summaryrefslogtreecommitdiffstats
path: root/module/plugins/crypter/LinkdecrypterCom.py
diff options
context:
space:
mode:
Diffstat (limited to 'module/plugins/crypter/LinkdecrypterCom.py')
-rw-r--r--module/plugins/crypter/LinkdecrypterCom.py91
1 files changed, 0 insertions, 91 deletions
diff --git a/module/plugins/crypter/LinkdecrypterCom.py b/module/plugins/crypter/LinkdecrypterCom.py
deleted file mode 100644
index 21f05b962..000000000
--- a/module/plugins/crypter/LinkdecrypterCom.py
+++ /dev/null
@@ -1,91 +0,0 @@
-# -*- coding: utf-8 -*-
-
-import re
-from module.plugins.Crypter import Crypter
-
-
-class LinkdecrypterCom(Crypter):
- __name__ = "LinkdecrypterCom"
- __type__ = "crypter"
- __version__ = "0.27"
-
- __pattern__ = None
-
- __description__ = """Linkdecrypter.com"""
- __author_name__ = ("zoidberg", "flowlee")
- __author_mail__ = ("zoidberg@mujmail.cz", "")
-
- TEXTAREA_PATTERN = r'<textarea name="links" wrap="off" readonly="1" class="caja_des">(.+)</textarea>'
- PASSWORD_PATTERN = r'<input type="text" name="password"'
- CAPTCHA_PATTERN = r'<img class="captcha" src="(.+?)"(.*?)>'
- REDIR_PATTERN = r'<i>(Click <a href="./">here</a> if your browser does not redirect you).</i>'
-
-
- def decrypt(self, pyfile):
-
- self.passwords = self.getPassword().splitlines()
-
- # API not working anymore
- self.urls = self.decryptHTML()
- if not self.urls:
- self.fail('Could not extract any links')
-
- def decryptAPI(self):
-
- get_dict = {"t": "link", "url": self.pyfile.url, "lcache": "1"}
- self.html = self.load('http://linkdecrypter.com/api', get=get_dict)
- if self.html.startswith('http://'):
- return self.html.splitlines()
-
- if self.html == 'INTERRUPTION(PASSWORD)':
- for get_dict['pass'] in self.passwords:
- self.html = self.load('http://linkdecrypter.com/api', get=get_dict)
- if self.html.startswith('http://'):
- return self.html.splitlines()
-
- self.logError('API', self.html)
- if self.html == 'INTERRUPTION(PASSWORD)':
- self.fail("No or incorrect password")
-
- return None
-
- def decryptHTML(self):
-
- retries = 5
-
- post_dict = {"link_cache": "on", "pro_links": self.pyfile.url, "modo_links": "text"}
- self.html = self.load('http://linkdecrypter.com/', post=post_dict, cookies=True, decode=True)
-
- while self.passwords or retries:
- m = re.search(self.TEXTAREA_PATTERN, self.html, flags=re.DOTALL)
- if m:
- return [x for x in m.group(1).splitlines() if '[LINK-ERROR]' not in x]
-
- m = re.search(self.CAPTCHA_PATTERN, self.html)
- if m:
- captcha_url = 'http://linkdecrypter.com/' + m.group(1)
- result_type = "positional" if "getPos" in m.group(2) else "textual"
-
- m = re.search(r"<p><i><b>([^<]+)</b></i></p>", self.html)
- msg = m.group(1) if m else ""
- self.logInfo("Captcha protected link", result_type, msg)
-
- captcha = self.decryptCaptcha(captcha_url, result_type=result_type)
- if result_type == "positional":
- captcha = "%d|%d" % captcha
- self.html = self.load('http://linkdecrypter.com/', post={"captcha": captcha}, decode=True)
- retries -= 1
-
- elif self.PASSWORD_PATTERN in self.html:
- if self.passwords:
- password = self.passwords.pop(0)
- self.logInfo("Password protected link, trying " + password)
- self.html = self.load('http://linkdecrypter.com/', post={'password': password}, decode=True)
- else:
- self.fail("No or incorrect password")
-
- else:
- retries -= 1
- self.html = self.load('http://linkdecrypter.com/', cookies=True, decode=True)
-
- return None