| [01:20] <Keybuk> sadmac_: yes | |
| [01:21] <Keybuk> though in 0.3.x it will probably contain the event that stopped the job | |
| [01:21] <Keybuk> whereas in trunk, UPSTART_EVENTS will contain the events that originally started it | |
| [12:08] <keesj> Hi | |
| [12:16] <Keybuk> hi | |
| [12:24] <keesj> The chance I wil use upstart are growing every day :p | |
| [12:25] <keesj> I am now looking into the kernel->upstart communication | |
| [12:28] <keesj> special I am looking at the udev from the replacement-initscripts http://paste-it.net/7162/raw/ | |
| [12:29] <keesj> that script has a "emits block-device-added block-device-removed" section | |
| [12:32] <keesj> what it that good for? | |
| [12:32] <Keybuk> "emits" is just a documentation stanza | |
| [12:33] <Keybuk> you'd use it in the job definitions of services that are known to emit their own events | |
| [12:33] <Keybuk> and that way other frontend tools can draw pretty dependency graphs using things like dotty | |
| [12:33] <keesj> sexy :p | |
| [12:33] <Keybuk> it's not actually used by Upstart in any way, other than being available as a property | |
| [12:34] <Keybuk> the theory there is that that distribution would ship udev rules that looked something like: | |
| [12:34] <Keybuk> SUBSYSTEM=="block", ACTION=="add", RUN+="/sbin/initctl emit block-device-added DEVPATH" | |
| [12:34] <Keybuk> SUBSYSTEM=="block", ACTION=="remove", RUN+="/sbin/initctl emit block-device-removed DEVPATH" | |
| [12:35] <Keybuk> SUBSYSTEM=="net", ACTION=="add", RUN+="/sbin/initctl emit network-device-added INTERFACE" | |
| [12:35] <Keybuk> SUBSYSTEM=="net", ACTION=="remove", RUN+="/sbin/initctl emit network-device-removed INTERFACE" | |
| [12:35] <Keybuk> etc. | |
| [12:35] <Keybuk> alternately we may want an upstart-addon-udev or something so we could have a generic RUN+="socket:/com/ubuntu/upstart/udev" at the end, and let the addon daemon turn those into events | |
| [12:35] <Keybuk> OR we might just patch udev directly to emit events into upstart | |
| [12:37] <keesj> my system does not use udev, i currently use <linux/netlink.h> to recieve NETLINK_KOBJECT_UEVENT events | |
| [12:38] <keesj> I could just create a mini app that gets started and emits the kernel uevent | |
| [12:40] <keesj> hmm I will look into it | |
| [12:43] <Keybuk> right | |
| [12:43] <Keybuk> you could have a little daemon to listen on the netlink socket and inject into upstart | |
| [12:44] <keesj> I do have the same kind of problems that are solved by udev , being that if a device get plugged before the daemon is started I will never know about | |
| [12:45] <keesj> I then have to write custom code to look at the status . | |
| [12:56] <Keybuk> why not just write to the uevent files like udev does? | |
| [12:56] <Keybuk> for each object under /sys that a uevent was originally emitted for, there will be a uevent file | |
| [12:56] <Keybuk> write "add" (no \0, no \n, etc.) to that, and the kernel will resend it | |
| [13:03] <keesj> Thanks for the tip , I will have to look into how that works! | |
| [19:12] <Keybuk> ion_: interesting thought of the day | |
| [19:12] <Keybuk> all of the dbus_bus_ calls block | |
| [19:17] <ion_> keybuk: Is that a problem? | |
| [19:18] <Keybuk> I'm not sure | |
| [19:20] <Keybuk> the fact that dbus_bus_get() can never return worries me enough :p | |
| [19:20] <ion_> Heh | |
| [19:23] <Keybuk> I guess it depends at what level of blocking we have to worry about in Upstart | |
| [19:24] <Keybuk> in some senses, blocking is actually not necessarily bad | |
| [19:24] <Keybuk> since it reduces system load | |
| [19:24] <Keybuk> e.g. upstart blocks while it sets up a process, even though that's happening the child | |
| [19:24] <Keybuk> and doesn't move onto the next until exec() completes | |
| [19:24] <Keybuk> so arguably, after starting dbus, blocking to connect to it and say Hello isn't too bad | |
| [19:40] <ion_> Yep... | |
| [19:56] <ion_> Don’t they have a separate DBus (client) implementation in C#? Perhaps someone should NIH a DBus implementation that doesn’t suck in C. ;-) | |
| [19:56] <Keybuk> ndesk-dbus | |
| [20:01] <Keybuk> the C API isn't really that bad | |
| [20:01] <Keybuk> the main problem is just figuring out how to dispatch methods | |
| [20:14] <keesj> ion_: that would be great | |
| [21:17] <Amaranth> @btlogin | |
| [21:17] <Amaranth> err | |