Re: Report from another try of running ostree

[Colin Walters <walters verbum org>]

On Wed, 2012-12-26 at 20:38 +0100, Giovanni Campagna wrote:

> Then I encountered a problem with dracut not being updated for the
> brave new multi-os world (bug 690742). I had to patch
> $(ostreeroot)/lib/dracut/modules.d/99base/init.sh to fix the variable
> NEWROOT.

I forgot to push the updated dracut patch, done now:
http://git.gnome.org/browse/gnome-ostree/commit/?id=46757138650a573e3c3d83cd306972480870c228

> After that I finally got it to boot - except that there was no GUI.
> Turns out that the yocto kernel is missing the whole drm subsystem,

Yeah, I've been making some progress.

> But things are actually worse: gnome-session-check-accelerated fails,
> which means that gdm-shell fails to gdm-fallback, but metacity is not
> installed at all, so it just fails in a loop (5 times, then it gives
> up). Obviously, given that fallback mode is no more, IsRunnableHelper
> checks need to be dropped from gnome-session (bug 690745)

Ah, thanks for that.  Yes, there's still various vestiges of the
fallback mode in the system.  See also:

https://bugzilla.gnome.org/show_bug.cgi?id=687586

> Starting it from terminal, I could only get segmentation faults. No
> other output, and my various attempts of getting a core file went no
> where - until I rebooted in Fedora today and found them in
> $(ostreeroot). Weird. Also, systemd-coredumpd appears to be installed
> and running, but I could not find the core files anywhere under /var.

You weren't getting cores because journald didn't have enough space to
store them in /run/journal (as opposed to /var/log/journal).  See:
http://git.gnome.org/browse/gnome-ostree-integration/commit/?id=ad37d32589f0861f51ca650e91669b36b3560db2
but the problem with this hack is that journald has been started
already.

If you rebooted again, it would work.  Hmm, possibly a special service
which creates /var/log/journal and  sends sighup to journald would work.

I suppose I could just ensure /var/log/journal exists when ostree admin
os-init is run, like we do for /var/log and /var/run, but it's a lot
more sustainable in general if services are responsible for the contents
of /var themselves.

> indeed swrast_dri.so is depending on
> a number of missing LLVM libraries. I filed bug 690747 for this.

I think I wasn't seeing this one as I was booting -devel, not -runtime,
and llvm-dev was pulling in all of the sublibraries.  Thanks for the
patches here!

> That's all for it. I'll try again in a week, or when you tell me it's
> usable again.

Cool.  So far I've personally been booting in qemu, not on baremetal.
And things have been working for that even with the current Yocto
standard kernel.

But I expect early next year to have a kernel, and we should be more
useful on baremetal again.