mirror of
https://github.com/TeamPGM/PagerMaid_Plugins.git
synced 2024-11-24 11:53:36 +00:00
1108 lines
46 KiB
Python
1108 lines
46 KiB
Python
import re, time, asyncio, requests, os, json, random
|
||
from io import BytesIO
|
||
from os import path, remove, makedirs, chdir
|
||
from shutil import copyfile, move, rmtree
|
||
from uuid import uuid4
|
||
from base64 import b64encode, b64decode
|
||
from importlib import import_module
|
||
from pagermaid import bot, redis, log, redis_status, working_dir, version
|
||
from pagermaid import user_id as me_id
|
||
from pagermaid.listener import listener
|
||
from pagermaid.utils import alias_command
|
||
from telethon.errors.rpcerrorlist import StickersetInvalidError
|
||
from telethon.tl.functions.messages import GetStickerSetRequest
|
||
from telethon.tl.types import InputStickerSetID, Channel
|
||
|
||
msg_freq = 1
|
||
group_last_time = {}
|
||
read_context = {}
|
||
|
||
|
||
def is_num(x: str):
|
||
try:
|
||
x = int(x)
|
||
return isinstance(x, int)
|
||
except ValueError:
|
||
return False
|
||
|
||
|
||
def encode(s: str):
|
||
return str(b64encode(s.encode("utf-8")), "utf-8")
|
||
|
||
|
||
def decode(s: str):
|
||
return str(b64decode(s.encode("utf-8")), "utf-8")
|
||
|
||
|
||
def random_str():
|
||
return str(uuid4()).replace("-", "")
|
||
|
||
|
||
def parse_rules(rules: str):
|
||
n_rules = {}
|
||
rules_parse = rules.split(";")
|
||
for p in rules_parse:
|
||
d = p.split(":")
|
||
if len(d) == 2:
|
||
key = decode(d[0])
|
||
value = decode(d[1])
|
||
n_rules[key] = value
|
||
return n_rules
|
||
|
||
|
||
def save_rules(rules: dict, placeholder: str):
|
||
n_rules = ""
|
||
for k, v in rules.items():
|
||
if placeholder:
|
||
k = k.replace(placeholder, "'")
|
||
v = v.replace(placeholder, "'")
|
||
n_rules += encode(k) + ":" + encode(v) + ";"
|
||
return n_rules
|
||
|
||
|
||
def validate(user_id: str, mode: int, user_list: list):
|
||
if mode == 0:
|
||
return user_id not in user_list
|
||
elif mode == 1:
|
||
return user_id in user_list
|
||
else:
|
||
return False
|
||
|
||
|
||
def validsent(trig: int, tmp):
|
||
if tmp:
|
||
return int(tmp.get("trig", "0"))
|
||
else:
|
||
return trig
|
||
|
||
|
||
def get_redis(db_key: str):
|
||
byte_data = redis.get(db_key)
|
||
byte_data = byte_data if byte_data else b""
|
||
byte_data = str(byte_data, "ascii")
|
||
return parse_rules(byte_data)
|
||
|
||
|
||
def parse_multi(rule: str):
|
||
sep_ph = random_str()
|
||
col_ph = random_str()
|
||
rule = rule.replace(r"\||", sep_ph)
|
||
rule = rule.replace(r"\::", col_ph)
|
||
rule = rule.split("||")
|
||
n_rule = []
|
||
for r in rule:
|
||
p = r.split("::")
|
||
p = [i.replace(sep_ph, "||") for i in p]
|
||
p = [i.replace(col_ph, "::") for i in p]
|
||
data = ["plain", ""]
|
||
if len(p) == 2:
|
||
data = p
|
||
else:
|
||
data[1] = p[0]
|
||
n_rule.append(data)
|
||
return n_rule
|
||
|
||
|
||
def get_capture(search_data, group_name: str):
|
||
try:
|
||
capture_data = search_data.group(group_name)
|
||
return capture_data
|
||
except:
|
||
return None
|
||
|
||
|
||
def get_rule(chat_id, rule_type, rule_index):
|
||
rule_index = int(rule_index)
|
||
rule_data = get_redis(f"keyword.{chat_id}.{rule_type}")
|
||
index = 0
|
||
for k in rule_data.keys():
|
||
if index == rule_index:
|
||
return encode(k)
|
||
index += 1
|
||
return None
|
||
|
||
|
||
def valid_time(chat_id):
|
||
global msg_freq, group_last_time
|
||
cus_freq = get_redis(f"keyword.{chat_id}.settings").get("freq", msg_freq)
|
||
try:
|
||
cus_freq = float(cus_freq)
|
||
except:
|
||
cus_freq = msg_freq
|
||
n_time = time.time()
|
||
chat_id = int(chat_id)
|
||
if chat_id in group_last_time:
|
||
if n_time - group_last_time[chat_id] >= cus_freq:
|
||
return True
|
||
else:
|
||
return False
|
||
else:
|
||
return True
|
||
|
||
|
||
def has_cache(chat_id, mode, trigger, filename):
|
||
basepath = f"data/keyword_cache/{chat_id}/{mode}:{encode(trigger)}"
|
||
filepath = f"{basepath}/{filename}"
|
||
if not path.exists(basepath):
|
||
makedirs(basepath)
|
||
return (False, filepath)
|
||
if not path.exists(filepath):
|
||
return (False, filepath)
|
||
return (True, filepath)
|
||
|
||
|
||
def cache_opened(chat_id, mode, trigger):
|
||
rule_data = get_redis(f"keyword.{chat_id}.single"
|
||
f".{mode}.{encode(trigger)}").get("cache", None)
|
||
chat_data = get_redis(f"keyword.{chat_id}.settings").get("cache", None)
|
||
global_data = get_redis("keyword.settings").get("cache", None)
|
||
if rule_data:
|
||
return True if rule_data == "1" else False
|
||
elif chat_data:
|
||
return True if chat_data == "1" else False
|
||
elif global_data:
|
||
return True if global_data == "1" else False
|
||
return False
|
||
|
||
|
||
def getsetting(chat_id, mode, trigger, name, default):
|
||
g_settings = get_redis("keyword.settings")
|
||
n_settings = get_redis(f"keyword.{chat_id}.settings")
|
||
s_settings = get_redis(f"keyword.{chat_id}.single.{mode}.{encode(trigger)}")
|
||
final = default
|
||
if s_settings.get(name, None):
|
||
final = s_settings[name]
|
||
elif n_settings.get(name, None):
|
||
final = n_settings[name]
|
||
elif g_settings.get(name, None):
|
||
final = g_settings[name]
|
||
return final
|
||
|
||
|
||
async def aexec(code, *args, **kwargs):
|
||
exec(
|
||
f"async def func(*args, **kwargs):" +
|
||
"".join(f"\n {p}" for p in code.split("\n"))
|
||
)
|
||
return await locals()["func"](*args, **kwargs)
|
||
|
||
|
||
async def del_msg(context, t_lim):
|
||
await asyncio.sleep(t_lim)
|
||
try:
|
||
await context.delete()
|
||
except:
|
||
pass
|
||
|
||
|
||
async def send_reply(chat_id, trigger, mode, reply_msg, context):
|
||
try:
|
||
# 拷贝一份数据
|
||
real_chat_id = chat_id
|
||
chat = context.chat
|
||
sender = context.sender
|
||
sender_channel = True if isinstance(sender, Channel) else False
|
||
replace_data = {}
|
||
if chat_id < 0:
|
||
replace_data = {
|
||
"chat_id": chat.id,
|
||
"chat_name": chat.title
|
||
}
|
||
if sender and not sender_channel:
|
||
replace_data["user_id"] = sender.id
|
||
replace_data["first_name"] = sender.first_name if sender.first_name else ""
|
||
replace_data["last_name"] = sender.last_name if sender.last_name else ""
|
||
if sender and sender_channel:
|
||
replace_data["user_id"] = sender.id
|
||
replace_data["first_name"] = sender.title
|
||
else:
|
||
replace_data["user_id"] = chat_id
|
||
if sender and not sender_channel:
|
||
replace_data["first_name"] = sender.first_name if sender.first_name else ""
|
||
replace_data["last_name"] = sender.last_name if sender.last_name else ""
|
||
if chat:
|
||
replace_data["chat_id"] = chat.id
|
||
last_name = chat.last_name
|
||
if not last_name:
|
||
last_name = ""
|
||
replace_data["chat_name"] = f"{chat.first_name} {last_name}"
|
||
# 校验能否发送消息
|
||
could_send_msg = valid_time(chat_id)
|
||
if not could_send_msg:
|
||
return
|
||
message_list = []
|
||
for re_type, re_msg in reply_msg:
|
||
try:
|
||
update_last_time = False
|
||
catch_pattern = r"\$\{func_(?P<str>((?!\}).)+)\}"
|
||
count = 0
|
||
bracket_str = random_str()
|
||
re_msg = re_msg.replace(r"\}", bracket_str)
|
||
# 函数功能相关
|
||
while re.search(catch_pattern, re_msg) and count < 20:
|
||
func_exec = re.search(catch_pattern, re_msg).group("str")
|
||
try:
|
||
func_name = func_exec
|
||
func_args = None
|
||
if func_exec.strip().endswith(")"):
|
||
arg_index = func_exec.find("(")
|
||
func_name = func_exec[0:arg_index].replace(bracket_str, "}")
|
||
func_args = func_exec[arg_index + 1:-1].replace(bracket_str, "}")
|
||
module = f"import_module('data.keyword_func.{func_name}').main"
|
||
parameter = f"context{', %s' % func_args if func_args else ''}"
|
||
func_data = await eval(f"{module}({parameter})")
|
||
except:
|
||
func_data = "[RE]"
|
||
chdir(working_dir)
|
||
re_msg = re_msg.replace("${func_%s}" % func_exec, str(func_data))
|
||
count += 1
|
||
re_msg = re_msg.replace(bracket_str, "}")
|
||
for k, v in replace_data.items():
|
||
re_type = re_type.replace(f"${k}", str(v))
|
||
re_msg = re_msg.replace(f"${k}", str(v))
|
||
type_parse = re_type.split(",")
|
||
edit_id = -1
|
||
for s in type_parse:
|
||
if len(s) >= 5 and "ext_" == s[0:4] and is_num(s[4:]):
|
||
chat_id = int(s[4:])
|
||
type_parse.remove(s)
|
||
elif len(s) >= 6 and "edit_" == s[0:5] and is_num(s[5:]):
|
||
edit_id = int(s[5:])
|
||
type_parse.remove(s)
|
||
# 处理file和photo
|
||
if ("file" in type_parse or "photo" in type_parse or "sticker" in type_parse) and len(re_msg.split()) >= 2:
|
||
update_last_time = True
|
||
re_data = re_msg.split(" ")
|
||
cache_exists, filename = has_cache(chat_id, mode, trigger, re_data[0])
|
||
is_opened = cache_opened(chat_id, mode, trigger)
|
||
if "sticker" not in type_parse:
|
||
if is_opened:
|
||
# 开启缓存,处理缓存
|
||
if not cache_exists:
|
||
if re_data[1][0:7] == "file://":
|
||
re_data[1] = re_data[1][7:]
|
||
copyfile(" ".join(re_data[1:]), filename)
|
||
else:
|
||
fileget = requests.get(" ".join(re_data[1:]))
|
||
with open(filename, "wb") as f:
|
||
f.write(fileget.content)
|
||
else:
|
||
# 未开启缓存
|
||
if re_data[1][0:7] == "file://":
|
||
re_data[1] = re_data[1][7:]
|
||
copyfile(" ".join(re_data[1:]), filename)
|
||
else:
|
||
fileget = requests.get(" ".join(re_data[1:]))
|
||
with open(filename, "wb") as f:
|
||
f.write(fileget.content)
|
||
reply_to = None
|
||
if "reply" in type_parse:
|
||
reply_to = context.id
|
||
redir = getsetting(chat_id, mode, trigger, "redir", "0")
|
||
reply = await context.get_reply_message()
|
||
if redir == "1" and reply:
|
||
reply_to = reply.id
|
||
if "sticker" in type_parse:
|
||
# 单独处理回复贴纸
|
||
sticker_params = re_data[0].split(",")
|
||
sticker_idx = re_data[1].split(",")
|
||
try:
|
||
stickers = await context.client(
|
||
GetStickerSetRequest(
|
||
stickerset=InputStickerSetID(
|
||
id=int(sticker_params[0]), access_hash=int(sticker_params[1]))))
|
||
except StickersetInvalidError:
|
||
await log('贴纸配置错误。')
|
||
return
|
||
try:
|
||
i = random.randint(0, len(sticker_idx) - 1)
|
||
message_list.append(await bot.send_file(
|
||
chat_id,
|
||
stickers.documents[int(sticker_idx[i])],
|
||
reply_to=reply_to
|
||
))
|
||
except:
|
||
pass
|
||
else:
|
||
if edit_id == -1:
|
||
message_list.append(await bot.send_file(
|
||
chat_id,
|
||
filename,
|
||
reply_to=reply_to,
|
||
force_document=("file" in type_parse)
|
||
))
|
||
else:
|
||
edit_file = await bot.upload_file(filename)
|
||
message_list[edit_id] = await message_list[edit_id].edit(
|
||
file=edit_file,
|
||
force_document=("file" in type_parse)
|
||
)
|
||
# 未开启缓存,删除文件
|
||
if not is_opened:
|
||
remove(filename)
|
||
# 处理tgfile和tgphoto
|
||
elif ("tgfile" in type_parse or "tgphoto" in type_parse) and len(re_msg.split()) >= 2:
|
||
update_last_time = True
|
||
re_data = re_msg.split(" ")
|
||
re_data[0] = " ".join(re_data[0:-1])
|
||
re_data[1] = re_data[-1:][0].split("/")[-2:]
|
||
cache_exists, filename = has_cache(chat_id, mode, trigger, re_data[0])
|
||
is_opened = cache_opened(chat_id, mode, trigger)
|
||
_data = BytesIO()
|
||
try:
|
||
msg_chat_id = int(re_data[1][0])
|
||
except:
|
||
async with bot.conversation(re_data[1][0]) as conversation:
|
||
msg_chat_id = conversation.chat_id
|
||
msg_id_inchat = int(re_data[1][1])
|
||
if is_opened:
|
||
if not cache_exists:
|
||
media_msg = await bot.get_messages(msg_chat_id, ids=msg_id_inchat, offset_id=0)
|
||
if media_msg and media_msg.media:
|
||
try:
|
||
await bot.download_file(media_msg.media.document, _data)
|
||
except:
|
||
await bot.download_file(media_msg.photo, _data)
|
||
with open(filename, "wb") as f:
|
||
f.write(_data.getvalue())
|
||
else:
|
||
media_msg = await bot.get_messages(msg_chat_id, ids=msg_id_inchat, offset_id=0)
|
||
if media_msg and media_msg.media:
|
||
try:
|
||
await bot.download_file(media_msg.media.document, _data)
|
||
except:
|
||
await bot.download_file(media_msg.photo, _data)
|
||
with open(filename, "wb") as f:
|
||
f.write(_data.getvalue())
|
||
reply_to = None
|
||
if "reply" in type_parse:
|
||
reply_to = context.id
|
||
redir = getsetting(chat_id, mode, trigger, "redir", "0")
|
||
reply = await context.get_reply_message()
|
||
if redir == "1" and reply:
|
||
reply_to = reply.id
|
||
if edit_id == -1:
|
||
message_list.append(await bot.send_file(
|
||
chat_id,
|
||
filename,
|
||
reply_to=reply_to,
|
||
force_document=("tgfile" in type_parse)
|
||
))
|
||
else:
|
||
edit_file = await bot.upload_file(filename)
|
||
message_list[edit_id] = await message_list[edit_id].edit(
|
||
file=edit_file,
|
||
force_document=("tgfile" in type_parse)
|
||
)
|
||
if not is_opened:
|
||
remove(filename)
|
||
elif "plain" in type_parse:
|
||
update_last_time = True
|
||
if edit_id == -1:
|
||
message_list.append(await bot.send_message(
|
||
chat_id,
|
||
re_msg,
|
||
link_preview=("nopreview" not in type_parse)
|
||
))
|
||
else:
|
||
message_list[edit_id] = await message_list[edit_id].edit(
|
||
re_msg,
|
||
link_preview=("nopreview" not in type_parse)
|
||
)
|
||
elif "reply" in type_parse and chat_id == real_chat_id:
|
||
update_last_time = True
|
||
if edit_id == -1:
|
||
reply_to = context.id
|
||
redir = getsetting(chat_id, mode, trigger, "redir", "0")
|
||
reply = await context.get_reply_message()
|
||
if redir == "1" and reply:
|
||
reply_to = reply.id
|
||
message_list.append(await bot.send_message(
|
||
chat_id,
|
||
re_msg,
|
||
reply_to=reply_to,
|
||
link_preview=("nopreview" not in type_parse)
|
||
))
|
||
else:
|
||
message_list[edit_id] = await message_list[edit_id].edit(
|
||
re_msg,
|
||
link_preview=("nopreview" not in type_parse)
|
||
)
|
||
elif "op" in type_parse:
|
||
if re_msg == "delete":
|
||
await context.delete()
|
||
elif re_msg.split()[0] == "sleep" and len(re_msg.split()) == 2:
|
||
sleep_time = re_msg.split()[1]
|
||
await asyncio.sleep(float(sleep_time))
|
||
elif re_msg.split()[0] == "delself" and len(re_msg.split()) == 2:
|
||
await message_list[int(re_msg.split()[1])].delete()
|
||
elif re_msg.split()[0] == "trigger" and len(re_msg.split()) == 2:
|
||
await auto_reply(message_list[int(re_msg.split()[1])])
|
||
elif re_msg.split("\n")[0].startswith("exec") and len(re_msg.split("\n")) >= 2:
|
||
args = [
|
||
"\n".join(re_msg.split("\n")[1:]),
|
||
" ".join(re_msg.split("\n")[0].split(" ")[1:])
|
||
]
|
||
await eval(f"aexec(args[0]{f', {args[1]}' if args[1] else ''})")
|
||
chdir(working_dir)
|
||
if update_last_time:
|
||
global group_last_time
|
||
group_last_time[int(chat_id)] = time.time()
|
||
except Exception as e:
|
||
pass
|
||
chat_id = real_chat_id
|
||
except Exception as e:
|
||
pass
|
||
|
||
|
||
@listener(is_plugin=True, outgoing=True, command=alias_command("keyword"),
|
||
description="关键词自动回复 [教程](https://telegra.ph/Keyword-插件使用教程-02-07)",
|
||
parameters="``new <plain|regex> '<规则>' '<回复信息>'` 或者 `del <plain|regex> '<规则>'` 或者 `list` 或者 "
|
||
"`clear <plain|regex>")
|
||
async def reply(context):
|
||
if not redis_status():
|
||
await context.edit("出错了呜呜呜 ~ Redis 离线,无法运行")
|
||
await del_msg(context, 5)
|
||
return
|
||
chat_id = context.chat_id
|
||
plain_dict = get_redis(f"keyword.{chat_id}.plain")
|
||
regex_dict = get_redis(f"keyword.{chat_id}.regex")
|
||
params = context.parameter
|
||
params = " ".join(params)
|
||
placeholder = random_str()
|
||
params = params.replace(r"\'", placeholder)
|
||
tmp_parse = params.split("'")
|
||
parse = []
|
||
for i in range(len(tmp_parse)):
|
||
if len(tmp_parse[i].split()) != 0:
|
||
parse.append(tmp_parse[i])
|
||
if len(parse) == 0 or (
|
||
len(parse[0].split()) == 1 and parse[0].split()[0] in ("new", "del", "delid", "clear")) or len(
|
||
parse[0].split()) > 2:
|
||
await context.edit(
|
||
"[Code: -1] 格式错误,格式为 `-keyword` 加上 `new <plain|regex> '<规则>' '<回复信息>'` 或者 "
|
||
"`del <plain|regex> '<规则>'` 或者 `list` 或者 `clear <plain|regex>`")
|
||
await del_msg(context, 10)
|
||
return
|
||
else:
|
||
parse[0] = parse[0].split()
|
||
if parse[0][0] == "new" and len(parse) == 3:
|
||
if parse[0][1] == "plain":
|
||
plain_dict[parse[1]] = parse[2]
|
||
redis.set(f"keyword.{chat_id}.plain", save_rules(plain_dict, placeholder))
|
||
elif parse[0][1] == "regex":
|
||
regex_dict[parse[1]] = parse[2]
|
||
redis.set(f"keyword.{chat_id}.regex", save_rules(regex_dict, placeholder))
|
||
else:
|
||
await context.edit(
|
||
"[Code: -1] 格式错误,格式为 `-keyword` 加上 `new <plain|regex> '<规则>' '<回复信息>'` 或者 "
|
||
"`del <plain|regex> '<规则>'` 或者 `list` 或者 `clear <plain|regex>`")
|
||
await del_msg(context, 10)
|
||
return
|
||
await context.edit("设置成功")
|
||
await del_msg(context, 5)
|
||
elif parse[0][0] in ("del", "delid") and len(parse) == 2:
|
||
if parse[0][0] == "delid":
|
||
parse[1] = get_rule(chat_id, parse[0][1], parse[1])
|
||
if parse[1]:
|
||
parse[1] = decode(parse[1])
|
||
if parse[0][1] == "plain":
|
||
if parse[1] and parse[1] in plain_dict:
|
||
redis.delete(f"keyword.{chat_id}.single.plain.{encode(parse[1])}")
|
||
plain_dict.pop(parse[1])
|
||
redis.set(f"keyword.{chat_id}.plain", save_rules(plain_dict, placeholder))
|
||
else:
|
||
await context.edit("规则不存在")
|
||
await del_msg(context, 5)
|
||
return
|
||
elif parse[0][1] == "regex":
|
||
if parse[1] and parse[1] in regex_dict:
|
||
redis.delete(f"keyword.{chat_id}.single.regex.{encode(parse[1])}")
|
||
regex_dict.pop(parse[1])
|
||
redis.set(f"keyword.{chat_id}.regex", save_rules(regex_dict, placeholder))
|
||
else:
|
||
await context.edit("规则不存在")
|
||
await del_msg(context, 5)
|
||
return
|
||
else:
|
||
await context.edit(
|
||
"[Code: -1] 格式错误,格式为 `-keyword` 加上 `new <plain|regex> '<规则>' '<回复信息>'` 或者 "
|
||
"`del <plain|regex> '<规则>'` 或者 `list` 或者 `clear <plain|regex>`")
|
||
await del_msg(context, 10)
|
||
return
|
||
await context.edit("删除成功")
|
||
await del_msg(context, 5)
|
||
elif parse[0][0] == "list" and len(parse) == 1:
|
||
plain_msg = "Plain: \n"
|
||
index = 0
|
||
for k, v in plain_dict.items():
|
||
plain_msg += f"`{index}`: `{k}` -> `{v}`\n"
|
||
index += 1
|
||
regex_msg = "Regex: \n"
|
||
index = 0
|
||
for k, v in regex_dict.items():
|
||
regex_msg += f"`{index}`: `{k}` -> `{v}`\n"
|
||
index += 1
|
||
await context.edit(plain_msg + "\n" + regex_msg)
|
||
elif parse[0][0] == "clear" and len(parse) == 1:
|
||
if parse[0][1] == "plain":
|
||
for k in plain_dict.keys():
|
||
redis.delete(f"keyword.{chat_id}.single.plain.{encode(k)}")
|
||
redis.set(f"keyword.{chat_id}.plain", "")
|
||
elif parse[0][1] == "regex":
|
||
for k in regex_dict.keys():
|
||
redis.delete(f"keyword.{chat_id}.single.regex.{encode(k)}")
|
||
redis.set(f"keyword.{chat_id}.regex", "")
|
||
else:
|
||
await context.edit("参数错误")
|
||
await del_msg(context, 5)
|
||
return
|
||
await context.edit("清除成功")
|
||
await del_msg(context, 5)
|
||
else:
|
||
await context.edit(
|
||
"[Code: -1] 格式错误,格式为 `-keyword` 加上 `new <plain|regex> '<规则>' '<回复信息>'` 或者 "
|
||
"`del <plain|regex> '<规则>'` 或者 `list` 或者 `clear <plain|regex>`")
|
||
await del_msg(context, 10)
|
||
return
|
||
|
||
|
||
@listener(outgoing=True, command=alias_command("replyset"),
|
||
description="自动回复设置",
|
||
parameters="help")
|
||
async def reply_set(context):
|
||
if not redis_status():
|
||
await context.edit("出错了呜呜呜 ~ Redis 离线,无法运行")
|
||
await del_msg(context, 5)
|
||
return
|
||
chat_id = context.chat_id
|
||
params = context.parameter
|
||
redis_data = f"keyword.{chat_id}.settings"
|
||
if len(params) >= 1 and params[0] == "global":
|
||
redis_data = "keyword.settings"
|
||
del params[0]
|
||
elif len(params) >= 2 and params[0] in ("plain", "regex") and is_num(params[1]):
|
||
rule_data = get_rule(chat_id, params[0], params[1])
|
||
if rule_data:
|
||
redis_data = f"keyword.{chat_id}.single.{params[0]}.{rule_data}"
|
||
del params[0:2]
|
||
settings_dict = get_redis(redis_data)
|
||
cmd_dict = {
|
||
"help": (1,),
|
||
"mode": (2,),
|
||
"list": (2, 3),
|
||
"freq": (2,),
|
||
"trig": (2,),
|
||
"show": (1,),
|
||
"cache": (2,),
|
||
"redir": (2,),
|
||
"status": (2,),
|
||
"clear": (1,)
|
||
}
|
||
if len(params) < 1:
|
||
await context.edit("参数错误")
|
||
await del_msg(context, 5)
|
||
return
|
||
if params[0] in cmd_dict and len(params) in cmd_dict[params[0]]:
|
||
if params[0] == "help":
|
||
await context.edit('''
|
||
`-replyset show` 或
|
||
`-replyset clear` 或
|
||
`-replyset mode <0/1/clear>` ( 0 表示黑名单,1 表示白名单 ) 或
|
||
`-replyset list <add/del/show/clear> [user_id]` 或
|
||
`-replyset freq <float/clear>` ( float 表示一个正的浮点数,clear 为清除 ) 或
|
||
`-replyset trig <0/1/clear>` ( 0 为关闭,1 为开启,clear 为清除 ) 或
|
||
`-replyset cache <0/1/clear>` ( 0 为关闭,1 为开启 ) 或
|
||
`-replyset status <0/1/clear>` ( 0 为关闭,1 为开启 ) 。
|
||
在 `-replyset` 后面加上 `global` 即为全局设置。
|
||
在 `-replyset` 后面加上 `plain/regex` 规则序号 可以单独设置一条规则。''')
|
||
await del_msg(context, 15)
|
||
return
|
||
elif params[0] == "show":
|
||
defaults = {
|
||
"mode": "未设置 (默认黑名单)",
|
||
"list": "未设置 (默认为空)",
|
||
"freq": "未设置 (默认为 1)",
|
||
"trig": "未设置 (默认关闭)",
|
||
"cache": "未设置 (默认关闭)",
|
||
"redir": "未设置 (默认关闭)",
|
||
"status": "未设置 (默认开启)"
|
||
}
|
||
msg = "Settings: \n"
|
||
for k, v in defaults.items():
|
||
msg += f"`{k}` -> `{settings_dict[k] if k in settings_dict else v}`\n"
|
||
await context.edit(msg)
|
||
return
|
||
elif params[0] == "mode":
|
||
if params[1] in ("0", "1"):
|
||
settings_dict["mode"] = params[1]
|
||
redis.set(redis_data, save_rules(settings_dict, None))
|
||
if params[1] == "0":
|
||
await context.edit("模式已更改为黑名单")
|
||
elif params[1] == "1":
|
||
await context.edit("模式已更改为白名单")
|
||
await del_msg(context, 5)
|
||
return
|
||
elif params[1] == "clear":
|
||
if "mode" in settings_dict:
|
||
del settings_dict["mode"]
|
||
redis.set(redis_data, save_rules(settings_dict, None))
|
||
await context.edit("清除成功")
|
||
await del_msg(context, 5)
|
||
return
|
||
else:
|
||
await context.edit("参数错误")
|
||
await del_msg(context, 5)
|
||
return
|
||
elif params[0] == "list":
|
||
if params[1] == "show" and len(params) == 2:
|
||
user_list = settings_dict.get("list", None)
|
||
if user_list:
|
||
msg = "List: \n"
|
||
for p in user_list.split(","):
|
||
msg += f"`{p}`\n"
|
||
await context.edit(msg)
|
||
return
|
||
else:
|
||
await context.edit("列表为空")
|
||
await del_msg(context, 5)
|
||
return
|
||
elif params[1] == "add" and len(params) == 3:
|
||
if is_num(params[2]):
|
||
tmp = settings_dict.get("list", None)
|
||
if not tmp:
|
||
settings_dict["list"] = params[2]
|
||
else:
|
||
settings_dict["list"] += f",{params[2]}"
|
||
redis.set(redis_data, save_rules(settings_dict, None))
|
||
await context.edit("添加成功")
|
||
await del_msg(context, 5)
|
||
return
|
||
else:
|
||
await context.edit("user_id 需为整数")
|
||
await del_msg(context, 5)
|
||
return
|
||
elif params[1] == "del" and len(params) == 3:
|
||
if is_num(params[2]):
|
||
tmp = settings_dict.get("list", None)
|
||
if tmp:
|
||
user_list = settings_dict["list"].split(",")
|
||
if params[2] in user_list:
|
||
user_list.remove(params[2])
|
||
settings_dict["list"] = ",".join(user_list)
|
||
redis.set(redis_data, save_rules(settings_dict, None))
|
||
await context.edit("删除成功")
|
||
await del_msg(context, 5)
|
||
return
|
||
else:
|
||
await context.edit("user_id 不在列表")
|
||
await del_msg(context, 5)
|
||
return
|
||
else:
|
||
await context.edit("列表为空")
|
||
await del_msg(context, 5)
|
||
return
|
||
else:
|
||
await context.edit("user_id 需为整数")
|
||
await del_msg(context, 5)
|
||
return
|
||
elif params[1] == "clear" and len(params) == 2:
|
||
if "list" in settings_dict:
|
||
del settings_dict["list"]
|
||
redis.set(redis_data, save_rules(settings_dict, None))
|
||
await context.edit("清除成功")
|
||
await del_msg(context, 5)
|
||
return
|
||
else:
|
||
await context.edit("参数错误")
|
||
await del_msg(context, 5)
|
||
return
|
||
elif params[0] == "freq":
|
||
if redis_data == f"keyword.{chat_id}.settings":
|
||
if params[1] == "clear":
|
||
if "freq" in settings_dict:
|
||
del settings_dict["freq"]
|
||
redis.set(redis_data, save_rules(settings_dict, None))
|
||
await context.edit("清除成功")
|
||
await del_msg(context, 5)
|
||
return
|
||
else:
|
||
try:
|
||
tmp = float(params[1])
|
||
if tmp > 0:
|
||
settings_dict["freq"] = params[1]
|
||
redis.set(redis_data, save_rules(settings_dict, None))
|
||
await context.edit("设置成功")
|
||
await del_msg(context, 5)
|
||
return
|
||
else:
|
||
await context.edit("频率需为正数")
|
||
await del_msg(context, 5)
|
||
return
|
||
except:
|
||
await context.edit("频率需为正数")
|
||
await del_msg(context, 5)
|
||
return
|
||
else:
|
||
await context.edit("此项无法使用全局设置和单独设置")
|
||
return
|
||
elif params[0] == "trig":
|
||
if params[1] == "0":
|
||
settings_dict["trig"] = "0"
|
||
redis.set(redis_data, save_rules(settings_dict, None))
|
||
await context.edit("已关闭自我触发")
|
||
await del_msg(context, 5)
|
||
return
|
||
elif params[1] == "1":
|
||
settings_dict["trig"] = "1"
|
||
redis.set(redis_data, save_rules(settings_dict, None))
|
||
await context.edit("已开启自我触发")
|
||
await del_msg(context, 5)
|
||
return
|
||
elif params[1] == "clear":
|
||
if "trig" in settings_dict:
|
||
del settings_dict["trig"]
|
||
redis.set(redis_data, save_rules(settings_dict, None))
|
||
await context.edit("已清除自我触发设置")
|
||
await del_msg(context, 5)
|
||
return
|
||
else:
|
||
await context.edit("参数错误")
|
||
await del_msg(context, 5)
|
||
return
|
||
elif params[0] == "cache":
|
||
if params[1] == "0":
|
||
settings_dict["cache"] = "0"
|
||
redis.set(redis_data, save_rules(settings_dict, None))
|
||
await context.edit("已关闭缓存功能")
|
||
await del_msg(context, 5)
|
||
return
|
||
elif params[1] == "1":
|
||
settings_dict["cache"] = "1"
|
||
redis.set(redis_data, save_rules(settings_dict, None))
|
||
await context.edit("已开启缓存功能")
|
||
await del_msg(context, 5)
|
||
return
|
||
elif params[1] == "remove":
|
||
if redis_data == "keyword.settings":
|
||
rmtree("data/keyword_cache")
|
||
elif redis_data.split(".")[2] == "single":
|
||
rmtree(f"data/keyword_cache/{chat_id}/"
|
||
f"{redis_data.split('.')[3]}:{redis_data.split('.')[4]}")
|
||
else:
|
||
rmtree(f"data/keyword_cache/{chat_id}")
|
||
await context.edit("已删除缓存")
|
||
await del_msg(context, 5)
|
||
return
|
||
elif params[1] == "clear":
|
||
if "cache" in settings_dict:
|
||
del settings_dict["cache"]
|
||
redis.set(redis_data, save_rules(settings_dict, None))
|
||
await context.edit("清除成功")
|
||
await del_msg(context, 5)
|
||
return
|
||
else:
|
||
await context.edit("参数错误")
|
||
await del_msg(context, 5)
|
||
return
|
||
elif params[0] == "redir":
|
||
if params[1] == "0":
|
||
settings_dict["redir"] = "0"
|
||
redis.set(redis_data, save_rules(settings_dict, None))
|
||
await context.edit("已关闭回复穿透")
|
||
await del_msg(context, 5)
|
||
return
|
||
elif params[1] == "1":
|
||
settings_dict["redir"] = "1"
|
||
redis.set(redis_data, save_rules(settings_dict, None))
|
||
await context.edit("已开启回复穿透")
|
||
await del_msg(context, 5)
|
||
return
|
||
elif params[1] == "clear":
|
||
if "redir" in settings_dict:
|
||
del settings_dict["redir"]
|
||
redis.set(redis_data, save_rules(settings_dict, None))
|
||
await context.edit("已清除回复穿透设置")
|
||
await del_msg(context, 5)
|
||
return
|
||
else:
|
||
await context.edit("参数错误")
|
||
await del_msg(context, 5)
|
||
return
|
||
elif params[0] == "status":
|
||
if redis_data == f"keyword.{chat_id}.settings":
|
||
if params[1] == "0":
|
||
settings_dict["status"] = "0"
|
||
redis.set(redis_data, save_rules(settings_dict, None))
|
||
await context.edit("已关闭此聊天的关键词回复")
|
||
await del_msg(context, 5)
|
||
return
|
||
elif params[1] == "1":
|
||
settings_dict["status"] = "1"
|
||
redis.set(redis_data, save_rules(settings_dict, None))
|
||
await context.edit("已开启此聊天的关键词回复")
|
||
await del_msg(context, 5)
|
||
return
|
||
elif params[1] == "clear":
|
||
if "status" in settings_dict:
|
||
del settings_dict["status"]
|
||
redis.set(redis_data, save_rules(settings_dict, None))
|
||
await context.edit("已清除此设置")
|
||
await del_msg(context, 5)
|
||
return
|
||
else:
|
||
await context.edit("参数错误")
|
||
await del_msg(context, 5)
|
||
return
|
||
else:
|
||
await context.edit("此项无法使用全局设置和单独设置")
|
||
return
|
||
elif params[0] == "clear":
|
||
redis.delete(redis_data)
|
||
await context.edit("清除成功")
|
||
await del_msg(context, 5)
|
||
return
|
||
else:
|
||
await context.edit("参数错误")
|
||
await del_msg(context, 5)
|
||
return
|
||
|
||
|
||
@listener(outgoing=True, command=alias_command("funcset"),
|
||
description="设置自定义函数",
|
||
parameters="help")
|
||
async def funcset(context):
|
||
if not path.exists("data/keyword_func"):
|
||
makedirs("data/keyword_func")
|
||
try:
|
||
params = context.parameter
|
||
params = " ".join(params).split("\n")
|
||
cmd = []
|
||
if len(params) >= 1:
|
||
cmd = params[0].split()
|
||
if len(cmd) > 0:
|
||
if len(cmd) == 1 and cmd[0] == "ls":
|
||
send_msg = "Functions:\n"
|
||
count = 1
|
||
for p in os.listdir("data/keyword_func"):
|
||
if path.isfile(f"data/keyword_func/{p}"):
|
||
try:
|
||
send_msg += f"{count}: `{p[:-3]}`\n"
|
||
count += 1
|
||
except:
|
||
pass
|
||
await context.edit(send_msg)
|
||
return
|
||
elif len(cmd) == 2 and cmd[0] == "show":
|
||
file_path = f"data/keyword_func/{cmd[1]}.py"
|
||
if path.exists(file_path) and path.isfile(file_path):
|
||
await bot.send_file(context.chat_id, file_path)
|
||
await context.edit("发送成功")
|
||
await del_msg(context, 5)
|
||
else:
|
||
await context.edit("函数不存在")
|
||
await del_msg(context, 5)
|
||
return
|
||
elif len(cmd) == 2 and cmd[0] == "del":
|
||
file_path = f"data/keyword_func/{cmd[1]}.py"
|
||
if path.exists(file_path) and path.isfile(file_path):
|
||
remove(file_path)
|
||
await context.edit("删除成功,PagerMaid-Modify 正在重新启动。")
|
||
await bot.disconnect()
|
||
else:
|
||
await context.edit("函数不存在")
|
||
await del_msg(context, 5)
|
||
return
|
||
elif len(cmd) == 2 and cmd[0] == "new":
|
||
message = await context.get_reply_message()
|
||
if context.media:
|
||
message = context
|
||
cmd[1] = cmd[1].replace(".py", "")
|
||
if message and message.media:
|
||
try:
|
||
data = BytesIO()
|
||
await bot.download_file(message.media.document, data)
|
||
with open(f"data/keyword_func/{cmd[1]}.py", "wb") as f:
|
||
f.write(data.getvalue())
|
||
await context.edit(f"函数 {cmd[1]} 已添加,PagerMaid-Modify 正在重新启动。")
|
||
await bot.disconnect()
|
||
except:
|
||
await context.edit("函数添加失败")
|
||
await del_msg(context, 5)
|
||
else:
|
||
|
||
await context.edit("未回复消息或回复的消息中不包含文件")
|
||
await del_msg(context, 5)
|
||
return
|
||
elif len(cmd) == 2 and cmd[0] == "install":
|
||
func_name = cmd[1]
|
||
func_online = \
|
||
json.loads(
|
||
requests.get("https://raw.githubusercontent.com/xtaodada/PagerMaid_Plugins/master"
|
||
"/keyword_func/list.json").content)["list"]
|
||
if func_name in func_online:
|
||
func_directory = f"data/keyword_func/"
|
||
file_path = func_name + ".py"
|
||
func_content = requests.get(
|
||
f"https://raw.githubusercontent.com/xtaodada/PagerMaid_Plugins/master"
|
||
f"/keyword_func/{func_name}.py").content
|
||
with open(file_path, "wb") as f:
|
||
f.write(func_content)
|
||
if path.exists(f"{func_directory}{file_path}"):
|
||
remove(f"{func_directory}{file_path}")
|
||
move(file_path, func_directory)
|
||
else:
|
||
move(file_path, func_directory)
|
||
await context.edit(f"函数 {path.basename(file_path)[:-3]} 已添加,PagerMaid-Modify 正在重新启动。")
|
||
await log(f"成功安装函数 {path.basename(file_path)[:-3]}.")
|
||
await bot.disconnect()
|
||
else:
|
||
await context.edit(f"{func_name} 函数不存在")
|
||
await del_msg(context, 5)
|
||
return
|
||
elif len(cmd) == 1 and cmd[0] == "help":
|
||
await context.edit("""
|
||
`-funcset new <func_name>` (要回复带有文件的信息或自己附带文件)
|
||
`-funcset install <func_name>` (云端获取函数文件)
|
||
`-funcset del <func_name>`
|
||
`-funcset show <func_name>` (发送文件)
|
||
`-funcset ls` (列出所有函数)""")
|
||
else:
|
||
await context.edit("参数错误")
|
||
await del_msg(context, 5)
|
||
return
|
||
else:
|
||
await context.edit("参数错误")
|
||
await del_msg(context, 5)
|
||
return
|
||
except:
|
||
pass
|
||
|
||
|
||
@listener(outgoing=True, command=alias_command("keydata"),
|
||
description="设置规则数据",
|
||
parameters="dump / load")
|
||
async def setdata(context):
|
||
try:
|
||
chat_id = context.chat_id
|
||
params = context.parameter
|
||
if params[0] == "dump":
|
||
data = redis.get(f"keyword.{chat_id}.{params[1]}")
|
||
if not data:
|
||
await context.edit("无规则数据")
|
||
await del_msg(context, 5)
|
||
return
|
||
data = str(data, "ascii")
|
||
await context.edit(data)
|
||
return
|
||
elif params[0] == "load":
|
||
redis.set(f"keyword.{chat_id}.{params[1]}", params[2])
|
||
await context.edit("设置成功")
|
||
await del_msg(context, 5)
|
||
return
|
||
else:
|
||
await context.edit("参数错误")
|
||
await del_msg(context, 5)
|
||
return
|
||
except:
|
||
await context.edit("运行错误")
|
||
await del_msg(context, 5)
|
||
return
|
||
|
||
|
||
@listener(incoming=True, outgoing=True, ignore_edited=True)
|
||
async def auto_reply(context):
|
||
global read_context
|
||
await asyncio.sleep(random.randint(0, 100) / 1000)
|
||
# 解决重复回复,不稳定修复
|
||
if (context.chat_id, context.id) in read_context:
|
||
return
|
||
read_context[(context.chat_id, context.id)] = True
|
||
|
||
# 判断redis状态
|
||
if not redis_status():
|
||
return
|
||
try:
|
||
chat_id = context.chat_id
|
||
sender_id = context.sender_id
|
||
# 读取keyword当前会话的设置
|
||
n_settings = get_redis(f"keyword.{chat_id}.settings")
|
||
if n_settings.get("status", "1") == "0":
|
||
return
|
||
# 设置本人id
|
||
self_id = me_id
|
||
# 获取全局设置
|
||
g_settings = get_redis("keyword.settings")
|
||
# 获取当前会话的2种回复配置
|
||
plain_dict = get_redis(f"keyword.{chat_id}.plain")
|
||
regex_dict = get_redis(f"keyword.{chat_id}.regex")
|
||
# 获取全局和当前会话设置的mode
|
||
g_mode = g_settings.get("mode", None)
|
||
n_mode = n_settings.get("mode", None)
|
||
mode = "0"
|
||
# 获取全局和当前会话设置的list
|
||
g_list = g_settings.get("list", None)
|
||
n_list = n_settings.get("list", None)
|
||
user_list = []
|
||
# 获取全局和当前会话设置的trig
|
||
g_trig = g_settings.get("trig", None)
|
||
n_trig = n_settings.get("trig", None)
|
||
trig = "0"
|
||
# 初始化各种设置
|
||
if g_mode and n_mode:
|
||
mode = n_mode
|
||
elif g_mode or n_mode:
|
||
mode = g_mode if g_mode else n_mode
|
||
if g_list and n_list:
|
||
user_list = n_list
|
||
elif g_list or n_list:
|
||
user_list = g_list if g_list else n_list
|
||
if g_trig and n_trig:
|
||
trig = n_trig
|
||
elif g_trig or n_trig:
|
||
trig = g_trig if g_trig else n_trig
|
||
send_text = context.text
|
||
if not send_text:
|
||
send_text = ""
|
||
# 是否自我触发
|
||
self_sent = (self_id == sender_id)
|
||
# 处理文本配置
|
||
for k, v in plain_dict.items():
|
||
# 接收到的文本符合配置
|
||
if k in send_text:
|
||
# 获取要回复的内容配置
|
||
tmp = get_redis(f"keyword.{chat_id}.single.plain.{encode(k)}")
|
||
# 判断是否要回复
|
||
could_reply = validate(str(sender_id), int(mode), user_list)
|
||
if tmp:
|
||
could_reply = validate(str(sender_id), int(tmp.get("mode", "0")), tmp.get("list", []))
|
||
if could_reply and (not self_sent or validsent(int(trig), tmp)):
|
||
read_context[(chat_id, context.id)] = None
|
||
# 发送回复
|
||
await send_reply(chat_id, k, "plain", parse_multi(v), context)
|
||
# 处理正则配置
|
||
for k, v in regex_dict.items():
|
||
pattern = re.compile(k)
|
||
if pattern.search(send_text):
|
||
tmp = get_redis(f"keyword.{chat_id}.single.regex.{encode(k)}")
|
||
could_reply = validate(str(sender_id), int(mode), user_list)
|
||
if tmp:
|
||
could_reply = validate(str(sender_id), int(tmp.get("mode", "0")), tmp.get("list", []))
|
||
if could_reply and (not self_sent or validsent(int(trig), tmp)):
|
||
read_context[(chat_id, context.id)] = None
|
||
catch_pattern = r"\$\{regex_(?P<str>((?!\}).)+)\}"
|
||
count = 0
|
||
while re.search(catch_pattern, v) and count < 20:
|
||
search_data = re.search(k, send_text)
|
||
group_name = re.search(catch_pattern, v).group("str")
|
||
capture_data = get_capture(search_data, group_name)
|
||
if not capture_data:
|
||
capture_data = ""
|
||
if re.search(catch_pattern, capture_data):
|
||
capture_data = ""
|
||
v = v.replace("${regex_%s}" % group_name, capture_data)
|
||
count += 1
|
||
await send_reply(chat_id, k, "regex", parse_multi(v), context)
|
||
except:
|
||
pass
|