You can not select more than 25 topics
Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
114 lines
4.2 KiB
114 lines
4.2 KiB
#!/usr/bin/env python
|
|
# -*- coding: utf-8 -*-
|
|
"""
|
|
Website property.
|
|
"""
|
|
try:
|
|
from importlib.resources import files # @UnresolvedImport
|
|
except ImportError:
|
|
from importlib_resources import files # @UnresolvedImport
|
|
|
|
from rebulk.remodule import re
|
|
|
|
from rebulk import Rebulk, Rule, RemoveMatch
|
|
from ..common import seps
|
|
from ..common.formatters import cleanup
|
|
from ..common.pattern import is_disabled
|
|
from ..common.validators import seps_surround
|
|
from ...reutils import build_or_pattern
|
|
|
|
|
|
def website(config):
|
|
"""
|
|
Builder for rebulk object.
|
|
|
|
:param config: rule configuration
|
|
:type config: dict
|
|
:return: Created Rebulk object
|
|
:rtype: Rebulk
|
|
"""
|
|
rebulk = Rebulk(disabled=lambda context: is_disabled(context, 'website'))
|
|
rebulk = rebulk.regex_defaults(flags=re.IGNORECASE).string_defaults(ignore_case=True)
|
|
rebulk.defaults(name="website")
|
|
|
|
with files('guessit.data') as data_files:
|
|
tld_file = data_files.joinpath('tlds-alpha-by-domain.txt').read_text(encoding='utf-8')
|
|
tlds = [
|
|
tld.strip()
|
|
for tld in tld_file.split('\n')
|
|
if '--' not in tld
|
|
][1:] # All registered domain extension
|
|
|
|
safe_tlds = config['safe_tlds'] # For sure a website extension
|
|
safe_subdomains = config['safe_subdomains'] # For sure a website subdomain
|
|
safe_prefix = config['safe_prefixes'] # Those words before a tlds are sure
|
|
website_prefixes = config['prefixes']
|
|
|
|
rebulk.regex(r'(?:[^a-z0-9]|^)((?:'+build_or_pattern(safe_subdomains) +
|
|
r'\.)+(?:[a-z-0-9-]+\.)+(?:'+build_or_pattern(tlds) +
|
|
r'))(?:[^a-z0-9]|$)',
|
|
children=True)
|
|
rebulk.regex(r'(?:[^a-z0-9]|^)((?:'+build_or_pattern(safe_subdomains) +
|
|
r'\.)*[a-z0-9-]+\.(?:'+build_or_pattern(safe_tlds) +
|
|
r'))(?:[^a-z0-9]|$)',
|
|
safe_subdomains=safe_subdomains, safe_tlds=safe_tlds, children=True)
|
|
rebulk.regex(r'(?:[^a-z0-9]|^)((?:'+build_or_pattern(safe_subdomains) +
|
|
r'\.)*[a-z0-9-]+\.(?:'+build_or_pattern(safe_prefix) +
|
|
r'\.)+(?:'+build_or_pattern(tlds) +
|
|
r'))(?:[^a-z0-9]|$)',
|
|
safe_subdomains=safe_subdomains, safe_prefix=safe_prefix, tlds=tlds, children=True)
|
|
|
|
rebulk.string(*website_prefixes,
|
|
validator=seps_surround, private=True, tags=['website.prefix'])
|
|
|
|
class PreferTitleOverWebsite(Rule):
|
|
"""
|
|
If found match is more likely a title, remove website.
|
|
"""
|
|
consequence = RemoveMatch
|
|
|
|
@staticmethod
|
|
def valid_followers(match):
|
|
"""
|
|
Validator for next website matches
|
|
"""
|
|
return match.named('season', 'episode', 'year')
|
|
|
|
def when(self, matches, context):
|
|
to_remove = []
|
|
for website_match in matches.named('website'):
|
|
safe = False
|
|
for safe_start in safe_subdomains + safe_prefix:
|
|
if website_match.value.lower().startswith(safe_start):
|
|
safe = True
|
|
break
|
|
if not safe:
|
|
suffix = matches.next(website_match, PreferTitleOverWebsite.valid_followers, 0)
|
|
if suffix:
|
|
group = matches.markers.at_match(website_match, lambda marker: marker.name == 'group', 0)
|
|
if not group:
|
|
to_remove.append(website_match)
|
|
return to_remove
|
|
|
|
rebulk.rules(PreferTitleOverWebsite, ValidateWebsitePrefix)
|
|
|
|
return rebulk
|
|
|
|
|
|
class ValidateWebsitePrefix(Rule):
|
|
"""
|
|
Validate website prefixes
|
|
"""
|
|
priority = 64
|
|
consequence = RemoveMatch
|
|
|
|
def when(self, matches, context):
|
|
to_remove = []
|
|
for prefix in matches.tagged('website.prefix'):
|
|
website_match = matches.next(prefix, predicate=lambda match: match.name == 'website', index=0)
|
|
if (not website_match or
|
|
matches.holes(prefix.end, website_match.start,
|
|
formatter=cleanup, seps=seps, predicate=lambda match: match.value)):
|
|
to_remove.append(prefix)
|
|
return to_remove
|