diff options
author | Daniel Baumann <daniel.baumann@progress-linux.org> | 2024-08-05 09:06:10 +0000 |
---|---|---|
committer | Daniel Baumann <daniel.baumann@progress-linux.org> | 2024-08-05 09:06:10 +0000 |
commit | 7e21328585afda6d66f98ca476301680eeffac32 (patch) | |
tree | 266a4e7b20443f94572748781d71fc0375a15037 /yt_dlp/extractor/rtnews.py | |
parent | Adding upstream version 2024.05.27. (diff) | |
download | yt-dlp-7e21328585afda6d66f98ca476301680eeffac32.tar.xz yt-dlp-7e21328585afda6d66f98ca476301680eeffac32.zip |
Adding upstream version 2024.07.01.upstream/2024.07.01
Signed-off-by: Daniel Baumann <daniel.baumann@progress-linux.org>
Diffstat (limited to 'yt_dlp/extractor/rtnews.py')
-rw-r--r-- | yt_dlp/extractor/rtnews.py | 60 |
1 files changed, 30 insertions, 30 deletions
diff --git a/yt_dlp/extractor/rtnews.py b/yt_dlp/extractor/rtnews.py index 6be9945..558511f 100644 --- a/yt_dlp/extractor/rtnews.py +++ b/yt_dlp/extractor/rtnews.py @@ -14,7 +14,7 @@ class RTNewsIE(InfoExtractor): 'id': '546301', 'title': 'Crowds gather to greet deported Djokovic as he returns to Serbia (VIDEO)', 'description': 'md5:1d5bfe1a988d81fd74227cfdf93d314d', - 'thumbnail': 'https://cdni.rt.com/files/2022.01/article/61e587a085f540102c3386c1.png' + 'thumbnail': 'https://cdni.rt.com/files/2022.01/article/61e587a085f540102c3386c1.png', }, }, { 'url': 'https://www.rt.com/shows/in-question/535980-plot-to-assassinate-julian-assange/', @@ -23,7 +23,7 @@ class RTNewsIE(InfoExtractor): 'id': '535980', 'title': 'The plot to assassinate Julian Assange', 'description': 'md5:55279ce5e4441dc1d16e2e4a730152cd', - 'thumbnail': 'https://cdni.rt.com/files/2021.09/article/615226f42030274e8879b53d.png' + 'thumbnail': 'https://cdni.rt.com/files/2021.09/article/615226f42030274e8879b53d.png', }, 'playlist': [{ 'info_dict': { @@ -31,7 +31,7 @@ class RTNewsIE(InfoExtractor): 'ext': 'mp4', 'title': '6152271d85f5400464496162', }, - }] + }], }] def _entries(self, webpage): @@ -46,12 +46,12 @@ class RTNewsIE(InfoExtractor): } def _real_extract(self, url): - id = self._match_id(url) - webpage = self._download_webpage(url, id) + playlist_id = self._match_id(url) + webpage = self._download_webpage(url, playlist_id) return { '_type': 'playlist', - 'id': id, + 'id': playlist_id, 'entries': self._entries(webpage), 'title': self._og_search_title(webpage), 'description': self._og_search_description(webpage), @@ -71,9 +71,9 @@ class RTDocumentryIE(InfoExtractor): 'description': 'md5:647c76984b7cb9a8b52a567e87448d88', 'thumbnail': 'https://cdni.rt.com/rtd-files/films/escobars-hitman/escobars-hitman_11.jpg', 'average_rating': 8.53, - 'duration': 3134.0 + 'duration': 3134.0, }, - 'params': {'skip_download': True} + 'params': {'skip_download': True}, }, { 'url': 'https://rtd.rt.com/shows/the-kalashnikova-show-military-secrets-anna-knishenko/iskander-tactical-system-natos-headache/', 'info_dict': { @@ -86,9 +86,9 @@ class RTDocumentryIE(InfoExtractor): 'duration': 274.0, 'timestamp': 1605726000, 'view_count': int, - 'upload_date': '20201118' + 'upload_date': '20201118', }, - 'params': {'skip_download': True} + 'params': {'skip_download': True}, }, { 'url': 'https://rtd.rt.com/series/i-am-hacked-trailer/introduction-to-safe-digital-life-ep2/', 'info_dict': { @@ -101,29 +101,29 @@ class RTDocumentryIE(InfoExtractor): 'duration': 1524.0, 'timestamp': 1636977600, 'view_count': int, - 'upload_date': '20211115' + 'upload_date': '20211115', }, - 'params': {'skip_download': True} + 'params': {'skip_download': True}, }] def _real_extract(self, url): - id = self._match_id(url) - webpage = self._download_webpage(url, id) + video_id = self._match_id(url) + webpage = self._download_webpage(url, video_id) ld_json = self._search_json_ld(webpage, None, fatal=False) if not ld_json: self.raise_no_formats('No video/audio found at the provided url.', expected=True) media_json = self._parse_json( self._search_regex(r'(?s)\'Med\'\s*:\s*\[\s*({.+})\s*\]\s*};', webpage, 'media info'), - id, transform_source=js_to_json) + video_id, transform_source=js_to_json) if 'title' not in ld_json and 'title' in media_json: ld_json['title'] = media_json['title'] formats = [{'url': src['file']} for src in media_json.get('sources') or [] if src.get('file')] return { - 'id': id, + 'id': video_id, 'thumbnail': media_json.get('image'), 'formats': formats, - **ld_json + **ld_json, } @@ -144,23 +144,23 @@ class RTDocumentryPlaylistIE(InfoExtractor): }, }] - def _entries(self, webpage, id): + def _entries(self, webpage, playlist_id): video_urls = set(re.findall(r'list-2__link\s*"\s*href="([^"]+)"', webpage)) for v_url in video_urls: - if id not in v_url: + if playlist_id not in v_url: continue yield self.url_result( - 'https://rtd.rt.com%s' % v_url, + f'https://rtd.rt.com{v_url}', ie=RTDocumentryIE.ie_key()) def _real_extract(self, url): - id = self._match_id(url) - webpage = self._download_webpage(url, id) + playlist_id = self._match_id(url) + webpage = self._download_webpage(url, playlist_id) return { '_type': 'playlist', - 'id': id, - 'entries': self._entries(webpage, id), + 'id': playlist_id, + 'entries': self._entries(webpage, playlist_id), } @@ -174,20 +174,20 @@ class RuptlyIE(InfoExtractor): 'ext': 'mp4', 'title': 'Japan: Double trouble! Tokyo zoo presents adorable panda twins | Video Ruptly', 'description': 'md5:85a8da5fdb31486f0562daf4360ce75a', - 'thumbnail': 'https://storage.ruptly.tv/thumbnails/20220112-020/i6JQKnTNpYuqaXsR/i6JQKnTNpYuqaXsR.jpg' + 'thumbnail': 'https://storage.ruptly.tv/thumbnails/20220112-020/i6JQKnTNpYuqaXsR/i6JQKnTNpYuqaXsR.jpg', }, - 'params': {'skip_download': True} + 'params': {'skip_download': True}, }] def _real_extract(self, url): - id = self._match_id(url) - webpage = self._download_webpage(url, id) + video_id = self._match_id(url) + webpage = self._download_webpage(url, video_id) m3u8_url = self._search_regex(r'preview_url"\s?:\s?"(https?://storage\.ruptly\.tv/video_projects/.+\.m3u8)"', webpage, 'm3u8 url', fatal=False) if not m3u8_url: self.raise_no_formats('No video/audio found at the provided url.', expected=True) - formats, subs = self._extract_m3u8_formats_and_subtitles(m3u8_url, id, ext='mp4') + formats, subs = self._extract_m3u8_formats_and_subtitles(m3u8_url, video_id, ext='mp4') return { - 'id': id, + 'id': video_id, 'formats': formats, 'subtitles': subs, 'title': self._og_search_title(webpage), |