summaryrefslogtreecommitdiffstats
path: root/module/plugins/hoster/UploadheroCom.py
diff options
context:
space:
mode:
Diffstat (limited to 'module/plugins/hoster/UploadheroCom.py')
-rw-r--r--module/plugins/hoster/UploadheroCom.py19
1 files changed, 7 insertions, 12 deletions
diff --git a/module/plugins/hoster/UploadheroCom.py b/module/plugins/hoster/UploadheroCom.py
index 189079017..ac0b4b562 100644
--- a/module/plugins/hoster/UploadheroCom.py
+++ b/module/plugins/hoster/UploadheroCom.py
@@ -11,7 +11,7 @@ from module.plugins.internal.SimpleHoster import SimpleHoster, create_getInfo
class UploadheroCom(SimpleHoster):
__name__ = "UploadheroCom"
__type__ = "hoster"
- __version__ = "0.16"
+ __version__ = "0.17"
__pattern__ = r'http://(?:www\.)?uploadhero\.com?/dl/\w+'
@@ -31,11 +31,12 @@ class UploadheroCom(SimpleHoster):
IP_WAIT_PATTERN = r'<span id="minutes">(\d+)</span>.*\s*<span id="seconds">(\d+)</span>'
CAPTCHA_PATTERN = r'"(/captchadl\.php\?\w+)"'
- FREE_URL_PATTERN = r'var magicomfg = \'<a href="(http://[^<>"]*?)"|"(http://storage\d+\.uploadhero\.co/\?d=\w+/[^<>"/]+)"'
- PREMIUM_URL_PATTERN = r'<a href="([^"]+)" id="downloadnow"'
+ LINK_FREE_PATTERN = r'var magicomfg = \'<a href="(http://[^<>"]*?)"|"(http://storage\d+\.uploadhero\.co/\?d=\w+/[^<>"/]+)"'
+ LINK_PREMIUM_PATTERN = r'<a href="([^"]+)" id="downloadnow"'
- def handleFree(self):
+
+ def handleFree(self, pyfile):
self.checkErrors()
m = re.search(self.CAPTCHA_PATTERN, self.html)
@@ -45,8 +46,8 @@ class UploadheroCom(SimpleHoster):
for _i in xrange(5):
captcha = self.decryptCaptcha(captcha_url)
- self.html = self.load(self.pyfile.url, get={"code": captcha})
- m = re.search(self.FREE_URL_PATTERN, self.html)
+ self.html = self.load(pyfile.url, get={"code": captcha})
+ m = re.search(self.LINK_FREE_PATTERN, self.html)
if m:
self.correctCaptcha()
download_url = m.group(1) or m.group(2)
@@ -59,12 +60,6 @@ class UploadheroCom(SimpleHoster):
self.download(download_url)
- def handlePremium(self):
- self.logDebug("%s: Use Premium Account" % self.__name__)
- link = re.search(self.PREMIUM_URL_PATTERN, self.html).group(1)
- self.download(link)
-
-
def checkErrors(self):
m = re.search(self.IP_BLOCKED_PATTERN, self.html)
if m: