Added extractors for smotri.com

This commit is contained in:
dst 2013-12-02 20:10:19 +07:00
parent 55a10eab48
commit 5270d8cb13
3 changed files with 265 additions and 1 deletions

View file

@ -22,7 +22,9 @@ from youtube_dl.extractor import (
LivestreamIE, LivestreamIE,
NHLVideocenterIE, NHLVideocenterIE,
BambuserChannelIE, BambuserChannelIE,
BandcampAlbumIE BandcampAlbumIE,
SmotriCommunityIE,
SmotriUserIE
) )
@ -120,5 +122,23 @@ class TestPlaylists(unittest.TestCase):
self.assertEqual(result['title'], u'Nightmare Night EP') self.assertEqual(result['title'], u'Nightmare Night EP')
self.assertTrue(len(result['entries']) >= 4) self.assertTrue(len(result['entries']) >= 4)
def test_smotri_community(self):
dl = FakeYDL()
ie = SmotriCommunityIE(dl)
result = ie.extract('http://smotri.com/community/video/kommuna')
self.assertIsPlaylist(result)
self.assertEqual(result['id'], u'kommuna')
self.assertEqual(result['title'], u'КПРФ')
self.assertTrue(len(result['entries']) >= 4)
def test_smotri_user(self):
dl = FakeYDL()
ie = SmotriUserIE(dl)
result = ie.extract('http://smotri.com/user/inspector')
self.assertIsPlaylist(result)
self.assertEqual(result['id'], u'inspector')
self.assertEqual(result['title'], u'Inspector')
self.assertTrue(len(result['entries']) >= 9)
if __name__ == '__main__': if __name__ == '__main__':
unittest.main() unittest.main()

View file

@ -121,6 +121,11 @@ from .rutube import RutubeIE
from .sina import SinaIE from .sina import SinaIE
from .slashdot import SlashdotIE from .slashdot import SlashdotIE
from .slideshare import SlideshareIE from .slideshare import SlideshareIE
from .smotri import (
SmotriIE,
SmotriCommunityIE,
SmotriUserIE,
)
from .sohu import SohuIE from .sohu import SohuIE
from .soundcloud import SoundcloudIE, SoundcloudSetIE, SoundcloudUserIE from .soundcloud import SoundcloudIE, SoundcloudSetIE, SoundcloudUserIE
from .southparkstudios import ( from .southparkstudios import (

View file

@ -0,0 +1,239 @@
# encoding: utf-8
import re
import json
import hashlib
from .common import InfoExtractor
from ..utils import (
determine_ext,
ExtractorError
)
class SmotriIE(InfoExtractor):
IE_DESC = u'Smotri.com'
IE_NAME = u'smotri'
_VALID_URL = r'^(?:http://)?(?:www\.)?(?P<url>smotri\.com/video/view/\?id=(?P<videoid>v(?P<realvideoid>[0-9]+)[a-z0-9]{4}))'
_TESTS = [
# real video id 2610366
{
u'url': u'http://smotri.com/video/view/?id=v261036632ab',
u'file': u'v261036632ab.mp4',
u'md5': u'46a72e83a6ad8862b64fa6953fa93f8a',
u'info_dict': {
u'title': u'катастрофа с камер видеонаблюдения',
u'uploader': u'rbc2008',
u'uploader_id': u'rbc08',
u'upload_date': u'20131118',
u'thumbnail': u'http://frame6.loadup.ru/8b/a9/2610366.3.3.jpg'
},
},
# real video id 57591
{
u'url': u'http://smotri.com/video/view/?id=v57591cb20',
u'file': u'v57591cb20.flv',
u'md5': u'9eae59f6dda7087bf39a140e2fff5757',
u'info_dict': {
u'title': u'test',
u'uploader': u'Support Photofile@photofile',
u'uploader_id': u'support-photofile',
u'upload_date': u'20070704',
u'thumbnail': u'http://frame4.loadup.ru/03/ed/57591.2.3.jpg'
},
},
# video-password
{
u'url': u'http://smotri.com/video/view/?id=v1390466a13c',
u'file': u'v1390466a13c.mp4',
u'md5': u'fe4dd9357558d5ee3c8fc0ef0d39de66',
u'info_dict': {
u'title': u'TOCCA_A_NOI_-_LE_COSE_NON_VANNO_CAMBIAMOLE_ORA-1',
u'uploader': u'timoxa40',
u'uploader_id': u'timoxa40',
u'upload_date': u'20100404',
u'thumbnail': u'http://frame7.loadup.ru/af/3f/1390466.3.3.jpg'
},
u'params': {
u'videopassword': u'qwerty',
},
},
# age limit + video-password
{
u'url': u'http://smotri.com/video/view/?id=v15408898bcf',
u'file': u'v15408898bcf.flv',
u'md5': u'c66a5d61379ac6fde06f07eebe436316',
u'info_dict': {
u'title': u'этот ролик не покажут по ТВ',
u'uploader': u'zzxxx',
u'uploader_id': u'ueggb',
u'upload_date': u'20101001',
u'thumbnail': u'http://frame3.loadup.ru/75/75/1540889.1.3.jpg',
u'age_limit': 18
},
u'params': {
u'videopassword': u'333'
}
}
]
_SUCCESS = 0
_PASSWORD_NOT_VERIFIED = 1
_PASSWORD_DETECTED = 2
_VIDEO_NOT_FOUND = 3
def _search_meta(self, name, html, display_name=None):
if display_name is None:
display_name = name
return self._html_search_regex(
r'<meta itemprop="%s" content="([^"]+)" />' % re.escape(name),
html, display_name, fatal=False)
def _real_extract(self, url):
mobj = re.match(self._VALID_URL, url)
video_id = mobj.group('videoid')
real_video_id = mobj.group('realvideoid')
# Download video JSON data
video_json_url = 'http://smotri.com/vt.php?id=%s' % real_video_id
video_json_page = self._download_webpage(video_json_url, video_id, u'Downloading video JSON')
video_json = json.loads(video_json_page)
status = video_json['status']
if status == self._VIDEO_NOT_FOUND:
raise ExtractorError(u'Video %s does not exist' % video_id, expected=True)
elif status == self._PASSWORD_DETECTED: # The video is protected by a password, retry with
# video-password set
video_password = self._downloader.params.get('videopassword', None)
if not video_password:
raise ExtractorError(u'This video is protected by a password, use the --video-password option', expected=True)
video_json_url += '&md5pass=%s' % hashlib.md5(video_password).hexdigest()
video_json_page = self._download_webpage(video_json_url, video_id, u'Downloading video JSON (video-password set)')
video_json = json.loads(video_json_page)
status = video_json['status']
if status == self._PASSWORD_NOT_VERIFIED:
raise ExtractorError(u'Video password is invalid', expected=True)
if status != self._SUCCESS:
raise ExtractorError(u'Unexpected status value %s' % status)
# Extract the URL of the video
video_url = video_json['file_data']
video_ext = determine_ext(video_url)
# Video JSON does not provide enough meta data
# We will extract some from the video web page instead
video_page_url = 'http://' + mobj.group('url')
video_page = self._download_webpage(video_page_url, video_id, u'Downloading video page')
# Adult content
if re.search(u'EroConfirmText">', video_page) is not None:
self.report_age_confirmation()
confirm_string = self._html_search_regex(
ur'<a href="/video/view/\?id=%s&confirm=([^"]+)" title="[^"]+">' % video_id,
video_page, u'confirm string')
confirm_url = video_page_url + '&confirm=%s' % confirm_string
video_page = self._download_webpage(confirm_url, video_id, u'Downloading video page (age confirmed)')
adult_content = True
else:
adult_content = False
# Extract the rest of meta data
video_title = self._search_meta(u'name', video_page, u'title')
if not video_title:
video_title = video_url.rsplit('/', 1)[-1]
video_description = self._search_meta(u'description', video_page)
video_thumbnail = self._search_meta(u'thumbnail', video_page)
upload_date_str = self._search_meta(u'uploadDate', video_page, u'upload date')
upload_date_m = re.search(r'(?P<year>\d{4})\.(?P<month>\d{2})\.(?P<day>\d{2})T', upload_date_str)
video_upload_date = (
(
upload_date_m.group('year') +
upload_date_m.group('month') +
upload_date_m.group('day')
)
if upload_date_m else None
)
duration_str = self._search_meta(u'duration', video_page)
duration_m = re.search(r'T(?P<hours>[0-9]{2})H(?P<minutes>[0-9]{2})M(?P<seconds>[0-9]{2})S', duration_str)
video_duration = (
(
(int(duration_m.group('hours')) * 60 * 60) +
(int(duration_m.group('minutes')) * 60) +
int(duration_m.group('seconds'))
)
if duration_m else None
)
video_uploader = self._html_search_regex(
ur'<div class="DescrUser"><div>Автор.*?onmouseover="popup_user_info[^"]+">(.*?)</a>',
video_page, u'uploader', fatal=False, flags=re.MULTILINE|re.DOTALL)
video_uploader_id = self._html_search_regex(
ur'<div class="DescrUser"><div>Автор.*?onmouseover="popup_user_info\(.*?\'([^\']+)\'\);">',
video_page, u'uploader id', fatal=False, flags=re.MULTILINE|re.DOTALL)
video_view_count = self._html_search_regex(
ur'Общее количество просмотров.*?<span class="Number">(\d+)</span>',
video_page, u'view count', fatal=False, flags=re.MULTILINE|re.DOTALL)
return {
'id': video_id,
'url': video_url,
'title': video_title,
'ext': video_ext,
'thumbnail': video_thumbnail,
'description': video_description,
'uploader': video_uploader,
'upload_date': video_upload_date,
'uploader_id': video_uploader_id,
'video_duration': video_duration,
'view_count': video_view_count,
'age_limit': 18 if adult_content else 0,
'video_page_url': video_page_url
}
class SmotriCommunityIE(InfoExtractor):
IE_DESC = u'Smotri.com community videos'
IE_NAME = u'smotri:community'
_VALID_URL = r'^(?:http://)?(?:www\.)?smotri\.com/community/video/(?P<communityid>[0-9A-Za-z_\'-]+)'
def _real_extract(self, url):
mobj = re.match(self._VALID_URL, url)
community_id = mobj.group('communityid')
url = 'http://smotri.com/export/rss/video/by/community/-/%s/video.xml' % community_id
rss = self._download_xml(url, community_id, u'Downloading community RSS')
entries = [self.url_result(video_url.text, 'Smotri')
for video_url in rss.findall('./channel/item/link')]
community_title = self._html_search_regex(
ur'^Видео сообщества "([^"]+)"$', rss.find('./channel/description').text, u'community title')
return self.playlist_result(entries, community_id, community_title)
class SmotriUserIE(InfoExtractor):
IE_DESC = u'Smotri.com user videos'
IE_NAME = u'smotri:user'
_VALID_URL = r'^(?:http://)?(?:www\.)?smotri\.com/user/(?P<userid>[0-9A-Za-z_\'-]+)'
def _real_extract(self, url):
mobj = re.match(self._VALID_URL, url);
user_id = mobj.group('userid')
url = 'http://smotri.com/export/rss/user/video/-/%s/video.xml' % user_id
rss = self._download_xml(url, user_id, u'Downloading user RSS')
entries = [self.url_result(video_url.text, 'Smotri')
for video_url in rss.findall('./channel/item/link')]
user_nickname = self._html_search_regex(
ur'^Видео режиссера (.*)$', rss.find('./channel/description').text, u'user nickname')
return self.playlist_result(entries, user_id, user_nickname)