[16:39:33]  <mezcalero> halfline: oh, have a minute for talk about the
   bootlog stuff?
   [16:39:43]  <halfline> well i have a few minutes
   [16:39:47]  <halfline> there's an office party soon though
   [16:40:12]  <mezcalero> notting: well, i'd like to reite runlevels as much
   as possible
   [16:40:19]  <mezcalero> retire
   [16:40:33]  <mezcalero> halfline: so
   [16:40:36]  <mezcalero> what i discussed with kay
   [16:40:49]  <mezcalero> was mostly that boot logging is awful to do, but
   kinda hard to do without
   [16:40:53]  <notting^ ok, not runlevel.. but if we can do it with an env
   variable vs. having three service files...
   [16:41:06]  <davidz> halfline: interesting - I'll try that in a bit
   [16:41:29]  <halfline> mezcalero: yea have to have logging
   [16:41:31]  <mezcalero> notting: well, we could do this via template
   instantiation
   [16:41:41]  <halfline> people don't mind graphical boot as long as they
   can read boot messages somewhere
   [16:41:59]  <mezcalero> halfline: so, for the sysinit stuff we have two
   options:
   [16:42:06]  <halfline> but if they can't read boot messages somewhere then
   they get angry about boot "hiding" them
   [16:42:37]  <mezcalero> 1) connect sysinit to kmsg and consider
   interactive bootup services a thing of the past and everything has to use
   plymouth
   [16:43:11]  <mezcalero> 2) admit that people might always need interactive
   stuff because people are crazy and connect sysinit to the console, but do
   the boot logging stuff
   [16:43:18]  <mezcalero> #1 would be trivial to do
   [16:43:30]  <mezcalero> #2 would probably means that we need to keep a
   boot logger of some kind
   [16:43:33]  <mezcalero> question is where this lives
   [16:43:38]  ***  nils (~leila@dslb-092-075-167-074.pools.arcor-ip.net) has
   joined chat #fedora-desktop.
   [16:43:40]  ***  jbiebel has left chat #fedora-desktop.
   [16:43:46]  <mezcalero> since you already have the code in plymouth
   [16:43:50]  <mezcalero> i'd say to keep it there
   [16:43:56]  <halfline> that's fine
   [16:44:00]  <nils> hej! does anyone know, how to install the gnome
   language selector?
   [16:44:12]  <mezcalero> which also has the advantage that it could work
   even before we handover from the initrd
   [16:44:21]  <halfline> we can dump it to syslog instead of manually
   writing boot.log though
   [16:44:25]  <mezcalero> halfline: however, here's one idea to make it a
   little less ugly:
   [16:44:35]  <mezcalero> connect things directly to kmsg
   [16:44:45]  <mezcalero> don't buffer it and flush it to boot.log
   [16:44:57]  <mezcalero> just write everything to kmsg, and syslog will
   flush it for you
   [16:45:04]  <halfline> well i still have to buffer it
   [16:45:10]  <halfline> since i have to show it when the user hits escape
   [16:45:19]  <mezcalero> halfline: you can read it from kmsg
   [16:45:27]  <mezcalero> just consider kmsg your buffer
   [16:45:30]  <mezcalero> there only needs to be one
   [16:45:46]  <halfline> oh true
   [16:45:58]  <notting> halfline: whee, a boot logger that dumps to syslogd.
   i remember that!
   [16:46:18]  <mezcalero> so the idea is basically that everything goes to
   kmsg until syslog is started. from the on everything goes to syslog,
   except what the kernel itself logs
   [16:46:23]  <halfline> well not syslogd
   [16:46:40]  <halfline> yea so that's an interesting complication
   [16:46:55]  <halfline> syslogd gets started somewhere in the middle of
   boot
   [16:46:59]  <mezcalero> halfline: i think by simply using kmsg a lot of
   the ugliness of boot logging goes away, and things even become more robust
   and discoverable
   [16:47:23]  <mezcalero> well, syslog flushes kmsg to syslog as soon as it
   starts up
   [16:47:42]  <mezcalero> and from then on will write verything coming from
   kmsg to it as well
   [16:47:58]  <mezcalero> so if you log to kmsg it will end  up in syslog in
   any case
   [16:48:04]  <halfline> oh okay
   [16:48:30]  <halfline> it's weird
   [16:48:39]  <mezcalero> make sure to prefix your lines when you write them
   to kmsg though. otherwise people will be pissed ;-)
   [16:49:06]  <mezcalero> and the kernel log buffer we use these days is
   gigantic anyway, so this should not be a problem
   [16:49:15]  <notting> although if you're writing to kmsg when you exit, it
   means there will be some interleaving/time skew in the logs
   [16:49:51]  <halfline> i guess <6> ?
   [16:50:38]  <halfline> notting: well i'd write to kmsg the entire time i
   guess
   [16:50:50]  <halfline> not just dump a big buffer at the end or anything
   like that
   [16:51:19]  <mezcalero> halfline: i was thinking about "<6>plymouth: " as
   a prefix
   [16:51:23]  <halfline> mezcalero: so one thing i'm worried about
   [16:51:34]  <mezcalero> i.e. when systemd logs to kmsg it uses
   "<n>systemd: "
   [16:51:38]  <mezcalero> and udev uses "udev: "
   [16:51:45]  <mezcalero> or you might even use "console: "
   [16:51:46]  <mezcalero> or so
   [16:51:46]  <halfline> if i read from kmsg
   [16:52:04]  <mezcalero> notting: time skew? explain?
   [16:52:04]  <halfline> then will i be swallowing syslogd's data?
   [16:52:27]  <halfline> like syslog(2) has a command "return number of
   unread characters"
   [16:52:28]  <mezcalero> halfline: there are two interfaces to kmsg
   [16:52:33]  <halfline> i wonder if syslogd uses that
   [16:52:45]  <mezcalero> one is /proc/kmsg
   [16:52:50]  <mezcalero> which is what syslog uses
   [16:52:57]  <notting^ if he's dumping when plymouth ends, then the
   sequence is 1) boot 2) syslog starts 3) syslog dumps kmsg so far 4) syslog
   logs other random things logged to syslog during boot 5) plymouth ends 6)
   plymouth stuff goes to kmsg and ends up in log files
   [16:52:57]  ***  nils has left chat #fedora-desktop.
   [16:52:59]  <mezcalero> the other is syslog(2)
   [16:53:30]  <mezcalero> halfline: when you want to show the logs in the UI
   use syslog(2) and you will not interfere with /proc/kmsg
   [16:53:53]  <mezcalero> notting: no, the idea is that plymouth writes to
   kmsg right-aaway
   [16:53:55]  <mezcalero> no queuing
   [16:54:17]  <halfline> are you sure?
   [16:54:22]  <mezcalero> just line-by-line copying from the pty to kmsg as
   it happens
   [16:54:24]  <halfline> (they don't interfere)
   [16:54:47]  <mezcalero> halfline: pretty sure
   [16:54:52]  <halfline> well i'll try it
   [16:54:59]  <mezcalero> not 200% sure though ;-)
   [16:55:09]  <halfline> and if i hit a problem i'll just buffer myself like
   i am now
   [16:55:35]  <halfline> i definitely like the idea of getting my messages
   from kmsg though
   [16:55:40]  <halfline> because i'll get a more full set of messages
   [16:55:45]  <halfline> than just the console messages
   [16:56:10]  <mezcalero> halfline: see the comments on syslog(2) regarding
   this btw. operation 2 is what syslog does. operation 3 is what you'd do
   for the UI