IRC, freenode, #hurd, 2013-07-22
<teythoon> ok, so back to the drawing board for the next big issue, the
potential proc and init merge
<teythoon> Roland had some harsh words for that proposal, but noone else
raised concerns
<youpi> noone else does not mean much
<youpi> I guess only Roland actually understands the matter
<youpi> so I'd tend to believe him
<teythoon> even though, his criticism was so superficial, he could at least
be a bit more specific...
<braunr> i agree that the argument, being simply based on vague principle,
isn't very convincing
<teythoon> so, what should I do?
<braunr> you can either keep them separate, or fight with roland
<teythoon> common braunr, I need a little more guidance in these kind of
social issues
<teythoon> a statement like this is of little use ;)
<braunr> that's the best i can give you
<teythoon> :/
<braunr> i have one patch "fixing" HZ on the hurd, and i even get to fight
about it
<teythoon> I understand Roland has been around forever and keeps an eye on
stuff
<teythoon> but could/would he block a patch for hurd if e.g. youpi would
accept it
<teythoon> i.e. how much control has he in practice?
<teythoon> me fighting with him over a patch is of little value for anyone
and I don't care to do so
<braunr> not much i suppose now
<braunr> but we also have to agree with the change
<braunr> with *real* arguments
<braunr> (well, if it was up to me, i'd even merge exec with proc so ..)
<teythoon> ok, so I whip up a patch to see how it goes in practice and
present it so we could talk about the issue with something to look at
first
<braunr> although maybe not ;p
<braunr> you'll hit the same reaction
<teythoon> from Roland?
<braunr> yes
<braunr> and youpi said he tends to trust what roland says
<braunr> so let's discuss the pros and cons a bit more
<teythoon> yes, but I'd honor his concerns if they were properly
presented. just telling me to hack on linux instead even though I think I
have demonstrated that I do want to work on Hurd is so childish in my
eyes that I do not consider that a valid argument at the moment
<teythoon> sure, shoot
<braunr> well, functionally, they're unrelated
<teythoon> head -n1 init/init.c
<teythoon> /* Start and maintain hurd core servers and system run state
<youpi> and thus it makes sense to make them separate, even if it does not
seem to bring anything useful now
<youpi> history has shown that it makes a bed for nice things later
<braunr> teythoon: that's not what proc is about
<teythoon> braunr: I know
<teythoon> braunr: that's what init is about in its own words ;)
<youpi> teythoon: also, "simplifying the code" is not necessarily an
argument that would be considered
<youpi> depending on the simplification
<youpi> linux made it all simple by using a monolithic kernel :)
<youpi> separating concerns is complex
<youpi> but in the end it usually pays off on the Hurd
<youpi> personally, I'd be fine with Guillem's solution, and renumbering
init's pid in Debian
<youpi> there's a pending question from Roland actually: what information
is exchanged between init and proc in the end?
<youpi> that's actually the point of the discussion: is that information
really big or not
<teythoon> I'm sorry, you lost me, where did he ask that question?
<pinotree> $ git grep proc_getmsgport | egrep '[0-9]' ← /hurd/init as pid 1
is hardcoded in few places
<youpi> teythoon: he didn't ask it this way, but that's the question I had
to be able to answer his
<youpi> Date: Mon, 15 Jul 2013 10:36:35 -0700 (PDT)
<youpi> > That's not what he said. He said there is a lot of information
<youpi> > propagated from init to proc, and thus the separation is
questionable.
<youpi> Are you talking about bootstrap, or what?
<youpi> as I haven't investigated much, I couldn't answer this
<youpi> pinotree: right. We could patch these in Debian
<teythoon> youpi: so, shall I refresh, test and refine Guillems patch and
resend it?
<youpi> it's probably an easier way
<teythoon> ok, I start by doing that
IRC, freenode, #hurd, 2013-07-25
<teythoon> pinotree: btw, there are two /sbin/init processes even with my
hacked up init/proc variant where /sbin/init gets to be pid 1
<pinotree> never seen that
<pinotree> what are their parents?
<teythoon> pinotree: well, pid 1 is /sbin/init now, pid 13 or something has
the parent 1
<teythoon> looks like init forks or something
<pinotree> i guess your sysvinit is compiled without INITDEBUG?
<pinotree> nothing in syslog either?
<teythoon> pinotree: it's compiled like the sysvinit shipped with debian
<pinotree> teythoon: do you have custom additions in inittab?
<teythoon> pinotree: a terminal for my serial console
<teythoon> *getty
<pinotree> are the getty started correctly for you, btw?
<teythoon> pinotree: yes
<pinotree> interesting
<pinotree> teythoon: back then, they were costantly respawning, with hurd's
getty's failing to start when exec'ed by (sysv)init
<pinotree> wonder what changed
<teythoon> pinotree: cool, magically went away then :)
IRC, freenode, #hurd, 2013-07-29
<teythoon> youpi: I need some feedback on the not freezing translators
issue, more specifically whether I understood you correctly in your mail
from wednesday (20130724131552.GG9576@type.bordeaux.inria.fr)
<teythoon> oh yeah, and I had some questions yesterday too, about rpctrace
and dead-name notifications, specifically why /hurd/init is not receiving
any for the root translator and the exec server
<braunr> teythoon: more details please
<teythoon> ok, so /hurd/init is registering for dead name notifications for
essential tasks
<teythoon> the rootfs and exec both register as essential tasks at init and
init requests successfully dead name notifications for them
<teythoon> if you e.g. kill the auth server, /hurd/init will notice and
crash the system
<teythoon> if you kill exec or the rootfs, /hurd/init does not get notified
<teythoon> I verified this with gdb and an subhurd
<teythoon> I'm puzzled by this, as the kernel is the one who sends the
notifications, right?
<braunr> yes
<braunr> teythoon: where is the problem ?
<teythoon> and it is not that the system is not sending any messages, it
is, I see the msgcount increase over time
<teythoon> braunr: dunno, as far as I can tell the kernel does not deliver
the notification for rootfs and exec
<braunr> oh
<teythoon> those are the two processes loaded by grub, maybe they are
different somehow
<braunr> is that affecting your work ?
<teythoon> no, not directly, I strayed around at the weekend, trying to
think of cool stuff hurd could do
<teythoon> youpi: I need some feedback on the not freezing translators
issue, more specifically whether I understood you correctly in your mail
from wednesday (20130724131552.GG9576@type.bordeaux.inria.fr)
<youpi> teythoon: ok, now I'm available for the not-freezing-translators
thing :)
IRC, freenode, #hurd, 2013-08-05
<teythoon> youpi: I'm in the process of producing a unified
sysvinit-as-pid1 and please-dont-kill-important-processes patch series
<teythoon> youpi: there is one issue with changing /hurd/inits pid, libcs
reboot() also assumes that it has the pid 1
<youpi> argl
<youpi> that's bad, because it's then an ABI, not just an internal thing
<teythoon> hardcoding the pid is the worst way of getting a handle of any
server :/
<teythoon> I've been thinking to make it explicit by binding it to
/servers/startup or something
<youpi> that would be more hurdish than using a pid, yes
<teythoon> yes, and not only does it break the abi, but in a bad way
too. if the libc is updated before the hurd, the shutdown sequence is
broken in a way that the translators aren't synced :/
<teythoon> youpi: as a workaround, we could make reboot() signal both pid 1
and 2
<youpi> at worse pid 1 shouldn't get harmed by receiving a startup_reboot
RPC indeed
<teythoon> yes
IRC, freenode, #hurd, 2013-08-16
<teythoon> grml, the procfs hardcodes the kernels pid :/
<teythoon> there's always one more thing to fix...
<teythoon> uh, and we made pids.h a private header, so no nice constant for
the procfs translator :/
<teythoon> server lookup by hardcoding the pid should be banned...
IRC, freenode, #hurd, 2013-09-16
<teythoon> youpi: I'm thinking about splitting /hurd/init into /hurd/init
and /hurd/startup
<teythoon> that way, you could also merge the init as pid1 patches
<teythoon> that should be doable within the week
<youpi> that would probably be better received by Roland than merging init
into proc :)
<teythoon> yes, I suppose so :D
<youpi> perhaps you should start the discussion on the list about it
already, with just a sketch of which would do what
<teythoon> ok
<teythoon> fwiw I like the name startup b/c it speaks the startup protocol
<braunr> teythoon: +1 startup
IRC, freenode, #hurd, 2013-09-23
<teythoon> I've been hacking on init/startup, I've looked into cleaning it
up
IRC, freenode, #hurd, 2013-10-07
<teythoon> braunr: btw, what do you think of my /hurd/startup proposal?
<braunr> i haven't read it in detail yet
<braunr> it's about separating init right ?
<teythoon> yes
