summaryrefslogtreecommitdiffstats
path: root/yt_dlp/extractor/ruv.py
diff options
context:
space:
mode:
Diffstat (limited to 'yt_dlp/extractor/ruv.py')
-rw-r--r--yt_dlp/extractor/ruv.py186
1 files changed, 186 insertions, 0 deletions
diff --git a/yt_dlp/extractor/ruv.py b/yt_dlp/extractor/ruv.py
new file mode 100644
index 0000000..12499d6
--- /dev/null
+++ b/yt_dlp/extractor/ruv.py
@@ -0,0 +1,186 @@
+from .common import InfoExtractor
+from ..utils import (
+ determine_ext,
+ parse_duration,
+ traverse_obj,
+ unified_timestamp,
+)
+
+
+class RuvIE(InfoExtractor):
+ _VALID_URL = r'https?://(?:www\.)?ruv\.is/(?:sarpurinn/[^/]+|node)/(?P<id>[^/]+(?:/\d+)?)'
+ _TESTS = [{
+ # m3u8
+ 'url': 'http://ruv.is/sarpurinn/ruv-aukaras/fh-valur/20170516',
+ 'md5': '66347652f4e13e71936817102acc1724',
+ 'info_dict': {
+ 'id': '1144499',
+ 'display_id': 'fh-valur/20170516',
+ 'ext': 'mp4',
+ 'title': 'FH - Valur',
+ 'description': 'Bein útsending frá 3. leik FH og Vals í úrslitum Olísdeildar karla í handbolta.',
+ 'timestamp': 1494963600,
+ 'upload_date': '20170516',
+ },
+ }, {
+ # mp3
+ 'url': 'http://ruv.is/sarpurinn/ras-2/morgunutvarpid/20170619',
+ 'md5': '395ea250c8a13e5fdb39d4670ef85378',
+ 'info_dict': {
+ 'id': '1153630',
+ 'display_id': 'morgunutvarpid/20170619',
+ 'ext': 'mp3',
+ 'title': 'Morgunútvarpið',
+ 'description': 'md5:a4cf1202c0a1645ca096b06525915418',
+ 'timestamp': 1497855000,
+ 'upload_date': '20170619',
+ },
+ }, {
+ 'url': 'http://ruv.is/sarpurinn/ruv/frettir/20170614',
+ 'only_matching': True,
+ }, {
+ 'url': 'http://www.ruv.is/node/1151854',
+ 'only_matching': True,
+ }, {
+ 'url': 'http://ruv.is/sarpurinn/klippa/secret-soltice-hefst-a-morgun',
+ 'only_matching': True,
+ }, {
+ 'url': 'http://ruv.is/sarpurinn/ras-1/morgunvaktin/20170619',
+ 'only_matching': True,
+ }]
+
+ def _real_extract(self, url):
+ display_id = self._match_id(url)
+
+ webpage = self._download_webpage(url, display_id)
+
+ title = self._og_search_title(webpage)
+
+ FIELD_RE = r'video\.%s\s*=\s*(["\'])(?P<url>(?:(?!\1).)+)\1'
+
+ media_url = self._html_search_regex(
+ FIELD_RE % 'src', webpage, 'video URL', group='url')
+
+ video_id = self._search_regex(
+ r'<link\b[^>]+\bhref=["\']https?://www\.ruv\.is/node/(\d+)',
+ webpage, 'video id', default=display_id)
+
+ ext = determine_ext(media_url)
+
+ if ext == 'm3u8':
+ formats = self._extract_m3u8_formats(
+ media_url, video_id, 'mp4', entry_protocol='m3u8_native',
+ m3u8_id='hls')
+ elif ext == 'mp3':
+ formats = [{
+ 'format_id': 'mp3',
+ 'url': media_url,
+ 'vcodec': 'none',
+ }]
+ else:
+ formats = [{
+ 'url': media_url,
+ }]
+
+ description = self._og_search_description(webpage, default=None)
+ thumbnail = self._og_search_thumbnail(
+ webpage, default=None) or self._search_regex(
+ FIELD_RE % 'poster', webpage, 'thumbnail', fatal=False)
+ timestamp = unified_timestamp(self._html_search_meta(
+ 'article:published_time', webpage, 'timestamp', fatal=False))
+
+ return {
+ 'id': video_id,
+ 'display_id': display_id,
+ 'title': title,
+ 'description': description,
+ 'thumbnail': thumbnail,
+ 'timestamp': timestamp,
+ 'formats': formats,
+ }
+
+
+class RuvSpilaIE(InfoExtractor):
+ IE_NAME = 'ruv.is:spila'
+ _VALID_URL = r'https?://(?:www\.)?ruv\.is/(?:(?:sjon|ut)varp|(?:krakka|ung)ruv)/spila/.+/(?P<series_id>[0-9]+)/(?P<id>[a-z0-9]+)'
+ _TESTS = [{
+ 'url': 'https://www.ruv.is/sjonvarp/spila/ithrottir/30657/9jcnd4',
+ 'info_dict': {
+ 'id': '9jcnd4',
+ 'ext': 'mp4',
+ 'title': '01.02.2022',
+ 'chapters': 'count:4',
+ 'timestamp': 1643743500,
+ 'upload_date': '20220201',
+ 'thumbnail': 'https://d38kdhuogyllre.cloudfront.net/fit-in/1960x/filters:quality(65)/hd_posters/94boog-iti3jg.jpg',
+ 'description': 'Íþróttafréttir.',
+ 'age_limit': 0,
+ },
+ }, {
+ 'url': 'https://www.ruv.is/utvarp/spila/i-ljosi-sogunnar/23795/7hqkre',
+ 'info_dict': {
+ 'id': '7hqkre',
+ 'ext': 'mp3',
+ 'thumbnail': 'https://d38kdhuogyllre.cloudfront.net/fit-in/1960x/filters:quality(65)/hd_posters/7hqkre-7uepao.jpg',
+ 'description': 'md5:8d7046549daff35e9a3190dc9901a120',
+ 'chapters': [],
+ 'upload_date': '20220204',
+ 'timestamp': 1643965500,
+ 'title': 'Nellie Bly II',
+ 'age_limit': 0,
+ },
+ }, {
+ 'url': 'https://www.ruv.is/ungruv/spila/ungruv/28046/8beuph',
+ 'only_matching': True
+ }, {
+ 'url': 'https://www.ruv.is/krakkaruv/spila/krakkafrettir/30712/9jbgb0',
+ 'only_matching': True
+ }]
+
+ def _real_extract(self, url):
+ display_id, series_id = self._match_valid_url(url).group('id', 'series_id')
+ program = self._download_json(
+ 'https://www.ruv.is/gql/', display_id, query={'query': '''{
+ Program(id: %s){
+ title image description short_description
+ episodes(id: {value: "%s"}) {
+ rating title duration file image firstrun description
+ clips {
+ time text
+ }
+ subtitles {
+ name value
+ }
+ }
+ }
+ }''' % (series_id, display_id)})['data']['Program']
+ episode = program['episodes'][0]
+
+ subs = {}
+ for trk in episode.get('subtitles'):
+ if trk.get('name') and trk.get('value'):
+ subs.setdefault(trk['name'], []).append({'url': trk['value'], 'ext': 'vtt'})
+
+ media_url = episode['file']
+ if determine_ext(media_url) == 'm3u8':
+ formats = self._extract_m3u8_formats(media_url, display_id)
+ else:
+ formats = [{'url': media_url}]
+
+ clips = [
+ {'start_time': parse_duration(c.get('time')), 'title': c.get('text')}
+ for c in episode.get('clips') or []]
+
+ return {
+ 'id': display_id,
+ 'title': traverse_obj(program, ('episodes', 0, 'title'), 'title'),
+ 'description': traverse_obj(
+ program, ('episodes', 0, 'description'), 'description', 'short_description',
+ expected_type=lambda x: x or None),
+ 'subtitles': subs,
+ 'thumbnail': episode.get('image', '').replace('$$IMAGESIZE$$', '1960') or None,
+ 'timestamp': unified_timestamp(episode.get('firstrun')),
+ 'formats': formats,
+ 'age_limit': episode.get('rating'),
+ 'chapters': clips
+ }