aboutsummaryrefslogtreecommitdiff
path: root/youtube_dl/extractor/vk.py
diff options
context:
space:
mode:
authorSergey M․ <dstftw@gmail.com>2016-07-13 21:51:44 +0700
committerSergey M․ <dstftw@gmail.com>2016-07-13 21:51:44 +0700
commit2d19fb50725489d362d13e7347f90e0d2de10939 (patch)
treef4de829e308860a4adbb14f1e6451147586ba6a6 /youtube_dl/extractor/vk.py
parent9d865a1af6a509fc8546a049d996b7a7d28313cf (diff)
[vk:wallpost] Add extractor
Diffstat (limited to 'youtube_dl/extractor/vk.py')
-rw-r--r--youtube_dl/extractor/vk.py223
1 files changed, 175 insertions, 48 deletions
diff --git a/youtube_dl/extractor/vk.py b/youtube_dl/extractor/vk.py
index 758d9c86b..bcb7df83d 100644
--- a/youtube_dl/extractor/vk.py
+++ b/youtube_dl/extractor/vk.py
@@ -6,11 +6,18 @@ import json
import sys
from .common import InfoExtractor
-from ..compat import compat_str
+from ..compat import (
+ compat_str,
+ compat_urlparse,
+)
from ..utils import (
+ clean_html,
ExtractorError,
+ get_element_by_class,
int_or_none,
orderedSet,
+ parse_duration,
+ remove_start,
str_to_int,
unescapeHTML,
unified_strdate,
@@ -20,7 +27,54 @@ from .vimeo import VimeoIE
from .pladform import PladformIE
-class VKIE(InfoExtractor):
+class VKBaseIE(InfoExtractor):
+ _NETRC_MACHINE = 'vk'
+
+ def _login(self):
+ (username, password) = self._get_login_info()
+ if username is None:
+ return
+
+ login_page, url_handle = self._download_webpage_handle(
+ 'https://vk.com', None, 'Downloading login page')
+
+ login_form = self._hidden_inputs(login_page)
+
+ login_form.update({
+ 'email': username.encode('cp1251'),
+ 'pass': password.encode('cp1251'),
+ })
+
+ # https://new.vk.com/ serves two same remixlhk cookies in Set-Cookie header
+ # and expects the first one to be set rather than second (see
+ # https://github.com/rg3/youtube-dl/issues/9841#issuecomment-227871201).
+ # As of RFC6265 the newer one cookie should be set into cookie store
+ # what actually happens.
+ # We will workaround this VK issue by resetting the remixlhk cookie to
+ # the first one manually.
+ cookies = url_handle.headers.get('Set-Cookie')
+ if sys.version_info[0] >= 3:
+ cookies = cookies.encode('iso-8859-1')
+ cookies = cookies.decode('utf-8')
+ remixlhk = re.search(r'remixlhk=(.+?);.*?\bdomain=(.+?)(?:[,;]|$)', cookies)
+ if remixlhk:
+ value, domain = remixlhk.groups()
+ self._set_cookie(domain, 'remixlhk', value)
+
+ login_page = self._download_webpage(
+ 'https://login.vk.com/?act=login', None,
+ note='Logging in as %s' % username,
+ data=urlencode_postdata(login_form))
+
+ if re.search(r'onLoginFailed', login_page):
+ raise ExtractorError(
+ 'Unable to login, incorrect username and/or password', expected=True)
+
+ def _real_initialize(self):
+ self._login()
+
+
+class VKIE(VKBaseIE):
IE_NAME = 'vk'
IE_DESC = 'VK'
_VALID_URL = r'''(?x)
@@ -38,8 +92,6 @@ class VKIE(InfoExtractor):
(?P<videoid>-?\d+_\d+)(?:.*\blist=(?P<list_id>[\da-f]+))?
)
'''
- _NETRC_MACHINE = 'vk'
-
_TESTS = [
{
'url': 'http://vk.com/videos-77521?z=video-77521_162222515%2Fclub77521',
@@ -189,49 +241,6 @@ class VKIE(InfoExtractor):
}
]
- def _login(self):
- (username, password) = self._get_login_info()
- if username is None:
- return
-
- login_page, url_handle = self._download_webpage_handle(
- 'https://vk.com', None, 'Downloading login page')
-
- login_form = self._hidden_inputs(login_page)
-
- login_form.update({
- 'email': username.encode('cp1251'),
- 'pass': password.encode('cp1251'),
- })
-
- # https://new.vk.com/ serves two same remixlhk cookies in Set-Cookie header
- # and expects the first one to be set rather than second (see
- # https://github.com/rg3/youtube-dl/issues/9841#issuecomment-227871201).
- # As of RFC6265 the newer one cookie should be set into cookie store
- # what actually happens.
- # We will workaround this VK issue by resetting the remixlhk cookie to
- # the first one manually.
- cookies = url_handle.headers.get('Set-Cookie')
- if sys.version_info[0] >= 3:
- cookies = cookies.encode('iso-8859-1')
- cookies = cookies.decode('utf-8')
- remixlhk = re.search(r'remixlhk=(.+?);.*?\bdomain=(.+?)(?:[,;]|$)', cookies)
- if remixlhk:
- value, domain = remixlhk.groups()
- self._set_cookie(domain, 'remixlhk', value)
-
- login_page = self._download_webpage(
- 'https://login.vk.com/?act=login', None,
- note='Logging in as %s' % username,
- data=urlencode_postdata(login_form))
-
- if re.search(r'onLoginFailed', login_page):
- raise ExtractorError(
- 'Unable to login, incorrect username and/or password', expected=True)
-
- def _real_initialize(self):
- self._login()
-
def _real_extract(self, url):
mobj = re.match(self._VALID_URL, url)
video_id = mobj.group('videoid')
@@ -355,7 +364,7 @@ class VKIE(InfoExtractor):
}
-class VKUserVideosIE(InfoExtractor):
+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)(?:[/?#&]|$)'
@@ -396,3 +405,121 @@ class VKUserVideosIE(InfoExtractor):
webpage, 'title', default=page_id))
return self.playlist_result(entries, page_id, title)
+
+
+class VKWallPostIE(VKBaseIE):
+ IE_NAME = 'vk:wallpost'
+ _VALID_URL = r'https?://(?:(?:(?:(?:m|new)\.)?vk\.com/(?:[^?]+\?.*\bw=)?wall(?P<id>-?\d+_\d+)))'
+ _TESTS = [{
+ # public page URL, audio playlist
+ 'url': 'https://vk.com/bs.official?w=wall-23538238_35',
+ 'info_dict': {
+ 'id': '23538238_35',
+ 'title': 'Black Shadow - Wall post 23538238_35',
+ 'description': 'md5:3f84b9c4f9ef499731cf1ced9998cc0c',
+ },
+ 'playlist': [{
+ 'md5': '5ba93864ec5b85f7ce19a9af4af080f6',
+ 'info_dict': {
+ 'id': '135220665_111806521',
+ 'ext': 'mp3',
+ 'title': 'Black Shadow - Слепое Верование',
+ 'duration': 370,
+ 'uploader': 'Black Shadow',
+ 'artist': 'Black Shadow',
+ 'track': 'Слепое Верование',
+ },
+ }, {
+ 'md5': '4cc7e804579122b17ea95af7834c9233',
+ 'info_dict': {
+ 'id': '135220665_111802303',
+ 'ext': 'mp3',
+ 'title': 'Black Shadow - Война - Негасимое Бездны Пламя!',
+ 'duration': 423,
+ 'uploader': 'Black Shadow',
+ 'artist': 'Black Shadow',
+ 'track': 'Война - Негасимое Бездны Пламя!',
+ },
+ 'params': {
+ 'skip_download': True,
+ },
+ }],
+ 'skip': 'Requires vk account credentials',
+ }, {
+ # single YouTube embed, no leading -
+ 'url': 'https://vk.com/wall85155021_6319',
+ 'info_dict': {
+ 'id': '85155021_6319',
+ 'title': 'Sergey Gorbunov - Wall post 85155021_6319',
+ },
+ 'playlist_count': 1,
+ 'skip': 'Requires vk account credentials',
+ }, {
+ # wall page URL
+ 'url': 'https://vk.com/wall-23538238_35',
+ 'only_matching': True,
+ }, {
+ # mobile wall page URL
+ 'url': 'https://m.vk.com/wall-23538238_35',
+ 'only_matching': True,
+ }]
+
+ def _real_extract(self, url):
+ post_id = self._match_id(url)
+
+ wall_url = 'https://vk.com/wall%s' % post_id
+
+ post_id = remove_start(post_id, '-')
+
+ webpage = self._download_webpage(wall_url, post_id)
+
+ error = self._html_search_regex(
+ r'>Error</div>\s*<div[^>]+class=["\']body["\'][^>]*>([^<]+)',
+ webpage, 'error', default=None)
+ if error:
+ raise ExtractorError('VK said: %s' % error, expected=True)
+
+ description = clean_html(get_element_by_class('wall_post_text', webpage))
+ uploader = clean_html(get_element_by_class(
+ 'fw_post_author', webpage)) or self._og_search_description(webpage)
+ thumbnail = self._og_search_thumbnail(webpage)
+
+ entries = []
+
+ for audio in re.finditer(r'''(?sx)
+ <input[^>]+
+ id=(?P<q1>["\'])audio_info(?P<id>\d+_\d+).*?(?P=q1)[^>]+
+ value=(?P<q2>["\'])(?P<url>http.+?)(?P=q2)
+ .+?
+ </table>''', webpage):
+ audio_html = audio.group(0)
+ audio_id = audio.group('id')
+ duration = parse_duration(get_element_by_class('duration', audio_html))
+ track = self._html_search_regex(
+ r'<span[^>]+id=["\']title%s[^>]*>([^<]+)' % audio_id,
+ audio_html, 'title', default=None)
+ artist = self._html_search_regex(
+ r'>([^<]+)</a></b>\s*&ndash', audio_html,
+ 'artist', default=None)
+ entries.append({
+ 'id': audio_id,
+ 'url': audio.group('url'),
+ 'title': '%s - %s' % (artist, track) if artist and track else audio_id,
+ 'thumbnail': thumbnail,
+ 'duration': duration,
+ 'uploader': uploader,
+ 'artist': artist,
+ 'track': track,
+ })
+
+ for video in re.finditer(
+ r'<a[^>]+href=(["\'])(?P<url>/video(?:-?[\d_]+).*?)\1', webpage):
+ entries.append(self.url_result(
+ compat_urlparse.urljoin(url, video.group('url')), VKIE.ie_key()))
+
+ title = 'Wall post %s' % post_id
+
+ return self.playlist_result(
+ orderedSet(entries), post_id,
+ '%s - %s' % (uploader, title) if uploader else title,
+ description)