288 lines
11 KiB
Python
288 lines
11 KiB
Python
"""冒险系统游戏模块"""
|
||
import random
|
||
import time
|
||
import logging
|
||
from datetime import datetime
|
||
from games.base import BaseGame
|
||
from utils.parser import CommandParser
|
||
from core.database import get_db
|
||
from typing import *
|
||
|
||
logger = logging.getLogger(__name__)
|
||
|
||
|
||
class AdventureGame(BaseGame):
|
||
"""冒险系统游戏"""
|
||
|
||
def __init__(self):
|
||
"""初始化游戏"""
|
||
super().__init__()
|
||
self.db = get_db()
|
||
|
||
# 奖品池配置
|
||
self.prize_pool: List[Tuple[int, float, str]] = [
|
||
# (权重, 倍率, 描述)
|
||
(500, 0.5, "少量积分"),
|
||
(350, 1, "中等积分"),
|
||
(200, 2, "大量积分"),
|
||
(100, 5, "丰厚积分"),
|
||
(50, 10, "丰厚积分"),
|
||
(10, 100, "🌟 巨额积分"),
|
||
(1, 1000, "💎 传说积分"),
|
||
]
|
||
|
||
self.total_weight: int = 0
|
||
for weight,_,_ in self.prize_pool:
|
||
self.total_weight += weight
|
||
|
||
async def handle(self, command: str, chat_id: int, user_id: int) -> str:
|
||
"""处理冒险相关指令
|
||
|
||
Args:
|
||
command: 指令,如 ".adventure", ".adventure 10", ".adventure stats"
|
||
chat_id: 会话ID
|
||
user_id: 用户ID
|
||
|
||
Returns:
|
||
回复消息
|
||
"""
|
||
try:
|
||
# 提取参数
|
||
_, args = CommandParser.extract_command_args(command)
|
||
args = args.strip().lower()
|
||
|
||
|
||
# 冒险说明
|
||
if args in ['help', '帮助', 'info']:
|
||
return self._get_adventure_help()
|
||
|
||
# 放弃当前冒险(按最低倍率结算已冒险时间)
|
||
if args in ['abandon', '放弃']:
|
||
return await self._abandon_adventure(chat_id, user_id)
|
||
|
||
# 默认:冒险耗时1分钟
|
||
else:
|
||
# 解析消耗时间
|
||
cost_time = 1 # 默认消耗1分钟
|
||
if args.isdigit():
|
||
cost_time = int(args)
|
||
|
||
return await self._perform_adventure(chat_id, user_id, cost_time)
|
||
|
||
except Exception as e:
|
||
logger.error(f"处理冒险指令错误: {e}", exc_info=True)
|
||
return f"❌ 处理指令出错: {str(e)}"
|
||
|
||
async def _perform_adventure(self, chat_id: int, user_id: int, cost_time: int) -> str:
|
||
"""执行冒险耗时
|
||
|
||
Args:
|
||
chat_id: 会话ID(使用0作为用户级标识)
|
||
user_id: 用户ID
|
||
cost_time: 消耗时间(分钟)
|
||
|
||
Returns:
|
||
抽奖结果消息
|
||
"""
|
||
# 参数验证
|
||
if cost_time < 1:
|
||
return "❌ 冒险时间至少需要1分钟!"
|
||
|
||
# 查询冒险状态(使用chat_id=0表示用户级状态)
|
||
state = self.db.get_game_state(0, user_id, 'adventure')
|
||
current_time = int(time.time())
|
||
|
||
# 情况1:用户已有未完成的冒险
|
||
if state:
|
||
try:
|
||
state_data = state['state_data']
|
||
start_time = state_data.get('start_time', 0)
|
||
saved_cost_time = state_data.get('cost_time', 1)
|
||
end_time = start_time + saved_cost_time * 60
|
||
remaining_seconds = end_time - current_time
|
||
|
||
# 情况1.1:冒险已完成(时间已到或过期)
|
||
if remaining_seconds <= 0:
|
||
# 选择奖品池
|
||
prize_pool = self.prize_pool
|
||
|
||
# 执行抽奖
|
||
reward = self._draw_prize(prize_pool)
|
||
reward_points = int(reward['value'] * saved_cost_time)
|
||
|
||
# 处理奖励
|
||
self.db.add_points(user_id, reward_points, "adventure", f"冒险奖励")
|
||
|
||
# 删除冒险状态
|
||
self.db.delete_game_state(0, user_id, 'adventure')
|
||
|
||
# 获取更新后的积分信息
|
||
updated_points = self.db.get_user_points(user_id)
|
||
|
||
# 格式化输出
|
||
text = f"## ⚡️ 冒险结果\n\n"
|
||
text += f"**消耗时间**: {saved_cost_time} 分钟\n\n"
|
||
text += f"**{reward['description']}**: 获得 {reward_points} 积分\n\n"
|
||
text += f"**当前积分**: {updated_points['points']} 分\n\n"
|
||
text += "---\n\n"
|
||
text += "💡 提示:冒险进行时不能炼金!"
|
||
|
||
return text
|
||
|
||
# 情况1.2:冒险未完成,返回等待提示
|
||
remaining_minutes = remaining_seconds // 60
|
||
remaining_secs = remaining_seconds % 60
|
||
|
||
if remaining_minutes > 0:
|
||
wait_msg = f"{remaining_minutes} 分 {remaining_secs} 秒"
|
||
else:
|
||
wait_msg = f"{remaining_secs} 秒"
|
||
|
||
text = f"## ⚡️ 冒险进行中\n\n"
|
||
text += f"你正在进行一次冒险,还需等待 **{wait_msg}** 才能完成。\n\n"
|
||
text += f"**当前冒险时长**: {saved_cost_time} 分钟\n\n"
|
||
text += "---\n\n"
|
||
text += "💡 提示:冒险期间无法进行炼金,请耐心等待!"
|
||
|
||
return text
|
||
|
||
except Exception as e:
|
||
# 状态数据异常,删除损坏状态并允许重新开始
|
||
logger.error(f"冒险状态数据异常: {e}", exc_info=True)
|
||
self.db.delete_game_state(0, user_id, 'adventure')
|
||
# 继续执行到情况3(开始新冒险)
|
||
|
||
# 情况3:用户没有冒险状态,开始新的冒险
|
||
state_data = {
|
||
'start_time': current_time,
|
||
'cost_time': cost_time
|
||
}
|
||
|
||
# 保存冒险状态
|
||
self.db.save_game_state(0, user_id, 'adventure', state_data)
|
||
|
||
# 计算预计完成时间
|
||
end_time = current_time + cost_time * 60
|
||
end_datetime = datetime.fromtimestamp(end_time)
|
||
end_time_str = end_datetime.strftime('%H:%M:%S')
|
||
|
||
text = f"## ⚡️ 冒险开始\n\n"
|
||
text += f"你已经开始了冒险之旅,本次冒险将持续 **{cost_time}** 分钟。\n\n"
|
||
text += f"**预计完成时间**: {end_time_str}\n\n"
|
||
text += "---\n\n"
|
||
text += "💡 提示:冒险期间无法进行炼金,完成后使用 `.adventure` 获取奖励!"
|
||
|
||
return text
|
||
|
||
async def _abandon_adventure(self, chat_id: int, user_id: int) -> str:
|
||
"""放弃当前冒险,按最低倍率结算已冒险时间
|
||
|
||
Args:
|
||
chat_id: 会话ID(使用0作为用户级标识)
|
||
user_id: 用户ID
|
||
|
||
Returns:
|
||
放弃结果消息
|
||
"""
|
||
try:
|
||
# 查询冒险状态
|
||
state = self.db.get_game_state(0, user_id, 'adventure')
|
||
if not state:
|
||
return "❌ 当前没有进行中的冒险,可使用 `.adventure` 开始新的冒险。"
|
||
|
||
state_data = state.get('state_data', {})
|
||
start_time = state_data.get('start_time')
|
||
cost_time = state_data.get('cost_time')
|
||
if start_time is None or cost_time is None:
|
||
# 状态异常,清理并提示
|
||
self.db.delete_game_state(0, user_id, 'adventure')
|
||
return "⚠️ 冒险状态异常已清理,请使用 `.adventure` 重新开始。"
|
||
|
||
current_time = int(time.time())
|
||
elapsed_seconds = max(0, current_time - int(start_time))
|
||
elapsed_minutes = elapsed_seconds // 60
|
||
if elapsed_minutes < 1:
|
||
elapsed_minutes = 1
|
||
|
||
# 计算最低倍率
|
||
try:
|
||
min_multiplier = min(m for _, m, _ in self.prize_pool)
|
||
except Exception:
|
||
# 兜底:若奖池异常,按0.5处理
|
||
min_multiplier = 0.5
|
||
|
||
reward_points = int(min_multiplier * elapsed_minutes)
|
||
if reward_points < 0:
|
||
reward_points = 0
|
||
|
||
# 发放奖励并清理状态
|
||
if reward_points > 0:
|
||
self.db.add_points(user_id, reward_points, "adventure", "冒险放弃奖励")
|
||
self.db.delete_game_state(0, user_id, 'adventure')
|
||
|
||
# 查询当前积分
|
||
updated_points = self.db.get_user_points(user_id)
|
||
|
||
# 输出
|
||
text = f"## ⚡️ 冒险放弃\n\n"
|
||
text += f"**已计入时间**: {elapsed_minutes} 分钟\n\n"
|
||
text += f"**最低倍率**: {min_multiplier} 倍\n\n"
|
||
text += f"**获得积分**: {reward_points} 分\n\n"
|
||
text += f"**当前积分**: {updated_points['points']} 分\n\n"
|
||
text += "---\n\n"
|
||
text += "💡 提示:可随时使用 `.adventure` 再次踏上冒险之旅!"
|
||
return text
|
||
except Exception as e:
|
||
logger.error(f"放弃冒险时出错: {e}", exc_info=True)
|
||
# 失败时不影响原状态,返回提示
|
||
return f"❌ 放弃冒险失败:{str(e)}"
|
||
|
||
def _draw_prize(self, prize_pool: list) -> dict:
|
||
"""从奖品池中抽取奖品
|
||
|
||
Args:
|
||
prize_pool: 奖品池,格式为 (权重, 倍率, 描述)
|
||
|
||
Returns:
|
||
奖品信息,格式为 {'value': 倍率, 'description': 描述}
|
||
"""
|
||
# 生成随机数
|
||
rand = random.random()*self.total_weight
|
||
cumulative_prob = 0.0
|
||
|
||
for weight, multiplier, description in prize_pool:
|
||
cumulative_prob += weight
|
||
if rand <= cumulative_prob:
|
||
return {
|
||
'value': multiplier,
|
||
'description': description
|
||
}
|
||
|
||
# 兜底返回第一个奖品
|
||
return {
|
||
'value': prize_pool[0][1],
|
||
'description': prize_pool[0][2]
|
||
}
|
||
|
||
def _get_adventure_help(self) -> str:
|
||
"""获取冒险帮助信息
|
||
|
||
Returns:
|
||
帮助信息消息
|
||
"""
|
||
text = f"## ⚡️ 冒险系统\n\n"
|
||
text += f"### 基础用法\n"
|
||
text += f"- `.adventure` - 消耗1分钟进行冒险\n"
|
||
text += f"- `.adventure time` - 消耗time分钟进行冒险, 最少一分钟\n"
|
||
|
||
text += f"### 其他功能\n"
|
||
text += f"- `.adventure abandon` - 放弃当前冒险,按最低倍率结算已冒险时间\n"
|
||
text += f"- `.adventure 放弃` - 放弃当前冒险,按最低倍率结算已冒险时间\n"
|
||
text += f"- `.adventure help` - 查看帮助\n\n"
|
||
|
||
return text
|
||
|
||
def get_help(self) -> str:
|
||
"""获取帮助信息"""
|
||
return self._get_adventure_help()
|