One of the things compat_svr4(8) says is missing -- a glaring one at that --
is a pty subsystem.  This can be corrected, because the pty multiplexer is
almost trivial to implement (only two ioctls, and a little code to steal
from your plain ol' BSD pty allocator subroutine).  Then, add a /dev/pts
directory with a bunch of numbered symlinks back to /dev/tty[pqrst]*.  Maybe
a pty ioctl or three to intercept before they get to sys_ioctl()....

So... I think it should be put on the svr4_net pseudo-device, since the
major number is allocated, but what is the numbering scheme or minor devices
on svr4_net?  Who to contact to make sure it doesn't clash? 

Note of progress:  I can run CDE dynamically linked things, including
dtterm, but dtterm can't open a pty to put a shell in.  I'm going to try
Netscape for Solaris 2.5.1 tomorrow to see if it will run....  Nothing in
CDE, incidentally, should require kernel threads because this CDE release
runs on Solaris 2.4. 

Goal:  get CDE running all the way down to the dtlogin stage, such that I
can have a CDE login screen on an xterm with a "NetBSD" pixmap in place of
the usual "Solaris" pixmap.  :)  Secondary goal:  strip down the binaries
and extraneous files to make a very lean and mean Solaris emulation
subsystem and distribute it as package-plus-script-to-yank-binaries-from-CD.

...sigh, and only two weeks ago I thought of serious kernel hacking as
untouchable.  If only I'd realized how much easier it was beforehand....

=====
== Todd Vierling (Personal tv@pobox.com; Business tv@lucent.com)
== I know you like the Internet, Bobby.  Now go eat your Frosted Flakes.