akg-factor-bridge/test_judgement_snapshot.py

"""日频行业观点快照的离线单测（不连库）。

覆盖：从研判结论视图正常取数并归一（只要产业研判与环节评析两类、自我校验三值、校验问题条数、
指纹与生成日期）；视图读失败返回空列表不抛错；同一计划日重跑覆盖当日行不追加重复行；材料指纹
未变时陈旧天数累加、采信倾向的上一版值带出来；材料指纹变化时算作一次迁移且陈旧天数归零；
第一次见到的主题不判迁移；版本史里摘迁移的入口函数。

取数、读上一版、落库三处都接受注入的函数，落库这一路用一张内存里的假表接住真实的 SQL
（DELETE 当日行 + 批插），所以幂等测的是真语句不是桩。

开发机没有 pandas 与数据库驱动时，只给缺席的模块装最小桩（与 test_market_context.py 同一约定：
仅在模块缺席时装桩，不覆盖真实模块）。

跑法：python3 test_judgement_snapshot.py   或   pytest test_judgement_snapshot.py
"""
import datetime as dt
import sys
import types

_STUBS = ("pandas", "psycopg", "pymysql", "dotenv")
for _n in _STUBS:
    if _n not in sys.modules:
        try:
            __import__(_n)
        except ImportError:
            _m = types.ModuleType(_n)
            if _n == "pandas":                      # db.py 的函数签名在定义时引用这两个名字
                _m.DataFrame = type("DataFrame", (), {})
                _m.Series = type("Series", (), {})
            sys.modules[_n] = _m

import config      # noqa: E402
import judgement   # noqa: E402
import sources     # noqa: E402


def t(name, cond):
    assert cond, name
    print("  ok", name)


DAY = "2026-09-03"
TABLE = "t_akg_judgement_snapshot_test"


# ---------------------------------------------------------------- 假数据
def _view_rows(fp_a="fp-a-1", leaning_a="偏多"):
    """研判结论视图的两行产业研判、一行环节评析，外加两行该被挡掉的（个股评析、簇键为空）。"""
    return [
        {"scope": "industry", "subject_name": "液冷散热", "segment_name": None,
         "cluster_key": "Industry::液冷散热", "leaning": leaning_a, "n_bull": 4, "n_bear": 2,
         "n_flags": 1, "verified": True, "verify_problems": [], "n_materials": 37,
         "model": "m1", "rounds": 3, "review_date": dt.date(2026, 8, 19),
         "reviewed_at": dt.datetime(2026, 8, 19, 10, 30, 5), "input_version": fp_a,
         "review_id": "rv1"},
        {"scope": "industry", "subject_name": "固态电池", "segment_name": None,
         "cluster_key": "Industry::固态电池", "leaning": "证据不足", "n_bull": "1", "n_bear": 0,
         "n_flags": 0, "verified": "false", "verify_problems": '["缺少产能口径", "价格来源不一致"]',
         "n_materials": 12, "model": "m1", "rounds": 3, "review_date": "2026-08-11",
         "reviewed_at": "2026-08-11T09:00:00", "input_version": "fp-b-1", "review_id": "rv2"},
        {"scope": "segment", "subject_name": "正极材料", "segment_name": "正极材料",
         "cluster_key": "Segment:正极材料", "leaning": "中性", "n_bull": 2, "n_bear": 2,
         "n_flags": None, "verified": None, "verify_problems": None, "n_materials": 8,
         "model": "m1", "rounds": 2, "review_date": "2026-08-20", "reviewed_at": None,
         "input_version": None, "review_id": "rv3"},
        {"scope": "industry", "subject_name": None, "segment_name": None,
         "cluster_key": "Industry::", "leaning": "偏多", "n_bull": 1, "n_bear": 0,
         "n_flags": 0, "verified": True, "verify_problems": None, "n_materials": 3,
         "model": "m1", "rounds": 3, "review_date": "2026-08-01", "reviewed_at": None,
         "input_version": "fp-x", "review_id": "rv4"},
    ]


def _pg_ok(sql, params=None):
    s = " ".join(sql.split())
    assert "v_factor_judgement" in s, s
    assert set(params) == {"industry", "segment"}, params    # 只要这两类
    return _view_rows()


def _boom(*a, **k):
    raise OSError("connection refused")


class _FakeCursor:
    """只认本模块会发的三种语句的假游标：建表、按计划日删、批插。"""

    def __init__(self, store):
        self.store = store

    def __enter__(self):
        return self

    def __exit__(self, *a):
        return False

    def execute(self, sql, params=None):
        s = " ".join(sql.split())
        if s.upper().startswith("CREATE TABLE"):
            self.store["created"] += 1
            assert TABLE in s, s
            return
        if s.upper().startswith("DELETE"):
            assert "WHERE plan_date = %s" in s, s
            day = params[0]
            self.store["deleted"].append(day)
            self.store["rows"] = [r for r in self.store["rows"] if r[0] != day]
            return
        raise AssertionError(f"意外的语句: {s}")

    def executemany(self, sql, rows):
        s = " ".join(sql.split())
        assert s.upper().startswith("INSERT INTO") and TABLE in s, s
        assert s.count("%s") == len(judgement.COLUMNS), s
        for r in rows:
            assert len(r) == len(judgement.COLUMNS)
        self.store["rows"].extend(list(rows))


class _FakeConn:
    def __init__(self, store):
        self.store = store

    def __enter__(self):
        return self

    def __exit__(self, *a):
        return False

    def cursor(self):
        return _FakeCursor(self.store)

    def commit(self):
        self.store["commits"] += 1


def _fake_store():
    return {"rows": [], "deleted": [], "created": 0, "commits": 0}


def _as_dicts(store):
    return [dict(zip(judgement.COLUMNS, r)) for r in store["rows"]]


# ---------------------------------------------------------------- 用例
def test_fetch():
    config.JUDGEMENT_SCOPES = {"industry", "segment"}
    rows = sources.judgement_rows(read_pg=_pg_ok)
    t("只出簇键与主题名齐全的行（空主题那行被挡掉）", len(rows) == 3)
    a = next(r for r in rows if r["cluster_key"] == "Industry::液冷散热")
    t("产业研判字段齐：采信倾向、多空条数、材料指纹、生成日期",
      a["scope"] == "industry" and a["subject_name"] == "液冷散热" and a["leaning"] == "偏多"
      and a["n_bull"] == 4 and a["n_bear"] == 2 and a["input_version"] == "fp-a-1"
      and a["review_date"] == "2026-08-19" and a["reviewed_at"] == "2026-08-19 10:30:05")
    t("自我校验为真、校验问题零条", a["verified"] is True and a["n_verify_problems"] == 0)
    b = next(r for r in rows if r["cluster_key"] == "Industry::固态电池")
    t("字符串形态的数字与真假值都归一、JSON 串的校验问题按条数",
      b["n_bull"] == 1 and b["verified"] is False and b["n_verify_problems"] == 2
      and b["reviewed_at"] == "2026-08-11 09:00:00")
    c = next(r for r in rows if r["scope"] == "segment")
    t("环节评析带环节名；空的自我校验保持为空不当成假",
      c["segment_name"] == "正极材料" and c["verified"] is None
      and c["n_verify_problems"] is None and c["input_version"] is None)
    t("scope 传参可覆盖旋钮",
      sources.judgement_rows(scopes=("industry", "segment"), read_pg=_pg_ok) == rows)


def test_fetch_failure():
    t("视图读失败返回空列表不抛错", sources.judgement_rows(read_pg=_boom) == [])
    config.JUDGEMENT_SCOPES = set()
    t("一类都不抄时不查库", sources.judgement_rows(read_pg=_boom) == [])
    config.JUDGEMENT_SCOPES = {"industry", "segment"}


def test_prev_read_failure():
    config.JUDGEMENT_SNAPSHOT_TABLE = TABLE
    t("表还不存在时上一版为空字典、不抛错", judgement.load_previous(DAY, read_mysql=_boom) == {})
    t("计划日不合法时也只是没有上一版",
      judgement.load_previous("不是日期", read_mysql=_boom) == {})


def test_same_day_idempotent():
    config.JUDGEMENT_SNAPSHOT_TABLE = TABLE
    store = _fake_store()

    def _write(day, rows):
        judgement.save(day, rows, conn_factory=lambda: _FakeConn(store))

    def _fetch():
        return sources.judgement_rows(read_pg=_pg_ok)

    r1 = judgement.snapshot(DAY, fetch=_fetch, load_prev=lambda d: {}, write=_write)
    t("首日写三行、每个主题一行", r1["rows"] == 3 and len(store["rows"]) == 3)
    t("按 scope 计数：产业研判两个、环节评析一个",
      r1["by_scope"] == {"industry": 2, "segment": 1})
    judgement.snapshot(DAY, fetch=_fetch, load_prev=lambda d: {}, write=_write)
    t("同一计划日重跑仍是三行，不追加重复行", len(store["rows"]) == 3)
    t("重跑先删当日行（删的正是这个计划日）", store["deleted"] == [DAY, DAY])
    keys = [r["cluster_key"] for r in _as_dicts(store)]
    t("三个簇键各一行、无重复", sorted(keys) == sorted(set(keys)) and len(keys) == 3)
    t("每行都带计划日与写入时刻",
      all(r["plan_date"] == DAY and len(r["snapshot_at"] or "") == 19 for r in _as_dicts(store)))
    t("首次见到不判迁移、陈旧天数从零起算",
      all(r["migrated"] is None and r["stale_days"] == 0 for r in _as_dicts(store))
      and r1["first_seen"] == 3 and r1["migrated"] == 0)

    # 视图这天一行都读不到：当日行照样先删，不留上一次重跑的残行。
    judgement.snapshot(DAY, fetch=lambda: [], load_prev=lambda d: {}, write=_write)
    t("视图无行时当日行被清空，不留残行", store["rows"] == [])


def test_stale_days_accumulate():
    """指纹未变：陈旧天数按距上一个计划日的自然日数累加，采信倾向的上一版值带出来。"""
    prev = {"Industry::液冷散热": {"plan_date": "2026-09-01", "cluster_key": "Industry::液冷散热",
                                   "leaning": "偏多", "input_version": "fp-a-1", "stale_days": 5}}
    rows = judgement.build_rows(DAY, sources.judgement_rows(read_pg=_pg_ok), prev, now="x")
    a = next(r for r in rows if r["cluster_key"] == "Industry::液冷散热")
    t("指纹未变：不算迁移", a["migrated"] == 0)
    t("陈旧天数 = 上一版 5 天 + 两个计划日相隔 2 天 = 7", a["stale_days"] == 7)
    t("上一版采信倾向带出来", a["leaning_prev"] == "偏多" and a["leaning"] == "偏多")
    other = next(r for r in rows if r["cluster_key"] == "Industry::固态电池")
    t("上一版里没有的主题按第一次见到", other["migrated"] is None and other["stale_days"] == 0)

    # 指纹本身缺失（环节评析那行没有指纹）：不判迁移，但日子照样变老。
    prev2 = {"Segment:正极材料": {"plan_date": "2026-09-02", "leaning": "中性",
                                  "input_version": None, "stale_days": 3}}
    rows = judgement.build_rows(DAY, sources.judgement_rows(read_pg=_pg_ok), prev2, now="x")
    seg = next(r for r in rows if r["cluster_key"] == "Segment:正极材料")
    t("指纹缺失：不判迁移，陈旧天数照常累加", seg["migrated"] is None and seg["stale_days"] == 4)

    # 计划日之间隔了一整个周末也照样按自然日数累加。
    prev3 = {"Industry::液冷散热": {"plan_date": "2026-08-28", "leaning": "偏多",
                                    "input_version": "fp-a-1", "stale_days": 0}}
    rows = judgement.build_rows("2026-08-31", sources.judgement_rows(read_pg=_pg_ok), prev3, now="x")
    a = next(r for r in rows if r["cluster_key"] == "Industry::液冷散热")
    t("跨周末按自然日数累加 3 天", a["stale_days"] == 3)


def test_migration_on_fingerprint_change():
    """指纹变化才算一次迁移：陈旧天数归零，上一版采信倾向留在 leaning_prev 里。"""
    prev = {"Industry::液冷散热": {"plan_date": "2026-09-02", "cluster_key": "Industry::液冷散热",
                                   "leaning": "偏多", "input_version": "fp-a-1", "stale_days": 9}}

    def _fetch_changed(sql=None, params=None):
        return _view_rows(fp_a="fp-a-2", leaning_a="偏空")

    rows = judgement.build_rows(DAY, sources.judgement_rows(read_pg=_fetch_changed), prev, now="x")
    a = next(r for r in rows if r["cluster_key"] == "Industry::液冷散热")
    t("指纹变化：算一次迁移", a["migrated"] == 1)
    t("迁移时陈旧天数归零", a["stale_days"] == 0)
    t("采信倾向从偏多变成偏空，两版都在行上",
      a["leaning_prev"] == "偏多" and a["leaning"] == "偏空")

    # 同一份材料指纹没变、只是采信倾向的文本被重新生成时，不算迁移（口径就是只认指纹）。
    def _fetch_same_fp(sql=None, params=None):
        return _view_rows(fp_a="fp-a-1", leaning_a="偏空")

    rows = judgement.build_rows(DAY, sources.judgement_rows(read_pg=_fetch_same_fp), prev, now="x")
    a = next(r for r in rows if r["cluster_key"] == "Industry::液冷散热")
    t("指纹没变就不算迁移，哪怕采信倾向的文本变了",
      a["migrated"] == 0 and a["stale_days"] == 10)

    hist = [{"plan_date": "2026-09-01", "migrated": None, "leaning": "偏多", "leaning_prev": None,
             "input_version": "fp-a-1", "review_date": "2026-08-19"},
            {"plan_date": "2026-09-02", "migrated": 0, "leaning": "偏多", "leaning_prev": "偏多",
             "input_version": "fp-a-1", "review_date": "2026-08-19"},
            {"plan_date": DAY, "migrated": 1, "leaning": "偏空", "leaning_prev": "偏多",
             "input_version": "fp-a-2", "review_date": "2026-09-02"}]
    got = judgement.migrations(hist)
    t("版本史里只摘出被判为迁移的那一天，带方向与出处",
      len(got) == 1 and got[0]["plan_date"] == DAY and got[0]["leaning_from"] == "偏多"
      and got[0]["leaning_to"] == "偏空" and got[0]["review_date"] == "2026-09-02")
    t("空版本史不报错", judgement.migrations([]) == [] and judgement.migrations(None) == [])


def main():
    test_fetch()
    test_fetch_failure()
    test_prev_read_failure()
    test_same_day_idempotent()
    test_stale_days_accumulate()
    test_migration_on_fingerprint_change()
    print("ALL OK — 行业观点快照：取数归一 / 读失败为空 / 同日重跑幂等 / 陈旧天数累加 / "
          "指纹变化算迁移 / 版本史摘迁移 全部通过")


if __name__ == "__main__":
    main()