Proxy support

[Josselin Mouette <joss debian org>]

Hi,

currently libsoup-gnome uses libproxy to obtain the proxy configuration.
Having had to deal recently with bugs in proxy support in our company,
we had a look at how libproxy works, and I can say I’m not impressed. At
all.
      * The most obvious issue is that libproxy reimplements parts of
        the HTTP protocol, in an awful way, for things like downloading
        proxy.pac files
      * It has extension modules that link directly to libmozjs,
        libwebkit and whatnot.
      * Since the modules link directly with the libraries, supporting
        libwebkit introduces a circular build-dependency (libproxy →
        libwebkit → libsoup → libproxy) that makes building webkit
        support very hard. This leads to an insane situation where, to
        have a webkit browser support libproxy, it needs indirectly to
        use libmozjs.
      * The GNOME proxy configuration is obtained through gconftool
        commands.

I’m starting to wonder whether it makes any sense to keep using such a
piece of software. Especially, if the proxy resolver was in libsoup
itself, it wouldn’t need to reinvent HTTP but could use the correct
implementation that we already have.

A correct implementation would lie in libsoup-gnome itself and:
      * directly read configuration settings using GConf (or now
        GSettings),
      * provide an extension point for parsing the JS in proxy.pac
        files,
      * ship such an extension along with libwebkit.

What do you think? Are there any ongoing evolutions of libsoup to
improve this situation?

-- 
 .''`.
: :' :     “You would need to ask a lawyer if you don't know
`. `'       that a handshake of course makes a valid contract.”
  `-        --  J???rg Schilling