mirror of
https://github.com/yt-dlp/yt-dlp.git
synced 2024-11-24 02:35:12 +00:00
add96eb9f8
Authored by: seproDev Reviewed-by: bashonly <88596187+bashonly@users.noreply.github.com> Reviewed-by: Simon Sawicki <contact@grub4k.xyz>
170 lines
6.4 KiB
Python
170 lines
6.4 KiB
Python
import re
|
|
|
|
from .common import InfoExtractor
|
|
from ..utils import (
|
|
ExtractorError,
|
|
extract_attributes,
|
|
get_element_by_class,
|
|
str_or_none,
|
|
)
|
|
|
|
|
|
class SteamIE(InfoExtractor):
|
|
_VALID_URL = r'''(?x)
|
|
https?://(?:store\.steampowered|steamcommunity)\.com/
|
|
(?:agecheck/)?
|
|
(?P<urltype>video|app)/ #If the page is only for videos or for a game
|
|
(?P<gameID>\d+)/?
|
|
(?P<videoID>\d*)(?P<extra>\??) # For urltype == video we sometimes get the videoID
|
|
|
|
|
https?://(?:www\.)?steamcommunity\.com/sharedfiles/filedetails/\?id=(?P<fileID>[0-9]+)
|
|
'''
|
|
_VIDEO_PAGE_TEMPLATE = 'http://store.steampowered.com/video/%s/'
|
|
_AGECHECK_TEMPLATE = 'http://store.steampowered.com/agecheck/video/%s/?snr=1_agecheck_agecheck__age-gate&ageDay=1&ageMonth=January&ageYear=1970'
|
|
_TESTS = [{
|
|
'url': 'http://store.steampowered.com/video/105600/',
|
|
'playlist': [
|
|
{
|
|
'md5': '695242613303ffa2a4c44c9374ddc067',
|
|
'info_dict': {
|
|
'id': '256785003',
|
|
'ext': 'mp4',
|
|
'title': 'Terraria video 256785003',
|
|
'thumbnail': r're:^https://cdn\.[^\.]+\.steamstatic\.com',
|
|
},
|
|
},
|
|
{
|
|
'md5': '6a294ee0c4b1f47f5bb76a65e31e3592',
|
|
'info_dict': {
|
|
'id': '2040428',
|
|
'ext': 'mp4',
|
|
'title': 'Terraria video 2040428',
|
|
'thumbnail': r're:^https://cdn\.[^\.]+\.steamstatic\.com',
|
|
},
|
|
},
|
|
],
|
|
'info_dict': {
|
|
'id': '105600',
|
|
'title': 'Terraria',
|
|
},
|
|
'params': {
|
|
'playlistend': 2,
|
|
},
|
|
}, {
|
|
'url': 'https://store.steampowered.com/app/271590/Grand_Theft_Auto_V/',
|
|
'info_dict': {
|
|
'id': '271590',
|
|
'title': 'Grand Theft Auto V',
|
|
},
|
|
'playlist_count': 23,
|
|
}]
|
|
|
|
def _real_extract(self, url):
|
|
m = self._match_valid_url(url)
|
|
file_id = m.group('fileID')
|
|
if file_id:
|
|
video_url = url
|
|
playlist_id = file_id
|
|
else:
|
|
game_id = m.group('gameID')
|
|
playlist_id = game_id
|
|
video_url = self._VIDEO_PAGE_TEMPLATE % playlist_id
|
|
|
|
self._set_cookie('steampowered.com', 'wants_mature_content', '1')
|
|
self._set_cookie('steampowered.com', 'birthtime', '944006401')
|
|
self._set_cookie('steampowered.com', 'lastagecheckage', '1-0-2000')
|
|
|
|
webpage = self._download_webpage(video_url, playlist_id)
|
|
|
|
if re.search('<div[^>]+>Please enter your birth date to continue:</div>', webpage) is not None:
|
|
video_url = self._AGECHECK_TEMPLATE % playlist_id
|
|
self.report_age_confirmation()
|
|
webpage = self._download_webpage(video_url, playlist_id)
|
|
|
|
videos = re.findall(r'(<div[^>]+id=[\'"]highlight_movie_(\d+)[\'"][^>]+>)', webpage)
|
|
entries = []
|
|
playlist_title = get_element_by_class('apphub_AppName', webpage)
|
|
for movie, movie_id in videos:
|
|
if not movie:
|
|
continue
|
|
movie = extract_attributes(movie)
|
|
if not movie_id:
|
|
continue
|
|
entry = {
|
|
'id': movie_id,
|
|
'title': f'{playlist_title} video {movie_id}',
|
|
}
|
|
formats = []
|
|
if movie:
|
|
entry['thumbnail'] = movie.get('data-poster')
|
|
for quality in ('', '-hd'):
|
|
for ext in ('webm', 'mp4'):
|
|
video_url = movie.get(f'data-{ext}{quality}-source')
|
|
if video_url:
|
|
formats.append({
|
|
'format_id': ext + quality,
|
|
'url': video_url,
|
|
})
|
|
entry['formats'] = formats
|
|
entries.append(entry)
|
|
embedded_videos = re.findall(r'(<iframe[^>]+>)', webpage)
|
|
for evideos in embedded_videos:
|
|
evideos = extract_attributes(evideos).get('src')
|
|
video_id = self._search_regex(r'youtube\.com/embed/([0-9A-Za-z_-]{11})', evideos, 'youtube_video_id', default=None)
|
|
if video_id:
|
|
entries.append({
|
|
'_type': 'url_transparent',
|
|
'id': video_id,
|
|
'url': video_id,
|
|
'ie_key': 'Youtube',
|
|
})
|
|
if not entries:
|
|
raise ExtractorError('Could not find any videos')
|
|
|
|
return self.playlist_result(entries, playlist_id, playlist_title)
|
|
|
|
|
|
class SteamCommunityBroadcastIE(InfoExtractor):
|
|
_VALID_URL = r'https?://steamcommunity\.(?:com)/broadcast/watch/(?P<id>\d+)'
|
|
_TESTS = [{
|
|
'url': 'https://steamcommunity.com/broadcast/watch/76561199073851486',
|
|
'info_dict': {
|
|
'id': '76561199073851486',
|
|
'title': r're:Steam Community :: pepperm!nt :: Broadcast 2022-06-26 \d{2}:\d{2}',
|
|
'ext': 'mp4',
|
|
'uploader_id': '1113585758',
|
|
'uploader': 'pepperm!nt',
|
|
'live_status': 'is_live',
|
|
},
|
|
'skip': 'Stream has ended',
|
|
}]
|
|
|
|
def _real_extract(self, url):
|
|
video_id = self._match_id(url)
|
|
webpage = self._download_webpage(url, video_id)
|
|
json_data = self._download_json(
|
|
'https://steamcommunity.com/broadcast/getbroadcastmpd/',
|
|
video_id, query={'steamid': f'{video_id}'})
|
|
|
|
formats, subs = self._extract_m3u8_formats_and_subtitles(json_data['hls_url'], video_id)
|
|
|
|
''' # We cannot download live dash atm
|
|
mpd_formats, mpd_subs = self._extract_mpd_formats_and_subtitles(json_data['url'], video_id)
|
|
formats.extend(mpd_formats)
|
|
self._merge_subtitles(mpd_subs, target=subs)
|
|
'''
|
|
|
|
uploader_json = self._download_json(
|
|
'https://steamcommunity.com/actions/ajaxresolveusers',
|
|
video_id, query={'steamids': video_id})[0]
|
|
|
|
return {
|
|
'id': video_id,
|
|
'title': self._generic_title('', webpage),
|
|
'formats': formats,
|
|
'live_status': 'is_live',
|
|
'view_count': json_data.get('num_view'),
|
|
'uploader': uploader_json.get('persona_name'),
|
|
'uploader_id': str_or_none(uploader_json.get('accountid')),
|
|
'subtitles': subs,
|
|
}
|