On Mon, Mar 31, 2014 at 11:42:32PM +0100, Paul Sherwood wrote:
On 2014-03-31 18:19, Richard Maw wrote:
>On Mon, Mar 31, 2014 at 03:33:33PM +0100, Richard Maw wrote:
>>I'll fix the match rules and update the branch, I'll report the
>>results
>>after I've rebuilt and re-tested.
>
>Updating the splitting rules worked for me. The
>baserock/richardmaw/S10786-finer-splitting-v2 branch has been updated
>to include this fix.
I've built and booted this :-)
fwiw...
- rootfs now 23MB
- lots of stuff got rebuilt, so morph build took 76m in total (is
that right?)
The cache keys are partly derived from the splitting rules, so changing
the splitting rules for eglibc implies needing to rebuild anything that
depends on it, which is pretty much everything.
- should this really include core.morph? it includes stuff that
doesn't seem to be available in the actual result (e.g. nano)
It wouldn't be 23MB if we included core.morph; it rebuilds that, since
it's a dependency of the bsp stratum, since the kernel uses a lot of
perl in its build system, and perl is in core.
Build-time only dependencies do get rebuilt if the things they depend
on change, but they don't have to be included in the final system.
- seems to take longer to boot than a
Busybox init requires a lot of scripts, so a lot of things that could
be done in parallel in a systemd based system, can't.
- a few errors on boot, including
- can't access tty: job control turned off
This is from the shell, it's something to do with spawning a shell
directly on the console, when the shells usually expect the getty
processes to spawn them.
It's pretty harmless, so I mostly ignore it, though the cttyhack program
from busybox is supposed to fix this.
- can't run /etc/init.d/rcS: No such file or directory
This is because we don't have any services that need to be started on
boot in the minimal systems.
- do we need to include the deploy script?
If by that you mean the busybox-init.configure script, strictly: no,
we could put all this busybox config in the busybox chunk and rely on
systemd to replace the /sbin/init symlink itself, but Emmet convinced
me that this _is_ just configuration, and we may want to deploy a system
to init differently, even if it has got busybox available.