Keep-alive broken in 2.6.1 and 2.6.2 with chunked transfer-encoding

[lmikkelsen]

Starting in 2.6.1 an exception is thrown if a connection with chunked transfer-encoding is reused. I suspect that the bug is in the bundled urllib3 which was upgraded from 1.10.2 to 1.10.3.

#!/usr/bin/env python
import requests
s = requests.Session()
r = s.get('http://httpbin.org/stream/1')
r.json()
r = s.get('http://httpbin.org/stream/1')
r.json()

$ ./test.py
Traceback (most recent call last):
  File "./test.py", line 10, in <module>
    r = s.get('http://httpbin.org/stream/1')
  File "[...]/requests/requests/sessions.py", line 477, in get
    return self.request('GET', url, **kwargs)
  File "[...]/requests/requests/sessions.py", line 465, in request
    resp = self.send(prep, **send_kwargs)
  File "[...]/requests/requests/sessions.py", line 573, in send
    r = adapter.send(request, **kwargs)
  File "[...]/requests/requests/adapters.py", line 415, in send
    raise ConnectionError(err, request=request)
requests.exceptions.ConnectionError: ('Connection aborted.', ResponseNotReady())

[Lukasa]

I suspect that the bug is in the bundled urllib3 which was upgraded from 1.10.2 to 1.10.3.

As do I: can you open this bug over there, please? We'll fix it in that repository.

[sigmavirus24]

We should start preparing requests 2.7.0 with the fix for shazow/urllib3#598

[rfortress]

Is there any workaround for this? I've gone back to 2.6.0 for now.

[Lukasa]

@rfortress I'm afraid not. We're working on this, there are a number of other problems with the chunked change and we'd like to land them at all once if we possibly can.

I'm genuinely sorry for how long this is taking. =(

[ne0ark]

Version 2.5.1 is latest version now? Isn't this version affected by: http://www.cve.mitre.org/cgi-bin/cvename.cgi?name=2015-2296

[EugeneDae]

The latest and affected by the bug in question, is 2.6.2. The latest that is not affected by the bug in question, seems to be 2.6.0 -- and it works well for me.

[Lukasa]

Correct, 2.6.0 is the version to use to avoid this bug.

[Lukasa]

I have just cut requests 2.7.0, which I believe resolves this problem (and several related ones). Thank you all for your patience!