The patch for http://bugs.python.org/issue6312 that is included in the large backports patch in our patch set (rev 4) is bad (see my comment on the issue tracker). I also don't think it's a good idea to just blanket include all backports, precisely for this reason: there might be bad patches in there. Can we do something different here? At least the httplib patch causes a real regression (in couchdb-python, in this case).
2.6: r353 3.1: r351
Fixed in 2.6.5-r3 and 3.1.2-r4.