mirror of
https://github.com/sherlock-project/sherlock
synced 2025-02-26 12:07:15 +00:00
31 lines
1.3 KiB
Python
31 lines
1.3 KiB
Python
#!/usr/bin/env python
|
|
# This module generates the listing of supported sites which can be found in
|
|
# sites.md. It also organizes all the sites in alphanumeric order
|
|
import json
|
|
|
|
# Read the data.json file
|
|
with open("../sherlock/resources/data.json", "r", encoding="utf-8") as data_file:
|
|
data: dict = json.load(data_file)
|
|
|
|
# Removes schema-specific keywords for proper processing
|
|
social_networks: dict = dict(data)
|
|
social_networks.pop('$schema', None)
|
|
|
|
# Sort the social networks in alphanumeric order
|
|
social_networks: list = sorted(social_networks.items())
|
|
|
|
# Write the list of supported sites to sites.md
|
|
with open("../sites.md", "w") as site_file:
|
|
site_file.write(f"## List Of Supported Sites ({len(social_networks)} Sites In Total!)\n")
|
|
for social_network, info in social_networks:
|
|
url_main = info["urlMain"]
|
|
is_nsfw = "**(NSFW)**" if info.get("isNSFW") else ""
|
|
site_file.write(f"1.  [{social_network}]({url_main}) {is_nsfw}\n")
|
|
|
|
# Overwrite the data.json file with sorted data
|
|
with open("../sherlock/resources/data.json", "w") as data_file:
|
|
sorted_data = json.dumps(data, indent=2, sort_keys=True)
|
|
data_file.write(sorted_data)
|
|
data_file.write("\n")
|
|
|
|
print("Finished updating supported site listing!")
|