mirror of
https://github.com/ytdl-org/youtube-dl
synced 2024-11-27 13:06:49 +01:00
[dvtv] Fix extraction (closes #18514)
This commit is contained in:
parent
9f182c23ba
commit
b9aad6c427
@ -10,7 +10,9 @@ from ..utils import (
|
||||
int_or_none,
|
||||
js_to_json,
|
||||
mimetype2ext,
|
||||
try_get,
|
||||
unescapeHTML,
|
||||
parse_iso8601,
|
||||
)
|
||||
|
||||
|
||||
@ -28,6 +30,8 @@ class DVTVIE(InfoExtractor):
|
||||
'ext': 'mp4',
|
||||
'title': 'Vondra o Českém století: Při pohledu na Havla mi bylo trapně',
|
||||
'duration': 1484,
|
||||
'upload_date': '20141217',
|
||||
'timestamp': 1418792400,
|
||||
}
|
||||
}, {
|
||||
'url': 'http://video.aktualne.cz/dvtv/dvtv-16-12-2014-utok-talibanu-boj-o-kliniku-uprchlici/r~973eb3bc854e11e498be002590604f2e/',
|
||||
@ -84,6 +88,8 @@ class DVTVIE(InfoExtractor):
|
||||
'ext': 'mp4',
|
||||
'title': 'Zeman si jen léčí mindráky, Sobotku nenávidí a Babiš se mu teď hodí, tvrdí Kmenta',
|
||||
'duration': 1103,
|
||||
'upload_date': '20170511',
|
||||
'timestamp': 1494514200,
|
||||
},
|
||||
'params': {
|
||||
'skip_download': True,
|
||||
@ -91,19 +97,38 @@ class DVTVIE(InfoExtractor):
|
||||
}, {
|
||||
'url': 'http://video.aktualne.cz/v-cechach-poprve-zazni-zelenkova-zrestaurovana-mse/r~45b4b00483ec11e4883b002590604f2e/',
|
||||
'only_matching': True,
|
||||
}, {
|
||||
# Test live stream video (liveStarter) parsing
|
||||
'url': 'https://video.aktualne.cz/dvtv/zive-mistryne-sveta-eva-samkova-po-navratu-ze-sampionatu/r~182654c2288811e990fd0cc47ab5f122/',
|
||||
'md5': '2e552e483f2414851ca50467054f9d5d',
|
||||
'info_dict': {
|
||||
'id': '8d116360288011e98c840cc47ab5f122',
|
||||
'ext': 'mp4',
|
||||
'title': 'Živě: Mistryně světa Eva Samková po návratu ze šampionátu',
|
||||
'upload_date': '20190204',
|
||||
'timestamp': 1549289591,
|
||||
},
|
||||
'params': {
|
||||
# Video content is no longer available
|
||||
'skip_download': True,
|
||||
},
|
||||
}]
|
||||
|
||||
def _parse_video_metadata(self, js, video_id, live_js=None):
|
||||
def _parse_video_metadata(self, js, video_id, timestamp):
|
||||
|
||||
data = self._parse_json(js, video_id, transform_source=js_to_json)
|
||||
if live_js:
|
||||
data.update(self._parse_json(
|
||||
live_js, video_id, transform_source=js_to_json))
|
||||
|
||||
live_starter = try_get(data, lambda x: x['plugins']['liveStarter'], dict)
|
||||
if live_starter:
|
||||
data.update(live_starter)
|
||||
|
||||
title = unescapeHTML(data['title'])
|
||||
|
||||
formats = []
|
||||
for video in data['sources']:
|
||||
video_url = video.get('file')
|
||||
|
||||
for tracks in data.get('tracks', {}).values():
|
||||
for video in tracks:
|
||||
video_url = video.get('src')
|
||||
if not video_url:
|
||||
continue
|
||||
video_type = video.get('type')
|
||||
@ -136,7 +161,7 @@ class DVTVIE(InfoExtractor):
|
||||
'description': data.get('description'),
|
||||
'thumbnail': data.get('image'),
|
||||
'duration': int_or_none(data.get('duration')),
|
||||
'timestamp': int_or_none(data.get('pubtime')),
|
||||
'timestamp': int_or_none(timestamp),
|
||||
'formats': formats
|
||||
}
|
||||
|
||||
@ -145,32 +170,33 @@ class DVTVIE(InfoExtractor):
|
||||
|
||||
webpage = self._download_webpage(url, video_id)
|
||||
|
||||
# live content
|
||||
live_item = self._search_regex(
|
||||
r'(?s)embedData[0-9a-f]{32}\.asset\.liveStarter\s*=\s*(\{.+?\});',
|
||||
webpage, 'video', default=None)
|
||||
|
||||
# single video
|
||||
item = self._search_regex(
|
||||
r'(?s)embedData[0-9a-f]{32}\[["\']asset["\']\]\s*=\s*(\{.+?\});',
|
||||
webpage, 'video', default=None)
|
||||
|
||||
if item:
|
||||
return self._parse_video_metadata(item, video_id, live_item)
|
||||
timestamp = parse_iso8601(self._html_search_meta(
|
||||
'article:published_time', webpage, 'published time', default=None))
|
||||
|
||||
# playlist
|
||||
items = re.findall(
|
||||
r"(?s)BBX\.context\.assets\['[0-9a-f]{32}'\]\.push\(({.+?})\);",
|
||||
r"(?s)playlist\.push\(({.+?})\);",
|
||||
webpage)
|
||||
if not items:
|
||||
items = re.findall(r'(?s)var\s+asset\s*=\s*({.+?});\n', webpage)
|
||||
|
||||
if items:
|
||||
return {
|
||||
'_type': 'playlist',
|
||||
'id': video_id,
|
||||
'title': self._og_search_title(webpage),
|
||||
'entries': [self._parse_video_metadata(i, video_id) for i in items]
|
||||
'entries': [self._parse_video_metadata(i, video_id, timestamp) for i in items]
|
||||
}
|
||||
|
||||
# single video
|
||||
item = self._search_regex(
|
||||
r'(?s)BBXPlayer.setup\((.+?)\);',
|
||||
webpage, 'video', default=None)
|
||||
|
||||
if item:
|
||||
# remove function calls (ex. htmldeentitize)
|
||||
# TODO this should be fixed in a general way in the js_to_json
|
||||
item = re.sub(r'\w+?\((.+)\)', r'\1', item)
|
||||
|
||||
if item:
|
||||
return self._parse_video_metadata(item, video_id, timestamp)
|
||||
|
||||
raise ExtractorError('Could not find neither video nor playlist')
|
||||
|
Loading…
Reference in New Issue
Block a user