summaryrefslogtreecommitdiffstats
path: root/yt_dlp/extractor/islamchannel.py
diff options
context:
space:
mode:
Diffstat (limited to 'yt_dlp/extractor/islamchannel.py')
-rw-r--r--yt_dlp/extractor/islamchannel.py81
1 files changed, 81 insertions, 0 deletions
diff --git a/yt_dlp/extractor/islamchannel.py b/yt_dlp/extractor/islamchannel.py
new file mode 100644
index 0000000..253a846
--- /dev/null
+++ b/yt_dlp/extractor/islamchannel.py
@@ -0,0 +1,81 @@
+import re
+
+from .common import InfoExtractor
+from ..utils import traverse_obj, urljoin
+
+
+class IslamChannelIE(InfoExtractor):
+ _VALID_URL = r'https?://watch\.islamchannel\.tv/watch/(?P<id>\d+)'
+ _TESTS = [{
+ 'url': 'https://watch.islamchannel.tv/watch/38604310',
+ 'info_dict': {
+ 'id': '38604310',
+ 'title': 'Omar - Young Omar',
+ 'description': 'md5:5cc7ddecef064ea7afe52eb5e0e33b55',
+ 'thumbnail': r're:https?://.+',
+ 'ext': 'mp4',
+ }
+ }]
+
+ def _real_extract(self, url):
+ video_id = self._match_id(url)
+ webpage = self._download_webpage(url, video_id)
+
+ thumbnail = self._search_regex(
+ r'data-poster="([^"]+)"', webpage, 'data poster', fatal=False) or \
+ self._html_search_meta(('og:image', 'twitter:image'), webpage)
+
+ headers = {
+ 'Token': self._search_regex(r'data-token="([^"]+)"', webpage, 'data token'),
+ 'Token-Expiry': self._search_regex(r'data-expiry="([^"]+)"', webpage, 'data expiry'),
+ 'Uvid': video_id,
+ }
+ show_stream = self._download_json(
+ f'https://v2-streams-elb.simplestreamcdn.com/api/show/stream/{video_id}', video_id,
+ query={
+ 'key': self._search_regex(r'data-key="([^"]+)"', webpage, 'data key'),
+ 'platform': 'chrome',
+ }, headers=headers)
+ # TODO: show_stream['stream'] and show_stream['drm'] may contain something interesting
+ streams = self._download_json(
+ traverse_obj(show_stream, ('response', 'tokenization', 'url')), video_id,
+ headers=headers)
+ formats, subs = self._extract_m3u8_formats_and_subtitles(traverse_obj(streams, ('Streams', 'Adaptive')), video_id, 'mp4')
+
+ return {
+ 'id': video_id,
+ 'title': self._html_search_meta(('og:title', 'twitter:title'), webpage),
+ 'description': self._html_search_meta(('og:description', 'twitter:description', 'description'), webpage),
+ 'formats': formats,
+ 'subtitles': subs,
+ 'thumbnails': [{
+ 'id': 'unscaled',
+ 'url': thumbnail.split('?')[0],
+ 'ext': 'jpg',
+ 'preference': 2,
+ }, {
+ 'id': 'orig',
+ 'url': thumbnail,
+ 'ext': 'jpg',
+ 'preference': 1,
+ }] if thumbnail else None,
+ }
+
+
+class IslamChannelSeriesIE(InfoExtractor):
+ _VALID_URL = r'https?://watch\.islamchannel\.tv/series/(?P<id>[a-f\d-]+)'
+ _TESTS = [{
+ 'url': 'https://watch.islamchannel.tv/series/a6cccef3-3ef1-11eb-bc19-06b69c2357cd',
+ 'info_dict': {
+ 'id': 'a6cccef3-3ef1-11eb-bc19-06b69c2357cd',
+ },
+ 'playlist_mincount': 31,
+ }]
+
+ def _real_extract(self, url):
+ pl_id = self._match_id(url)
+ webpage = self._download_webpage(url, pl_id)
+
+ return self.playlist_from_matches(
+ re.finditer(r'<a\s+href="(/watch/\d+)"[^>]+?data-video-type="show">', webpage),
+ pl_id, getter=lambda x: urljoin(url, x.group(1)), ie=IslamChannelIE)