We have discovered that the handling of the <local> macro in urllib.proxy_bypass is broken. According to the Microsoft documentation for this macro, what should be checked is simply that the host name specified does not contain a period. Since urllib gets its proxy information directly from the Windows registry it would make sense to use the same definitions that Microsoft does. Attached is a patch that does this.

Here is a link to the documentation that specifies this:

http://msdn2.microsoft.com/en-gb/library/aa384098.aspx

Note the section discussing the proxy bypass parameter.

In addition, this patch adds code to strip the entries in the proxy bypass settings of any leading and trailing spaces. Internet Explorer supports this although it does not appear to be officially documented.

From the documentation:

pwszProxyBypass
    [in] A pointer to a string variable that contains an optional list of host names or IP addresses, or both, that should not be routed through the proxy when dwAccessType is set to WINHTTP_ACCESS_TYPE_NAMED_PROXY. The list can contain wildcard characters. Do not use an empty string, because the WinHttpOpen function uses it as the proxy bypass list. If this parameter specifies the "<local>" macro as the only entry, this function bypasses any host name that does not contain a period. If dwAccessType is not set to WINHTTP_ACCESS_TYPE_NAMED_PROXY, this parameter must be set to WINHTTP_NO_PROXY_BYPASS.

Yes,valid patch for urllib.py.

I have fix this issue, in the revision 72155 for py2.7 and revision
72156 for py3k. 
Unfortunately, we have no tests for any proxy handling condition for
Windows and Darwin. This is a very specific case in the proxy handling;
thought let the fix be in (as per the MSDN spec), unittest coverage can
dealt separately.