mirror of
https://github.com/TeamPGM/PagerMaid_Plugins.git
synced 2024-11-25 20:38:51 +00:00
neteasemusic大型更新(指令变动为nem)
This commit is contained in:
parent
c7ad074509
commit
64739829bd
@ -282,13 +282,13 @@
|
|||||||
},
|
},
|
||||||
{
|
{
|
||||||
"name": "neteasemusic",
|
"name": "neteasemusic",
|
||||||
"version": "1.0",
|
"version": "1.1",
|
||||||
"section": "daily",
|
"section": "daily",
|
||||||
"maintainer": "TNTcraftHIM",
|
"maintainer": "TNTcraftHIM",
|
||||||
"size": "6.6 kb",
|
"size": "9.9 kb",
|
||||||
"supported": true,
|
"supported": true,
|
||||||
"des-short": "网易云点歌",
|
"des-short": "网易云点歌",
|
||||||
"des": "可以在线搜索并且点歌,并且歌曲信息齐全,支持ID点歌。命令:music。"
|
"des": "可以在线搜索并且快速点歌,歌曲信息齐全,支持直接回复点歌。命令:nem。"
|
||||||
}
|
}
|
||||||
]
|
]
|
||||||
}
|
}
|
||||||
|
159
neteasemusic.py
159
neteasemusic.py
@ -3,53 +3,65 @@ import requests
|
|||||||
import re
|
import re
|
||||||
from time import sleep
|
from time import sleep
|
||||||
from pagermaid.listener import listener
|
from pagermaid.listener import listener
|
||||||
|
from pagermaid import bot
|
||||||
|
from pagermaid.utils import obtain_message
|
||||||
from os import remove, path
|
from os import remove, path
|
||||||
from collections import defaultdict
|
from collections import defaultdict
|
||||||
|
|
||||||
|
|
||||||
@listener(is_plugin=True, outgoing=True, command="music",
|
@listener(is_plugin=True, outgoing=True, command="nem",
|
||||||
description="网抑云搜歌。",
|
description="网易云搜/点歌。",
|
||||||
parameters="<指令> <关键词>")
|
parameters="<指令> <关键词>")
|
||||||
async def music(context):
|
async def nem(context):
|
||||||
|
headers = {'User-Agent': 'Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/52.0.2743.116 Safari/537.36 Edge/15.15063',
|
||||||
|
"Accept": "text/html,application/xhtml+xml,application/xml;q=0.9,image/webp,image/apng,*/*;q=0.8,application/signed-exchange;v=b3;q=0.9"}
|
||||||
if len(context.parameter) < 2:
|
if len(context.parameter) < 2:
|
||||||
await context.edit("使用方法:-music <指令> <关键词>\n(指令s为搜索,指令p为播放\n播放时关键词可填歌曲ID,此时为精准播放)")
|
await context.edit("使用方法:`-nem` `<指令>` `<关键词>`\n(指令s为搜索,指令p为播放\n关键词可填歌曲ID,或直接回复搜索结果 `-nem` `p` `<歌曲数字序号>`)")
|
||||||
return
|
return
|
||||||
else:
|
else:
|
||||||
keyword = ''
|
keyword = ''
|
||||||
for i in range(1, len(context.parameter)):
|
for i in range(1, len(context.parameter)):
|
||||||
keyword += context.parameter[i] + " "
|
keyword += context.parameter[i] + " "
|
||||||
if context.parameter[0] == "s": #搜索功能
|
keyword = keyword[:-1]
|
||||||
|
if context.parameter[0] == "s": # 搜索功能
|
||||||
await context.edit("搜索中 . . .")
|
await context.edit("搜索中 . . .")
|
||||||
headers = {'User-Agent': 'Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/52.0.2743.116 Safari/537.36 Edge/15.15063',
|
url = "http://music.163.com/api/search/pc?&s=" + \
|
||||||
"Accept": "text/html,application/xhtml+xml,application/xml;q=0.9,image/webp,image/apng,*/*;q=0.8,application/signed-exchange;v=b3;q=0.9"}
|
keyword + "&offset=0&limit=5&type=1"
|
||||||
url = "http://music.163.com/api/search/pc?&s=" + keyword + "&offset=0&limit=5&type=1"
|
for _ in range(20): # 最多尝试20次
|
||||||
for _ in range(20): #最多尝试20次
|
status = False
|
||||||
status=False
|
|
||||||
req = requests.request("GET", url, headers=headers)
|
req = requests.request("GET", url, headers=headers)
|
||||||
if req.status_code == 200:
|
if req.status_code == 200:
|
||||||
req = json.loads(req.content)
|
req = json.loads(req.content)
|
||||||
if req['result']:
|
if req['result']:
|
||||||
info=defaultdict()
|
info = defaultdict()
|
||||||
for i in range(len(req['result']['songs'])):
|
for i in range(len(req['result']['songs'])):
|
||||||
info[i]={'id':'', 'title':'', 'alias':'', 'album':'', 'artist':''}
|
info[i] = {'id': '', 'title': '', 'alias': '',
|
||||||
info[i]['id']=req['result']['songs'][i]['id']
|
'album': '', 'albumpic': '', 'artist': ''}
|
||||||
info[i]['title']=req['result']['songs'][i]['name']
|
info[i]['id'] = req['result']['songs'][i]['id']
|
||||||
info[i]['alias']=req['result']['songs'][i]['alias']
|
info[i]['title'] = req['result']['songs'][i]['name']
|
||||||
info[i]['album']=req['result']['songs'][i]['album']['name']
|
info[i]['alias'] = req['result']['songs'][i]['alias']
|
||||||
|
info[i]['album'] = req['result']['songs'][i]['album']['name']
|
||||||
|
info[i]['albumpic'] = req['result']['songs'][i]['album']['picUrl']
|
||||||
for j in range(len(req['result']['songs'][i]['artists'])):
|
for j in range(len(req['result']['songs'][i]['artists'])):
|
||||||
info[i]['artist']+=req['result']['songs'][i]['artists'][j]['name'] + " "
|
info[i]['artist'] += req['result']['songs'][i]['artists'][j]['name'] + " "
|
||||||
text="网易云搜索结果如下 \n"
|
text = f"<strong>关于【{keyword}】的结果如下</strong> \n"
|
||||||
for i in range(len(info)):
|
for i in range(len(info)):
|
||||||
text+=f"{i+1}: \n歌名: {info[i]['title']} \n"
|
text += f"#{i+1}: \n<strong>歌名</strong>: {info[i]['title']}\n"
|
||||||
if info[i]['alias']:
|
if info[i]['alias']:
|
||||||
text+=f"别名: {info[i]['alias'][0]} \n"
|
text += f"<strong>别名</strong>: {info[i]['alias'][0]} \n"
|
||||||
text+=f"专辑: {info[i]['album']} \n作者: {info[i]['artist']}\n歌曲ID: {info[i]['id']} \n\n"
|
if info[i]['album']:
|
||||||
await context.edit(text)
|
res = '<a href="' + \
|
||||||
status=True
|
info[i]['albumpic'] + '">' + \
|
||||||
|
info[i]['album'] + '</a>'
|
||||||
|
text += f"<strong>专辑</strong>: {res} \n"
|
||||||
|
text += f"<strong>作者</strong>: {info[i]['artist']}\n<strong>歌曲ID</strong>: <code>{info[i]['id']}</code>\n————————\n"
|
||||||
|
text += "回复此消息 <code>-nem p <歌曲序号></code> 即可点歌"
|
||||||
|
await context.edit(text, parse_mode='html', link_preview=True)
|
||||||
|
status = True
|
||||||
break
|
break
|
||||||
else:
|
else:
|
||||||
await context.edit("未搜索到结果")
|
await context.edit("**未搜索到结果**")
|
||||||
status=True
|
status = True
|
||||||
break
|
break
|
||||||
else:
|
else:
|
||||||
continue
|
continue
|
||||||
@ -59,55 +71,105 @@ async def music(context):
|
|||||||
await context.delete()
|
await context.delete()
|
||||||
return
|
return
|
||||||
|
|
||||||
if context.parameter[0] == "p": #播放功能
|
if context.parameter[0] == "p": # 播放功能
|
||||||
|
try:
|
||||||
|
reply = await context.get_reply_message()
|
||||||
|
except ValueError:
|
||||||
|
await context.edit("出错了呜呜呜 ~ 无效的参数。")
|
||||||
|
return
|
||||||
|
search = ""
|
||||||
|
title = ""
|
||||||
|
if reply:
|
||||||
|
msg = reply.message
|
||||||
|
search = re.findall(".*【(.*)】.*", msg)
|
||||||
|
if search:
|
||||||
|
try:
|
||||||
|
if int(context.parameter[1]) > 5:
|
||||||
|
await context.edit("出错了呜呜呜 ~ 无效的歌曲序号。")
|
||||||
|
return
|
||||||
|
else:
|
||||||
|
start = "#" + context.parameter[1] + ":"
|
||||||
|
search = ".*" + start + "(.*?)" + '————————' + ".*"
|
||||||
|
msg = re.findall(search, msg, re.S)[0]
|
||||||
|
search = ".*歌曲ID: (.*)\n.*"
|
||||||
|
title = ".*歌名: (.*?)\n.*"
|
||||||
|
title = "【"+re.findall(title, msg, re.S)[0]+"】"
|
||||||
|
keyword = re.findall(search, msg, re.S)[0]
|
||||||
|
if reply.sender.is_self:
|
||||||
|
await reply.edit(f"{title}点歌完成")
|
||||||
|
except:
|
||||||
|
await context.edit("出错了呜呜呜 ~ 无效的参数。")
|
||||||
|
return
|
||||||
|
else:
|
||||||
|
await context.edit("出错了呜呜呜 ~ 无效的参数。")
|
||||||
|
return
|
||||||
|
|
||||||
await context.edit("获取中 . . .")
|
await context.edit("获取中 . . .")
|
||||||
try:
|
try:
|
||||||
import eyed3
|
import eyed3
|
||||||
imported = True
|
imported = True
|
||||||
except ImportError:
|
except ImportError:
|
||||||
imported = False
|
imported = False
|
||||||
await context.edit("获取中 . . .\n(eyeD3支持库未安装,歌曲文件信息将无法导入\n请使用-sh pip3 install eyed3安装,或自行ssh安装)")
|
await bot.send_message(context.chat_id, '(`eyeD3`支持库未安装,歌曲文件信息将无法导入\n请使用 `-sh` `pip3` `install` `eyed3` 安装,或自行ssh安装)')
|
||||||
headers = {'User-Agent': 'Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/52.0.2743.116 Safari/537.36 Edge/15.15063',
|
url = "http://music.163.com/api/search/pc?&s=" + \
|
||||||
"Accept": "text/html,application/xhtml+xml,application/xml;q=0.9,image/webp,image/apng,*/*;q=0.8,application/signed-exchange;v=b3;q=0.9"}
|
keyword + "&offset=0&limit=1&type=1"
|
||||||
url = "http://music.163.com/api/search/pc?&s=" + keyword + "&offset=0&limit=1&type=1"
|
for _ in range(20): # 最多尝试20次
|
||||||
for _ in range(20): #最多尝试20次
|
status = False
|
||||||
status=False
|
|
||||||
req = requests.request("GET", url, headers=headers)
|
req = requests.request("GET", url, headers=headers)
|
||||||
if req.status_code == 200:
|
if req.status_code == 200:
|
||||||
req = json.loads(req.content)
|
req = json.loads(req.content)
|
||||||
if req['result']:
|
if req['result']:
|
||||||
info={'id':'', 'title':'', 'alias':'', 'album':'', 'albumpic':'', 'artist':''}
|
info = {'id': '', 'title': '', 'alias': '',
|
||||||
info['id']=req['result']['songs'][0]['id']
|
'album': '', 'albumpic': '', 'artist': ''}
|
||||||
info['title']=req['result']['songs'][0]['name']
|
info['id'] = req['result']['songs'][0]['id']
|
||||||
info['alias']=req['result']['songs'][0]['alias']
|
info['title'] = req['result']['songs'][0]['name']
|
||||||
info['album']=req['result']['songs'][0]['album']['name']
|
info['alias'] = req['result']['songs'][0]['alias']
|
||||||
info['albumpic']=req['result']['songs'][0]['album']['picUrl']
|
info['album'] = req['result']['songs'][0]['album']['name']
|
||||||
|
info['albumpic'] = req['result']['songs'][0]['album']['picUrl']
|
||||||
for j in range(len(req['result']['songs'][0]['artists'])):
|
for j in range(len(req['result']['songs'][0]['artists'])):
|
||||||
info['artist']+=req['result']['songs'][0]['artists'][j]['name'] + ";"
|
info['artist'] += req['result']['songs'][0]['artists'][j]['name'] + ";"
|
||||||
info['artist']=info['artist'][:-1]
|
info['artist'] = info['artist'][:-1]
|
||||||
music = requests.request("GET", "http://music.163.com/song/media/outer/url?id=" + str(info['id']) +".mp3", headers=headers)
|
if title:
|
||||||
name = info['title'] + ".mp3"
|
title = ""
|
||||||
cap = info['artist'].replace(';',', ') + " - " + info['title']
|
else:
|
||||||
|
title = f"【{info['title']}】"
|
||||||
|
await context.edit(f"{title}下载中 . . .")
|
||||||
|
# 下载
|
||||||
|
music = requests.request(
|
||||||
|
"GET", "http://music.163.com/song/media/outer/url?id=" + str(info['id']) + ".mp3", headers=headers)
|
||||||
|
name = info['title'].replace('/', " ") + ".mp3"
|
||||||
|
cap = info['artist'].replace(
|
||||||
|
';', ', ') + " - " + "**" + info['title'] + "**"
|
||||||
pic = requests.get(info['albumpic'])
|
pic = requests.get(info['albumpic'])
|
||||||
with open(name, 'wb') as f:
|
with open(name, 'wb') as f:
|
||||||
f.write(music.content)
|
f.write(music.content)
|
||||||
if (path.getsize(name) / 1024) < 100:
|
if (path.getsize(name) / 1024) < 100:
|
||||||
remove(name)
|
remove(name)
|
||||||
continue
|
if reply.sender.is_self:
|
||||||
|
await reply.delete()
|
||||||
|
await context.delete()
|
||||||
|
res = '你可以点击<a href="https://music.163.com/#/song?id=' + \
|
||||||
|
str(info['id']) + '">' + \
|
||||||
|
' <strong>这里</strong> ' + '</a>' + '前往网页版收听'
|
||||||
|
await bot.send_message(context.chat_id, f"<strong>【{info['title']}】</strong>\n" + "歌曲获取失败,可能歌曲为VIP专属,或受到地区版权限制。\n" + res, parse_mode='html', link_preview=True)
|
||||||
|
return
|
||||||
if imported is True:
|
if imported is True:
|
||||||
tag = eyed3.load(name).tag
|
tag = eyed3.load(name).tag
|
||||||
tag.encoding = '\x01'
|
|
||||||
tag.artist = info['artist']
|
tag.artist = info['artist']
|
||||||
tag.title = info['title']
|
tag.title = info['title']
|
||||||
tag.album = info['album']
|
tag.album = info['album']
|
||||||
tag.images.set(3, pic.content, "image/jpeg", u'')
|
tag.images.set(3, pic.content, "image/jpeg", u'')
|
||||||
tag.save()
|
tag.save(
|
||||||
|
version=eyed3.id3.ID3_DEFAULT_VERSION, encoding='utf-8')
|
||||||
|
await context.edit(f"{title}上传中 . . .")
|
||||||
await context.client.send_file(
|
await context.client.send_file(
|
||||||
context.chat_id,
|
context.chat_id,
|
||||||
name,
|
name,
|
||||||
caption=cap,
|
caption=cap,
|
||||||
link_preview=False,
|
link_preview=False,
|
||||||
force_document=False)
|
force_document=False)
|
||||||
|
if reply.sender.is_self:
|
||||||
|
await reply.delete()
|
||||||
try:
|
try:
|
||||||
remove(name)
|
remove(name)
|
||||||
except:
|
except:
|
||||||
@ -116,11 +178,12 @@ async def music(context):
|
|||||||
status = True
|
status = True
|
||||||
break
|
break
|
||||||
else:
|
else:
|
||||||
await context.edit("未搜索到结果")
|
await context.edit("**未搜索到结果**")
|
||||||
status=True
|
status = True
|
||||||
break
|
break
|
||||||
else:
|
else:
|
||||||
continue
|
continue
|
||||||
|
|
||||||
if status is False:
|
if status is False:
|
||||||
await context.edit("出错了呜呜呜 ~ 试了好多好多次都无法访问到 API 服务器 。")
|
await context.edit("出错了呜呜呜 ~ 试了好多好多次都无法访问到 API 服务器 。")
|
||||||
sleep(2)
|
sleep(2)
|
||||||
|
Loading…
Reference in New Issue
Block a user