2024-07-09 18:11:08 +08:00

156 lines
5.3 KiB
Python

import os
import unicodedata
from typing import Optional, Dict, List, Union
from pathlib import Path
from sqlalchemy import or_, select
from sqlalchemy.sql import ColumnElement
from nonebot.log import logger
from nonebot.params import Arg, Depends
from nonebot.typing import T_State
from nonebot.matcher import Matcher
from nonebot.adapters import Message
from nonebot_plugin_orm import get_session
from nonebot_plugin_session import Session, SessionLevel, extract_session
from nonebot_plugin_session_orm import SessionModel
from nonebot_plugin_userinfo import EventUserInfo, UserInfo
from nonebot_plugin_htmlrender import html_to_pic,template_to_pic
from nonebot_plugin_apscheduler import scheduler
from nonebot_plugin_chatrecorder import MessageRecord
from nonebot_plugin_localstore import get_cache_dir
from nonebot_plugin_alconna import AlconnaMatcher
from .config import plugin_config
from .model import UserRankInfo
cache_path = get_cache_dir("nonebot_plugin_dialectlist")
# 暂时不做考虑
# def admin_permission():
# permission = SUPERUSER
# with contextlib.suppress(ImportError):
# from nonebot.adapters.onebot.v11.permission import GROUP_ADMIN, GROUP_OWNER
# permission = permission | GROUP_ADMIN | GROUP_OWNER
# return permission
async def ensure_group(matcher: Matcher, session: Session = Depends(extract_session)):
"""确保在群组中使用"""
if session.level not in [SessionLevel.LEVEL2, SessionLevel.LEVEL3]:
await matcher.finish("请在群组中使用!")
async def persist_id2user_id(ids: List) -> List[str]:
whereclause: List[ColumnElement[bool]] = []
whereclause.append(or_(*[SessionModel.id == id for id in ids]))
statement = (
select(SessionModel).where(*whereclause)
# .join(SessionModel, SessionModel.id == MessageRecord.session_persist_id)
)
async with get_session() as db_session:
records = (await db_session.scalars(statement)).all()
return [i.id1 for i in records]
async def user_id2persist_id(id: str) -> int:
whereclause: List[ColumnElement[bool]] = []
whereclause.append(or_(*[SessionModel.id2 == id]))
statement = (
select(SessionModel).where(*whereclause)
# .join(SessionModel, SessionModel.id == MessageRecord.session_persist_id)
)
async with get_session() as db_session:
records = (await db_session.scalars(statement)).all()
return records[0].id
def msg_counter(msg_list: List[MessageRecord]) -> Dict[str, int]:
"""### 计算每个人的消息量
Args:
msg_list (list[MessageRecord]): 需要处理的消息列表
Returns:
(dict[str,int]): 处理后的消息数量字典,键为用户,值为消息数量
"""
lst: Dict[str, int] = {}
msg_len = len(msg_list)
logger.info("wow , there are {} msgs to count !!!".format(msg_len))
for i in msg_list:
try:
lst[str(i.session_persist_id)] += 1
except KeyError:
lst[str(i.session_persist_id)] = 1
logger.debug(f"finish counting, result is {lst}")
return lst
def got_rank(msg_dict: Dict[str, int]) -> List:
"""### 获得排行榜
Args:
msg_dict (Dict[str,int]): 要处理的字典
Returns:
List[Tuple[str,int]]: 排行榜列表(已排序)
"""
rank = []
while len(rank) < plugin_config.get_num:
try:
max_key = max(msg_dict.items(), key=lambda x: x[1])
rank.append(list(max_key))
msg_dict.pop(max_key[0])
except ValueError:
logger.error(
"群内拥有聊天记录的人数不足,无法获取到长度为{}的排行榜,已将长度变化为:{}".format(
plugin_config.get_num, len(rank)
)
)
break
return rank
def remove_control_characters(string: str) -> str:
"""### 将字符串中的控制符去除
Args:
string (str): 需要去除的字符串
Returns:
(str): 经过处理的字符串
"""
return "".join(ch for ch in string if unicodedata.category(ch)[0] != "C")
async def get_rank_image(rank: List[UserRankInfo]) -> bytes:
for i in rank:
if i.user_avatar:
try:
user_avatar = i.user_avatar_bytes
except NotImplementedError:
user_avatar = open(os.path.dirname(os.path.abspath(__file__))+"/template/avatar/default.jpg", "rb").read()
# if not os.path.exists(cache_path / str(i.user_id)):
with open(cache_path / (str(i.user_id) + ".jpg"), "wb") as f:
f.write(user_avatar)
if plugin_config.template_path[:2] == './':
path = os.path.dirname(os.path.abspath(__file__)) + plugin_config.template_path[1:]
else:
path = plugin_config.template_path
path_dir, filename = os.path.split(path)
logger.debug(os.path.dirname(os.path.abspath(__file__)) + plugin_config.template_path[1:])
return await template_to_pic(path_dir,
filename,
{'users': rank,
'cache_path': cache_path,
'file_path': os.path.dirname(os.path.abspath(__file__))},
pages={"viewport": {"width": 1366, "height": 10}})