|
123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300 |
- # -*- coding: utf-8 -*-
- import web
- import time
- import math
- import hashlib
- import textwrap
- from channel.channel import Channel
- import channel.wechatmp.reply as reply
- import channel.wechatmp.receive as receive
- from common.log import logger
- from config import conf
- from bridge.reply import *
- from bridge.context import *
- from plugins import *
- import traceback
- import redis
-
- class WechatMPServer():
- def __init__(self):
- pass
-
- def startup(self):
- urls = (
- '/wx', 'WechatMPChannel',
- )
- app = web.application(urls, globals())
- app.run()
-
- cache_dict = dict()
- query1 = dict()
- query2 = dict()
- query3 = dict()
-
- from concurrent.futures import ThreadPoolExecutor
- thread_pool = ThreadPoolExecutor(max_workers=8)
-
- class WechatMPChannel(Channel):
-
- def GET(self):
- try:
- data = web.input()
- if len(data) == 0:
- return "hello, this is handle view"
- signature = data.signature
- timestamp = data.timestamp
- nonce = data.nonce
- echostr = data.echostr
- token = conf().get('wechatmp_token') #请按照公众平台官网\基本配置中信息填写
-
- data_list = [token, timestamp, nonce]
- data_list.sort()
- sha1 = hashlib.sha1()
- # map(sha1.update, data_list) #python2
- sha1.update("".join(data_list).encode('utf-8'))
- hashcode = sha1.hexdigest()
- print("handle/GET func: hashcode, signature: ", hashcode, signature)
- if hashcode == signature:
- return echostr
- else:
- return ""
- except Exception as Argument:
- return Argument
-
-
- def _do_build_reply(self, cache_key, fromUser, message):
- context = dict()
- context['session_id'] = fromUser
- reply_text = super().build_reply_content(message, context)
- # The query is done, record the cache
- logger.info("[threaded] Get reply for {}: {} \nA: {}".format(fromUser, message, reply_text))
- global cache_dict
- reply_cnt = math.ceil(len(reply_text) / 600)
- cache_dict[cache_key] = (reply_cnt, reply_text)
-
-
- def send(self, reply : Reply, cache_key):
- global cache_dict
- reply_cnt = math.ceil(len(reply.content) / 600)
- cache_dict[cache_key] = (reply_cnt, reply.content)
-
-
- def handle(self, context):
- global cache_dict
- try:
- reply = Reply()
- logger.debug('[wechatmp] ready to handle context: {}'.format(context))
-
- # reply的构建步骤
- e_context = PluginManager().emit_event(EventContext(Event.ON_HANDLE_CONTEXT, {'channel' : self, 'context': context, 'reply': reply}))
- reply = e_context['reply']
- if not e_context.is_pass():
- logger.debug('[wechatmp] ready to handle context: type={}, content={}'.format(context.type, context.content))
- if context.type == ContextType.TEXT or context.type == ContextType.IMAGE_CREATE:
- reply = super().build_reply_content(context.content, context)
- # elif context.type == ContextType.VOICE:
- # msg = context['msg']
- # file_name = TmpDir().path() + context.content
- # msg.download(file_name)
- # reply = super().build_voice_to_text(file_name)
- # if reply.type != ReplyType.ERROR and reply.type != ReplyType.INFO:
- # context.content = reply.content # 语音转文字后,将文字内容作为新的context
- # context.type = ContextType.TEXT
- # reply = super().build_reply_content(context.content, context)
- # if reply.type == ReplyType.TEXT:
- # if conf().get('voice_reply_voice'):
- # reply = super().build_text_to_voice(reply.content)
- else:
- logger.error('[wechatmp] unknown context type: {}'.format(context.type))
- return
-
- logger.debug('[wechatmp] ready to decorate reply: {}'.format(reply))
-
- # reply的包装步骤
- if reply and reply.type:
- e_context = PluginManager().emit_event(EventContext(Event.ON_DECORATE_REPLY, {'channel' : self, 'context': context, 'reply': reply}))
- reply=e_context['reply']
- if not e_context.is_pass() and reply and reply.type:
- if reply.type == ReplyType.TEXT:
- pass
- elif reply.type == ReplyType.ERROR or reply.type == ReplyType.INFO:
- reply.content = str(reply.type)+":\n" + reply.content
- elif reply.type == ReplyType.IMAGE_URL or reply.type == ReplyType.VOICE or reply.type == ReplyType.IMAGE:
- pass
- else:
- logger.error('[wechatmp] unknown reply type: {}'.format(reply.type))
- return
-
- # reply的发送步骤
- if reply and reply.type:
- e_context = PluginManager().emit_event(EventContext(Event.ON_SEND_REPLY, {'channel' : self, 'context': context, 'reply': reply}))
- reply=e_context['reply']
- if not e_context.is_pass() and reply and reply.type:
- logger.debug('[wechatmp] ready to send reply: {} to {}'.format(reply, context['receiver']))
- self.send(reply, context['receiver'])
- else:
- cache_dict[context['receiver']] = (1, "No reply")
-
- logger.info("[threaded] Get reply for {}: {} \nA: {}".format(context['receiver'], context.content, reply.content))
- except Exception as exc:
- print(traceback.format_exc())
- cache_dict[context['receiver']] = (1, "ERROR")
-
-
-
- def POST(self):
- try:
- queryTime = time.time()
- webData = web.data()
- # logger.debug("[wechatmp] Receive request:\n" + webData.decode("utf-8"))
- recMsg = receive.parse_xml(webData)
- if isinstance(recMsg, receive.Msg) and recMsg.MsgType == 'text':
- fromUser = recMsg.FromUserName
- toUser = recMsg.ToUserName
- createTime = recMsg.CreateTime
- message = recMsg.Content.decode("utf-8")
- message_id = recMsg.MsgId
-
- logger.info("[wechatmp] {}:{} Receive post query {} {}: {}".format(web.ctx.env.get('REMOTE_ADDR'), web.ctx.env.get('REMOTE_PORT'), fromUser, message_id, message))
-
- global cache_dict
- global query1
- global query2
- global query3
- cache_key = fromUser
- cache = cache_dict.get(cache_key)
-
- reply_text = ""
- # New request
- if cache == None:
- # The first query begin, reset the cache
- cache_dict[cache_key] = (0, "")
- # thread_pool.submit(self._do_build_reply, cache_key, fromUser, message)
-
- context = Context()
- context.kwargs = {'isgroup': False, 'receiver': fromUser, 'session_id': fromUser}
-
- R = redis.Redis(host='localhost', port=6379, db=0)
- user_openai_api_key = "openai_api_key_" + fromUser
- api_key = R.get(user_openai_api_key)
- if api_key != None:
- api_key = api_key.decode("utf-8")
- context['openai_api_key'] = api_key # None or user openai_api_key
-
- img_match_prefix = check_prefix(message, conf().get('image_create_prefix'))
- if img_match_prefix:
- message = message.replace(img_match_prefix, '', 1).strip()
- context.type = ContextType.IMAGE_CREATE
- else:
- context.type = ContextType.TEXT
- context.content = message
- thread_pool.submit(self.handle, context)
-
- query1[cache_key] = False
- query2[cache_key] = False
- query3[cache_key] = False
- # Request again
- elif cache[0] == 0 and query1.get(cache_key) == True and query2.get(cache_key) == True and query3.get(cache_key) == True:
- query1[cache_key] = False #To improve waiting experience, this can be set to True.
- query2[cache_key] = False #To improve waiting experience, this can be set to True.
- query3[cache_key] = False
- elif cache[0] >= 1:
- # Skip the waiting phase
- query1[cache_key] = True
- query2[cache_key] = True
- query3[cache_key] = True
-
-
- cache = cache_dict.get(cache_key)
- if query1.get(cache_key) == False:
- # The first query from wechat official server
- logger.debug("[wechatmp] query1 {}".format(cache_key))
- query1[cache_key] = True
- cnt = 0
- while cache[0] == 0 and cnt < 45:
- cnt = cnt + 1
- time.sleep(0.1)
- cache = cache_dict.get(cache_key)
- if cnt == 45:
- # waiting for timeout (the POST query will be closed by wechat official server)
- time.sleep(5)
- # and do nothing
- return
- else:
- pass
- elif query2.get(cache_key) == False:
- # The second query from wechat official server
- logger.debug("[wechatmp] query2 {}".format(cache_key))
- query2[cache_key] = True
- cnt = 0
- while cache[0] == 0 and cnt < 45:
- cnt = cnt + 1
- time.sleep(0.1)
- cache = cache_dict.get(cache_key)
- if cnt == 45:
- # waiting for timeout (the POST query will be closed by wechat official server)
- time.sleep(5)
- # and do nothing
- return
- else:
- pass
- elif query3.get(cache_key) == False:
- # The third query from wechat official server
- logger.debug("[wechatmp] query3 {}".format(cache_key))
- query3[cache_key] = True
- cnt = 0
- while cache[0] == 0 and cnt < 45:
- cnt = cnt + 1
- time.sleep(0.1)
- cache = cache_dict.get(cache_key)
- if cnt == 45:
- # Have waiting for 3x5 seconds
- # return timeout message
- reply_text = "【正在响应中,回复任意文字尝试获取回复】"
- logger.info("[wechatmp] Three queries has finished For {}: {}".format(fromUser, message_id))
- replyPost = reply.TextMsg(fromUser, toUser, reply_text).send()
- return replyPost
- else:
- pass
-
- if float(time.time()) - float(queryTime) > 4.8:
- logger.info("[wechatmp] Timeout for {} {}".format(fromUser, message_id))
- return
-
-
- if cache[0] > 1:
- reply_text = cache[1][:600] + " 【未完待续,回复任意文字以继续】" #wechatmp auto_reply length limit
- cache_dict[cache_key] = (cache[0] - 1, cache[1][600:])
- elif cache[0] == 1:
- reply_text = cache[1]
- cache_dict.pop(cache_key)
- logger.info("[wechatmp] {}:{} Do send {}".format(web.ctx.env.get('REMOTE_ADDR'), web.ctx.env.get('REMOTE_PORT'), reply_text))
- replyPost = reply.TextMsg(fromUser, toUser, reply_text).send()
- return replyPost
-
- elif isinstance(recMsg, receive.Event) and recMsg.MsgType == 'event':
- logger.info("[wechatmp] Event {} from {}".format(recMsg.Event, recMsg.FromUserName))
- content = textwrap.dedent("""\
- 感谢您的关注!
- 这里是ChatGPT,可以自由对话。
- 资源有限,回复较慢,请勿着急。
- 支持通用表情输入。
- 暂时不支持图片输入。
- 支持图片输出,画字开头的问题将回复图片链接。
- 支持角色扮演和文字冒险两种定制模式对话。
- 输入'#帮助' 查看详细指令。""")
- replyMsg = reply.TextMsg(recMsg.FromUserName, recMsg.ToUserName, content)
- return replyMsg.send()
- else:
- print("暂且不处理")
- return "success"
- except Exception as exc:
- print(exc)
- return exc
-
-
- def check_prefix(content, prefix_list):
- for prefix in prefix_list:
- if content.startswith(prefix):
- return prefix
- return None
|