Spaces:
Sleeping
Sleeping
import requests | |
import tldextract | |
from pathlib import Path | |
from utils.enver import enver | |
from utils.logger import logger | |
from networks.filepath_converter import UrlToFilepathConverter | |
from networks.network_configs import IGNORE_HOSTS, REQUESTS_HEADERS | |
class WebpageFetcher: | |
def __init__(self): | |
self.enver = enver | |
self.enver.set_envs(proxies=True) | |
self.filepath_converter = UrlToFilepathConverter() | |
def is_ignored_host(self, url): | |
self.host = tldextract.extract(url).registered_domain | |
if self.host in IGNORE_HOSTS: | |
return True | |
else: | |
return False | |
def send_request(self): | |
self.request_response = requests.get( | |
url=self.url, | |
headers=REQUESTS_HEADERS, | |
proxies=self.enver.requests_proxies, | |
) | |
def save_response(self): | |
if not self.output_path.exists(): | |
self.output_path.parent.mkdir(parents=True, exist_ok=True) | |
logger.success(f"Saving to: [{self.output_path}]") | |
with open(self.output_path, "wb") as wf: | |
wf.write(self.request_response.content) | |
def fetch(self, url, overwrite=False, output_parent=None): | |
self.url = url | |
logger.note(f"Fetching: [{self.url}]") | |
self.output_path = self.filepath_converter.convert( | |
self.url, parent=output_parent | |
) | |
if self.is_ignored_host(self.url): | |
logger.warn(f"Ignore host: [{self.host}]") | |
return self.output_path | |
if self.output_path.exists() and not overwrite: | |
logger.success(f"HTML existed: [{self.output_path}]") | |
else: | |
self.send_request() | |
self.save_response() | |
return self.output_path | |
if __name__ == "__main__": | |
url = ( | |
# "https://stackoverflow.com/questions/295135/turn-a-string-into-a-valid-filename" | |
# "https://www.liaoxuefeng.com/wiki/1016959663602400/1017495723838528" | |
"https://docs.python.org/zh-cn/3/tutorial/interpreter.html" | |
) | |
fetcher = WebpageFetcher() | |
fetcher.fetch(url) | |