mirror of
https://codeberg.org/polarisfm/youtube-dl
synced 2024-11-13 21:34:31 +01:00
144 lines
4.6 KiB
Python
144 lines
4.6 KiB
Python
# coding: utf-8
|
|
from __future__ import unicode_literals
|
|
|
|
import re
|
|
|
|
from .common import InfoExtractor
|
|
|
|
from ..utils import (
|
|
determine_ext,
|
|
float_or_none,
|
|
int_or_none,
|
|
parse_filesize,
|
|
)
|
|
|
|
|
|
class LibraryOfCongressIE(InfoExtractor):
|
|
IE_NAME = 'loc'
|
|
IE_DESC = 'Library of Congress'
|
|
_VALID_URL = r'https?://(?:www\.)?loc\.gov/(?:item/|today/cyberlc/feature_wdesc\.php\?.*\brec=)(?P<id>[0-9]+)'
|
|
_TESTS = [{
|
|
# embedded via <div class="media-player"
|
|
'url': 'http://loc.gov/item/90716351/',
|
|
'md5': '353917ff7f0255aa6d4b80a034833de8',
|
|
'info_dict': {
|
|
'id': '90716351',
|
|
'ext': 'mp4',
|
|
'title': "Pa's trip to Mars",
|
|
'thumbnail': r're:^https?://.*\.jpg$',
|
|
'duration': 0,
|
|
'view_count': int,
|
|
},
|
|
}, {
|
|
# webcast embedded via mediaObjectId
|
|
'url': 'https://www.loc.gov/today/cyberlc/feature_wdesc.php?rec=5578',
|
|
'info_dict': {
|
|
'id': '5578',
|
|
'ext': 'mp4',
|
|
'title': 'Help! Preservation Training Needs Here, There & Everywhere',
|
|
'duration': 3765,
|
|
'view_count': int,
|
|
'subtitles': 'mincount:1',
|
|
},
|
|
'params': {
|
|
'skip_download': True,
|
|
},
|
|
}, {
|
|
# with direct download links
|
|
'url': 'https://www.loc.gov/item/78710669/',
|
|
'info_dict': {
|
|
'id': '78710669',
|
|
'ext': 'mp4',
|
|
'title': 'La vie et la passion de Jesus-Christ',
|
|
'duration': 0,
|
|
'view_count': int,
|
|
'formats': 'mincount:4',
|
|
},
|
|
'params': {
|
|
'skip_download': True,
|
|
},
|
|
}]
|
|
|
|
def _real_extract(self, url):
|
|
video_id = self._match_id(url)
|
|
webpage = self._download_webpage(url, video_id)
|
|
|
|
media_id = self._search_regex(
|
|
(r'id=(["\'])media-player-(?P<id>.+?)\1',
|
|
r'<video[^>]+id=(["\'])uuid-(?P<id>.+?)\1',
|
|
r'<video[^>]+data-uuid=(["\'])(?P<id>.+?)\1',
|
|
r'mediaObjectId\s*:\s*(["\'])(?P<id>.+?)\1'),
|
|
webpage, 'media id', group='id')
|
|
|
|
data = self._download_json(
|
|
'https://media.loc.gov/services/v1/media?id=%s&context=json' % media_id,
|
|
video_id)['mediaObject']
|
|
|
|
derivative = data['derivatives'][0]
|
|
media_url = derivative['derivativeUrl']
|
|
|
|
title = derivative.get('shortName') or data.get('shortName') or self._og_search_title(
|
|
webpage)
|
|
|
|
# Following algorithm was extracted from setAVSource js function
|
|
# found in webpage
|
|
media_url = media_url.replace('rtmp', 'https')
|
|
|
|
is_video = data.get('mediaType', 'v').lower() == 'v'
|
|
ext = determine_ext(media_url)
|
|
if ext not in ('mp4', 'mp3'):
|
|
media_url += '.mp4' if is_video else '.mp3'
|
|
|
|
if 'vod/mp4:' in media_url:
|
|
formats = [{
|
|
'url': media_url.replace('vod/mp4:', 'hls-vod/media/') + '.m3u8',
|
|
'format_id': 'hls',
|
|
'ext': 'mp4',
|
|
'protocol': 'm3u8_native',
|
|
'quality': 1,
|
|
}]
|
|
elif 'vod/mp3:' in media_url:
|
|
formats = [{
|
|
'url': media_url.replace('vod/mp3:', ''),
|
|
'vcodec': 'none',
|
|
}]
|
|
|
|
download_urls = set()
|
|
for m in re.finditer(
|
|
r'<option[^>]+value=(["\'])(?P<url>.+?)\1[^>]+data-file-download=[^>]+>\s*(?P<id>.+?)(?:(?: |\s+)\((?P<size>.+?)\))?\s*<', webpage):
|
|
format_id = m.group('id').lower()
|
|
if format_id == 'gif':
|
|
continue
|
|
download_url = m.group('url')
|
|
if download_url in download_urls:
|
|
continue
|
|
download_urls.add(download_url)
|
|
formats.append({
|
|
'url': download_url,
|
|
'format_id': format_id,
|
|
'filesize_approx': parse_filesize(m.group('size')),
|
|
})
|
|
|
|
self._sort_formats(formats)
|
|
|
|
duration = float_or_none(data.get('duration'))
|
|
view_count = int_or_none(data.get('viewCount'))
|
|
|
|
subtitles = {}
|
|
cc_url = data.get('ccUrl')
|
|
if cc_url:
|
|
subtitles.setdefault('en', []).append({
|
|
'url': cc_url,
|
|
'ext': 'ttml',
|
|
})
|
|
|
|
return {
|
|
'id': video_id,
|
|
'title': title,
|
|
'thumbnail': self._og_search_thumbnail(webpage, default=None),
|
|
'duration': duration,
|
|
'view_count': view_count,
|
|
'formats': formats,
|
|
'subtitles': subtitles,
|
|
}
|