mirror of
https://github.com/yt-dlp/yt-dlp.git
synced 2024-11-30 07:28:19 +01:00
Closes #3736 Authored by: lksj, puc9, pukkandan Co-authored-by: puc9 <51006296+puc9@users.noreply.github.com>
This commit is contained in:
parent
c66ed4e2e5
commit
c9bd65185c
|
@ -122,7 +122,8 @@ def print_skipping(reason):
|
||||||
params['outtmpl'] = tname + '_' + params['outtmpl']
|
params['outtmpl'] = tname + '_' + params['outtmpl']
|
||||||
if is_playlist and 'playlist' not in test_case:
|
if is_playlist and 'playlist' not in test_case:
|
||||||
params.setdefault('extract_flat', 'in_playlist')
|
params.setdefault('extract_flat', 'in_playlist')
|
||||||
params.setdefault('playlistend', test_case.get('playlist_mincount'))
|
params.setdefault('playlistend', test_case.get(
|
||||||
|
'playlist_mincount', test_case.get('playlist_count', -2) + 1))
|
||||||
params.setdefault('skip_download', True)
|
params.setdefault('skip_download', True)
|
||||||
|
|
||||||
ydl = YoutubeDL(params, auto_init=False)
|
ydl = YoutubeDL(params, auto_init=False)
|
||||||
|
|
|
@ -1,5 +1,4 @@
|
||||||
import itertools
|
import itertools
|
||||||
import re
|
|
||||||
|
|
||||||
from .common import InfoExtractor
|
from .common import InfoExtractor
|
||||||
from ..utils import (
|
from ..utils import (
|
||||||
|
@ -7,6 +6,8 @@
|
||||||
extract_attributes,
|
extract_attributes,
|
||||||
int_or_none,
|
int_or_none,
|
||||||
lowercase_escape,
|
lowercase_escape,
|
||||||
|
parse_qs,
|
||||||
|
traverse_obj,
|
||||||
try_get,
|
try_get,
|
||||||
url_or_none,
|
url_or_none,
|
||||||
)
|
)
|
||||||
|
@ -23,7 +24,6 @@ class YandexVideoIE(InfoExtractor):
|
||||||
'''
|
'''
|
||||||
_TESTS = [{
|
_TESTS = [{
|
||||||
'url': 'https://yandex.ru/portal/video?stream_id=4dbb36ec4e0526d58f9f2dc8f0ecf374',
|
'url': 'https://yandex.ru/portal/video?stream_id=4dbb36ec4e0526d58f9f2dc8f0ecf374',
|
||||||
'md5': 'e02a05bfaf0d9615ef07ae3a10f4faf4',
|
|
||||||
'info_dict': {
|
'info_dict': {
|
||||||
'id': '4dbb36ec4e0526d58f9f2dc8f0ecf374',
|
'id': '4dbb36ec4e0526d58f9f2dc8f0ecf374',
|
||||||
'ext': 'mp4',
|
'ext': 'mp4',
|
||||||
|
@ -38,6 +38,7 @@ class YandexVideoIE(InfoExtractor):
|
||||||
'like_count': int,
|
'like_count': int,
|
||||||
'dislike_count': int,
|
'dislike_count': int,
|
||||||
},
|
},
|
||||||
|
'params': {'skip_download': 'm3u8'},
|
||||||
}, {
|
}, {
|
||||||
'url': 'https://yandex.ru/portal/efir?stream_id=4dbb262b4fe5cf15a215de4f34eee34d&from=morda',
|
'url': 'https://yandex.ru/portal/efir?stream_id=4dbb262b4fe5cf15a215de4f34eee34d&from=morda',
|
||||||
'only_matching': True,
|
'only_matching': True,
|
||||||
|
@ -188,34 +189,35 @@ def _real_extract(self, url):
|
||||||
|
|
||||||
|
|
||||||
class ZenYandexIE(InfoExtractor):
|
class ZenYandexIE(InfoExtractor):
|
||||||
_VALID_URL = r'https?://zen\.yandex\.ru(?:/video)?/(media|watch)/(?:(?:id/[^/]+/|[^/]+/)(?:[a-z0-9-]+)-)?(?P<id>[a-z0-9-]+)'
|
_VALID_URL = r'https?://(zen\.yandex|dzen)\.ru(?:/video)?/(media|watch)/(?:(?:id/[^/]+/|[^/]+/)(?:[a-z0-9-]+)-)?(?P<id>[a-z0-9-]+)'
|
||||||
_TESTS = [{
|
_TESTS = [{
|
||||||
'url': 'https://zen.yandex.ru/media/popmech/izverjenie-vulkana-iz-spichek-zreliscnyi-opyt-6002240ff8b1af50bb2da5e3',
|
|
||||||
'info_dict': {
|
|
||||||
'id': '6002240ff8b1af50bb2da5e3',
|
|
||||||
'ext': 'mp4',
|
|
||||||
'title': 'Извержение вулкана из спичек: зрелищный опыт',
|
|
||||||
'description': 'md5:053ad3c61b5596d510c9a199dc8ee633',
|
|
||||||
'thumbnail': 're:^https://avatars.mds.yandex.net/',
|
|
||||||
'uploader': 'Популярная механика',
|
|
||||||
},
|
|
||||||
'params': {
|
|
||||||
'skip_download': 'm3u8',
|
|
||||||
},
|
|
||||||
}, {
|
|
||||||
'url': 'https://zen.yandex.ru/media/id/606fd806cc13cb3c58c05cf5/vot-eto-focus-dedy-morozy-na-gidrociklah-60c7c443da18892ebfe85ed7',
|
'url': 'https://zen.yandex.ru/media/id/606fd806cc13cb3c58c05cf5/vot-eto-focus-dedy-morozy-na-gidrociklah-60c7c443da18892ebfe85ed7',
|
||||||
'info_dict': {
|
'info_dict': {
|
||||||
'id': '60c7c443da18892ebfe85ed7',
|
'id': '60c7c443da18892ebfe85ed7',
|
||||||
'ext': 'mp4',
|
'ext': 'mp4',
|
||||||
'title': 'ВОТ ЭТО Focus. Деды Морозы на гидроциклах',
|
'title': 'ВОТ ЭТО Focus. Деды Морозы на гидроциклах',
|
||||||
'description': 'md5:f3db3d995763b9bbb7b56d4ccdedea89',
|
'description': 'md5:f3db3d995763b9bbb7b56d4ccdedea89',
|
||||||
'thumbnail': 're:^https://avatars.mds.yandex.net/',
|
'thumbnail': 're:^https://avatars.dzeninfra.ru/',
|
||||||
'uploader': 'AcademeG DailyStream'
|
'uploader': 'AcademeG DailyStream'
|
||||||
},
|
},
|
||||||
'params': {
|
'params': {
|
||||||
'skip_download': 'm3u8',
|
'skip_download': 'm3u8',
|
||||||
'format': 'bestvideo',
|
'format': 'bestvideo',
|
||||||
},
|
},
|
||||||
|
'skip': 'The page does not exist',
|
||||||
|
}, {
|
||||||
|
'url': 'https://dzen.ru/media/id/606fd806cc13cb3c58c05cf5/vot-eto-focus-dedy-morozy-na-gidrociklah-60c7c443da18892ebfe85ed7',
|
||||||
|
'info_dict': {
|
||||||
|
'id': '60c7c443da18892ebfe85ed7',
|
||||||
|
'ext': 'mp4',
|
||||||
|
'title': 'ВОТ ЭТО Focus. Деды Морозы на гидроциклах',
|
||||||
|
'description': 'md5:f3db3d995763b9bbb7b56d4ccdedea89',
|
||||||
|
'thumbnail': r're:^https://avatars\.dzeninfra\.ru/',
|
||||||
|
'uploader': 'AcademeG DailyStream',
|
||||||
|
'upload_date': '20191111',
|
||||||
|
'timestamp': 1573465585,
|
||||||
|
},
|
||||||
|
'params': {'skip_download': 'm3u8'},
|
||||||
}, {
|
}, {
|
||||||
'url': 'https://zen.yandex.ru/video/watch/6002240ff8b1af50bb2da5e3',
|
'url': 'https://zen.yandex.ru/video/watch/6002240ff8b1af50bb2da5e3',
|
||||||
'info_dict': {
|
'info_dict': {
|
||||||
|
@ -223,21 +225,42 @@ class ZenYandexIE(InfoExtractor):
|
||||||
'ext': 'mp4',
|
'ext': 'mp4',
|
||||||
'title': 'Извержение вулкана из спичек: зрелищный опыт',
|
'title': 'Извержение вулкана из спичек: зрелищный опыт',
|
||||||
'description': 'md5:053ad3c61b5596d510c9a199dc8ee633',
|
'description': 'md5:053ad3c61b5596d510c9a199dc8ee633',
|
||||||
'uploader': 'Популярная механика',
|
'thumbnail': r're:^https://avatars\.dzeninfra\.ru/',
|
||||||
|
'uploader': 'TechInsider',
|
||||||
|
'timestamp': 1611378221,
|
||||||
|
'upload_date': '20210123',
|
||||||
},
|
},
|
||||||
'params': {
|
'params': {'skip_download': 'm3u8'},
|
||||||
'skip_download': 'm3u8',
|
}, {
|
||||||
|
'url': 'https://dzen.ru/video/watch/6002240ff8b1af50bb2da5e3',
|
||||||
|
'info_dict': {
|
||||||
|
'id': '6002240ff8b1af50bb2da5e3',
|
||||||
|
'ext': 'mp4',
|
||||||
|
'title': 'Извержение вулкана из спичек: зрелищный опыт',
|
||||||
|
'description': 'md5:053ad3c61b5596d510c9a199dc8ee633',
|
||||||
|
'thumbnail': 're:^https://avatars.dzeninfra.ru/',
|
||||||
|
'uploader': 'TechInsider',
|
||||||
|
'upload_date': '20210123',
|
||||||
|
'timestamp': 1611378221,
|
||||||
},
|
},
|
||||||
|
'params': {'skip_download': 'm3u8'},
|
||||||
}, {
|
}, {
|
||||||
'url': 'https://zen.yandex.ru/media/id/606fd806cc13cb3c58c05cf5/novyi-samsung-fold-3-moskvich-barahlit-612f93b7f8d48e7e945792a2?from=channel&rid=2286618386.482.1630817595976.42360',
|
'url': 'https://zen.yandex.ru/media/id/606fd806cc13cb3c58c05cf5/novyi-samsung-fold-3-moskvich-barahlit-612f93b7f8d48e7e945792a2?from=channel&rid=2286618386.482.1630817595976.42360',
|
||||||
'only_matching': True,
|
'only_matching': True,
|
||||||
|
}, {
|
||||||
|
'url': 'https://dzen.ru/media/id/606fd806cc13cb3c58c05cf5/novyi-samsung-fold-3-moskvich-barahlit-612f93b7f8d48e7e945792a2?from=channel&rid=2286618386.482.1630817595976.42360',
|
||||||
|
'only_matching': True,
|
||||||
}]
|
}]
|
||||||
|
|
||||||
def _real_extract(self, url):
|
def _real_extract(self, url):
|
||||||
id = self._match_id(url)
|
video_id = self._match_id(url)
|
||||||
webpage = self._download_webpage(url, id)
|
webpage = self._download_webpage(url, video_id)
|
||||||
data_json = self._parse_json(
|
redirect = self._search_json(r'var it\s*=\s*', webpage, 'redirect', id, default={}).get('retpath')
|
||||||
self._search_regex(r'data\s*=\s*({["\']_*serverState_*video.+?});', webpage, 'metadata'), id)
|
if redirect:
|
||||||
|
video_id = self._match_id(redirect)
|
||||||
|
webpage = self._download_webpage(redirect, video_id, note='Redirecting')
|
||||||
|
data_json = self._search_json(
|
||||||
|
r'data\s*=', webpage, 'metadata', video_id, contains_pattern=r'{["\']_*serverState_*video.+}')
|
||||||
serverstate = self._search_regex(r'(_+serverState_+video-site_[^_]+_+)',
|
serverstate = self._search_regex(r'(_+serverState_+video-site_[^_]+_+)',
|
||||||
webpage, 'server state').replace('State', 'Settings')
|
webpage, 'server state').replace('State', 'Settings')
|
||||||
uploader = self._search_regex(r'(<a\s*class=["\']card-channel-link[^"\']+["\'][^>]+>)',
|
uploader = self._search_regex(r'(<a\s*class=["\']card-channel-link[^"\']+["\'][^>]+>)',
|
||||||
|
@ -254,11 +277,12 @@ def _real_extract(self, url):
|
||||||
formats.extend(self._extract_m3u8_formats(s_url, id, 'mp4'))
|
formats.extend(self._extract_m3u8_formats(s_url, id, 'mp4'))
|
||||||
self._sort_formats(formats)
|
self._sort_formats(formats)
|
||||||
return {
|
return {
|
||||||
'id': id,
|
'id': video_id,
|
||||||
'title': video_json.get('title') or self._og_search_title(webpage),
|
'title': video_json.get('title') or self._og_search_title(webpage),
|
||||||
'formats': formats,
|
'formats': formats,
|
||||||
'duration': int_or_none(video_json.get('duration')),
|
'duration': int_or_none(video_json.get('duration')),
|
||||||
'view_count': int_or_none(video_json.get('views')),
|
'view_count': int_or_none(video_json.get('views')),
|
||||||
|
'timestamp': int_or_none(video_json.get('publicationDate')),
|
||||||
'uploader': uploader_name or data_json.get('authorName') or try_get(data_json, lambda x: x['publisher']['name']),
|
'uploader': uploader_name or data_json.get('authorName') or try_get(data_json, lambda x: x['publisher']['name']),
|
||||||
'description': self._og_search_description(webpage) or try_get(data_json, lambda x: x['og']['description']),
|
'description': self._og_search_description(webpage) or try_get(data_json, lambda x: x['og']['description']),
|
||||||
'thumbnail': self._og_search_thumbnail(webpage) or try_get(data_json, lambda x: x['og']['imageUrl']),
|
'thumbnail': self._og_search_thumbnail(webpage) or try_get(data_json, lambda x: x['og']['imageUrl']),
|
||||||
|
@ -266,40 +290,99 @@ def _real_extract(self, url):
|
||||||
|
|
||||||
|
|
||||||
class ZenYandexChannelIE(InfoExtractor):
|
class ZenYandexChannelIE(InfoExtractor):
|
||||||
_VALID_URL = r'https?://zen\.yandex\.ru/(?!media|video)(?:id/)?(?P<id>[a-z0-9-_]+)'
|
_VALID_URL = r'https?://(zen\.yandex|dzen)\.ru/(?!media|video)(?:id/)?(?P<id>[a-z0-9-_]+)'
|
||||||
_TESTS = [{
|
_TESTS = [{
|
||||||
'url': 'https://zen.yandex.ru/tok_media',
|
'url': 'https://zen.yandex.ru/tok_media',
|
||||||
'info_dict': {
|
'info_dict': {
|
||||||
'id': 'tok_media',
|
'id': 'tok_media',
|
||||||
|
'title': 'СПЕКТР',
|
||||||
|
'description': 'md5:a9e5b3c247b7fe29fd21371a428bcf56',
|
||||||
|
},
|
||||||
|
'playlist_mincount': 169,
|
||||||
|
}, {
|
||||||
|
'url': 'https://dzen.ru/tok_media',
|
||||||
|
'info_dict': {
|
||||||
|
'id': 'tok_media',
|
||||||
|
'title': 'СПЕКТР',
|
||||||
|
'description': 'md5:a9e5b3c247b7fe29fd21371a428bcf56',
|
||||||
},
|
},
|
||||||
'playlist_mincount': 169,
|
'playlist_mincount': 169,
|
||||||
}, {
|
}, {
|
||||||
'url': 'https://zen.yandex.ru/id/606fd806cc13cb3c58c05cf5',
|
'url': 'https://zen.yandex.ru/id/606fd806cc13cb3c58c05cf5',
|
||||||
'info_dict': {
|
'info_dict': {
|
||||||
'id': '606fd806cc13cb3c58c05cf5',
|
'id': '606fd806cc13cb3c58c05cf5',
|
||||||
|
'description': 'md5:517b7c97d8ca92e940f5af65448fd928',
|
||||||
|
'title': 'AcademeG DailyStream',
|
||||||
|
},
|
||||||
|
'playlist_mincount': 657,
|
||||||
|
}, {
|
||||||
|
# Test that the playlist extractor finishes extracting when the
|
||||||
|
# channel has less than one page
|
||||||
|
'url': 'https://zen.yandex.ru/jony_me',
|
||||||
|
'info_dict': {
|
||||||
|
'id': 'jony_me',
|
||||||
|
'description': 'md5:a2c62b4ef5cf3e3efb13d25f61f739e1',
|
||||||
|
'title': 'JONY ',
|
||||||
|
},
|
||||||
|
'playlist_count': 20,
|
||||||
|
}, {
|
||||||
|
# Test that the playlist extractor finishes extracting when the
|
||||||
|
# channel has more than one page of entries
|
||||||
|
'url': 'https://zen.yandex.ru/tatyanareva',
|
||||||
|
'info_dict': {
|
||||||
|
'id': 'tatyanareva',
|
||||||
|
'description': 'md5:296b588d60841c3756c9105f237b70c6',
|
||||||
|
'title': 'Татьяна Рева',
|
||||||
|
'entries': 'maxcount:200',
|
||||||
|
},
|
||||||
|
'playlist_count': 46,
|
||||||
|
}, {
|
||||||
|
'url': 'https://dzen.ru/id/606fd806cc13cb3c58c05cf5',
|
||||||
|
'info_dict': {
|
||||||
|
'id': '606fd806cc13cb3c58c05cf5',
|
||||||
|
'title': 'AcademeG DailyStream',
|
||||||
|
'description': 'md5:517b7c97d8ca92e940f5af65448fd928',
|
||||||
},
|
},
|
||||||
'playlist_mincount': 657,
|
'playlist_mincount': 657,
|
||||||
}]
|
}]
|
||||||
|
|
||||||
def _entries(self, id, url):
|
def _entries(self, item_id, server_state_json, server_settings_json):
|
||||||
webpage = self._download_webpage(url, id)
|
items = (traverse_obj(server_state_json, ('feed', 'items', ...))
|
||||||
data_json = self._parse_json(re.findall(r'var\s?data\s?=\s?({.+?})\s?;', webpage)[-1], id)
|
or traverse_obj(server_settings_json, ('exportData', 'items', ...)))
|
||||||
for key in data_json.keys():
|
|
||||||
if key.startswith('__serverState__'):
|
more = (traverse_obj(server_state_json, ('links', 'more'))
|
||||||
data_json = data_json[key]
|
or traverse_obj(server_settings_json, ('exportData', 'more', 'link')))
|
||||||
items = list(try_get(data_json, lambda x: x['feed']['items'], dict).values())
|
|
||||||
more = try_get(data_json, lambda x: x['links']['more']) or None
|
next_page_id = None
|
||||||
for page in itertools.count(1):
|
for page in itertools.count(1):
|
||||||
for item in items:
|
for item in items or []:
|
||||||
video_id = item.get('publication_id') or item.get('publicationId')
|
if item.get('type') != 'gif':
|
||||||
video_url = item.get('link')
|
continue
|
||||||
yield self.url_result(video_url, ie=ZenYandexIE.ie_key(), video_id=video_id.split(':')[-1])
|
video_id = traverse_obj(item, 'publication_id', 'publicationId') or ''
|
||||||
if not more:
|
yield self.url_result(item['link'], ZenYandexIE, video_id.split(':')[-1])
|
||||||
|
|
||||||
|
current_page_id = next_page_id
|
||||||
|
next_page_id = traverse_obj(parse_qs(more), ('next_page_id', -1))
|
||||||
|
if not all((more, items, next_page_id, next_page_id != current_page_id)):
|
||||||
break
|
break
|
||||||
data_json = self._download_json(more, id, note='Downloading Page %d' % page)
|
|
||||||
items = data_json.get('items', [])
|
data = self._download_json(more, item_id, note=f'Downloading Page {page}')
|
||||||
more = try_get(data_json, lambda x: x['more']['link']) or None
|
items, more = data.get('items'), traverse_obj(data, ('more', 'link'))
|
||||||
|
|
||||||
def _real_extract(self, url):
|
def _real_extract(self, url):
|
||||||
id = self._match_id(url)
|
item_id = self._match_id(url)
|
||||||
return self.playlist_result(self._entries(id, url), playlist_id=id)
|
webpage = self._download_webpage(url, item_id)
|
||||||
|
redirect = self._search_json(
|
||||||
|
r'var it\s*=\s*', webpage, 'redirect', item_id, default={}).get('retpath')
|
||||||
|
if redirect:
|
||||||
|
item_id = self._match_id(redirect)
|
||||||
|
webpage = self._download_webpage(redirect, item_id, note='Redirecting')
|
||||||
|
data = self._search_json(
|
||||||
|
r'var\s+data\s*=', webpage, 'channel data', item_id, contains_pattern=r'{\"__serverState__.+}')
|
||||||
|
server_state_json = traverse_obj(data, lambda k, _: k.startswith('__serverState__'), get_all=False)
|
||||||
|
server_settings_json = traverse_obj(data, lambda k, _: k.startswith('__serverSettings__'), get_all=False)
|
||||||
|
|
||||||
|
return self.playlist_result(
|
||||||
|
self._entries(item_id, server_state_json, server_settings_json),
|
||||||
|
item_id, traverse_obj(server_state_json, ('channel', 'source', 'title')),
|
||||||
|
traverse_obj(server_state_json, ('channel', 'source', 'description')))
|
||||||
|
|
Loading…
Reference in New Issue
Block a user