summaryrefslogtreecommitdiffstats
path: root/module/plugins/hoster/FourSharedCom.py
diff options
context:
space:
mode:
Diffstat (limited to 'module/plugins/hoster/FourSharedCom.py')
-rw-r--r--module/plugins/hoster/FourSharedCom.py76
1 files changed, 59 insertions, 17 deletions
diff --git a/module/plugins/hoster/FourSharedCom.py b/module/plugins/hoster/FourSharedCom.py
index df3d498bc..03da15fb0 100644
--- a/module/plugins/hoster/FourSharedCom.py
+++ b/module/plugins/hoster/FourSharedCom.py
@@ -2,45 +2,87 @@
# -*- coding: utf-8 -*-
from module.plugins.Hoster import Hoster
+from module.network.RequestFactory import getURL
+import re
+
+def getInfo(urls):
+ result = []
+
+ for url in urls:
+ html = getURL(url, decode=True)
+ if re.search(FourSharedCom.FILE_OFFLINE_PATTERN, html):
+ # File offline
+ result.append((url, 0, 1, url))
+ else:
+ # Get file info
+ name, size = url, 0
+
+ found = re.search(FourSharedCom.FILE_SIZE_PATTERN, html)
+ if found is not None:
+ size, units = float(found.group(1).replace(',','')), found.group(2)
+ size = size * 1024 ** {'KB': 1, 'MB': 2, 'GB': 3}[units]
+
+ found = re.search(FourSharedCom.FILE_NAME_PATTERN, html)
+ if found is not None:
+ name = re.sub(r"&#(\d+).", lambda m: unichr(int(m.group(1))), found.group(1))
+
+ if found or size > 0:
+ result.append((name, size, 2, url))
+ yield result
+
class FourSharedCom(Hoster):
__name__ = "FourSharedCom"
__type__ = "hoster"
__pattern__ = r"http://[\w\.]*?4shared(-china)?\.com/(account/)?(download|get|file|document|photo|video|audio)/.+?/.*"
- __version__ = "0.1"
+ __version__ = "0.2"
__description__ = """4Shared Download Hoster"""
- __author_name__ = ("jeix")
- __author_mail__ = ("jeix@hasnomail.de")
+ __author_name__ = ("jeix", "zoidberg")
+ __author_mail__ = ("jeix@hasnomail.de", "zoidberg@mujmail.cz")
+
+ FILE_NAME_PATTERN = '<meta name="title" content="([^"]+)" />'
+ FILE_SIZE_PATTERN = '<span title="Size: ([0-9,.]+) (KB|MB|GB)">'
+ FILE_OFFLINE_PATTERN = 'The file link that you requested is not valid\.'
def setup(self):
self.multiDL = False
def process(self, pyfile):
- html = self.load(pyfile.url)
- tmp_link = ""
- link = ""
- name = ""
+ self.html = self.load(pyfile.url, decode=True)
+ self.getFileInfo(pyfile)
+ self.handleFree(pyfile)
+
+ def getFileInfo(self, pyfile):
+ if re.search(self.FILE_OFFLINE_PATTERN, self.html): self.offline()
+
+ found = re.search(self.FILE_NAME_PATTERN, self.html)
+ if not found: self.fail("Parse error (file name)")
+ pyfile.name = re.sub(r"&#(\d+).", lambda m: unichr(int(m.group(1))), found.group(1))
+
+ found = re.search(self.FILE_SIZE_PATTERN, self.html)
+ if found is None: self.fail("Parse error (file size)")
+ size, units = float(found.group(1).replace(',','')), found.group(2)
+ pyfile.size = size * 1024 ** {'KB': 1, 'MB': 2, 'GB': 3}[units]
+
+ def handleFree(self, pyfile):
+ tmp_link = link = ""
wait = 20
- for line in html.splitlines():
+ for line in self.html.splitlines():
if "dbtn" in line:
tmp_link = line.split('href="')[1].split('"')[0]
- if '<span id="fileNameTextSpan">' in line:
- name = line.split('<span id="fileNameTextSpan">')[1].split('</span>')[0].strip()
- pyfile.name = name
-
if tmp_link:
- html = self.load(tmp_link).splitlines()
- for i, line in enumerate(html):
+ self.html = self.load(tmp_link).splitlines()
+ for i, line in enumerate(self.html):
if "id='divDLStart'" in line:
- link = html[i+1].split("<a href='")[1].split("'")[0]
+ link = self.html[i+2].split("<a href='")[1].split("'")[0]
elif '<div class="sec">' in line:
wait = int(line.split(">")[1].split("<")[0])
- self.setWait(wait)
- self.wait()
+ self.setWait(wait)
+ self.wait()
if link:
self.download(link)