mirror of
https://github.com/yt-dlp/yt-dlp.git
synced 2024-11-30 12:01:28 +00:00
add96eb9f8
Authored by: seproDev Reviewed-by: bashonly <88596187+bashonly@users.noreply.github.com> Reviewed-by: Simon Sawicki <contact@grub4k.xyz>
123 lines
4.3 KiB
Python
123 lines
4.3 KiB
Python
import re
|
|
import urllib.parse
|
|
|
|
from .common import InfoExtractor
|
|
from ..utils import (
|
|
ExtractorError,
|
|
merge_dicts,
|
|
)
|
|
|
|
|
|
class EroProfileIE(InfoExtractor):
|
|
_VALID_URL = r'https?://(?:www\.)?eroprofile\.com/m/videos/view/(?P<id>[^/]+)'
|
|
_LOGIN_URL = 'http://www.eroprofile.com/auth/auth.php?'
|
|
_NETRC_MACHINE = 'eroprofile'
|
|
_TESTS = [{
|
|
'url': 'http://www.eroprofile.com/m/videos/view/sexy-babe-softcore',
|
|
'md5': 'c26f351332edf23e1ea28ce9ec9de32f',
|
|
'info_dict': {
|
|
'id': '3733775',
|
|
'display_id': 'sexy-babe-softcore',
|
|
'ext': 'm4v',
|
|
'title': 'sexy babe softcore',
|
|
'thumbnail': r're:https?://.*\.jpg',
|
|
'age_limit': 18,
|
|
},
|
|
'skip': 'Video not found',
|
|
}, {
|
|
'url': 'http://www.eroprofile.com/m/videos/view/Try-It-On-Pee_cut_2-wmv-4shared-com-file-sharing-download-movie-file',
|
|
'md5': '1baa9602ede46ce904c431f5418d8916',
|
|
'info_dict': {
|
|
'id': '1133519',
|
|
'ext': 'm4v',
|
|
'title': 'Try It On Pee_cut_2.wmv - 4shared.com - file sharing - download movie file',
|
|
'thumbnail': r're:https?://.*\.jpg',
|
|
'age_limit': 18,
|
|
},
|
|
'skip': 'Requires login',
|
|
}]
|
|
|
|
def _perform_login(self, username, password):
|
|
query = urllib.parse.urlencode({
|
|
'username': username,
|
|
'password': password,
|
|
'url': 'http://www.eroprofile.com/',
|
|
})
|
|
login_url = self._LOGIN_URL + query
|
|
login_page = self._download_webpage(login_url, None, False)
|
|
|
|
m = re.search(r'Your username or password was incorrect\.', login_page)
|
|
if m:
|
|
raise ExtractorError(
|
|
'Wrong username and/or password.', expected=True)
|
|
|
|
self.report_login()
|
|
redirect_url = self._search_regex(
|
|
r'<script[^>]+?src="([^"]+)"', login_page, 'login redirect url')
|
|
self._download_webpage(redirect_url, None, False)
|
|
|
|
def _real_extract(self, url):
|
|
display_id = self._match_id(url)
|
|
|
|
webpage = self._download_webpage(url, display_id)
|
|
|
|
m = re.search(r'You must be logged in to view this video\.', webpage)
|
|
if m:
|
|
self.raise_login_required('This video requires login')
|
|
|
|
video_id = self._search_regex(
|
|
[r"glbUpdViews\s*\('\d*','(\d+)'", r'p/report/video/(\d+)'],
|
|
webpage, 'video id', default=None)
|
|
|
|
title = self._html_search_regex(
|
|
(r'Title:</th><td>([^<]+)</td>', r'<h1[^>]*>(.+?)</h1>'),
|
|
webpage, 'title')
|
|
|
|
info = self._parse_html5_media_entries(url, webpage, video_id)[0]
|
|
|
|
return merge_dicts(info, {
|
|
'id': video_id,
|
|
'display_id': display_id,
|
|
'title': title,
|
|
'age_limit': 18,
|
|
})
|
|
|
|
|
|
class EroProfileAlbumIE(InfoExtractor):
|
|
_VALID_URL = r'https?://(?:www\.)?eroprofile\.com/m/videos/album/(?P<id>[^/]+)'
|
|
IE_NAME = 'EroProfile:album'
|
|
|
|
_TESTS = [{
|
|
'url': 'https://www.eroprofile.com/m/videos/album/BBW-2-893',
|
|
'info_dict': {
|
|
'id': 'BBW-2-893',
|
|
'title': 'BBW 2',
|
|
},
|
|
'playlist_mincount': 486,
|
|
},
|
|
]
|
|
|
|
def _extract_from_page(self, page):
|
|
for url in re.findall(r'href=".*?(/m/videos/view/[^"]+)"', page):
|
|
yield self.url_result(f'https://www.eroprofile.com{url}', EroProfileIE.ie_key())
|
|
|
|
def _entries(self, playlist_id, first_page):
|
|
yield from self._extract_from_page(first_page)
|
|
|
|
page_urls = re.findall(rf'href=".*?(/m/videos/album/{playlist_id}\?pnum=(\d+))"', first_page)
|
|
max_page = max(int(n) for _, n in page_urls)
|
|
|
|
for n in range(2, max_page + 1):
|
|
url = f'https://www.eroprofile.com/m/videos/album/{playlist_id}?pnum={n}'
|
|
yield from self._extract_from_page(
|
|
self._download_webpage(url, playlist_id,
|
|
note=f'Downloading playlist page {int(n) - 1}'))
|
|
|
|
def _real_extract(self, url):
|
|
playlist_id = self._match_id(url)
|
|
first_page = self._download_webpage(url, playlist_id, note='Downloading playlist')
|
|
playlist_title = self._search_regex(
|
|
r'<title>Album: (.*) - EroProfile</title>', first_page, 'playlist_title')
|
|
|
|
return self.playlist_result(self._entries(playlist_id, first_page), playlist_id, playlist_title)
|