Updating porntrex
This commit is contained in:
parent
1ad6e81451
commit
26eaa4e762
@ -15,29 +15,33 @@ class PornTrexIE(InfoExtractor):
|
||||
_NETRC_MACHINE = 'porntrex'
|
||||
_VALID_URL = r'https?://(?:www\.)?porntrex\.com/video/(?P<id>[0-9]+)/.*'
|
||||
_TEST = {
|
||||
'url': 'https://www.porntrex.com/video/519351/be-ariyana-adin-breaking-and-entering-this-pussy',
|
||||
# 'md5': 'TODO: md5 sum of the first 10241 bytes of the video file (use --test)',
|
||||
'url': 'https://www.porntrex.com/video/519351/\
|
||||
be-ariyana-adin-breaking-and-entering-this-pussy',
|
||||
'info_dict': {
|
||||
'id': '519351',
|
||||
'ext': 'mp4',
|
||||
'title': 'BE - Ariyana Adin - Breaking And Entering This Pussy',
|
||||
'uploader': 'brand95',
|
||||
'description': 'BE - Ariyana Adin - Breaking And Entering This Pussy',
|
||||
'description': 'BE - Ariyana Adin - Breaking And Entering This \
|
||||
Pussy',
|
||||
}
|
||||
}
|
||||
|
||||
def get_resolution(self, url):
|
||||
'''Video resolution extraction from url'''
|
||||
try:
|
||||
resolution = ((url.split('.')[2])).split('_')[-1]
|
||||
except:
|
||||
resolution = ((url.split('.')[2])).split('_')[2]
|
||||
except IndexError:
|
||||
resolution = '480p'
|
||||
return resolution
|
||||
|
||||
|
||||
def get_protocol(self, url):
|
||||
return url.split('/')[0]
|
||||
'''Video protocol extraction from url'''
|
||||
return url.split(':')[0]
|
||||
|
||||
def get_thumbnails(self, html):
|
||||
thumbnails_regex = re.compile('href="(http.*?/screenshots/\d+.jpg/)"')
|
||||
'''Each video has 10 thumbnails - extracted here.'''
|
||||
thumbnails_regex = re.compile(r'href="(http.*?/screenshots/\d+.jpg/)"')
|
||||
thumbnails_list = re.findall(thumbnails_regex, html)
|
||||
thumbnails = []
|
||||
for thumbs in thumbnails_list:
|
||||
@ -57,7 +61,7 @@ class PornTrexIE(InfoExtractor):
|
||||
login_form.update({
|
||||
'username': username.encode('utf-8'),
|
||||
'pass': password.encode('utf-8'),
|
||||
})
|
||||
})
|
||||
|
||||
login_page = self._download_webpage(
|
||||
'https://www.porntrex.com/ajax-login/', None,
|
||||
@ -66,8 +70,9 @@ class PornTrexIE(InfoExtractor):
|
||||
|
||||
if re.search(r'generic-error hidden', login_page):
|
||||
raise ExtractorError(
|
||||
'Unable to login, incorrect username and/or password', expected=True)
|
||||
|
||||
'Unable to login, incorrect username and/or password',
|
||||
expected=True)
|
||||
|
||||
def _real_initialize(self):
|
||||
self._login()
|
||||
|
||||
@ -75,14 +80,18 @@ class PornTrexIE(InfoExtractor):
|
||||
video_id = self._match_id(url)
|
||||
webpage = self._download_webpage(url, video_id)
|
||||
|
||||
is_video_private_regex = re.compile('Only active members can watch private videos.')
|
||||
private_string = 'Only active members can watch private videos.'
|
||||
is_video_private_regex = re.compile(private_string)
|
||||
if (re.findall(is_video_private_regex, webpage)):
|
||||
self.raise_login_required()
|
||||
|
||||
title = self._html_search_regex(r'<title>(.+?)</title>', webpage, 'title',)
|
||||
title = self._html_search_regex(
|
||||
r'<title>(.+?)</title>', webpage, 'title',)
|
||||
url2_regex = re.compile("'(https://www.porntrex.com/get_file/.*?)/'")
|
||||
url2 = re.findall(url2_regex, webpage)
|
||||
uploader_regex = re.compile(r'<a href="https://www.porntrex.com/members/[0-9]+?/">(.+?)</a>', re.DOTALL)
|
||||
uploader_regex = re.compile(
|
||||
r'<a href="https://www.porntrex.com/members/[0-9]+?/">(.+?)</a>',
|
||||
re.DOTALL)
|
||||
uploader = re.findall(uploader_regex, webpage)[0].strip()
|
||||
formats = []
|
||||
for x, _ in enumerate(url2):
|
||||
@ -93,6 +102,7 @@ class PornTrexIE(InfoExtractor):
|
||||
})
|
||||
# self.get_thumbnails(webpage)
|
||||
self._sort_formats(formats)
|
||||
print(formats)
|
||||
|
||||
return {
|
||||
'id': video_id,
|
||||
@ -105,24 +115,18 @@ class PornTrexIE(InfoExtractor):
|
||||
|
||||
|
||||
class PornTrexPlayListIE(InfoExtractor):
|
||||
'''Class for downloading Porntrex video playlists.'''
|
||||
'''Class for downloading Porntrex video playlists.'''
|
||||
_NETRC_MACHINE = 'porntrex'
|
||||
_VALID_URL = r'https?://(?:www\.)?porntrex\.com/playlists/(?P<id>[0-9]+)/.*'
|
||||
_VALID_URL = \
|
||||
r'https?://(?:www\.)?porntrex\.com/playlists/(?P<id>[0-9]+)/.*'
|
||||
_TEST = {
|
||||
'url': 'https://www.porntrex.com/playlists/60671/les45/',
|
||||
# 'md5': 'TODO: md5 sum of the first 10241 bytes of the video file (use --test)',
|
||||
'info_dict': {
|
||||
'id': '477697',
|
||||
'ext': 'mp4',
|
||||
'uploader': 'tarpi',
|
||||
'title': '4. Kelly Divine, Tiffany Minx (1080p)',
|
||||
'description': '4. Kelly Divine, Tiffany Minx (1080p)'
|
||||
# 'thumbnail': r're:^https?://.*\.jpg$',
|
||||
# TODO more properties, either as:
|
||||
# * A value
|
||||
# * MD5 checksum; start the string with md5:
|
||||
# * A regular expression; start the string with re:
|
||||
# * Any Python type (for example int or float)
|
||||
}
|
||||
}
|
||||
|
||||
|
Loading…
x
Reference in New Issue
Block a user