0
0
Fork 0
mirror of https://github.com/yt-dlp/yt-dlp.git synced 2024-11-22 02:15:12 +00:00

[dropout] Add extractor (#2045)

Authored-by: TwoThousandHedgehogs, pukkandan
This commit is contained in:
Sonic 2021-12-24 06:49:33 -05:00 committed by GitHub
parent a709d87335
commit e9efb99f66
No known key found for this signature in database
GPG key ID: 4AEE18F83AFDEB23
2 changed files with 216 additions and 0 deletions

212
yt_dlp/extractor/dropout.py Normal file
View file

@ -0,0 +1,212 @@
# coding: utf-8
from .common import InfoExtractor
from .vimeo import VHXEmbedIE
from ..utils import (
clean_html,
ExtractorError,
get_element_by_class,
get_element_by_id,
get_elements_by_class,
int_or_none,
join_nonempty,
unified_strdate,
urlencode_postdata,
)
class DropoutIE(InfoExtractor):
_LOGIN_URL = 'https://www.dropout.tv/login'
_NETRC_MACHINE = 'dropout'
_VALID_URL = r'https?://(?:www\.)?dropout\.tv/(?:[^/]+/)*videos/(?P<id>[^/]+)/?$'
_TESTS = [
{
'url': 'https://www.dropout.tv/game-changer/season:2/videos/yes-or-no',
'note': 'Episode in a series',
'md5': '5e000fdfd8d8fa46ff40456f1c2af04a',
'info_dict': {
'id': '738153',
'display_id': 'yes-or-no',
'ext': 'mp4',
'title': 'Yes or No',
'description': 'Ally, Brennan, and Zac are asked a simple question, but is there a correct answer?',
'release_date': '20200508',
'thumbnail': 'https://vhx.imgix.net/chuncensoredstaging/assets/351e3f24-c4a3-459a-8b79-dc80f1e5b7fd.jpg',
'series': 'Game Changer',
'season_number': 2,
'season': 'Season 2',
'episode_number': 6,
'episode': 'Yes or No',
'duration': 1180,
'uploader_id': 'user80538407',
'uploader_url': 'https://vimeo.com/user80538407',
'uploader': 'OTT Videos'
},
'expected_warnings': ['Ignoring subtitle tracks found in the HLS manifest']
},
{
'url': 'https://www.dropout.tv/dimension-20-fantasy-high/season:1/videos/episode-1',
'note': 'Episode in a series (missing release_date)',
'md5': '712caf7c191f1c47c8f1879520c2fa5c',
'info_dict': {
'id': '320562',
'display_id': 'episode-1',
'ext': 'mp4',
'title': 'The Beginning Begins',
'description': 'The cast introduces their PCs, including a neurotic elf, a goblin PI, and a corn-worshipping cleric.',
'thumbnail': 'https://vhx.imgix.net/chuncensoredstaging/assets/4421ed0d-f630-4c88-9004-5251b2b8adfa.jpg',
'series': 'Dimension 20: Fantasy High',
'season_number': 1,
'season': 'Season 1',
'episode_number': 1,
'episode': 'The Beginning Begins',
'duration': 6838,
'uploader_id': 'user80538407',
'uploader_url': 'https://vimeo.com/user80538407',
'uploader': 'OTT Videos'
},
'expected_warnings': ['Ignoring subtitle tracks found in the HLS manifest']
},
{
'url': 'https://www.dropout.tv/videos/misfits-magic-holiday-special',
'note': 'Episode not in a series',
'md5': 'c30fa18999c5880d156339f13c953a26',
'info_dict': {
'id': '1915774',
'display_id': 'misfits-magic-holiday-special',
'ext': 'mp4',
'title': 'Misfits & Magic Holiday Special',
'description': 'The magical misfits spend Christmas break at Gowpenny, with an unwelcome visitor.',
'release_date': '20211215',
'thumbnail': 'https://vhx.imgix.net/chuncensoredstaging/assets/d91ea8a6-b250-42ed-907e-b30fb1c65176-8e24b8e5.jpg',
'duration': 11698,
'uploader_id': 'user80538407',
'uploader_url': 'https://vimeo.com/user80538407',
'uploader': 'OTT Videos'
},
'expected_warnings': ['Ignoring subtitle tracks found in the HLS manifest']
}
]
def _get_authenticity_token(self, display_id):
signin_page = self._download_webpage(
self._LOGIN_URL, display_id, note='Getting authenticity token')
return self._html_search_regex(
r'name=["\']authenticity_token["\'] value=["\'](.+?)["\']',
signin_page, 'authenticity_token')
def _login(self, display_id):
username, password = self._get_login_info()
if not (username and password):
self.raise_login_required(method='password')
response = self._download_webpage(
self._LOGIN_URL, display_id, note='Logging in', data=urlencode_postdata({
'email': username,
'password': password,
'authenticity_token': self._get_authenticity_token(display_id),
'utf8': True
}))
user_has_subscription = self._search_regex(
r'user_has_subscription:\s*["\'](.+?)["\']', response, 'subscription status', default='none')
if user_has_subscription.lower() == 'true':
return response
elif user_has_subscription.lower() == 'false':
raise ExtractorError('Account is not subscribed')
else:
raise ExtractorError('Incorrect username/password')
def _real_extract(self, url):
display_id = self._match_id(url)
try:
self._login(display_id)
webpage = self._download_webpage(url, display_id, note='Downloading video webpage')
finally:
self._download_webpage('https://www.dropout.tv/logout', display_id, note='Logging out')
embed_url = self._search_regex(r'embed_url:\s*["\'](.+?)["\']', webpage, 'embed url')
thumbnail = self._og_search_thumbnail(webpage)
watch_info = get_element_by_id('watch-info', webpage) or ''
title = clean_html(get_element_by_class('video-title', watch_info))
season_episode = get_element_by_class(
'site-font-secondary-color', get_element_by_class('text', watch_info))
episode_number = int_or_none(self._search_regex(
r'Episode (\d+)', season_episode or '', 'episode', default=None))
return {
'_type': 'url_transparent',
'ie_key': VHXEmbedIE.ie_key(),
'url': embed_url,
'id': self._search_regex(r'embed.vhx.tv/videos/(.+?)\?', embed_url, 'id'),
'display_id': display_id,
'title': title,
'description': self._html_search_meta('description', webpage, fatal=False),
'thumbnail': thumbnail.split('?')[0] if thumbnail else None, # Ignore crop/downscale
'series': clean_html(get_element_by_class('series-title', watch_info)),
'episode_number': episode_number,
'episode': title if episode_number else None,
'season_number': int_or_none(self._search_regex(
r'Season (\d+),', season_episode or '', 'season', default=None)),
'release_date': unified_strdate(self._search_regex(
r'data-meta-field-name=["\']release_dates["\'] data-meta-field-value=["\'](.+?)["\']',
watch_info, 'release date', default=None)),
}
class DropoutSeasonIE(InfoExtractor):
_VALID_URL = r'https?://(?:www\.)?dropout\.tv/(?P<id>[^\/$&?#]+)(?:/?$|/season:[0-9]+/?$)'
_TESTS = [
{
'url': 'https://www.dropout.tv/dimension-20-fantasy-high/season:1',
'note': 'Multi-season series with the season in the url',
'playlist_count': 17,
'info_dict': {
'id': 'dimension-20-fantasy-high-season-1',
'title': 'Dimension 20 Fantasy High - Season 1'
}
},
{
'url': 'https://www.dropout.tv/dimension-20-fantasy-high',
'note': 'Multi-season series with the season not in the url',
'playlist_count': 17,
'info_dict': {
'id': 'dimension-20-fantasy-high-season-1',
'title': 'Dimension 20 Fantasy High - Season 1'
}
},
{
'url': 'https://www.dropout.tv/dimension-20-shriek-week',
'note': 'Single-season series',
'playlist_count': 4,
'info_dict': {
'id': 'dimension-20-shriek-week-season-1',
'title': 'Dimension 20 Shriek Week - Season 1'
}
}
]
def _real_extract(self, url):
season_id = self._match_id(url)
season_title = season_id.replace('-', ' ').title()
webpage = self._download_webpage(url, season_id)
entries = [
self.url_result(
url=self._search_regex(r'<a href=["\'](.+?)["\'] class=["\']browse-item-link["\']',
item, 'item_url'),
ie=DropoutIE.ie_key()
) for item in get_elements_by_class('js-collection-item', webpage)
]
seasons = (get_element_by_class('select-dropdown-wrapper', webpage) or '').strip().replace('\n', '')
current_season = self._search_regex(r'<option[^>]+selected>([^<]+)</option>',
seasons, 'current_season', default='').strip()
return {
'_type': 'playlist',
'id': join_nonempty(season_id, current_season.lower().replace(' ', '-')),
'title': join_nonempty(season_title, current_season, delim=' - '),
'entries': entries
}

View file

@ -385,6 +385,10 @@
from .dispeak import DigitallySpeakingIE
from .doodstream import DoodStreamIE
from .dropbox import DropboxIE
from .dropout import (
DropoutSeasonIE,
DropoutIE
)
from .dw import (
DWIE,
DWArticleIE,