204 lines
8.7 KiB
Python
204 lines
8.7 KiB
Python
import random
|
|
import re
|
|
from random import choice, choices
|
|
from typing import List, Optional, Union
|
|
|
|
from .config.const import *
|
|
|
|
valid_username_chars = 'a-zA-Z0-9_\-'
|
|
valid_username_regex = re.compile("^[a-zA-Z0-9_\-]{3,25}$", flags=re.A)
|
|
mention_regex = re.compile('(^|\s|>)@([a-zA-Z0-9_\-]{1,30})(?![^<]*<\/(code|pre|a)>)', flags=re.A)
|
|
|
|
group_mention_regex = re.compile('(^|\s|>)!([a-z0-9_\-]{3,25})(?![^<]*<\/(code|pre|a)>)', flags=re.A)
|
|
|
|
everyone_regex = re.compile('(^|\s|>)!(everyone)(?![^<]*<\/(code|pre|a)>)', flags=re.A)
|
|
|
|
valid_password_regex = re.compile("^.{8,100}$", flags=re.A)
|
|
|
|
marseyaward_body_regex = re.compile(">[^<\s+]|[^>\s+]<", flags=re.A)
|
|
|
|
marseyaward_title_regex = re.compile("( *<img[^>]+>)+", flags=re.A)
|
|
|
|
|
|
marsey_regex = re.compile("marsey[a-z0-9]{1,24}", flags=re.A)
|
|
tags_regex = re.compile("[a-z0-9: ]{1,200}", flags=re.A)
|
|
hat_regex = re.compile("[a-zA-Z0-9\-() ,_]{1,50}", flags=re.A)
|
|
description_regex = re.compile("[^<>&\n\t]{1,300}", flags=re.A)
|
|
|
|
badge_name_regex = re.compile("[A-Za-z0-9 ]+", flags=re.A)
|
|
|
|
valid_sub_regex = re.compile("^[a-zA-Z0-9_\-]{3,25}$", flags=re.A)
|
|
|
|
query_regex = re.compile("(\w+):(\S+)", flags=re.A)
|
|
|
|
poll_regex = re.compile("\s*\$\$([^\$\n]+)\$\$\s*(?!([^<]*<\/(code|pre|a)>|[^`]*`))", flags=re.A)
|
|
bet_regex = re.compile("\s*!!([^\$\n]+)!!\s*(?!([^<]*<\/(code|pre|a)>|[^`]*`))", flags=re.A)
|
|
choice_regex = re.compile("\s*&&([^\$\n]+)&&\s*(?!([^<]*<\/(code|pre|a)>|[^`]*`))", flags=re.A)
|
|
|
|
html_comment_regex = re.compile("<!--.*-->", flags=re.A)
|
|
|
|
title_regex = re.compile("[^\w ]", flags=re.A)
|
|
|
|
controversial_regex = re.compile('["> ](https:\/\/old\.reddit\.com/r/[a-zA-Z0-9_]{3,20}\/comments\/[\w\-.#&/=\?@%+]{5,250})["< ]', flags=re.A)
|
|
|
|
fishylinks_regex = re.compile("(https?:\/\/)?[\w\-.#&/=\?@%;+,:]{2,10}\.[\w\-.#&/=\?@%;+,:]{2,250}", flags=re.A)
|
|
|
|
spoiler_regex = re.compile('\|\|(.+?)\|\|(?![^<]*<\/(code|pre|a)>)', flags=re.A)
|
|
reddit_regex = re.compile('(^|\s|<p>)\/?((r|u)\/(\w|-){3,25})(?![^<]*<\/(code|pre|a)>)', flags=re.A)
|
|
sub_regex = re.compile('(^|\s|<p>)\/?(h\/(\w|-){3,25})(?![^<]*<\/(code|pre|a)>)', flags=re.A)
|
|
|
|
strikethrough_regex = re.compile('(^|\s|>)~{1,2}([^~]+)~{1,2}(?![^<]*<\/(code|pre|a)>)', flags=re.A)
|
|
|
|
mute_regex = re.compile("\/mute @?([a-z0-9_\-]{3,30}) ([0-9]+)", flags=re.A|re.I)
|
|
|
|
emoji_regex = re.compile(f"<p>\s*(:[!#@]{{0,3}}[{valid_username_chars}]+:\s*)+<\/p>", flags=re.A)
|
|
emoji_regex2 = re.compile(f'(?<!"):([!#@{valid_username_chars}]{{1,36}}?):(?![^<]*<\/(code|pre|a)>)', flags=re.A)
|
|
emoji_regex3 = re.compile(f'(?<!"):([!@{valid_username_chars}]{{1,35}}?):', flags=re.A)
|
|
|
|
snappy_url_regex = re.compile('<a href="(https?:\/\/.+?)".*?>(.+?)<\/a>', flags=re.A)
|
|
snappy_youtube_regex = re.compile('<lite-youtube videoid="(.+?)" params="autoplay=1', flags=re.A)
|
|
|
|
email_regex = re.compile(EMAIL_REGEX_PATTERN, flags=re.A)
|
|
|
|
utm_regex = re.compile('utm_[0-z]+=[0-z_]+&', flags=re.A)
|
|
utm_regex2 = re.compile('[?&]utm_[0-z]+=[0-z_]+', flags=re.A)
|
|
|
|
slur_regex = re.compile(f"<[^>]*>|{slur_single_words}", flags=re.I|re.A)
|
|
slur_regex_upper = re.compile(f"<[^>]*>|{slur_single_words.upper()}", flags=re.A)
|
|
profanity_regex = re.compile(f"<[^>]*>|{profanity_single_words}", flags=re.I|re.A)
|
|
profanity_regex_upper = re.compile(f"<[^>]*>|{profanity_single_words.upper()}", flags=re.A)
|
|
|
|
torture_regex = re.compile('(^|\s)(i|me) ', flags=re.I|re.A)
|
|
torture_regex2 = re.compile("(^|\s)i'm ", flags=re.I|re.A)
|
|
torture_regex_exclude = re.compile('^\s*(>|`|<blockquote>|<codeblock>|<pre>)', flags=re.A)
|
|
|
|
|
|
image_check_regex = re.compile(f'!\[\]\(((?!(https:\/\/([a-z0-9-]+\.)*({hosts})\/|\/)).*?)\)', flags=re.A)
|
|
|
|
video_regex_extensions = '|'.join(VIDEO_FORMATS)
|
|
video_sub_regex = re.compile(f'(<p>[^<]*)(https:\/\/([a-z0-9-]+\.)*({hosts})\/[\w:~,()\-.#&\/=?@%;+]*?\.({video_regex_extensions}))', flags=re.A)
|
|
|
|
audio_regex_extensions = '|'.join(AUDIO_FORMATS)
|
|
audio_sub_regex = re.compile(f'(<p>[^<]*)(https:\/\/([a-z0-9-]+\.)*({hosts})\/[\w:~,()\-.#&\/=?@%;+]*?\.({audio_regex_extensions}))', flags=re.A)
|
|
|
|
image_regex_extensions = '|'.join(IMAGE_FORMATS)
|
|
image_regex = re.compile(f"(^|\s)(https:\/\/[\w\-.#&/=\?@%;+,:]{{5,250}}(\.|\?format=)({image_regex_extensions})((\?|&)[\w\-.#&/=\?@%;+,:]*)?)($|\s)", flags=re.I|re.A)
|
|
|
|
image_regex_extensions_no_gif = image_regex_extensions.replace('|gif', '')
|
|
imgur_regex = re.compile(f'(https:\/\/i\.imgur\.com\/[a-z0-9]+)\.({image_regex_extensions_no_gif})', flags=re.I|re.A)
|
|
|
|
giphy_regex = re.compile('(https:\/\/media\.giphy\.com\/media\/[a-z0-9]+\/giphy)\.gif', flags=re.I|re.A)
|
|
|
|
youtube_regex = re.compile('(<p>[^<]*)(https:\/\/youtube\.com\/watch\?[\w\-.#&/=?@%+]{7,})', flags=re.I|re.A)
|
|
yt_id_regex = re.compile('[a-z0-9-_]{5,20}', flags=re.I|re.A)
|
|
|
|
link_fix_regex = re.compile("(\[.*?\]\()(?!http|\/)(.*?\))(?!([^<]*<\/(code|pre|a)>|[^`]*`))", flags=re.A)
|
|
|
|
css_url_regex = re.compile('url\(\s*[\'"]?(.*?)[\'"]?\s*\)', flags=re.I|re.A)
|
|
|
|
linefeeds_regex = re.compile("([^\n])\n([^\n])", flags=re.A)
|
|
|
|
greentext_regex = re.compile("(\n|^)>([^ >][^\n]*)", flags=re.A)
|
|
|
|
ascii_only_regex = re.compile("[ -~]+", flags=re.A)
|
|
|
|
reddit_to_vreddit_regex = re.compile('(^|>|")https:\/\/old.reddit.com\/(r|u)\/', flags=re.A)
|
|
reddit_domain_regex = re.compile("(^|\s|\()https?:\/\/(reddit\.com|(?:(?:[A-z]{2})(?:-[A-z]{2})" "?|beta|i|m|pay|ssl|www|new|alpha)\.reddit\.com|libredd\.it|teddit\.net)\/(r|u)\/", flags=re.A)
|
|
|
|
color_regex = re.compile("[a-f0-9]{6}", flags=re.A)
|
|
|
|
# lazy match on the .*?, only match if there is trailing stuff
|
|
# Specifically match Snappy's way of formatting, this might break some losers' comments.
|
|
showmore_regex = re.compile(r"^(.*?</p>(?:</li></ul>)?)(\s*<p>.*)", flags=re.A|re.DOTALL)
|
|
|
|
search_token_regex = re.compile('"([^"]*)"|(\S+)', flags=re.A)
|
|
|
|
git_regex = re.compile("ref: (refs/.+)", flags=re.A)
|
|
|
|
pronouns_regex = re.compile("([a-z]{1,5})/[a-z]{1,5}(/[a-z]{1,5})?", flags=re.A|re.I)
|
|
|
|
knowledgebase_page_regex = re.compile("[a-zA-Z0-9_\-]+", flags=re.A)
|
|
|
|
html_title_regex = re.compile("<title>(.{1,200})</title>", flags=re.I)
|
|
|
|
def sub_matcher(match:re.Match, upper=False, replace_with:Union[dict[str, str], dict[str, List[str]]]=SLURS_FOR_REPLACING):
|
|
group_num = 0
|
|
match_str = match.group(group_num)
|
|
if match_str.startswith('<'):
|
|
return match_str
|
|
else:
|
|
repl = replace_with[match_str.lower()]
|
|
return repl if not upper or "<img" in repl else repl.upper()
|
|
|
|
def sub_matcher_upper(match, replace_with:Union[dict[str, str], dict[str, List[str]]]=SLURS_FOR_REPLACING):
|
|
return sub_matcher(match, upper=True, replace_with=replace_with)
|
|
|
|
|
|
# TODO: make censoring a bit better
|
|
def sub_matcher_slurs(match, upper=False):
|
|
return sub_matcher(match, upper, replace_with=SLURS_FOR_REPLACING)
|
|
|
|
def sub_matcher_slurs_upper(match):
|
|
return sub_matcher_slurs(match, upper=True)
|
|
|
|
def sub_matcher_profanities(match, upper=False):
|
|
return sub_matcher(match, upper, replace_with=PROFANITIES)
|
|
|
|
def sub_matcher_profanities_upper(match):
|
|
return sub_matcher_profanities(match, upper=True)
|
|
|
|
def censor_slurs(body:Optional[str], logged_user):
|
|
if not body: return ""
|
|
def replace_re(body:str, regex:re.Pattern, regex_upper:re.Pattern, sub_func, sub_func_upper):
|
|
body = regex_upper.sub(sub_func_upper, body)
|
|
return regex.sub(sub_func, body)
|
|
|
|
if not logged_user or logged_user == 'chat' or logged_user.slurreplacer:
|
|
body = replace_re(body, slur_regex, slur_regex_upper, sub_matcher_slurs, sub_matcher_slurs_upper)
|
|
if SITE_NAME == 'rDrama':
|
|
if not logged_user or logged_user == 'chat' or logged_user.profanityreplacer:
|
|
body = replace_re(body, profanity_regex, profanity_regex_upper, sub_matcher_profanities, sub_matcher_profanities_upper)
|
|
|
|
return body
|
|
|
|
def torture_ap(body, username):
|
|
lines = body.splitlines(keepends=True)
|
|
|
|
for i in range(len(lines)):
|
|
if torture_regex_exclude.match(lines[i]):
|
|
continue
|
|
for k, l in AJ_REPLACEMENTS.items():
|
|
lines[i] = lines[i].replace(k, l)
|
|
lines[i] = torture_regex.sub(rf'\1@{username} ', lines[i])
|
|
lines[i] = torture_regex2.sub(rf'\1@{username} is ', lines[i])
|
|
|
|
return ''.join(lines).strip()
|
|
|
|
|
|
commands = {
|
|
"fortune": FORTUNE_REPLIES,
|
|
"factcheck": FACTCHECK_REPLIES,
|
|
"8ball": EIGHTBALL_REPLIES,
|
|
"roll": range(1, 10000)
|
|
}
|
|
|
|
command_regex = re.compile("(\s|^)#(fortune|factcheck|8ball|roll)", flags=re.A|re.I)
|
|
|
|
def command_regex_matcher(match, upper=False):
|
|
result = str(choice(commands[match.group(2).lower()]))
|
|
if match.group(2) == 'roll':
|
|
color = tuple(choices(range(256), k=3))
|
|
result = f'<b style="color:rgb{color}">Your roll: {result}</b>'
|
|
return match.group(1) + result
|
|
|
|
reason_regex_post = re.compile('(/post/[0-9]+)', flags=re.A)
|
|
reason_regex_comment = re.compile('(/comment/[0-9]+)', flags=re.A)
|
|
|
|
discord_username_regex = re.compile("(\s|^|>).{2,32}#[0-9]{4}(?=[^0-9]|$)", flags=re.A)
|
|
|
|
numbered_list_regex = re.compile('((\s|^)[0-9]+)\. ', flags=re.A)
|
|
|
|
comment_link_regex = re.compile("/[0-9]+$", flags=re.A)
|
|
|
|
media_deletion_regex = re.compile(f"{SITE_FULL}\/(chat_)?(images|videos)\/[0-9]{{11,17}}\.(webp|{video_regex_extensions})", flags=re.A)
|