1
0
Fork 0
mirror of https://github.com/ytdl-org/youtube-dl.git synced 2024-12-05 08:27:43 +00:00
youtube-dl/youtube_dl/extractor/mailru.py

330 lines
11 KiB
Python
Raw Normal View History

2016-10-02 11:39:18 +00:00
# coding: utf-8
2014-03-01 09:34:38 +00:00
from __future__ import unicode_literals
import itertools
import json
2014-03-01 09:34:38 +00:00
import re
from .common import InfoExtractor
from ..compat import compat_urllib_parse_unquote
2016-02-09 15:30:48 +00:00
from ..utils import (
int_or_none,
parse_duration,
2016-02-09 15:30:48 +00:00
remove_end,
try_get,
2016-02-09 15:30:48 +00:00
)
2014-03-01 09:34:38 +00:00
class MailRuIE(InfoExtractor):
IE_NAME = 'mailru'
IE_DESC = 'Видео@Mail.Ru'
_VALID_URL = r'''(?x)
https?://
2019-12-24 21:28:34 +00:00
(?:(?:www|m)\.)?my\.mail\.ru/+
(?:
video/.*\#video=/?(?P<idv1>(?:[^/]+/){3}\d+)|
2019-12-24 21:28:34 +00:00
(?:(?P<idv2prefix>(?:[^/]+/+){2})video/(?P<idv2suffix>[^/]+/\d+))\.html|
(?:video/embed|\+/video/meta)/(?P<metaid>\d+)
)
'''
_TESTS = [
{
'url': 'http://my.mail.ru/video/top#video=/mail/sonypicturesrus/75/76',
'md5': 'dea205f03120046894db4ebb6159879a',
'info_dict': {
'id': '46301138_76',
'ext': 'mp4',
'title': 'Новый Человек-Паук. Высокое напряжение. Восстание Электро',
'timestamp': 1393235077,
'upload_date': '20140224',
'uploader': 'sonypicturesrus',
'uploader_id': 'sonypicturesrus@mail.ru',
'duration': 184,
},
2015-08-25 15:08:38 +00:00
'skip': 'Not accessible from Travis CI server',
},
{
'url': 'http://my.mail.ru/corp/hitech/video/news_hi-tech_mail_ru/1263.html',
'md5': '00a91a58c3402204dcced523777b475f',
'info_dict': {
'id': '46843144_1263',
'ext': 'mp4',
'title': 'Samsung Galaxy S5 Hammer Smash Fail Battery Explosion',
2016-02-09 15:31:51 +00:00
'timestamp': 1397039888,
'upload_date': '20140409',
'uploader': 'hitech',
'uploader_id': 'hitech@corp.mail.ru',
'duration': 245,
},
2015-08-25 15:08:38 +00:00
'skip': 'Not accessible from Travis CI server',
},
{
# only available via metaUrl API
'url': 'http://my.mail.ru/mail/720pizle/video/_myvideo/502.html',
'md5': '3b26d2491c6949d031a32b96bd97c096',
'info_dict': {
'id': '56664382_502',
'ext': 'mp4',
'title': ':8336',
'timestamp': 1449094163,
'upload_date': '20151202',
'uploader': '720pizle@mail.ru',
'uploader_id': '720pizle@mail.ru',
'duration': 6001,
},
'skip': 'Not accessible from Travis CI server',
},
{
'url': 'http://m.my.mail.ru/mail/3sktvtr/video/_myvideo/138.html',
'only_matching': True,
2017-12-04 08:26:09 +00:00
},
{
'url': 'https://my.mail.ru/video/embed/7949340477499637815',
'only_matching': True,
},
{
'url': 'http://my.mail.ru/+/video/meta/7949340477499637815',
'only_matching': True,
2019-12-24 21:28:34 +00:00
},
{
'url': 'https://my.mail.ru//list/sinyutin10/video/_myvideo/4.html',
'only_matching': True,
},
{
'url': 'https://my.mail.ru//list//sinyutin10/video/_myvideo/4.html',
'only_matching': True,
}
]
2014-03-01 09:34:38 +00:00
def _real_extract(self, url):
mobj = re.match(self._VALID_URL, url)
meta_id = mobj.group('metaid')
2017-12-04 08:26:09 +00:00
video_id = None
if meta_id:
meta_url = 'https://my.mail.ru/+/video/meta/%s' % meta_id
else:
2017-12-04 08:26:09 +00:00
video_id = mobj.group('idv1')
if not video_id:
video_id = mobj.group('idv2prefix') + mobj.group('idv2suffix')
webpage = self._download_webpage(url, video_id)
page_config = self._parse_json(self._search_regex(
r'(?s)<script[^>]+class="sp-video__page-config"[^>]*>(.+?)</script>',
webpage, 'page config', default='{}'), video_id, fatal=False)
if page_config:
meta_url = page_config.get('metaUrl') or page_config.get('video', {}).get('metaUrl')
else:
meta_url = None
video_data = None
if meta_url:
video_data = self._download_json(
meta_url, video_id or meta_id, 'Downloading video meta JSON',
fatal=not video_id)
# Fallback old approach
if not video_data:
video_data = self._download_json(
'http://api.video.mail.ru/videos/%s.json?new=1' % video_id,
video_id, 'Downloading video JSON')
2014-03-01 09:34:38 +00:00
headers = {}
video_key = self._get_cookies('https://my.mail.ru').get('video_key')
if video_key:
headers['Cookie'] = 'video_key=%s' % video_key.value
2016-02-09 15:30:48 +00:00
formats = []
for f in video_data['videos']:
video_url = f.get('url')
if not video_url:
continue
format_id = f.get('key')
height = int_or_none(self._search_regex(
r'^(\d+)[pP]$', format_id, 'height', default=None)) if format_id else None
formats.append({
'url': video_url,
'format_id': format_id,
'height': height,
'http_headers': headers,
2016-02-09 15:30:48 +00:00
})
self._sort_formats(formats)
2014-03-01 09:34:38 +00:00
meta_data = video_data['meta']
2016-02-09 15:30:48 +00:00
title = remove_end(meta_data['title'], '.mp4')
author = video_data.get('author')
uploader = author.get('name')
uploader_id = author.get('id') or author.get('email')
view_count = int_or_none(video_data.get('viewsCount') or video_data.get('views_count'))
acc_id = meta_data.get('accId')
item_id = meta_data.get('itemId')
content_id = '%s_%s' % (acc_id, item_id) if acc_id and item_id else video_id
thumbnail = meta_data.get('poster')
duration = int_or_none(meta_data.get('duration'))
timestamp = int_or_none(meta_data.get('timestamp'))
2014-03-01 09:34:38 +00:00
return {
'id': content_id,
'title': title,
'thumbnail': thumbnail,
'timestamp': timestamp,
2014-03-01 09:34:38 +00:00
'uploader': uploader,
'uploader_id': uploader_id,
'duration': duration,
'view_count': view_count,
'formats': formats,
}
class MailRuMusicSearchBaseIE(InfoExtractor):
def _search(self, query, url, audio_id, limit=100, offset=0):
search = self._download_json(
'https://my.mail.ru/cgi-bin/my/ajax', audio_id,
'Downloading songs JSON page %d' % (offset // limit + 1),
headers={
'Referer': url,
'X-Requested-With': 'XMLHttpRequest',
}, query={
'xemail': '',
'ajax_call': '1',
'func_name': 'music.search',
'mna': '',
'mnb': '',
'arg_query': query,
'arg_extended': '1',
'arg_search_params': json.dumps({
'music': {
'limit': limit,
'offset': offset,
},
}),
'arg_limit': limit,
'arg_offset': offset,
})
return next(e for e in search if isinstance(e, dict))
@staticmethod
def _extract_track(t, fatal=True):
audio_url = t['URL'] if fatal else t.get('URL')
if not audio_url:
return
audio_id = t['File'] if fatal else t.get('File')
if not audio_id:
return
thumbnail = t.get('AlbumCoverURL') or t.get('FiledAlbumCover')
uploader = t.get('OwnerName') or t.get('OwnerName_Text_HTML')
uploader_id = t.get('UploaderID')
duration = int_or_none(t.get('DurationInSeconds')) or parse_duration(
t.get('Duration') or t.get('DurationStr'))
view_count = int_or_none(t.get('PlayCount') or t.get('PlayCount_hr'))
track = t.get('Name') or t.get('Name_Text_HTML')
artist = t.get('Author') or t.get('Author_Text_HTML')
if track:
title = '%s - %s' % (artist, track) if artist else track
else:
title = audio_id
return {
'extractor_key': MailRuMusicIE.ie_key(),
'id': audio_id,
'title': title,
'thumbnail': thumbnail,
'uploader': uploader,
'uploader_id': uploader_id,
'duration': duration,
'view_count': view_count,
'vcodec': 'none',
'abr': int_or_none(t.get('BitRate')),
'track': track,
'artist': artist,
'album': t.get('Album'),
'url': audio_url,
}
class MailRuMusicIE(MailRuMusicSearchBaseIE):
IE_NAME = 'mailru:music'
IE_DESC = 'Музыка@Mail.Ru'
2019-12-24 21:28:34 +00:00
_VALID_URL = r'https?://my\.mail\.ru/+music/+songs/+[^/?#&]+-(?P<id>[\da-f]+)'
_TESTS = [{
'url': 'https://my.mail.ru/music/songs/%D0%BC8%D0%BB8%D1%82%D1%85-l-a-h-luciferian-aesthetics-of-herrschaft-single-2017-4e31f7125d0dfaef505d947642366893',
'md5': '0f8c22ef8c5d665b13ac709e63025610',
'info_dict': {
'id': '4e31f7125d0dfaef505d947642366893',
'ext': 'mp3',
'title': 'L.A.H. (Luciferian Aesthetics of Herrschaft) single, 2017 - М8Л8ТХ',
'uploader': 'Игорь Мудрый',
'uploader_id': '1459196328',
'duration': 280,
'view_count': int,
'vcodec': 'none',
'abr': 320,
'track': 'L.A.H. (Luciferian Aesthetics of Herrschaft) single, 2017',
'artist': 'М8Л8ТХ',
},
}]
def _real_extract(self, url):
audio_id = self._match_id(url)
webpage = self._download_webpage(url, audio_id)
title = self._og_search_title(webpage)
music_data = self._search(title, url, audio_id)['MusicData']
t = next(t for t in music_data if t.get('File') == audio_id)
info = self._extract_track(t)
info['title'] = title
return info
class MailRuMusicSearchIE(MailRuMusicSearchBaseIE):
IE_NAME = 'mailru:music:search'
IE_DESC = 'Музыка@Mail.Ru'
2019-12-24 21:28:34 +00:00
_VALID_URL = r'https?://my\.mail\.ru/+music/+search/+(?P<id>[^/?#&]+)'
_TESTS = [{
'url': 'https://my.mail.ru/music/search/black%20shadow',
'info_dict': {
'id': 'black shadow',
},
'playlist_mincount': 532,
}]
def _real_extract(self, url):
query = compat_urllib_parse_unquote(self._match_id(url))
entries = []
LIMIT = 100
offset = 0
for _ in itertools.count(1):
search = self._search(query, url, query, LIMIT, offset)
music_data = search.get('MusicData')
if not music_data or not isinstance(music_data, list):
break
for t in music_data:
track = self._extract_track(t, fatal=False)
if track:
entries.append(track)
total = try_get(
search, lambda x: x['Results']['music']['Total'], int)
if total is not None:
if offset > total:
break
offset += LIMIT
return self.playlist_result(entries, query)