yt-dlp/youtube_dl/extractor/ellentube.py
2017-12-09 02:16:48 +07:00

141 lines
5.3 KiB
Python

# coding: utf-8
from __future__ import unicode_literals
from .common import InfoExtractor
from ..utils import (
clean_html,
int_or_none,
)
class EllenTubeIE(InfoExtractor):
_VALID_URL = r'''(?x)
(?:
https://api-prod\.ellentube\.com/ellenapi/api/item/
|ellentube:
)
(?P<id>
[\da-f]{8}-[\da-f]{4}-[\da-f]{4}-[\da-f]{4}-[\da-f]{12}
)'''
_TESTS = [{
'url': 'https://api-prod.ellentube.com/ellenapi/api/item/75c64c16-aefd-4558-b4f5-3de09b22e6fc',
'match_only': True,
}, {
'url': 'ellentube:734a3353-f697-4e79-9ca9-bfc3002dc1e0',
'match_only': True,
}]
def _real_extract(self, url):
video_id = self._match_id(url)
data = self._download_json(
'https://api-prod.ellentube.com/ellenapi/api/item/%s' % video_id, video_id)
title = data['title']
description = data.get('description')
publish_time = int_or_none(data.get('publishTime'))
thumbnail = data.get('thumbnail')
formats = []
duration = None
for entry in data.get('media'):
if entry.get('id') == 'm3u8':
formats = self._extract_m3u8_formats(
entry.get('url'), video_id, 'mp4', entry_protocol='m3u8_native', m3u8_id='hls')
duration = int_or_none(entry.get('duration'))
break
self._sort_formats(formats)
return {
'id': video_id,
'title': title,
'description': description,
'duration': duration,
'thumbnail': thumbnail,
'timestamp': publish_time,
'formats': formats,
}
class EllenTubeVideoIE(InfoExtractor):
_VALID_URL = r'https?://(?:www\.)?ellentube\.com/video/(?P<id>.+)\.html'
_TEST = {
'url': 'https://www.ellentube.com/video/ellen-meets-las-vegas-survivors-jesus-campos-and-stephen-schuck.html',
'md5': '2fabc277131bddafdd120e0fc0f974c9',
'info_dict': {
'id': '0822171c-3829-43bf-b99f-d77358ae75e3',
'ext': 'mp4',
'title': 'Ellen Meets Las Vegas Survivors Jesus Campos and Stephen Schuck',
'description': 'md5:76e3355e2242a78ad9e3858e5616923f',
'duration': 514,
'timestamp': 1508505120000,
'thumbnail': 'https://warnerbros-h.assetsadobe.com/is/image/content/dam/ellen/videos/episodes/season15/32/video--2728751654987218111',
}
}
def _real_extract(self, url):
display_id = self._match_id(url)
webpage = self._download_webpage(url, display_id)
video_id = self._html_search_regex(
r'(?s)<!--\s*CONTENT\s*-->.*data-config.+([\da-f]{8}-[\da-f]{4}-[\da-f]{4}-[\da-f]{4}-[\da-f]{12})',
webpage, 'video id')
return self.url_result('ellentube:%s' % video_id, 'EllenTube')
class EllenTubePlaylistIE(InfoExtractor):
def _extract_videos_from_json(self, data, display_id):
return [self.url_result('ellentube:%s' % elem['id'], 'EllenTube')
for elem in data if elem.get('type') == 'VIDEO']
def _extract_playlist(self, url, display_id, extract_description=True):
webpage = self._download_webpage(url, display_id)
playlist_data = self._html_search_regex(
r'<div\s+data-component\s*=\s*"Details"(.+)</div>', webpage, 'playlist data')
playlist_title = self._search_regex(
r'"title"\s*:\s*"(.+?)"', playlist_data, 'playlist title')
playlist_description = clean_html(self._search_regex(
r'"description"\s*:\s*"(.+?)"', playlist_data, 'playlist description',
fatal=False)) if extract_description else None
api_search = self._search_regex(
r'"filter"\s*:\s*"(.+?)"', playlist_data, 'playlist api request')
api_data = self._download_json(
'https://api-prod.ellentube.com/ellenapi/api/feed/?%s' % api_search,
display_id)
return self.playlist_result(
self._extract_videos_from_json(api_data, display_id),
display_id, playlist_title, playlist_description)
class EllenTubeEpisodeIE(EllenTubePlaylistIE):
_VALID_URL = r'https?://(?:www\.)?ellentube\.com/episode/(?P<id>.+)\.html'
_TEST = {
'url': 'https://www.ellentube.com/episode/dax-shepard-jordan-fisher-haim.html',
'info_dict': {
'id': 'dax-shepard-jordan-fisher-haim',
'title': 'Dax Shepard, \'DWTS\' Team Jordan Fisher & Lindsay Arnold, HAIM',
'description': 'md5:aed85d42892f6126e71ec5ed2aea2a0d'
},
'playlist_count': 6,
}
def _real_extract(self, url):
display_id = self._match_id(url)
return self._extract_playlist(url, display_id)
class EllenTubeStudioIE(EllenTubePlaylistIE):
_VALID_URL = r'https?://(?:www\.)?ellentube\.com/studios/(?P<id>.+)\.html'
_TEST = {
'url': 'https://www.ellentube.com/studios/macey-goes-rving0.html',
'info_dict': {
'id': 'macey-goes-rving0',
'title': 'Macey Goes RVing',
},
'playlist_mincount': 3,
}
def _real_extract(self, url):
display_id = self._match_id(url)
return self._extract_playlist(url, display_id, False)