| keesj | Hi | 11:19 |
|---|---|---|
| Keybuk | hey | 11:19 |
| keesj | I am playing with upstart. I am trying to implement a state machine for my system | 11:22 |
| keesj | I wonder if it makes sence :). the big idea would be to emit events from userland or the kernel and handle upon them in upstart | 11:23 |
| keesj | I have a 3 gui apps that will run (not at the same time) | 11:24 |
| keesj | I would like to be able to emit "high" level events "start_mass-storage-app" and translate that into a few command (insmods perhaps) and starting a gui | 11:27 |
| keesj | can I emit such event from within a job | 11:27 |
| keesj | simply saying.. can I emit events from withing a jobs | 11:28 |
| Keybuk | sure | 11:36 |
| keesj | by calling initctl ? | 11:38 |
| Keybuk | right | 11:40 |
| keesj | there is no "emit" command" | 11:41 |
| Keybuk | there is | 11:46 |
| Keybuk | initctl emit | 11:46 |
| keesj | thanks | 12:12 |
| keesj | I installed upstart in a different directory... I need to prefix the initctl command :( | 12:12 |
| Keybuk | you can make an "emit" command by just symlinking to initctl :p | 12:12 |
| Keybuk | right | 12:12 |
| Keybuk | fixing that is on the TODO | 12:13 |
| keesj | that's nice | 12:13 |
| keesj | and would there be a way to store the current states so I can ask it again later? | 12:21 |
| Keybuk | how do you mean? | 12:22 |
| keesj | just like runlevels work ( that you can ask the current runlevel) | 12:23 |
| Keybuk | still not 100% following | 12:23 |
| keesj | I would like to keep some variable to implement my state | 12:23 |
| keesj | if (in_mass_storage) and usb unplugged -> rmmod ... | 12:24 |
| keesj | in_mass_storage would be a state | 12:24 |
| keesj | in the sample scripts "they" do set $(runlevel --set 1 || true) | 12:26 |
| keesj | to determine if an event gets handled based on the previous runlevel (this is sample code named rc1) | 12:27 |
| Keybuk | right, I'm kinda trying to get to that point | 12:28 |
| Keybuk | where upstart has states that are true or false | 12:28 |
| keesj | you are one of the developers, cool | 12:28 |
| Keybuk | heh, the upstart developer | 12:29 |
| keesj | yo! | 12:29 |
| keesj | I am currently testing by using a qemu (I first tried a chroot ) but the apparently was not the best way to do things | 12:30 |
| Keybuk | I just have it installed by default | 12:30 |
| Keybuk | it replaces sysvinit well enough | 12:31 |
| Keybuk | you can carry on running the old init scripts while playing/developing | 12:31 |
| keesj | I am playing with it for mamona so it is an embedded platform (under arm) | 12:32 |
| keesj | do you happen to know this package? http://smc.sourceforge.net/ (state machine compiler) | 12:33 |
| Keybuk | I don't, no | 12:39 |
| keesj | I used it a lot to create state machines. but basicaly you need "previous_state" + event to implement this | 12:41 |
| Keybuk | right | 12:42 |
| Keybuk | depends what you want to do, really | 12:42 |
| keesj | the "quick" hack would be to misues the initctl to tell init to change it's own "environement" | 12:43 |
| keesj | thanks for your time btw (I really apreaciate it) | 12:44 |
| Keybuk | right, but I'm not sure I fully understand what you want to achieve | 12:44 |
| keesj | hmm . perhaps I am just thinking the wrong way around. let me try to give a simple example | 12:47 |
| keesj | my system can have twoo states "video playback" and "mass storage". | 12:51 |
| keesj | if somebody presses the power button while in mass storage I don't like that so I would like to ignore a "power" event when in storage mode. | 12:52 |
| keesj | I was thinking in the line of creating a "button" event handler that listend to power_on power_off event | 12:53 |
| keesj | other events would be something like usb_connected (being a good reason to switch to mass storage) | 12:54 |
| keesj | I think I am taling about level events | 13:00 |
| Keybuk | what other kinds of decision would you make based on these states though/ | 13:00 |
| keesj | whell I have a few of those event flying in (usb state changes, button presses, system update event) I was also thinking of using this state when going in "suspend" mode | 13:04 |
| keesj | so that when I wakeup I know what program to start again | 13:04 |
| Keybuk | when suspending why do you need to stop programs? | 13:05 |
| keesj | good questions :p | 13:06 |
| keesj | currenty I had to do it because I don't get events when in suspend , so if the device is in video mode , I suspend , and plug it to a computer , I miss the "connected" event | 13:09 |
| keesj | (just like udev meesd to cache the events generated by the kernel , I guess this is not the scop of upstart :p | 13:09 |
| keesj | I am getting there ! | 13:13 |
| Keybuk | not entirely sure yet what the scope of upstart is ;) | 13:14 |
| keesj | haha | 13:16 |
| keesj | the code looks very clean | 13:17 |
| keesj | are you at fosdem? | 14:12 |
| Keybuk | no | 14:20 |
| Keybuk | I've never been to one of those | 14:23 |
| Keybuk | http://upstart.ubuntu.com/wiki/RemoveEventArguments | 14:37 |
| keesj | I can be nice :op | 14:37 |
| Keybuk | Any objection? no? good! :o) | 14:37 |
| keesj | haha | 14:39 |
| keesj | it | 14:39 |
| keesj | I really just should al be programmed in lua | 14:41 |
| Keybuk | why? | 14:42 |
| keesj | I see plenty discussion about creating, "combined events" or other other structs that would be possible if upstart used a scripting language for the events definitions | 14:44 |
| keesj | but the current aproach is also pretty clean i guess | 14:44 |
| keesj | >>from (tty-added or cua-added) and foo<< does that get interpreted every time an event would come? | 14:46 |
| Keybuk | the string? no | 14:52 |
| Keybuk | it's compiled into a match tree in memory | 14:52 |
| keesj | hmm that you can not replace by lua :) | 14:53 |
| Keybuk | what would lua gain though? | 14:53 |
| keesj | in this case nothing. but just like the current script command you could script the behaviour (if an event is consummed by an handler) | 14:56 |
| keesj | my system looks a little like the stoplight (the left image) http://smc.sourceforge.net/Images/SmcEx4_Stoplight.png I get the same type of events (like a timeout ). I see very whell how to perform the "start,turn light etc" . | 15:04 |
| keesj | you get the same "message" but want to perform something different(if you want me to stop just say so , I am really trying to find a awy to do it) | 15:05 |
| keesj | I have put the different events in an event_handler that performs the few functions | 15:06 |
| keesj | so I can not emit "switch_from_a_to_b" | 15:06 |
| Keybuk | hmm | 15:08 |
| Keybuk | currently Upstart's event matching is more functional than procedural | 15:08 |
| Keybuk | ie. you describe the state in which the job is running | 15:09 |
| Keybuk | inside the job itself, you have a shell script, so it's procedural | 15:09 |
| keesj | Yes , I Have the feeling I really am missing the point :p | 15:09 |
| Keybuk | I'm not sure I've got the point ;) | 15:09 |
| keesj | AM I trying to misue the system? | 15:09 |
| keesj | missue | 15:09 |
| Keybuk | I'm not quite sure what you're trying to do yet | 15:09 |
| keesj | in a perfect work I would like to say "emit go_into_mass_storage" where ever you come from | 15:11 |
| Keybuk | you wouldn't use an event like that | 15:12 |
| Keybuk | I'd define mass_storage as a state/job | 15:12 |
| Keybuk | so "start mass_storage" | 15:12 |
| keesj | yes I have this part http://paste-it.net/6595 | 15:14 |
| keesj | so do I need to create a userland progress saying "emit stop_mass_storage" "emit start_video" | 15:15 |
| keesj | with just the start/stop the system looks like /etc/init.d/mass_storage [start|stop] | 15:17 |
| keesj | or would start mass_storage stop video? | 15:18 |
| keesj | the backwards compatible scripts use wmtp to "know" what the current runlevel is | 15:19 |
| keesj | I am getting there ... /initctl list | 15:29 |
| keesj | shows a list of active tasks, I can only have one state active at the time..... | 15:29 |
| Keybuk | so you'd make your states start and stop each other? | 15:30 |
| Keybuk | or be started and stopped by the same events | 15:30 |
| keesj | I will listen for event use the initctl to know the current state and react as supposed | 15:31 |
| keesj | I currently implemented it a separate jobs that are triggered by events (the switching between states) | 15:32 |
| keesj | So I guess the second option | 15:33 |
| keesj | perhaps I might even use the list to do sanity checks | 15:34 |
| keesj | I will try to hack a version and try to post it to the list | 15:53 |
| Keybuk | shiny | 16:01 |
| keesj | I hope it wil work. (that there are no race cases...) | 16:05 |
| keesj | it requires a state to keep running apparently a jobs that is started and exits is in a "stop" state again | 16:06 |
| Keybuk | (gdb) p *((EventOperator *)job->start_on->node->left) | 16:14 |
| Keybuk | $5 = {node = {parent = 0x662720, left = 0x0, right = 0x0}, type = EVENT_MATCH, | 16:14 |
| Keybuk | value = 0, name = 0x662670 "wibble=foo bar", env = 0x0, event = 0x0, | 16:14 |
| Keybuk | blocked = 0} | 16:14 |
| Keybuk | (gdb) p *((EventOperator *)job->start_on->node->right) | 16:14 |
| Keybuk | $6 = {node = {parent = 0x662720, left = 0x0, right = 0x0}, type = EVENT_MATCH, | 16:14 |
| Keybuk | value = 0, name = 0x6628a0 "wobble=foo bar", env = 0x0, event = 0x0, | 16:14 |
| Keybuk | blocked = 0} | 16:14 |
| Keybuk | (tangent: | 16:14 |
| Keybuk | damn I was on good form when I wrote this, I predicted I'd need something a year down the line) | 16:14 |
| Keybuk | that was from | 16:14 |
| Keybuk | "wibble=foo bar" or wobble="foo bar" | 16:15 |
| Keybuk | so it handles quoting sensibly | 16:15 |
| keesj | can I get the event name inside my scripts? | 16:21 |
| keesj | UPSTART_EVENT... | 16:22 |
| Keybuk | right | 16:22 |
| Keybuk | $UPSTART_EVENTS in trunk | 16:22 |
| keesj | week-end here. thanks a lot for the help! | 16:33 |
| Keybuk | enjoy | 16:33 |
| keesj | thanks | 16:33 |
| === ion___ is now known as ion_ | ||
Generated by irclog2html.py 2.7 by Marius Gedminas - find it at mg.pov.lt!
