diff options
Diffstat (limited to 'pyload/plugins/hoster/FilerNet.py')
-rw-r--r-- | pyload/plugins/hoster/FilerNet.py | 99 |
1 files changed, 99 insertions, 0 deletions
diff --git a/pyload/plugins/hoster/FilerNet.py b/pyload/plugins/hoster/FilerNet.py new file mode 100644 index 000000000..e34a5799e --- /dev/null +++ b/pyload/plugins/hoster/FilerNet.py @@ -0,0 +1,99 @@ +# -*- coding: utf-8 -*- +# +# Test links: +# http://filer.net/get/ivgf5ztw53et3ogd +# http://filer.net/get/hgo14gzcng3scbvv + +import pycurl +import re + +from urlparse import urljoin + +from pyload.plugins.internal.CaptchaService import ReCaptcha +from pyload.plugins.internal.SimpleHoster import SimpleHoster, create_getInfo + + +class FilerNet(SimpleHoster): + __name__ = "FilerNet" + __type__ = "hoster" + __version__ = "0.07" + + __pattern__ = r'https?://(?:www\.)?filer\.net/get/\w+' + + __description__ = """Filer.net hoster plugin""" + __license__ = "GPLv3" + __authors__ = [("stickell", "l.stickell@yahoo.it")] + + + INFO_PATTERN = r'<h1 class="page-header">Free Download (?P<N>\S+) <small>(?P<S>[\w.]+) (?P<U>[\w^_]+)</small></h1>' + OFFLINE_PATTERN = r'Nicht gefunden' + + LINK_PATTERN = r'href="([^"]+)">Get download</a>' + + + def handleFree(self): + # Wait between downloads + m = re.search(r'musst du <span id="time">(\d+)</span> Sekunden warten', self.html) + if m: + self.retry(wait_time=int(m.group(1)), reason=_("Wait between free downloads")) + + self.html = self.load(self.pyfile.url, decode=True) + + inputs = self.parseHtmlForm(input_names='token')[1] + if 'token' not in inputs: + self.error(_("Unable to detect token")) + token = inputs['token'] + self.logDebug("Token: " + token) + + self.html = self.load(self.pyfile.url, post={'token': token}, decode=True) + + inputs = self.parseHtmlForm(input_names='hash')[1] + if 'hash' not in inputs: + self.error(_("Unable to detect hash")) + hash_data = inputs['hash'] + self.logDebug("Hash: " + hash_data) + + downloadURL = r'' + recaptcha = ReCaptcha(self) + + for _i in xrange(5): + challenge, response = recaptcha.challenge() + post_data = {'recaptcha_challenge_field': challenge, + 'recaptcha_response_field': response, + 'hash': hash_data} + + # Workaround for 0.4.9 just_header issue. In 0.5 clean the code using just_header + self.req.http.c.setopt(pycurl.FOLLOWLOCATION, 0) + self.load(self.pyfile.url, post=post_data) + self.req.http.c.setopt(pycurl.FOLLOWLOCATION, 1) + + if 'location' in self.req.http.header.lower(): + location = re.search(r'location: (\S+)', self.req.http.header, re.I).group(1) + downloadURL = urljoin('http://filer.net', location) + self.correctCaptcha() + break + else: + self.invalidCaptcha() + + if not downloadURL: + self.fail(_("No Download url retrieved/all captcha attempts failed")) + + self.download(downloadURL, disposition=True) + + + def handlePremium(self): + header = self.load(self.pyfile.url, just_header=True) + if 'location' in header: # Direct Download ON + dl = self.pyfile.url + else: # Direct Download OFF + html = self.load(self.pyfile.url) + m = re.search(self.LINK_PATTERN, html) + if m is None: + self.error(_("LINK_PATTERN not found")) + dl = 'http://filer.net' + m.group(1) + + self.logDebug("Direct link: " + dl) + self.download(dl, disposition=True) + + +getInfo = create_getInfo(FilerNet) |