|
|
@ -1,58 +1,67 @@ |
|
|
|
# coding: utf-8 |
|
|
|
from __future__ import unicode_literals |
|
|
|
|
|
|
|
import re |
|
|
|
|
|
|
|
from .common import InfoExtractor |
|
|
|
from ..utils import int_or_none |
|
|
|
from ..utils import ( |
|
|
|
int_or_none, |
|
|
|
unified_strdate, |
|
|
|
) |
|
|
|
|
|
|
|
|
|
|
|
class UstudioIE(InfoExtractor): |
|
|
|
IE_NAME = 'uStudio' |
|
|
|
_VALID_URL = r'http://(?:www\.|v1\.)?ustudio.com/video/(?P<id>[\w\d]+)/.+' |
|
|
|
_TESTS = [ |
|
|
|
{ |
|
|
|
'url': 'http://ustudio.com/video/Uxu2my9bgSph/san_francisco_golden_gate_bridge', |
|
|
|
'md5': '58bbfca62125378742df01fc2abbdef6', |
|
|
|
'info_dict': { |
|
|
|
'id': 'Uxu2my9bgSph', |
|
|
|
'ext': 'mp4', |
|
|
|
'title': 'San Francisco: Golden Gate Bridge', |
|
|
|
'thumbnail': 're:^https?://.*\.jpg$', |
|
|
|
'description': 'md5:23925500697f2c6d4830e387ba51a9be', |
|
|
|
'uploader': 'Tony Farley', |
|
|
|
} |
|
|
|
}, |
|
|
|
] |
|
|
|
_VALID_URL = r'https?://(?:(?:www|v1)\.)?ustudio\.com/video/(?P<id>[^/]+)/(?P<display_id>[^/?#&]+)' |
|
|
|
_TEST = { |
|
|
|
'url': 'http://ustudio.com/video/Uxu2my9bgSph/san_francisco_golden_gate_bridge', |
|
|
|
'md5': '58bbfca62125378742df01fc2abbdef6', |
|
|
|
'info_dict': { |
|
|
|
'id': 'Uxu2my9bgSph', |
|
|
|
'display_id': 'san_francisco_golden_gate_bridge', |
|
|
|
'ext': 'mp4', |
|
|
|
'title': 'San Francisco: Golden Gate Bridge', |
|
|
|
'description': 'md5:23925500697f2c6d4830e387ba51a9be', |
|
|
|
'thumbnail': 're:^https?://.*\.jpg$', |
|
|
|
'upload_date': '20111107', |
|
|
|
'uploader': 'Tony Farley', |
|
|
|
} |
|
|
|
} |
|
|
|
|
|
|
|
def _real_extract(self, url): |
|
|
|
video_id = self._match_id(url) |
|
|
|
|
|
|
|
webpage = self._download_webpage(url, video_id) |
|
|
|
|
|
|
|
doc = self._download_xml( |
|
|
|
'http://v1.ustudio.com/embed/{0}/ustudio/config.xml'.format( |
|
|
|
video_id), |
|
|
|
video_id, |
|
|
|
note='Downloading video info', |
|
|
|
errnote='Failed to download video info') |
|
|
|
|
|
|
|
formats = [ |
|
|
|
{ |
|
|
|
'url': quality.attrib['url'], |
|
|
|
'width': int_or_none(quality.attrib.get('width')), |
|
|
|
'height': int_or_none(quality.attrib.get('height')), |
|
|
|
} for quality in doc.findall('./qualities/quality/video') |
|
|
|
] |
|
|
|
mobj = re.match(self._VALID_URL, url) |
|
|
|
video_id = mobj.group('id') |
|
|
|
display_id = mobj.group('display_id') |
|
|
|
|
|
|
|
config = self._download_xml( |
|
|
|
'http://v1.ustudio.com/embed/%s/ustudio/config.xml' % video_id, |
|
|
|
display_id) |
|
|
|
|
|
|
|
def extract(kind): |
|
|
|
return [{ |
|
|
|
'url': item.attrib['url'], |
|
|
|
'width': int_or_none(item.get('width')), |
|
|
|
'height': int_or_none(item.get('height')), |
|
|
|
} for item in config.findall('./qualities/quality/%s' % kind) if item.get('url')] |
|
|
|
|
|
|
|
formats = extract('video') |
|
|
|
self._sort_formats(formats) |
|
|
|
|
|
|
|
webpage = self._download_webpage(url, display_id) |
|
|
|
|
|
|
|
title = self._og_search_title(webpage) |
|
|
|
upload_date = unified_strdate(self._search_regex( |
|
|
|
r'(?s)Uploaded by\s*.+?\s*on\s*<span>([^<]+)</span>', |
|
|
|
webpage, 'upload date', fatal=False)) |
|
|
|
uploader = self._search_regex( |
|
|
|
r'Uploaded by\s*<a[^>]*>([^<]+)<', |
|
|
|
webpage, 'uploader', fatal=False) |
|
|
|
|
|
|
|
return { |
|
|
|
'id': video_id, |
|
|
|
'title': self._og_search_title(webpage), |
|
|
|
'thumbnail': self._og_search_thumbnail(webpage), |
|
|
|
'formats': formats, |
|
|
|
'display_id': display_id, |
|
|
|
'title': title, |
|
|
|
'description': self._og_search_description(webpage), |
|
|
|
'uploader': self._html_search_regex( |
|
|
|
r'<a href=".*/user/.+">(.+)</a> on', |
|
|
|
webpage, |
|
|
|
'uploader', |
|
|
|
fatal=False), |
|
|
|
'thumbnails': extract('image'), |
|
|
|
'upload_date': upload_date, |
|
|
|
'uploader': uploader, |
|
|
|
'formats': formats, |
|
|
|
} |