# -*- coding: utf-8 -*-
"""AI分析P1：确定性经营异常扫描与角色化洞察。

异常由后端规则计算，LLM只负责后续报告归纳，避免把关键数字交给模型臆测。
"""
from collections import defaultdict
from datetime import date, datetime
import hashlib
import re

from services.ai_datasources import AI_DATASOURCES, datasource_meta
from services.simple_cache import get as cache_get, set as cache_set, delete as cache_delete

CACHE_KEY = 'ai_p1_operational_insights_v1'
CACHE_TTL = 300


def _num(value, default=0.0):
    try:
        return float(value or 0)
    except (TypeError, ValueError):
        return default


def _int(value):
    return int(round(_num(value)))


def _parse_date(value):
    if isinstance(value, (datetime, date)):
        return value.date() if isinstance(value, datetime) else value
    text = str(value or '').strip()
    if not text:
        return None
    # 订单历史数据包含 YYYY-MM-DD、YYYY-MM-DD HH:mm、YYYY/M/D 等形式。
    match = re.search(r'(20\d{2})[-/年](\d{1,2})[-/月](\d{1,2})', text)
    if not match:
        return None
    try:
        return date(int(match.group(1)), int(match.group(2)), int(match.group(3)))
    except ValueError:
        return None


def _severity_rank(level):
    return {'高': 3, '中': 2, '低': 1, '信息': 0}.get(level, 0)


def _stable_id(kind, *parts):
    raw = '|'.join([kind] + [str(x or '') for x in parts])
    return kind + '-' + hashlib.md5(raw.encode('utf-8')).hexdigest()[:12]


def _insight(kind, category, severity, title, detail, question, source_keys,
             product_code='', supplier='', metrics=None, source_page=''):
    return {
        'id': _stable_id(kind, product_code, supplier, title),
        'type': kind,
        'category': category,
        'severity': severity,
        'title': title,
        'detail': detail,
        'question': question,
        'source_keys': source_keys,
        'product_code': product_code,
        'supplier': supplier,
        'metrics': metrics or {},
        'source_page': source_page,
    }


def _fetch(key):
    ds = AI_DATASOURCES[key]
    result = ds['handler']({'_no_limit': True}) or {}
    return result.get('rows') or []


def _build_raw_insights():
    planning = _fetch('planning_metrics')
    profiles = _fetch('product_profile')
    sales = _fetch('product_sales')
    orders = _fetch('order_detail')
    suppliers = _fetch('supplier_delivery')
    bom = _fetch('bom_supply_chain')
    values = _fetch('inventory_value')

    profile_map = {str(r.get('product_code') or ''): r for r in profiles}
    sales_map = {str(r.get('product_code') or ''): r for r in sales}
    planning_map = {str(r.get('product_code') or ''): r for r in planning}
    value_map = {str(r.get('product_code') or ''): r for r in values}
    today = date.today()
    items = []

    # 1. 缺货/补货风险、积压与销量异常。
    for row in planning:
        code = str(row.get('product_code') or '')
        if not code:
            continue
        profile = profile_map.get(code) or {}
        sale = sales_map.get(code) or {}
        daily = _num(row.get('日销量预测'))
        support = _num(row.get('总支撑天数'), 999)
        inventory_support = _num(row.get('库存可支撑天数'), 999)
        suggested = _int(row.get('建议下单数量'))
        inventory = _int(row.get('库存')) + _int(row.get('加工厂库存'))
        lead = _int(profile.get('生产货期'))
        logistics = _int(profile.get('物流时间'))
        product_type = str(profile.get('产品类型') or '')
        product_name = str(profile.get('货品名称') or '')
        # 物料/内料/标签等非销售SKU不进入成品销量与积压预警，避免历史分类不规范造成噪音。
        material_terms = ('【内料】', '【瓶子】', '【瓶盖】', '【泵头】', '【彩盒】', '【标签】', '防伪码标签')
        if product_type in ('包材', '原材料', '其他') or any(term in product_name for term in material_terms):
            continue
        threshold = lead + logistics + (0 if product_type == '现货' else 7)
        avg7 = _num(sale.get('7天平均销量'))
        avg30 = _num(sale.get('30天平均销量'))

        if daily > 0 and (support < threshold or suggested > 0):
            severity = '高' if support < max(lead + logistics, 1) else '中'
            items.append(_insight(
                'stockout', '缺货与补货', severity,
                f'{code} 存在补货风险',
                f'总支撑天数{support:.1f}天，补货阈值{threshold}天，建议下单{suggested}件，日销量预测{daily:g}件。',
                f'分析{code}为什么需要下单，并检查库存、在途和BOM包材是否足够',
                ['planning_metrics', 'product_profile'],
                product_code=code,
                metrics={'总支撑天数': support, '补货阈值': threshold, '建议下单数量': suggested,
                         '日销量预测': daily}, source_page='/planning'))

        if daily <= 0 and inventory >= 1000:
            severity = '高' if inventory >= 10000 else '中'
            items.append(_insight(
                'overstock', '库存健康', severity,
                f'{code} 无销量但仍有库存',
                f'日销量预测为0，当前库存及加工厂库存合计{inventory}件，存在滞销或库存积压风险。',
                f'分析{code}为什么近期无销量，库存应该如何处理',
                ['planning_metrics'],
                product_code=code, metrics={'库存合计': inventory, '日销量预测': daily},
                source_page='/planning'))
        elif daily > 0 and support >= 120 and inventory >= daily * 60:
            value = value_map.get(code) or {}
            severity = '高' if support >= 180 else '中'
            items.append(_insight(
                'overstock', '库存健康', severity,
                f'{code} 支撑天数偏高',
                f'总支撑天数{support:.1f}天，库存及加工厂库存合计{inventory}件，明显高于近期销售需要。',
                f'分析{code}库存积压原因和去库存建议',
                ['planning_metrics'],
                product_code=code,
                metrics={'总支撑天数': support, '库存合计': inventory,
                         }, source_page='/planning'))

        if avg30 >= 3 and avg7 <= avg30 * 0.6:
            decline = round((1 - avg7 / avg30) * 100, 1) if avg30 else 0
            items.append(_insight(
                'sales_decline', '销量异常', '高' if decline >= 60 else '中',
                f'{code} 近7天销量明显下降',
                f'7天平均销量{avg7:g}件，30天平均销量{avg30:g}件，下降约{decline}%。',
                f'分析{code}近7天销量下降的原因及库存风险',
                ['product_sales'],
                product_code=code, metrics={'7天平均销量': avg7, '30天平均销量': avg30, '下降幅度': decline},
                source_page='/outbound'))
        elif avg30 >= 1 and avg7 >= max(avg30 * 1.5, avg30 + 3):
            growth = round((avg7 / avg30 - 1) * 100, 1) if avg30 else 0
            severity = '高' if inventory_support < max(threshold, 15) else '中'
            items.append(_insight(
                'sales_growth', '销量异常', severity,
                f'{code} 近7天销量快速增长',
                f'7天平均销量{avg7:g}件，30天平均销量{avg30:g}件，增长约{growth}%；库存可支撑{inventory_support:.1f}天。',
                f'分析{code}销量增长后是否需要提前补货',
                ['product_sales', 'planning_metrics'],
                product_code=code, metrics={'7天平均销量': avg7, '30天平均销量': avg30,
                                             '增长幅度': growth, '库存支撑天数': inventory_support},
                source_page='/planning'))

    # 2. 已逾期但仍未到货的订单。
    for row in orders:
        status = str(row.get('状态') or '')
        outstanding = _int(row.get('未到货数量'))
        expected = _parse_date(row.get('预计到货日期'))
        if status in ('已完结', '已完成') or outstanding <= 0 or not expected or expected >= today:
            continue
        overdue_days = (today - expected).days
        severity = '高' if overdue_days >= 7 or outstanding >= 5000 else '中'
        code = str(row.get('product_code') or '')
        supplier = str(row.get('供应商') or '')
        items.append(_insight(
            'overdue_order', '订单交付', severity,
            f'{row.get("订单类型") or "订单"} {row.get("订单ID")} 已逾期{overdue_days}天',
            f'{code or row.get("货品名称") or "该商品"}未到货{outstanding}件，预计到货日期{expected.isoformat()}，供应商{supplier or "未填写"}。',
            f'分析订单{row.get("订单ID")}延期风险及对{code or "相关商品"}库存的影响',
            ['order_detail'], product_code=code,
            supplier=supplier, metrics={'逾期天数': overdue_days, '未到货数量': outstanding,
                                       '预计到货日期': expected.isoformat()},
            source_page={'国内订单': '/orders/domestic', '海外订单': '/orders/overseas',
                         '包材订单': '/orders/packaging'}.get(row.get('订单类型'), '/orders/domestic')))

    # 3. 供应商履约风险。
    for row in suppliers:
        supplier = str(row.get('供应商') or '')
        open_orders = _int(row.get('未完结订单数'))
        outstanding = _int(row.get('未到货数量'))
        completion = _num(row.get('到货完成率'))
        defect = _num(row.get('不良品率'))
        if open_orders < 2 and outstanding < 3000 and defect < 3:
            continue
        if completion < 50 or outstanding >= 20000 or defect >= 5:
            severity = '高'
        elif completion < 75 or outstanding >= 5000 or defect >= 2:
            severity = '中'
        else:
            continue
        items.append(_insight(
            'supplier_risk', '供应商风险', severity,
            f'{supplier or "未填写供应商"} 交付风险偏高',
            f'未完结订单{open_orders}单，未到货{outstanding}件，到货完成率{completion:.2f}%，不良品率{defect:.2f}%。',
            f'分析{supplier}供应商的未完结订单和交付风险',
            ['supplier_delivery'], supplier=supplier,
            metrics={'未完结订单数': open_orders, '未到货数量': outstanding,
                     '到货完成率': completion, '不良品率': defect}, source_page='/suppliers'))

    # 4. BOM瓶颈：按成品取最小可支撑数量，与建议下单/30天需求对比。
    bom_by_product = defaultdict(list)
    for row in bom:
        code = str(row.get('product_code') or '')
        if code:
            bom_by_product[code].append(row)
    for code, materials in bom_by_product.items():
        plan = planning_map.get(code) or {}
        daily = _num(plan.get('日销量预测'))
        suggested = _int(plan.get('建议下单数量'))
        # P1只把系统已经建议下单的成品纳入BOM预警，避免将低销量/历史BOM误报为即时缺口。
        target = suggested
        if target <= 0:
            continue
        usable = [m for m in materials if _num(m.get('单位用量')) > 0]
        if not usable:
            continue
        bottleneck = min(usable, key=lambda m: _num(m.get('可支撑成品数量')))
        support_qty = _int(bottleneck.get('可支撑成品数量'))
        if support_qty >= target:
            continue
        gap = target - support_qty
        severity = '高' if support_qty < target * 0.7 else '中'
        material = str(bottleneck.get('物料名称') or bottleneck.get('material_code') or 'BOM物料')
        items.append(_insight(
            'bom_shortage', 'BOM物料', severity,
            f'{code} 的{material}可能不足',
            f'目标需求{target}件成品，当前该物料仅可支撑{support_qty}件，预计缺口{gap}件。',
            f'分析{code}的BOM包材缺口并给出补货建议',
            ['bom_supply_chain', 'planning_metrics'], product_code=code,
            supplier=str(bottleneck.get('物料供应商') or ''),
            metrics={'目标需求': target, '可支撑成品数量': support_qty, '预计缺口': gap,
                     '瓶颈物料': material}, source_page='/bom'))

    # 同一类/对象保留最高严重度，避免相同规则重复刷屏。
    unique = {}
    for item in items:
        key = (item['type'], item.get('product_code'), item.get('supplier'), item['title'])
        current = unique.get(key)
        if not current or _severity_rank(item['severity']) > _severity_rank(current['severity']):
            unique[key] = item
    items = list(unique.values())
    items.sort(key=lambda x: (-_severity_rank(x['severity']), x['category'], x['title']))
    return {
        'generated_at': datetime.now().strftime('%Y-%m-%d %H:%M:%S'),
        'items': items,
        'data_dates': {
            key: datasource_meta(key).get('data_date') or ''
            for key in ('planning_metrics', 'product_sales', 'product_inventory', 'inventory_value')
        },
    }


def get_operational_insights(force=False):
    if force:
        cache_delete(CACHE_KEY)
    data = None if force else cache_get(CACHE_KEY)
    if data is None:
        data = _build_raw_insights()
        cache_set(CACHE_KEY, data, ttl=CACHE_TTL)
    return data


def position_profile(position, role=''):
    position = str(position or '')
    profiles = {
        'procurement_planner': {
            'name': '采购计划', 'focus_types': ['stockout', 'bom_shortage', 'overstock', 'sales_growth'],
            'questions': ['哪些商品最需要优先补货？', '哪些建议下单商品存在BOM包材缺口？',
                          '哪些商品库存支撑天数过高？', '哪些销量增长商品需要提前备货？']},
        'procurement_follower': {
            'name': '采购跟单', 'focus_types': ['overdue_order', 'supplier_risk', 'bom_shortage'],
            'questions': ['当前有哪些逾期未到货订单？', '哪些供应商未完结订单最多？',
                          '哪些延期订单会造成商品缺货？', '供应商到货完成率和不良品情况如何？']},
        'operator': {
            'name': '运营专员', 'focus_types': ['sales_decline', 'sales_growth', 'overstock', 'stockout'],
            'questions': ['哪些商品近7天销量明显下降？', '哪些商品销量增长但库存不足？',
                          '哪些商品无销量但仍有大量库存？', '分析各品牌最近30天出库表现']},
        'product_developer': {
            'name': '产品开发', 'focus_types': ['bom_shortage', 'supplier_risk', 'overstock'],
            'questions': ['哪些成品存在BOM物料瓶颈？', '哪些包材供应商交付风险较高？',
                          '新品相关产品是否存在库存积压？', '查询某个产品的BOM和供应商信息']},
        'finance_specialist': {
            'name': '财务专员', 'focus_types': ['overstock', 'supplier_risk'],
            'questions': ['当前库存金额及主要构成是什么？', '哪些高库存金额商品动销较差？',
                          '供应商未到货数量和资金占用情况如何？', '各品牌库存金额和动销SKU数如何？']},
        'bp': {
            'name': 'BP', 'focus_types': ['stockout', 'overstock', 'sales_decline', 'sales_growth',
                                         'overdue_order', 'supplier_risk', 'bom_shortage'],
            'questions': ['生成今日经营风险分析报告', '当前最高风险的商品和订单有哪些？',
                          '库存、销量、订单和供应商有哪些异常？', '哪些问题需要管理层优先推动？']},
        'procurement_trainee': {
            'name': '采购管培生', 'focus_types': ['stockout', 'overdue_order', 'supplier_risk', 'bom_shortage'],
            'questions': ['当前有哪些需要关注的采购风险？', '哪些订单已经逾期？',
                          '哪些商品需要补货？', '哪些BOM物料可能不足？']},
    }
    if position in profiles:
        return profiles[position]
    if role in ('admin', 'super_admin'):
        return profiles['bp'] | {'name': '管理人员'}
    return {
        'name': '业务人员',
        'focus_types': ['stockout', 'overstock', 'sales_decline', 'overdue_order', 'bom_shortage'],
        'questions': ['当前有哪些高风险业务异常？', '哪些商品需要补货？',
                      '哪些订单已经延期？', '哪些商品库存积压？']
    }


def _impact_score(item):
    """在同一严重度内按业务影响排序，防止大量低影响异常淹没真正风险。"""
    metrics = item.get('metrics') or {}
    qty = max(_num(metrics.get(k)) for k in (
        '预计缺口', '未到货数量', '建议下单数量', '库存合计', '下单数量'
    ))
    days = max(_num(metrics.get(k)) for k in ('逾期天数', '补货阈值'))
    ratio = max(_num(metrics.get(k)) for k in ('下降幅度', '增长幅度', '不良品率'))
    amount = _num(metrics.get('库存金额')) / 100.0
    return qty + days * 100 + ratio * 10 + amount

def filter_insights(raw, allowed_sources, profile, limit=80):
    allowed = set(allowed_sources or [])
    focus = profile.get('focus_types') or []
    items = []
    for item in raw.get('items') or []:
        source_keys = item.get('source_keys') or []
        # 生成该异常实际用到的数据源必须全部有权限，避免跨菜单侧漏。
        if not source_keys or any(k not in allowed for k in source_keys):
            continue
        copy = dict(item)
        copy['source_keys'] = list(source_keys)
        copy['metrics'] = dict(copy.get('metrics') or {})
        if 'inventory_value' not in allowed:
            copy['metrics'].pop('库存金额', None)
        copy['focused'] = copy.get('type') in focus
        items.append(copy)
    items.sort(key=lambda x: (0 if x.get('focused') else 1, -_severity_rank(x['severity']),
                              -_impact_score(x), x['title']))
    # 按异常类型和风险等级均衡取样，避免高风险数量太多导致中风险完全看不到。
    max_items = max(1, min(int(limit or 80), 200))
    grouped = defaultdict(list)
    for item in items:
        grouped[item['type']].append(item)
    quota = max(4, (max_items + max(len(grouped), 1) - 1) // max(len(grouped), 1))
    balanced = []
    for group_items in grouped.values():
        by_severity = defaultdict(list)
        for item in group_items:
            by_severity[item['severity']].append(item)
        chosen = by_severity['高'][:max(1, int(quota * 0.65))]
        chosen += by_severity['中'][:max(1, int(quota * 0.3))]
        chosen += by_severity['低'][:max(1, quota - len(chosen))]
        if len(chosen) < quota:
            seen_ids = {x['id'] for x in chosen}
            chosen += [x for x in group_items if x['id'] not in seen_ids][:quota-len(chosen)]
        balanced.extend(chosen[:quota])
    balanced.sort(key=lambda x: (0 if x.get('focused') else 1, -_severity_rank(x['severity']),
                                 -_impact_score(x), x['title']))
    items = balanced[:max_items]
    counts = defaultdict(int)
    categories = defaultdict(int)
    for item in items:
        counts[item['severity']] += 1
        categories[item['category']] += 1
    return {
        'generated_at': raw.get('generated_at'),
        'data_dates': raw.get('data_dates') or {},
        'items': items,
        'counts': {'high': counts['高'], 'medium': counts['中'], 'low': counts['低'], 'total': len(items)},
        'categories': dict(categories),
        'profile': profile,
    }