from __future__ import unicode_literals
import collections
+import functools
import re
from .common import InfoExtractor
ExtractorError,
get_element_by_class,
int_or_none,
+ OnDemandPagedList,
orderedSet,
str_or_none,
str_to_int,
class VKUserVideosIE(VKBaseIE):
IE_NAME = 'vk:uservideos'
IE_DESC = "VK - User's Videos"
- _VALID_URL = r'https?://(?:(?:m|new)\.)?vk\.com/videos(?P<id>-?[0-9]+)(?!\?.*\bz=video)(?:[/?#&]|$)'
+ _VALID_URL = r'https?://(?:(?:m|new)\.)?vk\.com/videos(?P<id>-?[0-9]+)(?!\?.*\bz=video)(?:[/?#&](?:.*?\bsection=(?P<section>\w+))?|$)'
_TEMPLATE_URL = 'https://vk.com/videos'
_TESTS = [{
- 'url': 'http://vk.com/videos205387401',
+ 'url': 'https://vk.com/videos-767561',
+ 'info_dict': {
+ 'id': '-767561_all',
+ },
+ 'playlist_mincount': 1150,
+ }, {
+ 'url': 'https://vk.com/videos-767561?section=uploaded',
'info_dict': {
- 'id': '205387401',
+ 'id': '-767561_uploaded',
},
- 'playlist_mincount': 4,
+ 'playlist_mincount': 425,
+ }, {
+ 'url': 'http://vk.com/videos205387401',
+ 'only_matching': True,
}, {
'url': 'http://vk.com/videos-77521',
'only_matching': True,
'url': 'http://new.vk.com/videos205387401',
'only_matching': True,
}]
- _VIDEO = collections.namedtuple(
- 'Video', ['owner_id', 'id', 'thumb', 'title', 'flags', 'duration', 'hash', 'moder_acts', 'owner', 'date', 'views', 'platform', 'blocked', 'music_video_meta'])
-
- def _real_extract(self, url):
- page_id = self._match_id(url)
+ _PAGE_SIZE = 1000
+ _VIDEO = collections.namedtuple('Video', ['owner_id', 'id'])
+ def _fetch_page(self, page_id, section, page):
l = self._download_payload('al_video', page_id, {
'act': 'load_videos_silent',
+ 'offset': page * self._PAGE_SIZE,
'oid': page_id,
- })[0]['']['list']
+ 'section': section,
+ })[0][section]['list']
- entries = []
for video in l:
- v = self._VIDEO._make(video)
+ v = self._VIDEO._make(video[:2])
video_id = '%d_%d' % (v.owner_id, v.id)
- entries.append(self.url_result(
- 'http://vk.com/video' + video_id, 'VK', video_id=video_id))
+ yield self.url_result(
+ 'http://vk.com/video' + video_id, VKIE.ie_key(), video_id)
+
+ def _real_extract(self, url):
+ page_id, section = re.match(self._VALID_URL, url).groups()
+ if not section:
+ section = 'all'
+
+ entries = OnDemandPagedList(
+ functools.partial(self._fetch_page, page_id, section),
+ self._PAGE_SIZE)
- return self.playlist_result(entries, page_id)
+ return self.playlist_result(entries, '%s_%s' % (page_id, section))
class VKWallPostIE(VKBaseIE):
'only_matching': True,
}]
_BASE64_CHARS = 'abcdefghijklmnopqrstuvwxyzABCDEFGHIJKLMN0PQRSTUVWXYZO123456789+/='
- _AUDIO = collections.namedtuple(
- 'Audio', ['id', 'owner_id', 'url', 'title', 'performer', 'duration', 'album_id', 'unk', 'author_link', 'lyrics', 'flags', 'context', 'extra', 'hashes', 'cover_url', 'ads', 'subtitle', 'main_artists', 'feat_artists', 'album', 'track_code', 'restriction', 'album_part', 'new_stats', 'access_key'])
+ _AUDIO = collections.namedtuple('Audio', ['id', 'owner_id', 'url', 'title', 'performer', 'duration', 'album_id', 'unk', 'author_link', 'lyrics', 'flags', 'context', 'extra', 'hashes', 'cover_url', 'ads'])
def _decode(self, enc):
dec = ''
for audio in re.findall(r'data-audio="([^"]+)', webpage):
audio = self._parse_json(unescapeHTML(audio), post_id)
- a = self._AUDIO._make(audio)
+ a = self._AUDIO._make(audio[:16])
if not a.url:
continue
title = unescapeHTML(a.title)