mirror of
https://github.com/yt-dlp/yt-dlp.git
synced 2024-11-04 23:43:20 +00:00
705f5b84de
Closes #10418 Authored by: seproDev
119 lines
4.7 KiB
Python
119 lines
4.7 KiB
Python
import json
|
|
import urllib.parse
|
|
|
|
from .common import InfoExtractor
|
|
from ..utils import (
|
|
ExtractorError,
|
|
parse_iso8601,
|
|
update_url_query,
|
|
url_or_none,
|
|
)
|
|
from ..utils.traversal import traverse_obj
|
|
|
|
|
|
class BoxIE(InfoExtractor):
|
|
_VALID_URL = r'https?://(?:[^.]+\.)?(?P<service>app|ent)\.box\.com/s/(?P<shared_name>[^/?#]+)(?:/file/(?P<id>\d+))?'
|
|
_TESTS = [{
|
|
'url': 'https://mlssoccer.app.box.com/s/0evd2o3e08l60lr4ygukepvnkord1o1x/file/510727257538',
|
|
'md5': '1f81b2fd3960f38a40a3b8823e5fcd43',
|
|
'info_dict': {
|
|
'id': '510727257538',
|
|
'ext': 'mp4',
|
|
'title': 'Garber St. Louis will be 28th MLS team +scarving.mp4',
|
|
'uploader': '',
|
|
'timestamp': 1566320259,
|
|
'upload_date': '20190820',
|
|
'uploader_id': '235196876',
|
|
},
|
|
'params': {'skip_download': 'dash fragment too small'},
|
|
}, {
|
|
'url': 'https://utexas.app.box.com/s/2x6vanv85fdl8j2eqlcxmv0gp1wvps6e',
|
|
'info_dict': {
|
|
'id': '787379022466',
|
|
'ext': 'mp4',
|
|
'title': 'Webinar recording: Take the Leap!.mp4',
|
|
'uploader': 'Patricia Mosele',
|
|
'timestamp': 1615824864,
|
|
'upload_date': '20210315',
|
|
'uploader_id': '239068974',
|
|
},
|
|
'params': {'skip_download': 'dash fragment too small'},
|
|
}, {
|
|
'url': 'https://thejacksonlaboratory.ent.box.com/s/2x09dm6vcg6y28o0oox1so4l0t8wzt6l/file/1536173056065',
|
|
'info_dict': {
|
|
'id': '1536173056065',
|
|
'ext': 'mp4',
|
|
'uploader_id': '18523128264',
|
|
'uploader': 'Lexi Hennigan',
|
|
'title': 'iPSC Symposium recording part 1.mp4',
|
|
'timestamp': 1716228343,
|
|
'upload_date': '20240520',
|
|
},
|
|
'params': {'skip_download': 'dash fragment too small'},
|
|
}]
|
|
|
|
def _real_extract(self, url):
|
|
shared_name, file_id, service = self._match_valid_url(url).group('shared_name', 'id', 'service')
|
|
webpage = self._download_webpage(url, file_id or shared_name)
|
|
|
|
if not file_id:
|
|
post_stream_data = self._search_json(
|
|
r'Box\.postStreamData\s*=', webpage, 'Box post-stream data', shared_name)
|
|
shared_item = traverse_obj(
|
|
post_stream_data, ('/app-api/enduserapp/shared-item', {dict})) or {}
|
|
if shared_item.get('itemType') != 'file':
|
|
raise ExtractorError('The requested resource is not a file', expected=True)
|
|
|
|
file_id = str(shared_item['itemID'])
|
|
|
|
request_token = self._search_json(
|
|
r'Box\.config\s*=', webpage, 'Box config', file_id)['requestToken']
|
|
access_token = self._download_json(
|
|
f'https://{service}.box.com/app-api/enduserapp/elements/tokens', file_id,
|
|
'Downloading token JSON metadata',
|
|
data=json.dumps({'fileIDs': [file_id]}).encode(), headers={
|
|
'Content-Type': 'application/json',
|
|
'X-Request-Token': request_token,
|
|
'X-Box-EndUser-API': 'sharedName=' + shared_name,
|
|
})[file_id]['read']
|
|
shared_link = f'https://{service}.box.com/s/{shared_name}'
|
|
f = self._download_json(
|
|
'https://api.box.com/2.0/files/' + file_id, file_id,
|
|
'Downloading file JSON metadata', headers={
|
|
'Authorization': 'Bearer ' + access_token,
|
|
'BoxApi': 'shared_link=' + shared_link,
|
|
'X-Rep-Hints': '[dash]', # TODO: extract `hls` formats
|
|
}, query={
|
|
'fields': 'authenticated_download_url,created_at,created_by,description,extension,is_download_available,name,representations,size',
|
|
})
|
|
title = f['name']
|
|
|
|
query = {
|
|
'access_token': access_token,
|
|
'shared_link': shared_link,
|
|
}
|
|
|
|
formats = []
|
|
|
|
for url_tmpl in traverse_obj(f, (
|
|
'representations', 'entries', lambda _, v: v['representation'] == 'dash',
|
|
'content', 'url_template', {url_or_none},
|
|
)):
|
|
manifest_url = update_url_query(url_tmpl.replace('{+asset_path}', 'manifest.mpd'), query)
|
|
fmts = self._extract_mpd_formats(manifest_url, file_id)
|
|
for fmt in fmts:
|
|
fmt['extra_param_to_segment_url'] = urllib.parse.urlparse(manifest_url).query
|
|
formats.extend(fmts)
|
|
|
|
creator = f.get('created_by') or {}
|
|
|
|
return {
|
|
'id': file_id,
|
|
'title': title,
|
|
'formats': formats,
|
|
'description': f.get('description') or None,
|
|
'uploader': creator.get('name'),
|
|
'timestamp': parse_iso8601(f.get('created_at')),
|
|
'uploader_id': creator.get('id'),
|
|
}
|