"""项目能力 — sd_projects 的状态机语义化迁移（通用、产线无关）。

定位：项目 CRUD 走 xls2ui 生成的端点；本模块只做「生命周期状态流转」——
每个迁移 CAS 原子（防并发/越权）+ 租户隔离 + 审计。

项目的生命周期/流转规则在 project skill 里（LLM 读 skill 判断合法性）；本模块只固化操作原语。

角色规范：role 参数用 agent.{role}（无前缀自动补 agent.）；人角色用 {orgtype}.{role}。

scope 约定：sd_projects 是顶层容器，用 id 唯一标识（CAS WHERE 带 id 即可），
org_id 用于 list 过滤（租户隔离）；audit 的 tenant_id 用 project_id（项目即最外层容器）。
"""

import logging
import os
import json
import re
import tarfile
import tempfile
import shutil
from datetime import datetime
from sqlor.dbpools import DBPools
from appPublic.uniqueID import getID
from .audit import record_audit

DBNAME = "pipeline"
logger = logging.getLogger("pipeline.project_capability")

TABLE = "sd_projects"

# 项目状态（SDLC 默认，状态机语义见 project skill）
S_DRAFT = "draft"            # 草稿（已创建，未启动）
S_ACTIVE = "active"          # 进行中
S_PAUSED = "paused"          # 已暂停（暂停推进，PM 不再推进项目，除非恢复）
S_COMPLETED = "completed"    # 已完成
S_ARCHIVED = "archived"      # 已归档


def _get_db():
    db = DBPools()
    if not db.databases:
        from appPublic.jsonConfig import getConfig
        config = getConfig()
        if config.databases:
            db.databases = config.databases
    return db, DBNAME


def _normalize_role(role):
    """角色规范：agent 角色补 agent. 前缀；人角色 {orgtype}.{role} 保留原样。"""
    role = (role or "").strip()
    if not role:
        return ""
    if "." in role:
        return role
    return f"agent.{role}"


async def _transition(project_id, from_state, to_state, action,
                      who=None, agent_id=None, detail=None):
    """CAS 状态迁移 + 审计。返回 (ok, message)。"""
    if not project_id:
        return False, "缺少 project_id"
    db, dbname = _get_db()
    async with db.sqlorContext(dbname) as sor:
        await sor.sqlExe(
            f"UPDATE {TABLE} SET status=${{to}}$, updated_at=NOW() "
            "WHERE id=${pid}$ AND status=${from}$",
            {"to": to_state, "pid": project_id, "from": from_state})
        recs = await sor.R(TABLE, {'id': project_id})
        await sor.sqlExe("COMMIT", {})
        if not recs:
            return False, "项目不存在"
        cur = getattr(recs[0], 'status', '')
        if cur != to_state:
            return False, f"状态迁移失败(CAS): 期望 from={from_state} 实际 status={cur}"
        await record_audit(project_id, TABLE, project_id, action,
                           from_state=from_state, to_state=to_state,
                           who=who, agent_id=agent_id, detail=detail, sor=sor)
        return True, to_state


async def create_project(name, project_type="web_app", description="",
                         org_id="0", created_by="", who=None, agent_id=None):
    """创建项目：新建记录，status=draft。返回 (ok, project_id_or_message)。"""
    if not name or not name.strip():
        return False, "缺少项目名称"
    db, dbname = _get_db()
    async with db.sqlorContext(dbname) as sor:
        pid = getID()
        await sor.C(TABLE, {
            'id': pid,
            'name': name.strip(),
            'description': description or '',
            'project_type': project_type or 'web_app',
            'status': S_DRAFT,
            'org_id': org_id or '0',
            'created_by': created_by or '',
        })
        await record_audit(pid, TABLE, pid, 'create',
                           to_state=S_DRAFT, who=_normalize_role(who),
                           agent_id=agent_id, sor=sor)
        logger.info("create_project: %s name=%s", pid, name.strip())
        return True, pid


async def check_project_owner(project_id, user_id, sor=None):
    """校验 user_id 是否为项目 owner（sd_projects.created_by）。

    独立开 context（供 pipeline-task 等 dspy 直接调用）；传入 sor 时复用它（供
    pipeline-sdlc dspy 在已有 get_sor_context 内调用，避免嵌套 context 的 MDL 锁）。
    返回 (True, '') 或 (False, 错误信息)。
    """
    if not project_id:
        return False, "缺少 project_id"
    if not user_id:
        return False, "未登录"

    async def _check(_sor):
        recs = await _sor.sqlExe(
            "SELECT created_by, name FROM sd_projects WHERE id=${pid}$",
            {"pid": project_id})
        await _sor.sqlExe("COMMIT", {})
        if not recs:
            return False, "项目不存在"
        owner = getattr(recs[0], 'created_by', '') or ''
        if owner == user_id:
            return True, ''
        return False, "仅项目 owner 可执行此操作"

    if sor is not None:
        return await _check(sor)
    db, dbname = _get_db()
    async with db.sqlorContext(dbname) as _s:
        return await _check(_s)


async def check_task_owner(task_id, user_id):
    """校验 user_id 是否为 task 所属 SDLC 项目的 owner。

    SDLC 任务 tenant_id=project_id，须校验 owner；通用引擎任务 tenant_id=org_id（非
    项目 id），放行返回 (True, '')。独立开 context（供 pipeline-task dspy 调用）。
    """
    if not task_id:
        return False, "缺少 task_id"
    if not user_id:
        return False, "未登录"
    db, dbname = _get_db()
    async with db.sqlorContext(dbname) as sor:
        recs = await sor.sqlExe(
            "SELECT tenant_id FROM pipeline_tasks WHERE id=${tid}$", {"tid": task_id})
        await sor.sqlExe("COMMIT", {})
        if not recs:
            return False, "任务不存在"
        tenant = getattr(recs[0], 'tenant_id', '') or ''
        p = await sor.sqlExe(
            "SELECT id FROM sd_projects WHERE id=${pid}$", {"pid": tenant})
        await sor.sqlExe("COMMIT", {})
        if not p:
            # 通用引擎任务（tenant_id 非 SDLC 项目），无 owner 概念，放行
            return True, ''
        return await check_project_owner(tenant, user_id, sor)


async def check_tenant_owner(tenant_id, user_id):
    """校验 user_id 是否为 tenant_id 所属 SDLC 项目的 owner。非项目 tenant 放行。

    独立开 context（供 pipeline-task 的 task_submit 等 dspy 调用）。
    """
    if not user_id:
        return False, "未登录"
    if not tenant_id:
        return True, ''
    db, dbname = _get_db()
    async with db.sqlorContext(dbname) as sor:
        p = await sor.sqlExe(
            "SELECT id FROM sd_projects WHERE id=${pid}$", {"pid": tenant_id})
        await sor.sqlExe("COMMIT", {})
        if not p:
            return True, ''
        return await check_project_owner(tenant_id, user_id, sor)


async def start_project(project_id, who=None, agent_id=None):
    """启动项目：draft → active（也兼容 archived → active 重新激活）。"""
    ok, msg = await _transition(project_id, S_DRAFT, S_ACTIVE, 'start',
                                who=who, agent_id=agent_id)
    if ok:
        return ok, msg
    return await _transition(project_id, S_ARCHIVED, S_ACTIVE, 'start',
                             who=who, agent_id=agent_id)


async def complete_project(project_id, who=None, agent_id=None):
    """完成项目：active → completed。"""
    return await _transition(project_id, S_ACTIVE, S_COMPLETED, 'complete',
                             who=who, agent_id=agent_id)


async def archive_project(project_id, who=None, agent_id=None):
    """归档项目：completed → archived（也兼容 active → archived 强制归档）。"""
    ok, msg = await _transition(project_id, S_COMPLETED, S_ARCHIVED, 'archive',
                                who=who, agent_id=agent_id)
    if ok:
        return ok, msg
    return await _transition(project_id, S_ACTIVE, S_ARCHIVED, 'archive',
                             who=who, agent_id=agent_id)


async def reopen_project(project_id, who=None, agent_id=None):
    """重新打开：archived → active。"""
    return await _transition(project_id, S_ARCHIVED, S_ACTIVE, 'reopen',
                             who=who, agent_id=agent_id)


async def pause_project(project_id, who=None, agent_id=None):
    """暂停推进：active → paused。暂停后 PM 不再推进项目（不认领审核任务、不创建后续任务），
    直到 resume_project 恢复。仅用户明确指令「暂停推进」时才调用——默认必须推进，无需指令。"""
    return await _transition(project_id, S_ACTIVE, S_PAUSED, 'pause',
                             who=who, agent_id=agent_id)


async def resume_project(project_id, who=None, agent_id=None):
    """恢复推进：paused → active。"""
    return await _transition(project_id, S_PAUSED, S_ACTIVE, 'resume',
                             who=who, agent_id=agent_id)


async def set_project_state(project_id, from_state, to_state,
                            who=None, agent_id=None, detail=None):
    """通用 CAS 状态迁移兜底（跨产线自定义状态机用）。"""
    return await _transition(project_id, from_state, to_state, 'set_state',
                             who=who, agent_id=agent_id, detail=detail)


async def list_projects(org_id=None, status=None, limit=50) -> list:
    """列出项目（可按组织/状态过滤；org_id 空则列全部）。"""
    db, dbname = _get_db()
    async with db.sqlorContext(dbname) as sor:
        conditions = ["1=1"]
        params = {}
        if org_id:
            conditions.append("org_id=${org}$")
            params["org"] = org_id
        if status:
            conditions.append("status=${status}$")
            params["status"] = status
        where = " AND ".join(conditions)
        try:
            limit = int(limit)
        except (TypeError, ValueError):
            limit = 50
        sql = (f"SELECT * FROM {TABLE} WHERE {where} "
               f"ORDER BY created_at DESC LIMIT {limit}")
        recs = await sor.sqlExe(sql, params)
        # 释放 SELECT 元数据锁
        await sor.sqlExe("COMMIT", {})
        result = []
        for rec in (recs or []):
            result.append(_rec_to_dict(rec))
        return result


def _rec_to_dict(rec):
    """把 sqlor 记录对象转成 dict（sqlor 行是 DictObject，必须 dict(rec) 取列）。"""
    if isinstance(rec, dict):
        return dict(rec)
    try:
        return dict(rec)
    except (TypeError, ValueError):
        if hasattr(rec, 'to_dict'):
            try:
                return rec.to_dict()
            except Exception:
                pass
        return {}


async def _dump_project_records(sor, project_id):
    """导出项目及关联表全部记录（归档备份 JSON 用）。"""
    dump = {}
    # 直接关联（project_id / tenant_id）
    direct = {
        'sd_projects': "SELECT * FROM sd_projects WHERE id=${pid}$",
        'sd_iterations': "SELECT * FROM sd_iterations WHERE project_id=${pid}$",
        'pipeline_tasks': "SELECT * FROM pipeline_tasks WHERE tenant_id=${pid}$",
        'pipeline_deliverables': "SELECT * FROM pipeline_deliverables WHERE project_id COLLATE utf8mb4_unicode_ci=${pid}$",
        'pipeline_project_agents': "SELECT * FROM pipeline_project_agents WHERE project_id COLLATE utf8mb4_unicode_ci=${pid}$",
        'pipeline_agent_questions': "SELECT * FROM pipeline_agent_questions WHERE tenant_id=${pid}$",
        'sd_deploy_envs': "SELECT * FROM sd_deploy_envs WHERE project_id=${pid}$",
        'sd_features': "SELECT * FROM sd_features WHERE project_id=${pid}$",
        'sd_project_repos': "SELECT * FROM sd_project_repos WHERE project_id COLLATE utf8mb4_unicode_ci=${pid}$",
    }
    for tbl, sql in direct.items():
        try:
            recs = await sor.sqlExe(sql, {"pid": project_id})
            dump[tbl] = [_rec_to_dict(r) for r in (recs or [])]
        except Exception as e:
            dump[tbl] = {"_error": f"{type(e).__name__}: {str(e)[:200]}"}

    # 间接关联（iteration_id 通过子查询；pipeline_conversations 的 iteration_id 存的是 project_id）
    indirect = {
        'sd_bugs': "SELECT * FROM sd_bugs WHERE iteration_id IN (SELECT id FROM sd_iterations WHERE project_id=${pid}$)",
        'sd_test_plans': "SELECT * FROM sd_test_plans WHERE iteration_id IN (SELECT id FROM sd_iterations WHERE project_id=${pid}$)",
        'sd_test_cases': "SELECT * FROM sd_test_cases WHERE plan_id IN (SELECT id FROM sd_test_plans WHERE iteration_id IN (SELECT id FROM sd_iterations WHERE project_id=${pid}$))",
        'sd_conversations': "SELECT * FROM sd_conversations WHERE iteration_id COLLATE utf8mb4_unicode_ci IN (SELECT id FROM sd_iterations WHERE project_id=${pid}$)",
        'pipeline_conversations': "SELECT * FROM pipeline_conversations WHERE iteration_id COLLATE utf8mb4_unicode_ci=${pid}$ OR iteration_id COLLATE utf8mb4_unicode_ci IN (SELECT id FROM sd_iterations WHERE project_id=${pid}$)",
        'audit_log': "SELECT * FROM audit_log WHERE tenant_id COLLATE utf8mb4_unicode_ci=${pid}$ ORDER BY created_at ASC",
    }
    for tbl, sql in indirect.items():
        try:
            recs = await sor.sqlExe(sql, {"pid": project_id})
            dump[tbl] = [_rec_to_dict(r) for r in (recs or [])]
        except Exception as e:
            dump[tbl] = {"_error": f"{type(e).__name__}: {str(e)[:200]}"}

    return dump


async def backup_project(project_id, who=None, agent_id=None):
    """归档备份：打包项目工作目录 + 导出数据库记录 → tgz 到「机构工作目录/_archive/」。

    返回 (ok, tgz_path_or_message)。删除项目前必须调用，保证可回退。
    """
    db, dbname = _get_db()
    async with db.sqlorContext(dbname) as sor:
        recs = await sor.R(TABLE, {'id': project_id})
        if not recs:
            await sor.sqlExe("COMMIT", {})
            return False, "项目不存在"
        p = recs[0]
        pname = getattr(p, 'name', '') or project_id
        org_id = getattr(p, 'org_id', '0') or '0'
        ws = getattr(p, 'workspace_dir', '') or ''
        status = getattr(p, 'status', '')

        from .workspace import get_workspace_base
        workspace_base = await get_workspace_base(sor)
        if not (ws or '').startswith('/'):
            ws = os.path.join(workspace_base, str(org_id), pname)

        # 机构工作目录下的备份归档目录
        archive_dir = os.path.join(workspace_base, str(org_id), '_archive')
        os.makedirs(archive_dir, exist_ok=True)

        timestamp = datetime.now().strftime('%Y%m%d_%H%M%S')
        safe_name = re.sub(r'[^\w\u4e00-\u9fff.-]', '_', pname)
        tgz_path = os.path.join(archive_dir, f"{safe_name}_{timestamp}.tgz")

        # 导出数据库记录
        dump = await _dump_project_records(sor, project_id)

        try:
            with tarfile.open(tgz_path, 'w:gz') as tar:
                # 数据库记录 JSON
                with tempfile.NamedTemporaryFile('w', suffix='.json', delete=False, encoding='utf-8') as f:
                    json.dump(dump, f, ensure_ascii=False, default=str)
                    tmp_json = f.name
                tar.add(tmp_json, arcname='project_data.json')
                os.unlink(tmp_json)
                # 工作目录文件（若存在）
                if ws and os.path.isdir(ws):
                    tar.add(ws, arcname='workspace')
        except Exception as e:
            logger.error("backup_project tar failed: %s err=%s", project_id, e)
            return False, f"打包失败: {type(e).__name__}: {str(e)[:200]}"

        await record_audit(project_id, TABLE, project_id, 'backup',
                           to_state=status, who=_normalize_role(who),
                           agent_id=agent_id, detail=tgz_path, sor=sor)
        logger.info("backup_project: %s -> %s", project_id, tgz_path)
        return True, tgz_path


async def delete_project(project_id, who=None, agent_id=None, confirm=False):
    """删除项目：先归档备份，再删除关联表记录 + sd_projects + 工作目录。

    confirm 必须为 True（删除前二次确认）。备份失败则中止删除。
    返回 (ok, message)。
    """
    if not confirm:
        return False, "删除项目需二次确认（confirm=true），请先确认再执行"
    db, dbname = _get_db()
    async with db.sqlorContext(dbname) as sor:
        # 1. 归档备份（失败则中止删除）
        ok, backup_msg = await backup_project(project_id, who=who, agent_id=agent_id)
        if not ok:
            return False, f"归档备份失败，已中止删除：{backup_msg}"

        # 读工作目录（删除数据库后清理物理文件）
        recs = await sor.R(TABLE, {'id': project_id})
        if not recs:
            await sor.sqlExe("COMMIT", {})
            return False, "项目不存在"
        p = recs[0]
        pname = getattr(p, 'name', '') or project_id
        org_id = getattr(p, 'org_id', '0') or '0'
        ws = getattr(p, 'workspace_dir', '') or ''
        from .workspace import get_workspace_base
        workspace_base = await get_workspace_base(sor)
        if not (ws or '').startswith('/'):
            ws = os.path.join(workspace_base, str(org_id), pname)

        # 2. 删除间接关联表（依赖迭代）
        indirect = [
            "DELETE FROM sd_test_cases WHERE plan_id IN (SELECT id FROM sd_test_plans WHERE iteration_id IN (SELECT id FROM sd_iterations WHERE project_id=${pid}$))",
            "DELETE FROM sd_bugs WHERE iteration_id IN (SELECT id FROM sd_iterations WHERE project_id=${pid}$)",
            "DELETE FROM sd_test_plans WHERE iteration_id IN (SELECT id FROM sd_iterations WHERE project_id=${pid}$)",
            "DELETE FROM sd_conversations WHERE iteration_id COLLATE utf8mb4_unicode_ci IN (SELECT id FROM sd_iterations WHERE project_id=${pid}$)",
            "DELETE FROM pipeline_conversations WHERE iteration_id COLLATE utf8mb4_unicode_ci=${pid}$ OR iteration_id COLLATE utf8mb4_unicode_ci IN (SELECT id FROM sd_iterations WHERE project_id=${pid}$)",
        ]
        for sql in indirect:
            await sor.sqlExe(sql, {"pid": project_id})

        # 2b. 任务关联表：先取本项目的任务 id，再删 task_id 指向它们的行
        #     （pipeline_deliverables 等同时有 project_id 和 task_id 两列，仅按 project_id
        #      删会漏掉「项目删除时 task 已先被删、project_id 已失效」的残留）
        tid_rows = await sor.sqlExe(
            "SELECT id FROM pipeline_tasks WHERE tenant_id=${pid}$", {"pid": project_id})
        await sor.sqlExe("COMMIT", {})
        task_ids = [getattr(r, 'id', '') for r in (tid_rows or []) if getattr(r, 'id', '')]
        if task_ids:
            tid_list = ",".join("'" + t.replace("'", "") + "'" for t in task_ids)
            for sql in [
                "DELETE FROM pipeline_deliverables WHERE task_id COLLATE utf8mb4_unicode_ci IN (" + tid_list + ")",
                "DELETE FROM pipeline_artifacts WHERE task_id IN (" + tid_list + ")",
                "DELETE FROM pipeline_task_steps WHERE task_id IN (" + tid_list + ")",
                "DELETE FROM pipeline_human_tasks WHERE task_id IN (" + tid_list + ")",
            ]:
                try:
                    await sor.sqlExe(sql, {})
                except Exception as e:
                    logger.warning("delete_project task-linked cleanup failed: %s err=%s", sql, e)

        # 3. 删除直接关联表（project_id / tenant_id）
        direct = [
            "DELETE FROM pipeline_deliverables WHERE project_id COLLATE utf8mb4_unicode_ci=${pid}$",
            "DELETE FROM pipeline_project_agents WHERE project_id COLLATE utf8mb4_unicode_ci=${pid}$",
            "DELETE FROM pipeline_agent_questions WHERE tenant_id=${pid}$",
            "DELETE FROM sd_deploy_envs WHERE project_id=${pid}$",
            "DELETE FROM sd_features WHERE project_id=${pid}$",
            "DELETE FROM sd_project_repos WHERE project_id COLLATE utf8mb4_unicode_ci=${pid}$",
            "DELETE FROM sd_project_role_models WHERE project_id=${pid}$",
            "DELETE FROM pipeline_tasks WHERE tenant_id=${pid}$",
            "DELETE FROM audit_log WHERE tenant_id COLLATE utf8mb4_unicode_ci=${pid}$",
            "DELETE FROM sd_iterations WHERE project_id=${pid}$",
            # 会话/全局「当前项目」指针：清空而非删行（行里还有 iteration/llm 等其他设置）。
            # 不清理会产生悬空引用：下次打开工作空间/菜单时按会话解析仍命中已删项目，
            # 报「请先在会话中切换项目」——这正是本 bug 的源头。
            "UPDATE pipeline_session_settings SET current_project_id='' WHERE current_project_id=${pid}$",
            "UPDATE pipeline_agent_settings SET current_project_id='' WHERE current_project_id=${pid}$",
            "DELETE FROM sd_projects WHERE id=${pid}$",
        ]
        for sql in direct:
            await sor.sqlExe(sql, {"pid": project_id})
        await sor.sqlExe("COMMIT", {})

    # 4. 删除工作目录（物理文件，已备份）
    if ws and os.path.isdir(ws):
        try:
            shutil.rmtree(ws, ignore_errors=True)
        except Exception as e:
            logger.warning("delete_project rmtree failed: %s err=%s", ws, e)

    logger.info("delete_project: %s done, backup=%s", project_id, backup_msg)
    return True, f"项目已删除，归档备份在 {backup_msg}"


# ── 孤儿清扫 ────────────────────────────────────────────────────────
#
# 历史缺陷与手工测试会留下「指向已删项目/迭代/任务的孤儿行」。本函数逐表清扫。
#
# 孤儿判据 = 引用列【非空】且指向的主键在父表中已不存在。
#   注意：空串 '' 与 NULL 都表示「未挂载」（合法），不是孤儿——若不加 `<> ''` 守卫，
#   `NOT IN` 会把空串当成孤儿误删（实测 sd_conversations 32 条 / sd_features 23 条 iteration_id='' 被误判）。
#
# 列 collation 不一（sd_projects.id / sd_iterations.id / pipeline_tasks.id 是 unicode_ci；
#   sd_project_repos / pipeline_deliverables / sd_conversations 的引用列是 general_ci），
#   跨表比较处显式 COLLATE utf8mb4_unicode_ci 规避 1267。

# (label, 表, 引用列, 父表, 是否需 COLLATE)
_ORPHAN_CLEANUPS = [
    ("sd_project_repos(无项目)", "sd_project_repos", "project_id", "sd_projects", True),
    ("sd_features(无项目)", "sd_features", "project_id", "sd_projects", False),
    ("sd_project_role_models(无项目)", "sd_project_role_models", "project_id", "sd_projects", False),
    ("sd_deploy_envs(无项目)", "sd_deploy_envs", "project_id", "sd_projects", False),
    ("pipeline_deliverables(无项目)", "pipeline_deliverables", "project_id", "sd_projects", True),
    ("sd_iterations(无项目)", "sd_iterations", "project_id", "sd_projects", False),
    ("sd_features(无迭代)", "sd_features", "iteration_id", "sd_iterations", False),
    ("sd_bugs(无迭代)", "sd_bugs", "iteration_id", "sd_iterations", False),
    ("sd_test_plans(无迭代)", "sd_test_plans", "iteration_id", "sd_iterations", False),
    ("sd_conversations(无迭代)", "sd_conversations", "iteration_id", "sd_iterations", True),
    ("pipeline_deliverables(无迭代)", "pipeline_deliverables", "iteration_id", "sd_iterations", True),
    ("sd_test_cases(无计划)", "sd_test_cases", "plan_id", "sd_test_plans", False),
    ("pipeline_deliverables(无任务)", "pipeline_deliverables", "task_id", "pipeline_tasks", True),
    ("pipeline_artifacts(无任务)", "pipeline_artifacts", "task_id", "pipeline_tasks", False),
    ("pipeline_task_steps(无任务)", "pipeline_task_steps", "task_id", "pipeline_tasks", False),
    ("pipeline_human_tasks(无任务)", "pipeline_human_tasks", "task_id", "pipeline_tasks", False),
    ("pipeline_agent_questions(无任务)", "pipeline_agent_questions", "task_id", "pipeline_tasks", False),
    ("sd_features(无任务)", "sd_features", "task_id", "pipeline_tasks", False),
    ("sd_iterations(无任务)", "sd_iterations", "task_id", "pipeline_tasks", False),
    ("sd_conversations(无任务)", "sd_conversations", "task_id", "pipeline_tasks", True),
    # 「当前项目」指针：指向已删项目的会话/全局设置记录。解析层已能自愈回退，
    # 这里清扫是存量脏数据的批量兜底（历史 CRUD 删除不清理指针留下的）。
    ("pipeline_session_settings(无项目)", "pipeline_session_settings", "current_project_id", "sd_projects", True),
    ("pipeline_agent_settings(无项目)", "pipeline_agent_settings", "current_project_id", "sd_projects", True),
]


def _orphan_sql(label, table, col, parent, use_collate):
    """生成 (计数SQL, 删除SQL)。守卫：引用列非空（<> '' 排除 NULL 与空串）。"""
    lhs = col + " COLLATE utf8mb4_unicode_ci" if use_collate else col
    guard = f"{lhs} <> '' AND {lhs} NOT IN (SELECT id FROM {parent})"
    cnt = f"SELECT COUNT(*) AS rcnt FROM {table} WHERE {guard}"
    dlt = f"DELETE FROM {table} WHERE {guard}"
    return cnt, dlt


def _count_rows(recs):
    """从 SELECT COUNT(*) AS rcnt 结果取整数；兼容 DictObject/dict/list。"""
    if not recs:
        return 0
    r = recs[0]
    if isinstance(r, dict):
        v = r.get('rcnt') or r.get('count') or r.get('COUNT(*)')
        return int(v or 0)
    if isinstance(r, (list, tuple)):
        return int(r[0] if r and r[0] is not None else 0)
    v = getattr(r, 'rcnt', None)
    if v is None:
        v = getattr(r, 'count', None)
    return int(v or 0)


async def cleanup_orphans(confirm=False, dry_run=False):
    """清扫指向已删项目/迭代/任务的孤儿行（幂等，可反复执行）。

    confirm 必须为 True。dry_run=True 时只统计不删除。
    逐表执行，单表失败不中断其余表。
    返回 (ok, message)，message 含每表行数汇总。
    """
    if not confirm:
        return False, "孤儿清扫需二次确认（confirm=true）"
    db, dbname = _get_db()
    async with db.sqlorContext(dbname) as sor:
        results = []
        total = 0
        for label, table, col, parent, use_collate in _ORPHAN_CLEANUPS:
            cnt_sql, del_sql = _orphan_sql(label, table, col, parent, use_collate)
            try:
                recs = await sor.sqlExe(cnt_sql, {})
                await sor.sqlExe("COMMIT", {})
                n = _count_rows(recs)
            except Exception as e:
                results.append(f"{label}: 跳过({type(e).__name__})")
                continue
            if n and not dry_run:
                try:
                    await sor.sqlExe(del_sql, {})
                    await sor.sqlExe("COMMIT", {})
                except Exception as e:
                    results.append(f"{label}: 删除失败({type(e).__name__})")
                    continue
            total += n
            results.append(f"{label}: {n}")
    verb = "发现" if dry_run else "清除"
    logger.info("cleanup_orphans(%s): %s %d 行孤儿记录。明细: %s",
                "dry_run" if dry_run else "执行", verb, total, "; ".join(results))
    return True, f"孤儿清扫完成，{verb} {total} 行。明细：{'；'.join(results)}"