Re: powerpc kernel-patch 2.6.6-5 in incoming since over a month !!!
On Sat, Jun 19, 2004 at 01:02:39PM +0200, Jens Schmalzing wrote:
> Hi,
>
> Sven Luther writes:
>
> > Please tell me exactly in how far you feel that this would be so
> > much of a hindrance compared to the benefit of having kernel-image
> > installation work out of the box on non-pmac subarches ?
>
> mkvmlinuz serves the purpose of making some systems bootable given an
> uncompressed kernel and optionally an initrd. For other systems, the
> packages yaboot and quik serve the same purpose. Yet other systems
> use programs that are not packaged at all, such as BootX or a diskless
> setup or whatever.
The difference being that the vmlinux used by yaboot can be used
directly from a yaboot enabled CD or other rescue media, while for a
vmlinuz using system, it is not possible to boot the system if the
kernel was not prepared before hand, thus making the whole system
unusable to the unsuspecting user.
> While you seem to care mostly for your Pegasos board, I want PowerPC
> kernel-image packages that work equally well on all subarchs.
Well, i don't see how you are caring for anything exept new world pmac
with the current choice of things.
> Following your logic, the way to achieve this would be to
> automatically pull in all packages mentioned above and let the
> postinst pick and configure the correct one. This has bloat written
> all over it, and is unacceptable.
Well, there is not yet subarch specific depends, so there is not much we
can do here. A boot-loader or similar virtual package provided by
mkvmlinuz, yaboot or quik would be a solution here, but still. I hardly
see how a 28K unpacked script package could be considered cruft.
> So I am suggesting to let the installation system and the user take
> care of a reasonable setup. If you think the postinst should fail on
> CHRP if it doesn't detect one, fine with me. Go implement it.
No, no, no, you don't understand. It _HAS_ to work out of the box,
without further user intervention, unless you are volunteering to do
the user support i will then be facing. And anyway, even on non pegasos,
it is probable that it will only work out because more advanced users
are there.
> > If you would have told me that the real problem is that mkvmlinuz pulls
> > in binutils, which is 6.6Mo or something such, then that would be a real
> > argument
>
> It's not an argument at all. Have you forgotten that one could do
> without binutils and create very small wrapper scripts if size was an
> issue?
So, what exactly is the reason for not depending on mkvmlinuz ? you seem
to make a fanatical issue out of it, but i haven't heard yet one single
valable reason as to why to do that. And claiming that a 28k unpacked
package is cruft doesn't seem very valid to me, unless you give me some
argumentation for supporting your claims. And what would you tell me
about the 4Mo big vmlinux, which is nothing but cruft on a non yaboot
using system ?
> > Well, you were the one proposing to use the postinst_hook, where you not ?
>
> As a workaround, pending a real solution in kernel-package.
Well, let's follow up on this part in the other mail.
Friendly,
Sven Luther
Reply to: