# coding: utf-8
from __future__ import unicode_literals
+import re
+
from .common import InfoExtractor
+from ..compat import (
+ compat_parse_qs,
+ compat_str,
+ compat_urllib_parse_urlparse,
+)
from ..utils import (
- qualities,
- int_or_none,
- mimetype2ext,
determine_ext,
+ int_or_none,
+ try_get,
+ qualities,
)
class SixPlayIE(InfoExtractor):
- _VALID_URL = r'(?:6play:|https?://(?:www\.)?6play\.fr/.+?-c_)(?P<id>[0-9]+)'
- _TEST = {
- 'url': 'http://www.6play.fr/jamel-et-ses-amis-au-marrakech-du-rire-p_1316/jamel-et-ses-amis-au-marrakech-du-rire-2015-c_11495320',
- 'md5': '42310bffe4ba3982db112b9cd3467328',
+ IE_NAME = '6play'
+ _VALID_URL = r'(?:6play:|https?://(?:www\.)?(?P<domain>6play\.fr|rtlplay\.be|play\.rtl\.hr|rtlmost\.hu)/.+?-c_)(?P<id>[0-9]+)'
+ _TESTS = [{
+ 'url': 'https://www.6play.fr/minute-par-minute-p_9533/le-but-qui-a-marque-lhistoire-du-football-francais-c_12041051',
+ 'md5': '31fcd112637baa0c2ab92c4fcd8baf27',
'info_dict': {
- 'id': '11495320',
+ 'id': '12041051',
'ext': 'mp4',
- 'title': 'Jamel et ses amis au Marrakech du rire 2015',
- 'description': 'md5:ba2149d5c321d5201b78070ee839d872',
+ 'title': 'Le but qui a marqué l\'histoire du football français !',
+ 'description': 'md5:b59e7e841d646ef1eb42a7868eb6a851',
},
- }
+ }, {
+ 'url': 'https://www.rtlplay.be/rtl-info-13h-p_8551/les-titres-du-rtlinfo-13h-c_12045869',
+ 'only_matching': True,
+ }, {
+ 'url': 'https://play.rtl.hr/pj-masks-p_9455/epizoda-34-sezona-1-catboyevo-cudo-na-dva-kotaca-c_11984989',
+ 'only_matching': True,
+ }, {
+ 'url': 'https://www.rtlmost.hu/megtorve-p_14167/megtorve-6-resz-c_12397787',
+ 'only_matching': True,
+ }]
def _real_extract(self, url):
- video_id = self._match_id(url)
- clip_data = self._download_json(
- 'https://player.m6web.fr/v2/video/config/6play-auth/FR/%s.json' % video_id,
- video_id)
- video_data = clip_data['videoInfo']
+ domain, video_id = re.search(self._VALID_URL, url).groups()
+ service, consumer_name = {
+ '6play.fr': ('6play', 'm6web'),
+ 'rtlplay.be': ('rtlbe_rtl_play', 'rtlbe'),
+ 'play.rtl.hr': ('rtlhr_rtl_play', 'rtlhr'),
+ 'rtlmost.hu': ('rtlhu_rtl_most', 'rtlhu'),
+ }.get(domain, ('6play', 'm6web'))
+
+ data = self._download_json(
+ 'https://pc.middleware.6play.fr/6play/v2/platforms/m6group_web/services/%s/videos/clip_%s' % (service, video_id),
+ video_id, headers={
+ 'x-customer-name': consumer_name
+ }, query={
+ 'csa': 5,
+ 'with': 'clips',
+ })
+
+ clip_data = data['clips'][0]
+ title = clip_data['title']
+ urls = []
quality_key = qualities(['lq', 'sd', 'hq', 'hd'])
formats = []
- for source in clip_data['sources']:
- source_type, source_url = source.get('type'), source.get('src')
- if not source_url or source_type == 'hls/primetime':
+ subtitles = {}
+ assets = clip_data.get('assets') or []
+ for asset in assets:
+ asset_url = asset.get('full_physical_path')
+ protocol = asset.get('protocol')
+ if not asset_url or ((protocol == 'primetime' or asset.get('type') == 'usp_hlsfp_h264') and not ('_drmnp.ism/' in asset_url or '_unpnp.ism/' in asset_url)) or asset_url in urls:
continue
- ext = mimetype2ext(source_type) or determine_ext(source_url)
- if ext == 'm3u8':
- formats.extend(self._extract_m3u8_formats(
- source_url, video_id, 'mp4', 'm3u8_native',
- m3u8_id='hls', fatal=False))
- formats.extend(self._extract_f4m_formats(
- source_url.replace('.m3u8', '.f4m'),
- video_id, f4m_id='hds', fatal=False))
- elif ext == 'mp4':
- quality = source.get('quality')
+ urls.append(asset_url)
+ container = asset.get('video_container')
+ ext = determine_ext(asset_url)
+ if protocol == 'http_subtitle' or ext == 'vtt':
+ subtitles.setdefault('fr', []).append({'url': asset_url})
+ continue
+ if container == 'm3u8' or ext == 'm3u8':
+ if protocol == 'usp':
+ if compat_parse_qs(compat_urllib_parse_urlparse(asset_url).query).get('token', [None])[0]:
+ urlh = self._request_webpage(
+ asset_url, video_id, fatal=False,
+ headers=self.geo_verification_headers())
+ if not urlh:
+ continue
+ asset_url = urlh.geturl()
+ asset_url = asset_url.replace('_drmnp.ism/', '_unpnp.ism/')
+ for i in range(3, 0, -1):
+ asset_url = asset_url = asset_url.replace('_sd1/', '_sd%d/' % i)
+ m3u8_formats = self._extract_m3u8_formats(
+ asset_url, video_id, 'mp4', 'm3u8_native',
+ m3u8_id='hls', fatal=False)
+ formats.extend(m3u8_formats)
+ formats.extend(self._extract_mpd_formats(
+ asset_url.replace('.m3u8', '.mpd'),
+ video_id, mpd_id='dash', fatal=False))
+ if m3u8_formats:
+ break
+ else:
+ formats.extend(self._extract_m3u8_formats(
+ asset_url, video_id, 'mp4', 'm3u8_native',
+ m3u8_id='hls', fatal=False))
+ elif container == 'mp4' or ext == 'mp4':
+ quality = asset.get('video_quality')
formats.append({
- 'url': source_url,
+ 'url': asset_url,
'format_id': quality,
'quality': quality_key(quality),
'ext': ext,
})
self._sort_formats(formats)
+ def get(getter):
+ for src in (data, clip_data):
+ v = try_get(src, getter, compat_str)
+ if v:
+ return v
+
return {
'id': video_id,
- 'title': video_data['title'].strip(),
- 'description': video_data.get('description'),
- 'duration': int_or_none(video_data.get('duration')),
- 'series': video_data.get('titlePgm'),
+ 'title': title,
+ 'description': get(lambda x: x['description']),
+ 'duration': int_or_none(clip_data.get('duration')),
+ 'series': get(lambda x: x['program']['title']),
'formats': formats,
+ 'subtitles': subtitles,
}