mirror of
https://github.com/minoplhy/filters-maker.git
synced 2024-12-22 18:18:13 +00:00
121 lines
4.2 KiB
Python
121 lines
4.2 KiB
Python
import os
|
|
import requests
|
|
import re
|
|
|
|
def clear_old_files(incoming):
|
|
try:
|
|
os.remove(incoming)
|
|
except OSError:
|
|
pass
|
|
|
|
def download_filters(url,incoming):
|
|
print("downloading: ",url)
|
|
|
|
get = requests.get(url)
|
|
if get.status_code == requests.codes.ok:
|
|
with open(incoming, 'ab') as f:
|
|
for data in get:
|
|
f.write(data)
|
|
return url
|
|
|
|
def filtering(filters_welcome):
|
|
unwanted = ['#',';','@','$',' NS',' NS','@@||','!']
|
|
print("filtering . . .")
|
|
with open(filters_welcome, 'r') as f:
|
|
lines = f.read().splitlines()
|
|
with open(filters_welcome, 'w') as f:
|
|
for line in lines:
|
|
if not line.startswith((tuple(unwanted))) and line.strip():
|
|
f.write('\n'.join([line + '\n']))
|
|
print("++ successful!")
|
|
f.close()
|
|
|
|
def filteringcon(filters_regex_one):
|
|
print("filtering . . .")
|
|
with open(filters_regex_one) as f:
|
|
file = f.read().split('\n')
|
|
for i in range(len(file)):
|
|
file[i] = re.sub('\s\s+#.*', '', file[i])
|
|
file[i] = re.sub(' CNAME .$', '', file[i])
|
|
file[i] = re.sub(' CNAME . $', '', file[i])
|
|
file[i] = re.sub('^\*.', '', file[i])
|
|
with open(filters_regex_one, 'w') as f1:
|
|
f1.writelines(["%s\n" % item for item in file])
|
|
print("++ successful!")
|
|
f.close()
|
|
|
|
a = ['||','^','|','0.0.0.0 ','0.0.0.0','::1 ','127.0.0.1 ','0','::','::1','127.0.0.1','0','::']
|
|
lst = []
|
|
with open(filters_regex_one, 'r') as f:
|
|
for line in f:
|
|
for word in a:
|
|
if word in line and not line.startswith('#') and line.startswith((tuple(word))) and not '.' in line:
|
|
line = line.replace(word,'')
|
|
elif '.' in line and not line.startswith('.'):
|
|
line = line.replace(line, line)
|
|
lst.append(line)
|
|
with open(filters_regex_one, 'w') as f:
|
|
for line in lst:
|
|
f.write(line)
|
|
f.close()
|
|
|
|
remove_words = ['localhost','localhost.localdomain','local','broadcasthost','loopback','ip6-localnet','ip6-mcastprefix','ip6-allnodes','ip6-allrouters','ip6-allhosts','ip6-loopback',' CNAME rpz-passthru.']
|
|
|
|
with open(filters_regex_one, 'r') as f:
|
|
lines = f.read().splitlines()
|
|
with open(filters_regex_one, 'w') as f:
|
|
for line in lines:
|
|
if not line.endswith((tuple(remove_words))):
|
|
f.write('\n'.join([line + '\n']))
|
|
|
|
with open(filters_regex_one) as f:
|
|
file = f.read().split('\n')
|
|
for i in range(len(file)):
|
|
file[i] = re.sub('\s\s+', ' ', file[i])
|
|
file[i] = re.sub('#..*', '', file[i])
|
|
file[i] = re.sub('CNAME . ;..*', '', file[i])
|
|
file[i] = re.sub(';..*', '', file[i])
|
|
file[i] = re.sub('\A^\.' ,'' ,file[i])
|
|
with open(filters_regex_one, 'w') as f1:
|
|
f1.writelines(["%s\n" % item for item in file])
|
|
f.close()
|
|
|
|
|
|
def killingdup(duplicated_file):
|
|
print('Getting rid of duplicated line')
|
|
with open(duplicated_file, 'r') as f:
|
|
lines = set(f.readlines())
|
|
with open(duplicated_file, 'w') as f:
|
|
f.writelines(set(lines))
|
|
print("++ successful!")
|
|
f.close()
|
|
|
|
def excluded(excluded ,incoming):
|
|
exline = [';','$','@',' IN']
|
|
with open(excluded ,'r') as f:
|
|
exclude = f.read().split()
|
|
with open(incoming ,'r') as f:
|
|
lines = f.read().splitlines() # read lines
|
|
with open(incoming ,'w') as f:
|
|
for line in lines:
|
|
if line.strip() and not line in exclude and not line.startswith(';'):
|
|
f.write('\n'.join([line + '\n']))
|
|
elif line.startswith((tuple(exline))):
|
|
f.write('\n'.join([line + '\n']))
|
|
elif not line.strip():
|
|
f.write('\n'.join([line + '\n']))
|
|
|
|
def blankremover(incoming):
|
|
with open(incoming ,'r') as f:
|
|
lines = f.read().split()
|
|
with open(incoming ,'w') as f:
|
|
for line in lines:
|
|
if line.strip():
|
|
f.write('\n'.join([line + '\n']))
|
|
|
|
def sort(incoming):
|
|
with open(incoming, 'r') as f:
|
|
lines = sorted(f.readlines())
|
|
with open(incoming, 'w') as f:
|
|
for line in lines:
|
|
f.write(line) |