Jonathan Stone wrote:

> What will a multboot CD look like?  Will we have N kernels on the
> root, each with a ramdisk image stuffed in its belly?

That's what I've been working towards.

> [...]  If that's the
> plan, then we *cant* do what Jason wants.  if we go lookign for
> "/netbsd" in the multiboot CD, we can't tell if it's a kernel for
> Alpha, or i386, or sparc, or sun3, or whatever.
> 
> netbsd.{MACHINE} could go at the end of the searchlist, but then the
> install would get noise from every earlier element. Not good, if
> that's every new user's first handson experience with NetBSD.
> 
> Jason, what're you after? Making failure when looking for
> netbsd.${MACHINE} be silent, as the `normal' post-install case?  Or
> special bootblocks with a specail kernel searchpath for install CDs?

One idea here that takes advantage of the fact (on pmax at least) that
the cdfs filesystem code is a lot smaller than the {u,l}fs code is to
add some bootinfo-type information passing from the first stage to
the second stage.  If the second stage knows that it was booted by
bootxx_cd9660 then it could try "netbsd.${MACHINE}" first instead of
last.

I was already thinking of something similar to this so that sysinst
could realise it was booted off an install CD and could automatically
look for the sets on the boot media instead of prompting the user for
stuff that it should already know (with maybe a utility menu option to
disable this automatic behaviour).

Simon.