shaguabijia-app-server/app/repositories/store_mapping.py

"""平台店铺映射落库:一次比价(trace)一行,各目标平台解析出 id 就 upsert 进同一行。

合并键 = trace_id(唯一约束)。一次跨平台比价 = 一个 trace = 一个真实店铺:淘宝腿解析出
shopId 先建行(填淘宝列),京东腿(将来)解析出 id 再 upsert 进**同一行**(填京东列)。

合并策略 = 填空(fill-the-blanks):只写该行当前为 NULL 的列,绝不覆盖已有非空值。保证后到
的平台只填自己那几列、动不了先到平台的数据;共享列(geo / source / 溯源)先到先得。
不依赖 ON CONFLICT,跨方言(PG / SQLite dev)都安全;并发撞唯一约束则回滚后转走合并路径。

⚠️ 一行里 id_taobao 与 id_jd 共存只表示"两条腿搜同一个源店名各自匹配到了某家店",是
name-match 置信度、非已核实同一实体。作为 append-only 原始资产留存,跨平台精确匹配由下游做。
"""
from __future__ import annotations

import logging
import math

from sqlalchemy import select
from sqlalchemy.exc import IntegrityError
from sqlalchemy.orm import Session

from app.models.store_mapping import StoreMapping
from app.schemas.store_mapping import StoreMappingIn

logger = logging.getLogger("shagua.store_mapping")

# 源平台 → 该平台店名所在列(缓存查询的匹配键)。镜像 pricebot reporter 的同名表。
_SOURCE_NAME_COLUMN = {
    "meituan": "name_meituan",
    "meituan_waimai": "name_meituan",
    "jd_waimai": "name_jd",
    "jd_waimai_standalone": "name_jd",
    "taobao_flash": "name_taobao",
}

# 源平台 → 它自己对应的目标 key(查缓存时排除"源平台自己", 不会复用源平台的店铺 id)。
_SOURCE_TARGET_KEY = {
    "meituan": "meituan", "meituan_waimai": "meituan",
    "jd_waimai": "jd", "jd_waimai_standalone": "jd",
    "taobao_flash": "taobao",
}

# upsert 填空时可写的列。不含: trace_id(合并键)/ business_type(非空默认)/
# id(主键)/ created_at(server_default)。各平台只会带自己那几列非空, 其余为 None 不动。
_MERGE_COLUMNS = (
    "source_platform",
    "id_taobao", "name_taobao", "id_meituan", "name_meituan", "id_jd", "name_jd",
    "city", "geohash", "lng", "lat", "taobao_address",
    "source_device_id", "source_user_id",
    "taobao_share_url", "taobao_resolved_url", "taobao_deeplink",
    "meituan_poi_id_str", "meituan_share_url", "meituan_resolved_url", "meituan_deeplink",
    "jd_vender_id", "jd_share_url", "jd_resolved_url", "jd_deeplink",
    "attrs",
)


def _find(db: Session, trace_id: str) -> StoreMapping | None:
    return db.execute(
        select(StoreMapping).where(StoreMapping.trace_id == trace_id)
    ).scalar_one_or_none()


def _merge_fill_blanks(existing: StoreMapping, payload: StoreMappingIn) -> list[str]:
    """把 payload 里非空、且 existing 当前为 NULL 的列填进去。返回被填的列名(空=无变化)。"""
    filled: list[str] = []
    for col in _MERGE_COLUMNS:
        new = getattr(payload, col)
        if new is not None and getattr(existing, col) is None:
            setattr(existing, col, new)
            filled.append(col)
    return filled


def upsert(db: Session, payload: StoreMappingIn) -> tuple[int, int | None]:
    """落一行跨平台店铺映射,返回 (created, row_id)。
    created=1 新建该 trace 的行 / 0 合并进已存在行(填空,不覆盖)。"""
    existing = _find(db, payload.trace_id)
    if existing is None:
        # 首写: 把 payload 全部可合并列灌进去(逐列 setattr 而非硬编码构造器, 否则首写的是
        # 美团/京东腿时它们的列会漏 —— 不在硬编码列表里就丢)。trace_id/business_type 是键/默认, 显式给。
        row = StoreMapping(
            trace_id=payload.trace_id,
            business_type=payload.business_type,
        )
        for col in _MERGE_COLUMNS:
            setattr(row, col, getattr(payload, col))
        db.add(row)
        try:
            db.commit()
        except IntegrityError:
            # 并发: 另一个请求刚插了同 trace → 撞唯一约束。回滚后转合并路径填空。
            db.rollback()
            existing = _find(db, payload.trace_id)
            if existing is None:
                raise
            logger.warning("store_mapping 并发冲突 trace=%s, 转填空合并", payload.trace_id)
        else:
            db.refresh(row)
            return 1, row.id

    # 已存在(或并发回退到此): 填空合并, 只写当前 NULL 的列
    filled = _merge_fill_blanks(existing, payload)
    if filled:
        db.commit()
        db.refresh(existing)
        logger.info("store_mapping 合并 trace=%s 填列=%s", payload.trace_id, filled)
    return 0, existing.id


# ============================================================
# 缓存查询: 比价前按"源平台店名"反查已沉淀的各目标平台店铺 id, 命中就让 pricebot 直接
# deeplink 跳店内搜索, 省掉"开平台→进店→分享反查"整段。
# ============================================================

def _haversine_km(lat1: float, lng1: float, lat2: float, lng2: float) -> float:
    """两点球面距离(km)。仅用于同名候选里挑最近, 精度够用。"""
    r = 6371.0
    p1, p2 = math.radians(lat1), math.radians(lat2)
    dp = math.radians(lat2 - lat1)
    dl = math.radians(lng2 - lng1)
    a = math.sin(dp / 2) ** 2 + math.cos(p1) * math.cos(p2) * math.sin(dl / 2) ** 2
    return 2 * r * math.asin(math.sqrt(a))


def _pick_best(rows: list[StoreMapping], lat: float | None, lng: float | None) -> StoreMapping:
    """同名 + 含目标 id 的候选里挑一条:有入参 geo 且有候选带 geo → 取最近;否则取 created_at 最新。"""
    if lat is not None and lng is not None:
        geod = [r for r in rows if r.lat is not None and r.lng is not None]
        if geod:
            return min(geod, key=lambda r: _haversine_km(lat, lng, r.lat, r.lng))
    return max(rows, key=lambda r: r.created_at)


# 目标 key → (该平台店铺 id 列, 组装返回 payload 的函数)。pricebot 拿 id 现拼 deeplink。
_TARGETS = {
    "taobao": ("id_taobao", lambda r: {"shop_id": r.id_taobao, "deeplink": r.taobao_deeplink}),
    "jd": ("id_jd", lambda r: {"store_id": r.id_jd, "vender_id": r.jd_vender_id, "deeplink": r.jd_deeplink}),
    "meituan": ("meituan_poi_id_str", lambda r: {"poi_id_str": r.meituan_poi_id_str, "deeplink": r.meituan_deeplink}),
}


def lookup_nearest(
    db: Session, source_platform: str, store_name: str,
    lat: float | None = None, lng: float | None = None,
) -> dict:
    """按"源平台店名"反查各目标平台已沉淀的店铺 id。返回 {target_key: {id..., deeplink, row_id, ...}}。
    - 匹配键 = 源平台对应的 name 列 == store_name(精确)。
    - 每个目标**分别**取"含该目标 id 的同名候选里最近一条"(淘宝 id / 京东 id 可能在不同行)。
    - 排除源平台自己(不复用源平台的店铺 id)。命中为空 = 没缓存, pricebot 走现场反查老路。"""
    name_col = _SOURCE_NAME_COLUMN.get(source_platform)
    if not name_col or not store_name:
        return {}
    rows = db.execute(
        select(StoreMapping).where(getattr(StoreMapping, name_col) == store_name)
    ).scalars().all()
    if not rows:
        return {}
    src_key = _SOURCE_TARGET_KEY.get(source_platform)
    out: dict = {}
    for tgt, (id_attr, make_payload) in _TARGETS.items():
        if tgt == src_key:
            continue  # 不返回源平台自己
        cands = [r for r in rows if getattr(r, id_attr)]
        if not cands:
            continue
        best = _pick_best(cands, lat, lng)
        payload = make_payload(best)
        payload["row_id"] = best.id
        if best.lat is not None and best.lng is not None and lat is not None and lng is not None:
            payload["dist_km"] = round(_haversine_km(lat, lng, best.lat, best.lng), 3)
        out[tgt] = payload
    return out