On Fri, 23 Jul 1999 nigel@ind.tansu.com.au wrote:

> > Why do you need to use the actual FFS code?  If this is for a boot
> > program, use the FFS reading package in libsa.
> 
> 	I also want UFS, and ext2fs, and maybe LFS in the future.
> Having the booter do a mountroot() in exactly the same way as the
> kernel does it will be best, since device names will match, et c.

ufs isn't a real, on-disk fs anymore. It represents the code common to
ffs, lfs, and ext2fs. For instance, ffs and lfs use the same directory
format, and both use the ufs_lookup routine.

If we're talking the booter, I'd vote going for libsa. If we're talking
installer, then I'd vote doing both the same, and going your present
routie. :-)

> > You could certainly also add simple write support to the libsa FFS
> > package.  I mean, it's not like you have to interact with the buffer
> > cache, deal with the vnode layer, etc. in a boot block :-)
> 
> 	Looking at the (lack of) speed of the current Installer, the
> buffer cache is _exactly_ what we need!

How is the install-under-NetBSD track looking? I know Collin did a lot
with it for 1.4. :-)

Take care,

Bill