mirror of
https://github.com/ytdl-org/youtube-dl.git
synced 2024-12-23 09:17:48 +00:00
140 lines
4.4 KiB
Python
140 lines
4.4 KiB
Python
from __future__ import unicode_literals
|
|
|
|
import re
|
|
|
|
from .common import InfoExtractor
|
|
from ..compat import compat_str
|
|
from ..utils import (
|
|
try_get,
|
|
unescapeHTML,
|
|
url_or_none,
|
|
urljoin,
|
|
)
|
|
|
|
|
|
class WWEBaseIE(InfoExtractor):
|
|
_SUBTITLE_LANGS = {
|
|
'English': 'en',
|
|
'Deutsch': 'de',
|
|
}
|
|
|
|
def _extract_entry(self, data, url, video_id=None):
|
|
video_id = compat_str(video_id or data['nid'])
|
|
title = data['title']
|
|
|
|
formats = self._extract_m3u8_formats(
|
|
data['file'], video_id, 'mp4', entry_protocol='m3u8_native',
|
|
m3u8_id='hls')
|
|
|
|
description = data.get('description')
|
|
thumbnail = urljoin(url, data.get('image'))
|
|
series = data.get('show_name')
|
|
episode = data.get('episode_name')
|
|
|
|
subtitles = {}
|
|
tracks = data.get('tracks')
|
|
if isinstance(tracks, list):
|
|
for track in tracks:
|
|
if not isinstance(track, dict):
|
|
continue
|
|
if track.get('kind') != 'captions':
|
|
continue
|
|
track_file = url_or_none(track.get('file'))
|
|
if not track_file:
|
|
continue
|
|
label = track.get('label')
|
|
lang = self._SUBTITLE_LANGS.get(label, label) or 'en'
|
|
subtitles.setdefault(lang, []).append({
|
|
'url': track_file,
|
|
})
|
|
|
|
return {
|
|
'id': video_id,
|
|
'title': title,
|
|
'description': description,
|
|
'thumbnail': thumbnail,
|
|
'series': series,
|
|
'episode': episode,
|
|
'formats': formats,
|
|
'subtitles': subtitles,
|
|
}
|
|
|
|
|
|
class WWEIE(WWEBaseIE):
|
|
_VALID_URL = r'https?://(?:[^/]+\.)?wwe\.com/(?:[^/]+/)*videos/(?P<id>[^/?#&]+)'
|
|
_TESTS = [{
|
|
'url': 'https://www.wwe.com/videos/daniel-bryan-vs-andrade-cien-almas-smackdown-live-sept-4-2018',
|
|
'md5': '92811c6a14bfc206f7a6a9c5d9140184',
|
|
'info_dict': {
|
|
'id': '40048199',
|
|
'ext': 'mp4',
|
|
'title': 'Daniel Bryan vs. Andrade "Cien" Almas: SmackDown LIVE, Sept. 4, 2018',
|
|
'description': 'md5:2d7424dbc6755c61a0e649d2a8677f67',
|
|
'thumbnail': r're:^https?://.*\.jpg$',
|
|
}
|
|
}, {
|
|
'url': 'https://de.wwe.com/videos/gran-metalik-vs-tony-nese-wwe-205-live-sept-4-2018',
|
|
'only_matching': True,
|
|
}]
|
|
|
|
def _real_extract(self, url):
|
|
display_id = self._match_id(url)
|
|
webpage = self._download_webpage(url, display_id)
|
|
|
|
landing = self._parse_json(
|
|
self._html_search_regex(
|
|
r'(?s)Drupal\.settings\s*,\s*({.+?})\s*\)\s*;',
|
|
webpage, 'drupal settings'),
|
|
display_id)['WWEVideoLanding']
|
|
|
|
data = landing['initialVideo']['playlist'][0]
|
|
video_id = landing.get('initialVideoId')
|
|
|
|
info = self._extract_entry(data, url, video_id)
|
|
info['display_id'] = display_id
|
|
return info
|
|
|
|
|
|
class WWEPlaylistIE(WWEBaseIE):
|
|
_VALID_URL = r'https?://(?:[^/]+\.)?wwe\.com/(?:[^/]+/)*(?P<id>[^/?#&]+)'
|
|
_TESTS = [{
|
|
'url': 'https://www.wwe.com/shows/raw/2018-11-12',
|
|
'info_dict': {
|
|
'id': '2018-11-12',
|
|
},
|
|
'playlist_mincount': 11,
|
|
}, {
|
|
'url': 'http://www.wwe.com/article/walk-the-prank-wwe-edition',
|
|
'only_matching': True,
|
|
}, {
|
|
'url': 'https://www.wwe.com/shows/wwenxt/article/matt-riddle-interview',
|
|
'only_matching': True,
|
|
}]
|
|
|
|
@classmethod
|
|
def suitable(cls, url):
|
|
return False if WWEIE.suitable(url) else super(WWEPlaylistIE, cls).suitable(url)
|
|
|
|
def _real_extract(self, url):
|
|
display_id = self._match_id(url)
|
|
webpage = self._download_webpage(url, display_id)
|
|
|
|
entries = []
|
|
for mobj in re.finditer(
|
|
r'data-video\s*=\s*(["\'])(?P<data>{.+?})\1', webpage):
|
|
video = self._parse_json(
|
|
mobj.group('data'), display_id, transform_source=unescapeHTML,
|
|
fatal=False)
|
|
if not video:
|
|
continue
|
|
data = try_get(video, lambda x: x['playlist'][0], dict)
|
|
if not data:
|
|
continue
|
|
try:
|
|
entry = self._extract_entry(data, url)
|
|
except Exception:
|
|
continue
|
|
entry['extractor_key'] = WWEIE.ie_key()
|
|
entries.append(entry)
|
|
|
|
return self.playlist_result(entries, display_id)
|