import subprocess
import PyPDF2
import os
import shutil
import sys
import qrcode
import json
import string
import random
import requests
def clear_workspace(files):
if not files or not isinstance(files, list) or len(files) == 0:
return
for file in files:
if os.path.exists(file):
os.remove(file)
def delete_folder(folder_path):
try:
shutil.rmtree(folder_path)
except Exception as e:
print(f"Error deleting the folder {folder_path}: {e}")
def merge_pdfs(pdf1_path, pdf2_path, output_path):
# open the two PDF files
with open(pdf1_path, 'rb') as file1, open(pdf2_path, 'rb') as file2:
# create PDF-Reader-Objects
pdf1_reader = PyPDF2.PdfReader(file1)
pdf2_reader = PyPDF2.PdfReader(file2)
# create PDF-Writer-Object
pdf_writer = PyPDF2.PdfWriter()
# add pages from the first PDF
for page_num in range(len(pdf1_reader.pages)):
page = pdf1_reader.pages[page_num]
pdf_writer.add_page(page)
# add pages from the second PDF
for page_num in range(len(pdf2_reader.pages)):
page = pdf2_reader.pages[page_num]
pdf_writer.add_page(page)
# save the combined PDF
with open(output_path, 'wb') as output_file:
pdf_writer.write(output_file)
# If a script failes the files are moved to the old_files directory so we need to move them back so that the paths defined in the files are correct
def move_files_back_from_old_files():
# check if old_files directory exists and move files back
if os.path.exists("oldFiles"):
for file in os.listdir("oldFiles"):
shutil.move(os.path.join("oldFiles", file), file)
os.rmdir("oldFiles")
# copy the tasks files from another task to the current runningTasks to execute it
# be aware of that in your work directory the python script has not the same name as the target script because it will be overwriten
def execute_another_group_task(folder_path, script_file_name):
work_directory = os.getcwd() # get current folder path
copy_files(folder_path, work_directory)
clear_workspace(["index.json", "requirements.txt"])
execute_python_file(script_file_name)
def execute_python_file(file_path):
if not os.path.exists(file_path):
print(f"The file {file_path} does not exist.")
sys.exit(1)
if not file_path.endswith('.py'):
print("The specified file is not a Python file.")
sys.exit(1)
try:
# base_dir = os.path.dirname(os.path.abspath(file_path))
# os.chdir(base_dir) # Change to the directory of the script
subprocess.run(['python3.9', os.path.basename(
file_path)] + sys.argv[1:], check=True)
except subprocess.CalledProcessError as e:
print(f"Error executing the file: {e}")
sys.exit(1)
except Exception as e:
print(f"An unexpected error occurred: {e}")
sys.exit(1)
def copy_files(source_folder, destination_folder):
if not os.path.exists(source_folder):
print(f"The source folder {source_folder} does not exist.")
sys.exit(1)
if not os.path.exists(destination_folder):
print(f"The destination folder {destination_folder} does not exist.")
sys.exit(1)
try:
# List all files in the source folder
files = os.listdir(source_folder)
# Copy each file to the destination folder
for file in files:
source_path = os.path.join(source_folder, file)
destination_path = os.path.join(destination_folder, file)
shutil.copy2(source_path, destination_path)
except Exception as e:
print(f"An error occurred while copying files: {e}")
sys.exit(1)
# this will extract the product id from the string which is provided as arg from the select dropdown where the user can select a product
def extract_product_id(product_type_id):
return product_type_id.split(" ")[0].split("#")[1]
def extract_filament_id(filament_type_id):
print("extract_filament_id", filament_type_id)
return filament_type_id.split(" ")[0]
def create_qrcode(qr_code_url, save_path, back_color):
data = qr_code_url
# Generate QR code
qr = qrcode.QRCode(
version=1,
error_correction=qrcode.constants.ERROR_CORRECT_L,
box_size=10,
border=0,
)
qr.add_data(data)
qr.make(fit=True)
# Create an image from the QR Code instance
img = qr.make_image(fill_color="black", back_color=back_color)
# Save image to a file
# like ./test.png
img.save(save_path)
def get_secrets():
with open('../../secrets/secrets.json') as content:
return json.load(content)
# create shopify discount code
# this function will retry 3 times to create a new discount code if the random generate code already exists
def create_shopify_discount_code(shopify_discount_code_price_rule_name):
# load data from secrets
secrets = get_secrets()
shopify_secrets = secrets["shopify"]
shopify_url = shopify_secrets["store_url"]
x_shopify_app_access_token = shopify_secrets["x_shopify_app_access_token"]
shopify_discount_code_price_rule = shopify_secrets["discount_price_rules"][shopify_discount_code_price_rule_name]
characters = string.ascii_uppercase + string.digits
# _createShopifyDiscountCode(secrets, characters, shopify_discount_code_price_rule, 0)
generated_discount_code = None
retries = 0
while (generated_discount_code is None or retries >= 3):
# generate random code
discount_code = "".join(random.choice(characters) for _ in range(12))
# shopify_discount_code_price_rule = shopify_secrets["discount_price_rules"]["order_voucher_10_percent"]
response = requests.post(
url=f"{shopify_url}/admin/api/2024-01/price_rules/{shopify_discount_code_price_rule}/discount_codes.json",
headers={
"X-Shopify-Access-Token": x_shopify_app_access_token,
"Content-Type": "application/json"
},
json={
"discount_code": {"code": discount_code}
})
if response.status_code == 422:
print("Failed to create code, response status code 422. Trying to generate new discount code")
if response.status_code != 201:
sys.exit(f"Failed to create discount code, status code: {response.status_code}")
generated_discount_code = discount_code
retries += 1
if retries >= 3:
sys.exit("Error, tried 3 times to create discount code")
print(f"Created discount code: {generated_discount_code}. Retries: {retries}")
return generated_discount_code
def get_email_footer_html(language, unsubscribe_link):
languageJson = None
with open("../../groupsData/jnx-email-template/lang.json", "r") as file:
languageJson = json.load(file)
with open("../../groupsData/jnx-email-template/footer.html", "r") as file:
content = file.read()
def replace_unsubscribe_text_button(content):
return content.replace("{{unsubscribe_text_button}}", f"{languageJson[language]['unsubscribe_text_button']}")
# loop through all keys in the language json and replace the placeholders in the html file
for key in languageJson[language]:
if(key == "unsubscribe_text_button"):
content = replace_unsubscribe_text_button(content)
else:
value = languageJson[language][key]
value = value.replace("\n", "
")
content = content.replace(f"{{{{{key}}}}}", value)
content = replace_unsubscribe_text_button(content)
content = content.replace("{{unsubscribeLink}}", unsubscribe_link)
return content
def get_email_footer_text(language, unsubscribe_link):
languageJson = None
with open("../../groupsData/jnx-email-template/lang.json", "r") as file:
languageJson = json.load(file)
with open("../../groupsData/jnx-email-template/footer.txt", "r") as file:
content = file.read()
# loop through all keys in the language json and replace the placeholders in the html file
for key in languageJson[language]:
content = content.replace(f"{{{{{key}}}}}", languageJson[language][key])
content = content.replace("{{unsubscribe_text_button}}", languageJson[language]['unsubscribe_text_button'])
content = content.replace("{{unsubscribeLink}}", unsubscribe_link)
return content
def generate_unsubscribe_link(customerId):
print(f"Using CRM unsubscribe link for customer {customerId}")
_linkName = f"JNX: ⛔️⛔️⛔️ WARN - E-Mail unsubscribed ⛔️⛔️⛔️"
_secrets = get_secrets()
crm_endpoint_url = _secrets['admin_dashboard']['url'] + "/api/v1/crm"
headers = {
"X-Api-Key": _secrets["admin_dashboard"]["x_api_key"],
"Content-Type": "application/json"
}
def get_link():
response = requests.get(
url=f"{crm_endpoint_url}/customer/view/{customerId}",
headers=headers,
)
print(f"GetCustomerActivityLinks response status code {response.status_code}")
if response.status_code != 200:
print(f"GetCustomerActivityLinks req error {response.status_code}")
sys.exit(1)
data = response.json()["Links"]
if len(data) == 0:
print("No links found")
return None
# sort data by creation date
data = sorted(data, key=lambda x: x["CreatedAt"], reverse=True)
# find the first link that has a name that starts with "Shopify Order #orderId - price € Gutschein"
createdLink = None
for link in data:
if link["Name"].startswith(_linkName):
createdLink = link
break
if(createdLink is None):
print(f"Unsubscribe link not found, creating new one")
return None
return _secrets["admin_dashboard"]["qr_code_url"] + createdLink['Id']
link = get_link()
if(link is not None):
print(f"Using existing unsubscribe link for customer {customerId}")
return link
print(f"Creating CRM unsubscribe link for customer {customerId}")
response = requests.post(
url=f"{crm_endpoint_url}/links",
headers=headers,
json={
"CustomerId": customerId,
"Name": _linkName,
"Url": "https://jannex.de/unsubscribed"
})
print(f"CreateCrmActivityLink response status code {response.status_code} for {type}")
if response.status_code != 200:
print(f"CreateCrmActivityLink req error {response.status_code}")
sys.exit(1)
link = get_link()
if(link is None):
print("Error creating unsubscribe link")
sys.exit(1)
return link