perf(portfolio): G5-P2收尾—all_weather三调用点向量化:①filter_roic逐只循环→一次批量+fields=['roic']短路(9×)②_trend_mean get_price长表+pivot→get_closes_panel宽表(340×,count=N→start-N*2自然日+.tail(N)同momentum口径,fq均raw)③prepare_stock_list get_price(close+high_limit)→get_limit_status_batch精确涨跌停口径;测试fixture补panel/limit_batch mock(原MagicMock碰巧truthy蒙混paused剔除语义,默认改全正常);+6回归测试,portfolio全套302绿 [vps]

sanguo_portfolio/strategies/all_weather.py

@@ -141,26 +141,18 @@ class AllWeatherStrategy:
         previous_date = _previous_date_str(context)
         if not previous_date:
             return
-        try:
-            df = self.provider.get_price(
-                self.hold_list,
-                end_date=previous_date,
-                frequency="daily",
-                fields=["close", "high_limit"],
-                count=1,
-                panel=False,
-                fill_paused=False,
-            )
-        except Exception as exc:
-            logger.warning("prepare_stock_list get_price 失败: %s", exc)
-            return
-        if df is None or len(df) == 0:
+        fn = getattr(self.provider, "get_limit_status_batch", None)
+        if fn is None:
             return
         try:
-            hit = df[df["close"] == df["high_limit"]]
-            self.yesterday_hl_list = list(hit.get("code", []))
+            status_map = fn(self.hold_list, previous_date) or {}
         except Exception as exc:
-            logger.debug("prepare_stock_list 解析涨停失败: %s", exc)
+            logger.warning("prepare_stock_list get_limit_status_batch 失败: %s", exc)
+            return
+        self.yesterday_hl_list = [
+            s for s in self.hold_list
+            if (status_map.get(s) or {}).get("is_limit_up")
+        ]
 
     # =================== stop_loss ===================
     def stop_loss(self, context: Any) -> None:
@@ -375,18 +367,26 @@ class AllWeatherStrategy:
         """聚宽 filter_roic:只保留 ROIC > 0.08。
 
         聚宽原版用 ``get_factor_values(stock, 'roic_ttm')``,这里走我们自算的 roic
-        (provider 已合并到 df['roic'])。逐只查比 df 慢,但保留原签名便于实证对账。
+        (provider 已合并到 df['roic'])。批量 + ``fields=['roic']`` 按需短路
+        (只读 income/balance 源表,fields 短路 9 倍速,见 provider f416a17)。
         """
+        if not stock_list:
+            return []
         threshold = self.config.roic_threshold
-        out: List[str] = []
-        for stock in stock_list:
-            df = self.provider.get_fundamentals_df([stock], date=previous_date)
-            if df.empty:
-                continue
-            roic = float(df["roic"].iloc[0]) if "roic" in df.columns else float("nan")
-            if roic == roic and roic > threshold:
-                out.append(stock)
-        return out
+        try:
+            df = self.provider.get_fundamentals_df(
+                stock_list, date=previous_date, fields=["roic"],
+            )
+        except Exception as exc:
+            logger.warning("filter_roic get_fundamentals_df 失败: %s", exc)
+            return []
+        if df is None or df.empty or "roic" not in df.columns:
+            return []
+        roic = df["roic"]
+        return [
+            s for s in stock_list
+            if s in roic.index and roic[s] == roic[s] and roic[s] > threshold
+        ]
 
     # =================== 调仓辅助 ===================
     def _close_position(self, code: str) -> bool:
@@ -430,30 +430,28 @@ class AllWeatherStrategy:
         return list(df.index)[:n]
 
     def _trend_mean(self, stocks: List[str], end_date: str, n: int) -> float:
-        """N 日涨幅(% )的均值。聚宽原版取 close 涨幅,这里同样。"""
+        """N 日涨幅(% )的均值。聚宽原版取 close 涨幅,这里同样。
+
+        批量取数:``get_closes_panel`` 宽表(UNION ALL 走复合索引 340×,只取 close)
+        替 ``get_price`` 长表+pivot;``count=n`` → ``start=end-n*2 自然日`` + ``.tail(n)``
+        切片(避免自然日 vs 交易日换算歧义,同 momentum_timing 口径)。fq 均为 raw 不变。
+        """
         if not stocks:
             return 0.0
+        start_date = _shift_date(end_date, -n * 2)
         try:
-            df = self.provider.get_price(
-                stocks,
-                end_date=end_date,
-                frequency="1d",
-                fields=["close"],
-                count=n,
-                panel=False,
+            panel = self.provider.get_closes_panel(
+                stocks, start_date, end_date, fq="raw",
             )
         except Exception as exc:
-            logger.warning("get_price trend 失败: %s", exc)
+            logger.warning("get_closes_panel trend 失败: %s", exc)
             return 0.0
-        if df is None or df.empty:
+        if panel is None or panel.empty:
             return 0.0
-        try:
-            pivot = df.pivot(index="time", columns="code", values="close")
-        except Exception:
+        panel = panel.tail(n)
+        if len(panel) < 2:
             return 0.0
-        if len(pivot) < 2:
-            return 0.0
-        change = (pivot.iloc[-1] / pivot.iloc[0] - 1) * 100
+        change = (panel.iloc[-1] / panel.iloc[0] - 1) * 100
         arr = np.nan_to_num(change.to_numpy())
         return float(np.mean(arr))
 
@@ -578,4 +576,16 @@ def _dedup(seq: Sequence[str]) -> List[str]:
     return out
 
 
+def _shift_date(date_str: str, days: int) -> str:
+    """字符串日期加减自然日,返回 YYYY-MM-DD(供 count=N → start=end-N*2 换算)。
+
+    同 momentum_timing._shift_date 口径;本地定义避免互相 import 循环。
+    """
+    try:
+        dt = datetime.datetime.strptime(date_str[:10], "%Y-%m-%d")
+    except (ValueError, TypeError):
+        return date_str
+    return (dt + datetime.timedelta(days=days)).strftime("%Y-%m-%d")
+
+
 __all__ = ["AllWeatherStrategy", "AllWeatherConfig", "BrokerFacade"]

tests/portfolio/test_all_weather.py

@@ -68,6 +68,30 @@ def make_strategy(
 
     provider.get_price.side_effect = _get_price
 
+    # get_closes_panel:从 price_df_map 同源构造 close 宽表(忽略 count 差异)
+    def _get_closes_panel(symbols, start=None, end=None, interval="d", fq="raw"):
+        frames = [
+            df[["time", "code", "close"]]
+            for (sec, fields, _cnt), df in price_df_map.items()
+            if sec == str(symbols) and tuple(fields) == ("close",)
+            and df is not None and len(df)
+        ]
+        if not frames:
+            return pd.DataFrame()
+        long_df = pd.concat(frames)
+        return long_df.pivot(index="time", columns="code", values="close")
+
+    provider.get_closes_panel.side_effect = _get_closes_panel
+
+    # 涨跌停/停牌批量:默认全正常(filter_paused 无bar=剔除语义,MagicMock/{} 会误杀)
+    def _get_limit_status_batch(codes, date=None):
+        return {
+            c: {"is_limit_up": False, "is_limit_down": False, "is_paused": False}
+            for c in codes
+        }
+
+    provider.get_limit_status_batch.side_effect = _get_limit_status_batch
+
     broker = BrokerFacade()
     broker.order_target_value = MagicMock(return_value=MagicMock(filled=100))
     broker.order_value = MagicMock(return_value=MagicMock(filled=100))
@@ -136,23 +160,69 @@ class TestPrepareStockList:
 
     def test_records_yesterday_limit_up(self):
         s = make_strategy()
-        # 清掉 make_strategy 设置的 side_effect,直接用 return_value
-        s.provider.get_price.side_effect = None
         pos = MagicMock(); pos.security = "600519.XSHG"
         ctx = MagicMock()
         ctx.portfolio.positions = {"600519.XSHG": pos}
         ctx.previous_date = "2024-09-30"
-        # close == high_limit 视为涨停
-        s.provider.get_price.return_value = pd.DataFrame({
-            "code": ["600519.XSHG"],
-            "close": [10.0],
-            "high_limit": [10.0],
-        })
+        # get_limit_status_batch 返回昨日涨停(G5-P2:原 get_price close==high_limit 改批量口径)
+        s.provider.get_limit_status_batch.side_effect = \
+            lambda codes, date=None: {
+                c: {"is_limit_up": True, "is_limit_down": False, "is_paused": False}
+                for c in codes
+            }
         s.prepare_stock_list(ctx)
         assert "600519.XSHG" in s.yesterday_hl_list
 
+    def test_uses_batch_endpoint_once(self):
+        """G5-P2:原 get_price(close+high_limit) → get_limit_status_batch 一次批量。"""
+        s = make_strategy()
+        pos = MagicMock(); pos.security = "600519.XSHG"
+        ctx = MagicMock()
+        ctx.portfolio.positions = {"600519.XSHG": pos, "000001.XSHE": MagicMock(security="000001.XSHE")}
+        ctx.previous_date = "2024-09-30"
+        s.prepare_stock_list(ctx)
+        assert s.provider.get_limit_status_batch.call_count == 1
+        args, _kwargs = s.provider.get_limit_status_batch.call_args
+        assert sorted(args[0]) == ["000001.XSHE", "600519.XSHG"]
+        s.provider.get_price.assert_not_called()
 
-# =================== stop_loss ===================
+
+# =================== trend_mean ===================
+class TestTrendMean:
+    def _panel_seed(self):
+        return pd.DataFrame({
+            "time": pd.to_datetime(["2024-09-20", "2024-09-30"]),
+            "code": ["600519.XSHG"] * 2,
+            "close": [10.0, 15.0],  # 涨 50%
+        })
+
+    def test_uses_closes_panel_not_get_price(self):
+        """G5-P2:原 get_price 长表+pivot → get_closes_panel 宽表。"""
+        s = make_strategy(price_df_map={
+            ("['600519.XSHG']", ("close",), 10): self._panel_seed(),
+        })
+        out = s._trend_mean(["600519.XSHG"], "2024-09-30", 10)
+        assert out == 50.0
+        s.provider.get_closes_panel.assert_called_once()
+        s.provider.get_price.assert_not_called()
+
+    def test_empty_or_short_panel_returns_zero(self):
+        """panel 缺失/不足 2 行 → 0.0 不崩(降级同原 get_price 空df语义)。"""
+        s = make_strategy()
+        assert s._trend_mean([], "2024-09-30", 10) == 0.0
+        assert s._trend_mean(["600519.XSHG"], "2024-09-30", 10) == 0.0  # 无种子数据
+
+    def test_nan_only_change_not_propagated(self):
+        """某列 NaN(close 缺失)→ nan_to_num 归 0,不产出 NaN。"""
+        wide = pd.DataFrame(
+            {"600519.XSHG": [np.nan, 15.0], "000001.XSHE": [10.0, 11.0]},
+            index=pd.to_datetime(["2024-09-20", "2024-09-30"]),
+        )
+        s = make_strategy()
+        s.provider.get_closes_panel.side_effect = None
+        s.provider.get_closes_panel.return_value = wide
+        # 600519 NaN→0%, 000001 +10% → 均值 5(浮点容差)
+        assert s._trend_mean(["600519.XSHG", "000001.XSHE"], "2024-09-30", 10) == pytest.approx(5.0)
 class TestStopLoss:
     def test_stop_loss_triggers_when_price_drops_8pct(self):
         """avg_cost=100, price=91 (< 100*0.92=92) → 止损。"""
@@ -467,3 +537,15 @@ class TestFilterRoic:
         s = make_strategy()
         out = s.filter_roic([], previous_date="2024-09-30")
         assert out == []
+
+    def test_batch_single_call_with_fields_shortcut(self):
+        """G5-P2:原逐只循环 → 一次批量 + fields=['roic'] 短路。"""
+        df = make_fund_df([{"code": "A.XSHG", "roic": 0.15}])
+        s = make_strategy()
+        s.provider.get_fundamentals_df.return_value = df
+        s.filter_roic(["A.XSHG", "B.XSHG"], previous_date="2024-09-30")
+        # 一次批量调用(非逐只 N 次),且带 fields 短路
+        assert s.provider.get_fundamentals_df.call_count == 1
+        args, kwargs = s.provider.get_fundamentals_df.call_args
+        assert list(args[0]) == ["A.XSHG", "B.XSHG"]
+        assert kwargs.get("fields") == ["roic"]