diff options
Diffstat (limited to 'pyload/plugin/hoster/FreakshareCom.py')
-rw-r--r-- | pyload/plugin/hoster/FreakshareCom.py | 182 |
1 files changed, 182 insertions, 0 deletions
diff --git a/pyload/plugin/hoster/FreakshareCom.py b/pyload/plugin/hoster/FreakshareCom.py new file mode 100644 index 000000000..64d8f8308 --- /dev/null +++ b/pyload/plugin/hoster/FreakshareCom.py @@ -0,0 +1,182 @@ +# -*- coding: utf-8 -*- + +import re + +from pyload.plugin.Hoster import Hoster +from pyload.plugin.captcha.ReCaptcha import ReCaptcha +from pyload.plugin.internal.SimpleHoster import secondsToMidnight + + +class FreakshareCom(Hoster): + __name = "FreakshareCom" + __type = "hoster" + __version = "0.40" + + __pattern = r'http://(?:www\.)?freakshare\.(net|com)/files/\S*?/' + + __description = """Freakshare.com hoster plugin""" + __license = "GPLv3" + __authors = [("sitacuisses", "sitacuisses@yahoo.de"), + ("spoob", "spoob@pyload.org"), + ("mkaay", "mkaay@mkaay.de"), + ("Toilal", "toilal.dev@gmail.com")] + + + def setup(self): + self.multiDL = False + self.req_opts = [] + + + def process(self, pyfile): + self.pyfile = pyfile + + pyfile.url = pyfile.url.replace("freakshare.net/", "freakshare.com/") + + if self.account: + self.html = self.load(pyfile.url, cookies=False) + pyfile.name = self.get_file_name() + self.download(pyfile.url) + + else: + self.prepare() + self.get_file_url() + + self.download(pyfile.url, post=self.req_opts) + + check = self.checkDownload({"bad" : "bad try", + "paralell" : "> Sorry, you cant download more then 1 files at time. <", + "empty" : "Warning: Unknown: Filename cannot be empty", + "wrong_captcha" : "Wrong Captcha!", + "downloadserver": "No Downloadserver. Please try again later!"}) + + if check == "bad": + self.fail(_("Bad Try")) + + elif check == "paralell": + self.setWait(300, True) + self.wait() + self.retry() + + elif check == "empty": + self.fail(_("File not downloadable")) + + elif check == "wrong_captcha": + self.invalidCaptcha() + self.retry() + + elif check == "downloadserver": + self.retry(5, 15 * 60, _("No Download server")) + + + def prepare(self): + pyfile = self.pyfile + + self.download_html() + + if not self.file_exists(): + self.offline() + + self.setWait(self.get_waiting_time()) + + pyfile.name = self.get_file_name() + pyfile.size = self.get_file_size() + + self.wait() + + return True + + + def download_html(self): + self.load("http://freakshare.com/index.php", {"language": "EN"}) # Set english language in server session + self.html = self.load(self.pyfile.url) + + + def get_file_url(self): + """ returns the absolute downloadable filepath + """ + if not self.html: + self.download_html() + if not self.wantReconnect: + self.req_opts = self.get_download_options() # get the Post options for the Request + #file_url = self.pyfile.url + #return file_url + else: + self.offline() + + + def get_file_name(self): + if not self.html: + self.download_html() + + if not self.wantReconnect: + m = re.search(r"<h1\sclass=\"box_heading\"\sstyle=\"text-align:center;\">([^ ]+)", self.html) + if m: + file_name = m.group(1) + else: + file_name = self.pyfile.url + + return file_name + else: + return self.pyfile.url + + + def get_file_size(self): + size = 0 + if not self.html: + self.download_html() + + if not self.wantReconnect: + m = re.search(r"<h1\sclass=\"box_heading\"\sstyle=\"text-align:center;\">[^ ]+ - ([^ ]+) (\w\w)yte", self.html) + if m: + units = float(m.group(1).replace(",", "")) + pow = {'KB': 1, 'MB': 2, 'GB': 3}[m.group(2)] + size = int(units * (2 ** 20) ** pow) + + return size + + + def get_waiting_time(self): + if not self.html: + self.download_html() + + if "Your Traffic is used up for today" in self.html: + self.wantReconnect = True + return secondsToMidnight(gmt=2) + + timestring = re.search('\s*var\s(?:downloadWait|time)\s=\s(\d*)[\d.]*;', self.html) + if timestring: + return int(timestring.group(1)) + else: + return 60 + + + def file_exists(self): + """ returns True or False + """ + if not self.html: + self.download_html() + if re.search(r"This file does not exist!", self.html): + return False + else: + return True + + + def get_download_options(self): + re_envelope = re.search(r".*?value=\"Free\sDownload\".*?\n*?(.*?<.*?>\n*)*?\n*\s*?</form>", + self.html).group(0) # get the whole request + to_sort = re.findall(r"<input\stype=\"hidden\"\svalue=\"(.*?)\"\sname=\"(.*?)\"\s\/>", re_envelope) + request_options = dict((n, v) for (v, n) in to_sort) + + herewego = self.load(self.pyfile.url, None, request_options) # the actual download-Page + + to_sort = re.findall(r"<input\stype=\".*?\"\svalue=\"(\S*?)\".*?name=\"(\S*?)\"\s.*?\/>", herewego) + request_options = dict((n, v) for (v, n) in to_sort) + + challenge = re.search(r"http://api\.recaptcha\.net/challenge\?k=(\w+)", herewego) + + if challenge: + re_captcha = ReCaptcha(self) + (request_options['recaptcha_challenge_field'], + request_options['recaptcha_response_field']) = re_captcha.challenge(challenge.group(1)) + + return request_options |