mirror of
https://github.com/Grasscutters/mitmproxy.git
synced 2024-11-25 18:03:50 +00:00
08895e9ba6
- restructure examples (fix #4031) - remove example dependencies from setup.py, we do not need special dependencies for our supported addons. - unify how we generate docs from code - improve example docs
28 lines
1.4 KiB
Python
28 lines
1.4 KiB
Python
# This script determines if request is an HTML webpage and if so seeks out
|
|
# relative links (<a href="./about.html">) and expands them to absolute links
|
|
# In practice this can be used to front an indexing spider that may not have the capability to expand relative page links.
|
|
# Usage: mitmdump -s link_expander.py or mitmproxy -s link_expander.py
|
|
|
|
import re
|
|
from urllib.parse import urljoin
|
|
|
|
|
|
def response(flow):
|
|
|
|
if "Content-Type" in flow.response.headers and flow.response.headers["Content-Type"].find("text/html") != -1:
|
|
pageUrl = flow.request.url
|
|
pageText = flow.response.text
|
|
pattern = (r"<a\s+(?:[^>]*?\s+)?href=(?P<delimiter>[\"'])"
|
|
r"(?P<link>(?!https?:\/\/|ftps?:\/\/|\/\/|#|javascript:|mailto:).*?)(?P=delimiter)")
|
|
rel_matcher = re.compile(pattern, flags=re.IGNORECASE)
|
|
rel_matches = rel_matcher.finditer(pageText)
|
|
map_dict = {}
|
|
for match_num, match in enumerate(rel_matches):
|
|
(delimiter, rel_link) = match.group("delimiter", "link")
|
|
abs_link = urljoin(pageUrl, rel_link)
|
|
map_dict["{0}{1}{0}".format(delimiter, rel_link)] = "{0}{1}{0}".format(delimiter, abs_link)
|
|
for map in map_dict.items():
|
|
pageText = pageText.replace(*map)
|
|
# Uncomment the following to print the expansion mapping
|
|
# print("{0} -> {1}".format(*map))
|
|
flow.response.text = pageText |