mirror of
https://github.com/yt-dlp/yt-dlp.git
synced 2024-12-05 02:53:00 +01:00
add96eb9f8
Authored by: seproDev Reviewed-by: bashonly <88596187+bashonly@users.noreply.github.com> Reviewed-by: Simon Sawicki <contact@grub4k.xyz>
180 lines
7.2 KiB
Python
180 lines
7.2 KiB
Python
import re
|
|
|
|
from .common import InfoExtractor
|
|
from ..utils import (
|
|
clean_html,
|
|
determine_ext,
|
|
get_element_by_class,
|
|
int_or_none,
|
|
merge_dicts,
|
|
parse_bitrate,
|
|
parse_resolution,
|
|
remove_end,
|
|
str_or_none,
|
|
url_or_none,
|
|
urlencode_postdata,
|
|
)
|
|
|
|
|
|
class IcareusIE(InfoExtractor):
|
|
_DOMAINS = '|'.join(map(re.escape, (
|
|
'asahitv.fi',
|
|
'helsinkikanava.fi',
|
|
'hyvinvointitv.fi',
|
|
'inez.fi',
|
|
'permanto.fi',
|
|
'suite.icareus.com',
|
|
'videos.minifiddlers.org',
|
|
)))
|
|
_VALID_URL = rf'(?P<base_url>https?://(?:www\.)?(?:{_DOMAINS}))/[^?#]+/player/[^?#]+\?(?:[^#]+&)?(?:assetId|eventId)=(?P<id>\d+)'
|
|
_TESTS = [{
|
|
'url': 'https://www.helsinkikanava.fi/fi_FI/web/helsinkikanava/player/vod?assetId=68021894',
|
|
'md5': 'ca0b62ffc814a5411dfa6349cf5adb8a',
|
|
'info_dict': {
|
|
'id': '68021894',
|
|
'ext': 'mp4',
|
|
'title': 'Perheiden parhaaksi',
|
|
'description': 'md5:295785ea408e5ac00708766465cc1325',
|
|
'thumbnail': 'https://www.helsinkikanava.fi/image/image_gallery?img_id=68022501',
|
|
'upload_date': '20200924',
|
|
'timestamp': 1600938300,
|
|
},
|
|
}, { # Recorded livestream
|
|
'url': 'https://www.helsinkikanava.fi/fi/web/helsinkikanava/player/event/view?eventId=76241489',
|
|
'md5': '014327e69dfa7b949fcc861f6d162d6d',
|
|
'info_dict': {
|
|
'id': '76258304',
|
|
'ext': 'mp4',
|
|
'title': 'Helsingin kaupungin ja HUSin tiedotustilaisuus koronaepidemiatilanteesta 24.11.2020',
|
|
'description': 'md5:3129d041c6fbbcdc7fe68d9a938fef1c',
|
|
'thumbnail': 'https://icareus-suite.secure2.footprint.net/image/image_gallery?img_id=76288630',
|
|
'upload_date': '20201124',
|
|
'timestamp': 1606206600,
|
|
},
|
|
}, { # Non-m3u8 stream
|
|
'url': 'https://suite.icareus.com/fi/web/westend-indians/player/vod?assetId=47567389',
|
|
'md5': '72fc04ee971bbedc44405cdf16c990b6',
|
|
'info_dict': {
|
|
'id': '47567389',
|
|
'ext': 'mp4',
|
|
'title': 'Omatoiminen harjoittelu - Laukominen',
|
|
'description': '',
|
|
'thumbnail': 'https://suite.icareus.com/image/image_gallery?img_id=47568162',
|
|
'upload_date': '20200319',
|
|
'timestamp': 1584658080,
|
|
},
|
|
}, {
|
|
'url': 'https://asahitv.fi/fi/web/asahi/player/vod?assetId=89415818',
|
|
'only_matching': True,
|
|
}, {
|
|
'url': 'https://hyvinvointitv.fi/fi/web/hyvinvointitv/player/vod?assetId=89149730',
|
|
'only_matching': True,
|
|
}, {
|
|
'url': 'https://inez.fi/fi/web/inez-media/player/vod?assetId=71328822',
|
|
'only_matching': True,
|
|
}, {
|
|
'url': 'https://www.permanto.fi/fi/web/alfatv/player/vod?assetId=135497515',
|
|
'only_matching': True,
|
|
}, {
|
|
'url': 'https://videos.minifiddlers.org/web/international-minifiddlers/player/vod?assetId=1982759',
|
|
'only_matching': True,
|
|
}]
|
|
|
|
def _real_extract(self, url):
|
|
base_url, temp_id = self._match_valid_url(url).groups()
|
|
webpage = self._download_webpage(url, temp_id)
|
|
|
|
video_id = self._search_regex(r"_icareus\['itemId'\]\s*=\s*'(\d+)'", webpage, 'video_id')
|
|
organization_id = self._search_regex(r"_icareus\['organizationId'\]\s*=\s*'(\d+)'", webpage, 'organization_id')
|
|
|
|
assets = self._download_json(
|
|
self._search_regex(r'var\s+publishingServiceURL\s*=\s*"(http[^"]+)";', webpage, 'api_base'),
|
|
video_id, data=urlencode_postdata({
|
|
'version': '03',
|
|
'action': 'getAssetPlaybackUrls',
|
|
'organizationId': organization_id,
|
|
'assetId': video_id,
|
|
'token': self._search_regex(r"_icareus\['token'\]\s*=\s*'([a-f0-9]+)'", webpage, 'icareus_token'),
|
|
}))
|
|
|
|
subtitles = {
|
|
remove_end(sdesc.split(' ')[0], ':'): [{'url': url_or_none(surl)}]
|
|
for _, sdesc, surl in assets.get('subtitles') or []
|
|
}
|
|
|
|
formats = [{
|
|
'format': item.get('name'),
|
|
'format_id': 'audio',
|
|
'vcodec': 'none',
|
|
'url': url_or_none(item['url']),
|
|
'tbr': int_or_none(self._search_regex(
|
|
r'\((\d+)\s*k\)', item.get('name') or '', 'audio bitrate', default=None)),
|
|
} for item in assets.get('audio_urls') or [] if url_or_none(item.get('url'))]
|
|
|
|
for item in assets.get('urls') or []:
|
|
video_url = url_or_none(item.get('url'))
|
|
if video_url is None:
|
|
continue
|
|
ext = determine_ext(video_url)
|
|
if ext == 'm3u8':
|
|
fmts, subs = self._extract_m3u8_formats_and_subtitles(
|
|
video_url, video_id, 'mp4', m3u8_id='hls', fatal=False)
|
|
formats.extend(fmts)
|
|
self._merge_subtitles(subs, target=subtitles)
|
|
else:
|
|
fmt = item.get('name')
|
|
formats.append({
|
|
'url': video_url,
|
|
'format': fmt,
|
|
'tbr': parse_bitrate(fmt),
|
|
'format_id': str_or_none(item.get('id')),
|
|
**parse_resolution(fmt),
|
|
})
|
|
|
|
info, token, live_title = self._search_json_ld(webpage, video_id, default={}), None, None
|
|
if not info:
|
|
token = self._search_regex(
|
|
r'data\s*:\s*{action:"getAsset".*?token:\'([a-f0-9]+)\'}', webpage, 'token', default=None)
|
|
if not token:
|
|
live_title = get_element_by_class('unpublished-info-item future-event-title', webpage)
|
|
|
|
if token:
|
|
metadata = self._download_json(
|
|
f'{base_url}/icareus-suite-api-portlet/publishing',
|
|
video_id, fatal=False, data=urlencode_postdata({
|
|
'version': '03',
|
|
'action': 'getAsset',
|
|
'organizationId': organization_id,
|
|
'assetId': video_id,
|
|
'languageId': 'en_US',
|
|
'userId': '0',
|
|
'token': token,
|
|
})) or {}
|
|
info = {
|
|
'title': metadata.get('name'),
|
|
'description': metadata.get('description'),
|
|
'timestamp': int_or_none(metadata.get('date'), scale=1000),
|
|
'duration': int_or_none(metadata.get('duration')),
|
|
'thumbnail': url_or_none(metadata.get('thumbnailMedium')),
|
|
}
|
|
elif live_title: # Recorded livestream
|
|
info = {
|
|
'title': live_title,
|
|
'description': get_element_by_class('unpublished-info-item future-event-description', webpage),
|
|
'timestamp': int_or_none(self._search_regex(
|
|
r'var startEvent\s*=\s*(\d+);', webpage, 'uploadDate', fatal=False), scale=1000),
|
|
}
|
|
|
|
thumbnails = info.get('thumbnails') or [{
|
|
'url': url_or_none(info.get('thumbnail') or assets.get('thumbnail')),
|
|
}]
|
|
|
|
return merge_dicts({
|
|
'id': video_id,
|
|
'title': None,
|
|
'formats': formats,
|
|
'subtitles': subtitles,
|
|
'description': clean_html(info.get('description')),
|
|
'thumbnails': thumbnails if thumbnails[0]['url'] else None,
|
|
}, info)
|