mirror of https://github.com/yt-dlp/yt-dlp.git
[extractor/duboku] Fix for hostname change (#3891)
Authored by: mozbugbox
This commit is contained in:
parent
2c6dcb65fb
commit
5bbe631e04
|
@ -51,31 +51,39 @@ def _get_element_by_tag_and_attrib(html, tag=None, attribute=None, value=None, e
|
||||||
|
|
||||||
class DubokuIE(InfoExtractor):
|
class DubokuIE(InfoExtractor):
|
||||||
IE_NAME = 'duboku'
|
IE_NAME = 'duboku'
|
||||||
IE_DESC = 'www.duboku.co'
|
IE_DESC = 'www.duboku.io'
|
||||||
|
|
||||||
_VALID_URL = r'(?:https?://[^/]+\.duboku\.co/vodplay/)(?P<id>[0-9]+-[0-9-]+)\.html.*'
|
_VALID_URL = r'(?:https?://[^/]+\.duboku\.io/vodplay/)(?P<id>[0-9]+-[0-9-]+)\.html.*'
|
||||||
_TESTS = [{
|
_TESTS = [{
|
||||||
'url': 'https://www.duboku.co/vodplay/1575-1-1.html',
|
'url': 'https://w.duboku.io/vodplay/1575-1-1.html',
|
||||||
'info_dict': {
|
'info_dict': {
|
||||||
'id': '1575-1-1',
|
'id': '1575-1-1',
|
||||||
'ext': 'ts',
|
'ext': 'mp4',
|
||||||
'series': '白色月光',
|
'series': '白色月光',
|
||||||
'title': 'contains:白色月光',
|
'title': 'contains:白色月光',
|
||||||
'season_number': 1,
|
'season_number': 1,
|
||||||
'episode_number': 1,
|
'episode_number': 1,
|
||||||
|
'season': 'Season 1',
|
||||||
|
'episode_id': '1',
|
||||||
|
'season_id': '1',
|
||||||
|
'episode': 'Episode 1',
|
||||||
},
|
},
|
||||||
'params': {
|
'params': {
|
||||||
'skip_download': 'm3u8 download',
|
'skip_download': 'm3u8 download',
|
||||||
},
|
},
|
||||||
}, {
|
}, {
|
||||||
'url': 'https://www.duboku.co/vodplay/1588-1-1.html',
|
'url': 'https://w.duboku.io/vodplay/1588-1-1.html',
|
||||||
'info_dict': {
|
'info_dict': {
|
||||||
'id': '1588-1-1',
|
'id': '1588-1-1',
|
||||||
'ext': 'ts',
|
'ext': 'mp4',
|
||||||
'series': '亲爱的自己',
|
'series': '亲爱的自己',
|
||||||
'title': 'contains:预告片',
|
'title': 'contains:第1集',
|
||||||
'season_number': 1,
|
'season_number': 1,
|
||||||
'episode_number': 1,
|
'episode_number': 1,
|
||||||
|
'episode': 'Episode 1',
|
||||||
|
'season': 'Season 1',
|
||||||
|
'episode_id': '1',
|
||||||
|
'season_id': '1',
|
||||||
},
|
},
|
||||||
'params': {
|
'params': {
|
||||||
'skip_download': 'm3u8 download',
|
'skip_download': 'm3u8 download',
|
||||||
|
@ -91,7 +99,7 @@ class DubokuIE(InfoExtractor):
|
||||||
season_id = temp[1]
|
season_id = temp[1]
|
||||||
episode_id = temp[2]
|
episode_id = temp[2]
|
||||||
|
|
||||||
webpage_url = 'https://www.duboku.co/vodplay/%s.html' % video_id
|
webpage_url = 'https://w.duboku.io/vodplay/%s.html' % video_id
|
||||||
webpage_html = self._download_webpage(webpage_url, video_id)
|
webpage_html = self._download_webpage(webpage_url, video_id)
|
||||||
|
|
||||||
# extract video url
|
# extract video url
|
||||||
|
@ -124,12 +132,13 @@ class DubokuIE(InfoExtractor):
|
||||||
data_from = player_data.get('from')
|
data_from = player_data.get('from')
|
||||||
|
|
||||||
# if it is an embedded iframe, maybe it's an external source
|
# if it is an embedded iframe, maybe it's an external source
|
||||||
|
headers = {'Referer': webpage_url}
|
||||||
if data_from == 'iframe':
|
if data_from == 'iframe':
|
||||||
# use _type url_transparent to retain the meaningful details
|
# use _type url_transparent to retain the meaningful details
|
||||||
# of the video.
|
# of the video.
|
||||||
return {
|
return {
|
||||||
'_type': 'url_transparent',
|
'_type': 'url_transparent',
|
||||||
'url': smuggle_url(data_url, {'http_headers': {'Referer': webpage_url}}),
|
'url': smuggle_url(data_url, {'http_headers': headers}),
|
||||||
'id': video_id,
|
'id': video_id,
|
||||||
'title': title,
|
'title': title,
|
||||||
'series': series_title,
|
'series': series_title,
|
||||||
|
@ -139,7 +148,7 @@ class DubokuIE(InfoExtractor):
|
||||||
'episode_id': episode_id,
|
'episode_id': episode_id,
|
||||||
}
|
}
|
||||||
|
|
||||||
formats = self._extract_m3u8_formats(data_url, video_id, 'mp4')
|
formats = self._extract_m3u8_formats(data_url, video_id, 'mp4', headers=headers)
|
||||||
|
|
||||||
return {
|
return {
|
||||||
'id': video_id,
|
'id': video_id,
|
||||||
|
@ -150,36 +159,29 @@ class DubokuIE(InfoExtractor):
|
||||||
'episode_number': int_or_none(episode_id),
|
'episode_number': int_or_none(episode_id),
|
||||||
'episode_id': episode_id,
|
'episode_id': episode_id,
|
||||||
'formats': formats,
|
'formats': formats,
|
||||||
'http_headers': {'Referer': 'https://www.duboku.co/static/player/videojs.html'}
|
'http_headers': headers
|
||||||
}
|
}
|
||||||
|
|
||||||
|
|
||||||
class DubokuPlaylistIE(InfoExtractor):
|
class DubokuPlaylistIE(InfoExtractor):
|
||||||
IE_NAME = 'duboku:list'
|
IE_NAME = 'duboku:list'
|
||||||
IE_DESC = 'www.duboku.co entire series'
|
IE_DESC = 'www.duboku.io entire series'
|
||||||
|
|
||||||
_VALID_URL = r'(?:https?://[^/]+\.duboku\.co/voddetail/)(?P<id>[0-9]+)\.html.*'
|
_VALID_URL = r'(?:https?://[^/]+\.duboku\.io/voddetail/)(?P<id>[0-9]+)\.html.*'
|
||||||
_TESTS = [{
|
_TESTS = [{
|
||||||
'url': 'https://www.duboku.co/voddetail/1575.html',
|
'url': 'https://w.duboku.io/voddetail/1575.html',
|
||||||
'info_dict': {
|
'info_dict': {
|
||||||
'id': 'startswith:1575',
|
'id': 'startswith:1575',
|
||||||
'title': '白色月光',
|
'title': '白色月光',
|
||||||
},
|
},
|
||||||
'playlist_count': 12,
|
'playlist_count': 12,
|
||||||
}, {
|
}, {
|
||||||
'url': 'https://www.duboku.co/voddetail/1554.html',
|
'url': 'https://w.duboku.io/voddetail/1554.html',
|
||||||
'info_dict': {
|
'info_dict': {
|
||||||
'id': 'startswith:1554',
|
'id': 'startswith:1554',
|
||||||
'title': '以家人之名',
|
'title': '以家人之名',
|
||||||
},
|
},
|
||||||
'playlist_mincount': 30,
|
'playlist_mincount': 30,
|
||||||
}, {
|
|
||||||
'url': 'https://www.duboku.co/voddetail/1554.html#playlist2',
|
|
||||||
'info_dict': {
|
|
||||||
'id': '1554#playlist2',
|
|
||||||
'title': '以家人之名',
|
|
||||||
},
|
|
||||||
'playlist_mincount': 27,
|
|
||||||
}]
|
}]
|
||||||
|
|
||||||
def _real_extract(self, url):
|
def _real_extract(self, url):
|
||||||
|
@ -189,7 +191,7 @@ class DubokuPlaylistIE(InfoExtractor):
|
||||||
series_id = mobj.group('id')
|
series_id = mobj.group('id')
|
||||||
fragment = compat_urlparse.urlparse(url).fragment
|
fragment = compat_urlparse.urlparse(url).fragment
|
||||||
|
|
||||||
webpage_url = 'https://www.duboku.co/voddetail/%s.html' % series_id
|
webpage_url = 'https://w.duboku.io/voddetail/%s.html' % series_id
|
||||||
webpage_html = self._download_webpage(webpage_url, series_id)
|
webpage_html = self._download_webpage(webpage_url, series_id)
|
||||||
|
|
||||||
# extract title
|
# extract title
|
||||||
|
@ -234,6 +236,6 @@ class DubokuPlaylistIE(InfoExtractor):
|
||||||
# return url results
|
# return url results
|
||||||
return self.playlist_result([
|
return self.playlist_result([
|
||||||
self.url_result(
|
self.url_result(
|
||||||
compat_urlparse.urljoin('https://www.duboku.co', x['href']),
|
compat_urlparse.urljoin('https://w.duboku.io', x['href']),
|
||||||
ie=DubokuIE.ie_key(), video_title=x.get('title'))
|
ie=DubokuIE.ie_key(), video_title=x.get('title'))
|
||||||
for x in playlist], series_id + '#' + playlist_id, title)
|
for x in playlist], series_id + '#' + playlist_id, title)
|
||||||
|
|
Loading…
Reference in New Issue