This commit replaces our WSGI implementation with a new ASGI one,
which then uses `asgiref`'s compatibility mode to still support WSGI applications.
The ASGI implementation is a bit bare-bone, but good enough for our purposes.
The major changes are:
- We now support ASGI apps.
- Instead of taking connections out of mitmproxy's normal processing,
we now just set flow.response and let things continue as usual.
This allows users to see responses in mitmproxy, use the response hook
to modify app responses, etc. Also important for us,
this makes the new implementation work for shenanigans like sans-io.
This allow's trailers without the initial Trailer header announcement. In HTTP/2 the stream ends with any frame containing END_SREAM. In the case of trailers, it is a final HEADERS frame after all the DATA frames. Therefore we do not need to explicitly check for the trailer announcement header, but can simply wait until the response message / stream has ended.
- 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
the HAR file spec (http://www.softwareishard.com/blog/har-12-spec/#timings) states that timings that do not apply for a certain requests should be set to -1, this example may set -1000 as a timings value for certain requests.
This ends up producing invalid HAR files in many cases.
My proposed fix is to assign -1 into the dic and only multiply by 1000 for other values
test_xss_scanner.py was utterly failing because of a trouble (bug?)
with the `monkeypatch` fixture failing to replace `gethostbyname`
with the correct mock function.
Indeed, when stepping through the code, the `gethostbyname` presumably
mocked was reported as a builtin python function. The problem could
then come from the fact that it is hard to monkeypatch builtin function
in python.
Using absolute imports seems to resolve the problem.
