Improve no_proxy support

[cveilleux]

Both the py2 and new py3 support for the no_proxy env var use a simple endswith check in bypass_host.

• https://github.com/httplib2/httplib2/blob/master/python2/httplib2/__init__.py#L843
• https://github.com/httplib2/httplib2/blob/master/python3/httplib2/__init__.py#L800

There are no standards about the behaviour of no_proxy env var, but endswith is probably not what you want here.

If domain starts with a dot (example: .example.com):

• Use endswith to match any subdomain (foo.example.com should match)
• Remove the dot and do an exact match (example.com should also match)

If domain does not start with a dot (example: example.com):

• It should be an exact match.

My understanding is that the current implementation of the check will match wrong cases:

no_proxy='bar.com'
bypass_host('foobar.com') # TRUE! 

Might be a good idea to add a test for it.

See this thread: curl/curl#1208

Even the curl people are struggling with understanding / documenting the correct behaviour. There does not seem to be an authoritative document about it.

[temoto]

Fix is merged in 12b26f9 and released on PyPI in 0.11.0