classification
Title: httplib mishandles unknown 1XX status codes
Type: behavior Stage:
Components: Library (Lib) Versions: Python 3.7
process
Status: open Resolution:
Dependencies: Superseder:
Assigned To: Nosy List: Lukasa, SilentGhost, martin.panter
Priority: normal Keywords: patch

Created on 2016-10-31 19:51 by Lukasa, last changed 2016-10-31 23:57 by martin.panter.

Files
File name Uploaded Description Edit
28570.diff SilentGhost, 2016-10-31 20:38 review
Messages (4)
msg279823 - (view) Author: Cory Benfield (Lukasa) * Date: 2016-10-31 19:51
Long story short
~~~~~~~~~~~~~~~~

http.client does not tolerate non-100 or 101 status codes from the 1XX block in a sensible manner: it treats them as final responses, rather than as provisional ones.


Expected behaviour
~~~~~~~~~~~~~~~~~~

When http.client receives a 1XX status code that it does not recognise, it should either surface these to a user that expects them by means of a callback, or it should ignore them and only show the user the eventual final status code. This is required by RFC 7231 Section 6.2 (Informational 1xx), which reads:

> A client MUST be able to parse one or more 1xx responses received prior to a final response, even if the client does not expect one. A user agent MAY ignore unexpected 1xx responses.


Actual behaviour
~~~~~~~~~~~~~~~~

http.client treats the 1XX status code as final. It parses the header block as though it were a response in the 2XX or higher range. http.client will assume that there is no content in the 1XX response, and so will leave the following final response unconsumed in the socket buffer. This means that if the HTTPConnection is re-used, the user will see the final response following the 1XX as the response to the next request, even though it is not.


Steps to reproduce
~~~~~~~~~~~~~~~~~~

The following "server" can demonstrate the issue:

import socket
import time

document = b'''<!DOCTYPE html>
<html lang="en">
  <head>
    <meta charset="utf-8">
    <title>title</title>
    <link rel="stylesheet" href="/other/styles.css">
    <script src="/other/action.js"></script>
  </head>
  <body>
    <h1>Hello, world!</h1>
  </body>
</html>
'''

s = socket.socket()
s.setsockopt(socket.SOL_SOCKET, socket.SO_REUSEADDR, 1)
s.bind(('localhost', 8080))
s.listen(5)

while True:
    new_socket, _ = s.accept()
    data = b''

    while not data.endswith(b'\r\n\r\n'):
        data += new_socket.recv(8192)

    new_socket.sendall(
        b'HTTP/1.1 103 Early Hints\r\n'
        b'Server: socketserver/1.0.0\r\n'
        b'Link: </other/styles.css>; rel=preload; as=style\r\n'
        b'Link: </other/action.js>; rel=preload; as=script\r\n'
        b'\r\n'
    )
    time.sleep(1)
    new_socket.sendall(
        b'HTTP/1.1 200 OK\r\n'
        b'Server: socketserver/1.0.0\r\n'
        b'Content-Type: text/html\r\n'
        b'Content-Length: %s\r\n'
        b'Link: </other/styles.css>; rel=preload; as=style\r\n'
        b'Link: </other/action.js>; rel=preload; as=script\r\n'
        b'Connection: close\r\n'
        b'\r\n' % len(document)
    )
    new_socket.sendall(document)
    new_socket.close()


If this server is run directly, the following client can be used to test it:

from http.client import HTTPConnection

c = HTTPConnection('localhost', 8080)
c.request('GET', '/')
r = c.getresponse()

print("Status: {} {}".format(r.status, r.reason))
print("Headers: {}".format(r.getheaders()))
print("Body: {}".format(r.read()))


This client will print

Status: 103 Early Hints
Headers: [('Content-Length', '0'), ('Server', 'socketserver/1.0.0'), ('Link', '</other/styles.css>; rel=preload; as=style'), ('Link', '</other/action.js>; rel=preload; as=script')]
Body: b''


This response is wrong: the 200 header block is hidden from the client. Unfortunately, http.client doesn't allow us to ask for the next response: another call to "getresponse()" raises a "ResponseNotReady: Idle" exception.
msg279826 - (view) Author: SilentGhost (SilentGhost) * (Python triager) Date: 2016-10-31 20:38
The fix could be as small as the attached patched, though I'm not sure that is the correct way of handling 101 code.
msg279827 - (view) Author: Cory Benfield (Lukasa) * Date: 2016-10-31 20:49
101 should probably be special-cased, because in that case the underlying protocol is being totally changed.
msg279841 - (view) Author: Martin Panter (martin.panter) * (Python committer) Date: 2016-10-31 23:57
Consuming and ignoring 1xx responses seems reasonable for general cases. Treating 101 (Switching Protocols) as a special case also seems reasonable.

I also agree it would be good to provide an API to return these responses (at least after the request is completely sent). Perhaps a new flag, c.getresponse(informational=True), or a new method, c.get_informational_response(timeout=...). The timeout is for servers that do not support “Expect: 100-continue” mode; see Issue 1346874.

There is also Issue 25919 discussing how to handle responses (including 1xx) while concurrently sending the request. That is a harder problem, but may be solvable with a few new APIs.
History
Date User Action Args
2016-10-31 23:57:08martin.pantersetnosy: + martin.panter
messages: + msg279841
2016-10-31 20:49:50Lukasasetmessages: + msg279827
2016-10-31 20:38:01SilentGhostsetfiles: + 28570.diff

nosy: + SilentGhost
messages: + msg279826

keywords: + patch
2016-10-31 19:51:23Lukasacreate