
-for _i in range(8):
-    _p = os.path.dirname(_THIS)
-    if os.path.isfile(os.path.join(_p, 'foundation', 'bootstrap.py')):
-        sys.path.insert(0, _p); break
-    _THIS = _p
-
-# 复用 kg_build_db 的配置
-MYSQL_HOST = os.environ.get('MYSQL_HOST', '10.54.6.9')
-MYSQL_PORT = int(os.environ.get('MYSQL_PORT', '3306'))
-MYSQL_USER = os.environ.get('MYSQL_USER', 'fywl_ics_test')
-MYSQL_PASSWORD = os.environ.get('MYSQL_PASSWORD', 'Fywl_ics_test56476')
-DB_NAME_PREFIX = os.environ.get('DB_NAME_PREFIX', 'fywl_ics_')
-
-
-def _can_connect(timeout=3):
-    """探测内网 MySQL 是否可达。"""
-    try:
-        with socket.create_connection((MYSQL_HOST, MYSQL_PORT), timeout=timeout):
-            return True
-    except Exception:
-        return False
-
-
-def extract_all_schema(repo_id: str = 'fywl-ics', logger=print):
-    """连 MySQL → 灌 db_tables + db_columns。
-
-    Returns: (tables: list[dict], columns: list[dict]) 连不上返回 ([], [])
-    """
-    if not _can_connect():
-        logger('  [db] ⚠️ 内网 MySQL 不可达 (%s:%s), 跳过 schema 提取' % (MYSQL_HOST, MYSQL_PORT))
-        return [], []
-    try:
-        import pymysql
-    except ImportError:
-        logger('  [db] pymysql 未装, 跳过')
-        return [], []
-
-    logger('  [db] 连接 MySQL %s:%s ...' % (MYSQL_HOST, MYSQL_PORT))
-    conn = pymysql.connect(host=MYSQL_HOST, port=MYSQL_PORT, user=MYSQL_USER,
-                           password=MYSQL_PASSWORD, charset='utf8mb4',
-                           cursorclass=pymysql.cursors.DictCursor,
-                           connect_timeout=5, read_timeout=30)
-    tables, columns = [], []
-    try:
-        cur = conn.cursor()
-        cur.execute("SELECT schema_name FROM information_schema.schemata "
-                    "WHERE schema_name LIKE %s ORDER BY schema_name",
-                    (DB_NAME_PREFIX + '%'))
-        dbs = [r['schema_name'] for r in cur.fetchall()]
-        logger('  [db] 发现 %d 个业务库: %s' % (len(dbs), dbs[:3]))
-
-        for db in dbs:
-            # 表
-            cur.execute("SELECT table_name, table_comment, table_rows "
-                        "FROM information_schema.tables WHERE table_schema=%s", (db,))
-            for r in cur.fetchall():
-                tables.append({
-                    'repo_id': repo_id, 'schema_name': db,
-                    'table_name': r['table_name'], 'table_comment': r.get('table_comment') or '',
-                    'rows_count': r.get('table_rows'),
-                })
-            # 列
-            cur.execute("SELECT table_name, column_name, data_type, is_nullable, "
-                        "column_comment, ordinal_position "
-                        "FROM information_schema.columns WHERE table_schema=%s "
-                        "ORDER BY table_name, ordinal_position", (db,))
-            for r in cur.fetchall():
-                columns.append({
-                    'repo_id': repo_id, 'schema_name': db,
-                    'table_name': r['table_name'], 'column_name': r['column_name'],
-                    'data_type': r['data_type'], 'is_nullable': r['is_nullable'],
-                    'column_comment': r.get('column_comment') or '',
-                    'ordinal_position': r.get('ordinal_position'),
-                })
-        logger('  [db] 完成: %d 表 / %d 列' % (len(tables), len(columns)))
-    finally:
-        conn.close()
-    return tables, columns

package/templates/qoder/scripts/domain/kg/extract/inference/__init__.py

@@ -1,9 +0,0 @@
-# -*- coding: utf-8 -*-
-"""L3 推断层 (GOAL 5): 跨源锚点关联。
-
-子模块:
-  weak_link     embedding 弱关联推理 (向量召回, 需 pgvector; 不可用时降级命名匹配)
-  naming_matcher 命名同源匹配 (AssetController ↔ asset ↔ 资产管理)
-  recompute     C 方案指纹重算 (30天过期弱边对比指纹)
-  promote       弱边升格 (出现强关联时弱边删除)
-"""

package/templates/qoder/scripts/domain/kg/extract/inference/community_summarizer.py

@@ -1,206 +0,0 @@
-# -*- coding: utf-8 -*-
-"""D2: 补社区摘要 + 摘要向量。
-
-对 graph_communities 表里 summary 为空的社区:
-  ① 读 member_ids → 取成员的 canonical+cn → LLM 生成摘要
-  ② 对摘要算 embedding → summary_vec
-
-独立于现有 summarize_communities (不改动它), 直接补 PG。
-"""
-from __future__ import annotations
-import os
-import sys
-import json
-import time
-import urllib.request
-from concurrent.futures import ThreadPoolExecutor
-
-_THIS = os.path.dirname(os.path.abspath(__file__))
-for _i in range(8):
-    _p = os.path.dirname(_THIS)
-    if os.path.isfile(os.path.join(_p, 'foundation', 'bootstrap.py')):
-        sys.path.insert(0, _p); break
-    _THIS = _p
-
-_SUMMARY_PROMPT = """你是软件架构师。下面是一个代码社区(聚类在一起的相关功能)的成员列表。
-请用1-2句话概括这个社区负责什么业务功能。
-
-成员列表(类名/端点/按钮, 括号内是中文含义):
-%s
-
-只返回摘要文本, 不要解释。"""
-
-
-def _call_qwen(prompt: str, timeout: int = 20) -> str:
-    key = os.environ.get('DASHSCOPE_API_KEY', '')
-    base = os.environ.get('ALI_CHAT_BASE', '').rstrip('/')
-    model = os.environ.get('ALI_CHAT_MODEL', 'qwen-plus')
-    if not key or not base:
-        return ''
-    req = urllib.request.Request(
-        base + '/chat/completions',
-        data=json.dumps({'model': model, 'messages': [{'role': 'user', 'content': prompt}],
-                         'max_tokens': 200, 'temperature': 0.2}).encode('utf-8'),
-        headers={'Authorization': 'Bearer ' + key, 'Content-Type': 'application/json'},
-    )
-    try:
-        with urllib.request.urlopen(req, timeout=timeout) as resp:
-            return json.loads(resp.read())['choices'][0]['message']['content'].strip()
-    except Exception:
-        return ''
-
-
-def _embed(text: str) -> list:
-    """调 DashScope embedding, 返回 1024 维向量。"""
-    key = os.environ.get('DASHSCOPE_API_KEY', '')
-    if not key or not text:
-        return []
-    req = urllib.request.Request(
-        'https://dashscope.aliyuncs.com/compatible-mode/v1/embeddings',
-        data=json.dumps({'model': 'text-embedding-v3', 'input': [text[:500]], 'dimensions': 1024}).encode('utf-8'),
-        headers={'Authorization': 'Bearer ' + key, 'Content-Type': 'application/json'},
-    )
-    try:
-        with urllib.request.urlopen(req, timeout=15) as resp:
-            return json.loads(resp.read())['data'][0]['embedding']
-    except Exception:
-        return []
-
-
-def summarize_missing(limit: int = 2500, do_embed: bool = True, logger=print):
-    """补全缺摘要的社区。
-
-    Args:
-        limit: 上限 (默认 2500, 覆盖全部 2206 个缺的)
-        do_embed: True 同时补 summary_vec
-    """
-    from sqlalchemy import text
-    from domain.kg.extract.java import pg_upsert
-
-    engine, _ = pg_upsert._get_pg_engine()
-    if engine is None:
-        return {'error': 'PG 不可用'}
-
-    # 取缺摘要的社区
-    with engine.connect() as conn:
-        rows = conn.execute(text("""
-            SELECT id, level, size, member_ids FROM graph_communities
-            WHERE summary IS NULL OR summary = ''
-            LIMIT :lim
-        """), {'lim': limit}).all()
-    logger('  [d2] %d 个社区待补摘要' % len(rows))
-    if not rows:
-        return {'summarized': 0}
-
-    # 批量取成员的 canonical+cn (一次查)
-    all_member_ids = set()
-    for r in rows:
-        try:
-            mids = json.loads(r[3]) if isinstance(r[3], str) else (r[3] or [])
-            all_member_ids.update(mids[:20])  # 每社区最多取20个成员
-        except Exception:
-            pass
-
-    member_info = {}  # {entity_id: "canonical(cn)"}
-    if all_member_ids:
-        member_list = list(all_member_ids)[:5000]  # 限 5000 避免查询太大
-        # 分 chunk 查
-        for ci in range(0, len(member_list), 200):
-            chunk = member_list[ci:ci+200]
-            arr = "ARRAY[" + ','.join("'" + eid.replace("'", "''") + "'" for eid in chunk) + "]"
-            with engine.connect() as conn:
-                mrows = conn.execute(text(
-                    f"SELECT id, canonical, cn FROM entities WHERE id = ANY({arr})"
-                )).all()
-            for eid, canonical, cn in mrows:
-                desc = canonical or eid
-                if cn:
-                    desc += '(' + cn[:20] + ')'
-                member_info[eid] = desc[:50]
-
-    logger('  [d2] 成员信息: %d 条' % len(member_info))
-
-    # 批量生成摘要 (4线程并发)
-    def make_summary(row):
-        cid, level, size, member_ids_raw = row
-        try:
-            mids = json.loads(member_ids_raw) if isinstance(member_ids_raw, str) else (member_ids_raw or [])
-        except Exception:
-            mids = []
-        members_desc = [member_info.get(mid, mid) for mid in mids[:15]]
-        if not members_desc:
-            members_desc = [mid for mid in mids[:15]]
-        if not members_desc:
-            return (cid, '空社区')
-        prompt = _SUMMARY_PROMPT % '\n'.join(members_desc)
-        summary = _call_qwen(prompt)
-        if not summary:
-            # 兜底: 实体名拼接
-            summary = '、'.join(members_desc[:5]) + '等相关功能'
-        return (cid, summary[:200])
-
-    # 并发跑
-    t0 = time.time()
-    batches = [rows[i:i+10] for i in range(0, len(rows), 10)]
-    all_summaries = []
-    for bi in range(0, len(batches), 5):
-        chunk = batches[bi:bi+5]
-        with ThreadPoolExecutor(max_workers=4) as ex:
-            results = list(ex.map(make_summary, [r for b in chunk for r in b]))
-        all_summaries += results
-        if (bi//5+1) % 10 == 0:
-            logger('  [d2] %d/%d 社区摘要生成 (%d)' % (len(all_summaries), len(rows), len(all_summaries)))
-
-    logger('  [d2] 摘要生成完成: %d / %.1fs' % (len(all_summaries), time.time()-t0))
-
-    # 写 PG (分块提交)
-    ok = 0
-    vec_ok = 0
-    for i in range(0, len(all_summaries), 100):
-        chunk = all_summaries[i:i+100]
-        with engine.begin() as conn:
-            for cid, summary in chunk:
-                try:
-                    # 算摘要向量 (可选)
-                    vec_str = ''
-                    if do_embed:
-                        vec = _embed(summary)
-                        if vec:
-                            # summary_vec 存成 JSONB (兼容无 pgvector)
-                            vec_str = json.dumps(vec)
-                    if vec_str:
-                        # summary_vec 是 ARRAY 类型, 用 FLOAT[] 写入
-                        arr_str = '{' + ','.join(str(float(v)) for v in vec[:1024]) + '}'
-                        conn.execute(text("""
-                            UPDATE graph_communities SET summary=:s, summary_vec=CAST(:arr AS FLOAT[])
-                            WHERE id=:id
-                        """), {'s': summary, 'arr': arr_str, 'id': cid})
-                        vec_ok += 1
-                    else:
-                        conn.execute(text("""
-                            UPDATE graph_communities SET summary=:s WHERE id=:id
-                        """), {'s': summary, 'id': cid})
-                    ok += 1
-                except Exception:
-                    pass
-        if (i//100+1) % 5 == 0:
-            logger('  [d2] 写入 %d/%d (vec=%d)' % (ok, len(all_summaries), vec_ok))
-
-    dt = time.time() - t0
-    logger('  [d2] 完成: %d 摘要 / %d 向量 / %.1fs' % (ok, vec_ok, dt))
-    return {'summarized': ok, 'vectorized': vec_ok, 'seconds': round(dt, 1)}
-
-
-if __name__ == '__main__':
-    import argparse
-    ap = argparse.ArgumentParser()
-    ap.add_argument('--limit', type=int, default=2500)
-    ap.add_argument('--no-embed', action='store_true')
-    args = ap.parse_args()
-    try:
-        from dotenv import load_dotenv
-        for p in ['wlinkj-workspace/backend/.env', '.env']:
-            if os.path.isfile(p): load_dotenv(p); break
-    except: pass
-    r = summarize_missing(limit=args.limit, do_embed=not args.no_embed)
-    print(json.dumps(r, ensure_ascii=False, indent=2))

package/templates/qoder/scripts/domain/kg/extract/inference/embed_builder.py

@@ -1,132 +0,0 @@
-# -*- coding: utf-8 -*-
-"""批量灌入实体向量 → embeddings 表 (GOAL A 核心)。
-
-对关键实体 (anchors + SERVICE + TABLE + PRD) 算 DashScope embedding,
-存进 embeddings.vec_json (JSONB 数组, 不依赖 pgvector)。
-
-被 build_goal_a.py 调用, 也可独立 CLI。
-"""
-from __future__ import annotations
-import os
-import sys
-import json
-import time
-import urllib.request
-
-_THIS = os.path.dirname(os.path.abspath(__file__))
-for _i in range(8):
-    _p = os.path.dirname(_THIS)
-    if os.path.isfile(os.path.join(_p, 'foundation', 'bootstrap.py')):
-        sys.path.insert(0, _p); break
-    _THIS = _p
-
-
-def embed_batch(texts: list, batch_size: int = 10) -> list:
-    """批量调 DashScope text-embedding-v3 → [[float]]。失败返回空。"""
-    key = os.environ.get('DASHSCOPE_API_KEY', '')
-    if not key or not texts:
-        return []
-    all_vecs = []
-    for i in range(0, len(texts), batch_size):
-        batch = texts[i:i + batch_size]
-        payload = json.dumps({
-            'model': 'text-embedding-v3', 'input': batch, 'dimensions': 1024,
-        }).encode('utf-8')
-        req = urllib.request.Request(
-            'https://dashscope.aliyuncs.com/compatible-mode/v1/embeddings',
-            data=payload,
-            headers={'Authorization': 'Bearer ' + key, 'Content-Type': 'application/json'},
-        )
-        try:
-            with urllib.request.urlopen(req, timeout=30) as resp:
-                r = json.loads(resp.read())
-            for d in sorted(r['data'], key=lambda x: x['index']):
-                all_vecs.append(d['embedding'])
-        except Exception as e:
-            # 这批失败 → 填 None 占位 (保持顺序)
-            all_vecs.extend([None] * len(batch))
-    return all_vecs
-
-
-def build_embeddings_for_entities(entity_filter: str = None, limit: int = 500,
-                                   kinds: list = None, logger=print):
-    """对实体算 embedding → 灌 embeddings.vec_json。
-
-    Args:
-        entity_filter: SQL WHERE 条件 (如 "type='ANCHOR'")
-        limit: 上限 (控成本)
-        kinds: 实体类型白名单 (默认 anchors + SERVICE + TABLE + PRD)
-    """
-    from domain.kg.extract.java import pg_upsert
-    from sqlalchemy import text
-
-    engine, _ = pg_upsert._get_pg_engine()
-    if engine is None:
-        return {'error': 'PG 不可用'}
-
-    if kinds is None:
-        kinds = ['ANCHOR', 'SERVICE', 'TABLE', 'PRD', 'FEATURE']
-
-    # 取实体 (跳过已 embed 过的)
-    where = "WHERE e.type IN :kinds AND e.id NOT IN (SELECT entity_id FROM embeddings)"
-    if entity_filter:
-        where += " AND " + entity_filter
-    sql = f"""
-        SELECT e.id, e.repo_id, e.type, e.canonical, e.cn
-        FROM entities e
-        {where.replace(':kinds', ':kinds')}
-        LIMIT :lim
-    """
-    # 兼容 IN 参数
-    with engine.connect() as conn:
-        rows = conn.execute(text(sql.replace(':kinds', "('" + "','".join(kinds) + "')")),
-                            {'lim': limit}).all()
-
-    logger('  [embed] %d 实体待向量化 (类型=%s)' % (len(rows), kinds))
-    if not rows:
-        return {'embedded': 0}
-
-    # 批量 embed
-    descs = [(r[3] or r[4] or r[0])[:200] for r in rows]  # canonical 优先
-    t0 = time.time()
-    vecs = embed_batch(descs)
-    embed_time = time.time() - t0
-    valid = sum(1 for v in vecs if v)
-    logger('  [embed] embedding 完成: %d/%d 有效 / %.1fs' % (valid, len(vecs), embed_time))
-
-    # 灌 PG
-    ok = 0
-    with engine.begin() as conn:
-        for (eid, repo_id, etype, canonical, cn), vec in zip(rows, vecs):
-            if not vec:
-                continue
-            try:
-                conn.execute(text("""
-                    INSERT INTO embeddings (entity_id, repo_id, kind, content, vec_json, model, embedded_at)
-                    VALUES (:id, :repo, :kind, :content, CAST(:vec AS JSONB), 'text-embedding-v3', now())
-                    ON CONFLICT (entity_id) DO UPDATE SET
-                        vec_json=EXCLUDED.vec_json, content=EXCLUDED.content, embedded_at=now()
-                """), {
-                    'id': eid, 'repo': repo_id, 'kind': etype,
-                    'content': (canonical or cn or eid)[:200],
-                    'vec': json.dumps(vec),
-                })
-                ok += 1
-            except Exception:
-                pass
-    logger('  [embed] 灌入 %d 条向量' % ok)
-    return {'embedded': ok, 'time_sec': round(embed_time, 1)}
-
-
-if __name__ == '__main__':
-    import argparse
-    ap = argparse.ArgumentParser()
-    ap.add_argument('--limit', type=int, default=500)
-    args = ap.parse_args()
-    try:
-        from dotenv import load_dotenv
-        for p in ['wlinkj-workspace/backend/.env', '.env']:
-            if os.path.isfile(p): load_dotenv(p); break
-    except: pass
-    r = build_embeddings_for_entities(limit=args.limit)
-    print(json.dumps(r, ensure_ascii=False, indent=2))