mirror of https://git.sr.ht/~tsileo/microblog.pub
63 lines
1.7 KiB
Python
63 lines
1.7 KiB
Python
|
from urllib.parse import urlparse
|
||
|
|
||
|
import httpx
|
||
|
from bs4 import BeautifulSoup # type: ignore
|
||
|
from loguru import logger
|
||
|
|
||
|
from app import config
|
||
|
from app.utils.url import is_url_valid
|
||
|
|
||
|
|
||
|
def _make_abs(url: str | None, parent: str) -> str | None:
|
||
|
if url is None:
|
||
|
return None
|
||
|
|
||
|
if url.startswith("http"):
|
||
|
return url
|
||
|
|
||
|
return (
|
||
|
urlparse(parent)._replace(path=url, params="", query="", fragment="").geturl()
|
||
|
)
|
||
|
|
||
|
|
||
|
async def _discover_webmention_endoint(url: str) -> str | None:
|
||
|
async with httpx.AsyncClient() as client:
|
||
|
try:
|
||
|
resp = await client.get(
|
||
|
url,
|
||
|
headers={
|
||
|
"User-Agent": config.USER_AGENT,
|
||
|
},
|
||
|
follow_redirects=True,
|
||
|
)
|
||
|
resp.raise_for_status()
|
||
|
except (httpx.HTTPError, httpx.HTTPStatusError):
|
||
|
logger.exception(f"Failed to discover webmention endpoint for {url}")
|
||
|
return None
|
||
|
|
||
|
for k, v in resp.links.items():
|
||
|
if k and "webmention" in k:
|
||
|
return _make_abs(resp.links[k].get("url"), url)
|
||
|
|
||
|
soup = BeautifulSoup(resp.text, "html5lib")
|
||
|
wlinks = soup.find_all(["link", "a"], attrs={"rel": "webmention"})
|
||
|
for wlink in wlinks:
|
||
|
if "href" in wlink.attrs:
|
||
|
return _make_abs(wlink.attrs["href"], url)
|
||
|
|
||
|
return None
|
||
|
|
||
|
|
||
|
async def discover_webmention_endpoint(url: str) -> str | None:
|
||
|
"""Discover the Webmention endpoint of a given URL, if any.
|
||
|
|
||
|
Passes all the tests at https://webmention.rocks!
|
||
|
|
||
|
"""
|
||
|
wurl = await _discover_webmention_endoint(url)
|
||
|
if wurl is None:
|
||
|
return None
|
||
|
if not is_url_valid(wurl):
|
||
|
return None
|
||
|
return wurl
|