fix AttributeError on connectivity error. closes #38

[deronnax]

No description provided.

[coveralls]

Coverage decreased (-0.21%) to 92.18% when pulling 37dc518 on deronnax:fix_attribute_error_on_error into 7f8c181 on linkedin:master.

[deronnax]

Hu, it exploses on python3, don't know why. I'll investigate this quickly

[catermelon]

Sorry for the delay on getting back to you, lemme know if you need anything - I'm going to be out of town for the next week or so, so responses might be a bit slow. Sorry about that.

[mosbasik]

I'll be porting a project to Python 3 that uses this library and ran into the bug it fixes. I know it's a long shot, but I'd like to see this PR or something like it passing tests. Don't like the verify=False fix (link) very much.

Was going to try and pin down the issue this PR has with Python 3.x myself, but less than a third of the tests passed when run against master or tag 0.6 using Python 2.7.6 and 3.6.2. I guess that's probably the unpinned requirements files, given the age of the project? If anyone has an environment that passes the tests, could you post the working version numbers of both reqs files' libraries?

Haven't dug too deep and am not very pro at Python, but I found two places start looking into the PR's issue.

1. IOError was merged into OSError starting in 3.3 (link), and requests.exceptions.RequestException subclasses IOError (link) so maybe an interface changed there?
2. The tests that failed for 3.3 all involved an HTTPError, and request.exceptions imports urllib3.exceptions.HTTPError as BaseHTTPError (link) because it defines its own HTTPError class as well. (link) I'm not sure how aliases are handled in tracebacks so I'm not sure which of the two classes the failing tests are referring to.