keesj | anybody having a 5.0+ system running | 16:19 |
---|---|---|
keesj | 0.5 | 16:31 |
sadmac | keesj: Keybuk might've had one | 16:32 |
sadmac | he had a boot chart for a mostly pure upstart system recently. forget which version it was | 16:32 |
keesj | I have something that kils mine http://www.paste-it.net/raw/public/l01613a/ | 16:32 |
keesj | http://www.paste-it.net/public/kc01d79/ (result) | 16:33 |
Keybuk | traceback please | 16:33 |
keesj | somehow becasue of the respawn stanza things go wrong | 16:34 |
keesj | Keybuk: I will work on that :p | 16:35 |
Keybuk | it should dump core from a child | 16:35 |
Keybuk | it'll be in / | 16:35 |
Keybuk | you may need to ulimit -c unlimited in the initramfs, of course | 16:35 |
keesj | I have a RO system | 16:35 |
Keybuk | actually | 16:35 |
Keybuk | ignore the last bit, I patched upstart to always set ulimit to INFINITY :) | 16:35 |
sadmac | Keybuk: what are your thoughts on dealing with a system where multiple display managers are installed, and we want to run one of them? | 16:36 |
sadmac | (but never both) | 16:36 |
Keybuk | don't install two | 16:37 |
Keybuk | or use a system like alternatives | 16:37 |
sadmac | there's a pretty large chunk of Fedora users with full GNOME and KDE on the same system, so I'd like to get this right. | 16:38 |
Keybuk | how do you do it now? | 16:38 |
sadmac | we run a script called prefdm that checks for each of them and runs the first it finds. | 16:39 |
sadmac | and its ugly and bad and I hate it. | 16:39 |
Keybuk | you have alternatives in Fedora? | 16:39 |
sadmac | yeah. I don't know why we don't use it. | 16:40 |
sadmac | though I never understood how package managers worked with it. | 16:40 |
Keybuk | I don't think this is an upstart problem | 16:40 |
sadmac | Keybuk: what if each display manager wants different service management settings? (I wouldn't put it past the various desktop groups to collectively fail in this manner) | 16:42 |
keesj | http://www.paste-it.net/public/ge70991/ (without debugging symbols) but it's a raise | 16:42 |
Keybuk | sadmac: alternatives handles all this already | 16:42 |
Keybuk | keesj: 6 | 16:42 |
Keybuk | #1 0x4008c160 in abort () from /mnt/lib/libc.so.6 | 16:42 |
Keybuk | Backtrace stopped: frame did not save the PC | 16:42 |
Keybuk | meh | 16:42 |
sadmac | Keybuk: from what I understand alternatives just changes the */bin symlink. it doesn't swap out configuration. especially not for other programs. | 16:43 |
Keybuk | it works today in Ubuntu | 16:44 |
Keybuk | if you install kdm and gdm, both have init scripts | 16:44 |
Keybuk | but only one of them runs | 16:44 |
sadmac | I'll look at that then. | 16:45 |
keesj | that part is not handled using the alternatives | 16:45 |
Keybuk | I thought it was handled by the file they write | 16:45 |
keesj | but generaly alternatives work pretty well , I can imagine the /etc/init.d/job.d/service_gdm is a symlink | 16:46 |
sadmac | my first thought was "well, Upstart has a way of reflecting the 'requires' relationship already. 'provides' and 'conflicts' seem like they'd follow." | 16:46 |
Keybuk | I think that overcomplicates things | 16:47 |
sadmac | depends. I'd want to see a complete 0.10 first | 16:48 |
sadmac | if its a fairly small change on top of that I'd say its worth having. | 16:48 |
Keybuk | I don't buy that argument | 16:48 |
Keybuk | we should only do things where it makes sense :p | 16:49 |
Keybuk | next you'll be asking for some kind of config handling in Upstart because your package manager sucks at managing /etc :p | 16:49 |
sadmac | alternatives pushes the concern into the packages. we have to modify the distributions (renaming binaries etc) and we have to do it in a very distro-specific way | 16:50 |
sadmac | a simple conflict solution in upstart means that this can come upstream in all the display managers. | 16:51 |
keesj | Keybuk: http://www.paste-it.net/raw/public/i0ad7ab/ better? | 16:51 |
Keybuk | hmm | 16:52 |
Keybuk | that looks like a different problem | 16:52 |
Keybuk | in fact, you might want to run "cont" there | 16:52 |
Keybuk | I assume that wasn't the core file? | 16:52 |
keesj | no that's the init | 16:52 |
Keybuk | how do you mean? | 16:53 |
keesj | I run gdbserver and attach to pid 1 | 16:53 |
Keybuk | ah | 16:53 |
Keybuk | can you run "cont" | 16:53 |
Keybuk | SIGPIPE is peffectly normal | 16:53 |
keesj | I will do cont | 16:54 |
Keybuk | sadmac: again, I don't buy that | 16:56 |
Keybuk | rah state changed from stopping to killed | 17:01 |
Keybuk | rah state changed from killed to post-stop | 17:01 |
Keybuk | rah post-stop process (32487) | 17:01 |
Keybuk | rah pre-stop process (32484) exited normally | 17:01 |
Keybuk | init:job_process.c:1103: Assertion failed in job_process_terminated: job->state == JOB_PRE_STOP | 17:01 |
Keybuk | zsh: abort (core dumped) ./init | 17:01 |
Keybuk | eep | 17:01 |
Keybuk | 1103nih_assert (job->state == JOB_PRE_STOP); | 17:01 |
Keybuk | (gdb) p job->state | 17:01 |
Keybuk | $1 = JOB_POST_STOP | 17:01 |
keesj | I need to go | 17:02 |
Keybuk | it left the running state when the main process ended | 17:04 |
Keybuk | while the pre-stop process was still running | 17:04 |
Keybuk | so when the pre-stop process died, it asserted | 17:04 |
Keybuk | keesj: please file a bug when you get back | 17:07 |
Keybuk | attach the job, and your output, etc. | 17:07 |
Keybuk | I think that the code in job_process_terminated that checks for post-start or pre-stop still running needs to be higher up before the "break" statements ;) | 17:07 |
Keybuk | but I'll need to make test cases and suchlike | 17:07 |
keesj | I will | 17:08 |
ion_ | There’s a character in ST:TNG that looks just like MacGyver. http://heh.fi/tmp/tng-macgyver | 17:15 |
Keybuk | There's a character in Stargate SG-1 that looks just like MacGyver too | 17:15 |
ion_ | :-D | 17:16 |
=== sadmac_ is now known as sadmac |
Generated by irclog2html.py 2.7 by Marius Gedminas - find it at mg.pov.lt!