[16:04] <Keybuk> sadmac: Bill is using emit in a very strange way in #435368
[16:47] <sadmac_> Keybuk: what's the issue?
[16:50] <Keybuk> is it just a minimal test case that demonstrates the bug?
[16:50] <Keybuk> ie. why are you using emit?
[16:50] <sadmac_> exactly, just a test case
[16:50] <sadmac_> the point is a job started by an event with arguments doesn't get those arguments again when restarted
[16:53] <Keybuk> true...
[16:53] <Keybuk> but since events don't even *have* arguments anymore in trunk
[16:54] <Keybuk> and even while they still did, they weren't passed to the job anyway ...
[16:54] <Keybuk> it's kinda a null test case :p
[16:54] <Keybuk> it's certainly a bug in 0.3 that the associated event information is lost for services once they're running, so not available after the respawn - that is fixed in trunk
[16:57] <sadmac_> Ah cool
[16:57] <Keybuk> the trunk equivalent would be something like:
[16:58] <Keybuk>   initctl emit foo DEV=bar PATH=baz
[16:58] <Keybuk> and then you'd use $DEV and $PATH in the script
[16:58] <Keybuk> and they survive respawn
[16:58] <Keybuk> until a different event starts the job
[16:58] <Keybuk> or you change them with start
[16:58] <Keybuk> ie.
[16:58] <Keybuk>   initctl start jobname DEV=wibble PATH=wobble
[16:59] <sadmac_> cool
[17:00] <sadmac_> Keybuk: have you had any thoughts on lod and making it work?
[17:00] <Keybuk> lod?
[17:00] <sadmac_> s/lod/logd/
[17:01] <Keybuk> not really
[17:01] <Keybuk> the only thing I've come up with would be making logd have a pty
[17:01] <Keybuk> and telling upstart to divert output from jobs to that pty while running
[17:01] <Keybuk> that doesn't fix logd dying though
[17:02] <Keybuk> so has the same basic problems as the socket
[17:02] <sadmac_> hm
[17:03] <sadmac_> Keybuk: you could have upstart itself hold a pipe to each program, and then just bridge it through sunrpc to logd (since sunrpc has that lovely property of just sitting and buffering when the listener goes away)
[17:03] <Keybuk> interesting
[17:04] <Keybuk> would you need the bridge in that instance?
[17:04] <sadmac_> the one big limitation of sunrpc is it has a custom interface (no file descriptors) so you can't just bolt it right onto stdio
[17:04] <sadmac_> so yes
[17:04] <Keybuk> ah
[17:04] <sadmac_> oh, not sunrpc
[17:05] <sadmac_> sunipc
[17:05] <sadmac_> or sun ipc
[17:05] <Keybuk> lol
[17:05] <Keybuk> the problem with the bootlogd approach, btw (stealing console output via ioctl) is it doesn't work with our usplash
[17:06] <Keybuk> and I suspect it wouldn't work with your X rhgb either
[17:06] <Keybuk> since both *change* the console to be a graphical one, with text diverted elsewhere
[17:07] <sadmac_> you could alter usplash of course. (I may rewrite rhgb once we get off the ground. right now I'm patching some stuff into upstart to deal with it)
[17:08] <Keybuk> it isn't usplash that's the problem
[17:08] <Keybuk> it's svgalib
[17:08] <Keybuk> and I want to fiddle with that about as much as I want to wake up without my kidneys
[17:09] <sadmac_> ah
[17:20] <sadmac_> Keybuk: it looked like in trunk the event files had moved, is this the case?
[17:21] <Keybuk> event files?
[17:21] <sadmac_> /etc/event.d/*
[17:23] <Keybuk> possibly moving to /etc/init/jobs.d
[17:23] <sadmac_> yeah, any particular reason?
[17:24] <Keybuk> to stop people calling them "event files" :)
[17:24] <Keybuk> since they're not
[17:24] <Keybuk> they're job files
[17:24] <Keybuk> jobs and events are different things
[17:24] <Keybuk> but everyone confuses the two because of the directory name
[17:24] <sadmac_> ah.
[17:26] <Keybuk> e.g. people think that if you have /etc/event.d/foo
[17:27] <Keybuk> then initctl emit foo will start it
[17:27] <sadmac_> yeah.
[17:28] <Keybuk> which isn't true at all
[17:32] <sadmac_> Keybuk: one more correction, it wasn't sun ipc. it was sysv.
[17:32]  * sadmac_ should drink coffee while reading things
[17:34] <sadmac_> specifically message queues