So, the following ports implement md_copy_filesystem():

	alpha	arm32	bebox
	i386	mac68k	pmax

All do approximately the same set of stuff: copy the root file
system's contents over to the hard disk.  In the case of many/most
installations, that's the contents of the ramdisk in the booted
installation kernel.

Is there any reason that this is done, for any of these architectures,
any more?


This type of operation made sense when the old scripts-based floppy
install method was used.  In that case, your root was on an actual
file system on the floppy and you needed to get those bits on to the
disk and reboot from it so that you could free up the floppy drive.
However, the existing ramdisk-in-kernel approach no longer ties down
the boot media, so it's not necessary for that particular situation.

Further, that approach is actually _harmful_ when combined with the
ramdisk-in-kernel approach.  a bunch of utilities built for the latest
system, but no kernel, is copied over the hard disk.  This has the
effect of rendering an otherwise happy hard disk unusable.  (see my
PR#7608.)  Now, granted, you were in the process of clobbering the
contents of the disk with an install/upgrade, so you expect that if an
error happens at a bad time you'll lose, but this seems to make the
problem much worse.

Even further, this behaviour has (or often has had, in my experience)
the annoying result of leaving some installation turds on the file
system.


Is there any reason why this function is still implemented to copy
bits to the disk on _any_ of the ports listed above?!



cgd
-- 
Chris Demetriou - cgd@netbsd.org - http://www.netbsd.org/People/Pages/cgd.html
Disclaimer: Not speaking for NetBSD, just expressing my own opinion.