import sqlite3
import json
from datetime import datetime, timedelta
from typing import List, Dict, Optional, Tuple


# =========================
# 1) 状态持久化：按“最近交易日记录”统计
# =========================
class StatePersistence:
    """状态持久化管理（最小改动增强版）"""

    def __init__(self, db_path="state_persistence.db"):
        self.db_path = db_path
        self.init_database()

    def init_database(self):
        conn = sqlite3.connect(self.db_path)
        cursor = conn.cursor()
        cursor.execute(
            '''
            CREATE TABLE IF NOT EXISTS daily_state (
                date TEXT PRIMARY KEY,
                final_state TEXT,
                raw_state TEXT,
                score REAL,
                consecutive_days INTEGER,
                switch_count_last_5days INTEGER,
                previous_position REAL,
                alerts TEXT
            )
            '''
        )
        conn.commit()
        conn.close()

    def save_daily_state(
        self,
        date: str,
        final_state: str,
        raw_state: str,
        score: float,
        consecutive_days: int,
        switch_count_last_5days: int,
        previous_position: float,
        alerts=None,
    ):
        conn = sqlite3.connect(self.db_path)
        cursor = conn.cursor()
        alerts_json = json.dumps(alerts) if alerts else "[]"
        cursor.execute(
            '''
            INSERT OR REPLACE INTO daily_state
            (date, final_state, raw_state, score, consecutive_days,
             switch_count_last_5days, previous_position, alerts)
            VALUES (?, ?, ?, ?, ?, ?, ?, ?)
            ''',
            (
                date,
                final_state,
                raw_state,
                score,
                consecutive_days,
                switch_count_last_5days,
                previous_position,
                alerts_json,
            ),
        )
        conn.commit()
        conn.close()

    def load_previous_state(self, date: str) -> Optional[Dict]:
        conn = sqlite3.connect(self.db_path)
        cursor = conn.cursor()
        cursor.execute('SELECT * FROM daily_state WHERE date = ?', (date,))
        result = cursor.fetchone()
        conn.close()

        if result:
            return {
                'date': result[0],
                'final_state': result[1],
                'raw_state': result[2],
                'score': result[3],
                'consecutive_days': result[4],
                'switch_count_last_5days': result[5],
                'previous_position': result[6],
                'alerts': json.loads(result[7]) if result[7] else []
            }
        return None

    def load_latest_state_before(self, current_date: str) -> Optional[Dict]:
        """加载 current_date 之前最近一个交易日状态，而不是机械地 current_date-1。"""
        conn = sqlite3.connect(self.db_path)
        cursor = conn.cursor()
        cursor.execute(
            '''
            SELECT *
            FROM daily_state
            WHERE date < ?
            ORDER BY date DESC
            LIMIT 1
            ''',
            (current_date,),
        )
        result = cursor.fetchone()
        conn.close()

        if result:
            return {
                'date': result[0],
                'final_state': result[1],
                'raw_state': result[2],
                'score': result[3],
                'consecutive_days': result[4],
                'switch_count_last_5days': result[5],
                'previous_position': result[6],
                'alerts': json.loads(result[7]) if result[7] else []
            }
        return None

    def get_recent_states(self, current_date: str, limit: int = 10) -> List[Dict]:
        """取 current_date 之前最近 N 个已保存交易日状态。"""
        conn = sqlite3.connect(self.db_path)
        cursor = conn.cursor()
        cursor.execute(
            '''
            SELECT *
            FROM daily_state
            WHERE date < ?
            ORDER BY date DESC
            LIMIT ?
            ''',
            (current_date, limit),
        )
        rows = cursor.fetchall()
        conn.close()

        out = []
        for r in rows:
            out.append({
                'date': r[0],
                'final_state': r[1],
                'raw_state': r[2],
                'score': r[3],
                'consecutive_days': r[4],
                'switch_count_last_5days': r[5],
                'previous_position': r[6],
                'alerts': json.loads(r[7]) if r[7] else []
            })
        return out

    def calculate_consecutive_days(self, current_state: str, current_date: str, max_lookback: int = 20) -> int:
        """按最近交易日记录统计连续同状态天数，不受周末/节假日影响。"""
        recent = self.get_recent_states(current_date, limit=max_lookback)
        consecutive_days = 1  # 包含今天
        for row in recent:
            if row['final_state'] == current_state:
                consecutive_days += 1
            else:
                break
        return consecutive_days

    def calculate_switch_count_last_5days(
        self,
        current_date: str,
        candidate_state: Optional[str] = None,
        window: int = 5,
    ) -> int:
        """
        统计最近 window 个“交易日状态”的切换次数。
        支持把今天候选状态 candidate_state 纳入统计，避免慢一天。
        """
        recent = self.get_recent_states(current_date, limit=max(window - 1, 1))
        states = []
        if candidate_state is not None:
            states.append(candidate_state)
        states.extend([x['final_state'] for x in recent])

        if len(states) <= 1:
            return 0

        switch_count = 0
        prev = states[0]
        for s in states[1:window]:
            if s != prev:
                switch_count += 1
            prev = s
        return switch_count


# =========================
# 2) 轻量确认门：少改框架，减少碎切并提升准确率
# =========================
STATE_ORDER = {
    "RISK_SEVERE": 0,
    "RISK_MILD": 1,
    "RANGE": 2,
    "TREND_NORMAL": 3,
    "TREND_BULL": 4,
}


def _is_upgrade(prev_state: str, new_state: str) -> bool:
    return STATE_ORDER.get(new_state, -999) > STATE_ORDER.get(prev_state, -999)


def _is_downgrade(prev_state: str, new_state: str) -> bool:
    return STATE_ORDER.get(new_state, -999) < STATE_ORDER.get(prev_state, -999)


def stabilize_state_transition(
    candidate_state: str,
    current_date: str,
    persistence: StatePersistence,
    severe_state: str = "RISK_SEVERE",
    confirm_upgrade_days: int = 2,
    confirm_downgrade_days: int = 1,
    confirm_exit_severe_days: int = 2,
    min_hold_days: int = 2,
) -> str:
    """
    在不改核心评分框架前提下，对 final_state 再包一层轻量确认门。

    原则：
    - 严重风险即时进入
    - 严重风险退出需要更高确认
    - 升级慢（减少追涨误判）
    - 降级可快（保留风控能力）
    - 最短持有期（防碎切）
    """
    prev = persistence.load_latest_state_before(current_date)
    if not prev:
        return candidate_state

    prev_state = prev['final_state']
    prev_consecutive = int(prev.get('consecutive_days') or 1)

    if candidate_state == prev_state:
        return candidate_state

    # 严重风险：允许即时进入
    if candidate_state == severe_state:
        return candidate_state

    # 严重风险退出：要求更高确认，避免刚出险就来回打脸
    if prev_state == severe_state and candidate_state != severe_state:
        recent = persistence.get_recent_states(current_date, limit=max(confirm_exit_severe_days, 5))
        # 需要最近若干个“原已落地状态”都不是 severe，才放行退出
        non_severe_streak = 0
        for row in recent:
            if row['final_state'] != severe_state:
                non_severe_streak += 1
            else:
                break
        if non_severe_streak < (confirm_exit_severe_days - 1):
            return prev_state
        return candidate_state

    # 普通状态最短持有期：优先减少一天一切
    if prev_consecutive < min_hold_days:
        # 允许朝更危险方向快速切，但不允许朝更乐观方向太快切
        if _is_upgrade(prev_state, candidate_state):
            return prev_state

    recent = persistence.get_recent_states(current_date, limit=max(confirm_upgrade_days, confirm_downgrade_days, 5))
    same_as_candidate_streak = 0
    for row in recent:
        if row['final_state'] == candidate_state:
            same_as_candidate_streak += 1
        else:
            break

    if _is_upgrade(prev_state, candidate_state):
        # 升级慢：要求最近已有连续候选方向痕迹，降低假突破
        need = max(confirm_upgrade_days - 1, 0)
        if same_as_candidate_streak < need:
            return prev_state
        return candidate_state

    if _is_downgrade(prev_state, candidate_state):
        # 降级快：一般放行；如果你想更稳，可把 confirm_downgrade_days 设成 2
        need = max(confirm_downgrade_days - 1, 0)
        if same_as_candidate_streak < need:
            return prev_state if confirm_downgrade_days > 1 else candidate_state
        return candidate_state

    return candidate_state


# =========================
# 3) 仓位限速：0仓也限速，降低误判伤害
# =========================
def apply_position_limit(new_position: float, daily_change_limit: float, previous_position: Optional[float]):
    """
    只对“真正首日（None）”不设限。
    previous_position == 0 也要限速，防止误判时一步跳太大。
    """
    if previous_position is None:
        return new_position

    max_change = daily_change_limit
    actual_change = new_position - previous_position

    if abs(actual_change) > max_change:
        if actual_change > 0:
            return previous_position + max_change
        return previous_position - max_change
    return new_position


# =========================
# 4) 主流程：最小侵入式接入
# =========================
def run_daily_pipeline_fixed(date, core_stocks, config):
    """
    最小侵入增强版：
    - 前一日状态改为“最近交易日状态”
    - final_state 后加轻量确认门
    - 切换次数把今天候选状态纳入统计
    """
    state_persistence = StatePersistence()

    # 最近一个已保存交易日，而不是机械地 date-1
    previous_state = state_persistence.load_latest_state_before(date)

    market_data = get_market_data(date)
    state_machine = V13StateMachine(config)

    score = state_machine.calculate_score(market_data)
    raw_state = state_machine.determine_state(score)
    cooled_state = state_machine.apply_cooling_period(raw_state, score)

    # 轻量确认门：不改核心评分框架，只改“状态落地”
    final_state = stabilize_state_transition(
        candidate_state=cooled_state,
        current_date=date,
        persistence=state_persistence,
        severe_state="RISK_SEVERE",
        confirm_upgrade_days=config.get("STATE_STABILITY_CONFIG", {}).get("confirm_upgrade_days", 2),
        confirm_downgrade_days=config.get("STATE_STABILITY_CONFIG", {}).get("confirm_downgrade_days", 1),
        confirm_exit_severe_days=config.get("STATE_STABILITY_CONFIG", {}).get("confirm_exit_severe_days", 2),
        min_hold_days=config.get("STATE_STABILITY_CONFIG", {}).get("min_hold_days", 2),
    )

    consecutive_days = state_persistence.calculate_consecutive_days(final_state, date)
    switch_count_last_5days = state_persistence.calculate_switch_count_last_5days(
        current_date=date,
        candidate_state=final_state,
        window=5,
    )

    base_confidence = state_machine.calculate_confidence(final_state, score)
    adjusted_confidence, _, _ = adjust_confidence_by_stability(
        base_confidence,
        consecutive_days,
        switch_count_last_5days,
    )

    previous_position = previous_state['previous_position'] if previous_state else None
    suggested_position = calculate_continuous_position_fixed(
        final_state,
        score,
        config["POSITION_MAPPING"],
        config["POSITION_LIMIT_CONFIG"]["daily_change_limit"],
        previous_position,
    )

    execution_permissions = calculate_execution_permissions(final_state, adjusted_confidence)

    trade_candidates = []
    for stock in core_stocks:
        stock_data = get_stock_data(stock, date)
        alpha_score = calculate_alpha_score(stock_data, final_state)
        threshold = get_state_threshold(final_state)

        if alpha_score >= threshold and execution_permissions['allow_new_positions']:
            action, conf = calibrate_confidence_by_bucket(
                adjusted_confidence,
                config["EXECUTION_ACTIONS"],
            )
            position = calculate_stock_position(action, suggested_position)
            trade_candidates.append({
                "stock": stock,
                "alpha_score": alpha_score,
                "threshold": threshold,
                "action": action,
                "position": position,
                "intercepted": False,
            })
        else:
            intercept_reason = []
            if alpha_score < threshold:
                intercept_reason.append(
                    f"Alpha score {alpha_score:.1f} < threshold {threshold:.1f}"
                )
            if not execution_permissions['allow_new_positions']:
                intercept_reason.append("New positions not allowed in current state")

            trade_candidates.append({
                "stock": stock,
                "alpha_score": alpha_score,
                "threshold": threshold,
                "action": "intercepted",
                "position": 0,
                "intercepted": True,
                "reason": "; ".join(intercept_reason),
            })

    state_persistence.save_daily_state(
        date,
        final_state,
        raw_state,
        score,
        consecutive_days,
        switch_count_last_5days,
        suggested_position,
        [],
    )

    report = {
        "date": date,
        "market_state": final_state,
        "raw_state": raw_state,
        "cooled_state": cooled_state,
        "confidence": adjusted_confidence,
        "suggested_position": suggested_position,
        "execution_permissions": execution_permissions,
        "trade_candidates": trade_candidates,
        "consecutive_days": consecutive_days,
        "switch_count_last_5days": switch_count_last_5days,
    }
    return report


# =========================
# 5) 配置：只加很少参数，避免过拟合
# =========================
DEFAULT_STATE_STABILITY_CONFIG = {
    # 只用粗颗粒参数，避免调出“历史最优”细节
    "confirm_upgrade_days": 2,
    "confirm_downgrade_days": 1,
    "confirm_exit_severe_days": 2,
    "min_hold_days": 2,
}


# =========================
# 6) 建议验证方法（写在注释里，方便回测时照着看）
# =========================
# A. 先固定这4个参数，不要来回精调：2 / 1 / 2 / 2
# B. 只做 very small grid：
#    - confirm_upgrade_days: [2, 3]
#    - confirm_downgrade_days: [1, 2]
#    - confirm_exit_severe_days: [2, 3]
#    - min_hold_days: [2, 3]
# C. 评价顺序：
#    1) 切换次数是否明显下降
#    2) 最大回撤是否不恶化
#    3) 收益是否改善
# D. 若只有某一组参数特别好、邻近参数都差，则视为过拟合嫌疑