I haven't tried to bootstrap a Netbsd/pmax kernel, but thought that I
might throw in a few comments anyhow:-)

First off, I would like to say that it seems to me that Jonathan et. al.
have been doing a great job and that everyone should resist doing anything
that might discourage them.

I think that, for the time being, the core team should be very liberal
w.r.t. installing changes in the pmax subtree of the kernel, even if
they aren't in philosophical agreement with the changes. (Eventually,
when the system self hosts and bootstraps without Ultrix easily, so
of the Ultrix compat stuff can come back out..) In the meantime, anything
that will make it easier for people to get a system bootstrapped is a
good idea. I would suggest that if anyone has a GNU tool chain that
works for the pmax, that they put this toolchain somewhere in the tree
until the standard NetBSD tool chain is up to the task. (Ideally the
toolchain would include binaries that run under Ultrix as well as
native NetBSD ones, but I am dreaming.)

Eventually, a program akin to Robitussin(sp?) that converts a bsd a.out
header to ECOFF may be all that is required to satisfy bootstrapping
hassles.

One final observation is that the Ultrix as knows more about "how many
delay slots are needed for special instructions like mtprX" and as such,
it may be best to keep the .s files structured so that they can be run
through the Ultrix "as", if only to check that locore.s has enough nops'
in the right places. (The macros sounded like a good idea.)

Anyhow, this got longer than I intended, so good luck, rick