On Mon, 2010-05-24 at 15:21 +0100, Martyn Russell wrote: > On Mon, 2010-05-24 at 13:59 +0200, Philip Van Hoof wrote: > > On Mon, 2010-05-24 at 13:56 +0200, Alexandre Mazari wrote: > > You can look here how it might look (DBus object is called > > TrackerSteroids in this experiment). Note the "experiment" in the branch > > name. No promises if you start working with this! > > > > http://git.gnome.org/browse/tracker/log/?h=dbus-fd-experiment > > Note, unless there are serious advantages to not using D-Bus, I wouldn't > advocate this. D-Bus is quite well known and people know how to use it. > Creating our own IPC also has the added risk of being error prone (until > stabilised), lacking security features and tested much less than D-Bus.
SmallNote that this branch uses a D-Bus feature, better known as file descriptor passing, supported since D-Bus 1.3.x, combined with pipe2(). So Adrien isn't creating his own IPC here. He's using a feature of D-Bus and sticking to D-Bus. The feature keeps D-Bus' toys in place. It's a technique _in_ D-Bus to pass larger amounts of data way faster than the traditional way of letting D-Bus marshal everything back and forth. Note that gvfs _also_ uses this technique for the same purpose. So it's _also_ not the case that we'd be the only ones in the GNOME stack doing this. It was in fact Alexander Larsson himself suggesting using this technique over the custom UNIX domain socket experiment we first did: http://pvanhoof.be/blog/index.php/2010/05/13/ipc-performance-the-report#comment-2206 Cheers, Philip -- Philip Van Hoof freelance software developer Codeminded BVBA - http://codeminded.be _______________________________________________ tracker-list mailing list tracker-list@gnome.org http://mail.gnome.org/mailman/listinfo/tracker-list
