2022-08-15 00:57:07 +00:00
|
|
|
import base64
|
|
|
|
import io
|
|
|
|
import json
|
|
|
|
import re
|
2022-08-18 02:47:04 +00:00
|
|
|
from typing import Callable, TypeVar
|
2022-08-15 00:57:07 +00:00
|
|
|
import meme_generator
|
2022-08-18 02:47:04 +00:00
|
|
|
from meme_generator import WebcomicPanel, OneCharacterWebcomicPanel, TwoCharacterWebcomicPanel, TitleCardWebcomicPanel, add_watermark, create_webcomic
|
2022-08-15 00:57:07 +00:00
|
|
|
from RDramaAPIInterface import RDramaAPIInterface
|
|
|
|
from datetime import datetime
|
|
|
|
from os.path import exists, join, realpath, split
|
|
|
|
from threading import Timer
|
|
|
|
from BotModels import Base, Comment, Post, User
|
|
|
|
from sqlalchemy import create_engine
|
|
|
|
from sqlalchemy.orm import Session
|
|
|
|
import os
|
|
|
|
from markdown import markdown
|
|
|
|
from bs4 import BeautifulSoup
|
2022-08-18 02:47:04 +00:00
|
|
|
from utils import get_real_filename
|
2022-08-15 00:57:07 +00:00
|
|
|
|
|
|
|
TEST_MODE = True
|
|
|
|
TEST_AUTH_TOKEN = "ED3eURMKP9FKBFbi-JUxo8MPGWkEihuyIlAUGtVL7xwx0NEy4Nf6J_mxWYTPgAQx1iy1X91hx7PPHyEBS79hvKVIy5DMEzOyAe9PAc5pmqSJlLGq_-ROewMwFzGrqer4"
|
|
|
|
MINUTES_BEFORE_FORCED_SHUTDOWN = 10
|
|
|
|
DB_FILENAME = "automeme_database.db"
|
|
|
|
PAGES_TO_SCAN = 5
|
2022-08-18 02:47:04 +00:00
|
|
|
AUTOMEME_ID = 13427
|
2022-08-15 00:57:07 +00:00
|
|
|
ALLOWED_COMMENTS_PER_POST = 20
|
|
|
|
ALLOWED_COMMENTS_PER_USER_PER_DAY = 20
|
|
|
|
|
2022-08-18 02:47:04 +00:00
|
|
|
EMOJI_REGEX = r":[^ ]*:"
|
|
|
|
IMAGE_REGEX = r"!\[\]\(/images/([1234567890]*)\.webp\)"
|
|
|
|
PARSED_IMAGE_REGEX = r"IMAGE:/images/[1234567890]*\.webp"
|
|
|
|
INJECTABLE_IMAGE_REGEX = r"IMAGE:/images/\1\.webp"
|
2022-08-15 00:57:07 +00:00
|
|
|
|
|
|
|
# rdrama = RDramaAPIInterface(TEST_AUTH_TOKEN, "localhost", sleep=5, https=False)
|
|
|
|
# print(open('emoji_cache/klanjak.webp', 'rb'))
|
|
|
|
|
|
|
|
# image = meme_generator.create_classic_meme_from_emoji("marseyannoyed", "THAT FACE WHEN YOU", "FORGET TO TAKE OUT THE TRASH")
|
|
|
|
# output = io.BytesIO()
|
|
|
|
# image.save(output, format="webp")
|
|
|
|
# file = {'file': ('based.webp', output.getvalue(), 'image/webp')}
|
|
|
|
# rdrama.reply_to_comment_easy(175, 1, "assddsfssdssdsd", file=file)
|
|
|
|
|
2022-08-18 02:47:04 +00:00
|
|
|
def comment_with_image(message, image, comment_id, post_id):
|
2022-08-21 21:46:07 +00:00
|
|
|
file = {'file': ('based.webp', image.get_binary(), 'image/webp')}
|
2022-08-18 02:47:04 +00:00
|
|
|
return rdrama.reply_to_comment_easy(comment_id, post_id, message, file=file)['id']
|
|
|
|
|
|
|
|
class TextLine:
|
|
|
|
def __init__(self, string):
|
|
|
|
self.line = text_elements(string)
|
|
|
|
|
|
|
|
@property
|
|
|
|
def text(self) -> str:
|
|
|
|
text = [i.text for i in self.line if isinstance(i, Text)]
|
|
|
|
return " ".join(text)
|
|
|
|
|
|
|
|
@property
|
|
|
|
def captions(self) -> 'list[Text]':
|
|
|
|
return [i for i in self.line if isinstance(i, Text)]
|
|
|
|
|
|
|
|
@property
|
|
|
|
def images(self) -> 'list[Image]':
|
|
|
|
return [i for i in self.line if isinstance(i, Image)]
|
|
|
|
|
|
|
|
@property
|
|
|
|
def emojis(self) -> 'list[Emoji]':
|
|
|
|
return [i for i in self.line if isinstance(i, Emoji)]
|
|
|
|
|
|
|
|
@property
|
|
|
|
def is_dialogue_line(self):
|
|
|
|
return len(self.emojis) == 1 and len(self.captions) == 1
|
|
|
|
|
|
|
|
@property
|
|
|
|
def is_argument_line(self):
|
|
|
|
return len(self.emojis) == 2 and (len(self.captions) == 2 or len(self.captions) == 1)
|
|
|
|
|
|
|
|
@property
|
|
|
|
def is_pure_text_line(self):
|
|
|
|
return len(self.emojis) == 0 and len(self.images) == 0
|
|
|
|
|
|
|
|
@property
|
|
|
|
def is_big_marsey_line(self):
|
|
|
|
return len(self.emojis) == 1 and self.emojis[0].big and len(self.captions) == 0
|
|
|
|
|
|
|
|
@property
|
|
|
|
def is_image_line(self):
|
|
|
|
return len(self.images) == 1 and len(self.captions) == 0
|
2022-08-15 00:57:07 +00:00
|
|
|
|
|
|
|
class TextElement():
|
|
|
|
pass
|
|
|
|
|
|
|
|
class Text(TextElement):
|
|
|
|
def __init__(self, text):
|
|
|
|
self.text = text
|
|
|
|
|
|
|
|
def __repr__(self) -> str:
|
|
|
|
return f"Text({self.text})"
|
|
|
|
|
2022-08-18 02:47:04 +00:00
|
|
|
class Image(TextElement):
|
|
|
|
def __init__(self, link):
|
|
|
|
self.link = link
|
|
|
|
|
|
|
|
def __repr__(self) -> str:
|
|
|
|
return f"Link({self.link})"
|
|
|
|
|
2022-08-15 00:57:07 +00:00
|
|
|
class Emoji(TextElement):
|
|
|
|
def __init__(self, emoji, big):
|
|
|
|
self.emoji = emoji
|
|
|
|
self.big = big
|
|
|
|
|
|
|
|
def __repr__(self) -> str:
|
|
|
|
return f"Emoji({self.emoji}, big={self.big})"
|
|
|
|
|
|
|
|
def get_text_only(text_elements : list[TextElement]) -> str:
|
|
|
|
text = [i.text for i in text_elements if isinstance(i, Text)]
|
|
|
|
return " ".join(text)
|
|
|
|
|
|
|
|
def text_elements(string : str):
|
2022-08-18 02:47:04 +00:00
|
|
|
FULL_REGEX = rf"({EMOJI_REGEX})|({PARSED_IMAGE_REGEX})"
|
|
|
|
elements = re.split(FULL_REGEX, string)
|
2022-08-15 00:57:07 +00:00
|
|
|
to_return = []
|
|
|
|
for element in elements:
|
2022-08-18 02:47:04 +00:00
|
|
|
if element == None:
|
2022-08-15 00:57:07 +00:00
|
|
|
continue
|
2022-08-18 02:47:04 +00:00
|
|
|
if element.strip() == "":
|
|
|
|
continue
|
|
|
|
if re.match(EMOJI_REGEX, element):
|
2022-08-15 00:57:07 +00:00
|
|
|
if "#" in element:
|
|
|
|
big = True
|
|
|
|
element = element.replace("#","")
|
|
|
|
else:
|
|
|
|
big = False
|
|
|
|
element = element.strip(":")
|
|
|
|
to_return.append(Emoji(element, big))
|
2022-08-18 02:47:04 +00:00
|
|
|
elif re.match(PARSED_IMAGE_REGEX, element):
|
|
|
|
to_return.append(Image(element.strip()[6:]))
|
|
|
|
else:
|
|
|
|
to_return.append(Text(element.strip()))
|
2022-08-15 00:57:07 +00:00
|
|
|
return to_return
|
|
|
|
|
|
|
|
def strip_markdown(markdown_string):
|
2022-08-18 02:47:04 +00:00
|
|
|
markdown_string = re.sub(IMAGE_REGEX, INJECTABLE_IMAGE_REGEX, markdown_string)
|
2022-08-15 00:57:07 +00:00
|
|
|
markdown_string = re.sub(">.*\n", "", markdown_string)
|
|
|
|
try:
|
|
|
|
html = markdown(markdown_string)
|
|
|
|
except AttributeError:
|
|
|
|
html = markdown_string #if there is no markdown in the string you get an error
|
|
|
|
soup = BeautifulSoup(html, "html.parser")
|
|
|
|
text = ''.join(soup.findAll(text=True))
|
|
|
|
text = re.sub(r"!blackjack[^ ]*", "", text)
|
|
|
|
text = re.sub(r"fortune", "", text)
|
|
|
|
text = re.sub(r"factcheck", "", text)
|
|
|
|
text = re.sub("!slots.*?\s", "", text)
|
|
|
|
text = re.sub(r"\"", "'", text)
|
|
|
|
|
|
|
|
# make sure there are only letters in the string.
|
|
|
|
if len(set(list(text.lower())).intersection(set(["a",'b','c','d','e','f','g','h','i','j','k','l','m','n','o','p','q','r','s','t','u','v','w','x','y','z']))) == 0:
|
|
|
|
return ""
|
|
|
|
|
|
|
|
text = re.sub("@(.*?)\s", "", text)
|
|
|
|
text = re.sub("!slots.*?\s", "", text)
|
|
|
|
|
|
|
|
text = re.sub("(?i)trans lives matter", "", text)
|
|
|
|
|
|
|
|
return text
|
|
|
|
|
|
|
|
def remove_duplicates(list):
|
|
|
|
return [json.loads(j) for j in set([json.dumps(i) for i in list])]
|
|
|
|
|
|
|
|
def get_eligible_comments(rdrama : RDramaAPIInterface, session : Session):
|
|
|
|
comments = rdrama.get_comments(number_of_pages=PAGES_TO_SCAN)['data']
|
|
|
|
comments = [comment for comment in comments if not comment['is_bot']] #No bots
|
2022-08-18 02:47:04 +00:00
|
|
|
comments = [comment for comment in comments if not comment['author']['id'] == AUTOMEME_ID] #Don't reply to self
|
2022-08-15 00:57:07 +00:00
|
|
|
comments = [comment for comment in comments if Post.get_number_of_replies(comment['post_id'], session) < ALLOWED_COMMENTS_PER_POST] #Don't spam posts
|
|
|
|
comments = [comment for comment in comments if User.get_number_of_comments(comment['author']['id'], session) < ALLOWED_COMMENTS_PER_USER_PER_DAY] #Don't spam users
|
|
|
|
comments = [comment for comment in comments if Comment.get_comment(comment['id'], session) is None] #Double check that we haven't replied to the comment
|
|
|
|
comments = remove_duplicates(comments) #Remove the duplicates
|
|
|
|
return comments
|
|
|
|
|
2022-08-18 02:47:04 +00:00
|
|
|
T = TypeVar('T')
|
|
|
|
def lambda_count(list : list[T], predicate : 'Callable[[T], bool]' ):
|
|
|
|
return sum(1 for i in list if predicate(i))
|
|
|
|
|
|
|
|
def get_full_rdrama_image_url(partial_url) -> str:
|
|
|
|
if (TEST_MODE):
|
|
|
|
return f"http://localhost{partial_url}"
|
|
|
|
else:
|
|
|
|
return f"https://rdrama.net{partial_url}"
|
|
|
|
|
2022-08-15 00:57:07 +00:00
|
|
|
def main_processing_task(rdrama : RDramaAPIInterface, session : Session):
|
|
|
|
is_chudded = False #Do we have the chud award?
|
|
|
|
can_communicate = True #Can we send any message at all?
|
|
|
|
is_pizzad = False
|
|
|
|
rdrama.get_front_page()
|
2022-08-18 02:47:04 +00:00
|
|
|
automeme_information = rdrama.get_user_information(AUTOMEME_ID)
|
|
|
|
print(f"coins: {automeme_information['coins']} comments: {automeme_information['comment_count']}")
|
|
|
|
for badge in automeme_information['badges']:
|
2022-08-15 00:57:07 +00:00
|
|
|
if (badge['name'] == "Marsey Award"):
|
|
|
|
print("We have the marsey award. STOP.")
|
|
|
|
can_communicate = False
|
|
|
|
if (badge['name'] == "Chud"):
|
|
|
|
print("We have the CHUD award. CONTINUE")
|
|
|
|
is_chudded = True
|
|
|
|
if (badge['name'] == "Bird Site Award"):
|
|
|
|
print("We have the Bird Site Award. STOP.")
|
|
|
|
can_communicate = False
|
|
|
|
if (badge['name'] == "Pizzashill Award"):
|
|
|
|
print("We have the Pizzashill Award. CONTINUE.")
|
|
|
|
is_pizzad = True
|
|
|
|
|
2022-08-18 02:47:04 +00:00
|
|
|
if automeme_information['is_banned']:
|
2022-08-15 00:57:07 +00:00
|
|
|
print("We are banned. STOP.")
|
|
|
|
can_communicate = False
|
|
|
|
|
|
|
|
if can_communicate:
|
|
|
|
eligible_comments = get_eligible_comments(rdrama, session)
|
|
|
|
for eligible_comment in eligible_comments:
|
2022-08-20 22:18:44 +00:00
|
|
|
begin = datetime.now()
|
2022-08-18 02:47:04 +00:00
|
|
|
under_post_limit = Post.get_number_of_replies(eligible_comment['post_id'], session) < ALLOWED_COMMENTS_PER_POST
|
|
|
|
under_user_limit = User.get_number_of_comments(eligible_comment['author']['id'], session) < ALLOWED_COMMENTS_PER_USER_PER_DAY
|
|
|
|
has_not_replied_to_comment = Comment.get_comment(eligible_comment['id'], session) is None
|
|
|
|
|
|
|
|
if (not (under_post_limit and under_user_limit and has_not_replied_to_comment)):
|
|
|
|
continue
|
|
|
|
|
2022-08-15 00:57:07 +00:00
|
|
|
comment_text = eligible_comment['body']
|
|
|
|
cleaned_comment_text = strip_markdown(comment_text)
|
|
|
|
|
|
|
|
comment_lines = cleaned_comment_text.split("\n")
|
|
|
|
comment_lines = [comment_line for comment_line in comment_lines if comment_line != ""]
|
2022-08-18 02:47:04 +00:00
|
|
|
element_lines = [TextLine(line) for line in comment_lines]
|
2022-08-15 00:57:07 +00:00
|
|
|
|
2022-08-18 02:47:04 +00:00
|
|
|
argument_lines_count, dialog_lines_count, text_lines_count, big_marsey_lines_count = 0,0,0,0
|
|
|
|
|
|
|
|
dialog_lines = list(filter(lambda a : a.is_dialogue_line, element_lines))
|
|
|
|
argument_lines = list(filter(lambda a : a.is_argument_line, element_lines))
|
|
|
|
pure_text_lines = list(filter(lambda a : a.is_pure_text_line, element_lines))
|
|
|
|
big_marsey_lines = list(filter(lambda a : a.is_big_marsey_line, element_lines))
|
|
|
|
image_lines = list(filter(lambda a : a.is_image_line, element_lines))
|
2022-08-15 00:57:07 +00:00
|
|
|
|
2022-08-18 02:47:04 +00:00
|
|
|
argument_lines_count = len(argument_lines)
|
|
|
|
dialog_lines_count = len(dialog_lines)
|
|
|
|
pure_text_lines_count = len(pure_text_lines)
|
|
|
|
big_marsey_lines_count = len(big_marsey_lines)
|
|
|
|
image_lines_count = len(image_lines)
|
|
|
|
|
|
|
|
image = None
|
|
|
|
if (dialog_lines_count == 2):
|
|
|
|
#Soy vs Chad
|
|
|
|
line1 = dialog_lines[0]
|
|
|
|
line2 = dialog_lines[1]
|
|
|
|
|
|
|
|
emoji1 = line1.emojis[0].emoji
|
|
|
|
emoji2 = line2.emojis[0].emoji
|
|
|
|
caption1 = line1.text
|
|
|
|
caption2 = line2.text
|
|
|
|
|
|
|
|
image = meme_generator.create_soy_vs_chad_meme(emoji1, emoji2, caption1, caption2)
|
|
|
|
elif (big_marsey_lines_count == 1 and pure_text_lines_count == 1):
|
|
|
|
# Modern Meme with Marsey
|
|
|
|
text_line = pure_text_lines[0]
|
|
|
|
marsey_line = big_marsey_lines[0]
|
|
|
|
|
|
|
|
marsey = marsey_line.emojis[0].emoji
|
|
|
|
caption = text_line.text
|
|
|
|
|
|
|
|
image = meme_generator.create_modern_meme_from_emoji(marsey, caption)
|
|
|
|
elif (image_lines_count == 1 and pure_text_lines_count == 1):
|
|
|
|
# Modern Meme with Image
|
|
|
|
text_line = pure_text_lines[0]
|
|
|
|
image_line = image_lines[0]
|
|
|
|
|
|
|
|
image = image_line.images[0].link
|
|
|
|
full_image_url = get_full_rdrama_image_url(image)
|
|
|
|
caption = text_line.text
|
|
|
|
|
|
|
|
image = meme_generator.create_modern_meme_from_url(full_image_url, caption)
|
|
|
|
elif (big_marsey_lines_count == 1 and pure_text_lines_count == 2):
|
|
|
|
# Classic Meme with big marsey
|
|
|
|
top_text_line = pure_text_lines[0]
|
|
|
|
bottom_text_line = pure_text_lines[1]
|
|
|
|
marsey_line = big_marsey_lines[0]
|
|
|
|
|
|
|
|
emoji = marsey_line.emojis[0].emoji
|
|
|
|
top_caption = top_text_line.text
|
|
|
|
bottom_caption = bottom_text_line.text
|
|
|
|
image = meme_generator.create_classic_meme_from_emoji(emoji, top_caption, bottom_caption)
|
|
|
|
elif (image_lines_count == 1 and pure_text_lines_count == 2):
|
|
|
|
# Classic Meme with Image
|
|
|
|
top_text_line = pure_text_lines[0]
|
|
|
|
bottom_text_line = pure_text_lines[1]
|
|
|
|
image_line = image_lines[0]
|
|
|
|
|
|
|
|
image = image_line.images[0].link
|
|
|
|
full_image_url = get_full_rdrama_image_url(image)
|
|
|
|
top_caption = top_text_line.text
|
|
|
|
bottom_caption = bottom_text_line.text
|
|
|
|
|
|
|
|
image = meme_generator.create_classic_meme_from_url(full_image_url, top_caption, bottom_caption)
|
|
|
|
elif (argument_lines_count >= 1 or dialog_lines_count >= 1):
|
|
|
|
panels : 'list[WebcomicPanel]' = []
|
|
|
|
|
|
|
|
for element_line in element_lines:
|
|
|
|
if element_line.is_dialogue_line:
|
|
|
|
caption = element_line.text
|
|
|
|
emoji = element_line.emojis[0].emoji
|
|
|
|
if len(caption) > 100:
|
|
|
|
in_background = True
|
|
|
|
else:
|
|
|
|
in_background = False
|
|
|
|
oneCharacterWebcomicPanel = OneCharacterWebcomicPanel(emoji, caption, in_background)
|
|
|
|
panels.append(oneCharacterWebcomicPanel)
|
|
|
|
elif element_line.is_argument_line:
|
|
|
|
left_caption = element_line.captions[0].text
|
|
|
|
if len(element_line.captions) == 2:
|
|
|
|
right_caption = element_line.captions[1].text
|
|
|
|
else:
|
|
|
|
right_caption = ""
|
|
|
|
left_emoji = element_line.emojis[0].emoji
|
|
|
|
right_emoji = element_line.emojis[1].emoji
|
|
|
|
twoCharacterWebcomicPanel = TwoCharacterWebcomicPanel(left_emoji, left_caption, right_emoji, right_caption)
|
|
|
|
panels.append(twoCharacterWebcomicPanel)
|
|
|
|
elif element_line.is_pure_text_line:
|
|
|
|
panels.append(TitleCardWebcomicPanel(element_line.text))
|
|
|
|
|
|
|
|
image = create_webcomic(panels)
|
2022-08-15 00:57:07 +00:00
|
|
|
if image != None:
|
2022-08-18 02:47:04 +00:00
|
|
|
image = add_watermark(image, eligible_comment['author']['username'])
|
|
|
|
user_id = eligible_comment['author']['id']
|
|
|
|
parent_comment_id = eligible_comment['id']
|
|
|
|
post_id = eligible_comment['post_id']
|
|
|
|
automeme_comment_id = comment_with_image("yo got a meme for ya nigga", image, eligible_comment['id'], eligible_comment['post_id'])
|
|
|
|
Comment.create_new_comment(parent_comment_id, automeme_comment_id, session)
|
|
|
|
Post.increment_replies(post_id, session)
|
|
|
|
User.increase_number_of_comments(user_id, session)
|
2022-08-20 22:18:44 +00:00
|
|
|
end = datetime.now()
|
|
|
|
print(end-begin)
|
2022-08-15 00:57:07 +00:00
|
|
|
|
|
|
|
if __name__ == "__main__":
|
|
|
|
TEST_AUTH_TOKEN = "ED3eURMKP9FKBFbi-JUxo8MPGWkEihuyIlAUGtVL7xwx0NEy4Nf6J_mxWYTPgAQx1iy1X91hx7PPHyEBS79hvKVIy5DMEzOyAe9PAc5pmqSJlLGq_-ROewMwFzGrqer4"
|
|
|
|
print(f"======= RUNNING AT {datetime.now().hour}:{datetime.now().minute} ======= ")
|
|
|
|
if TEST_MODE:
|
|
|
|
website = "localhost"
|
|
|
|
auth = TEST_AUTH_TOKEN
|
|
|
|
https = False
|
|
|
|
timeout = 1
|
2022-08-18 02:47:04 +00:00
|
|
|
AUTOMEME_ID = 7
|
2022-08-15 00:57:07 +00:00
|
|
|
OPERATOR_ID = 9
|
|
|
|
ACTUALLY_CALL_OPEN_AI = False
|
|
|
|
else:
|
|
|
|
website = "rdrama.net"
|
|
|
|
with open(get_real_filename("rdrama_auth_token"), "r") as f:
|
|
|
|
auth = f.read()
|
|
|
|
https = True
|
|
|
|
timeout = 10
|
|
|
|
rdrama = RDramaAPIInterface(auth, website, timeout, https=https)
|
|
|
|
|
|
|
|
#Set up fail safe
|
|
|
|
def exitfunc():
|
|
|
|
print("*That's it, you're going in the retard squisher.*")
|
|
|
|
print("FAILSAFE FORCED SHUTDOWN", datetime.now())
|
|
|
|
os._exit(0)
|
|
|
|
|
|
|
|
timer = Timer(60*MINUTES_BEFORE_FORCED_SHUTDOWN, exitfunc)
|
|
|
|
timer.start()
|
|
|
|
engine = create_engine(f"sqlite:///{get_real_filename(DB_FILENAME)}")
|
|
|
|
Base.metadata.create_all(engine)
|
|
|
|
|
|
|
|
with Session(engine) as session:
|
|
|
|
main_processing_task(rdrama, session)
|
|
|
|
session.commit()
|
|
|
|
timer.cancel()
|