summaryrefslogtreecommitdiffstats
path: root/module/plugins/hoster/ShareRapidCom.py
diff options
context:
space:
mode:
authorGravatar Walter Purcaro <vuolter@gmail.com> 2014-09-08 00:29:57 +0200
committerGravatar Walter Purcaro <vuolter@gmail.com> 2014-09-14 11:02:23 +0200
commit68d662e689cd42687341c550fb6ebb74e6968d21 (patch)
tree486cef41bd928b8db704894233b2cef94a6e346f /module/plugins/hoster/ShareRapidCom.py
parentsave_join -> safe_join & save_path -> safe_filename (diff)
downloadpyload-68d662e689cd42687341c550fb6ebb74e6968d21.tar.xz
module -> pyload
Diffstat (limited to 'module/plugins/hoster/ShareRapidCom.py')
-rw-r--r--module/plugins/hoster/ShareRapidCom.py66
1 files changed, 0 insertions, 66 deletions
diff --git a/module/plugins/hoster/ShareRapidCom.py b/module/plugins/hoster/ShareRapidCom.py
deleted file mode 100644
index 414e92feb..000000000
--- a/module/plugins/hoster/ShareRapidCom.py
+++ /dev/null
@@ -1,66 +0,0 @@
-# -*- coding: utf-8 -*-
-
-import re
-
-from pycurl import HTTPHEADER
-
-from module.network.RequestFactory import getRequest
-from module.plugins.internal.SimpleHoster import SimpleHoster, parseFileInfo
-
-
-def getInfo(urls):
- h = getRequest()
- h.c.setopt(HTTPHEADER,
- ["Accept: text/html",
- "User-Agent: Mozilla/5.0 (Windows NT 6.1; Win64; x64; rv:25.0) Gecko/20100101 Firefox/25.0"])
- for url in urls:
- html = h.load(url, decode=True)
- file_info = parseFileInfo(ShareRapidCom, url, html)
- yield file_info
-
-
-class ShareRapidCom(SimpleHoster):
- __name__ = "ShareRapidCom"
- __type__ = "hoster"
- __version__ = "0.54"
-
- __pattern__ = r'http://(?:www\.)?(share|mega)rapid\.cz/soubor/\d+/.+'
-
- __description__ = """MegaRapid.cz hoster plugin"""
- __author_name__ = ("MikyWoW", "zoidberg", "stickell", "Walter Purcaro")
- __author_mail__ = ("mikywow@seznam.cz", "zoidberg@mujmail.cz", "l.stickell@yahoo.it", "vuolter@gmail.com")
-
- FILE_NAME_PATTERN = r'<h1[^>]*><span[^>]*>(?:<a[^>]*>)?(?P<N>[^<]+)'
- FILE_SIZE_PATTERN = r'<td class="i">Velikost:</td>\s*<td class="h"><strong>\s*(?P<S>[0-9.]+) (?P<U>[kKMG])i?B</strong></td>'
- OFFLINE_PATTERN = ur'Nastala chyba 404|Soubor byl smazán'
-
- SH_CHECK_TRAFFIC = True
-
- LINK_PATTERN = r'<a href="([^"]+)" title="Stahnout">([^<]+)</a>'
- ERR_LOGIN_PATTERN = ur'<div class="error_div"><strong>Stahování je přístupné pouze přihlášeným uživatelům'
- ERR_CREDIT_PATTERN = ur'<div class="error_div"><strong>Stahování zdarma je možné jen přes náš'
-
-
- def setup(self):
- self.chunkLimit = 1
-
- def handlePremium(self):
- try:
- self.html = self.load(self.pyfile.url, decode=True)
- except BadHeader, e:
- self.account.relogin(self.user)
- self.retry(max_tries=3, reason=str(e))
-
- m = re.search(self.LINK_PATTERN, self.html)
- if m:
- link = m.group(1)
- self.logDebug("Premium link: %s" % link)
- self.download(link, disposition=True)
- else:
- if re.search(self.ERR_LOGIN_PATTERN, self.html):
- self.relogin(self.user)
- self.retry(max_tries=3, reason="User login failed")
- elif re.search(self.ERR_CREDIT_PATTERN, self.html):
- self.fail("Not enough credit left")
- else:
- self.fail("Download link not found")