> Would it not be better to use the pax TMPDIR fix, and reduce the
   > size of the kernel instead? If we do come up with a better long-term
   
   Has it been verified the reduction is sufficient?
   
   In any case, I fear it will only be a question of time before we cross
   the line again. In -current already, the INET6 libc code has topped off
   all available room in the `MD' image.

this is too true.  i had worked hard to keep INSTALL small
enough to not blow up the loader but it was always going to
be a losing battle.  note that there is only about 300k of
gzipped space free on the floppy beyond what it was in
1.4-ish (or so i recall), so it won't be long until we're
at a two floppy install ... mmm ustarfs :-)

i think the only real way to fix this is to make /boot smart
enough to map kernel memory itself.  now, doesn't that sound
fun?  sys/arch/sparc/libmmu, anyone? :-)