1
0
mirror of https://github.com/moparisthebest/SickRage synced 2024-11-16 06:15:09 -05:00
SickRage/sickbeard/helpers.py

1418 lines
46 KiB
Python

# Author: Nic Wolfe <nic@wolfeden.ca>
# URL: http://code.google.com/p/sickbeard/
#
# This file is part of SickRage.
#
# SickRage is free software: you can redistribute it and/or modify
# it under the terms of the GNU General Public License as published by
# the Free Software Foundation, either version 3 of the License, or
# (at your option) any later version.
#
# SickRage is distributed in the hope that it will be useful,
# but WITHOUT ANY WARRANTY; without even the implied warranty of
# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
# GNU General Public License for more details.
#
# You should have received a copy of the GNU General Public License
# along with SickRage. If not, see <http://www.gnu.org/licenses/>.
from __future__ import with_statement
import getpass
import os
import re
import shutil
import socket
import stat
import tempfile
import time
import traceback
import urllib
import hashlib
import httplib
import urlparse
import uuid
import base64
import zipfile
import datetime
import sickbeard
import subliminal
import adba
import requests
import requests.exceptions
try:
import json
except ImportError:
from lib import simplejson as json
try:
import xml.etree.cElementTree as etree
except ImportError:
import elementtree.ElementTree as etree
from xml.dom.minidom import Node
from sickbeard.exceptions import MultipleShowObjectsException, ex
from sickbeard import logger, classes
from sickbeard.common import USER_AGENT, mediaExtensions, subtitleExtensions
from sickbeard import db
from sickbeard import encodingKludge as ek
from sickbeard import notifiers
from sickbeard import clients
from cachecontrol import CacheControl, caches
from itertools import izip, cycle
urllib._urlopener = classes.SickBeardURLopener()
def indentXML(elem, level=0):
'''
Does our pretty printing, makes Matt very happy
'''
i = "\n" + level * " "
if len(elem):
if not elem.text or not elem.text.strip():
elem.text = i + " "
if not elem.tail or not elem.tail.strip():
elem.tail = i
for elem in elem:
indentXML(elem, level + 1)
if not elem.tail or not elem.tail.strip():
elem.tail = i
else:
# Strip out the newlines from text
if elem.text:
elem.text = elem.text.replace('\n', ' ')
if level and (not elem.tail or not elem.tail.strip()):
elem.tail = i
def remove_extension(name):
"""
Remove download or media extension from name (if any)
"""
if name and "." in name:
base_name, sep, extension = name.rpartition('.') # @UnusedVariable
if base_name and extension.lower() in ['nzb', 'torrent'] + mediaExtensions:
name = base_name
return name
def remove_non_release_groups(name):
"""
Remove non release groups from name
"""
if name and "-" in name:
name_group = name.rsplit('-', 1)
if name_group[-1].upper() in ["RP", "NZBGEEK"]:
name = name_group[0]
return name
def replaceExtension(filename, newExt):
'''
>>> replaceExtension('foo.avi', 'mkv')
'foo.mkv'
>>> replaceExtension('.vimrc', 'arglebargle')
'.vimrc'
>>> replaceExtension('a.b.c', 'd')
'a.b.d'
>>> replaceExtension('', 'a')
''
>>> replaceExtension('foo.bar', '')
'foo.'
'''
sepFile = filename.rpartition(".")
if sepFile[0] == "":
return filename
else:
return sepFile[0] + "." + newExt
def isSyncFile(filename):
extension = filename.rpartition(".")[2].lower()
if extension == '!sync' or extension == 'lftp-pget-status':
return True
else:
return False
def isMediaFile(filename):
# ignore samples
if re.search('(^|[\W_])(sample\d*)[\W_]', filename, re.I):
return False
# ignore MAC OS's retarded "resource fork" files
if filename.startswith('._'):
return False
sepFile = filename.rpartition(".")
if re.search('extras?$', sepFile[0], re.I):
return False
if sepFile[2].lower() in mediaExtensions:
return True
else:
return False
def isRarFile(filename):
archive_regex = '(?P<file>^(?P<base>(?:(?!\.part\d+\.rar$).)*)\.(?:(?:part0*1\.)?rar)$)'
if re.search(archive_regex, filename):
return True
return False
def isBeingWritten(filepath):
# Return True if file was modified within 60 seconds. it might still be being written to.
ctime = max(ek.ek(os.path.getctime, filepath), ek.ek(os.path.getmtime, filepath))
if ctime > time.time() - 60:
return True
return False
def sanitizeFileName(name):
'''
>>> sanitizeFileName('a/b/c')
'a-b-c'
>>> sanitizeFileName('abc')
'abc'
>>> sanitizeFileName('a"b')
'ab'
>>> sanitizeFileName('.a.b..')
'a.b'
'''
# remove bad chars from the filename
name = re.sub(r'[\\/\*]', '-', name)
name = re.sub(r'[:"<>|?]', '', name)
# remove leading/trailing periods and spaces
name = name.strip(' .')
return name
def _remove_file_failed(file):
try:
ek.ek(os.remove, file)
except:
pass
def findCertainShow(showList, indexerid):
results = []
if showList and indexerid:
results = filter(lambda x: int(x.indexerid) == int(indexerid), showList)
if len(results) == 1:
return results[0]
elif len(results) > 1:
raise MultipleShowObjectsException()
def makeDir(path):
if not ek.ek(os.path.isdir, path):
try:
ek.ek(os.makedirs, path)
# do the library update for synoindex
notifiers.synoindex_notifier.addFolder(path)
except OSError:
return False
return True
def searchDBForShow(regShowName, log=False):
showNames = [re.sub('[. -]', ' ', regShowName)]
yearRegex = "([^()]+?)\s*(\()?(\d{4})(?(2)\))$"
myDB = db.DBConnection()
for showName in showNames:
sqlResults = myDB.select("SELECT * FROM tv_shows WHERE show_name LIKE ?",
[showName])
if len(sqlResults) == 1:
return int(sqlResults[0]["indexer_id"])
else:
# if we didn't get exactly one result then try again with the year stripped off if possible
match = re.match(yearRegex, showName)
if match and match.group(1):
if log:
logger.log(u"Unable to match original name but trying to manually strip and specify show year",
logger.DEBUG)
sqlResults = myDB.select(
"SELECT * FROM tv_shows WHERE (show_name LIKE ?) AND startyear = ?",
[match.group(1) + '%', match.group(3)])
if len(sqlResults) == 0:
if log:
logger.log(u"Unable to match a record in the DB for " + showName, logger.DEBUG)
continue
elif len(sqlResults) > 1:
if log:
logger.log(u"Multiple results for " + showName + " in the DB, unable to match show name",
logger.DEBUG)
continue
else:
return int(sqlResults[0]["indexer_id"])
def searchIndexerForShowID(regShowName, indexer=None, indexer_id=None, ui=None):
showNames = [re.sub('[. -]', ' ', regShowName)]
# Query Indexers for each search term and build the list of results
for i in sickbeard.indexerApi().indexers if not indexer else int(indexer or []):
# Query Indexers for each search term and build the list of results
lINDEXER_API_PARMS = sickbeard.indexerApi(i).api_params.copy()
if ui is not None: lINDEXER_API_PARMS['custom_ui'] = ui
t = sickbeard.indexerApi(i).indexer(**lINDEXER_API_PARMS)
for name in showNames:
logger.log(u"Trying to find " + name + " on " + sickbeard.indexerApi(i).name, logger.DEBUG)
try:
search = t[indexer_id] if indexer_id else t[name]
except:
continue
try:
seriesname = search.seriesname
except:
seriesname = None
try:
series_id = search.id
except:
series_id = None
if not (seriesname and series_id):
continue
if str(name).lower() == str(seriesname).lower and not indexer_id:
return (seriesname, i, int(series_id))
elif int(indexer_id) == int(series_id):
return (seriesname, i, int(indexer_id))
if indexer:
break
return (None, None, None)
def sizeof_fmt(num):
'''
>>> sizeof_fmt(2)
'2.0 bytes'
>>> sizeof_fmt(1024)
'1.0 KB'
>>> sizeof_fmt(2048)
'2.0 KB'
>>> sizeof_fmt(2**20)
'1.0 MB'
>>> sizeof_fmt(1234567)
'1.2 MB'
'''
for x in ['bytes', 'KB', 'MB', 'GB', 'TB']:
if num < 1024.0:
return "%3.1f %s" % (num, x)
num /= 1024.0
def listMediaFiles(path):
if not dir or not ek.ek(os.path.isdir, path):
return []
files = []
for curFile in ek.ek(os.listdir, path):
fullCurFile = ek.ek(os.path.join, path, curFile)
# if it's a folder do it recursively
if ek.ek(os.path.isdir, fullCurFile) and not curFile.startswith('.') and not curFile == 'Extras':
files += listMediaFiles(fullCurFile)
elif isMediaFile(curFile):
files.append(fullCurFile)
return files
def copyFile(srcFile, destFile):
ek.ek(shutil.copyfile, srcFile, destFile)
try:
ek.ek(shutil.copymode, srcFile, destFile)
except OSError:
pass
def moveFile(srcFile, destFile):
try:
ek.ek(os.rename, srcFile, destFile)
fixSetGroupID(destFile)
except OSError:
copyFile(srcFile, destFile)
ek.ek(os.unlink, srcFile)
def link(src, dst):
if os.name == 'nt':
import ctypes
if ctypes.windll.kernel32.CreateHardLinkW(unicode(dst), unicode(src), 0) == 0: raise ctypes.WinError()
else:
os.link(src, dst)
def hardlinkFile(srcFile, destFile):
try:
ek.ek(link, srcFile, destFile)
fixSetGroupID(destFile)
except:
logger.log(u"Failed to create hardlink of " + srcFile + " at " + destFile + ". Copying instead", logger.ERROR)
copyFile(srcFile, destFile)
def symlink(src, dst):
if os.name == 'nt':
import ctypes
if ctypes.windll.kernel32.CreateSymbolicLinkW(unicode(dst), unicode(src), 1 if os.path.isdir(src) else 0) in [0,
1280]: raise ctypes.WinError()
else:
os.symlink(src, dst)
def moveAndSymlinkFile(srcFile, destFile):
try:
ek.ek(os.rename, srcFile, destFile)
fixSetGroupID(destFile)
ek.ek(symlink, destFile, srcFile)
except:
logger.log(u"Failed to create symlink of " + srcFile + " at " + destFile + ". Copying instead", logger.ERROR)
copyFile(srcFile, destFile)
def make_dirs(path):
"""
Creates any folders that are missing and assigns them the permissions of their
parents
"""
logger.log(u"Checking if the path " + path + " already exists", logger.DEBUG)
if not ek.ek(os.path.isdir, path):
# Windows, create all missing folders
if os.name == 'nt' or os.name == 'ce':
try:
logger.log(u"Folder " + path + " didn't exist, creating it", logger.DEBUG)
ek.ek(os.makedirs, path)
except (OSError, IOError), e:
logger.log(u"Failed creating " + path + " : " + ex(e), logger.ERROR)
return False
# not Windows, create all missing folders and set permissions
else:
sofar = ''
folder_list = path.split(os.path.sep)
# look through each subfolder and make sure they all exist
for cur_folder in folder_list:
sofar += cur_folder + os.path.sep
# if it exists then just keep walking down the line
if ek.ek(os.path.isdir, sofar):
continue
try:
logger.log(u"Folder " + sofar + " didn't exist, creating it", logger.DEBUG)
ek.ek(os.mkdir, sofar)
# use normpath to remove end separator, otherwise checks permissions against itself
chmodAsParent(ek.ek(os.path.normpath, sofar))
# do the library update for synoindex
notifiers.synoindex_notifier.addFolder(sofar)
except (OSError, IOError), e:
logger.log(u"Failed creating " + sofar + " : " + ex(e), logger.ERROR)
return False
return True
def rename_ep_file(cur_path, new_path, old_path_length=0):
"""
Creates all folders needed to move a file to its new location, renames it, then cleans up any folders
left that are now empty.
cur_path: The absolute path to the file you want to move/rename
new_path: The absolute path to the destination for the file WITHOUT THE EXTENSION
old_path_length: The length of media file path (old name) WITHOUT THE EXTENSION
"""
new_dest_dir, new_dest_name = os.path.split(new_path) # @UnusedVariable
if old_path_length == 0 or old_path_length > len(cur_path):
# approach from the right
cur_file_name, cur_file_ext = os.path.splitext(cur_path) # @UnusedVariable
else:
# approach from the left
cur_file_ext = cur_path[old_path_length:]
cur_file_name = cur_path[:old_path_length]
if cur_file_ext[1:] in subtitleExtensions:
# Extract subtitle language from filename
sublang = os.path.splitext(cur_file_name)[1][1:]
# Check if the language extracted from filename is a valid language
try:
language = subliminal.language.Language(sublang, strict=True)
cur_file_ext = '.' + sublang + cur_file_ext
except ValueError:
pass
# put the extension on the incoming file
new_path += cur_file_ext
make_dirs(os.path.dirname(new_path))
# move the file
try:
logger.log(u"Renaming file from " + cur_path + " to " + new_path)
ek.ek(os.rename, cur_path, new_path)
except (OSError, IOError), e:
logger.log(u"Failed renaming " + cur_path + " to " + new_path + ": " + ex(e), logger.ERROR)
return False
# clean up any old folders that are empty
delete_empty_folders(ek.ek(os.path.dirname, cur_path))
return True
def delete_empty_folders(check_empty_dir, keep_dir=None):
"""
Walks backwards up the path and deletes any empty folders found.
check_empty_dir: The path to clean (absolute path to a folder)
keep_dir: Clean until this path is reached
"""
# treat check_empty_dir as empty when it only contains these items
ignore_items = []
logger.log(u"Trying to clean any empty folders under " + check_empty_dir)
# as long as the folder exists and doesn't contain any files, delete it
while ek.ek(os.path.isdir, check_empty_dir) and check_empty_dir != keep_dir:
check_files = ek.ek(os.listdir, check_empty_dir)
if not check_files or (len(check_files) <= len(ignore_items) and all(
[check_file in ignore_items for check_file in check_files])):
# directory is empty or contains only ignore_items
try:
logger.log(u"Deleting empty folder: " + check_empty_dir)
# need shutil.rmtree when ignore_items is really implemented
ek.ek(os.rmdir, check_empty_dir)
# do the library update for synoindex
notifiers.synoindex_notifier.deleteFolder(check_empty_dir)
except OSError, e:
logger.log(u"Unable to delete " + check_empty_dir + ": " + repr(e) + " / " + str(e), logger.WARNING)
break
check_empty_dir = ek.ek(os.path.dirname, check_empty_dir)
else:
break
def fileBitFilter(mode):
for bit in [stat.S_IXUSR, stat.S_IXGRP, stat.S_IXOTH, stat.S_ISUID, stat.S_ISGID]:
if mode & bit:
mode -= bit
return mode
def chmodAsParent(childPath):
if os.name == 'nt' or os.name == 'ce':
return
parentPath = ek.ek(os.path.dirname, childPath)
if not parentPath:
logger.log(u"No parent path provided in " + childPath + ", unable to get permissions from it", logger.DEBUG)
return
parentPathStat = ek.ek(os.stat, parentPath)
parentMode = stat.S_IMODE(parentPathStat[stat.ST_MODE])
childPathStat = ek.ek(os.stat, childPath)
childPath_mode = stat.S_IMODE(childPathStat[stat.ST_MODE])
if ek.ek(os.path.isfile, childPath):
childMode = fileBitFilter(parentMode)
else:
childMode = parentMode
if childPath_mode == childMode:
return
childPath_owner = childPathStat.st_uid
user_id = os.geteuid() # @UndefinedVariable - only available on UNIX
if user_id != 0 and user_id != childPath_owner:
logger.log(u"Not running as root or owner of " + childPath + ", not trying to set permissions", logger.DEBUG)
return
try:
ek.ek(os.chmod, childPath, childMode)
logger.log(u"Setting permissions for %s to %o as parent directory has %o" % (childPath, childMode, parentMode),
logger.DEBUG)
except OSError:
logger.log(u"Failed to set permission for %s to %o" % (childPath, childMode), logger.ERROR)
def fixSetGroupID(childPath):
if os.name == 'nt' or os.name == 'ce':
return
parentPath = ek.ek(os.path.dirname, childPath)
parentStat = ek.ek(os.stat, parentPath)
parentMode = stat.S_IMODE(parentStat[stat.ST_MODE])
if parentMode & stat.S_ISGID:
parentGID = parentStat[stat.ST_GID]
childStat = ek.ek(os.stat, childPath)
childGID = childStat[stat.ST_GID]
if childGID == parentGID:
return
childPath_owner = childStat.st_uid
user_id = os.geteuid() # @UndefinedVariable - only available on UNIX
if user_id != 0 and user_id != childPath_owner:
logger.log(u"Not running as root or owner of " + childPath + ", not trying to set the set-group-ID",
logger.DEBUG)
return
try:
ek.ek(os.chown, childPath, -1, parentGID) # @UndefinedVariable - only available on UNIX
logger.log(u"Respecting the set-group-ID bit on the parent directory for %s" % (childPath), logger.DEBUG)
except OSError:
logger.log(
u"Failed to respect the set-group-ID bit on the parent directory for %s (setting group ID %i)" % (
childPath, parentGID), logger.ERROR)
def is_anime_in_show_list():
for show in sickbeard.showList:
if show.is_anime:
return True
return False
def update_anime_support():
sickbeard.ANIMESUPPORT = is_anime_in_show_list()
def get_absolute_number_from_season_and_episode(show, season, episode):
absolute_number = None
if season and episode:
myDB = db.DBConnection()
sql = "SELECT * FROM tv_episodes WHERE showid = ? and season = ? and episode = ?"
sqlResults = myDB.select(sql, [show.indexerid, season, episode])
if len(sqlResults) == 1:
absolute_number = int(sqlResults[0]["absolute_number"])
logger.log(
"Found absolute_number:" + str(absolute_number) + " by " + str(season) + "x" + str(episode),
logger.DEBUG)
else:
logger.log(
"No entries for absolute number in show: " + show.name + " found using " + str(season) + "x" + str(
episode),
logger.DEBUG)
return absolute_number
def get_all_episodes_from_absolute_number(show, absolute_numbers, indexer_id=None):
episodes = []
season = None
if len(absolute_numbers):
if not show and indexer_id:
show = findCertainShow(sickbeard.showList, indexer_id)
if show:
for absolute_number in absolute_numbers:
ep = show.getEpisode(None, None, absolute_number=absolute_number)
if ep:
episodes.append(ep.episode)
season = ep.season # this will always take the last found seson so eps that cross the season border are not handeled well
return (season, episodes)
def sanitizeSceneName(name, ezrss=False):
"""
Takes a show name and returns the "scenified" version of it.
ezrss: If true the scenified version will follow EZRSS's cracksmoker rules as best as possible
Returns: A string containing the scene version of the show name given.
"""
if name:
if not ezrss:
bad_chars = u",:()'!?\u2019"
# ezrss leaves : and ! in their show names as far as I can tell
else:
bad_chars = u",()'?\u2019"
# strip out any bad chars
for x in bad_chars:
name = name.replace(x, "")
# tidy up stuff that doesn't belong in scene names
name = name.replace("- ", ".").replace(" ", ".").replace("&", "and").replace('/', '.')
name = re.sub("\.\.*", ".", name)
if name.endswith('.'):
name = name[:-1]
return name
else:
return ''
def create_https_certificates(ssl_cert, ssl_key):
"""
Create self-signed HTTPS certificares and store in paths 'ssl_cert' and 'ssl_key'
"""
try:
from OpenSSL import crypto # @UnresolvedImport
from lib.certgen import createKeyPair, createCertRequest, createCertificate, TYPE_RSA, \
serial # @UnresolvedImport
except Exception, e:
logger.log(u"pyopenssl module missing, please install for https access", logger.WARNING)
return False
# Create the CA Certificate
cakey = createKeyPair(TYPE_RSA, 1024)
careq = createCertRequest(cakey, CN='Certificate Authority')
cacert = createCertificate(careq, (careq, cakey), serial, (0, 60 * 60 * 24 * 365 * 10)) # ten years
cname = 'SickRage'
pkey = createKeyPair(TYPE_RSA, 1024)
req = createCertRequest(pkey, CN=cname)
cert = createCertificate(req, (cacert, cakey), serial, (0, 60 * 60 * 24 * 365 * 10)) # ten years
# Save the key and certificate to disk
try:
open(ssl_key, 'w').write(crypto.dump_privatekey(crypto.FILETYPE_PEM, pkey))
open(ssl_cert, 'w').write(crypto.dump_certificate(crypto.FILETYPE_PEM, cert))
except:
logger.log(u"Error creating SSL key and certificate", logger.ERROR)
return False
return True
if __name__ == '__main__':
import doctest
doctest.testmod()
def parse_json(data):
"""
Parse json data into a python object
data: data string containing json
Returns: parsed data as json or None
"""
try:
parsedJSON = json.loads(data)
except ValueError, e:
logger.log(u"Error trying to decode json data. Error: " + ex(e), logger.DEBUG)
return None
return parsedJSON
def parse_xml(data, del_xmlns=False):
"""
Parse data into an xml elementtree.ElementTree
data: data string containing xml
del_xmlns: if True, removes xmlns namesspace from data before parsing
Returns: parsed data as elementtree or None
"""
if del_xmlns:
data = re.sub(' xmlns="[^"]+"', '', data)
try:
parsedXML = etree.fromstring(data)
except Exception, e:
logger.log(u"Error trying to parse xml data. Error: " + ex(e), logger.DEBUG)
parsedXML = None
return parsedXML
def get_xml_text(element, mini_dom=False):
"""
Get all text inside a xml element
element: A xml element either created with elementtree.ElementTree or xml.dom.minidom
mini_dom: Default False use elementtree, True use minidom
Returns: text
"""
text = ""
if mini_dom:
node = element
for child in node.childNodes:
if child.nodeType in (Node.CDATA_SECTION_NODE, Node.TEXT_NODE):
text += child.data
else:
if element is not None:
for child in [element] + element.findall('.//*'):
if child.text:
text += child.text
return text.strip()
def backupVersionedFile(old_file, version):
numTries = 0
new_file = old_file + '.' + 'v' + str(version)
while not ek.ek(os.path.isfile, new_file):
if not ek.ek(os.path.isfile, old_file):
logger.log(u"Not creating backup, " + old_file + " doesn't exist", logger.DEBUG)
break
try:
logger.log(u"Trying to back up " + old_file + " to " + new_file, logger.DEBUG)
shutil.copy(old_file, new_file)
logger.log(u"Backup done", logger.DEBUG)
break
except Exception, e:
logger.log(u"Error while trying to back up " + old_file + " to " + new_file + " : " + ex(e), logger.WARNING)
numTries += 1
time.sleep(1)
logger.log(u"Trying again.", logger.DEBUG)
if numTries >= 10:
logger.log(u"Unable to back up " + old_file + " to " + new_file + " please do it manually.", logger.ERROR)
return False
return True
def restoreVersionedFile(backup_file, version):
numTries = 0
new_file, backup_version = os.path.splitext(backup_file)
restore_file = new_file + '.' + 'v' + str(version)
if not ek.ek(os.path.isfile, new_file):
logger.log(u"Not restoring, " + new_file + " doesn't exist", logger.DEBUG)
return False
try:
logger.log(
u"Trying to backup " + new_file + " to " + new_file + "." + "r" + str(version) + " before restoring backup",
logger.DEBUG)
shutil.move(new_file, new_file + '.' + 'r' + str(version))
except Exception, e:
logger.log(
u"Error while trying to backup DB file " + restore_file + " before proceeding with restore: " + ex(e),
logger.WARNING)
return False
while not ek.ek(os.path.isfile, new_file):
if not ek.ek(os.path.isfile, restore_file):
logger.log(u"Not restoring, " + restore_file + " doesn't exist", logger.DEBUG)
break
try:
logger.log(u"Trying to restore " + restore_file + " to " + new_file, logger.DEBUG)
shutil.copy(restore_file, new_file)
logger.log(u"Restore done", logger.DEBUG)
break
except Exception, e:
logger.log(u"Error while trying to restore " + restore_file + ": " + ex(e), logger.WARNING)
numTries += 1
time.sleep(1)
logger.log(u"Trying again.", logger.DEBUG)
if numTries >= 10:
logger.log(u"Unable to restore " + restore_file + " to " + new_file + " please do it manually.",
logger.ERROR)
return False
return True
# try to convert to int, if it fails the default will be returned
def tryInt(s, s_default=0):
try:
return int(s)
except:
return s_default
# generates a md5 hash of a file
def md5_for_file(filename, block_size=2 ** 16):
try:
with open(filename, 'rb') as f:
md5 = hashlib.md5()
while True:
data = f.read(block_size)
if not data:
break
md5.update(data)
f.close()
return md5.hexdigest()
except Exception:
return None
def get_lan_ip():
"""
Simple function to get LAN localhost_ip
http://stackoverflow.com/questions/11735821/python-get-localhost-ip
"""
if os.name != "nt":
import fcntl
import struct
def get_interface_ip(ifname):
s = socket.socket(socket.AF_INET, socket.SOCK_DGRAM)
return socket.inet_ntoa(fcntl.ioctl(s.fileno(), 0x8915, struct.pack('256s',
ifname[:15]))[20:24])
ip = socket.gethostbyname(socket.gethostname())
if ip.startswith("127.") and os.name != "nt":
interfaces = [
"eth0",
"eth1",
"eth2",
"wlan0",
"wlan1",
"wifi0",
"ath0",
"ath1",
"ppp0",
]
for ifname in interfaces:
try:
ip = get_interface_ip(ifname)
print ifname, ip
break
except IOError:
pass
return ip
def check_url(url):
"""
Check if a URL exists without downloading the whole file.
We only check the URL header.
"""
# see also http://stackoverflow.com/questions/2924422
# http://stackoverflow.com/questions/1140661
good_codes = [httplib.OK, httplib.FOUND, httplib.MOVED_PERMANENTLY]
host, path = urlparse.urlparse(url)[1:3] # elems [1] and [2]
try:
conn = httplib.HTTPConnection(host)
conn.request('HEAD', path)
return conn.getresponse().status in good_codes
except StandardError:
return None
"""
Encryption
==========
By Pedro Jose Pereira Vieito <pvieito@gmail.com> (@pvieito)
* If encryption_version==0 then return data without encryption
* The keys should be unique for each device
To add a new encryption_version:
1) Code your new encryption_version
2) Update the last encryption_version available in webserve.py
3) Remember to maintain old encryption versions and key generators for retrocompatibility
"""
# Key Generators
unique_key1 = hex(uuid.getnode() ** 2) # Used in encryption v1
# Encryption Functions
def encrypt(data, encryption_version=0, decrypt=False):
# Version 1: Simple XOR encryption (this is not very secure, but works)
if encryption_version == 1:
if decrypt:
return ''.join(chr(ord(x) ^ ord(y)) for (x, y) in izip(base64.decodestring(data), cycle(unique_key1)))
else:
return base64.encodestring(
''.join(chr(ord(x) ^ ord(y)) for (x, y) in izip(data, cycle(unique_key1)))).strip()
# Version 0: Plain text
else:
return data
def decrypt(data, encryption_version=0):
return encrypt(data, encryption_version, decrypt=True)
def full_sanitizeSceneName(name):
return re.sub('[. -]', ' ', sanitizeSceneName(name)).lower().lstrip()
def _check_against_names(nameInQuestion, show, season=-1):
showNames = []
if season in [-1, 1]:
showNames = [show.name]
showNames.extend(sickbeard.scene_exceptions.get_scene_exceptions(show.indexerid, season=season))
for showName in showNames:
nameFromList = full_sanitizeSceneName(showName)
if nameFromList == nameInQuestion:
return True
return False
def get_show(name, tryIndexers=False):
if not sickbeard.showList:
return
showObj = None
fromCache = False
try:
# check cache for show
cache = sickbeard.name_cache.retrieveNameFromCache(name)
if cache:
fromCache = True
showObj = findCertainShow(sickbeard.showList, int(cache))
if not showObj and tryIndexers:
showObj = findCertainShow(sickbeard.showList,
searchIndexerForShowID(full_sanitizeSceneName(name), ui=classes.ShowListUI)[2])
# add show to cache
if showObj and not fromCache:
sickbeard.name_cache.addNameToCache(name, showObj.indexerid)
except Exception as e:
logger.log(u"Error when attempting to find show: " + name + " in SickRage: " + str(e), logger.DEBUG)
return showObj
def is_hidden_folder(folder):
"""
Returns True if folder is hidden.
On Linux based systems hidden folders start with . (dot)
folder: Full path of folder to check
"""
if ek.ek(os.path.isdir, folder):
if ek.ek(os.path.basename, folder).startswith('.'):
return True
return False
def real_path(path):
"""
Returns: the canonicalized absolute pathname. The resulting path will have no symbolic link, '/./' or '/../' components.
"""
return ek.ek(os.path.normpath, ek.ek(os.path.normcase, ek.ek(os.path.realpath, path)))
def validateShow(show, season=None, episode=None):
indexer_lang = show.lang
try:
lINDEXER_API_PARMS = sickbeard.indexerApi(show.indexer).api_params.copy()
if indexer_lang and not indexer_lang == 'en':
lINDEXER_API_PARMS['language'] = indexer_lang
t = sickbeard.indexerApi(show.indexer).indexer(**lINDEXER_API_PARMS)
if season is None and episode is None:
return t
return t[show.indexerid][season][episode]
except (sickbeard.indexer_episodenotfound, sickbeard.indexer_seasonnotfound):
pass
def set_up_anidb_connection():
if not sickbeard.USE_ANIDB:
logger.log(u"Usage of anidb disabled. Skiping", logger.DEBUG)
return False
if not sickbeard.ANIDB_USERNAME and not sickbeard.ANIDB_PASSWORD:
logger.log(u"anidb username and/or password are not set. Aborting anidb lookup.", logger.DEBUG)
return False
if not sickbeard.ADBA_CONNECTION:
anidb_logger = lambda x: logger.log("ANIDB: " + str(x), logger.DEBUG)
sickbeard.ADBA_CONNECTION = adba.Connection(keepAlive=True, log=anidb_logger)
if not sickbeard.ADBA_CONNECTION.authed():
try:
sickbeard.ADBA_CONNECTION.auth(sickbeard.ANIDB_USERNAME, sickbeard.ANIDB_PASSWORD)
except Exception, e:
logger.log(u"exception msg: " + str(e))
return False
else:
return True
return sickbeard.ADBA_CONNECTION.authed()
def makeZip(fileList, archive):
"""
'fileList' is a list of file names - full path each name
'archive' is the file name for the archive with a full path
"""
try:
a = zipfile.ZipFile(archive, 'w', zipfile.ZIP_DEFLATED)
for f in fileList:
a.write(f)
a.close()
return True
except Exception as e:
logger.log(u"Zip creation error: " + str(e), logger.ERROR)
return False
def extractZip(archive, targetDir):
"""
'fileList' is a list of file names - full path each name
'archive' is the file name for the archive with a full path
"""
try:
if not os.path.exists(targetDir):
os.mkdir(targetDir)
zip_file = zipfile.ZipFile(archive, 'r')
for member in zip_file.namelist():
filename = os.path.basename(member)
# skip directories
if not filename:
continue
# copy file (taken from zipfile's extract)
source = zip_file.open(member)
target = file(os.path.join(targetDir, filename), "wb")
shutil.copyfileobj(source, target)
source.close()
target.close()
zip_file.close()
return True
except Exception as e:
logger.log(u"Zip extraction error: " + str(e), logger.ERROR)
return False
def mapIndexersToShow(showObj):
mapped = {}
# init mapped indexers object
for indexer in sickbeard.indexerApi().indexers:
mapped[indexer] = showObj.indexerid if int(indexer) == int(showObj.indexer) else 0
myDB = db.DBConnection()
sqlResults = myDB.select(
"SELECT * FROM indexer_mapping WHERE indexer_id = ? AND indexer = ?",
[showObj.indexerid, showObj.indexer])
# for each mapped entry
for curResult in sqlResults:
logger.log(u"Found indexer mapping in cache for show: " + showObj.name, logger.DEBUG)
mapped[int(curResult['mindexer'])] = int(curResult['mindexer_id'])
else:
sql_l = []
for indexer in sickbeard.indexerApi().indexers:
if indexer == showObj.indexer:
mapped[indexer] = showObj.indexerid
continue
lINDEXER_API_PARMS = sickbeard.indexerApi(indexer).api_params.copy()
lINDEXER_API_PARMS['custom_ui'] = classes.ShowListUI
t = sickbeard.indexerApi(indexer).indexer(**lINDEXER_API_PARMS)
try:
mapped_show = t[showObj.name]
except sickbeard.indexer_shownotfound:
logger.log(u"Unable to map " + sickbeard.indexerApi(showObj.indexer).name + "->" + sickbeard.indexerApi(
indexer).name + " for show: " + showObj.name + ", skipping it", logger.DEBUG)
continue
if mapped_show and len(mapped_show) == 1:
logger.log(u"Mapping " + sickbeard.indexerApi(showObj.indexer).name + "->" + sickbeard.indexerApi(
indexer).name + " for show: " + showObj.name, logger.DEBUG)
mapped[indexer] = int(mapped_show[0]['id'])
logger.log(u"Adding indexer mapping to DB for show: " + showObj.name, logger.DEBUG)
sql_l.append([
"INSERT OR IGNORE INTO indexer_mapping (indexer_id, indexer, mindexer_id, mindexer) VALUES (?,?,?,?)",
[showObj.indexerid, showObj.indexer, int(mapped_show[0]['id']), indexer]])
if len(sql_l) > 0:
myDB = db.DBConnection()
myDB.mass_action(sql_l)
return mapped
def touchFile(fname, atime=None):
if None != atime:
try:
with file(fname, 'a'):
os.utime(fname, (atime, atime))
return True
except:
logger.log(u"File air date stamping not available on your OS", logger.DEBUG)
pass
return False
def _getTempDir():
"""Returns the [system temp dir]/tvdb_api-u501 (or
tvdb_api-myuser)
"""
if hasattr(os, 'getuid'):
uid = "u%d" % (os.getuid())
else:
# For Windows
try:
uid = getpass.getuser()
except ImportError:
return os.path.join(tempfile.gettempdir(), "sickrage")
return os.path.join(tempfile.gettempdir(), "sickrage-%s" % (uid))
def getURL(url, post_data=None, params=None, headers=None, timeout=30, session=None, json=False):
"""
Returns a byte-string retrieved from the url provider.
"""
# request session
cache_dir = sickbeard.CACHE_DIR or _getTempDir()
session = CacheControl(sess=session, cache=caches.FileCache(os.path.join(cache_dir, 'sessions')))
# request session headers
req_headers = {'User-Agent': USER_AGENT, 'Accept-Encoding': 'gzip,deflate'}
if headers:
req_headers.update(headers)
session.headers.update(req_headers)
# request session ssl verify
session.verify = False
# request session paramaters
session.params = params
try:
# Remove double-slashes from url
parsed = list(urlparse.urlparse(url))
parsed[2] = re.sub("/{2,}", "/", parsed[2]) # replace two or more / with one
url = urlparse.urlunparse(parsed)
# request session proxies
if sickbeard.PROXY_SETTING:
logger.log("Using proxy for url: " + url, logger.DEBUG)
session.proxies = {
"http": sickbeard.PROXY_SETTING,
"https": sickbeard.PROXY_SETTING,
}
# decide if we get or post data to server
if post_data:
resp = session.post(url, data=post_data, timeout=timeout)
else:
resp = session.get(url, timeout=timeout)
if not resp.ok:
logger.log(u"Requested url " + url + " returned status code is " + str(
resp.status_code) + ': ' + clients.http_error_code[resp.status_code], logger.DEBUG)
return
except requests.exceptions.HTTPError, e:
logger.log(u"HTTP error " + str(e.errno) + " while loading URL " + url, logger.WARNING)
return
except requests.exceptions.ConnectionError, e:
logger.log(u"Connection error " + str(e.message) + " while loading URL " + url, logger.WARNING)
return
except requests.exceptions.Timeout, e:
logger.log(u"Connection timed out " + str(e.message) + " while loading URL " + url, logger.WARNING)
return
except Exception:
logger.log(u"Unknown exception while loading URL " + url + ": " + traceback.format_exc(), logger.WARNING)
return
if json:
return resp.json()
return resp.content
def download_file(url, filename, session=None):
# create session
cache_dir = sickbeard.CACHE_DIR or _getTempDir()
session = CacheControl(sess=session, cache=caches.FileCache(os.path.join(cache_dir, 'sessions')))
# request session headers
session.headers.update({'User-Agent': USER_AGENT, 'Accept-Encoding': 'gzip,deflate'})
# request session ssl verify
session.verify = False
# request session streaming
session.stream = True
# request session proxies
if sickbeard.PROXY_SETTING:
logger.log("Using proxy for url: " + url, logger.DEBUG)
session.proxies = {
"http": sickbeard.PROXY_SETTING,
"https": sickbeard.PROXY_SETTING,
}
try:
resp = session.get(url)
if not resp.ok:
logger.log(u"Requested url " + url + " returned status code is " + str(
resp.status_code) + ': ' + clients.http_error_code[resp.status_code], logger.DEBUG)
return False
with open(filename, 'wb') as fp:
for chunk in resp.iter_content(chunk_size=1024):
if chunk:
fp.write(chunk)
fp.flush()
chmodAsParent(filename)
except requests.exceptions.HTTPError, e:
_remove_file_failed(filename)
logger.log(u"HTTP error " + str(e.errno) + " while loading URL " + url, logger.WARNING)
return False
except requests.exceptions.ConnectionError, e:
_remove_file_failed(filename)
logger.log(u"Connection error " + str(e.message) + " while loading URL " + url, logger.WARNING)
return False
except requests.exceptions.Timeout, e:
_remove_file_failed(filename)
logger.log(u"Connection timed out " + str(e.message) + " while loading URL " + url, logger.WARNING)
return False
except EnvironmentError, e:
_remove_file_failed(filename)
logger.log(u"Unable to save the file: " + ex(e), logger.ERROR)
return False
except Exception:
_remove_file_failed(filename)
logger.log(u"Unknown exception while loading URL " + url + ": " + traceback.format_exc(), logger.WARNING)
return False
return True
def clearCache(force=False):
update_datetime = datetime.datetime.now()
# clean out cache directory, remove everything > 12 hours old
if sickbeard.CACHE_DIR:
logger.log(u"Trying to clean cache folder " + sickbeard.CACHE_DIR)
# Does our cache_dir exists
if not ek.ek(os.path.isdir, sickbeard.CACHE_DIR):
logger.log(u"Can't clean " + sickbeard.CACHE_DIR + " if it doesn't exist", logger.WARNING)
else:
max_age = datetime.timedelta(hours=12)
# Get all our cache files
exclude = ['rss', 'images']
for cache_root, cache_dirs, cache_files in os.walk(sickbeard.CACHE_DIR, topdown=True):
cache_dirs[:] = [d for d in cache_dirs if d not in exclude]
for file in cache_files:
cache_file = ek.ek(os.path.join, cache_root, file)
if ek.ek(os.path.isfile, cache_file):
cache_file_modified = datetime.datetime.fromtimestamp(
ek.ek(os.path.getmtime, cache_file))
if force or (update_datetime - cache_file_modified > max_age):
try:
ek.ek(os.remove, cache_file)
except OSError, e:
logger.log(u"Unable to clean " + cache_root + ": " + repr(e) + " / " + str(e),
logger.WARNING)
break
def human(size):
"""
format a size in bytes into a 'human' file size, e.g. bytes, KB, MB, GB, TB, PB
Note that bytes/KB will be reported in whole numbers but MB and above will have greater precision
e.g. 1 byte, 43 bytes, 443 KB, 4.3 MB, 4.43 GB, etc
"""
if size == 1:
# because I really hate unnecessary plurals
return "1 byte"
suffixes_table = [('bytes', 0), ('KB', 0), ('MB', 1), ('GB', 2),('TB', 2), ('PB', 2)]
num = float(size)
for suffix, precision in suffixes_table:
if num < 1024.0:
break
num /= 1024.0
if precision == 0:
formatted_size = "%d" % num
else:
formatted_size = str(round(num, ndigits=precision))
return "%s %s" % (formatted_size, suffix)
def get_size(start_path='.'):
total_size = 0
for dirpath, dirnames, filenames in ek.ek(os.walk, start_path):
for f in filenames:
fp = ek.ek(os.path.join, dirpath, f)
total_size += ek.ek(os.path.getsize, fp)
return total_size