0
0
Fork 0
mirror of https://github.com/yt-dlp/yt-dlp.git synced 2024-11-22 02:15:12 +00:00

[lnkgo] fix extraction(closes #16834)

This commit is contained in:
Remita Amine 2019-11-06 23:14:26 +01:00
parent 5d92b407e0
commit d4f53af482

View file

@ -5,24 +5,27 @@
from .common import InfoExtractor
from ..utils import (
clean_html,
compat_str,
int_or_none,
unified_strdate,
parse_iso8601,
)
class LnkGoIE(InfoExtractor):
_VALID_URL = r'https?://(?:www\.)?lnkgo\.(?:alfa\.)?lt/visi-video/(?P<show>[^/]+)/ziurek-(?P<id>[A-Za-z0-9-]+)'
_VALID_URL = r'https?://(?:www\.)?lnk(?:go)?\.(?:alfa\.)?lt/(?:visi-video/[^/]+|video)/(?P<id>[A-Za-z0-9-]+)(?:/(?P<episode_id>\d+))?'
_TESTS = [{
'url': 'http://lnkgo.alfa.lt/visi-video/yra-kaip-yra/ziurek-yra-kaip-yra-162',
'url': 'http://www.lnkgo.lt/visi-video/aktualai-pratesimas/ziurek-putka-trys-klausimai',
'info_dict': {
'id': '46712',
'id': '10809',
'ext': 'mp4',
'title': 'Yra kaip yra',
'upload_date': '20150107',
'description': 'md5:d82a5e36b775b7048617f263a0e3475e',
'age_limit': 7,
'duration': 3019,
'thumbnail': r're:^https?://.*\.jpg$'
'title': "Put'ka: Trys Klausimai",
'upload_date': '20161216',
'description': 'Seniai matytas Putka užduoda tris klausimėlius. Pabandykime surasti atsakymus.',
'age_limit': 18,
'duration': 117,
'thumbnail': r're:^https?://.*\.jpg$',
'timestamp': 1481904000,
},
'params': {
'skip_download': True, # HLS download
@ -30,20 +33,21 @@ class LnkGoIE(InfoExtractor):
}, {
'url': 'http://lnkgo.alfa.lt/visi-video/aktualai-pratesimas/ziurek-nerdas-taiso-kompiuteri-2',
'info_dict': {
'id': '47289',
'id': '10467',
'ext': 'mp4',
'title': 'Nėrdas: Kompiuterio Valymas',
'upload_date': '20150113',
'description': 'md5:7352d113a242a808676ff17e69db6a69',
'age_limit': 18,
'duration': 346,
'thumbnail': r're:^https?://.*\.jpg$'
'thumbnail': r're:^https?://.*\.jpg$',
'timestamp': 1421164800,
},
'params': {
'skip_download': True, # HLS download
},
}, {
'url': 'http://www.lnkgo.lt/visi-video/aktualai-pratesimas/ziurek-putka-trys-klausimai',
'url': 'https://lnk.lt/video/neigalieji-tv-bokste/37413',
'only_matching': True,
}]
_AGE_LIMITS = {
@ -51,66 +55,34 @@ class LnkGoIE(InfoExtractor):
'N-14': 14,
'S': 18,
}
_M3U8_TEMPL = 'https://vod.lnk.lt/lnk_vod/lnk/lnk/%s:%s/playlist.m3u8%s'
def _real_extract(self, url):
display_id = self._match_id(url)
display_id, video_id = re.match(self._VALID_URL, url).groups()
webpage = self._download_webpage(
url, display_id, 'Downloading player webpage')
video_id = self._search_regex(
r'data-ep="([^"]+)"', webpage, 'video ID')
title = self._og_search_title(webpage)
description = self._og_search_description(webpage)
upload_date = unified_strdate(self._search_regex(
r'class="[^"]*meta-item[^"]*air-time[^"]*">.*?<strong>([^<]+)</strong>', webpage, 'upload date', fatal=False))
thumbnail_w = int_or_none(
self._og_search_property('image:width', webpage, 'thumbnail width', fatal=False))
thumbnail_h = int_or_none(
self._og_search_property('image:height', webpage, 'thumbnail height', fatal=False))
thumbnail = {
'url': self._og_search_thumbnail(webpage),
}
if thumbnail_w and thumbnail_h:
thumbnail.update({
'width': thumbnail_w,
'height': thumbnail_h,
})
config = self._parse_json(self._search_regex(
r'episodePlayer\((\{.*?\}),\s*\{', webpage, 'sources'), video_id)
if config.get('pGeo'):
self.report_warning(
'This content might not be available in your country due to copyright reasons')
formats = [{
'format_id': 'hls',
'ext': 'mp4',
'url': config['EpisodeVideoLink_HLS'],
}]
m = re.search(r'^(?P<url>rtmp://[^/]+/(?P<app>[^/]+))/(?P<play_path>.+)$', config['EpisodeVideoLink'])
if m:
formats.append({
'format_id': 'rtmp',
'ext': 'flv',
'url': m.group('url'),
'play_path': m.group('play_path'),
'page_url': url,
})
video_info = self._download_json(
'https://lnk.lt/api/main/video-page/%s/%s/false' % (display_id, video_id or '0'),
display_id)['videoConfig']['videoInfo']
video_id = compat_str(video_info['id'])
title = video_info['title']
prefix = 'smil' if video_info.get('isQualityChangeAvailable') else 'mp4'
formats = self._extract_m3u8_formats(
self._M3U8_TEMPL % (prefix, video_info['videoUrl'], video_info.get('secureTokenParams') or ''),
video_id, 'mp4', 'm3u8_native')
self._sort_formats(formats)
poster_image = video_info.get('posterImage')
return {
'id': video_id,
'display_id': display_id,
'title': title,
'formats': formats,
'thumbnails': [thumbnail],
'duration': int_or_none(config.get('VideoTime')),
'description': description,
'age_limit': self._AGE_LIMITS.get(config.get('PGRating'), 0),
'upload_date': upload_date,
'thumbnail': 'https://lnk.lt/all-images/' + poster_image if poster_image else None,
'duration': int_or_none(video_info.get('duration')),
'description': clean_html(video_info.get('htmlDescription')),
'age_limit': self._AGE_LIMITS.get(video_info.get('pgRating'), 0),
'timestamp': parse_iso8601(video_info.get('airDate')),
'view_count': int_or_none(video_info.get('viewsCount')),
}