[viewster] Rewrite for new API (Closes #6317)
parent
28afa6e77a
commit
b68a2613f8
@ -1,129 +1,137 @@
|
||||
# coding: utf-8
|
||||
from __future__ import unicode_literals
|
||||
|
||||
from .common import InfoExtractor
|
||||
from ..compat import compat_urllib_request
|
||||
from ..compat import (
|
||||
compat_urllib_request,
|
||||
compat_urllib_parse,
|
||||
)
|
||||
from ..utils import (
|
||||
determine_ext,
|
||||
int_or_none,
|
||||
parse_iso8601,
|
||||
)
|
||||
|
||||
|
||||
class ViewsterIE(InfoExtractor):
|
||||
_VALID_URL = r'http://(?:www\.)?viewster\.com/movie/(?P<id>\d+-\d+-\d+)'
|
||||
_VALID_URL = r'http://(?:www\.)?viewster\.com/(?:serie|movie)/(?P<id>\d+-\d+-\d+)'
|
||||
_TESTS = [{
|
||||
# movielink, paymethod=fre
|
||||
'url': 'http://www.viewster.com/movie/1293-19341-000/hout-wood/',
|
||||
'playlist': [{
|
||||
'md5': '8f9d94b282d80c42b378dffdbb11caf3',
|
||||
'info_dict': {
|
||||
'id': '1293-19341-000-movie',
|
||||
'ext': 'flv',
|
||||
'title': "'Hout' (Wood) - Movie",
|
||||
},
|
||||
}],
|
||||
'info_dict': {
|
||||
'id': '1293-19341-000',
|
||||
'title': "'Hout' (Wood)",
|
||||
'description': 'md5:925733185a9242ef96f436937683f33b',
|
||||
}
|
||||
}, {
|
||||
# movielink, paymethod=adv
|
||||
# movie, Type=Movie
|
||||
'url': 'http://www.viewster.com/movie/1140-11855-000/the-listening-project/',
|
||||
'playlist': [{
|
||||
'md5': '77a005453ca7396cbe3d35c9bea30aef',
|
||||
'info_dict': {
|
||||
'id': '1140-11855-000-movie',
|
||||
'ext': 'flv',
|
||||
'title': "THE LISTENING PROJECT - Movie",
|
||||
},
|
||||
}],
|
||||
'md5': '14d3cfffe66d57b41ae2d9c873416f01',
|
||||
'info_dict': {
|
||||
'id': '1140-11855-000',
|
||||
'title': "THE LISTENING PROJECT",
|
||||
'description': 'md5:714421ae9957e112e672551094bf3b08',
|
||||
}
|
||||
'ext': 'flv',
|
||||
'title': 'The listening Project',
|
||||
'description': 'md5:bac720244afd1a8ea279864e67baa071',
|
||||
'timestamp': 1214870400,
|
||||
'upload_date': '20080701',
|
||||
'duration': 4680,
|
||||
},
|
||||
}, {
|
||||
# direct links, no movielink
|
||||
'url': 'http://www.viewster.com/movie/1198-56411-000/sinister/',
|
||||
'playlist': [{
|
||||
'md5': '0307b7eac6bfb21ab0577a71f6eebd8f',
|
||||
'info_dict': {
|
||||
'id': '1198-56411-000-trailer',
|
||||
'ext': 'mp4',
|
||||
'title': "Sinister - Trailer",
|
||||
},
|
||||
}, {
|
||||
'md5': '80b9ee3ad69fb368f104cb5d9732ae95',
|
||||
'info_dict': {
|
||||
'id': '1198-56411-000-behind-scenes',
|
||||
'ext': 'mp4',
|
||||
'title': "Sinister - Behind Scenes",
|
||||
},
|
||||
}, {
|
||||
'md5': '3b3ea897ecaa91fca57a8a94ac1b15c5',
|
||||
'info_dict': {
|
||||
'id': '1198-56411-000-scene-from-movie',
|
||||
'ext': 'mp4',
|
||||
'title': "Sinister - Scene from movie",
|
||||
},
|
||||
}],
|
||||
# series episode, Type=Episode
|
||||
'url': 'http://www.viewster.com/serie/1284-19427-001/the-world-and-a-wall/',
|
||||
'md5': 'd5434c80fcfdb61651cc2199a88d6ba3',
|
||||
'info_dict': {
|
||||
'id': '1198-56411-000',
|
||||
'title': "Sinister",
|
||||
'description': 'md5:014c40b0488848de9683566a42e33372',
|
||||
}
|
||||
'id': '1284-19427-001',
|
||||
'ext': 'flv',
|
||||
'title': 'The World and a Wall',
|
||||
'description': 'md5:24814cf74d3453fdf5bfef9716d073e3',
|
||||
'timestamp': 1428192000,
|
||||
'upload_date': '20150405',
|
||||
'duration': 1500,
|
||||
},
|
||||
}, {
|
||||
# serie, Type=Serie
|
||||
'url': 'http://www.viewster.com/serie/1303-19426-000/',
|
||||
'info_dict': {
|
||||
'id': '1303-19426-000',
|
||||
'title': 'Is It Wrong to Try to Pick up Girls in a Dungeon?',
|
||||
'description': 'md5:eeda9bef25b0d524b3a29a97804c2f11',
|
||||
},
|
||||
'playlist_count': 13,
|
||||
}, {
|
||||
# unfinished serie, no Type
|
||||
'url': 'http://www.viewster.com/serie/1284-19427-000/baby-steps-season-2/',
|
||||
'info_dict': {
|
||||
'id': '1284-19427-000',
|
||||
'title': 'Baby Steps—Season 2',
|
||||
'description': 'md5:e7097a8fc97151e25f085c9eb7a1cdb1',
|
||||
},
|
||||
'playlist_mincount': 16,
|
||||
}]
|
||||
|
||||
_ACCEPT_HEADER = 'application/json, text/javascript, */*; q=0.01'
|
||||
_AUTH_TOKEN = '/YqhSYsx8EaU9Bsta3ojlA=='
|
||||
|
||||
def _real_extract(self, url):
|
||||
video_id = self._match_id(url)
|
||||
|
||||
request = compat_urllib_request.Request(
|
||||
'http://api.live.viewster.com/api/v1/movie/%s' % video_id)
|
||||
def _download_json(self, url, video_id, note='Downloading JSON metadata', fatal=True):
|
||||
request = compat_urllib_request.Request(url)
|
||||
request.add_header('Accept', self._ACCEPT_HEADER)
|
||||
request.add_header('Auth-token', self._AUTH_TOKEN)
|
||||
return super(ViewsterIE, self)._download_json(request, video_id, note, fatal=fatal)
|
||||
|
||||
movie = self._download_json(
|
||||
request, video_id, 'Downloading movie metadata JSON')
|
||||
|
||||
title = movie.get('title') or movie['original_title']
|
||||
description = movie.get('synopsis')
|
||||
thumbnail = movie.get('large_artwork') or movie.get('artwork')
|
||||
|
||||
entries = []
|
||||
for clip in movie['play_list']:
|
||||
entry = None
|
||||
|
||||
# movielink api
|
||||
link_request = clip.get('link_request')
|
||||
if link_request:
|
||||
request = compat_urllib_request.Request(
|
||||
'http://api.live.viewster.com/api/v1/movielink?movieid=%(movieid)s&action=%(action)s&paymethod=%(paymethod)s&price=%(price)s¤cy=%(currency)s&language=%(language)s&subtitlelanguage=%(subtitlelanguage)s&ischromecast=%(ischromecast)s'
|
||||
% link_request)
|
||||
request.add_header('Accept', self._ACCEPT_HEADER)
|
||||
def _real_extract(self, url):
|
||||
video_id = self._match_id(url)
|
||||
|
||||
movie_link = self._download_json(
|
||||
request, video_id, 'Downloading movie link JSON', fatal=False)
|
||||
info = self._download_json(
|
||||
'https://public-api.viewster.com/search/%s' % video_id,
|
||||
video_id, 'Downloading entry JSON')
|
||||
|
||||
if movie_link:
|
||||
formats = self._extract_f4m_formats(
|
||||
movie_link['url'] + '&hdcore=3.2.0&plugin=flowplayer-3.2.0.1', video_id)
|
||||
self._sort_formats(formats)
|
||||
entry = {
|
||||
'formats': formats,
|
||||
}
|
||||
entry_id = info.get('Id') or info['id']
|
||||
|
||||
# direct link
|
||||
clip_url = clip.get('clip_data', {}).get('url')
|
||||
if clip_url:
|
||||
entry = {
|
||||
'url': clip_url,
|
||||
'ext': 'mp4',
|
||||
}
|
||||
# unfinished serie has no Type
|
||||
if info.get('Type') in ['Serie', None]:
|
||||
episodes = self._download_json(
|
||||
'https://public-api.viewster.com/series/%s/episodes' % entry_id,
|
||||
video_id, 'Downloading series JSON')
|
||||
entries = [
|
||||
self.url_result(
|
||||
'http://www.viewster.com/movie/%s' % episode['OriginId'], 'Viewster')
|
||||
for episode in episodes]
|
||||
title = info.get('Title') or info['Synopsis']['Title']
|
||||
description = info.get('Synopsis', {}).get('Detailed')
|
||||
return self.playlist_result(entries, video_id, title, description)
|
||||
|
||||
if entry:
|
||||
entry.update({
|
||||
'id': '%s-%s' % (video_id, clip['canonical_title']),
|
||||
'title': '%s - %s' % (title, clip['title']),
|
||||
formats = []
|
||||
for media_type in ('application/f4m+xml', 'application/x-mpegURL'):
|
||||
media = self._download_json(
|
||||
'https://public-api.viewster.com/movies/%s/video?mediaType=%s'
|
||||
% (entry_id, compat_urllib_parse.quote(media_type)),
|
||||
video_id, 'Downloading %s JSON' % media_type, fatal=False)
|
||||
if not media:
|
||||
continue
|
||||
video_url = media.get('Uri')
|
||||
if not video_url:
|
||||
continue
|
||||
ext = determine_ext(video_url)
|
||||
if ext == 'f4m':
|
||||
video_url += '&' if '?' in video_url else '?'
|
||||
video_url += 'hdcore=3.2.0&plugin=flowplayer-3.2.0.1'
|
||||
formats.extend(self._extract_f4m_formats(
|
||||
video_url, video_id, f4m_id='hds'))
|
||||
elif ext == 'm3u8':
|
||||
formats.extend(self._extract_m3u8_formats(
|
||||
video_url, video_id, 'mp4', m3u8_id='hls',
|
||||
fatal=False # m3u8 sometimes fail
|
||||
))
|
||||
else:
|
||||
formats.append({
|
||||
'url': video_url,
|
||||
})
|
||||
entries.append(entry)
|
||||
self._sort_formats(formats)
|
||||
|
||||
playlist = self.playlist_result(entries, video_id, title, description)
|
||||
playlist['thumbnail'] = thumbnail
|
||||
return playlist
|
||||
synopsis = info.get('Synopsis', {})
|
||||
# Prefer title outside synopsis since it's less messy
|
||||
title = info.get('Title') or synopsis['Title'].strip()
|
||||
description = synopsis.get('Detailed') or info.get('Synopsis', {}).get('Short')
|
||||
duration = int_or_none(info.get('Duration'))
|
||||
timestamp = parse_iso8601(info.get('ReleaseDate'))
|
||||
|
||||
return {
|
||||
'id': video_id,
|
||||
'title': title,
|
||||
'description': description,
|
||||
'timestamp': timestamp,
|
||||
'duration': duration,
|
||||
'formats': formats,
|
||||
}
|
||||
|
Loading…
Reference in New Issue