The proxy support code in urllib is imho too specific and cannot easily support dynamic proxy configuration using a proxy.pac file.

That is, the code assumes that there is at most one proxy per protocol, while this is not quite true in practice.  A proxy.pac can refer different URLs to different proxies (or exclude some from proxying).

This could be solved by adding a new proxy hook to urllib, something like:

def getproxies_for_url(url) -> [info, ...]

That is, the argument is the string for an URL, the result is a list of proxies that can be used to fetch the URL.

This hook would allow us to provide a hook on OSX that enables python scripts to use the system proxy settings, even when those are complex (such as by using proxy autoconfiguration).

I like the idea of having the proxy handler expanded. In fact I suggest to base the idea on libproxy ( http://code.google.com/p/libproxy ) which is available on Linux / openSolaris / Windows and Mac (currently).

Libproxy queries the correct settings to be used from KDE and gnome, depending on the currently running session. A fallback is of course env.

the API of libproxy is very simple and python bindings are available.