[13:20] <ion_> ADHOS: Attention-Deficit Hyperac... Ooh, Shiny!
[14:04] <mbiebl> Keybuk: hi
[14:12] <Keybuk> hay
[14:12] <Keybuk> oh hai
[14:14] <mbiebl> Keybuk: I was contemplating about moving libdbus to /lib on Debian as you already did on Ubuntu 
[14:15] <mbiebl> I wondered though, why you also moved dbus-daemon itself (and its utils), when its dependencies (like libexpat) are still on /usr
[14:16] <Keybuk> just a case of a partial move
[14:16] <Keybuk> expact is on the todo list
[14:17] <mbiebl> but is dbus-daemon on /bin really necessary?
[14:17] <mbiebl> Wouldn't libdbus be sufficient
[14:18] <Keybuk> a reasonable question
[14:18] <Keybuk> depends how we want to do filesystem mounting
[14:18] <mbiebl> as in?
[14:19] <mbiebl> what role would dbus-daemon have to play there?
[14:20] <Keybuk> notification of the availability of new block devices
[14:21] <Keybuk> which are then checked, and if their filesystem parents are mounted, mounted themselves
[14:21] <Keybuk> (otherwise queued)
[14:22] <ion_> The automount daemon i’ve been contemplating would handle that and wouldn’t use dbus-daemon.
[14:23] <Keybuk> how would you be notified?
[14:24] <ion_> I was thinking of something like SUBSYSTEM=="block", ACTION=="add|remove", RUN+="/sbin/automountc udev", with the client sending a simple message to automountd. It could always use libdbus but talk directly to the daemon’s UNIX socket.
[14:24] <mbiebl> Keybuk: but that would also require hal or DeviceKit, right?
[14:25] <Keybuk> ion_: that incurs quite a high cost per device
[14:25] <ion_> Hmm, okay.
[14:25] <Keybuk> you'd have to fork/exec the udev child, the match costs, then the cost of fork/execing your automount client
[14:25] <Keybuk> your automount client could just use udev-dbus-daemon to be notified of the new devices
[14:25] <Keybuk> (if we have an early dbus daemon)
[14:25] <ion_> True
[14:26] <Keybuk> (udev-dbus-daemon being the bit that used to be DeviceKit)
[14:26] <Keybuk> ie. d-bus access to the udevdb
[14:26] <mbiebl> oh yeah, davidz likes to move things around quite a bit ;-)
[14:27] <ion_> A random thought, SUBSYSTEM=="block", ACTION=="add|remove", RUN+="socket:@/.../automount/udev_event" :-P
[14:30] <Keybuk> ion_: why is that better than just using dbus?
[14:30] <Keybuk> (and it still costs you the udev child needed to process that)
[14:31] <ion_> Assuming every relevant distro will be running dbus before fscking /, it’s not better.
[14:32] <Keybuk> well... since the relevant distros are those using udev, and probably upstart
[14:32] <Keybuk> those that don't embrace such things are unlikely to embrace a "turn by turn" mount -a replacement <g>
[14:32] <ion_> True
[14:33] <ion_> Allright, i’m convinced, dbus is the way. :-)
[14:53] <damjan> good luck debuging that "mount by dbus" thing when it doesn't magically do what it's supposed to do :)
[15:15] <Keybuk> damjan: ?
[16:06] <ion_> I don’t see how that’s different from debugging anything else.