yt-dlc/yt_dlp/extractor/zee5.py

206 lines
8.3 KiB
Python

# coding: utf-8
from __future__ import unicode_literals
import re
from .common import InfoExtractor
from ..compat import compat_str
from ..utils import (
int_or_none,
parse_age_limit,
str_or_none,
try_get,
unified_strdate,
unified_timestamp,
url_or_none,
)
class Zee5IE(InfoExtractor):
_VALID_URL = r'''(?x)
(?:
zee5:|
(?:https?://)(?:www\.)?zee5\.com/(?:[^#?]+/)?
(?:
(?:tvshows|kids|zee5originals)(?:/[^#/?]+){3}
|movies/[^#/?]+
)/(?P<display_id>[^#/?]+)/
)
(?P<id>[^#/?]+)/?(?:$|[?#])
'''
_TESTS = [{
'url': 'https://www.zee5.com/movies/details/krishna-the-birth/0-0-63098',
'info_dict': {
'id': '0-0-63098',
'ext': 'mp4',
'display_id': 'krishna-the-birth',
'title': 'Krishna - The Birth',
'duration': 4368,
'average_rating': 4,
'description': str,
'alt_title': 'Krishna - The Birth',
'uploader': 'Zee Entertainment Enterprises Ltd',
'release_date': '20060101',
'upload_date': '20060101',
'timestamp': 1136073600,
'thumbnail': 'https://akamaividz.zee5.com/resources/0-0-63098/list/270x152/0063098_list_80888170.jpg',
'tags': list
},
'params': {
'format': 'bv',
},
}, {
'url': 'https://zee5.com/tvshows/details/krishna-balram/0-6-1871/episode-1-the-test-of-bramha/0-1-233402',
'info_dict': {
'id': '0-1-233402',
'ext': 'mp4',
'display_id': 'episode-1-the-test-of-bramha',
'title': 'Episode 1 - The Test Of Bramha',
'duration': 1336,
'average_rating': 4,
'description': str,
'alt_title': 'Episode 1 - The Test Of Bramha',
'uploader': 'Green Gold',
'release_date': '20090101',
'upload_date': '20090101',
'timestamp': 1230768000,
'thumbnail': 'https://akamaividz.zee5.com/resources/0-1-233402/list/270x152/01233402_list.jpg',
'series': 'Krishna Balram',
'season_number': 1,
'episode_number': 1,
'tags': list,
},
'params': {
'format': 'bv',
},
}, {
'url': 'https://www.zee5.com/hi/tvshows/details/kundali-bhagya/0-6-366/kundali-bhagya-march-08-2021/0-1-manual_7g9jv1os7730?country=IN',
'only_matching': True
}, {
'url': 'https://www.zee5.com/global/hi/tvshows/details/kundali-bhagya/0-6-366/kundali-bhagya-march-08-2021/0-1-manual_7g9jv1os7730',
'only_matching': True
}]
def _real_extract(self, url):
video_id, display_id = re.match(self._VALID_URL, url).group('id', 'display_id')
access_token_request = self._download_json(
'https://useraction.zee5.com/token/platform_tokens.php?platform_name=web_app',
video_id, note='Downloading access token')
token_request = self._download_json(
'https://useraction.zee5.com/tokennd',
video_id, note='Downloading video token')
json_data = self._download_json(
'https://gwapi.zee5.com/content/details/{}?translation=en&country=IN'.format(video_id),
video_id, headers={'X-Access-Token': access_token_request['token']})
m3u8_url = try_get(
json_data,
(lambda x: x['hls'][0], lambda x: x['video_details']['hls_url']),
str)
formats = self._extract_m3u8_formats(
'https://zee5vodnd.akamaized.net' + m3u8_url.replace('/drm1/', '/hls1/') + token_request['video_token'],
video_id, fatal=False)
mpd_url = try_get(
json_data,
(lambda x: x['video'][0], lambda x: x['video_details']['url']),
str)
formats += self._extract_mpd_formats(
'https://zee5vodnd.akamaized.net' + mpd_url + token_request['video_token'],
video_id, fatal=False)
self._sort_formats(formats)
return {
'id': video_id,
'display_id': display_id,
'title': json_data['title'],
'formats': formats,
'duration': int_or_none(json_data.get('duration')),
'average_rating': int_or_none(json_data.get('rating')),
'description': str_or_none(json_data.get('description')),
'alt_title': str_or_none(json_data.get('original_title')),
'uploader': str_or_none(json_data.get('content_owner')),
'age_limit': parse_age_limit(json_data.get('age_rating')),
'release_date': unified_strdate(json_data.get('release_date')),
'timestamp': unified_timestamp(json_data.get('release_date')),
'thumbnail': url_or_none(json_data.get('image_url')),
'series': try_get(json_data, lambda x: x['tvshow_details']['title'], str),
'season': try_get(json_data, lambda x: x['season_details']['title'], str),
'season_number': int_or_none(try_get(json_data, lambda x: x['season_details']['index'])),
'episode_number': int_or_none(try_get(json_data, lambda x: x['index'])),
'tags': try_get(json_data, lambda x: x['tags'], list)
}
class Zee5SeriesIE(InfoExtractor):
IE_NAME = 'zee5:series'
_VALID_URL = r'''(?x)
(?:
zee5:series:|
(?:https?://)(?:www\.)?zee5\.com/(?:[^#?]+/)?
(?:tvshows|kids|zee5originals)(?:/[^#/?]+){2}/
)
(?P<id>[^#/?]+)/?(?:$|[?#])
'''
_TESTS = [{
'url': 'https://www.zee5.com/kids/kids-shows/krishna-balram/0-6-1871',
'playlist_mincount': 43,
'info_dict': {
'id': '0-6-1871',
},
}, {
'url': 'https://www.zee5.com/tvshows/details/bhabi-ji-ghar-par-hai/0-6-199',
'playlist_mincount': 1500,
'info_dict': {
'id': '0-6-199',
},
}, {
'url': 'https://www.zee5.com/tvshows/details/agent-raghav-crime-branch/0-6-965',
'playlist_mincount': 25,
'info_dict': {
'id': '0-6-965',
},
}, {
'url': 'https://www.zee5.com/ta/tvshows/details/nagabhairavi/0-6-3201',
'playlist_mincount': 3,
'info_dict': {
'id': '0-6-3201',
},
}, {
'url': 'https://www.zee5.com/global/hi/tvshows/details/khwaabon-ki-zamin-par/0-6-270',
'playlist_mincount': 150,
'info_dict': {
'id': '0-6-270',
},
}
]
def _entries(self, show_id):
access_token_request = self._download_json(
'https://useraction.zee5.com/token/platform_tokens.php?platform_name=web_app',
show_id, note='Downloading access token')
headers = {
'X-Access-Token': access_token_request['token'],
'Referer': 'https://www.zee5.com/',
}
show_url = 'https://gwapi.zee5.com/content/tvshow/{}?translation=en&country=IN'.format(show_id)
page_num = 0
show_json = self._download_json(show_url, video_id=show_id, headers=headers)
for season in show_json.get('seasons') or []:
season_id = try_get(season, lambda x: x['id'], compat_str)
next_url = 'https://gwapi.zee5.com/content/tvshow/?season_id={}&type=episode&translation=en&country=IN&on_air=false&asset_subtype=tvshow&page=1&limit=100'.format(season_id)
while next_url:
page_num += 1
episodes_json = self._download_json(
next_url, video_id=show_id, headers=headers,
note='Downloading JSON metadata page %d' % page_num)
for episode in try_get(episodes_json, lambda x: x['episode'], list) or []:
video_id = episode.get('id')
yield self.url_result(
'zee5:%s' % video_id,
ie=Zee5IE.ie_key(), video_id=video_id)
next_url = url_or_none(episodes_json.get('next_episode_api'))
def _real_extract(self, url):
show_id = self._match_id(url)
return self.playlist_result(self._entries(show_id), playlist_id=show_id)